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/avx512-vpermv3-commute.ll

339 lines
16 KiB
LLVM
Raw Normal View History

[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=cannonlake | FileCheck %s
; These test cases demonstrate cases where vpermt2/vpermi2 could benefit from being commuted.
declare <16 x i32> @llvm.x86.avx512.mask.vpermi2var.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
define <16 x i32>@test_int_x86_avx512_mask_vpermi2var_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_d_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2d (%rdi), %zmm1, %zmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2 = load <16 x i32>, <16 x i32>* %x2p
%res = call <16 x i32> @llvm.x86.avx512.mask.vpermi2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
ret <16 x i32> %res
}
declare <8 x double> @llvm.x86.avx512.mask.vpermi2var.pd.512(<8 x double>, <8 x i64>, <8 x double>, i8)
define <8 x double>@test_int_x86_avx512_mask_vpermi2var_pd_512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_pd_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2pd %zmm2, %zmm1, %zmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask.vpermi2var.pd.512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 -1)
ret <8 x double> %res
}
declare <16 x float> @llvm.x86.avx512.mask.vpermi2var.ps.512(<16 x float>, <16 x i32>, <16 x float>, i16)
define <16 x float>@test_int_x86_avx512_mask_vpermi2var_ps_512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_ps_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask.vpermi2var.ps.512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 -1)
ret <16 x float> %res
}
declare <8 x i64> @llvm.x86.avx512.mask.vpermi2var.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
define <8 x i64>@test_int_x86_avx512_mask_vpermi2var_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_q_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2q %zmm2, %zmm1, %zmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <8 x i64> @llvm.x86.avx512.mask.vpermi2var.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
ret <8 x i64> %res
}
declare <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
define <16 x i32>@test_int_x86_avx512_maskz_vpermt2var_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_d_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %esi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2d (%rdi), %zmm1, %zmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2 = load <16 x i32>, <16 x i32>* %x2p
%res = call <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
ret <16 x i32> %res
}
declare <8 x double> @llvm.x86.avx512.maskz.vpermt2var.pd.512(<8 x i64>, <8 x double>, <8 x double>, i8)
define <8 x double>@test_int_x86_avx512_maskz_vpermt2var_pd_512(<8 x i64> %x0, <8 x double> %x1, double* %x2ptr, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_pd_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %esi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2pd (%rdi){1to8}, %zmm1, %zmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2s = load double, double* %x2ptr
%x2ins = insertelement <8 x double> undef, double %x2s, i32 0
%x2 = shufflevector <8 x double> %x2ins, <8 x double> undef, <8 x i32> zeroinitializer
%res = call <8 x double> @llvm.x86.avx512.maskz.vpermt2var.pd.512(<8 x i64> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3)
ret <8 x double> %res
}
declare <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32>, <16 x float>, <16 x float>, i16)
define <16 x float>@test_int_x86_avx512_maskz_vpermt2var_ps_512(<16 x i32> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_ps_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %edi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2ps %zmm2, %zmm1, %zmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3)
ret <16 x float> %res
}
declare <8 x i64> @llvm.x86.avx512.maskz.vpermt2var.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
define <8 x i64>@test_int_x86_avx512_maskz_vpermt2var_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_q_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %edi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2q %zmm2, %zmm1, %zmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <8 x i64> @llvm.x86.avx512.maskz.vpermt2var.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
ret <8 x i64> %res
}
declare <16 x i32> @llvm.x86.avx512.mask.vpermt2var.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
define <16 x i32>@test_int_x86_avx512_mask_vpermt2var_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_d_512:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2d %zmm2, %zmm1, %zmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <16 x i32> @llvm.x86.avx512.mask.vpermt2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
ret <16 x i32> %res
}
declare <4 x i32> @llvm.x86.avx512.mask.vpermt2var.d.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
define <4 x i32>@test_int_x86_avx512_mask_vpermt2var_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_d_128:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2d %xmm2, %xmm1, %xmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <4 x i32> @llvm.x86.avx512.mask.vpermt2var.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 -1)
ret <4 x i32> %res
}
declare <4 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
define <4 x i32>@test_int_x86_avx512_maskz_vpermt2var_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_d_128:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %edi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2d %xmm2, %xmm1, %xmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <4 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
