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
2009 lines
146 KiB
LLVM
2009 lines
146 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=SSE,SSE2
|
|
; 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=SSE,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=AVX,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=AVX,AVX2
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+xop,+avx | FileCheck %s --check-prefixes=XOP,XOPAVX1
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=XOP,XOPAVX2
|
|
; 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=AVX512,AVX512F
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512F
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512F
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VL
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
|
|
;
|
|
; 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=SSE,SSE42
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,SSE42
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX1
|
|
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mcpu=tigerlake | FileCheck %s --check-prefixes=AVX512,AVX512GFNI
|
|
|
|
; Verify the cost of vector arithmetic shift right instructions.
|
|
|
|
;
|
|
; Variable Shifts
|
|
;
|
|
|
|
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
|
|
; SSE2-LABEL: 'var_shift_v2i64'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:12 Lat:11 SizeLat:16 for: %shift = ashr <2 x i64> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v2i64'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:8 CodeSize:5 Lat:17 SizeLat:7 for: %shift = ashr <2 x i64> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v2i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:6 SizeLat:14 for: %shift = ashr <2 x i64> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v2i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:5 SizeLat:5 for: %shift = ashr <2 x i64> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v2i64'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <2 x i64> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'var_shift_v2i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <2 x i64> %a, %b
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
%shift = ashr <2 x i64> %a, %b
|
|
ret <2 x i64> %shift
|
|
}
|
|
|
|
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
|
|
; SSE2-LABEL: 'var_shift_v4i64'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:24 Lat:22 SizeLat:32 for: %shift = ashr <4 x i64> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v4i64'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:16 CodeSize:10 Lat:34 SizeLat:14 for: %shift = ashr <4 x i64> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v4i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:12 CodeSize:22 Lat:12 SizeLat:30 for: %shift = ashr <4 x i64> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v4i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:5 Lat:8 SizeLat:9 for: %shift = ashr <4 x i64> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v4i64'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:6 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <4 x i64> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'var_shift_v4i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i64> %a, %b
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
%shift = ashr <4 x i64> %a, %b
|
|
ret <4 x i64> %shift
|
|
}
|
|
|
|
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
|
|
; SSE2-LABEL: 'var_shift_v8i64'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:32 CodeSize:48 Lat:44 SizeLat:64 for: %shift = ashr <8 x i64> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v8i64'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:32 CodeSize:20 Lat:68 SizeLat:28 for: %shift = ashr <8 x i64> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v8i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:24 CodeSize:44 Lat:24 SizeLat:60 for: %shift = ashr <8 x i64> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v8i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:10 Lat:16 SizeLat:18 for: %shift = ashr <8 x i64> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v8i64'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:12 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <8 x i64> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'var_shift_v8i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i64> %a, %b
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
%shift = ashr <8 x i64> %a, %b
|
|
ret <8 x i64> %shift
|
|
}
|
|
|
|
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
|
|
; SSE2-LABEL: 'var_shift_v4i32'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:12 CodeSize:15 Lat:12 SizeLat:19 for: %shift = ashr <4 x i32> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v4i32'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:16 CodeSize:15 Lat:17 SizeLat:19 for: %shift = ashr <4 x i32> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v4i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:12 Lat:7 SizeLat:16 for: %shift = ashr <4 x i32> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v4i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:3 for: %shift = ashr <4 x i32> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v4i32'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <4 x i32> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'var_shift_v4i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i32> %a, %b
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
%shift = ashr <4 x i32> %a, %b
|
|
ret <4 x i32> %shift
|
|
}
|
|
|
|
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
|
|
; SSE2-LABEL: 'var_shift_v8i32'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:24 CodeSize:30 Lat:24 SizeLat:38 for: %shift = ashr <8 x i32> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v8i32'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:32 CodeSize:30 Lat:34 SizeLat:38 for: %shift = ashr <8 x i32> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v8i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:14 CodeSize:26 Lat:14 SizeLat:34 for: %shift = ashr <8 x i32> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v8i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:3 for: %shift = ashr <8 x i32> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'var_shift_v8i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <8 x i32> %a, %b
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'var_shift_v8i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:3 for: %shift = ashr <8 x i32> %a, %b
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'var_shift_v8i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i32> %a, %b
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
%shift = ashr <8 x i32> %a, %b
|
|
ret <8 x i32> %shift
|
|
}
|
|
|
|
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
|
|
; SSE2-LABEL: 'var_shift_v16i32'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:48 CodeSize:60 Lat:48 SizeLat:76 for: %shift = ashr <16 x i32> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v16i32'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:64 CodeSize:60 Lat:68 SizeLat:76 for: %shift = ashr <16 x i32> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v16i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:28 CodeSize:52 Lat:28 SizeLat:68 for: %shift = ashr <16 x i32> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v16i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:2 Lat:8 SizeLat:6 for: %shift = ashr <16 x i32> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'var_shift_v16i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:12 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <16 x i32> %a, %b
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'var_shift_v16i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:2 Lat:8 SizeLat:6 for: %shift = ashr <16 x i32> %a, %b
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'var_shift_v16i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i32> %a, %b
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
%shift = ashr <16 x i32> %a, %b
|
|
ret <16 x i32> %shift
|
|
}
|
|
|
|
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
|
|
; SSE2-LABEL: 'var_shift_v8i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:31 Lat:19 SizeLat:31 for: %shift = ashr <8 x i16> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v8i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:22 CodeSize:23 Lat:26 SizeLat:27 for: %shift = ashr <8 x i16> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v8i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:13 CodeSize:14 Lat:16 SizeLat:22 for: %shift = ashr <8 x i16> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v8i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:11 SizeLat:10 for: %shift = ashr <8 x i16> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v8i16'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <8 x i16> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'var_shift_v8i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:11 SizeLat:10 for: %shift = ashr <8 x i16> %a, %b
