[llvm] r329401 - [CostModel][X86] Regenerate integer division/remainder tests with update_analyze_test_checks.py
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 6 08:23:26 PDT 2018
Author: rksimon
Date: Fri Apr 6 08:23:26 2018
New Revision: 329401
URL: http://llvm.org/viewvc/llvm-project?rev=329401&view=rev
Log:
[CostModel][X86] Regenerate integer division/remainder tests with update_analyze_test_checks.py
Modified:
llvm/trunk/test/Analysis/CostModel/X86/div.ll
llvm/trunk/test/Analysis/CostModel/X86/rem.ll
llvm/trunk/test/Analysis/CostModel/X86/vdiv-cost.ll
Modified: llvm/trunk/test/Analysis/CostModel/X86/div.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/div.ll?rev=329401&r1=329400&r2=329401&view=diff
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/div.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/div.ll Fri Apr 6 08:23:26 2018
@@ -1,375 +1,645 @@
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE2
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+ssse3 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSSE3
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+sse4.2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE42
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX1
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX2
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512BW
+; 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=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSSE3
+; 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
-target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
-target triple = "x86_64-apple-macosx10.8.0"
-
-; CHECK-LABEL: 'sdiv'
define i32 @sdiv() {
- ; CHECK: cost of 1 {{.*}} %I64 = sdiv
+; CHECK-LABEL: 'sdiv'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i32 = sdiv <4 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i32 = sdiv <8 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i32 = sdiv <16 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i16 = sdiv <8 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i16 = sdiv <16 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i16 = sdiv <32 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = sdiv i64 undef, undef
- ; SSE: cost of 40 {{.*}} %V2i64 = sdiv
- ; AVX: cost of 40 {{.*}} %V2i64 = sdiv
%V2i64 = sdiv <2 x i64> undef, undef
- ; SSE: cost of 80 {{.*}} %V4i64 = sdiv
- ; AVX: cost of 80 {{.*}} %V4i64 = sdiv
%V4i64 = sdiv <4 x i64> undef, undef
- ; SSE: cost of 160 {{.*}} %V8i64 = sdiv
- ; AVX: cost of 160 {{.*}} %V8i64 = sdiv
%V8i64 = sdiv <8 x i64> undef, undef
- ; CHECK: cost of 1 {{.*}} %I32 = sdiv
%I32 = sdiv i32 undef, undef
- ; SSE: cost of 80 {{.*}} %V4i32 = sdiv
- ; AVX: cost of 80 {{.*}} %V4i32 = sdiv
%V4i32 = sdiv <4 x i32> undef, undef
- ; SSE: cost of 160 {{.*}} %V8i32 = sdiv
- ; AVX: cost of 160 {{.*}} %V8i32 = sdiv
%V8i32 = sdiv <8 x i32> undef, undef
- ; SSE: cost of 320 {{.*}} %V16i32 = sdiv
- ; AVX: cost of 320 {{.*}} %V16i32 = sdiv
%V16i32 = sdiv <16 x i32> undef, undef
- ; CHECK: cost of 1 {{.*}} %I16 = sdiv
%I16 = sdiv i16 undef, undef
- ; SSE: cost of 160 {{.*}} %V8i16 = sdiv
- ; AVX: cost of 160 {{.*}} %V8i16 = sdiv
%V8i16 = sdiv <8 x i16> undef, undef
- ; SSE: cost of 320 {{.*}} %V16i16 = sdiv
- ; AVX: cost of 320 {{.*}} %V16i16 = sdiv
%V16i16 = sdiv <16 x i16> undef, undef
- ; SSE: cost of 640 {{.*}} %V32i16 = sdiv
- ; AVX: cost of 640 {{.*}} %V32i16 = sdiv
%V32i16 = sdiv <32 x i16> undef, undef
- ; CHECK: cost of 1 {{.*}} %I8 = sdiv
%I8 = sdiv i8 undef, undef
- ; SSE: cost of 320 {{.*}} %V16i8 = sdiv
- ; AVX: cost of 320 {{.*}} %V16i8 = sdiv
%V16i8 = sdiv <16 x i8> undef, undef
- ; SSE: cost of 640 {{.*}} %V32i8 = sdiv
- ; AVX: cost of 640 {{.*}} %V32i8 = sdiv
%V32i8 = sdiv <32 x i8> undef, undef
- ; SSE: cost of 1280 {{.*}} %V64i8 = sdiv
- ; AVX: cost of 1280 {{.*}} %V64i8 = sdiv
%V64i8 = sdiv <64 x i8> undef, undef
ret i32 undef
}
-; CHECK-LABEL: 'udiv'
define i32 @udiv() {
- ; CHECK: cost of 1 {{.*}} %I64 = udiv
+; CHECK-LABEL: 'udiv'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i32 = udiv <4 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i32 = udiv <8 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i32 = udiv <16 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i16 = udiv <8 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i16 = udiv <16 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i16 = udiv <32 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = udiv i64 undef, undef
- ; SSE: cost of 40 {{.*}} %V2i64 = udiv
- ; AVX: cost of 40 {{.*}} %V2i64 = udiv
%V2i64 = udiv <2 x i64> undef, undef
- ; SSE: cost of 80 {{.*}} %V4i64 = udiv
- ; AVX: cost of 80 {{.*}} %V4i64 = udiv
%V4i64 = udiv <4 x i64> undef, undef
- ; SSE: cost of 160 {{.*}} %V8i64 = udiv