ret <4 x i32> %res
}
define <4 x i32>@test_int_x86_avx512_maskz_vpermt2var_d_128_broadcast(<4 x i32> %x0, <4 x i32> %x1, i32* %x2ptr, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_d_128_broadcast:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %esi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2d (%rdi){1to4}, %xmm1, %xmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2s = load i32, i32* %x2ptr
%x2ins = insertelement <4 x i32> undef, i32 %x2s, i32 0
%x2 = shufflevector <4 x i32> %x2ins, <4 x i32> undef, <4 x i32> zeroinitializer
%res = call <4 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %x3)
ret <4 x i32> %res
}
declare <8 x i32> @llvm.x86.avx512.mask.vpermt2var.d.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
define <8 x i32>@test_int_x86_avx512_mask_vpermt2var_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_d_256:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2d %ymm2, %ymm1, %ymm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <8 x i32> @llvm.x86.avx512.mask.vpermt2var.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 -1)
ret <8 x i32> %res
}
declare <8 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
define <8 x i32>@test_int_x86_avx512_maskz_vpermt2var_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_d_256:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %edi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2d %ymm2, %ymm1, %ymm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <8 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
ret <8 x i32> %res
}
declare <2 x double> @llvm.x86.avx512.mask.vpermi2var.pd.128(<2 x double>, <2 x i64>, <2 x double>, i8)
define <2 x double>@test_int_x86_avx512_mask_vpermi2var_pd_128(<2 x double> %x0, <2 x i64> %x1, <2 x double> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_pd_128:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2pd %xmm2, %xmm1, %xmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <2 x double> @llvm.x86.avx512.mask.vpermi2var.pd.128(<2 x double> %x0, <2 x i64> %x1, <2 x double> %x2, i8 -1)
ret <2 x double> %res
}
declare <4 x double> @llvm.x86.avx512.mask.vpermi2var.pd.256(<4 x double>, <4 x i64>, <4 x double>, i8)
define <4 x double>@test_int_x86_avx512_mask_vpermi2var_pd_256(<4 x double> %x0, <4 x i64> %x1, <4 x double> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_pd_256:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2pd %ymm2, %ymm1, %ymm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <4 x double> @llvm.x86.avx512.mask.vpermi2var.pd.256(<4 x double> %x0, <4 x i64> %x1, <4 x double> %x2, i8 -1)
ret <4 x double> %res
}
declare <4 x float> @llvm.x86.avx512.mask.vpermi2var.ps.128(<4 x float>, <4 x i32>, <4 x float>, i8)
define <4 x float>@test_int_x86_avx512_mask_vpermi2var_ps_128(<4 x float> %x0, <4 x i32> %x1, <4 x float> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_ps_128:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2ps %xmm2, %xmm1, %xmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <4 x float> @llvm.x86.avx512.mask.vpermi2var.ps.128(<4 x float> %x0, <4 x i32> %x1, <4 x float> %x2, i8 -1)
ret <4 x float> %res
}
declare <8 x float> @llvm.x86.avx512.mask.vpermi2var.ps.256(<8 x float>, <8 x i32>, <8 x float>, i8)
define <8 x float>@test_int_x86_avx512_mask_vpermi2var_ps_256(<8 x float> %x0, <8 x i32> %x1, <8 x float> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_ps_256:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2ps %ymm2, %ymm1, %ymm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <8 x float> @llvm.x86.avx512.mask.vpermi2var.ps.256(<8 x float> %x0, <8 x i32> %x1, <8 x float> %x2, i8 -1)
ret <8 x float> %res
}
define <8 x float>@test_int_x86_avx512_mask_vpermi2var_ps_256_load(<8 x float> %x0, <8 x i32> %x1, <8 x float>* %x2p) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_ps_256_load:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2ps (%rdi), %ymm1, %ymm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2 = load <8 x float>, <8 x float>* %x2p
%res = call <8 x float> @llvm.x86.avx512.mask.vpermi2var.ps.256(<8 x float> %x0, <8 x i32> %x1, <8 x float> %x2, i8 -1)
ret <8 x float> %res
}
define <8 x float>@test_int_x86_avx512_mask_vpermi2var_ps_256_broadcast(<8 x float> %x0, <8 x i32> %x1, float* %x2ptr) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_ps_256_broadcast:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2ps (%rdi){1to8}, %ymm1, %ymm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2s = load float, float* %x2ptr
%x2ins = insertelement <8 x float> undef, float %x2s, i32 0
%x2 = shufflevector <8 x float> %x2ins, <8 x float> undef, <8 x i32> zeroinitializer
%res = call <8 x float> @llvm.x86.avx512.mask.vpermi2var.ps.256(<8 x float> %x0, <8 x i32> %x1, <8 x float> %x2, i8 -1)
ret <8 x float> %res
}
declare <16 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
define <16 x i8>@test_int_x86_avx512_mask_vpermi2var_qi_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_qi_128:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2b %xmm2, %xmm1, %xmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <16 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 -1)
ret <16 x i8> %res
}
declare <32 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
define <32 x i8>@test_int_x86_avx512_mask_vpermi2var_qi_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_qi_256:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermt2b %ymm2, %ymm1, %ymm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <32 x i8> @llvm.x86.avx512.mask.vpermi2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
