[llvm] [CostModel][X86] merge bswap costs tests using -cost-kind=all (PR #131784)
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 18 04:18:41 PDT 2025
https://github.com/RKSimon created https://github.com/llvm/llvm-project/pull/131784
None
>From a111f15b296ee9eb5ef584398b1abd47e3dcf690 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Tue, 18 Mar 2025 11:17:56 +0000
Subject: [PATCH] [CostModel][X86] merge bswap costs tests using -cost-kind=all
---
.../Analysis/CostModel/X86/bswap-codesize.ll | 244 ---------------
.../Analysis/CostModel/X86/bswap-latency.ll | 244 ---------------
.../CostModel/X86/bswap-sizelatency.ll | 244 ---------------
llvm/test/Analysis/CostModel/X86/bswap.ll | 290 +++++++++---------
4 files changed, 145 insertions(+), 877 deletions(-)
delete mode 100644 llvm/test/Analysis/CostModel/X86/bswap-codesize.ll
delete mode 100644 llvm/test/Analysis/CostModel/X86/bswap-latency.ll
delete mode 100644 llvm/test/Analysis/CostModel/X86/bswap-sizelatency.ll
diff --git a/llvm/test/Analysis/CostModel/X86/bswap-codesize.ll b/llvm/test/Analysis/CostModel/X86/bswap-codesize.ll
deleted file mode 100644
index 1d7f19d1664e1..0000000000000
--- a/llvm/test/Analysis/CostModel/X86/bswap-codesize.ll
+++ /dev/null
@@ -1,244 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx | FileCheck %s --check-prefixes=AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=slm | FileCheck %s --check-prefixes=SLM
-
-;
-; bswap(X)
-;
-
-define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
-; SSE2-LABEL: 'cost_bswap_i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i64'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i64'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i64'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i64'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i64'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I64 = call i64 @llvm.bswap.i64(i64 %a64)
- %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
- %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
- %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
- ret void
-}
-
-define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
-; SSE2-LABEL: 'cost_bswap_i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i32'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i32'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i32'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i32'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I32 = call i32 @llvm.bswap.i32(i32 %a32)
- %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
- %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
- %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
- ret void
-}
-
-define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
-; SSE2-LABEL: 'cost_bswap_i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i16'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i16'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i16'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i16'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i16'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I16 = call i16 @llvm.bswap.i16(i16 %a16)
- %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
- %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
- %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
- ret void
-}
-
-declare i64 @llvm.bswap.i64(i64)
-declare i32 @llvm.bswap.i32(i32)
-declare i16 @llvm.bswap.i16(i16)
-
-declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
-declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
-declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
-
-declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>)
-declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>)
-declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
-
-declare <8 x i64> @llvm.bswap.v8i64(<8 x i64>)
-declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
-declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)
diff --git a/llvm/test/Analysis/CostModel/X86/bswap-latency.ll b/llvm/test/Analysis/CostModel/X86/bswap-latency.ll
deleted file mode 100644
index 592541f4816f2..0000000000000
--- a/llvm/test/Analysis/CostModel/X86/bswap-latency.ll
+++ /dev/null
@@ -1,244 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=SLM
-
-;
-; bswap(X)
-;
-
-define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
-; SSE2-LABEL: 'cost_bswap_i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i64'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i64'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i64'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i64'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i64'
-; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I64 = call i64 @llvm.bswap.i64(i64 %a64)
- %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
- %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
- %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
- ret void
-}
-
-define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
-; SSE2-LABEL: 'cost_bswap_i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i32'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i32'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i32'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i32'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I32 = call i32 @llvm.bswap.i32(i32 %a32)
- %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
- %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
- %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
- ret void
-}
-
-define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
-; SSE2-LABEL: 'cost_bswap_i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i16'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i16'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i16'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i16'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i16'
-; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I16 = call i16 @llvm.bswap.i16(i16 %a16)
- %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
- %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
- %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
- ret void
-}
-
-declare i64 @llvm.bswap.i64(i64)
-declare i32 @llvm.bswap.i32(i32)
-declare i16 @llvm.bswap.i16(i16)
-
-declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
-declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
-declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
-
-declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>)
-declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>)
-declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
-
-declare <8 x i64> @llvm.bswap.v8i64(<8 x i64>)
-declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
-declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)
diff --git a/llvm/test/Analysis/CostModel/X86/bswap-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/bswap-sizelatency.ll
deleted file mode 100644
index c6cfafb204620..0000000000000
--- a/llvm/test/Analysis/CostModel/X86/bswap-sizelatency.ll
+++ /dev/null
@@ -1,244 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx | FileCheck %s --check-prefixes=AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mcpu=slm | FileCheck %s --check-prefixes=SLM
-
-;
-; bswap(X)
-;
-
-define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
-; SSE2-LABEL: 'cost_bswap_i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i64'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i64'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i64'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i64'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i64'
-; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I64 = call i64 @llvm.bswap.i64(i64 %a64)
- %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
- %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
- %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
- ret void
-}
-
-define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
-; SSE2-LABEL: 'cost_bswap_i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i32'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i32'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i32'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i32'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I32 = call i32 @llvm.bswap.i32(i32 %a32)
- %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
- %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
- %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
- ret void
-}
-
-define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
-; SSE2-LABEL: 'cost_bswap_i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_bswap_i16'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_bswap_i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX1-LABEL: 'cost_bswap_i16'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX2-LABEL: 'cost_bswap_i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512F-LABEL: 'cost_bswap_i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512DQ-LABEL: 'cost_bswap_i16'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; AVX512BW-LABEL: 'cost_bswap_i16'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SLM-LABEL: 'cost_bswap_i16'
-; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
- %I16 = call i16 @llvm.bswap.i16(i16 %a16)
- %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
- %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
- %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
- ret void
-}
-
-declare i64 @llvm.bswap.i64(i64)
-declare i32 @llvm.bswap.i32(i32)
-declare i16 @llvm.bswap.i16(i16)
-
-declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
-declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
-declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
-
-declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>)
-declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>)
-declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
-
-declare <8 x i64> @llvm.bswap.v8i64(<8 x i64>)
-declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
-declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)
diff --git a/llvm/test/Analysis/CostModel/X86/bswap.ll b/llvm/test/Analysis/CostModel/X86/bswap.ll
index d1c7daa04cfa6..4466926c3483a 100644
--- a/llvm/test/Analysis/CostModel/X86/bswap.ll
+++ b/llvm/test/Analysis/CostModel/X86/bswap.ll
@@ -1,14 +1,14 @@
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm | FileCheck %s --check-prefixes=SLM
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
+;
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mcpu=slm | FileCheck %s --check-prefixes=SLM
;
; bswap(X)
@@ -16,67 +16,67 @@
define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
; SSE2-LABEL: 'cost_bswap_i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE2-NEXT: Cost Model: Found costs of RThru:5 CodeSize:11 Lat:6 SizeLat:11 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE2-NEXT: Cost Model: Found costs of RThru:10 CodeSize:22 Lat:12 SizeLat:22 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE2-NEXT: Cost Model: Found costs of RThru:20 CodeSize:44 Lat:24 SizeLat:44 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SSSE3-LABEL: 'cost_bswap_i64'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSSE3-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:5 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:6 SizeLat:10 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:12 SizeLat:20 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SSE42-LABEL: 'cost_bswap_i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE42-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:5 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE42-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:6 SizeLat:10 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE42-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:12 SizeLat:20 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX1-LABEL: 'cost_bswap_i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:3 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX1-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:6 SizeLat:10 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX1-NEXT: Cost Model: Found costs of RThru:10 CodeSize:10 Lat:12 SizeLat:20 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX2-LABEL: 'cost_bswap_i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:4 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512F-LABEL: 'cost_bswap_i64'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:5 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512DQ-LABEL: 'cost_bswap_i64'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:5 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512BW-LABEL: 'cost_bswap_i64'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SLM-LABEL: 'cost_bswap_i64'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SLM-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SLM-NEXT: Cost Model: Found costs of RThru:5 CodeSize:1 Lat:5 SizeLat:5 for: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SLM-NEXT: Cost Model: Found costs of RThru:10 CodeSize:2 Lat:10 SizeLat:10 for: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SLM-NEXT: Cost Model: Found costs of RThru:20 CodeSize:4 Lat:20 SizeLat:20 for: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SLM-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