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'var_shift_v8i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, %b
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'var_shift_v8i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:11 SizeLat:10 for: %shift = ashr <8 x i16> %a, %b
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'var_shift_v8i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, %b
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'var_shift_v8i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, %b
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
%shift = ashr <8 x i16> %a, %b
|
|
ret <8 x i16> %shift
|
|
}
|
|
|
|
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
|
|
; SSE2-LABEL: 'var_shift_v16i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:32 CodeSize:62 Lat:38 SizeLat:62 for: %shift = ashr <16 x i16> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v16i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:44 CodeSize:46 Lat:52 SizeLat:54 for: %shift = ashr <16 x i16> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v16i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:28 CodeSize:31 Lat:30 SizeLat:48 for: %shift = ashr <16 x i16> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v16i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:14 for: %shift = ashr <16 x i16> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v16i16'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:6 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <16 x i16> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'var_shift_v16i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:14 for: %shift = ashr <16 x i16> %a, %b
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'var_shift_v16i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, %b
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'var_shift_v16i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:14 for: %shift = ashr <16 x i16> %a, %b
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'var_shift_v16i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, %b
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'var_shift_v16i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, %b
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
%shift = ashr <16 x i16> %a, %b
|
|
ret <16 x i16> %shift
|
|
}
|
|
|
|
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
|
|
; SSE2-LABEL: 'var_shift_v32i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:64 CodeSize:124 Lat:76 SizeLat:124 for: %shift = ashr <32 x i16> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v32i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:88 CodeSize:92 Lat:104 SizeLat:108 for: %shift = ashr <32 x i16> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v32i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:56 CodeSize:62 Lat:60 SizeLat:96 for: %shift = ashr <32 x i16> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v32i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:20 Lat:20 SizeLat:28 for: %shift = ashr <32 x i16> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v32i16'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:12 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <32 x i16> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'var_shift_v32i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:11 CodeSize:11 Lat:16 SizeLat:15 for: %shift = ashr <32 x i16> %a, %b
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'var_shift_v32i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, %b
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'var_shift_v32i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:11 CodeSize:11 Lat:16 SizeLat:15 for: %shift = ashr <32 x i16> %a, %b
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'var_shift_v32i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, %b
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'var_shift_v32i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, %b
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
%shift = ashr <32 x i16> %a, %b
|
|
ret <32 x i16> %shift
|
|
}
|
|
|
|
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
|
|
; SSE2-LABEL: 'var_shift_v16i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:27 CodeSize:54 Lat:30 SizeLat:54 for: %shift = ashr <16 x i8> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v16i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:38 CodeSize:30 Lat:41 SizeLat:36 for: %shift = ashr <16 x i8> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v16i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:21 CodeSize:24 Lat:22 SizeLat:36 for: %shift = ashr <16 x i8> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v16i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:17 CodeSize:24 Lat:17 SizeLat:30 for: %shift = ashr <16 x i8> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v16i8'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <16 x i8> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'var_shift_v16i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:17 CodeSize:24 Lat:17 SizeLat:30 for: %shift = ashr <16 x i8> %a, %b
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'var_shift_v16i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, %b
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'var_shift_v16i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:17 CodeSize:24 Lat:17 SizeLat:30 for: %shift = ashr <16 x i8> %a, %b
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'var_shift_v16i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, %b
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'var_shift_v16i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, %b
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
%shift = ashr <16 x i8> %a, %b
|
|
ret <16 x i8> %shift
|
|
}
|
|
|
|
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
|
|
; SSE2-LABEL: 'var_shift_v32i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:54 CodeSize:108 Lat:60 SizeLat:108 for: %shift = ashr <32 x i8> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v32i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:76 CodeSize:60 Lat:82 SizeLat:72 for: %shift = ashr <32 x i8> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v32i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:44 CodeSize:51 Lat:45 SizeLat:76 for: %shift = ashr <32 x i8> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v32i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:18 CodeSize:24 Lat:20 SizeLat:43 for: %shift = ashr <32 x i8> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v32i8'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:6 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <32 x i8> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'var_shift_v32i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:18 CodeSize:24 Lat:20 SizeLat:43 for: %shift = ashr <32 x i8> %a, %b
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'var_shift_v32i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:6 CodeSize:24 Lat:13 SizeLat:30 for: %shift = ashr <32 x i8> %a, %b
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'var_shift_v32i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:18 CodeSize:24 Lat:20 SizeLat:43 for: %shift = ashr <32 x i8> %a, %b
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'var_shift_v32i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:6 CodeSize:24 Lat:13 SizeLat:30 for: %shift = ashr <32 x i8> %a, %b
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'var_shift_v32i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:6 CodeSize:24 Lat:13 SizeLat:30 for: %shift = ashr <32 x i8> %a, %b
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
%shift = ashr <32 x i8> %a, %b
|
|
ret <32 x i8> %shift
|
|
}
|
|
|
|
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
|
|
; SSE2-LABEL: 'var_shift_v64i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:108 CodeSize:216 Lat:120 SizeLat:216 for: %shift = ashr <64 x i8> %a, %b
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'var_shift_v64i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:152 CodeSize:120 Lat:164 SizeLat:144 for: %shift = ashr <64 x i8> %a, %b
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'var_shift_v64i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:88 CodeSize:102 Lat:90 SizeLat:152 for: %shift = ashr <64 x i8> %a, %b