- ; AVX: cost of 160 {{.*}} %V8i64 = udiv
%V8i64 = udiv <8 x i64> undef, undef
- ; CHECK: cost of 1 {{.*}} %I32 = udiv
%I32 = udiv i32 undef, undef
- ; SSE: cost of 80 {{.*}} %V4i32 = udiv
- ; AVX: cost of 80 {{.*}} %V4i32 = udiv
%V4i32 = udiv <4 x i32> undef, undef
- ; SSE: cost of 160 {{.*}} %V8i32 = udiv
- ; AVX: cost of 160 {{.*}} %V8i32 = udiv
%V8i32 = udiv <8 x i32> undef, undef
- ; SSE: cost of 320 {{.*}} %V16i32 = udiv
- ; AVX: cost of 320 {{.*}} %V16i32 = udiv
%V16i32 = udiv <16 x i32> undef, undef
- ; CHECK: cost of 1 {{.*}} %I16 = udiv
%I16 = udiv i16 undef, undef
- ; SSE: cost of 160 {{.*}} %V8i16 = udiv
- ; AVX: cost of 160 {{.*}} %V8i16 = udiv
%V8i16 = udiv <8 x i16> undef, undef
- ; SSE: cost of 320 {{.*}} %V16i16 = udiv
- ; AVX: cost of 320 {{.*}} %V16i16 = udiv
%V16i16 = udiv <16 x i16> undef, undef
- ; SSE: cost of 640 {{.*}} %V32i16 = udiv
- ; AVX: cost of 640 {{.*}} %V32i16 = udiv
%V32i16 = udiv <32 x i16> undef, undef
- ; CHECK: cost of 1 {{.*}} %I8 = udiv
%I8 = udiv i8 undef, undef
- ; SSE: cost of 320 {{.*}} %V16i8 = udiv
- ; AVX: cost of 320 {{.*}} %V16i8 = udiv
%V16i8 = udiv <16 x i8> undef, undef
- ; SSE: cost of 640 {{.*}} %V32i8 = udiv
- ; AVX: cost of 640 {{.*}} %V32i8 = udiv
%V32i8 = udiv <32 x i8> undef, undef
- ; SSE: cost of 1280 {{.*}} %V64i8 = udiv
- ; AVX: cost of 1280 {{.*}} %V64i8 = udiv
%V64i8 = udiv <64 x i8> undef, undef
ret i32 undef
}
-; CHECK-LABEL: 'sdiv_uniformconst'
define i32 @sdiv_uniformconst() {
- ; CHECK: cost of 1 {{.*}} %I64 = sdiv
+; SSE2-LABEL: 'sdiv_uniformconst'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 7
+; SSE2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 7
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 7
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 7
+; SSE2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'sdiv_uniformconst'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 7
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 7
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 7
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 7
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sdiv_uniformconst'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 7
+; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 7
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 7
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 7
+; SSE42-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'sdiv_uniformconst'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'sdiv_uniformconst'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'sdiv_uniformconst'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'sdiv_uniformconst'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = sdiv i64 undef, 7
- ; SSE: cost of 40 {{.*}} %V2i64 = sdiv
- ; AVX: cost of 40 {{.*}} %V2i64 = sdiv
%V2i64 = sdiv <2 x i64> undef, <i64 7, i64 7>
- ; SSE: cost of 80 {{.*}} %V4i64 = sdiv
- ; AVX: cost of 80 {{.*}} %V4i64 = sdiv
%V4i64 = sdiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
- ; SSE: cost of 160 {{.*}} %V8i64 = sdiv
- ; AVX: cost of 160 {{.*}} %V8i64 = sdiv
%V8i64 = sdiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
- ; CHECK: cost of 1 {{.*}} %I32 = sdiv
%I32 = sdiv i32 undef, 7
- ; SSE2: cost of 19 {{.*}} %V4i32 = sdiv
- ; SSSE3: cost of 19 {{.*}} %V4i32 = sdiv
- ; SSE42: cost of 15 {{.*}} %V4i32 = sdiv
- ; AVX: cost of 15 {{.*}} %V4i32 = sdiv
%V4i32 = sdiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
- ; SSE2: cost of 38 {{.*}} %V8i32 = sdiv
- ; SSSE3: cost of 38 {{.*}} %V8i32 = sdiv
- ; SSE42: cost of 30 {{.*}} %V8i32 = sdiv
- ; AVX1: cost of 32 {{.*}} %V8i32 = sdiv
- ; AVX2: cost of 15 {{.*}} %V8i32 = sdiv
- ; AVX512: cost of 15 {{.*}} %V8i32 = sdiv
%V8i32 = sdiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
- ; SSE2: cost of 76 {{.*}} %V16i32 = sdiv
- ; SSSE3: cost of 76 {{.*}} %V16i32 = sdiv
- ; SSE42: cost of 60 {{.*}} %V16i32 = sdiv
- ; AVX1: cost of 64 {{.*}} %V16i32 = sdiv
- ; AVX2: cost of 30 {{.*}} %V16i32 = sdiv
- ; AVX512: cost of 15 {{.*}} %V16i32 = sdiv
%V16i32 = sdiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
- ; CHECK: cost of 1 {{.*}} %I16 = sdiv
%I16 = sdiv i16 undef, 7
- ; SSE: cost of 6 {{.*}} %V8i16 = sdiv
- ; AVX: cost of 6 {{.*}} %V8i16 = sdiv
%V8i16 = sdiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
- ; SSE: cost of 12 {{.*}} %V16i16 = sdiv
- ; AVX1: cost of 14 {{.*}} %V16i16 = sdiv
- ; AVX2: cost of 6 {{.*}} %V16i16 = sdiv
- ; AVX512: cost of 6 {{.*}} %V16i16 = sdiv
%V16i16 = sdiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
- ; SSE: cost of 24 {{.*}} %V32i16 = sdiv
- ; AVX1: cost of 28 {{.*}} %V32i16 = sdiv
- ; AVX2: cost of 12 {{.*}} %V32i16 = sdiv
- ; AVX512F: cost of 12 {{.*}} %V32i16 = sdiv
