maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/X86/vector-shift-shl-512.ll

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

324 lines
15 KiB
LLVM
Raw Normal View History

Make utils/update_llc_test_checks.py note that the assertions are autogenerated. Also update existing test cases which appear to be generated by it and weren't modified (other than addition of the header) by rerunning it. llvm-svn: 253917
2015-11-24 05:33:58 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[CodeGen][X86] Remove unused check-prefixes from vector shift tests. NFCI.
2020-11-11 11:43:22 +08:00
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq | FileCheck %s --check-prefixes=ALL,AVX512DQ
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefixes=ALL,AVX512BW
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
;
; Variable Shifts
;
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) nounwind {
; ALL-LABEL: var_shift_v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpsllvq %zmm1, %zmm0, %zmm0
; ALL-NEXT: retq
%shift = shl <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) nounwind {
; ALL-LABEL: var_shift_v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpsllvd %zmm1, %zmm0, %zmm0
; ALL-NEXT: retq
%shift = shl <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: var_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vpmovzxwd {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
; AVX512DQ-NEXT: vpmovzxwd {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
; AVX512DQ-NEXT: vpsllvd %zmm2, %zmm3, %zmm2
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpmovdw %zmm2, %ymm2
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm1, %ymm1
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm0
[X86][AVX512DQ] Enable v16i16 vector shifts to use an extend+shift+truncate pattern. Use the existing AVX2 v8i16 vector shift lowering for v16i16 on AVX512 targets (AVX512BW will have already have lowered with vpsravw). Cost model updates to follow. llvm-svn: 291445
2017-01-09 23:15:45 +08:00
; AVX512DQ-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllvd %zmm1, %zmm0, %zmm0
[X86][AVX512DQ] Enable v16i16 vector shifts to use an extend+shift+truncate pattern. Use the existing AVX2 v8i16 vector shift lowering for v16i16 on AVX512 targets (AVX512BW will have already have lowered with vpsravw). Cost model updates to follow. llvm-svn: 291445
2017-01-09 23:15:45 +08:00
; AVX512DQ-NEXT: vpmovdw %zmm0, %ymm0
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
;
; AVX512BW-LABEL: var_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%shift = shl <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: var_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512DQ-NEXT: vpsllw $4, %ymm2, %ymm3
; AVX512DQ-NEXT: vmovdqa {{.*#+}} ymm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
; AVX512DQ-NEXT: vpand %ymm4, %ymm3, %ymm3
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm1, %ymm5
; AVX512DQ-NEXT: vpsllw $5, %ymm5, %ymm5
; AVX512DQ-NEXT: vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
; AVX512DQ-NEXT: vpsllw $2, %ymm2, %ymm3
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vpand %ymm6, %ymm3, %ymm3
; AVX512DQ-NEXT: vpaddb %ymm5, %ymm5, %ymm5
; AVX512DQ-NEXT: vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
; AVX512DQ-NEXT: vpaddb %ymm2, %ymm2, %ymm3
; AVX512DQ-NEXT: vpaddb %ymm5, %ymm5, %ymm5
; AVX512DQ-NEXT: vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllw $4, %ymm0, %ymm3
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vpand %ymm4, %ymm3, %ymm3
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllw $5, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
; AVX512DQ-NEXT: vpsllw $2, %ymm0, %ymm3
; AVX512DQ-NEXT: vpand %ymm6, %ymm3, %ymm3
; AVX512DQ-NEXT: vpaddb %ymm1, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
; AVX512DQ-NEXT: vpaddb %ymm0, %ymm0, %ymm3
; AVX512DQ-NEXT: vpaddb %ymm1, %ymm1, %ymm1
; AVX512DQ-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
