[llvm] r351682 - [CostModel][X86] Split icmp/fcmp costs tests and test all comparison codes

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Sun Jan 20 04:10:42 PST 2019


Author: rksimon
Date: Sun Jan 20 04:10:42 2019
New Revision: 351682

URL: http://llvm.org/viewvc/llvm-project?rev=351682&view=rev
Log:
[CostModel][X86] Split icmp/fcmp costs tests and test all comparison codes

Added:
    llvm/trunk/test/Analysis/CostModel/X86/fcmp.ll
      - copied, changed from r351681, llvm/trunk/test/Analysis/CostModel/X86/cmp.ll
    llvm/trunk/test/Analysis/CostModel/X86/icmp.ll
      - copied, changed from r351677, llvm/trunk/test/Analysis/CostModel/X86/cmp.ll
Removed:
    llvm/trunk/test/Analysis/CostModel/X86/cmp.ll

Removed: llvm/trunk/test/Analysis/CostModel/X86/cmp.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/cmp.ll?rev=351681&view=auto
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/cmp.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/cmp.ll (removed)
@@ -1,330 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+sse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSE3
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSSE3
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mcpu=btver2 | FileCheck %s --check-prefixes=BTVER2
-
-define i32 @cmp_float(i32 %arg) {
-; SSE2-LABEL: 'cmp_float'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE3-LABEL: 'cmp_float'
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSSE3-LABEL: 'cmp_float'
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE41-LABEL: 'cmp_float'
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE42-LABEL: 'cmp_float'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'cmp_float'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX512-LABEL: 'cmp_float'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; BTVER2-LABEL: 'cmp_float'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-  %V2F32 = fcmp olt <2 x float> undef, undef
-  %V4F32 = fcmp olt <4 x float> undef, undef
-  %V8F32 = fcmp olt <8 x float> undef, undef
-  %V16F32 = fcmp olt <16 x float> undef, undef
-
-  %V2F64 = fcmp olt <2 x double> undef, undef
-  %V4F64 = fcmp olt <4 x double> undef, undef
-  %V8F64 = fcmp olt <8 x double> undef, undef
-  %V16F64 = fcmp olt <16 x double> undef, undef
-
-  ret i32 undef
-}
-
-define i32 @cmp_int(i32 %arg) {
-; SSE2-LABEL: 'cmp_int'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE3-LABEL: 'cmp_int'
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSSE3-LABEL: 'cmp_int'
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE41-LABEL: 'cmp_int'
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE42-LABEL: 'cmp_int'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX1-LABEL: 'cmp_int'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX2-LABEL: 'cmp_int'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX512F-LABEL: 'cmp_int'
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX512BW-LABEL: 'cmp_int'
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; BTVER2-LABEL: 'cmp_int'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-  %V16I8 = icmp eq <16 x i8> undef, undef
-  %V32I8 = icmp eq <32 x i8> undef, undef
-  %V64I8 = icmp eq <64 x i8> undef, undef
-  %V128I8 = icmp eq <128 x i8> undef, undef
-
-  %V8I16  = icmp eq <8 x i16> undef, undef
-  %V16I16 = icmp eq <16 x i16> undef, undef
-  %V32I16 = icmp eq <32 x i16> undef, undef
-  %V64I16 = icmp eq <64 x i16> undef, undef
-
-  %V4I32 = icmp eq <4 x i32> undef, undef
-  %V8I32 = icmp eq <8 x i32> undef, undef
-  %V16I32 = icmp eq <16 x i32> undef, undef
-  %V32I32 = icmp eq <32 x i32> undef, undef
-
-  %V2I64 = icmp eq <2 x i64> undef, undef
-  %V4I64 = icmp eq <4 x i64> undef, undef
-  %V8I64 = icmp eq <8 x i64> undef, undef
-  %V16I64 = icmp eq <16 x i64> undef, undef
-
-  ret i32 undef
-}

Copied: llvm/trunk/test/Analysis/CostModel/X86/fcmp.ll (from r351681, llvm/trunk/test/Analysis/CostModel/X86/cmp.ll)
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/fcmp.ll?p2=llvm/trunk/test/Analysis/CostModel/X86/fcmp.ll&p1=llvm/trunk/test/Analysis/CostModel/X86/cmp.ll&r1=351681&r2=351682&rev=351682&view=diff
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/cmp.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/fcmp.ll Sun Jan 20 04:10:42 2019
@@ -13,100 +13,838 @@
 ; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
 ; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mcpu=btver2 | FileCheck %s --check-prefixes=BTVER2
 