- ; AVX512BW: cost of 6 {{.*}} %V32i16 = sdiv
%V32i16 = sdiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
- ; CHECK: cost of 1 {{.*}} %I8 = sdiv
%I8 = sdiv i8 undef, 7
- ; SSE: cost of 320 {{.*}} %V16i8 = sdiv
- ; AVX: cost of 320 {{.*}} %V16i8 = sdiv
%V16i8 = sdiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
- ; SSE: cost of 640 {{.*}} %V32i8 = sdiv
- ; AVX: cost of 640 {{.*}} %V32i8 = sdiv
%V32i8 = sdiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
- ; SSE: cost of 1280 {{.*}} %V64i8 = sdiv
- ; AVX: cost of 1280 {{.*}} %V64i8 = sdiv
%V64i8 = sdiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
ret i32 undef
}
-; CHECK-LABEL: 'udiv_uniformconst'
define i32 @udiv_uniformconst() {
- ; CHECK: cost of 1 {{.*}} %I64 = udiv
+; SSE-LABEL: 'udiv_uniformconst'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 7
+; SSE-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 7
+; SSE-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 7
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 7
+; SSE-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'udiv_uniformconst'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 7, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 7
+; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'udiv_uniformconst'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 7, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 7
+; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'udiv_uniformconst'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 7, i64 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 7
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'udiv_uniformconst'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 7, i64 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 7
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = udiv i64 undef, 7
- ; SSE: cost of 40 {{.*}} %V2i64 = udiv
- ; AVX: cost of 40 {{.*}} %V2i64 = udiv
%V2i64 = udiv <2 x i64> undef, <i64 7, i64 7>
- ; SSE: cost of 80 {{.*}} %V4i64 = udiv
- ; AVX: cost of 80 {{.*}} %V4i64 = udiv
%V4i64 = udiv <4 x i64> undef, <i64 7, i64 7, i64 7, i64 7>
- ; SSE: cost of 160 {{.*}} %V8i64 = udiv
- ; AVX: cost of 160 {{.*}} %V8i64 = udiv
%V8i64 = udiv <8 x i64> undef, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
- ; CHECK: cost of 1 {{.*}} %I32 = udiv
%I32 = udiv i32 undef, 7
- ; SSE: cost of 15 {{.*}} %V4i32 = udiv
- ; AVX: cost of 15 {{.*}} %V4i32 = udiv
%V4i32 = udiv <4 x i32> undef, <i32 7, i32 7, i32 7, i32 7>
- ; SSE: cost of 30 {{.*}} %V8i32 = udiv
- ; AVX1: cost of 32 {{.*}} %V8i32 = udiv
- ; AVX2: cost of 15 {{.*}} %V8i32 = udiv
- ; AVX512: cost of 15 {{.*}} %V8i32 = udiv
%V8i32 = udiv <8 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
- ; SSE: cost of 60 {{.*}} %V16i32 = udiv
- ; AVX1: cost of 64 {{.*}} %V16i32 = udiv
- ; AVX2: cost of 30 {{.*}} %V16i32 = udiv
- ; AVX512: cost of 15 {{.*}} %V16i32 = udiv
%V16i32 = udiv <16 x i32> undef, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
- ; CHECK: cost of 1 {{.*}} %I16 = udiv
%I16 = udiv i16 undef, 7
- ; SSE: cost of 6 {{.*}} %V8i16 = udiv
- ; AVX: cost of 6 {{.*}} %V8i16 = udiv
%V8i16 = udiv <8 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
- ; SSE: cost of 12 {{.*}} %V16i16 = udiv
- ; AVX1: cost of 14 {{.*}} %V16i16 = udiv
- ; AVX2: cost of 6 {{.*}} %V16i16 = udiv
- ; AVX512: cost of 6 {{.*}} %V16i16 = udiv
%V16i16 = udiv <16 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
- ; SSE: cost of 24 {{.*}} %V32i16 = udiv
- ; AVX1: cost of 28 {{.*}} %V32i16 = udiv
- ; AVX2: cost of 12 {{.*}} %V32i16 = udiv
- ; AVX512F: cost of 12 {{.*}} %V32i16 = udiv
- ; AVX512BW: cost of 6 {{.*}} %V32i16 = udiv
%V32i16 = udiv <32 x i16> undef, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
- ; CHECK: cost of 1 {{.*}} %I8 = udiv
%I8 = udiv i8 undef, 7
- ; SSE: cost of 320 {{.*}} %V16i8 = udiv
- ; AVX: cost of 320 {{.*}} %V16i8 = udiv
%V16i8 = udiv <16 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
- ; SSE: cost of 640 {{.*}} %V32i8 = udiv
- ; AVX: cost of 640 {{.*}} %V32i8 = udiv
%V32i8 = udiv <32 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
- ; SSE: cost of 1280 {{.*}} %V64i8 = udiv
- ; AVX: cost of 1280 {{.*}} %V64i8 = udiv
%V64i8 = udiv <64 x i8> undef, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
ret i32 undef
}
-; CHECK-LABEL: 'sdiv_uniformconstpow2'
define i32 @sdiv_uniformconstpow2() {
- ; CHECK: cost of 1 {{.*}} %I64 = sdiv
+; SSE2-LABEL: 'sdiv_uniformconstpow2'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSSE3-LABEL: 'sdiv_uniformconstpow2'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 16
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 16
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 16
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 16
