maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll

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

1181 lines
72 KiB
LLVM
Raw Normal View History

AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512f --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
declare <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i32)
declare <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i32)
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
define <16 x float> @test_x86_vfnmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_x86_vfnmadd_ps_z:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmadd213ps %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xac,0xc2]
; CHECK-NEXT: # zmm0 = -(zmm1 * zmm0) + zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %a1
%2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %a2)
ret <16 x float> %2
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
}
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
define <16 x float> @test_mask_vfnmadd_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_vfnmadd_ps:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfnmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmadd_ps:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %a1
%2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %a2)
%3 = bitcast i16 %mask to <16 x i1>
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %a0
ret <16 x float> %4
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
}
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
define <8 x double> @test_x86_vfnmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_x86_vfnmadd_pd_z:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmadd213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xac,0xc2]
; CHECK-NEXT: # zmm0 = -(zmm1 * zmm0) + zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %a2)
ret <8 x double> %2
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
}
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
define <8 x double> @test_mask_vfnmadd_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_vfnmadd_pd:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfnmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmadd_pd:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %a2)
%3 = bitcast i8 %mask to <8 x i1>
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %a0
ret <8 x double> %4
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
}
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
define <16 x float> @test_x86_vfnmsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_x86_vfnmsubps_z:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmsub213ps %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xae,0xc2]
; CHECK-NEXT: # zmm0 = -(zmm1 * zmm0) - zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %a1
%2 = fsub <16 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, 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 <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %2)
ret <16 x float> %3
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
}
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
define <16 x float> @test_mask_vfnmsub_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_vfnmsub_ps:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfnmsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmsub_ps:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %a1
%2 = fsub <16 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, 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 <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %2)
%4 = bitcast i16 %mask to <16 x i1>
%5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %a0
ret <16 x float> %5
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
}
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
define <8 x double> @test_x86_vfnmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_x86_vfnmsubpd_z:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmsub213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xae,0xc2]
; CHECK-NEXT: # zmm0 = -(zmm1 * zmm0) - zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
ret <8 x double> %3
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
}
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
define <8 x double> @test_mask_vfnmsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_vfnmsub_pd:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmsub_pd:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
%4 = bitcast i8 %mask to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
ret <8 x double> %5
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
}
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
define <16 x float> @test_x86_vfmaddsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_x86_vfmaddsubps_z:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmaddsub213ps %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xa6,0xc2]
; CHECK-NEXT: # zmm0 = (zmm1 * zmm0) +/- zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) #2
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
ret <16 x float> %res
}
[AVX512] Switch FMA intrinsics to the masking version This does the renaming and updates the lowering logic. Part of <rdar://problem/17688758> llvm-svn: 215664
2014-08-15 01:13:30 +08:00
[AVX512] Add test for FMA masking instrinsics llvm-svn: 215665
2014-08-15 01:13:33 +08:00
define <16 x float> @test_mask_fmaddsub_ps(<16 x float> %a, <16 x float> %b, <16 x float> %c, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_fmaddsub_ps:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfmaddsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_fmaddsub_ps:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmaddsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %a, <16 x float> %b, <16 x float> %c, i32 4)
%bc = bitcast i16 %mask to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a
ret <16 x float> %sel
[AVX512] Add test for FMA masking instrinsics llvm-svn: 215665
2014-08-15 01:13:33 +08:00
}
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
declare <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i32) nounwind readnone
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
define <8 x double> @test_x86_vfmaddsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_x86_vfmaddsubpd_z:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmaddsub213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xa6,0xc2]
; CHECK-NEXT: # zmm0 = (zmm1 * zmm0) +/- zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) #2
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
ret <8 x double> %res
}
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
declare <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) nounwind readnone
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
define <8 x double> @test_mask_vfmaddsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_vfmaddsub_pd:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfmaddsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfmaddsub_pd:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmaddsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) #2
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i8 %mask to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
ret <8 x double> %sel
Add AVX512 unmasked FMA intrinsics and support. llvm-svn: 194824
2013-11-16 01:01:14 +08:00
}
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
define <8 x double>@test_int_x86_avx512_mask_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfmaddsub_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmaddsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfmaddsub_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmaddsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i32 4)
%bc = bitcast i8 %x3 to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %x0
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
ret <8 x double> %sel
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <8 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmaddsub231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xb6,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmaddsub231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xb6,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %2)
%4 = shufflevector <8 x double> %3, <8 x double> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
%5 = bitcast i8 %x3 to <8 x i1>
%6 = select <8 x i1> %5, <8 x double> %4, <8 x double> %x2
ret <8 x double> %6
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <8 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmaddsub213pd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0xa6,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmaddsub213pd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0xa6,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %2)