-define i32 @cmp_float(i32 %arg) {
-; SSE2-LABEL: 'cmp_float'
+define i32 @cmp_float_oeq(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_oeq'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_oeq'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_oeq'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_oeq'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_oeq'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_oeq'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_oeq'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_oeq'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oeq float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oeq <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oeq <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp oeq <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp oeq <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oeq double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oeq <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp oeq <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp oeq <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp oeq <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp oeq float undef, undef
+  %V2F32 = fcmp oeq <2 x float> undef, undef
+  %V4F32 = fcmp oeq <4 x float> undef, undef
+  %V8F32 = fcmp oeq <8 x float> undef, undef
+  %V16F32 = fcmp oeq <16 x float> undef, undef
+
+  %F64 = fcmp oeq double undef, undef
+  %V2F64 = fcmp oeq <2 x double> undef, undef
+  %V4F64 = fcmp oeq <4 x double> undef, undef
+  %V8F64 = fcmp oeq <8 x double> undef, undef
+  %V16F64 = fcmp oeq <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_one(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_one'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_one'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_one'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_one'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_one'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_one'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_one'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_one'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp one float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp one <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp one <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp one <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp one <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp one double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp one <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp one <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp one <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp one <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp one float undef, undef
+  %V2F32 = fcmp one <2 x float> undef, undef
+  %V4F32 = fcmp one <4 x float> undef, undef
+  %V8F32 = fcmp one <8 x float> undef, undef
+  %V16F32 = fcmp one <16 x float> undef, undef
+
+  %F64 = fcmp one double undef, undef
+  %V2F64 = fcmp one <2 x double> undef, undef
+  %V4F64 = fcmp one <4 x double> undef, undef
+  %V8F64 = fcmp one <8 x double> undef, undef
+  %V16F64 = fcmp one <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_ord(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_ord'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_ord'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_ord'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_ord'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_ord'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_ord'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_ord'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_ord'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ord float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ord <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ord <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ord <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ord <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ord double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ord <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ord <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ord <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ord <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp ord float undef, undef
+  %V2F32 = fcmp ord <2 x float> undef, undef
+  %V4F32 = fcmp ord <4 x float> undef, undef
+  %V8F32 = fcmp ord <8 x float> undef, undef
+  %V16F32 = fcmp ord <16 x float> undef, undef
+
+  %F64 = fcmp ord double undef, undef
+  %V2F64 = fcmp ord <2 x double> undef, undef
+  %V4F64 = fcmp ord <4 x double> undef, undef
+  %V8F64 = fcmp ord <8 x double> undef, undef
+  %V16F64 = fcmp ord <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_oge(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_oge'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_oge'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_oge'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_oge'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_oge'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_oge'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_oge'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_oge'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp oge float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp oge <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp oge <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp oge <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp oge <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp oge double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp oge <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp oge <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp oge <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp oge <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp oge float undef, undef
+  %V2F32 = fcmp oge <2 x float> undef, undef
+  %V4F32 = fcmp oge <4 x float> undef, undef
+  %V8F32 = fcmp oge <8 x float> undef, undef
+  %V16F32 = fcmp oge <16 x float> undef, undef
+
+  %F64 = fcmp oge double undef, undef
+  %V2F64 = fcmp oge <2 x double> undef, undef
+  %V4F64 = fcmp oge <4 x double> undef, undef
+  %V8F64 = fcmp oge <8 x double> undef, undef
+  %V16F64 = fcmp oge <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_ogt(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_ogt'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_ogt'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_ogt'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_ogt'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_ogt'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_ogt'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_ogt'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_ogt'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ogt float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ogt <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ogt <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ogt <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ogt <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ogt double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ogt <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ogt <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ogt <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ogt <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp ogt float undef, undef
+  %V2F32 = fcmp ogt <2 x float> undef, undef
+  %V4F32 = fcmp ogt <4 x float> undef, undef
+  %V8F32 = fcmp ogt <8 x float> undef, undef
+  %V16F32 = fcmp ogt <16 x float> undef, undef
+
+  %F64 = fcmp ogt double undef, undef
+  %V2F64 = fcmp ogt <2 x double> undef, undef
+  %V4F64 = fcmp ogt <4 x double> undef, undef
+  %V8F64 = fcmp ogt <8 x double> undef, undef
+  %V16F64 = fcmp ogt <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_ole(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_ole'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_ole'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_ole'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_ole'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_ole'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_ole'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_ole'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_ole'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ole float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ole <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ole <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ole <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ole <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ole double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ole <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ole <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ole <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ole <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp ole float undef, undef
+  %V2F32 = fcmp ole <2 x float> undef, undef
+  %V4F32 = fcmp ole <4 x float> undef, undef
+  %V8F32 = fcmp ole <8 x float> undef, undef
+  %V16F32 = fcmp ole <16 x float> undef, undef
+
+  %F64 = fcmp ole double undef, undef
+  %V2F64 = fcmp ole <2 x double> undef, undef
+  %V4F64 = fcmp ole <4 x double> undef, undef
+  %V8F64 = fcmp ole <8 x double> undef, undef
+  %V16F64 = fcmp ole <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_olt(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_olt'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSE3-LABEL: 'cmp_float'
+; SSE3-LABEL: 'cmp_float_olt'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSSE3-LABEL: 'cmp_float'
+; SSSE3-LABEL: 'cmp_float_olt'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSE41-LABEL: 'cmp_float'
+; SSE41-LABEL: 'cmp_float_olt'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSE42-LABEL: 'cmp_float'
+; SSE42-LABEL: 'cmp_float_olt'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; AVX-LABEL: 'cmp_float'
+; AVX-LABEL: 'cmp_float_olt'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; AVX512-LABEL: 'cmp_float'
+; AVX512-LABEL: 'cmp_float_olt'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; BTVER2-LABEL: 'cmp_float'
+; BTVER2-LABEL: 'cmp_float_olt'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp olt float undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp olt double undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
+  %F32 = fcmp olt float undef, undef
   %V2F32 = fcmp olt <2 x float> undef, undef
   %V4F32 = fcmp olt <4 x float> undef, undef
   %V8F32 = fcmp olt <8 x float> undef, undef
   %V16F32 = fcmp olt <16 x float> undef, undef
 
