derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
245 lines
21 KiB
LLVM
245 lines
21 KiB
LLVM
; 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=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)
|
|
;
|
|
|
|
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 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 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 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 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 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 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 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 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 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)
|
|
%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 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 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 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 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 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 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 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 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 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)
|
|
%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 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 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 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 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 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 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 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 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 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)
|
|
%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>)
|