%4 = shufflevector <8 x double> %3, <8 x double> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
%5 = bitcast i8 %x3 to <8 x i1>
%6 = select <8 x i1> %5, <8 x double> %4, <8 x double> zeroinitializer
ret <8 x double> %6
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask_vfmaddsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfmaddsub_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmaddsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfmaddsub_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmaddsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x96,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i32 4)
%bc = bitcast i16 %x3 to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %x0
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
ret <16 x float> %sel
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmaddsub231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xb6,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmaddsub231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xb6,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) +/- zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
%2 = fsub <16 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, 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>, %x2
%3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %2)
%4 = shufflevector <16 x float> %3, <16 x float> %1, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
%5 = bitcast i16 %x3 to <16 x i1>
%6 = select <16 x i1> %5, <16 x float> %4, <16 x float> %x2
ret <16 x float> %6
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmaddsub213ps %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0xa6,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmaddsub213ps %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0xa6,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) +/- zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
%2 = fsub <16 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, 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>, %x2
%3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %2)
%4 = shufflevector <16 x float> %3, <16 x float> %1, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
%5 = bitcast i16 %x3 to <16 x i1>
%6 = select <16 x i1> %5, <16 x float> %4, <16 x float> zeroinitializer
ret <16 x float> %6
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <8 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmsubadd231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xb7,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) -/+ zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmsubadd231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xb7,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) -/+ zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %2)
%4 = shufflevector <8 x double> %1, <8 x double> %3, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
%5 = bitcast i8 %x3 to <8 x i1>
%6 = select <8 x i1> %5, <8 x double> %4, <8 x double> %x2
ret <8 x double> %6
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmsubadd231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xb7,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) -/+ zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmsubadd231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xb7,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) -/+ zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
%2 = fsub <16 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, 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>, %x2
%3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %2)
%4 = shufflevector <16 x float> %1, <16 x float> %3, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
%5 = bitcast i16 %x3 to <16 x i1>
%6 = select <16 x i1> %5, <16 x float> %4, <16 x float> %x2
ret <16 x float> %6
AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets by Asaf Badouh http://reviews.llvm.org/D6456 llvm-svn: 224764
2014-12-23 18:30:39 +08:00
}
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rne(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_ps_rrb_rne:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x19,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ps_rrb_rne:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132ps {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x19,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 8) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i16 %mask to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
ret <16 x float> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtn(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_ps_rrb_rtn:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfmadd132ps {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x39,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ps_rrb_rtn:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132ps {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x39,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 9) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i16 %mask to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
ret <16 x float> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtp(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_ps_rrb_rtp:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfmadd132ps {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x59,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ps_rrb_rtp:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132ps {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x59,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 10) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i16 %mask to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
ret <16 x float> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rtz(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_ps_rrb_rtz:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfmadd132ps {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x79,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ps_rrb_rtz:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132ps {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x79,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 11) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i16 %mask to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
ret <16 x float> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrb_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_ps_rrb_current:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vfmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ps_rrb_current:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) nounwind
%bc = bitcast i16 %mask to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %a0
ret <16 x float> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rne(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rne:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213ps {rn-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x18,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 8) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtn(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtn:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213ps {rd-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x38,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 9) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtp(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtp:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213ps {ru-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x58,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 10) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_rtz(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_rtz:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213ps {rz-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x78,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 11) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <16 x float> %res
}
define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_ps_rrbz_current:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xa8,0xc2]
; CHECK-NEXT: # zmm0 = (zmm1 * zmm0) + zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <16 x float> %res
}
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
define <8 x double>@test_int_x86_avx512_mask3_vfmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmsub231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xba,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmsub231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xba,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
%2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %1)
%3 = bitcast i8 %x3 to <8 x i1>
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %x2
ret <8 x double> %4
