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/fma-intrinsics-x86.ll

1163 lines
61 KiB
LLVM
Raw Normal View History

[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX512VL
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; RUN: llc < %s -mtriple=x86_64-pc-windows -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
; VFMADD
define <4 x float> @test_x86_fma_vfmadd_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfmadd132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x99,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a0, i64 0
%2 = extractelement <4 x float> %a1, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = call float @llvm.fma.f32(float %1, float %2, float %3)
%5 = insertelement <4 x float> %a0, float %4, i64 0
ret <4 x float> %5
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <4 x float> @test_x86_fma_vfmadd_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_bac_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xa9,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = (xmm0 * xmm1) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_bac_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xa9,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = (xmm0 * xmm1) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_bac_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfmadd132ss (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x99,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a1, i64 0
%2 = extractelement <4 x float> %a0, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = call float @llvm.fma.f32(float %1, float %2, float %3)
%5 = insertelement <4 x float> %a1, float %4, i64 0
ret <4 x float> %5
[X86] Add a missing FMA3 scalar intrinsic pattern. This allows us to use 231 form to fold an insertelement on the add input to the fma. There is technically no software intrinsic that can use this until AVX512F, but it can be manually built up from other intrinsics. llvm-svn: 337223
2018-07-17 07:10:58 +08:00
}
define <4 x float> @test_x86_fma_vfmadd_ss_231(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ss_231:
; CHECK-FMA: # %bb.0:
; CHECK-FMA-NEXT: vfmadd231ss %xmm1, %xmm0, %xmm2 # encoding: [0xc4,0xe2,0x79,0xb9,0xd1]
; CHECK-FMA-NEXT: # xmm2 = (xmm0 * xmm1) + xmm2
; CHECK-FMA-NEXT: vmovaps %xmm2, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc2]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
;
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_ss_231:
; CHECK-AVX512VL: # %bb.0:
; CHECK-AVX512VL-NEXT: vfmadd231ss %xmm1, %xmm0, %xmm2 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xb9,0xd1]
; CHECK-AVX512VL-NEXT: # xmm2 = (xmm0 * xmm1) + xmm2
; CHECK-AVX512VL-NEXT: vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ss_231:
; CHECK-FMA-WIN: # %bb.0:
; CHECK-FMA-WIN-NEXT: vmovaps (%r8), %xmm0 # encoding: [0xc4,0xc1,0x78,0x28,0x00]
; CHECK-FMA-WIN-NEXT: vmovss (%rcx), %xmm1 # encoding: [0xc5,0xfa,0x10,0x09]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfmadd231ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xb9,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * mem) + xmm0
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
%1 = extractelement <4 x float> %a0, i64 0
%2 = extractelement <4 x float> %a1, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = call float @llvm.fma.f32(float %1, float %2, float %3)
%5 = insertelement <4 x float> %a2, float %4, i64 0
ret <4 x float> %5
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <2 x double> @test_x86_fma_vfmadd_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa9,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfmadd132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x99,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a0, i64 0
%2 = extractelement <2 x double> %a1, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = call double @llvm.fma.f64(double %1, double %2, double %3)
%5 = insertelement <2 x double> %a0, double %4, i64 0
ret <2 x double> %5
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <2 x double> @test_x86_fma_vfmadd_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_bac_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xa9,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = (xmm0 * xmm1) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_bac_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xa9,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = (xmm0 * xmm1) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_bac_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfmadd132sd (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x99,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a1, i64 0
%2 = extractelement <2 x double> %a0, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = call double @llvm.fma.f64(double %1, double %2, double %3)
%5 = insertelement <2 x double> %a1, double %4, i64 0
ret <2 x double> %5
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <4 x float> @test_x86_fma_vfmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ps:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_ps:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ps:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa8,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
ret <4 x float> %1
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <2 x double> @test_x86_fma_vfmadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_pd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_pd:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_pd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa8,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
ret <2 x double> %1
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <8 x float> @test_x86_fma_vfmadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) + ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) + ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa8,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
ret <8 x float> %1
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <4 x double> @test_x86_fma_vfmadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmadd_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) + ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmadd_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) + ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa8,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
ret <4 x double> %1
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
; VFMSUB
define <4 x float> @test_x86_fma_vfmsub_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xab,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xab,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfmsub132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9b,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a0, i64 0
%2 = extractelement <4 x float> %a1, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = fsub float -0.000000e+00, %3
%5 = call float @llvm.fma.f32(float %1, float %2, float %4)
%6 = insertelement <4 x float> %a0, float %5, i64 0