[X86][AVX512BW] Updated tests to demonstrate AVX512BW's inability to vectorize v64i8 shifts llvm-svn: 278790
2016-08-16 19:05:47 +08:00
;
; AVX512BW-LABEL: var_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
[X86][AVX512BW] Vectorize v64i8 vector shifts Differential Revision: https://reviews.llvm.org/D28447 llvm-svn: 291665
2017-01-11 18:36:51 +08:00
; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm2
; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm2, %zmm2
; AVX512BW-NEXT: vpsllw $5, %zmm1, %zmm1
; AVX512BW-NEXT: vpmovb2m %zmm1, %k1
; AVX512BW-NEXT: vmovdqu8 %zmm2, %zmm0 {%k1}
; AVX512BW-NEXT: vpsllw $2, %zmm0, %zmm2
; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm2, %zmm2
; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1
; AVX512BW-NEXT: vpmovb2m %zmm1, %k1
; AVX512BW-NEXT: vmovdqu8 %zmm2, %zmm0 {%k1}
; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1
; AVX512BW-NEXT: vpmovb2m %zmm1, %k1
; AVX512BW-NEXT: vpaddb %zmm0, %zmm0, %zmm0 {%k1}
[X86][AVX512BW] Updated tests to demonstrate AVX512BW's inability to vectorize v64i8 shifts llvm-svn: 278790
2016-08-16 19:05:47 +08:00
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%shift = shl <64 x i8> %a, %b
ret <64 x i8> %shift
}
;
; Uniform Variable Shifts
;
define <8 x i64> @splatvar_shift_v8i64(<8 x i64> %a, <8 x i64> %b) nounwind {
; ALL-LABEL: splatvar_shift_v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpsllq %xmm1, %zmm0, %zmm0
; ALL-NEXT: retq
%splat = shufflevector <8 x i64> %b, <8 x i64> undef, <8 x i32> zeroinitializer
%shift = shl <8 x i64> %a, %splat
ret <8 x i64> %shift
}
define <16 x i32> @splatvar_shift_v16i32(<16 x i32> %a, <16 x i32> %b) nounwind {
; ALL-LABEL: splatvar_shift_v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86] Optimize vector shifts with variable but uniform shift amounts Summary: For instructions such as PSLLW/PSLLD/PSLLQ a variable shift amount may be passed in an XMM register. The lower 64-bits of the register are evaluated to determine the shift amount. This patch improves the construction of the vector containing the shift amount. Reviewers: craig.topper, delena, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D28353 llvm-svn: 291120
2017-01-05 23:11:43 +08:00
; ALL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpslld %xmm1, %zmm0, %zmm0
; ALL-NEXT: retq
%splat = shufflevector <16 x i32> %b, <16 x i32> undef, <16 x i32> zeroinitializer
%shift = shl <16 x i32> %a, %splat
ret <16 x i32> %shift
}
define <32 x i16> @splatvar_shift_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: splatvar_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm2
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllw %xmm1, %ymm2, %ymm2
; AVX512DQ-NEXT: vpsllw %xmm1, %ymm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
;
; AVX512BW-LABEL: splatvar_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
[X86] Optimize vector shifts with variable but uniform shift amounts Summary: For instructions such as PSLLW/PSLLD/PSLLQ a variable shift amount may be passed in an XMM register. The lower 64-bits of the register are evaluated to determine the shift amount. This patch improves the construction of the vector containing the shift amount. Reviewers: craig.topper, delena, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D28353 llvm-svn: 291120
2017-01-05 23:11:43 +08:00
; AVX512BW-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512BW-NEXT: vpsllw %xmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%splat = shufflevector <32 x i16> %b, <32 x i16> undef, <32 x i32> zeroinitializer
%shift = shl <32 x i16> %a, %splat
ret <32 x i16> %shift
}
define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: splatvar_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm2
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllw %xmm1, %ymm2, %ymm2
; AVX512DQ-NEXT: vpsllw %xmm1, %ymm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