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask3_vfmsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmsub231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xba,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmsub231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xba,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %x2
%2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %1)
%3 = bitcast i16 %x3 to <16 x i1>
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %x2
ret <16 x float> %4
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_pd_rrb_rne:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x19,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_pd_rrb_rne:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x19,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 8) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i8 %mask to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
ret <8 x double> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_pd_rrb_rtn:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfmadd132pd {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x39,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_pd_rrb_rtn:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132pd {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x39,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 9) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i8 %mask to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
ret <8 x double> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_pd_rrb_rtp:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfmadd132pd {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x59,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_pd_rrb_rtp:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132pd {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x59,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 10) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i8 %mask to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
ret <8 x double> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_pd_rrb_rtz:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfmadd132pd {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x79,0x98,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_pd_rrb_rtz:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132pd {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x79,0x98,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 11) nounwind
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%bc = bitcast i8 %mask to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
ret <8 x double> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrb_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfmadd512_pd_rrb_current:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_pd_rrb_current:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
%bc = bitcast i8 %mask to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
ret <8 x double> %sel
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rne:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x18,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 8) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtn:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213pd {rd-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x38,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 9) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtp:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213pd {ru-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x58,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 10) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_rtz:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213pd {rz-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x78,0xa8,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 11) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <8 x double> %res
}
define <8 x double> @test_mask_round_vfmadd512_pd_rrbz_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfmadd512_pd_rrbz_current:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xa8,0xc2]
; CHECK-NEXT: # zmm0 = (zmm1 * zmm0) + zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
ret <8 x double> %res
}
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
define <8 x double>@test_int_x86_avx512_mask_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfmadd_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfmadd_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i32 4)
%bc = bitcast i8 %x3 to <8 x i1>
%sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %x0
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
ret <8 x double> %sel
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <8 x double>@test_int_x86_avx512_mask3_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmadd231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xb8,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) + zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmadd231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xb8,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) + zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%3 = select <8 x i1> %2, <8 x double> %1, <8 x double> %x2
ret <8 x double> %3
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <8 x double>@test_int_x86_avx512_maskz_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0xa8,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0xa8,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%3 = select <8 x i1> %2, <8 x double> %1, <8 x double> zeroinitializer
ret <8 x double> %3
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfmadd_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfmadd_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x98,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = (zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Lowering FMA intrinsics to native IR (LLVM part) Support for Clang lowering of fused intrinsics. This patch: 1. Removes bindings to clang fma intrinsics. 2. Introduces new LLVM unmasked intrinsics with rounding mode: int_x86_avx512_vfmadd_pd_512 int_x86_avx512_vfmadd_ps_512 int_x86_avx512_vfmaddsub_pd_512 int_x86_avx512_vfmaddsub_ps_512 supported with a new intrinsic type (INTR_TYPE_3OP_RM). 3. Introduces new x86 fmaddsub/fmsubadd folding. 4. Introduces new tests for code emitted by sequentions introduced in Clang part. Patch by tkrupa Reviewers: craig.topper, sroland, spatel, RKSimon Reviewed By: craig.topper, RKSimon Differential Revision: https://reviews.llvm.org/D47443 llvm-svn: 333554
2018-05-30 23:25:16 +08:00
%res = call <16 x float> @llvm.x86.avx512.vfmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i32 4)
%bc = bitcast i16 %x3 to <16 x i1>
%sel = select <16 x i1> %bc, <16 x float> %res, <16 x float> %x0
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
ret <16 x float> %sel
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask3_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmadd231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xb8,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) + zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmadd231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xb8,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = (zmm0 * zmm1) + zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
%2 = bitcast i16 %x3 to <16 x i1>
%3 = select <16 x i1> %2, <16 x float> %1, <16 x float> %x2
ret <16 x float> %3
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
define <16 x float> @test_int_x86_avx512_maskz_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0xa8,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0xa8,0xc2]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} {z} = (zmm1 * zmm0) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
%2 = bitcast i16 %x3 to <16 x i1>
%3 = select <16 x i1> %2, <16 x float> %1, <16 x float> zeroinitializer
ret <16 x float> %3
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfnmsub512_pd_rrb_rne:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x19,0x9e,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfnmsub512_pd_rrb_rne:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x19,0x9e,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 8)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%4 = bitcast i8 %mask to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
ret <8 x double> %5
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtn:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfnmsub132pd {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x39,0x9e,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtn:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmsub132pd {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x39,0x9e,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 9)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%4 = bitcast i8 %mask to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