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'var_shift_v64i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:36 CodeSize:48 Lat:40 SizeLat:86 for: %shift = ashr <64 x i8> %a, %b
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; XOP-LABEL: 'var_shift_v64i8'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:12 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <64 x i8> %a, %b
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'var_shift_v64i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:37 CodeSize:51 Lat:37 SizeLat:63 for: %shift = ashr <64 x i8> %a, %b
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'var_shift_v64i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:15 CodeSize:30 Lat:15 SizeLat:30 for: %shift = ashr <64 x i8> %a, %b
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'var_shift_v64i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:37 CodeSize:51 Lat:37 SizeLat:63 for: %shift = ashr <64 x i8> %a, %b
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'var_shift_v64i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:15 CodeSize:30 Lat:15 SizeLat:30 for: %shift = ashr <64 x i8> %a, %b
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'var_shift_v64i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:15 CodeSize:30 Lat:15 SizeLat:30 for: %shift = ashr <64 x i8> %a, %b
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
%shift = ashr <64 x i8> %a, %b
|
|
ret <64 x i8> %shift
|
|
}
|
|
|
|
;
|
|
; Uniform Variable Shifts
|
|
;
|
|
|
|
define <2 x i64> @splatvar_shift_v2i64(<2 x i64> %a, i64 %b) {
|
|
; SSE-LABEL: 'splatvar_shift_v2i64'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <2 x i64> undef, i64 %b, i32 0
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:9 SizeLat:7 for: %shift = ashr <2 x i64> %a, %splat
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v2i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <2 x i64> undef, i64 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:4 SizeLat:7 for: %shift = ashr <2 x i64> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v2i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <2 x i64> undef, i64 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:5 Lat:4 SizeLat:7 for: %shift = ashr <2 x i64> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v2i64'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <2 x i64> undef, i64 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:4 SizeLat:7 for: %shift = ashr <2 x i64> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v2i64'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <2 x i64> undef, i64 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:5 Lat:4 SizeLat:7 for: %shift = ashr <2 x i64> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v2i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <2 x i64> undef, i64 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <2 x i64> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
%insert = insertelement <2 x i64> undef, i64 %b, i32 0
|
|
%splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
|
|
%shift = ashr <2 x i64> %a, %splat
|
|
ret <2 x i64> %shift
|
|
}
|
|
|
|
define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, i64 %b) {
|
|
; SSE-LABEL: 'splatvar_shift_v4i64'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i64> undef, i64 %b, i32 0
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:10 CodeSize:10 Lat:18 SizeLat:14 for: %shift = ashr <4 x i64> %a, %splat
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v4i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i64> undef, i64 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:7 SizeLat:13 for: %shift = ashr <4 x i64> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v4i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i64> undef, i64 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:6 SizeLat:9 for: %shift = ashr <4 x i64> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v4i64'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i64> undef, i64 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:7 SizeLat:13 for: %shift = ashr <4 x i64> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v4i64'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i64> undef, i64 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:6 SizeLat:9 for: %shift = ashr <4 x i64> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v4i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i64> undef, i64 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:2 for: %shift = ashr <4 x i64> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
%insert = insertelement <4 x i64> undef, i64 %b, i32 0
|
|
%splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
|
|
%shift = ashr <4 x i64> %a, %splat
|
|
ret <4 x i64> %shift
|
|
}
|
|
|
|
define <8 x i64> @splatvar_shift_v8i64(<8 x i64> %a, i64 %b) {
|
|
; SSE-LABEL: 'splatvar_shift_v8i64'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i64> undef, i64 %b, i32 0
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:20 CodeSize:20 Lat:36 SizeLat:28 for: %shift = ashr <8 x i64> %a, %splat
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v8i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i64> undef, i64 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:12 CodeSize:20 Lat:14 SizeLat:26 for: %shift = ashr <8 x i64> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v8i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i64> undef, i64 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:12 SizeLat:18 for: %shift = ashr <8 x i64> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v8i64'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i64> undef, i64 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:12 CodeSize:20 Lat:14 SizeLat:26 for: %shift = ashr <8 x i64> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v8i64'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i64> undef, i64 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:12 SizeLat:18 for: %shift = ashr <8 x i64> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v8i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i64> undef, i64 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:2 for: %shift = ashr <8 x i64> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
%insert = insertelement <8 x i64> undef, i64 %b, i32 0
|
|
%splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
|
|
%shift = ashr <8 x i64> %a, %splat
|
|
ret <8 x i64> %shift
|
|
}
|
|
|
|
define <4 x i32> @splatvar_shift_v4i32(<4 x i32> %a, i32 %b) {
|
|
; SSE-LABEL: 'splatvar_shift_v4i32'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i32> undef, i32 %b, i32 0
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <4 x i32> %a, %splat
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v4i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i32> undef, i32 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %shift = ashr <4 x i32> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v4i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i32> undef, i32 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <4 x i32> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v4i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i32> undef, i32 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %shift = ashr <4 x i32> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v4i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i32> undef, i32 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <4 x i32> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v4i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <4 x i32> undef, i32 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <4 x i32> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
%insert = insertelement <4 x i32> undef, i32 %b, i32 0
|
|
%splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
|
|
%shift = ashr <4 x i32> %a, %splat
|
|
ret <4 x i32> %shift
|
|
}
|
|
|
|
define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, i32 %b) {
|
|
; SSE-LABEL: 'splatvar_shift_v8i32'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i32> undef, i32 %b, i32 0
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %shift = ashr <8 x i32> %a, %splat
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v8i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i32> undef, i32 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:7 SizeLat:7 for: %shift = ashr <8 x i32> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v8i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i32> undef, i32 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <8 x i32> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v8i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i32> undef, i32 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:7 SizeLat:7 for: %shift = ashr <8 x i32> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v8i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i32> undef, i32 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <8 x i32> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v8i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i32> undef, i32 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <8 x i32> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