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sdiv_uniformconstpow2'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'sdiv_uniformconstpow2'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'sdiv_uniformconstpow2'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'sdiv_uniformconstpow2'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'sdiv_uniformconstpow2'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = sdiv i64 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = sdiv i32 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = sdiv i16 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = sdiv i8 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = sdiv i64 undef, 16
- ; SSE: cost of 40 {{.*}} %V2i64 = sdiv
- ; AVX: cost of 40 {{.*}} %V2i64 = sdiv
%V2i64 = sdiv <2 x i64> undef, <i64 16, i64 16>
- ; SSE: cost of 80 {{.*}} %V4i64 = sdiv
- ; AVX: cost of 80 {{.*}} %V4i64 = sdiv
%V4i64 = sdiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
- ; SSE: cost of 160 {{.*}} %V8i64 = sdiv
- ; AVX: cost of 160 {{.*}} %V8i64 = sdiv
%V8i64 = sdiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
- ; CHECK: cost of 1 {{.*}} %I32 = sdiv
%I32 = sdiv i32 undef, 16
- ; SSE2: cost of 19 {{.*}} %V4i32 = sdiv
- ; SSSE3: cost of 19 {{.*}} %V4i32 = sdiv
- ; SSE42: cost of 15 {{.*}} %V4i32 = sdiv
- ; AVX: cost of 15 {{.*}} %V4i32 = sdiv
%V4i32 = sdiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
- ; SSE2: cost of 38 {{.*}} %V8i32 = sdiv
- ; SSSE3: cost of 38 {{.*}} %V8i32 = sdiv
- ; SSE42: cost of 30 {{.*}} %V8i32 = sdiv
- ; AVX1: cost of 32 {{.*}} %V8i32 = sdiv
- ; AVX2: cost of 15 {{.*}} %V8i32 = sdiv
- ; AVX512: cost of 15 {{.*}} %V8i32 = sdiv
%V8i32 = sdiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
- ; SSE2: cost of 76 {{.*}} %V16i32 = sdiv
- ; SSSE3: cost of 76 {{.*}} %V16i32 = sdiv
- ; SSE42: cost of 60 {{.*}} %V16i32 = sdiv
- ; AVX1: cost of 64 {{.*}} %V16i32 = sdiv
- ; AVX2: cost of 30 {{.*}} %V16i32 = sdiv
- ; AVX512: cost of 15 {{.*}} %V16i32 = sdiv
%V16i32 = sdiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
- ; CHECK: cost of 1 {{.*}} %I16 = sdiv
%I16 = sdiv i16 undef, 16
- ; SSE: cost of 6 {{.*}} %V8i16 = sdiv
- ; AVX: cost of 6 {{.*}} %V8i16 = sdiv
%V8i16 = sdiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
- ; SSE: cost of 12 {{.*}} %V16i16 = sdiv
- ; AVX1: cost of 14 {{.*}} %V16i16 = sdiv
- ; AVX2: cost of 6 {{.*}} %V16i16 = sdiv
- ; AVX512: cost of 6 {{.*}} %V16i16 = sdiv
%V16i16 = sdiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
- ; SSE: cost of 24 {{.*}} %V32i16 = sdiv
- ; AVX1: cost of 28 {{.*}} %V32i16 = sdiv
- ; AVX2: cost of 12 {{.*}} %V32i16 = sdiv
- ; AVX512F: cost of 12 {{.*}} %V32i16 = sdiv
- ; AVX512BW: cost of 6 {{.*}} %V32i16 = sdiv
%V32i16 = sdiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
- ; CHECK: cost of 1 {{.*}} %I8 = sdiv
%I8 = sdiv i8 undef, 16
- ; SSE: cost of 320 {{.*}} %V16i8 = sdiv
- ; AVX: cost of 320 {{.*}} %V16i8 = sdiv
%V16i8 = sdiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
- ; SSE: cost of 640 {{.*}} %V32i8 = sdiv
- ; AVX: cost of 640 {{.*}} %V32i8 = sdiv
%V32i8 = sdiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
- ; SSE: cost of 1280 {{.*}} %V64i8 = sdiv
- ; AVX: cost of 1280 {{.*}} %V64i8 = sdiv
%V64i8 = sdiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
ret i32 undef
}
-; CHECK-LABEL: 'udiv_uniformconstpow2'
define i32 @udiv_uniformconstpow2() {
- ; CHECK: cost of 1 {{.*}} %I64 = udiv
+; SSE-LABEL: 'udiv_uniformconstpow2'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SSE-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 16, i64 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SSE-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SSE-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'udiv_uniformconstpow2'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 16, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'udiv_uniformconstpow2'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 16, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'udiv_uniformconstpow2'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 16, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'udiv_uniformconstpow2'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 16, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 640 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1280 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = udiv i64 undef, 16
- ; SSE: cost of 40 {{.*}} %V2i64 = udiv
- ; AVX: cost of 40 {{.*}} %V2i64 = udiv
%V2i64 = udiv <2 x i64> undef, <i64 16, i64 16>
- ; SSE: cost of 80 {{.*}} %V4i64 = udiv
- ; AVX: cost of 80 {{.*}} %V4i64 = udiv
%V4i64 = udiv <4 x i64> undef, <i64 16, i64 16, i64 16, i64 16>