[X86][AVX512] Fold concat(and(x,y),and(z,w)) -> and(concat(x,z),concat(y,w)) for 512-bit vectors Helps vpternlog folding on non-AVX512BW targets
2020-08-02 03:34:39 +08:00
; AVX512DQ-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512DQ-NEXT: vpsllw %xmm1, %xmm2, %xmm1
; AVX512DQ-NEXT: vpbroadcastb %xmm1, %ymm1
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm1, %zmm1, %zmm1
; AVX512DQ-NEXT: vpandq %zmm1, %zmm0, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
[X86][AVX512BW] Updated tests to demonstrate AVX512BW's inability to vectorize v64i8 shifts llvm-svn: 278790
2016-08-16 19:05:47 +08:00
;
; AVX512BW-LABEL: splatvar_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
[X86][SSE] Improve variable scalar shift of vXi8 vectors (PR34694) This patch creates the shift mask and actual shift using the vXi16 vector shift ops. Differential Revision: https://reviews.llvm.org/D51263 llvm-svn: 340813
2018-08-28 18:37:29 +08:00
; AVX512BW-NEXT: vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
; AVX512BW-NEXT: vpsllw %xmm1, %zmm0, %zmm0
[X86][SSE] Add SSE vector shift support to SimplifyDemandedVectorEltsForTargetNode vector splitting llvm-svn: 359680
2019-05-01 21:51:09 +08:00
; AVX512BW-NEXT: vpcmpeqd %xmm2, %xmm2, %xmm2
; AVX512BW-NEXT: vpsllw %xmm1, %xmm2, %xmm1
[X86][AVX512BW] Updated tests to demonstrate AVX512BW's inability to vectorize v64i8 shifts llvm-svn: 278790
2016-08-16 19:05:47 +08:00
; AVX512BW-NEXT: vpbroadcastb %xmm1, %zmm1
[X86][SSE] Improve variable scalar shift of vXi8 vectors (PR34694) This patch creates the shift mask and actual shift using the vXi16 vector shift ops. Differential Revision: https://reviews.llvm.org/D51263 llvm-svn: 340813
2018-08-28 18:37:29 +08:00
; AVX512BW-NEXT: vpandq %zmm1, %zmm0, %zmm0
[X86][AVX512BW] Updated tests to demonstrate AVX512BW's inability to vectorize v64i8 shifts llvm-svn: 278790
2016-08-16 19:05:47 +08:00
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%splat = shufflevector <64 x i8> %b, <64 x i8> undef, <64 x i32> zeroinitializer
%shift = shl <64 x i8> %a, %splat
ret <64 x i8> %shift
}
;
; Constant Shifts
;
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) nounwind {
; ALL-LABEL: constant_shift_v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpsllvq {{.*}}(%rip), %zmm0, %zmm0
; ALL-NEXT: retq
%shift = shl <8 x i64> %a, <i64 1, i64 7, i64 31, i64 62, i64 1, i64 7, i64 31, i64 62>
ret <8 x i64> %shift
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) nounwind {
; ALL-LABEL: constant_shift_v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpsllvd {{.*}}(%rip), %zmm0, %zmm0
; ALL-NEXT: retq
%shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 8, i32 7, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 8, i32 7>
ret <16 x i32> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: constant_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm1
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: vmovdqa {{.*#+}} ymm2 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768]
; AVX512DQ-NEXT: vpmullw %ymm2, %ymm1, %ymm1
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpmullw %ymm2, %ymm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
;
; AVX512BW-LABEL: constant_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512BW-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
ret <32 x i16> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: constant_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm1
; AVX512DQ-NEXT: vpsllw $4, %ymm1, %ymm2
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
; AVX512DQ-NEXT: vpand %ymm3, %ymm2, %ymm2
[X86][AVX] X86ISD::PERMV/PERMV3 node types can never fold index ops Improves codegen demonstrated by D60512 - instructions represented by X86ISD::PERMV/PERMV3 can never memory fold the operand used for their index register. This patch updates the 'isUseOfShuffle' helper into the more capable 'isFoldableUseOfShuffle' that recognises that the op is used for a X86ISD::PERMV/PERMV3 index mask and can't be folded - allowing us to use broadcast/subvector-broadcast ops to reduce the size of the mask constant pool data. Differential Revision: https://reviews.llvm.org/D60562 llvm-svn: 358516