ret <32 x i8> %res
}
declare <16 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
define <16 x i8>@test_int_x86_avx512_mask_vpermt2var_qi_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_128:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2b %xmm2, %xmm1, %xmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <16 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 -1)
ret <16 x i8> %res
}
define <16 x i8>@test_int_x86_avx512_mask_vpermt2var_qi_128_load(<16 x i8> %x0, <16 x i8> %x1, <16 x i8>* %x2p) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_128_load:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2b (%rdi), %xmm1, %xmm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2 = load <16 x i8>, <16 x i8>* %x2p
%res = call <16 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 -1)
ret <16 x i8> %res
}
declare <32 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
define <32 x i8>@test_int_x86_avx512_mask_vpermt2var_qi_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_qi_256:
[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
; CHECK: ## %bb.0:
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2b %ymm2, %ymm1, %ymm0
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <32 x i8> @llvm.x86.avx512.mask.vpermt2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
ret <32 x i8> %res
}
declare <16 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
define <16 x i8>@test_int_x86_avx512_maskz_vpermt2var_qi_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_qi_128:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %edi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2b %xmm2, %xmm1, %xmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <16 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3)
ret <16 x i8> %res
}
define <16 x i8>@test_int_x86_avx512_maskz_vpermt2var_qi_128_load(<16 x i8> %x0, <16 x i8> %x1, <16 x i8>* %x2p, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_qi_128_load:
[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
; CHECK: ## %bb.0:
[AVX-512] Fix accidental uses of AH/BH/CH/DH after copies to/from mask registers We've had several bugs(PR32256, PR32241) recently that resulted from usages of AH/BH/CH/DH either before or after a copy to/from a mask register. This ultimately occurs because we create COPY_TO_REGCLASS with VK1 and GR8. Then in CopyToFromAsymmetricReg in X86InstrInfo we find a 32-bit super register for the GR8 to emit the KMOV with. But as these tests are demonstrating, its possible for the GR8 register to be a high register and we end up doing an accidental extra or insert from bits 15:8. I think the best way forward is to stop making copies directly between mask registers and GR8/GR16. Instead I think we should restrict to only copies between mask registers and GR32/GR64 and use EXTRACT_SUBREG/INSERT_SUBREG to handle the conversion from GR32 to GR16/8 or vice versa. Unfortunately, this complicates fastisel a bit more now to create the subreg extracts where we used to create GR8 copies. We can probably make a helper function to bring down the repitition. This does result in KMOVD being used for copies when BWI is available because we don't know the original mask register size. This caused a lot of deltas on tests because we have to split the checks for KMOVD vs KMOVW based on BWI. Differential Revision: https://reviews.llvm.org/D30968 llvm-svn: 298928
2017-03-29 00:35:29 +08:00
; CHECK-NEXT: kmovd %esi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2b (%rdi), %xmm1, %xmm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2 = load <16 x i8>, <16 x i8>* %x2p
%res = call <16 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3)
ret <16 x i8> %res
}
declare <32 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
define <32 x i8>@test_int_x86_avx512_maskz_vpermt2var_qi_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_qi_256:
[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
; CHECK: ## %bb.0:
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: kmovd %edi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2b %ymm2, %ymm1, %ymm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%res = call <32 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
ret <32 x i8> %res
}
define <32 x i8>@test_int_x86_avx512_maskz_vpermt2var_qi_256_load(<32 x i8> %x0, <32 x i8> %x1, <32 x i8>* %x2p, i32 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_qi_256_load:
[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
; CHECK: ## %bb.0:
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: kmovd %esi, %k1
[AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD). Summary: The index and one of the table operands can be swapped by changing the opcode to the other version. Neither of these operands are the one that can load from memory so this can't be used to increase memory folding opportunities. We need to handle the unmasked forms and the kz forms. Since the load operand isn't being commuted we can commute the load and broadcast instructions too. Reviewers: igorb, delena, Ayal, Farhana, RKSimon Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D25652 llvm-svn: 287621
2016-11-22 12:57:34 +08:00
; CHECK-NEXT: vpermi2b (%rdi), %ymm1, %ymm0 {%k1} {z}
[AVX-512] Add tests to show opportunities for commuting vpermi2/vpermt2 instructions. Commuting will be added in a future commit. llvm-svn: 284808
2016-10-21 13:55:40 +08:00
; CHECK-NEXT: retq
%x2 = load <32 x i8>, <32 x i8>* %x2p
%res = call <32 x i8> @llvm.x86.avx512.maskz.vpermt2var.qi.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
ret <32 x i8> %res
}