+  %F64 = fcmp olt double undef, undef
   %V2F64 = fcmp olt <2 x double> undef, undef
   %V4F64 = fcmp olt <4 x double> undef, undef
   %V8F64 = fcmp olt <8 x double> undef, undef
@@ -115,216 +853,1082 @@ define i32 @cmp_float(i32 %arg) {
   ret i32 undef
 }
 
-define i32 @cmp_int(i32 %arg) {
-; SSE2-LABEL: 'cmp_int'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE3-LABEL: 'cmp_int'
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSSE3-LABEL: 'cmp_int'
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE41-LABEL: 'cmp_int'
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE42-LABEL: 'cmp_int'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX1-LABEL: 'cmp_int'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX2-LABEL: 'cmp_int'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX512F-LABEL: 'cmp_int'
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX512BW-LABEL: 'cmp_int'
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; BTVER2-LABEL: 'cmp_int'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-  %V16I8 = icmp eq <16 x i8> undef, undef
-  %V32I8 = icmp eq <32 x i8> undef, undef
-  %V64I8 = icmp eq <64 x i8> undef, undef
-  %V128I8 = icmp eq <128 x i8> undef, undef
-
-  %V8I16  = icmp eq <8 x i16> undef, undef
-  %V16I16 = icmp eq <16 x i16> undef, undef
-  %V32I16 = icmp eq <32 x i16> undef, undef
-  %V64I16 = icmp eq <64 x i16> undef, undef
-
-  %V4I32 = icmp eq <4 x i32> undef, undef
-  %V8I32 = icmp eq <8 x i32> undef, undef
-  %V16I32 = icmp eq <16 x i32> undef, undef
-  %V32I32 = icmp eq <32 x i32> undef, undef
-
-  %V2I64 = icmp eq <2 x i64> undef, undef
-  %V4I64 = icmp eq <4 x i64> undef, undef
-  %V8I64 = icmp eq <8 x i64> undef, undef
-  %V16I64 = icmp eq <16 x i64> undef, undef
+define i32 @cmp_float_ueq(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_ueq'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_ueq'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_ueq'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_ueq'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_ueq'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_ueq'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_ueq'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_ueq'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ueq float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ueq <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ueq <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ueq <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ueq <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ueq double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ueq <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ueq <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ueq <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ueq <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp ueq float undef, undef
+  %V2F32 = fcmp ueq <2 x float> undef, undef
+  %V4F32 = fcmp ueq <4 x float> undef, undef
+  %V8F32 = fcmp ueq <8 x float> undef, undef
+  %V16F32 = fcmp ueq <16 x float> undef, undef
+
+  %F64 = fcmp ueq double undef, undef
+  %V2F64 = fcmp ueq <2 x double> undef, undef
+  %V4F64 = fcmp ueq <4 x double> undef, undef
+  %V8F64 = fcmp ueq <8 x double> undef, undef
+  %V16F64 = fcmp ueq <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_une(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_une'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_une'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_une'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_une'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_une'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_une'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_une'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_une'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp une float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp une <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp une <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp une <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp une <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp une double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp une <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp une <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp une <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp une <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp une float undef, undef
+  %V2F32 = fcmp une <2 x float> undef, undef
+  %V4F32 = fcmp une <4 x float> undef, undef
+  %V8F32 = fcmp une <8 x float> undef, undef
+  %V16F32 = fcmp une <16 x float> undef, undef
+
+  %F64 = fcmp une double undef, undef
+  %V2F64 = fcmp une <2 x double> undef, undef
+  %V4F64 = fcmp une <4 x double> undef, undef
+  %V8F64 = fcmp une <8 x double> undef, undef
+  %V16F64 = fcmp une <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_uno(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_uno'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_uno'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_uno'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_uno'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_uno'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_uno'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_uno'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_uno'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uno float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uno <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uno <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp uno <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp uno <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uno double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uno <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp uno <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp uno <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp uno <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp uno float undef, undef
+  %V2F32 = fcmp uno <2 x float> undef, undef
+  %V4F32 = fcmp uno <4 x float> undef, undef
+  %V8F32 = fcmp uno <8 x float> undef, undef
+  %V16F32 = fcmp uno <16 x float> undef, undef
+
+  %F64 = fcmp uno double undef, undef
+  %V2F64 = fcmp uno <2 x double> undef, undef
+  %V4F64 = fcmp uno <4 x double> undef, undef
+  %V8F64 = fcmp uno <8 x double> undef, undef
+  %V16F64 = fcmp uno <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_uge(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_uge'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_uge'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_uge'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_uge'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_uge'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_uge'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_uge'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_uge'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp uge float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp uge <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp uge <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp uge <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp uge <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp uge double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp uge <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp uge <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp uge <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp uge <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp uge float undef, undef
+  %V2F32 = fcmp uge <2 x float> undef, undef
+  %V4F32 = fcmp uge <4 x float> undef, undef
+  %V8F32 = fcmp uge <8 x float> undef, undef
+  %V16F32 = fcmp uge <16 x float> undef, undef
+
+  %F64 = fcmp uge double undef, undef
+  %V2F64 = fcmp uge <2 x double> undef, undef
+  %V4F64 = fcmp uge <4 x double> undef, undef
+  %V8F64 = fcmp uge <8 x double> undef, undef