%insert = insertelement <8 x i32> undef, i32 %b, i32 0
|
|
%splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
|
|
%shift = ashr <8 x i32> %a, %splat
|
|
ret <8 x i32> %shift
|
|
}
|
|
|
|
define <16 x i32> @splatvar_shift_v16i32(<16 x i32> %a, i32 %b) {
|
|
; SSE-LABEL: 'splatvar_shift_v16i32'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i32> undef, i32 %b, i32 0
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %shift = ashr <16 x i32> %a, %splat
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v16i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i32> undef, i32 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:14 SizeLat:14 for: %shift = ashr <16 x i32> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v16i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i32> undef, i32 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:6 for: %shift = ashr <16 x i32> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v16i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i32> undef, i32 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:3 SizeLat:3 for: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:14 SizeLat:14 for: %shift = ashr <16 x i32> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v16i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i32> undef, i32 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:6 for: %shift = ashr <16 x i32> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v16i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i32> undef, i32 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <16 x i32> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
%insert = insertelement <16 x i32> undef, i32 %b, i32 0
|
|
%splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
|
|
%shift = ashr <16 x i32> %a, %splat
|
|
ret <16 x i32> %shift
|
|
}
|
|
|
|
define <8 x i16> @splatvar_shift_v8i16(<8 x i16> %a, i16 %b) {
|
|
; SSE2-LABEL: 'splatvar_shift_v8i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:2 SizeLat:2 for: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <8 x i16> %a, %splat
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'splatvar_shift_v8i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <8 x i16> %a, %splat
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v8i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %shift = ashr <8 x i16> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v8i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <8 x i16> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v8i16'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %shift = ashr <8 x i16> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v8i16'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <8 x i16> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v8i16'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <8 x i16> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
%insert = insertelement <8 x i16> undef, i16 %b, i32 0
|
|
%splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
|
|
%shift = ashr <8 x i16> %a, %splat
|
|
ret <8 x i16> %shift
|
|
}
|
|
|
|
define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, i16 %b) {
|
|
; SSE2-LABEL: 'splatvar_shift_v16i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:2 SizeLat:2 for: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %shift = ashr <16 x i16> %a, %splat
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'splatvar_shift_v16i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %shift = ashr <16 x i16> %a, %splat
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v16i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:4 for: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:7 SizeLat:7 for: %shift = ashr <16 x i16> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v16i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <16 x i16> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v16i16'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:4 for: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:7 SizeLat:7 for: %shift = ashr <16 x i16> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v16i16'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <16 x i16> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatvar_shift_v16i16'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <16 x i16> %a, %splat
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
%insert = insertelement <16 x i16> undef, i16 %b, i32 0
|
|
%splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
|
|
%shift = ashr <16 x i16> %a, %splat
|
|
ret <16 x i16> %shift
|
|
}
|
|
|
|
define <32 x i16> @splatvar_shift_v32i16(<32 x i16> %a, i16 %b) {
|
|
; SSE2-LABEL: 'splatvar_shift_v32i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:2 SizeLat:2 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %shift = ashr <32 x i16> %a, %splat
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'splatvar_shift_v32i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:8 CodeSize:4 Lat:8 SizeLat:8 for: %shift = ashr <32 x i16> %a, %splat
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v32i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:4 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:14 SizeLat:14 for: %shift = ashr <32 x i16> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v32i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:6 for: %shift = ashr <32 x i16> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v32i16'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:4 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:14 SizeLat:14 for: %shift = ashr <32 x i16> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v32i16'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:6 for: %shift = ashr <32 x i16> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatvar_shift_v32i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:10 SizeLat:7 for: %shift = ashr <32 x i16> %a, %splat
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatvar_shift_v32i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <32 x i16> %a, %splat
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatvar_shift_v32i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:10 SizeLat:7 for: %shift = ashr <32 x i16> %a, %splat
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatvar_shift_v32i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <32 x i16> %a, %splat
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatvar_shift_v32i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:4 SizeLat:3 for: %shift = ashr <32 x i16> %a, %splat
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
%insert = insertelement <32 x i16> undef, i16 %b, i32 0
|
|
%splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
|
|
%shift = ashr <32 x i16> %a, %splat
|
|
ret <32 x i16> %shift
|
|
}
|
|
|
|
define <16 x i8> @splatvar_shift_v16i8(<16 x i8> %a, i8 %b) {
|
|
; SSE2-LABEL: 'splatvar_shift_v16i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:4 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:11 CodeSize:9 Lat:15 SizeLat:13 for: %shift = ashr <16 x i8> %a, %splat
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'splatvar_shift_v16i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:11 CodeSize:9 Lat:15 SizeLat:13 for: %shift = ashr <16 x i8> %a, %splat
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v16i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:9 Lat:6 SizeLat:13 for: %shift = ashr <16 x i8> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v16i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:9 Lat:5 SizeLat:13 for: %shift = ashr <16 x i8> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v16i8'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <16 x i8> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v16i8'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:9 Lat:5 SizeLat:13 for: %shift = ashr <16 x i8> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatvar_shift_v16i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:4 CodeSize:9 Lat:5 SizeLat:13 for: %shift = ashr <16 x i8> %a, %splat
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatvar_shift_v16i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:4 CodeSize:8 Lat:12 SizeLat:12 for: %shift = ashr <16 x i8> %a, %splat