- ; SSE: cost of 160 {{.*}} %V8i64 = udiv
- ; AVX: cost of 160 {{.*}} %V8i64 = udiv
%V8i64 = udiv <8 x i64> undef, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
- ; CHECK: cost of 1 {{.*}} %I32 = udiv
%I32 = udiv i32 undef, 16
- ; SSE: cost of 15 {{.*}} %V4i32 = udiv
- ; AVX: cost of 15 {{.*}} %V4i32 = udiv
%V4i32 = udiv <4 x i32> undef, <i32 16, i32 16, i32 16, i32 16>
- ; SSE: cost of 30 {{.*}} %V8i32 = udiv
- ; AVX1: cost of 32 {{.*}} %V8i32 = udiv
- ; AVX2: cost of 15 {{.*}} %V8i32 = udiv
- ; AVX512: cost of 15 {{.*}} %V8i32 = udiv
%V8i32 = udiv <8 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
- ; SSE: cost of 60 {{.*}} %V16i32 = udiv
- ; AVX1: cost of 64 {{.*}} %V16i32 = udiv
- ; AVX2: cost of 30 {{.*}} %V16i32 = udiv
- ; AVX512: cost of 15 {{.*}} %V16i32 = udiv
%V16i32 = udiv <16 x i32> undef, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
- ; CHECK: cost of 1 {{.*}} %I16 = udiv
%I16 = udiv i16 undef, 16
- ; SSE: cost of 6 {{.*}} %V8i16 = udiv
- ; AVX: cost of 6 {{.*}} %V8i16 = udiv
%V8i16 = udiv <8 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
- ; SSE: cost of 12 {{.*}} %V16i16 = udiv
- ; AVX1: cost of 14 {{.*}} %V16i16 = udiv
- ; AVX2: cost of 6 {{.*}} %V16i16 = udiv
- ; AVX512: cost of 6 {{.*}} %V16i16 = udiv
%V16i16 = udiv <16 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
- ; SSE: cost of 24 {{.*}} %V32i16 = udiv
- ; AVX1: cost of 28 {{.*}} %V32i16 = udiv
- ; AVX2: cost of 12 {{.*}} %V32i16 = udiv
- ; AVX512F: cost of 12 {{.*}} %V32i16 = udiv
- ; AVX512BW: cost of 6 {{.*}} %V32i16 = udiv
%V32i16 = udiv <32 x i16> undef, <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>
- ; CHECK: cost of 1 {{.*}} %I8 = udiv
%I8 = udiv i8 undef, 16
- ; SSE: cost of 320 {{.*}} %V16i8 = udiv
- ; AVX: cost of 320 {{.*}} %V16i8 = udiv
%V16i8 = udiv <16 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
- ; SSE: cost of 640 {{.*}} %V32i8 = udiv
- ; AVX: cost of 640 {{.*}} %V32i8 = udiv
%V32i8 = udiv <32 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
- ; SSE: cost of 1280 {{.*}} %V64i8 = udiv
- ; AVX: cost of 1280 {{.*}} %V64i8 = udiv
%V64i8 = udiv <64 x i8> undef, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>
ret i32 undef
Modified: llvm/trunk/test/Analysis/CostModel/X86/rem.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/rem.ll?rev=329401&r1=329400&r2=329401&view=diff
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/rem.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/rem.ll Fri Apr 6 08:23:26 2018
@@ -1,115 +1,93 @@
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE2
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+ssse3 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSSE3
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+sse4.2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE --check-prefix=SSE42
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX1
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX2
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512BW
+; 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=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSSE3
+; 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
-target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
-target triple = "x86_64-apple-macosx10.8.0"
-
-; CHECK-LABEL: 'srem'
define i32 @srem() {
- ; CHECK: cost of 1 {{.*}} %I64 = srem
+; CHECK-LABEL: 'srem'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = srem i64 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2i64 = srem <2 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i64 = srem <4 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i64 = srem <8 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = srem i32 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i32 = srem <4 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i32 = srem <8 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i32 = srem <16 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = srem i16 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i16 = srem <8 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i16 = srem <16 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32i16 = srem <32 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = srem i8 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i8 = srem <16 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32i8 = srem <32 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 192 for instruction: %V64i8 = srem <64 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = srem i64 undef, undef
- ; SSE: cost of 6 {{.*}} %V2i64 = srem
- ; AVX: cost of 6 {{.*}} %V2i64 = srem
%V2i64 = srem <2 x i64> undef, undef
- ; SSE: cost of 12 {{.*}} %V4i64 = srem
- ; AVX: cost of 12 {{.*}} %V4i64 = srem