+  %V16F64 = fcmp uge <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_ugt(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_ugt'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_ugt'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_ugt'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_ugt'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_ugt'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_ugt'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_ugt'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_ugt'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ugt float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ugt <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ugt <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ugt <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ugt <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ugt double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ugt <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ugt <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ugt <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ugt <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp ugt float undef, undef
+  %V2F32 = fcmp ugt <2 x float> undef, undef
+  %V4F32 = fcmp ugt <4 x float> undef, undef
+  %V8F32 = fcmp ugt <8 x float> undef, undef
+  %V16F32 = fcmp ugt <16 x float> undef, undef
+
+  %F64 = fcmp ugt double undef, undef
+  %V2F64 = fcmp ugt <2 x double> undef, undef
+  %V4F64 = fcmp ugt <4 x double> undef, undef
+  %V8F64 = fcmp ugt <8 x double> undef, undef
+  %V16F64 = fcmp ugt <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_ule(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_ule'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_ule'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_ule'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_ule'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_ule'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_ule'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_ule'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_ule'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ule float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ule <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ule <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ule <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ule <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ule double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ule <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ule <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ule <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ule <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp ule float undef, undef
+  %V2F32 = fcmp ule <2 x float> undef, undef
+  %V4F32 = fcmp ule <4 x float> undef, undef
+  %V8F32 = fcmp ule <8 x float> undef, undef
+  %V16F32 = fcmp ule <16 x float> undef, undef
+
+  %F64 = fcmp ule double undef, undef
+  %V2F64 = fcmp ule <2 x double> undef, undef
+  %V4F64 = fcmp ule <4 x double> undef, undef
+  %V8F64 = fcmp ule <8 x double> undef, undef
+  %V16F64 = fcmp ule <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_ult(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_ult'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_ult'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_ult'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_ult'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_ult'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_ult'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_ult'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_ult'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp ult float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp ult <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp ult <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp ult <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp ult <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp ult double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp ult <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp ult <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp ult <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp ult <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp ult float undef, undef
+  %V2F32 = fcmp ult <2 x float> undef, undef
+  %V4F32 = fcmp ult <4 x float> undef, undef
+  %V8F32 = fcmp ult <8 x float> undef, undef
+  %V16F32 = fcmp ult <16 x float> undef, undef
+
+  %F64 = fcmp ult double undef, undef
+  %V2F64 = fcmp ult <2 x double> undef, undef
+  %V4F64 = fcmp ult <4 x double> undef, undef
+  %V8F64 = fcmp ult <8 x double> undef, undef
+  %V16F64 = fcmp ult <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_false(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_false'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_false'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_false'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_false'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_false'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_false'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_false'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_false'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp false float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp false <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp false <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp false <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp false <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp false double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp false <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp false <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp false <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp false <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp false float undef, undef
+  %V2F32 = fcmp false <2 x float> undef, undef
+  %V4F32 = fcmp false <4 x float> undef, undef
+  %V8F32 = fcmp false <8 x float> undef, undef
+  %V16F32 = fcmp false <16 x float> undef, undef
+
+  %F64 = fcmp false double undef, undef
+  %V2F64 = fcmp false <2 x double> undef, undef
+  %V4F64 = fcmp false <4 x double> undef, undef
+  %V8F64 = fcmp false <8 x double> undef, undef
+  %V16F64 = fcmp false <16 x double> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_float_true(i32 %arg) {
+; SSE2-LABEL: 'cmp_float_true'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_float_true'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_float_true'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_float_true'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_float_true'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX-LABEL: 'cmp_float_true'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512-LABEL: 'cmp_float_true'
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_float_true'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fcmp true float undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp true <2 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp true <4 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp true <8 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp true <16 x float> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = fcmp true double undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp true <2 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp true <4 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp true <8 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp true <16 x double> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %F32 = fcmp true float undef, undef
+  %V2F32 = fcmp true <2 x float> undef, undef
+  %V4F32 = fcmp true <4 x float> undef, undef
+  %V8F32 = fcmp true <8 x float> undef, undef
+  %V16F32 = fcmp true <16 x float> undef, undef
+
+  %F64 = fcmp true double undef, undef
+  %V2F64 = fcmp true <2 x double> undef, undef
+  %V4F64 = fcmp true <4 x double> undef, undef
+  %V8F64 = fcmp true <8 x double> undef, undef
+  %V16F64 = fcmp true <16 x double> undef, undef
 
   ret i32 undef
 }

Copied: llvm/trunk/test/Analysis/CostModel/X86/icmp.ll (from r351677, llvm/trunk/test/Analysis/CostModel/X86/cmp.ll)
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/icmp.ll?p2=llvm/trunk/test/Analysis/CostModel/X86/icmp.ll&p1=llvm/trunk/test/Analysis/CostModel/X86/cmp.ll&r1=351677&r2=351682&rev=351682&view=diff
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/cmp.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/icmp.ll Sun Jan 20 04:10:42 2019
@@ -13,314 +13,256 @@
 ; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
 ; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -cost-model -analyze -mcpu=btver2 | FileCheck %s --check-prefixes=BTVER2
 