ret <8 x double> %5
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtp:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfnmsub132pd {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x59,0x9e,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtp:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmsub132pd {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x59,0x9e,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 10)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%4 = bitcast i8 %mask to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
ret <8 x double> %5
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtz:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfnmsub132pd {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x79,0x9e,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfnmsub512_pd_rrb_rtz:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmsub132pd {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x79,0x9e,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 11)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%4 = bitcast i8 %mask to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
ret <8 x double> %5
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_mask_round_vfnmsub512_pd_rrb_current:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
; X86-NEXT: vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfnmsub512_pd_rrb_current:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
; X64-NEXT: vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
%4 = bitcast i8 %mask to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
ret <8 x double> %5
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rne:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x18,0xae,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 8)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
ret <8 x double> %3
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtn:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmsub213pd {rd-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x38,0xae,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 9)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
ret <8 x double> %3
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtp:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmsub213pd {ru-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x58,0xae,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 10)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
ret <8 x double> %3
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_rtz:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmsub213pd {rz-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x78,0xae,0xc2]
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
[X86] Make lowering of intrinsics with rounding mode stricter so that only valid rounding modes are lowered. Update tests accordingly Many of our tests were not using valid rounding mode immediates. Clang verifies this in the frontend when it creates the intrinsics from builtins, but the backend would still lower invalid immediates. With this change we will now leave them as intrinsics if the immediate is invalid. This will cause an isel selection failure. llvm-svn: 355789
2019-03-11 01:20:45 +08:00
%3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 11)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
ret <8 x double> %3
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
AVX512 : Change v8i1 bitconvert GR8 pattern, remove unnecessary movzbl instruction. code example , previous implementation. movzbl %dil, %eax kmovw %eax, %k0 new code kmovw %edi, %k0 Differential Revision: http://reviews.llvm.org/D16287 llvm-svn: 258045
2016-01-18 20:02:45 +08:00
; CHECK-LABEL: test_mask_round_vfnmsub512_pd_rrbz_current:
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; CHECK: # %bb.0:
; CHECK-NEXT: vfnmsub213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xae,0xc2]
; CHECK-NEXT: # zmm0 = -(zmm1 * zmm0) - zmm2
; CHECK-NEXT: ret{{[l|q]}} # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
ret <8 x double> %3
AVX-512: Added FMA intrinsics with rounding mode By Asaf Badouh and Elena Demikhovsky Added special nodes for rounding: FMADD_RND, FMSUB_RND.. It will prevent merge between nodes with rounding and other standard nodes. llvm-svn: 227303
2015-01-28 18:21:27 +08:00
}
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
define <8 x double>@test_int_x86_avx512_mask_vfnmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfnmsub_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmsub_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x1
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %1, <8 x double> %2)
%4 = bitcast i8 %x3 to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %x0
ret <8 x double> %5
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <8 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfnmsub231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xbe,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = -(zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfnmsub231pd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0xbe,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = -(zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x0
%2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
%3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %1, <8 x double> %x1, <8 x double> %2)
%4 = bitcast i8 %x3 to <8 x i1>
%5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %x2
ret <8 x double> %5
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask_vfnmsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfnmsub_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfnmsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmsub_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfnmsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9e,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) - zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %x1
%2 = fsub <16 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, 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>, %x2
%3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %1, <16 x float> %2)
%4 = bitcast i16 %x3 to <16 x i1>
%5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %x0
ret <16 x float> %5
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfnmsub231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xbe,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm2 {%k1} = -(zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfnmsub231ps %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0xbe,0xd1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm2 {%k1} = -(zmm0 * zmm1) - zmm2
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %x0
%2 = fsub <16 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, 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>, %x2
%3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %1, <16 x float> %x1, <16 x float> %2)
%4 = bitcast i16 %x3 to <16 x i1>
%5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %x2
ret <16 x float> %5
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <8 x double>@test_int_x86_avx512_mask_vfnmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfnmadd_pd_512:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfnmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmadd_pd_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfnmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x1
%2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %1, <8 x double> %x2)
%3 = bitcast i8 %x3 to <8 x i1>
%4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %x0
ret <8 x double> %4
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
define <16 x float>@test_int_x86_avx512_mask_vfnmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-LABEL: test_int_x86_avx512_mask_vfnmadd_ps_512:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X86-NEXT: vfnmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X86-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmadd_ps_512:
; X64: # %bb.0:
; X64-NEXT: kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
[X86] Cleanup some of the avx512 masked fma tests to prepare for removing and autoupgrading. -Split cases that call 2 intrinsics in the same case. -Remove testing mask3 and maskz intrinsics with an all ones mask. These won't be interesting after the upgrade. -Restore test cases for some intrinsics that are marked for deletion, but haven't been deleted yet. llvm-svn: 336408
2018-07-06 11:42:06 +08:00
; X64-NEXT: vfnmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9c,0xc1]
[X86] Print k-mask in FMA3 comments.