%V4i64 = srem <4 x i64> undef, undef
- ; SSE: cost of 24 {{.*}} %V8i64 = srem
- ; AVX: cost of 24 {{.*}} %V8i64 = srem
%V8i64 = srem <8 x i64> undef, undef
- ; CHECK: cost of 1 {{.*}} %I32 = srem
%I32 = srem i32 undef, undef
- ; SSE: cost of 12 {{.*}} %V4i32 = srem
- ; AVX: cost of 12 {{.*}} %V4i32 = srem
%V4i32 = srem <4 x i32> undef, undef
- ; SSE: cost of 24 {{.*}} %V8i32 = srem
- ; AVX: cost of 24 {{.*}} %V8i32 = srem
%V8i32 = srem <8 x i32> undef, undef
- ; SSE: cost of 48 {{.*}} %V16i32 = srem
- ; AVX: cost of 48 {{.*}} %V16i32 = srem
%V16i32 = srem <16 x i32> undef, undef
- ; CHECK: cost of 1 {{.*}} %I16 = srem
%I16 = srem i16 undef, undef
- ; SSE: cost of 24 {{.*}} %V8i16 = srem
- ; AVX: cost of 24 {{.*}} %V8i16 = srem
%V8i16 = srem <8 x i16> undef, undef
- ; SSE: cost of 48 {{.*}} %V16i16 = srem
- ; AVX: cost of 48 {{.*}} %V16i16 = srem
%V16i16 = srem <16 x i16> undef, undef
- ; SSE: cost of 96 {{.*}} %V32i16 = srem
- ; AVX: cost of 96 {{.*}} %V32i16 = srem
%V32i16 = srem <32 x i16> undef, undef
- ; CHECK: cost of 1 {{.*}} %I8 = srem
%I8 = srem i8 undef, undef
- ; SSE: cost of 48 {{.*}} %V16i8 = srem
- ; AVX: cost of 48 {{.*}} %V16i8 = srem
%V16i8 = srem <16 x i8> undef, undef
- ; SSE: cost of 96 {{.*}} %V32i8 = srem
- ; AVX: cost of 96 {{.*}} %V32i8 = srem
%V32i8 = srem <32 x i8> undef, undef
- ; SSE: cost of 192 {{.*}} %V64i8 = srem
- ; AVX: cost of 192 {{.*}} %V64i8 = srem
%V64i8 = srem <64 x i8> undef, undef
ret i32 undef
}
-; CHECK-LABEL: 'urem'
define i32 @urem() {
- ; CHECK: cost of 1 {{.*}} %I64 = urem
+; CHECK-LABEL: 'urem'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = urem i64 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2i64 = urem <2 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i64 = urem <4 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i64 = urem <8 x i64> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = urem i32 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i32 = urem <4 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i32 = urem <8 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i32 = urem <16 x i32> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = urem i16 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i16 = urem <8 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i16 = urem <16 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32i16 = urem <32 x i16> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = urem i8 undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i8 = urem <16 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32i8 = urem <32 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 192 for instruction: %V64i8 = urem <64 x i8> undef, undef
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
%I64 = urem i64 undef, undef
- ; SSE: cost of 6 {{.*}} %V2i64 = urem
- ; AVX: cost of 6 {{.*}} %V2i64 = urem
%V2i64 = urem <2 x i64> undef, undef
- ; SSE: cost of 12 {{.*}} %V4i64 = urem
- ; AVX: cost of 12 {{.*}} %V4i64 = urem
%V4i64 = urem <4 x i64> undef, undef
- ; SSE: cost of 24 {{.*}} %V8i64 = urem
- ; AVX: cost of 24 {{.*}} %V8i64 = urem
%V8i64 = urem <8 x i64> undef, undef
- ; CHECK: cost of 1 {{.*}} %I32 = urem
%I32 = urem i32 undef, undef
- ; SSE: cost of 12 {{.*}} %V4i32 = urem
- ; AVX: cost of 12 {{.*}} %V4i32 = urem
%V4i32 = urem <4 x i32> undef, undef
- ; SSE: cost of 24 {{.*}} %V8i32 = urem
- ; AVX: cost of 24 {{.*}} %V8i32 = urem
%V8i32 = urem <8 x i32> undef, undef
- ; SSE: cost of 48 {{.*}} %V16i32 = urem
- ; AVX: cost of 48 {{.*}} %V16i32 = urem
%V16i32 = urem <16 x i32> undef, undef
- ; CHECK: cost of 1 {{.*}} %I16 = urem
%I16 = urem i16 undef, undef
- ; SSE: cost of 24 {{.*}} %V8i16 = urem
- ; AVX: cost of 24 {{.*}} %V8i16 = urem
%V8i16 = urem <8 x i16> undef, undef
- ; SSE: cost of 48 {{.*}} %V16i16 = urem
- ; AVX: cost of 48 {{.*}} %V16i16 = urem
%V16i16 = urem <16 x i16> undef, undef
- ; SSE: cost of 96 {{.*}} %V32i16 = urem
- ; AVX: cost of 96 {{.*}} %V32i16 = urem
%V32i16 = urem <32 x i16> undef, undef
- ; CHECK: cost of 1 {{.*}} %I8 = urem
%I8 = urem i8 undef, undef
- ; SSE: cost of 48 {{.*}} %V16i8 = urem
- ; AVX: cost of 48 {{.*}} %V16i8 = urem
%V16i8 = urem <16 x i8> undef, undef
- ; SSE: cost of 96 {{.*}} %V32i8 = urem
- ; AVX: cost of 96 {{.*}} %V32i8 = urem
%V32i8 = urem <32 x i8> undef, undef
- ; SSE: cost of 192 {{.*}} %V64i8 = urem
- ; AVX: cost of 192 {{.*}} %V64i8 = urem
%V64i8 = urem <64 x i8> undef, undef
ret i32 undef
Modified: llvm/trunk/test/Analysis/CostModel/X86/vdiv-cost.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/vdiv-cost.ll?rev=329401&r1=329400&r2=329401&view=diff
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/vdiv-cost.ll (original)
+++ llvm/trunk/test/Analysis/CostModel/X86/vdiv-cost.ll Fri Apr 6 08:23:26 2018