-define i32 @cmp_float(i32 %arg) {
-; SSE2-LABEL: 'cmp_float'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE3-LABEL: 'cmp_float'
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSSE3-LABEL: 'cmp_float'
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE41-LABEL: 'cmp_float'
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; SSE42-LABEL: 'cmp_float'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'cmp_float'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX512-LABEL: 'cmp_float'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; BTVER2-LABEL: 'cmp_float'
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = fcmp olt <2 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = fcmp olt <4 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = fcmp olt <8 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = fcmp olt <16 x float> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = fcmp olt <2 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = fcmp olt <4 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F64 = fcmp olt <8 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F64 = fcmp olt <16 x double> undef, undef
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-  %V2F32 = fcmp olt <2 x float> undef, undef
-  %V4F32 = fcmp olt <4 x float> undef, undef
-  %V8F32 = fcmp olt <8 x float> undef, undef
-  %V16F32 = fcmp olt <16 x float> undef, undef
-
-  %V2F64 = fcmp olt <2 x double> undef, undef
-  %V4F64 = fcmp olt <4 x double> undef, undef
-  %V8F64 = fcmp olt <8 x double> undef, undef
-  %V16F64 = fcmp olt <16 x double> undef, undef
-
-  ret i32 undef
-}
-
-define i32 @cmp_int(i32 %arg) {
-; SSE2-LABEL: 'cmp_int'
+define i32 @cmp_int_eq(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_eq'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSE3-LABEL: 'cmp_int'
+; SSE3-LABEL: 'cmp_int_eq'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSSE3-LABEL: 'cmp_int'
+; SSSE3-LABEL: 'cmp_int_eq'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSE41-LABEL: 'cmp_int'
+; SSE41-LABEL: 'cmp_int_eq'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; SSE42-LABEL: 'cmp_int'
+; SSE42-LABEL: 'cmp_int_eq'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; AVX1-LABEL: 'cmp_int'
+; AVX1-LABEL: 'cmp_int_eq'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; AVX2-LABEL: 'cmp_int'
+; AVX2-LABEL: 'cmp_int_eq'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; AVX512F-LABEL: 'cmp_int'
+; AVX512F-LABEL: 'cmp_int_eq'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; AVX512BW-LABEL: 'cmp_int'
+; AVX512BW-LABEL: 'cmp_int_eq'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
-; BTVER2-LABEL: 'cmp_int'
+; BTVER2-LABEL: 'cmp_int_eq'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp eq i8 undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp eq <16 x i8> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp eq <32 x i8> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp eq <64 x i8> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp eq <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp eq i16 undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp eq <8 x i16> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp eq <16 x i16> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp eq <32 x i16> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp eq <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp eq i32 undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp eq <4 x i32> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp eq <8 x i32> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp eq <16 x i32> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp eq <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp eq i64 undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp eq <2 x i64> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp eq <4 x i64> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp eq <8 x i64> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp eq <16 x i64> undef, undef
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
+  %I8 = icmp eq i8 undef, undef
   %V16I8 = icmp eq <16 x i8> undef, undef
   %V32I8 = icmp eq <32 x i8> undef, undef
   %V64I8 = icmp eq <64 x i8> undef, undef
   %V128I8 = icmp eq <128 x i8> undef, undef
 
+  %I16 = icmp eq i16 undef, undef
   %V8I16  = icmp eq <8 x i16> undef, undef
   %V16I16 = icmp eq <16 x i16> undef, undef
   %V32I16 = icmp eq <32 x i16> undef, undef
   %V64I16 = icmp eq <64 x i16> undef, undef
 
+  %I32 = icmp eq i32 undef, undef
   %V4I32 = icmp eq <4 x i32> undef, undef
   %V8I32 = icmp eq <8 x i32> undef, undef
   %V16I32 = icmp eq <16 x i32> undef, undef
   %V32I32 = icmp eq <32 x i32> undef, undef
 