ret <4 x float> %6
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <4 x float> @test_x86_fma_vfmsub_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_bac_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xab,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = (xmm0 * xmm1) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_bac_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xab,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = (xmm0 * xmm1) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_bac_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfmsub132ss (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9b,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a1, i64 0
%2 = extractelement <4 x float> %a0, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = fsub float -0.000000e+00, %3
%5 = call float @llvm.fma.f32(float %1, float %2, float %4)
%6 = insertelement <4 x float> %a1, float %5, i64 0
ret <4 x float> %6
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <2 x double> @test_x86_fma_vfmsub_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xab,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xab,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfmsub132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9b,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a0, i64 0
%2 = extractelement <2 x double> %a1, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = fsub double -0.000000e+00, %3
%5 = call double @llvm.fma.f64(double %1, double %2, double %4)
%6 = insertelement <2 x double> %a0, double %5, i64 0
ret <2 x double> %6
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <2 x double> @test_x86_fma_vfmsub_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_bac_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xab,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = (xmm0 * xmm1) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_bac_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xab,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = (xmm0 * xmm1) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_bac_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfmsub132sd (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9b,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a1, i64 0
%2 = extractelement <2 x double> %a0, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = fsub double -0.000000e+00, %3
%5 = call double @llvm.fma.f64(double %1, double %2, double %4)
%6 = insertelement <2 x double> %a1, double %5, i64 0
ret <2 x double> %6
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <4 x float> @test_x86_fma_vfmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ps:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_ps:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ps:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xaa,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %1)
ret <4 x float> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <2 x double> @test_x86_fma_vfmsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_pd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_pd:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_pd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xaa,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
%2 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %1)
ret <2 x double> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <8 x float> @test_x86_fma_vfmsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) - ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) - ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xaa,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%2 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %1)
ret <8 x float> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <4 x double> @test_x86_fma_vfmsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsub_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) - ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsub_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xaa,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) - ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xaa,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%2 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %1)
ret <4 x double> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
; VFNMADD
define <4 x float> @test_x86_fma_vfnmadd_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xad,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xad,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfnmadd132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9d,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a0, i64 0
%2 = extractelement <4 x float> %a1, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = fsub float -0.000000e+00, %2
%5 = call float @llvm.fma.f32(float %1, float %4, float %3)
%6 = insertelement <4 x float> %a0, float %5, i64 0
ret <4 x float> %6
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <4 x float> @test_x86_fma_vfnmadd_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_bac_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xad,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = -(xmm0 * xmm1) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_bac_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xad,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = -(xmm0 * xmm1) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_bac_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfnmadd132ss (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9d,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a1, i64 0
%2 = extractelement <4 x float> %a0, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = fsub float -0.000000e+00, %2
%5 = call float @llvm.fma.f32(float %1, float %4, float %3)
%6 = insertelement <4 x float> %a1, float %5, i64 0
ret <4 x float> %6
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <2 x double> @test_x86_fma_vfnmadd_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xad,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xad,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfnmadd132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9d,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a0, i64 0
%2 = extractelement <2 x double> %a1, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = fsub double -0.000000e+00, %2
%5 = call double @llvm.fma.f64(double %1, double %4, double %3)
%6 = insertelement <2 x double> %a0, double %5, i64 0
ret <2 x double> %6
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <2 x double> @test_x86_fma_vfnmadd_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_bac_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xad,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = -(xmm0 * xmm1) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_bac_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xad,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = -(xmm0 * xmm1) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_bac_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfnmadd132sd (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9d,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) + xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a1, i64 0