@@ -1,122 +1,180 @@
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE2
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+ssse3 | FileCheck %s --check-prefix=CHECK --check-prefix=SSSE3
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+sse4.2 | FileCheck %s --check-prefix=CHECK --check-prefix=SSE42
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx | FileCheck %s --check-prefix=CHECK --check-prefix=AVX
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=AVX2
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512BW
-; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512DQ
+; 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=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSSE3
+; 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 -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512DQ
define <4 x i32> @test1(<4 x i32> %a) {
+; CHECK-LABEL: 'test1'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = udiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %div
+;
%div = udiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
ret <4 x i32> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test1':
-; SSE: Found an estimated cost of 15 for instruction: %div
-; AVX: Found an estimated cost of 15 for instruction: %div
-; AVX512: Found an estimated cost of 15 for instruction: %div
}
define <8 x i32> @test2(<8 x i32> %a) {
+; SSE-LABEL: 'test2'
+; SSE-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %div = udiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; AVX1-LABEL: 'test2'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %div = udiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; AVX2-LABEL: 'test2'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = udiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; AVX512-LABEL: 'test2'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = udiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
%div = udiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7,i32 7, i32 7, i32 7, i32 7>
ret <8 x i32> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test2':
-; SSE: Found an estimated cost of 30 for instruction: %div
-; AVX1: Found an estimated cost of 30 for instruction: %div
-; AVX2: Found an estimated cost of 15 for instruction: %div
-; AVX512: Found an estimated cost of 15 for instruction: %div
}
define <8 x i16> @test3(<8 x i16> %a) {
+; CHECK-LABEL: 'test3'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %div = udiv <8 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %div
+;
%div = udiv <8 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
ret <8 x i16> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test3':
-; SSE: Found an estimated cost of 6 for instruction: %div
-; AVX: Found an estimated cost of 6 for instruction: %div
-; AVX512: Found an estimated cost of 6 for instruction: %div
}
define <16 x i16> @test4(<16 x i16> %a) {
+; SSE-LABEL: 'test4'
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %div = udiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
+; AVX1-LABEL: 'test4'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %div = udiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
+; AVX2-LABEL: 'test4'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %div = udiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
+; AVX512-LABEL: 'test4'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %div = udiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
%div = udiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7>
ret <16 x i16> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test4':
-; SSE: Found an estimated cost of 12 for instruction: %div
-; AVX1: Found an estimated cost of 12 for instruction: %div
-; AVX2: Found an estimated cost of 6 for instruction: %div
-; AVX512: Found an estimated cost of 6 for instruction: %div
}
define <8 x i16> @test5(<8 x i16> %a) {
+; CHECK-LABEL: 'test5'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %div = sdiv <8 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %div
+;
%div = sdiv <8 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
ret <8 x i16> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test5':
-; SSE: Found an estimated cost of 6 for instruction: %div
-; AVX: Found an estimated cost of 6 for instruction: %div
-; AVX512: Found an estimated cost of 6 for instruction: %div
}
define <16 x i16> @test6(<16 x i16> %a) {
+; SSE-LABEL: 'test6'
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %div = sdiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
+; AVX1-LABEL: 'test6'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %div = sdiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