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatvar_shift_v16i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:4 CodeSize:9 Lat:5 SizeLat:13 for: %shift = ashr <16 x i8> %a, %splat
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatvar_shift_v16i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:4 CodeSize:8 Lat:12 SizeLat:12 for: %shift = ashr <16 x i8> %a, %splat
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatvar_shift_v16i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:4 CodeSize:8 Lat:12 SizeLat:12 for: %shift = ashr <16 x i8> %a, %splat
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
%insert = insertelement <16 x i8> undef, i8 %b, i32 0
|
|
%splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
|
|
%shift = ashr <16 x i8> %a, %splat
|
|
ret <16 x i8> %shift
|
|
}
|
|
|
|
define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
|
|
; SSE2-LABEL: 'splatvar_shift_v32i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:4 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:22 CodeSize:18 Lat:30 SizeLat:26 for: %shift = ashr <32 x i8> %a, %splat
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'splatvar_shift_v32i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:22 CodeSize:18 Lat:30 SizeLat:26 for: %shift = ashr <32 x i8> %a, %splat
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v32i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:3 Lat:4 SizeLat:6 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:10 CodeSize:16 Lat:11 SizeLat:21 for: %shift = ashr <32 x i8> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v32i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:6 CodeSize:11 Lat:9 SizeLat:16 for: %shift = ashr <32 x i8> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:3 Lat:4 SizeLat:6 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <32 x i8> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:6 CodeSize:11 Lat:9 SizeLat:16 for: %shift = ashr <32 x i8> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatvar_shift_v32i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:6 CodeSize:11 Lat:9 SizeLat:16 for: %shift = ashr <32 x i8> %a, %splat
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatvar_shift_v32i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:10 SizeLat:13 for: %shift = ashr <32 x i8> %a, %splat
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatvar_shift_v32i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:6 CodeSize:11 Lat:9 SizeLat:16 for: %shift = ashr <32 x i8> %a, %splat
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatvar_shift_v32i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:10 SizeLat:13 for: %shift = ashr <32 x i8> %a, %splat
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatvar_shift_v32i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:10 SizeLat:13 for: %shift = ashr <32 x i8> %a, %splat
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
%insert = insertelement <32 x i8> undef, i8 %b, i32 0
|
|
%splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
|
|
%shift = ashr <32 x i8> %a, %splat
|
|
ret <32 x i8> %shift
|
|
}
|
|
|
|
define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
|
|
; SSE2-LABEL: 'splatvar_shift_v64i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:4 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:44 CodeSize:36 Lat:60 SizeLat:52 for: %shift = ashr <64 x i8> %a, %splat
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'splatvar_shift_v64i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:44 CodeSize:36 Lat:60 SizeLat:52 for: %shift = ashr <64 x i8> %a, %splat
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatvar_shift_v64i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:3 Lat:4 SizeLat:6 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:20 CodeSize:32 Lat:22 SizeLat:42 for: %shift = ashr <64 x i8> %a, %splat
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatvar_shift_v64i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:12 CodeSize:22 Lat:18 SizeLat:32 for: %shift = ashr <64 x i8> %a, %splat
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:3 Lat:4 SizeLat:6 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:12 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <64 x i8> %a, %splat
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:2 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:12 CodeSize:22 Lat:18 SizeLat:32 for: %shift = ashr <64 x i8> %a, %splat
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatvar_shift_v64i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:37 CodeSize:51 Lat:37 SizeLat:63 for: %shift = ashr <64 x i8> %a, %splat
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:10 SizeLat:15 for: %shift = ashr <64 x i8> %a, %splat
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatvar_shift_v64i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:37 CodeSize:51 Lat:37 SizeLat:63 for: %shift = ashr <64 x i8> %a, %splat
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatvar_shift_v64i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:10 SizeLat:15 for: %shift = ashr <64 x i8> %a, %splat
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatvar_shift_v64i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:10 SizeLat:15 for: %shift = ashr <64 x i8> %a, %splat
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
%insert = insertelement <64 x i8> undef, i8 %b, i32 0
|
|
%splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
|
|
%shift = ashr <64 x i8> %a, %splat
|
|
ret <64 x i8> %shift
|
|
}
|
|
|
|
;
|
|
; Constant Shifts
|
|
;
|
|
|
|
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v2i64'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:12 Lat:11 SizeLat:16 for: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v2i64'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:8 CodeSize:5 Lat:17 SizeLat:7 for: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v2i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:5 CodeSize:10 Lat:6 SizeLat:14 for: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v2i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:5 SizeLat:5 for: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v2i64'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'constant_shift_v2i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
%shift = ashr <2 x i64> %a, <i64 1, i64 7>
|
|
ret <2 x i64> %shift
|
|
}
|
|
|
|
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v4i64'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:24 Lat:22 SizeLat:32 for: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v4i64'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:16 CodeSize:10 Lat:34 SizeLat:14 for: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v4i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:12 CodeSize:22 Lat:12 SizeLat:30 for: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v4i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:5 Lat:8 SizeLat:9 for: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v4i64'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'constant_shift_v4i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
%shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
|
|
ret <4 x i64> %shift
|
|
}
|
|
|
|
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v8i64'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:32 CodeSize:48 Lat:44 SizeLat:64 for: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v8i64'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:32 CodeSize:20 Lat:68 SizeLat:28 for: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v8i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:24 CodeSize:44 Lat:24 SizeLat:60 for: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v8i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:10 Lat:16 SizeLat:18 for: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v8i64'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'constant_shift_v8i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
%shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
|
|
ret <8 x i64> %shift
|
|
}
|
|
|
|
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v4i32'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:12 CodeSize:15 Lat:12 SizeLat:19 for: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v4i32'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:16 CodeSize:15 Lat:17 SizeLat:19 for: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v4i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:12 Lat:7 SizeLat:16 for: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v4i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:3 SizeLat:3 for: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v4i32'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'constant_shift_v4i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
%shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
|
|
ret <4 x i32> %shift
|
|
}
|
|
|
|
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v8i32'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:24 CodeSize:30 Lat:24 SizeLat:38 for: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v8i32'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:32 CodeSize:30 Lat:34 SizeLat:38 for: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v8i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:14 CodeSize:26 Lat:14 SizeLat:34 for: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v8i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:3 for: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'constant_shift_v8i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'constant_shift_v8i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:1 Lat:4 SizeLat:3 for: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'constant_shift_v8i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
%shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
ret <8 x i32> %shift
|
|
}
|
|
|
|
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v16i32'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:48 CodeSize:60 Lat:48 SizeLat:76 for: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v16i32'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:64 CodeSize:60 Lat:68 SizeLat:76 for: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v16i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:28 CodeSize:52 Lat:28 SizeLat:68 for: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v16i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:2 Lat:8 SizeLat:6 for: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'constant_shift_v16i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'constant_shift_v16i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:2 Lat:8 SizeLat:6 for: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'constant_shift_v16i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
%shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
|
|
ret <16 x i32> %shift
|
|
}
|
|
|
|
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v8i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:31 Lat:19 SizeLat:31 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v8i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:22 CodeSize:23 Lat:26 SizeLat:27 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v8i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:13 CodeSize:14 Lat:16 SizeLat:22 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v8i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:11 SizeLat:10 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v8i16'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'constant_shift_v8i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:11 SizeLat:10 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'constant_shift_v8i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'constant_shift_v8i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:11 SizeLat:10 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'constant_shift_v8i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'constant_shift_v8i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
%shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
ret <8 x i16> %shift
|
|
}
|
|
|
|
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v16i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:32 CodeSize:62 Lat:38 SizeLat:62 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v16i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:44 CodeSize:46 Lat:52 SizeLat:54 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v16i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:28 CodeSize:31 Lat:30 SizeLat:48 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v16i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:14 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v16i16'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'constant_shift_v16i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:14 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'constant_shift_v16i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'constant_shift_v16i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:14 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'constant_shift_v16i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'constant_shift_v16i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
%shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
ret <16 x i16> %shift
|
|
}
|
|
|
|
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v32i16'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:64 CodeSize:124 Lat:76 SizeLat:124 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v32i16'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:88 CodeSize:92 Lat:104 SizeLat:108 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v32i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:56 CodeSize:62 Lat:60 SizeLat:96 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v32i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:16 CodeSize:20 Lat:20 SizeLat:28 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v32i16'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'constant_shift_v32i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:11 CodeSize:11 Lat:16 SizeLat:15 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'constant_shift_v32i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'constant_shift_v32i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:11 CodeSize:11 Lat:16 SizeLat:15 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'constant_shift_v32i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'constant_shift_v32i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
%shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
|
|
ret <32 x i16> %shift
|
|
}
|
|
|
|
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v16i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:27 CodeSize:54 Lat:30 SizeLat:54 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v16i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:38 CodeSize:30 Lat:41 SizeLat:36 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v16i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:21 CodeSize:24 Lat:22 SizeLat:36 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v16i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:17 CodeSize:24 Lat:17 SizeLat:30 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v16i8'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'constant_shift_v16i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:17 CodeSize:24 Lat:17 SizeLat:30 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'constant_shift_v16i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'constant_shift_v16i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:17 CodeSize:24 Lat:17 SizeLat:30 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'constant_shift_v16i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'constant_shift_v16i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:4 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
%shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
ret <16 x i8> %shift
|
|
}
|
|
|
|
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v32i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:54 CodeSize:108 Lat:60 SizeLat:108 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v32i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:76 CodeSize:60 Lat:82 SizeLat:72 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v32i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:44 CodeSize:51 Lat:45 SizeLat:76 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v32i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:18 CodeSize:24 Lat:20 SizeLat:43 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v32i8'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'constant_shift_v32i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:18 CodeSize:24 Lat:20 SizeLat:43 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'constant_shift_v32i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:6 CodeSize:24 Lat:13 SizeLat:30 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'constant_shift_v32i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:18 CodeSize:24 Lat:20 SizeLat:43 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'constant_shift_v32i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:6 CodeSize:24 Lat:13 SizeLat:30 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'constant_shift_v32i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:6 CodeSize:24 Lat:13 SizeLat:30 for: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
%shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