%2 = extractelement <2 x double> %a0, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = fsub double -0.000000e+00, %2
%5 = call double @llvm.fma.f64(double %1, double %4, double %3)
%6 = insertelement <2 x double> %a1, double %5, i64 0
ret <2 x double> %6
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <4 x float> @test_x86_fma_vfnmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ps:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_ps:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ps:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xac,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm1 * xmm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a0
%2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %1, <4 x float> %a1, <4 x float> %a2)
ret <4 x float> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <2 x double> @test_x86_fma_vfnmadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_pd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_pd:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) + xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_pd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xac,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm1 * xmm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a0
%2 = call <2 x double> @llvm.fma.v2f64(<2 x double> %1, <2 x double> %a1, <2 x double> %a2)
ret <2 x double> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <8 x float> @test_x86_fma_vfnmadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = -(ymm1 * ymm0) + ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = -(ymm1 * ymm0) + ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xac,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = -(ymm1 * ymm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a0
%2 = call <8 x float> @llvm.fma.v8f32(<8 x float> %1, <8 x float> %a1, <8 x float> %a2)
ret <8 x float> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <4 x double> @test_x86_fma_vfnmadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = -(ymm1 * ymm0) + ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmadd_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = -(ymm1 * ymm0) + ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xac,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = -(ymm1 * ymm0) + mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a0
%2 = call <4 x double> @llvm.fma.v4f64(<4 x double> %1, <4 x double> %a1, <4 x double> %a2)
ret <4 x double> %2
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
; VFNMSUB
define <4 x float> @test_x86_fma_vfnmsub_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213ss %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xaf,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213ss %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xaf,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfnmsub132ss (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9f,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a0, i64 0
%2 = extractelement <4 x float> %a1, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = fsub float -0.000000e+00, %2
%5 = fsub float -0.000000e+00, %3
%6 = call float @llvm.fma.f32(float %1, float %4, float %5)
%7 = insertelement <4 x float> %a0, float %6, i64 0
ret <4 x float> %7
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <4 x float> @test_x86_fma_vfnmsub_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_bac_ss:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213ss %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0x79,0xaf,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = -(xmm0 * xmm1) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovaps %xmm1, %xmm0 # encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_bac_ss:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213ss %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xaf,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = -(xmm0 * xmm1) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_bac_ss:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovss (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7a,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero,zero,zero
; CHECK-FMA-WIN-NEXT: vfnmsub132ss (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0x9f,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <4 x float> %a1, i64 0
%2 = extractelement <4 x float> %a0, i64 0
%3 = extractelement <4 x float> %a2, i64 0
%4 = fsub float -0.000000e+00, %2
%5 = fsub float -0.000000e+00, %3
%6 = call float @llvm.fma.f32(float %1, float %4, float %5)
%7 = insertelement <4 x float> %a1, float %6, i64 0
ret <4 x float> %7
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <2 x double> @test_x86_fma_vfnmsub_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213sd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xaf,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213sd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xaf,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x01]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfnmsub132sd (%rdx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9f,0x02]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a0, i64 0
%2 = extractelement <2 x double> %a1, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = fsub double -0.000000e+00, %2
%5 = fsub double -0.000000e+00, %3
%6 = call double @llvm.fma.f64(double %1, double %4, double %5)
%7 = insertelement <2 x double> %a0, double %6, i64 0
ret <2 x double> %7
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
define <2 x double> @test_x86_fma_vfnmsub_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_bac_sd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213sd %xmm2, %xmm0, %xmm1 # encoding: [0xc4,0xe2,0xf9,0xaf,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm1 = -(xmm0 * xmm1) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vmovapd %xmm1, %xmm0 # encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_bac_sd:
[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-AVX512VL: # %bb.0:
[X86] Use EVEX encoded intrinsics for legacy FMA intrinsics when possible. llvm-svn: 317454
2017-11-06 13:48:26 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213sd %xmm2, %xmm0, %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xaf,0xca]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm1 = -(xmm0 * xmm1) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vmovapd %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc1]
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_bac_sd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
[X86] Remove X86 specific scalar FMA intrinsics and upgrade to tart independent FMA and extractelement/insertelement. llvm-svn: 336315
2018-07-05 14:52:55 +08:00
; CHECK-FMA-WIN-NEXT: vmovsd (%r8), %xmm1 # encoding: [0xc4,0xc1,0x7b,0x10,0x08]