2020-04-13 04:16:53 +08:00
; X64-NEXT: # zmm0 {%k1} = -(zmm0 * zmm1) + zmm2
[X86][AVX512] Cleanup intrinsics tests Ensure we test on 32-bit and 64-bit targets, and strip -mcpu usage. Part of ongoing work to ensure we test all intrinsic style tests on 32 and 64 bit targets where possible. llvm-svn: 333843
2018-06-03 22:56:04 +08:00
; X64-NEXT: retq # encoding: [0xc3]
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
%1 = fsub <16 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, 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>, %x1
%2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %1, <16 x float> %x2)
%3 = bitcast i16 %x3 to <16 x i1>
%4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %x0
ret <16 x float> %4
AVX-512: Implemented missing encoding and intrinsics for FMA instructions Added tests for DAG lowering ,encoding and intrinsics Differential Revision: http://reviews.llvm.org/D10796 llvm-svn: 240926
2015-06-29 17:10:00 +08:00
}
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
[X86] In combineFMA, make sure we bitcast the result of isFNEG back the expected type before creating the new FMA node. Previously, we were creating malformed SDNodes, but nothing noticed because the type constraints prevented isel from noticing. llvm-svn: 336566
2018-07-10 01:43:24 +08:00
; This test case used to crash due to combineFMA not bitcasting results of isFNEG.
define <4 x float> @foo() {
; X86-LABEL: foo:
; X86: # %bb.0: # %entry
; X86-NEXT: vmovss (%eax), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x00]
; X86-NEXT: # xmm0 = mem[0],zero,zero,zero
[X86] Add the rounding control operand to the printing for some scalar FMA instructions. llvm-svn: 358844
2019-04-21 15:12:56 +08:00
; X86-NEXT: vfmsub213ss {rd-sae}, %xmm0, %xmm0, %xmm0 # encoding: [0x62,0xf2,0x7d,0x38,0xab,0xc0]
[X86] In combineFMA, make sure we bitcast the result of isFNEG back the expected type before creating the new FMA node. Previously, we were creating malformed SDNodes, but nothing noticed because the type constraints prevented isel from noticing. llvm-svn: 336566
2018-07-10 01:43:24 +08:00
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: foo:
; X64: # %bb.0: # %entry
; X64-NEXT: vmovss (%rax), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x00]
; X64-NEXT: # xmm0 = mem[0],zero,zero,zero
[X86] Add the rounding control operand to the printing for some scalar FMA instructions. llvm-svn: 358844
2019-04-21 15:12:56 +08:00
; X64-NEXT: vfmsub213ss {rd-sae}, %xmm0, %xmm0, %xmm0 # encoding: [0x62,0xf2,0x7d,0x38,0xab,0xc0]
[X86] In combineFMA, make sure we bitcast the result of isFNEG back the expected type before creating the new FMA node. Previously, we were creating malformed SDNodes, but nothing noticed because the type constraints prevented isel from noticing. llvm-svn: 336566
2018-07-10 01:43:24 +08:00
; X64-NEXT: retq # encoding: [0xc3]
entry:
%0 = load <4 x float>, <4 x float>* undef, align 16
%sub = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %0
%1 = extractelement <4 x float> %sub, i64 0
%2 = call float @llvm.x86.avx512.vfmadd.f32(float undef, float undef, float %1, i32 9)
%3 = select i1 extractelement (<8 x i1> bitcast (<1 x i8> <i8 1> to <8 x i1>), i64 0), float %2, float undef
%4 = insertelement <4 x float> undef, float %3, i64 0
ret <4 x float> %4
}
; Function Attrs: nounwind readnone
declare float @llvm.x86.avx512.vfmadd.f32(float, float, float, i32)
[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode. This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
2018-07-06 11:42:09 +08:00
declare <16 x float> @llvm.fma.v16f32(<16 x float>, <16 x float>, <16 x float>)
declare <8 x double> @llvm.fma.v8f64(<8 x double>, <8 x double>, <8 x double>)