+  %I64 = icmp eq i64 undef, undef
   %V2I64 = icmp eq <2 x i64> undef, undef
   %V4I64 = icmp eq <4 x i64> undef, undef
   %V8I64 = icmp eq <8 x i64> undef, undef
@@ -328,3 +270,2326 @@ define i32 @cmp_int(i32 %arg) {
 
   ret i32 undef
 }
+
+define i32 @cmp_int_ne(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_ne'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_ne'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_ne'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_ne'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_ne'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_ne'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_ne'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_ne'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_ne'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_ne'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ne i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ne <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ne <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ne <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ne <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ne i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ne <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ne <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ne <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ne <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ne i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ne <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ne <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ne <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ne <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ne i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ne <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ne <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ne <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ne <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp ne i8 undef, undef
+  %V16I8 = icmp ne <16 x i8> undef, undef
+  %V32I8 = icmp ne <32 x i8> undef, undef
+  %V64I8 = icmp ne <64 x i8> undef, undef
+  %V128I8 = icmp ne <128 x i8> undef, undef
+
+  %I16 = icmp ne i16 undef, undef
+  %V8I16  = icmp ne <8 x i16> undef, undef
+  %V16I16 = icmp ne <16 x i16> undef, undef
+  %V32I16 = icmp ne <32 x i16> undef, undef
+  %V64I16 = icmp ne <64 x i16> undef, undef
+
+  %I32 = icmp ne i32 undef, undef
+  %V4I32 = icmp ne <4 x i32> undef, undef
+  %V8I32 = icmp ne <8 x i32> undef, undef
+  %V16I32 = icmp ne <16 x i32> undef, undef
+  %V32I32 = icmp ne <32 x i32> undef, undef
+
+  %I64 = icmp ne i64 undef, undef
+  %V2I64 = icmp ne <2 x i64> undef, undef
+  %V4I64 = icmp ne <4 x i64> undef, undef
+  %V8I64 = icmp ne <8 x i64> undef, undef
+  %V16I64 = icmp ne <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_sge(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_sge'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_sge'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_sge'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_sge'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_sge'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_sge'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_sge'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_sge'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_sge'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_sge'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sge i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sge <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp sge <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp sge <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp sge <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sge i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sge <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp sge <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp sge <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp sge <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sge i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sge <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp sge <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp sge <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp sge <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sge i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sge <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp sge <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp sge <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp sge <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp sge i8 undef, undef
+  %V16I8 = icmp sge <16 x i8> undef, undef
+  %V32I8 = icmp sge <32 x i8> undef, undef
+  %V64I8 = icmp sge <64 x i8> undef, undef
+  %V128I8 = icmp sge <128 x i8> undef, undef
+
+  %I16 = icmp sge i16 undef, undef
+  %V8I16  = icmp sge <8 x i16> undef, undef
+  %V16I16 = icmp sge <16 x i16> undef, undef
+  %V32I16 = icmp sge <32 x i16> undef, undef
+  %V64I16 = icmp sge <64 x i16> undef, undef
+
+  %I32 = icmp sge i32 undef, undef
+  %V4I32 = icmp sge <4 x i32> undef, undef
+  %V8I32 = icmp sge <8 x i32> undef, undef
+  %V16I32 = icmp sge <16 x i32> undef, undef
+  %V32I32 = icmp sge <32 x i32> undef, undef
+
+  %I64 = icmp sge i64 undef, undef
+  %V2I64 = icmp sge <2 x i64> undef, undef
+  %V4I64 = icmp sge <4 x i64> undef, undef
+  %V8I64 = icmp sge <8 x i64> undef, undef
+  %V16I64 = icmp sge <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_uge(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_uge'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_uge'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_uge'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_uge'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_uge'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_uge'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_uge'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_uge'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_uge'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_uge'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp uge i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp uge <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp uge <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp uge <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp uge <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp uge i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp uge <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp uge <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp uge <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp uge <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp uge i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp uge <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp uge <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp uge <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp uge <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp uge i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp uge <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp uge <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp uge <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp uge <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp uge i8 undef, undef
+  %V16I8 = icmp uge <16 x i8> undef, undef
+  %V32I8 = icmp uge <32 x i8> undef, undef
+  %V64I8 = icmp uge <64 x i8> undef, undef
+  %V128I8 = icmp uge <128 x i8> undef, undef
+
+  %I16 = icmp uge i16 undef, undef
+  %V8I16  = icmp uge <8 x i16> undef, undef
+  %V16I16 = icmp uge <16 x i16> undef, undef
+  %V32I16 = icmp uge <32 x i16> undef, undef
+  %V64I16 = icmp uge <64 x i16> undef, undef
+
+  %I32 = icmp uge i32 undef, undef
+  %V4I32 = icmp uge <4 x i32> undef, undef
+  %V8I32 = icmp uge <8 x i32> undef, undef
+  %V16I32 = icmp uge <16 x i32> undef, undef
+  %V32I32 = icmp uge <32 x i32> undef, undef
+
+  %I64 = icmp uge i64 undef, undef
+  %V2I64 = icmp uge <2 x i64> undef, undef
+  %V4I64 = icmp uge <4 x i64> undef, undef
+  %V8I64 = icmp uge <8 x i64> undef, undef