+; AVX2-LABEL: 'test6'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %div = sdiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
+; AVX512-LABEL: 'test6'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %div = sdiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %div
+;
%div = sdiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7>
ret <16 x i16> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test6':
-; SSE: Found an estimated cost of 12 for instruction: %div
-; AVX1: Found an estimated cost of 12 for instruction: %div
-; AVX2: Found an estimated cost of 6 for instruction: %div
-; AVX512: Found an estimated cost of 6 for instruction: %div
}
define <16 x i8> @test7(<16 x i8> %a) {
+; CHECK-LABEL: 'test7'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %div = sdiv <16 x i8> %a, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %div
+;
%div = sdiv <16 x i8> %a, <i8 7, i8 7, i8 7, i8 7,i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7,i8 7, i8 7, i8 7, i8 7>
ret <16 x i8> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test7':
-; SSE: Found an estimated cost of 320 for instruction: %div
-; AVX: Found an estimated cost of 320 for instruction: %div
-; AVX512: Found an estimated cost of 320 for instruction: %div
}
define <4 x i32> @test8(<4 x i32> %a) {
+; SSE2-LABEL: 'test8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %div = sdiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %div
+;
+; SSSE3-LABEL: 'test8'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %div = sdiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %div
+;
+; SSE42-LABEL: 'test8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = sdiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %div
+;
+; AVX-LABEL: 'test8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = sdiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %div
+;
+; AVX512-LABEL: 'test8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = sdiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %div
+;
%div = sdiv <4 x i32> %a, <i32 7, i32 7, i32 7, i32 7>
ret <4 x i32> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test8':
-; SSE: Found an estimated cost of 19 for instruction: %div
-; AVX: Found an estimated cost of 15 for instruction: %div
-; AVX512: Found an estimated cost of 15 for instruction: %div
}
define <8 x i32> @test9(<8 x i32> %a) {
+; SSE2-LABEL: 'test9'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %div = sdiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; SSSE3-LABEL: 'test9'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %div = sdiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; SSE42-LABEL: 'test9'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %div = sdiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; AVX1-LABEL: 'test9'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %div = sdiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; AVX2-LABEL: 'test9'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = sdiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
+; AVX512-LABEL: 'test9'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %div = sdiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
%div = sdiv <8 x i32> %a, <i32 7, i32 7, i32 7, i32 7,i32 7, i32 7, i32 7, i32 7>
ret <8 x i32> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test9':
-; SSE: Found an estimated cost of 38 for instruction: %div
-; AVX1: Found an estimated cost of 38 for instruction: %div
-; AVX2: Found an estimated cost of 15 for instruction: %div
-; AVX512: Found an estimated cost of 15 for instruction: %div
}
define <8 x i32> @test10(<8 x i32> %a) {
+; CHECK-LABEL: 'test10'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %div = sdiv <8 x i32> %a, <i32 8, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %div
+;
%div = sdiv <8 x i32> %a, <i32 8, i32 7, i32 7, i32 7,i32 7, i32 7, i32 7, i32 7>
ret <8 x i32> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test10':
-; SSE: Found an estimated cost of 160 for instruction: %div
-; AVX: Found an estimated cost of 160 for instruction: %div
-; AVX512: Found an estimated cost of 160 for instruction: %div
}
define <16 x i32> @test11(<16 x i32> %a) {
+; CHECK-LABEL: 'test11'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %div = sdiv <16 x i32> %a, <i32 8, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %div
+;
%div = sdiv <16 x i32> %a, <i32 8, i32 7, i32 7, i32 7,i32 7, i32 7, i32 7, i32 7, i32 8, i32 7, i32 7, i32 7,i32 7, i32 7, i32 7, i32 7>
ret <16 x i32> %div
-
-; CHECK: 'Cost Model Analysis' for function 'test11':
-; SSE: Found an estimated cost of 320 for instruction: %div
-; AVX: Found an estimated cost of 320 for instruction: %div
-; AVX512: Found an estimated cost of 320 for instruction: %div
}
More information about the llvm-commits
mailing list