%I64 = call i64 @llvm.bswap.i64(i64 %a64)
%V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
@@ -87,67 +87,67 @@ define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
; SSE2-LABEL: 'cost_bswap_i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE2-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE2-NEXT: Cost Model: Found costs of RThru:5 CodeSize:9 Lat:5 SizeLat:9 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE2-NEXT: Cost Model: Found costs of RThru:10 CodeSize:18 Lat:10 SizeLat:18 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE2-NEXT: Cost Model: Found costs of RThru:20 CodeSize:36 Lat:20 SizeLat:36 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SSSE3-LABEL: 'cost_bswap_i32'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSSE3-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:5 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:6 SizeLat:10 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:12 SizeLat:20 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SSE42-LABEL: 'cost_bswap_i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE42-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE42-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:5 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE42-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:6 SizeLat:10 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE42-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:12 SizeLat:20 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX1-LABEL: 'cost_bswap_i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX1-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:3 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX1-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:6 SizeLat:10 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX1-NEXT: Cost Model: Found costs of RThru:10 CodeSize:10 Lat:12 SizeLat:20 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX2-LABEL: 'cost_bswap_i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX2-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:4 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512F-LABEL: 'cost_bswap_i32'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512F-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:5 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512DQ-LABEL: 'cost_bswap_i32'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512DQ-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:5 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512BW-LABEL: 'cost_bswap_i32'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SLM-LABEL: 'cost_bswap_i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SLM-NEXT: Cost Model: Found costs of 1 for: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SLM-NEXT: Cost Model: Found costs of RThru:5 CodeSize:1 Lat:5 SizeLat:5 for: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SLM-NEXT: Cost Model: Found costs of RThru:10 CodeSize:2 Lat:10 SizeLat:10 for: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SLM-NEXT: Cost Model: Found costs of RThru:20 CodeSize:4 Lat:20 SizeLat:20 for: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SLM-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
%I32 = call i32 @llvm.bswap.i32(i32 %a32)
%V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
@@ -158,67 +158,67 @@ define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
; SSE2-LABEL: 'cost_bswap_i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE2-NEXT: Cost Model: Found costs of RThru:5 CodeSize:4 Lat:5 SizeLat:5 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE2-NEXT: Cost Model: Found costs of RThru:10 CodeSize:8 Lat:10 SizeLat:10 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE2-NEXT: Cost Model: Found costs of RThru:20 CodeSize:16 Lat:20 SizeLat:20 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SSSE3-LABEL: 'cost_bswap_i16'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSSE3-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:5 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:6 SizeLat:10 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:12 SizeLat:20 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SSE42-LABEL: 'cost_bswap_i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE42-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:5 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE42-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:6 SizeLat:10 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE42-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:12 SizeLat:20 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX1-LABEL: 'cost_bswap_i16'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:3 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX1-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:6 SizeLat:10 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX1-NEXT: Cost Model: Found costs of RThru:10 CodeSize:10 Lat:12 SizeLat:20 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX2-LABEL: 'cost_bswap_i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:4 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512F-LABEL: 'cost_bswap_i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:5 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512DQ-LABEL: 'cost_bswap_i16'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:5 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512DQ-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; AVX512BW-LABEL: 'cost_bswap_i16'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:1 SizeLat:2 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
; SLM-LABEL: 'cost_bswap_i16'
-; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
-; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SLM-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SLM-NEXT: Cost Model: Found costs of RThru:5 CodeSize:1 Lat:5 SizeLat:5 for: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SLM-NEXT: Cost Model: Found costs of RThru:10 CodeSize:2 Lat:10 SizeLat:10 for: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SLM-NEXT: Cost Model: Found costs of RThru:20 CodeSize:4 Lat:20 SizeLat:20 for: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SLM-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
;
%I16 = call i16 @llvm.bswap.i16(i16 %a16)
%V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
More information about the llvm-commits
mailing list