+  %V16I64 = icmp uge <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_sgt(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_sgt'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_sgt'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_sgt'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_sgt'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_sgt'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_sgt'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_sgt'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_sgt'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_sgt'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_sgt'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sgt i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sgt <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp sgt <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp sgt <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp sgt <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sgt i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sgt <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp sgt <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp sgt <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp sgt <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sgt i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sgt <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp sgt <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp sgt <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp sgt <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sgt i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sgt <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp sgt <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp sgt <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp sgt <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp sgt i8 undef, undef
+  %V16I8 = icmp sgt <16 x i8> undef, undef
+  %V32I8 = icmp sgt <32 x i8> undef, undef
+  %V64I8 = icmp sgt <64 x i8> undef, undef
+  %V128I8 = icmp sgt <128 x i8> undef, undef
+
+  %I16 = icmp sgt i16 undef, undef
+  %V8I16  = icmp sgt <8 x i16> undef, undef
+  %V16I16 = icmp sgt <16 x i16> undef, undef
+  %V32I16 = icmp sgt <32 x i16> undef, undef
+  %V64I16 = icmp sgt <64 x i16> undef, undef
+
+  %I32 = icmp sgt i32 undef, undef
+  %V4I32 = icmp sgt <4 x i32> undef, undef
+  %V8I32 = icmp sgt <8 x i32> undef, undef
+  %V16I32 = icmp sgt <16 x i32> undef, undef
+  %V32I32 = icmp sgt <32 x i32> undef, undef
+
+  %I64 = icmp sgt i64 undef, undef
+  %V2I64 = icmp sgt <2 x i64> undef, undef
+  %V4I64 = icmp sgt <4 x i64> undef, undef
+  %V8I64 = icmp sgt <8 x i64> undef, undef
+  %V16I64 = icmp sgt <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_ugt(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_ugt'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_ugt'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_ugt'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_ugt'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_ugt'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_ugt'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_ugt'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_ugt'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_ugt'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_ugt'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ugt i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ugt <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ugt <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ugt <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ugt <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ugt i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ugt <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ugt <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ugt <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ugt <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ugt i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ugt <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ugt <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ugt <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ugt <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ugt i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ugt <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ugt <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ugt <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ugt <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp ugt i8 undef, undef
+  %V16I8 = icmp ugt <16 x i8> undef, undef
+  %V32I8 = icmp ugt <32 x i8> undef, undef
+  %V64I8 = icmp ugt <64 x i8> undef, undef
+  %V128I8 = icmp ugt <128 x i8> undef, undef
+
+  %I16 = icmp ugt i16 undef, undef
+  %V8I16  = icmp ugt <8 x i16> undef, undef
+  %V16I16 = icmp ugt <16 x i16> undef, undef
+  %V32I16 = icmp ugt <32 x i16> undef, undef
+  %V64I16 = icmp ugt <64 x i16> undef, undef
+
+  %I32 = icmp ugt i32 undef, undef
+  %V4I32 = icmp ugt <4 x i32> undef, undef
+  %V8I32 = icmp ugt <8 x i32> undef, undef
+  %V16I32 = icmp ugt <16 x i32> undef, undef
+  %V32I32 = icmp ugt <32 x i32> undef, undef
+
+  %I64 = icmp ugt i64 undef, undef
+  %V2I64 = icmp ugt <2 x i64> undef, undef
+  %V4I64 = icmp ugt <4 x i64> undef, undef
+  %V8I64 = icmp ugt <8 x i64> undef, undef
+  %V16I64 = icmp ugt <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_sle(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_sle'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_sle'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_sle'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_sle'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_sle'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_sle'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_sle'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_sle'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_sle'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_sle'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp sle i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp sle <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp sle <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp sle <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp sle <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp sle i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp sle <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp sle <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp sle <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp sle <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp sle i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp sle <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp sle <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp sle <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp sle <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp sle i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp sle <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp sle <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp sle <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp sle <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp sle i8 undef, undef
+  %V16I8 = icmp sle <16 x i8> undef, undef
+  %V32I8 = icmp sle <32 x i8> undef, undef
+  %V64I8 = icmp sle <64 x i8> undef, undef
+  %V128I8 = icmp sle <128 x i8> undef, undef
+
+  %I16 = icmp sle i16 undef, undef
+  %V8I16  = icmp sle <8 x i16> undef, undef
+  %V16I16 = icmp sle <16 x i16> undef, undef
+  %V32I16 = icmp sle <32 x i16> undef, undef
+  %V64I16 = icmp sle <64 x i16> undef, undef
+
+  %I32 = icmp sle i32 undef, undef
+  %V4I32 = icmp sle <4 x i32> undef, undef
+  %V8I32 = icmp sle <8 x i32> undef, undef
+  %V16I32 = icmp sle <16 x i32> undef, undef
+  %V32I32 = icmp sle <32 x i32> undef, undef
+
+  %I64 = icmp sle i64 undef, undef
+  %V2I64 = icmp sle <2 x i64> undef, undef
+  %V4I64 = icmp sle <4 x i64> undef, undef
+  %V8I64 = icmp sle <8 x i64> undef, undef