; CHECK-FMA-WIN-NEXT: # xmm1 = mem[0],zero
; CHECK-FMA-WIN-NEXT: vfnmsub132sd (%rcx), %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0x9f,0x01]
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm0 * mem) - xmm1
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = extractelement <2 x double> %a1, i64 0
%2 = extractelement <2 x double> %a0, i64 0
%3 = extractelement <2 x double> %a2, i64 0
%4 = fsub double -0.000000e+00, %2
%5 = fsub double -0.000000e+00, %3
%6 = call double @llvm.fma.f64(double %1, double %4, double %5)
%7 = insertelement <2 x double> %a1, double %6, i64 0
ret <2 x double> %7
Created new X86 FMA3 opcodes (FMA*_Int) that are used now for lowering of scalar FMA intrinsics. Patch by Slava Klochkov The key difference between FMA* and FMA*_Int opcodes is that FMA*_Int opcodes are handled more conservatively. It is illegal to commute the 1st operand of FMA*_Int instructions as the upper bits of scalar FMA intrinsic result must be taken from the 1st operand, but such commute transformation would change those upper bits and invalidate the intrinsic's result. Reviewers: Quentin Colombet, Elena Demikhovsky Differential Revision: http://reviews.llvm.org/D13710 llvm-svn: 252060
2015-11-05 02:10:41 +08:00
}
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
define <4 x float> @test_x86_fma_vfnmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ps:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_ps:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ps:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xae,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm1 * xmm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a0
%2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %1, <4 x float> %a1, <4 x float> %2)
ret <4 x float> %3
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <2 x double> @test_x86_fma_vfnmsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_pd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_pd:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = -(xmm1 * xmm0) - xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_pd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xae,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = -(xmm1 * xmm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a0
%2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %1, <2 x double> %a1, <2 x double> %2)
ret <2 x double> %3
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <8 x float> @test_x86_fma_vfnmsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = -(ymm1 * ymm0) - ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = -(ymm1 * ymm0) - ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xae,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = -(ymm1 * ymm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a0
%2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %1, <8 x float> %a1, <8 x float> %2)
ret <8 x float> %3
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <4 x double> @test_x86_fma_vfnmsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfnmsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = -(ymm1 * ymm0) - ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfnmsub_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfnmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = -(ymm1 * ymm0) - ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfnmsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xae,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = -(ymm1 * ymm0) - mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a0
%2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %1, <4 x double> %a1, <4 x double> %2)
ret <4 x double> %3
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
; VFMADDSUB
define <4 x float> @test_x86_fma_vfmaddsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_ps:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) +/- xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_ps:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmaddsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) +/- xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_ps:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmaddsub213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa6,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) +/- mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
%2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %2)
%4 = shufflevector <4 x float> %3, <4 x float> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
ret <4 x float> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <2 x double> @test_x86_fma_vfmaddsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_pd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) +/- xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_pd:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmaddsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) +/- xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_pd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmaddsub213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa6,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) +/- mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
%2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %2)
%4 = shufflevector <2 x double> %3, <2 x double> %1, <2 x i32> <i32 0, i32 3>
ret <2 x double> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <8 x float> @test_x86_fma_vfmaddsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) +/- ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmaddsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) +/- ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmaddsub213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa6,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) +/- mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
%2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %2)
%4 = shufflevector <8 x float> %3, <8 x float> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
ret <8 x float> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <4 x double> @test_x86_fma_vfmaddsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) +/- ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmaddsub_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmaddsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) +/- ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmaddsub213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa6,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) +/- mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
%2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %2)
%4 = shufflevector <4 x double> %3, <4 x double> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
ret <4 x double> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
; VFMSUBADD