ret <32 x i8> %shift
|
|
}
|
|
|
|
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
|
|
; SSE2-LABEL: 'constant_shift_v64i8'
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:108 CodeSize:216 Lat:120 SizeLat:216 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; SSE42-LABEL: 'constant_shift_v64i8'
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:152 CodeSize:120 Lat:164 SizeLat:144 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'constant_shift_v64i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:88 CodeSize:102 Lat:90 SizeLat:152 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'constant_shift_v64i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:36 CodeSize:48 Lat:40 SizeLat:86 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; XOP-LABEL: 'constant_shift_v64i8'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'constant_shift_v64i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:37 CodeSize:51 Lat:37 SizeLat:63 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'constant_shift_v64i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:15 CodeSize:30 Lat:15 SizeLat:30 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'constant_shift_v64i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:37 CodeSize:51 Lat:37 SizeLat:63 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'constant_shift_v64i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:15 CodeSize:30 Lat:15 SizeLat:30 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'constant_shift_v64i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:15 CodeSize:30 Lat:15 SizeLat:30 for: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
%shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
|
|
ret <64 x i8> %shift
|
|
}
|
|
|
|
;
|
|
; Uniform Constant Shifts
|
|
;
|
|
|
|
define <2 x i64> @splatconstant_shift_v2i64(<2 x i64> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v2i64'
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:3 CodeSize:6 Lat:5 SizeLat:6 for: %shift = ashr <2 x i64> %a, splat (i64 7)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX-LABEL: 'splatconstant_shift_v2i64'
|
|
; AVX-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:3 for: %shift = ashr <2 x i64> %a, splat (i64 7)
|
|
; AVX-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; XOP-LABEL: 'splatconstant_shift_v2i64'
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:2 CodeSize:3 Lat:3 SizeLat:3 for: %shift = ashr <2 x i64> %a, splat (i64 7)
|
|
; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatconstant_shift_v2i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <2 x i64> %a, splat (i64 7)
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %shift
|
|
;
|
|
%shift = ashr <2 x i64> %a, <i64 7, i64 7>
|
|
ret <2 x i64> %shift
|
|
}
|
|
|
|
define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v4i64'
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:6 CodeSize:12 Lat:10 SizeLat:12 for: %shift = ashr <4 x i64> %a, splat (i64 7)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v4i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:5 CodeSize:8 Lat:7 SizeLat:9 for: %shift = ashr <4 x i64> %a, splat (i64 7)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v4i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:3 Lat:4 SizeLat:6 for: %shift = ashr <4 x i64> %a, splat (i64 7)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v4i64'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:5 CodeSize:8 Lat:7 SizeLat:9 for: %shift = ashr <4 x i64> %a, splat (i64 7)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v4i64'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:3 Lat:4 SizeLat:6 for: %shift = ashr <4 x i64> %a, splat (i64 7)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatconstant_shift_v4i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i64> %a, splat (i64 7)
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i64> %shift
|
|
;
|
|
%shift = ashr <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
|
|
ret <4 x i64> %shift
|
|
}
|
|
|
|
define <8 x i64> @splatconstant_shift_v8i64(<8 x i64> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v8i64'
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:12 CodeSize:24 Lat:20 SizeLat:24 for: %shift = ashr <8 x i64> %a, splat (i64 7)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v8i64'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:10 CodeSize:16 Lat:14 SizeLat:18 for: %shift = ashr <8 x i64> %a, splat (i64 7)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v8i64'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:6 Lat:8 SizeLat:12 for: %shift = ashr <8 x i64> %a, splat (i64 7)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v8i64'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:10 CodeSize:16 Lat:14 SizeLat:18 for: %shift = ashr <8 x i64> %a, splat (i64 7)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v8i64'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:6 Lat:8 SizeLat:12 for: %shift = ashr <8 x i64> %a, splat (i64 7)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatconstant_shift_v8i64'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i64> %a, splat (i64 7)
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i64> %shift
|
|
;
|
|
%shift = ashr <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
|
|
ret <8 x i64> %shift
|
|
}
|
|
|
|
define <4 x i32> @splatconstant_shift_v4i32(<4 x i32> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v4i32'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i32> %a, splat (i32 5)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v4i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:1 for: %shift = ashr <4 x i32> %a, splat (i32 5)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v4i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i32> %a, splat (i32 5)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v4i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:1 for: %shift = ashr <4 x i32> %a, splat (i32 5)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v4i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i32> %a, splat (i32 5)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatconstant_shift_v4i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <4 x i32> %a, splat (i32 5)
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %shift
|
|
;
|
|
%shift = ashr <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
|
|
ret <4 x i32> %shift
|
|
}
|
|
|
|
define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v8i32'
|
|
; SSE-NEXT: Cost Model: Found costs of 2 for: %shift = ashr <8 x i32> %a, splat (i32 5)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v8i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:4 Lat:6 SizeLat:5 for: %shift = ashr <8 x i32> %a, splat (i32 5)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v8i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <8 x i32> %a, splat (i32 5)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v8i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:4 Lat:6 SizeLat:5 for: %shift = ashr <8 x i32> %a, splat (i32 5)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v8i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <8 x i32> %a, splat (i32 5)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatconstant_shift_v8i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i32> %a, splat (i32 5)
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i32> %shift
|
|
;
|
|
%shift = ashr <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
|
|
ret <8 x i32> %shift
|
|
}
|
|
|
|
define <16 x i32> @splatconstant_shift_v16i32(<16 x i32> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v16i32'
|
|
; SSE-NEXT: Cost Model: Found costs of 4 for: %shift = ashr <16 x i32> %a, splat (i32 5)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v16i32'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:8 Lat:12 SizeLat:10 for: %shift = ashr <16 x i32> %a, splat (i32 5)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v16i32'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %shift = ashr <16 x i32> %a, splat (i32 5)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v16i32'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:8 Lat:12 SizeLat:10 for: %shift = ashr <16 x i32> %a, splat (i32 5)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v16i32'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %shift = ashr <16 x i32> %a, splat (i32 5)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatconstant_shift_v16i32'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i32> %a, splat (i32 5)
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i32> %shift
|
|
;
|
|
%shift = ashr <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
|
|
ret <16 x i32> %shift
|
|
}
|
|
|
|