+  %V16I64 = icmp sle <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_ule(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_ule'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_ule'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_ule'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_ule'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_ule'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_ule'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_ule'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_ule'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_ule'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_ule'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ule i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ule <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ule <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ule <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ule <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ule i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ule <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ule <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ule <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ule <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ule i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ule <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ule <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ule <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ule <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ule i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ule <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ule <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ule <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ule <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp ule i8 undef, undef
+  %V16I8 = icmp ule <16 x i8> undef, undef
+  %V32I8 = icmp ule <32 x i8> undef, undef
+  %V64I8 = icmp ule <64 x i8> undef, undef
+  %V128I8 = icmp ule <128 x i8> undef, undef
+
+  %I16 = icmp ule i16 undef, undef
+  %V8I16  = icmp ule <8 x i16> undef, undef
+  %V16I16 = icmp ule <16 x i16> undef, undef
+  %V32I16 = icmp ule <32 x i16> undef, undef
+  %V64I16 = icmp ule <64 x i16> undef, undef
+
+  %I32 = icmp ule i32 undef, undef
+  %V4I32 = icmp ule <4 x i32> undef, undef
+  %V8I32 = icmp ule <8 x i32> undef, undef
+  %V16I32 = icmp ule <16 x i32> undef, undef
+  %V32I32 = icmp ule <32 x i32> undef, undef
+
+  %I64 = icmp ule i64 undef, undef
+  %V2I64 = icmp ule <2 x i64> undef, undef
+  %V4I64 = icmp ule <4 x i64> undef, undef
+  %V8I64 = icmp ule <8 x i64> undef, undef
+  %V16I64 = icmp ule <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_slt(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_slt'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_slt'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_slt'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_slt'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_slt'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_slt'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_slt'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_slt'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_slt'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_slt'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp slt i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp slt <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp slt <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp slt <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp slt <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp slt i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp slt <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp slt <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp slt <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp slt <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp slt i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp slt <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp slt <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp slt <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp slt <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp slt i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp slt <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp slt <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp slt <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp slt <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp slt i8 undef, undef
+  %V16I8 = icmp slt <16 x i8> undef, undef
+  %V32I8 = icmp slt <32 x i8> undef, undef
+  %V64I8 = icmp slt <64 x i8> undef, undef
+  %V128I8 = icmp slt <128 x i8> undef, undef
+
+  %I16 = icmp slt i16 undef, undef
+  %V8I16  = icmp slt <8 x i16> undef, undef
+  %V16I16 = icmp slt <16 x i16> undef, undef
+  %V32I16 = icmp slt <32 x i16> undef, undef
+  %V64I16 = icmp slt <64 x i16> undef, undef
+
+  %I32 = icmp slt i32 undef, undef
+  %V4I32 = icmp slt <4 x i32> undef, undef
+  %V8I32 = icmp slt <8 x i32> undef, undef
+  %V16I32 = icmp slt <16 x i32> undef, undef
+  %V32I32 = icmp slt <32 x i32> undef, undef
+
+  %I64 = icmp slt i64 undef, undef
+  %V2I64 = icmp slt <2 x i64> undef, undef
+  %V4I64 = icmp slt <4 x i64> undef, undef
+  %V8I64 = icmp slt <8 x i64> undef, undef
+  %V16I64 = icmp slt <16 x i64> undef, undef
+
+  ret i32 undef
+}
+
+define i32 @cmp_int_ult(i32 %arg) {
+; SSE2-LABEL: 'cmp_int_ult'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE3-LABEL: 'cmp_int_ult'
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'cmp_int_ult'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE41-LABEL: 'cmp_int_ult'
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; SSE41-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'cmp_int_ult'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'cmp_int_ult'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'cmp_int_ult'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'cmp_int_ult'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'cmp_int_ult'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; BTVER2-LABEL: 'cmp_int_ult'
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = icmp ult i8 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = icmp ult <16 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = icmp ult <32 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = icmp ult <64 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V128I8 = icmp ult <128 x i8> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = icmp ult i16 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = icmp ult <8 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = icmp ult <16 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = icmp ult <32 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64I16 = icmp ult <64 x i16> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = icmp ult i32 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = icmp ult <4 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = icmp ult <8 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16I32 = icmp ult <16 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V32I32 = icmp ult <32 x i32> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = icmp ult i64 undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = icmp ult <2 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = icmp ult <4 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = icmp ult <8 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16I64 = icmp ult <16 x i64> undef, undef
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+  %I8 = icmp ult i8 undef, undef
+  %V16I8 = icmp ult <16 x i8> undef, undef
+  %V32I8 = icmp ult <32 x i8> undef, undef
+  %V64I8 = icmp ult <64 x i8> undef, undef
+  %V128I8 = icmp ult <128 x i8> undef, undef
+
+  %I16 = icmp ult i16 undef, undef
+  %V8I16  = icmp ult <8 x i16> undef, undef
+  %V16I16 = icmp ult <16 x i16> undef, undef
+  %V32I16 = icmp ult <32 x i16> undef, undef
+  %V64I16 = icmp ult <64 x i16> undef, undef
+
+  %I32 = icmp ult i32 undef, undef
+  %V4I32 = icmp ult <4 x i32> undef, undef
+  %V8I32 = icmp ult <8 x i32> undef, undef
+  %V16I32 = icmp ult <16 x i32> undef, undef
+  %V32I32 = icmp ult <32 x i32> undef, undef
+
+  %I64 = icmp ult i64 undef, undef
+  %V2I64 = icmp ult <2 x i64> undef, undef
+  %V4I64 = icmp ult <4 x i64> undef, undef
+  %V8I64 = icmp ult <8 x i64> undef, undef
+  %V16I64 = icmp ult <16 x i64> undef, undef
+
+  ret i32 undef
+}
+




More information about the llvm-commits mailing list