2019-04-17 03:18:53 +08:00
; AVX512DQ-NEXT: vbroadcasti128 {{.*#+}} ymm4 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
; AVX512DQ-NEXT: # ymm4 = mem[0,1,0,1]
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: vpblendvb %ymm4, %ymm2, %ymm1, %ymm1
; AVX512DQ-NEXT: vpsllw $2, %ymm1, %ymm2
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vmovdqa {{.*#+}} ymm5 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: vpand %ymm5, %ymm2, %ymm2
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpaddb %ymm4, %ymm4, %ymm6
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: vpblendvb %ymm6, %ymm2, %ymm1, %ymm1
; AVX512DQ-NEXT: vpaddb %ymm1, %ymm1, %ymm2
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpaddb %ymm6, %ymm6, %ymm7
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: vpblendvb %ymm7, %ymm2, %ymm1, %ymm1
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllw $4, %ymm0, %ymm2
; AVX512DQ-NEXT: vpand %ymm3, %ymm2, %ymm2
; AVX512DQ-NEXT: vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
; AVX512DQ-NEXT: vpsllw $2, %ymm0, %ymm2
; AVX512DQ-NEXT: vpand %ymm5, %ymm2, %ymm2
; AVX512DQ-NEXT: vpblendvb %ymm6, %ymm2, %ymm0, %ymm0
; AVX512DQ-NEXT: vpaddb %ymm0, %ymm0, %ymm2
; AVX512DQ-NEXT: vpblendvb %ymm7, %ymm2, %ymm0, %ymm0
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
[X86][AVX512BW] Updated tests to demonstrate AVX512BW's inability to vectorize v64i8 shifts llvm-svn: 278790
2016-08-16 19:05:47 +08:00
;
; AVX512BW-LABEL: constant_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
[X86][AVX] X86ISD::PERMV/PERMV3 node types can never fold index ops Improves codegen demonstrated by D60512 - instructions represented by X86ISD::PERMV/PERMV3 can never memory fold the operand used for their index register. This patch updates the 'isUseOfShuffle' helper into the more capable 'isFoldableUseOfShuffle' that recognises that the op is used for a X86ISD::PERMV/PERMV3 index mask and can't be folded - allowing us to use broadcast/subvector-broadcast ops to reduce the size of the mask constant pool data. Differential Revision: https://reviews.llvm.org/D60562 llvm-svn: 358516
2019-04-17 03:18:53 +08:00
; AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm1 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
; AVX512BW-NEXT: # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
[X86][AVX512BW] Vectorize v64i8 vector shifts Differential Revision: https://reviews.llvm.org/D28447 llvm-svn: 291665
2017-01-11 18:36:51 +08:00
; AVX512BW-NEXT: vpmovb2m %zmm1, %k1
; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm2
; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm2, %zmm2
; AVX512BW-NEXT: vmovdqu8 %zmm2, %zmm0 {%k1}
; AVX512BW-NEXT: vpsllw $2, %zmm0, %zmm2
; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm2, %zmm2
; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1
; AVX512BW-NEXT: vpmovb2m %zmm1, %k1
; AVX512BW-NEXT: vmovdqu8 %zmm2, %zmm0 {%k1}
; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1
; AVX512BW-NEXT: vpmovb2m %zmm1, %k1
; AVX512BW-NEXT: vpaddb %zmm0, %zmm0, %zmm0 {%k1}
[X86][AVX512BW] Updated tests to demonstrate AVX512BW's inability to vectorize v64i8 shifts llvm-svn: 278790
2016-08-16 19:05:47 +08:00
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%shift = shl <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 <8 x i64> @splatconstant_shift_v8i64(<8 x i64> %a) nounwind {
; ALL-LABEL: splatconstant_shift_v8i64:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpsllq $7, %zmm0, %zmm0
; ALL-NEXT: retq
%shift = shl <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 <16 x i32> @splatconstant_shift_v16i32(<16 x i32> %a) nounwind {
; ALL-LABEL: splatconstant_shift_v16i32:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; ALL: # %bb.0:
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
; ALL-NEXT: vpslld $5, %zmm0, %zmm0
; ALL-NEXT: retq
%shift = shl <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 <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: splatconstant_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vpsllw $3, %ymm0, %ymm1
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm0
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllw $3, %ymm0, %ymm0
[X86] Make v32i16/v64i8 legal types without avx512bw. Use custom splitting instead. This moves v32i16/v64i8 to a model consistent with how we treat integer types with avx1. This does change the ABI for types vXi16/vXi8 vectors larger than 512 bits to pass in multiple zmms instead of multiple ymms. We'd already hacked some code to make v64i8/v32i16 pass in zmm. Cost model is still a bit of a mess. In some place I tried to match existing behavior. But really we need to account for splitting and concating costs. Cost model for shuffles is especially pessimistic. Differential Revision: https://reviews.llvm.org/D76212
2020-04-16 03:03:39 +08:00
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_shift_v32i16:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512BW-NEXT: vpsllw $3, %zmm0, %zmm0
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%shift = shl <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 <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) nounwind {
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-LABEL: splatconstant_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512DQ: # %bb.0:
[X86][AVX512] Fold concat(and(x,y),and(z,w)) -> and(concat(x,z),concat(y,w)) for 512-bit vectors Helps vpternlog folding on non-AVX512BW targets
2020-08-02 03:34:39 +08:00
; AVX512DQ-NEXT: vpsllw $3, %ymm0, %ymm1
; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm0
[X86] Pass v32i16/v64i8 in zmm registers on KNL target. gcc and icc pass these types in zmm registers in zmm registers. This patch implements a quick hack to override the register type before calling convention handling to one that is legal. Longer term we might want to do something similar to 256-bit integer registers on AVX1 where we just split all the operations. Fixes PR42957 Differential Revision: https://reviews.llvm.org/D66708 llvm-svn: 370495
2019-08-31 01:35:08 +08:00
; AVX512DQ-NEXT: vpsllw $3, %ymm0, %ymm0
[X86][AVX512] Fold concat(and(x,y),and(z,w)) -> and(concat(x,z),concat(y,w)) for 512-bit vectors Helps vpternlog folding on non-AVX512BW targets
2020-08-02 03:34:39 +08:00
; AVX512DQ-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0
; AVX512DQ-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512DQ-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_shift_v64i8:
[CodeGen] Unify MBB reference format in both MIR and debug output As part of the unification of the debug format and the MIR format, print MBB references as '%bb.5'. The MIR printer prints the IR name of a MBB only for block definitions. * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g' * find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g' * find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g' * grep -nr 'BB#' and fix Differential Revision: https://reviews.llvm.org/D40422 llvm-svn: 319665
2017-12-05 01:18:51 +08:00
; AVX512BW: # %bb.0:
AVX512BW: Enable packed word shift for 512bit vector. Enable lowering scalar immidiate shift v64i8 .Fix predicate for AVX1/2 shifts. Differential Revision: http://reviews.llvm.org/D15713 llvm-svn: 256324
2015-12-23 16:06:50 +08:00
; AVX512BW-NEXT: vpsllw $3, %zmm0, %zmm0
; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: retq
[X86][AVX512] Added 512-bit vector shift tests. Only works for avx512f (dq) targets so far - need to add avx512bw tests once char/short shifts are supported. llvm-svn: 246943
2015-09-06 21:36:32 +08:00
%shift = shl <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
}