define <4 x float> @test_x86_fma_vfmsubadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_ps:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsubadd213ps %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x71,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) -/+ xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_ps:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsubadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) -/+ xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_ps:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %xmm1 # encoding: [0xc5,0xf8,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %xmm0 # encoding: [0xc5,0xf8,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsubadd213ps (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0x71,0xa7,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) -/+ mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
%2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %2)
%4 = shufflevector <4 x float> %1, <4 x float> %3, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
ret <4 x float> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <2 x double> @test_x86_fma_vfmsubadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_pd:
[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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsubadd213pd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0xf1,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # xmm0 = (xmm1 * xmm0) -/+ xmm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_pd:
[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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsubadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # xmm0 = (xmm1 * xmm0) -/+ xmm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_pd:
[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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %xmm1 # encoding: [0xc5,0xf9,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %xmm0 # encoding: [0xc5,0xf9,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsubadd213pd (%r8), %xmm1, %xmm0 # encoding: [0xc4,0xc2,0xf1,0xa7,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # xmm0 = (xmm1 * xmm0) -/+ mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
%2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %2)
%4 = shufflevector <2 x double> %1, <2 x double> %3, <2 x i32> <i32 0, i32 3>
ret <2 x double> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <8 x float> @test_x86_fma_vfmsubadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsubadd213ps %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x75,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) -/+ ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsubadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) -/+ ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovaps (%rcx), %ymm1 # encoding: [0xc5,0xfc,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovaps (%rdx), %ymm0 # encoding: [0xc5,0xfc,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsubadd213ps (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0x75,0xa7,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) -/+ mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
%2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
%3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %2)
%4 = shufflevector <8 x float> %1, <8 x float> %3, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
ret <8 x float> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
define <4 x double> @test_x86_fma_vfmsubadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_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-FMA: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: vfmsubadd213pd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0xf5,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-NEXT: # ymm0 = (ymm1 * ymm0) -/+ ymm2
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-NEXT: retq # encoding: [0xc3]
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
;
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-LABEL: test_x86_fma_vfmsubadd_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-AVX512VL: # %bb.0:
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: vfmsubadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa7,0xc2]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-AVX512VL-NEXT: # ymm0 = (ymm1 * ymm0) -/+ ymm2
[X86] Add avx512vl command line to fma-instrinsics-x86.ll Some of these demonstrate a missed EVEX to VEX compression because we aren't prefering EVEX instructions during isel. llvm-svn: 317452
2017-11-06 13:48:24 +08:00
; CHECK-AVX512VL-NEXT: retq # encoding: [0xc3]
;
[X86][FMA] Regenerate FMA tests llvm-svn: 312871
2017-09-10 03:25:59 +08:00
; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_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-FMA-WIN: # %bb.0:
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: vmovapd (%rcx), %ymm1 # encoding: [0xc5,0xfd,0x28,0x09]
; CHECK-FMA-WIN-NEXT: vmovapd (%rdx), %ymm0 # encoding: [0xc5,0xfd,0x28,0x02]
; CHECK-FMA-WIN-NEXT: vfmsubadd213pd (%r8), %ymm1, %ymm0 # encoding: [0xc4,0xc2,0xf5,0xa7,0x00]
[X86] Add comments to the end of FMA3 instructions to make the operation clear Summary: There are 3 different operand orders for FMA instructions so figuring out the exact operation being performed requires a lot of thought. This patch adds a comment to the end of the assembly line to print the exact operation. I think I've got all the instructions in here except the ones with builtin rounding. I didn't update all tests, but I assume we can get them as we regenerate tests in the future. Reviewers: spatel, v_klochkov, RKSimon Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D44345 llvm-svn: 327225
2018-03-11 05:30:46 +08:00
; CHECK-FMA-WIN-NEXT: # ymm0 = (ymm1 * ymm0) -/+ mem
[X86] Simplify command lines on the fma-instrinsics-x86.ll test and add -show-mc-encoding. Use feature names instead of CPU names. A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions. llvm-svn: 317451
2017-11-06 13:48:23 +08:00
; CHECK-FMA-WIN-NEXT: retq # encoding: [0xc3]
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
%1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
%2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %2)
%4 = shufflevector <4 x double> %1, <4 x double> %3, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
ret <4 x double> %4
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
}
[X86] Remove the last of the 'x86.fma.' intrinsics and autoupgrade them to 'llvm.fma'. Add upgrade tests for all. Still need to remove the AVX512 masked versions. llvm-svn: 336383
2018-07-06 02:43:58 +08:00
declare float @llvm.fma.f32(float, float, float)
declare double @llvm.fma.f64(double, double, double)
declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>)
declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>)
declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>)
declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>)
[X86] Cleanup fma tests a little bit. NFC. Reformat, isolate 213->231 xform, actually --check-prefix CHECK, and deduplicate the FMA intrinsic tests (FMA3 in AMD-land). llvm-svn: 240615
2015-06-25 08:40:25 +08:00
attributes #0 = { nounwind }