define <8 x i16> @splatconstant_shift_v8i16(<8 x i16> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v8i16'
|
|
; SSE-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, splat (i16 3)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v8i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:1 for: %shift = ashr <8 x i16> %a, splat (i16 3)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v8i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, splat (i16 3)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v8i16'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:2 SizeLat:1 for: %shift = ashr <8 x i16> %a, splat (i16 3)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v8i16'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, splat (i16 3)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
; AVX512-LABEL: 'splatconstant_shift_v8i16'
|
|
; AVX512-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <8 x i16> %a, splat (i16 3)
|
|
; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i16> %shift
|
|
;
|
|
%shift = ashr <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
|
|
ret <8 x i16> %shift
|
|
}
|
|
|
|
define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v16i16'
|
|
; SSE-NEXT: Cost Model: Found costs of 2 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v16i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:4 Lat:6 SizeLat:5 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v16i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v16i16'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:4 Lat:6 SizeLat:5 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v16i16'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatconstant_shift_v16i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:2 CodeSize:4 Lat:7 SizeLat:4 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatconstant_shift_v16i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatconstant_shift_v16i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:2 CodeSize:4 Lat:7 SizeLat:4 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatconstant_shift_v16i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatconstant_shift_v16i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <16 x i16> %a, splat (i16 3)
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i16> %shift
|
|
;
|
|
%shift = ashr <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
|
|
ret <16 x i16> %shift
|
|
}
|
|
|
|
define <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v32i16'
|
|
; SSE-NEXT: Cost Model: Found costs of 4 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v32i16'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:8 Lat:12 SizeLat:10 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v32i16'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v32i16'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:6 CodeSize:8 Lat:12 SizeLat:10 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v32i16'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:4 CodeSize:2 Lat:4 SizeLat:4 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatconstant_shift_v32i16'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:10 SizeLat:7 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatconstant_shift_v32i16'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatconstant_shift_v32i16'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:5 CodeSize:5 Lat:10 SizeLat:7 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatconstant_shift_v32i16'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatconstant_shift_v32i16'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of 1 for: %shift = ashr <32 x i16> %a, splat (i16 3)
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i16> %shift
|
|
;
|
|
%shift = ashr <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
|
|
ret <32 x i16> %shift
|
|
}
|
|
|
|
define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v16i8'
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:9 SizeLat:6 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v16i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:9 SizeLat:6 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v16i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:5 Lat:10 SizeLat:6 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v16i8'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v16i8'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:2 CodeSize:5 Lat:10 SizeLat:6 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatconstant_shift_v16i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:2 CodeSize:5 Lat:10 SizeLat:6 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatconstant_shift_v16i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatconstant_shift_v16i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:2 CodeSize:5 Lat:10 SizeLat:6 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatconstant_shift_v16i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:4 Lat:8 SizeLat:5 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatconstant_shift_v16i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:6 SizeLat:2 for: %shift = ashr <16 x i8> %a, splat (i8 3)
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <16 x i8> %shift
|
|
;
|
|
%shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
|
|
ret <16 x i8> %shift
|
|
}
|
|
|
|
define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v32i8'
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:18 SizeLat:12 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v32i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:7 CodeSize:12 Lat:7 SizeLat:13 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v32i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:10 SizeLat:9 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:4 CodeSize:5 Lat:7 SizeLat:6 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:10 SizeLat:9 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatconstant_shift_v32i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:10 SizeLat:9 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatconstant_shift_v32i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:4 Lat:9 SizeLat:5 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatconstant_shift_v32i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:10 SizeLat:9 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatconstant_shift_v32i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:4 Lat:9 SizeLat:5 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatconstant_shift_v32i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:6 SizeLat:2 for: %shift = ashr <32 x i8> %a, splat (i8 3)
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <32 x i8> %shift
|
|
;
|
|
%shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
|
|
ret <32 x i8> %shift
|
|
}
|
|
|
|
define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
|
|
; SSE-LABEL: 'splatconstant_shift_v64i8'
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:12 CodeSize:20 Lat:36 SizeLat:24 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX1-LABEL: 'splatconstant_shift_v64i8'
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:14 CodeSize:24 Lat:14 SizeLat:26 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX2-LABEL: 'splatconstant_shift_v64i8'
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:20 SizeLat:18 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:14 SizeLat:12 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; XOPAVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:6 CodeSize:10 Lat:20 SizeLat:18 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; XOPAVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512F-LABEL: 'splatconstant_shift_v64i8'
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:3 CodeSize:12 Lat:10 SizeLat:12 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BW-LABEL: 'splatconstant_shift_v64i8'
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:4 Lat:9 SizeLat:6 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512VL-LABEL: 'splatconstant_shift_v64i8'
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:3 CodeSize:12 Lat:10 SizeLat:12 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; AVX512VL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512BWVL-LABEL: 'splatconstant_shift_v64i8'
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:1 CodeSize:4 Lat:9 SizeLat:6 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; AVX512BWVL-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
; AVX512GFNI-LABEL: 'splatconstant_shift_v64i8'
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:6 SizeLat:2 for: %shift = ashr <64 x i8> %a, splat (i8 3)
|
|
; AVX512GFNI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <64 x i8> %shift
|
|
;
|
|
%shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
|
|
ret <64 x i8> %shift
|
|
}
|