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/avx512vbmi2vl-intrinsics.ll

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

1466 lines
82 KiB
LLVM
Raw Normal View History

[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +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=+avx512vbmi2,+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi2,+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define <8 x i16> @test_mask_expand_load_w_128(i8* %addr, <8 x i16> %data, i8 %mask) {
; X86-LABEL: test_mask_expand_load_w_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
; X86-NEXT: vpexpandw (%eax), %xmm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_load_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandw (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = bitcast i8 %mask to <8 x i1>
%3 = call <8 x i16> @llvm.masked.expandload.v8i16(i16* %1, <8 x i1> %2, <8 x i16> %data)
ret <8 x i16> %3
}
define <8 x i16> @test_maskz_expand_load_w_128(i8* %addr, i8 %mask) {
; X86-LABEL: test_maskz_expand_load_w_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
; X86-NEXT: vpexpandw (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_load_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandw (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = bitcast i8 %mask to <8 x i1>
%3 = call <8 x i16> @llvm.masked.expandload.v8i16(i16* %1, <8 x i1> %2, <8 x i16> zeroinitializer)
ret <8 x i16> %3
}
define <8 x i16> @test_expand_load_w_128(i8* %addr, <8 x i16> %data) {
; X86-LABEL: test_expand_load_w_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X86-NEXT: vpexpandw (%eax), %xmm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_expand_load_w_128:
; X64: # %bb.0:
; X64-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X64-NEXT: vpexpandw (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = call <8 x i16> @llvm.masked.expandload.v8i16(i16* %1, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <8 x i16> %data)
ret <8 x i16> %2
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <8 x i16> @test_expand_w_128(<8 x i16> %data) {
; CHECK-LABEL: test_expand_w_128:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <8 x i16> @llvm.x86.avx512.mask.expand.v8i16(<8 x i16> %data, <8 x i16> undef, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <8 x i16> %1
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <8 x i16> @test_mask_expand_w_128(<8 x i16> %data, <8 x i16> %passthru, 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_expand_w_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpexpandw %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x62,0xc8]
; X86-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandw %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x62,0xc8]
; X64-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i8 %mask to <8 x i1>
%2 = call <8 x i16> @llvm.x86.avx512.mask.expand.v8i16(<8 x i16> %data, <8 x i16> %passthru, <8 x i1> %1)
ret <8 x i16> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <8 x i16> @test_maskz_expand_w_128(<8 x i16> %data, 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_maskz_expand_w_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpexpandw %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x62,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandw %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x62,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i8 %mask to <8 x i1>
%2 = call <8 x i16> @llvm.x86.avx512.mask.expand.v8i16(<8 x i16> %data, <8 x i16> zeroinitializer, <8 x i1> %1)
ret <8 x i16> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define <16 x i8> @test_mask_expand_load_b_128(i8* %addr, <16 x i8> %data, i16 %mask) {
; X86-LABEL: test_mask_expand_load_b_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpexpandb (%eax), %xmm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_load_b_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandb (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i8> @llvm.masked.expandload.v16i8(i8* %addr, <16 x i1> %1, <16 x i8> %data)
ret <16 x i8> %2
}
define <16 x i8> @test_maskz_expand_load_b_128(i8* %addr, i16 %mask) {
; X86-LABEL: test_maskz_expand_load_b_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpexpandb (%eax), %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0x89,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_load_b_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandb (%rdi), %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0x89,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i8> @llvm.masked.expandload.v16i8(i8* %addr, <16 x i1> %1, <16 x i8> zeroinitializer)
ret <16 x i8> %2
}
define <16 x i8> @test_expand_load_b_128(i8* %addr, <16 x i8> %data) {
; X86-LABEL: test_expand_load_b_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X86-NEXT: vpexpandb (%eax), %xmm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_expand_load_b_128:
; X64: # %bb.0:
; X64-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X64-NEXT: vpexpandb (%rdi), %xmm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = call <16 x i8> @llvm.masked.expandload.v16i8(i8* %addr, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i8> %data)
ret <16 x i8> %1
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i8> @test_expand_b_128(<16 x i8> %data) {
; CHECK-LABEL: test_expand_b_128:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <16 x i8> @llvm.x86.avx512.mask.expand.v16i8(<16 x i8> %data, <16 x i8> undef, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <16 x i8> %1
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <16 x i8> @test_mask_expand_b_128(<16 x i8> %data, <16 x i8> %passthru, 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_expand_b_128:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpexpandb %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x62,0xc8]
; X86-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_b_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandb %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x62,0xc8]
; X64-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i8> @llvm.x86.avx512.mask.expand.v16i8(<16 x i8> %data, <16 x i8> %passthru, <16 x i1> %1)
ret <16 x i8> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <16 x i8> @test_maskz_expand_b_128(<16 x i8> %data, 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_maskz_expand_b_128:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpexpandb %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0x89,0x62,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_b_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandb %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0x89,0x62,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i8> @llvm.x86.avx512.mask.expand.v16i8(<16 x i8> %data, <16 x i8> zeroinitializer, <16 x i1> %1)
ret <16 x i8> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define void @test_mask_compress_store_w_128(i8* %addr, <8 x i16> %data, i8 %mask) {
; X86-LABEL: test_mask_compress_store_w_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
; X86-NEXT: vpcompressw %xmm0, (%eax) {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x63,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_store_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpcompressw %xmm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x63,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = bitcast i8 %mask to <8 x i1>
call void @llvm.masked.compressstore.v8i16(<8 x i16> %data, i16* %1, <8 x i1> %2)
ret void
}
define void @test_compress_store_w_128(i8* %addr, <8 x i16> %data) {
; X86-LABEL: test_compress_store_w_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X86-NEXT: vpcompressw %xmm0, (%eax) {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x63,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_compress_store_w_128:
; X64: # %bb.0:
; X64-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X64-NEXT: vpcompressw %xmm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x63,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
call void @llvm.masked.compressstore.v8i16(<8 x i16> %data, i16* %1, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret void
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <8 x i16> @test_mask_compress_w_128(<8 x i16> %data, <8 x i16> %passthru, 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_compress_w_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpcompressw %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x63,0xc1]
; X86-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressw %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x63,0xc1]
; X64-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i8 %mask to <8 x i1>
%2 = call <8 x i16> @llvm.x86.avx512.mask.compress.v8i16(<8 x i16> %data, <8 x i16> %passthru, <8 x i1> %1)
ret <8 x i16> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <8 x i16> @test_maskz_compress_w_128(<8 x i16> %data, 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_maskz_compress_w_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpcompressw %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x63,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_compress_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressw %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x63,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i8 %mask to <8 x i1>
%2 = call <8 x i16> @llvm.x86.avx512.mask.compress.v8i16(<8 x i16> %data, <8 x i16> zeroinitializer, <8 x i1> %1)
ret <8 x i16> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <8 x i16> @test_compress_w_128(<8 x i16> %data) {
; CHECK-LABEL: test_compress_w_128:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <8 x i16> @llvm.x86.avx512.mask.compress.v8i16(<8 x i16> %data, <8 x i16> undef, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <8 x i16> %1
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define void @test_mask_compress_store_b_128(i8* %addr, <16 x i8> %data, i16 %mask) {
; X86-LABEL: test_mask_compress_store_b_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpcompressb %xmm0, (%eax) {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x63,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_store_b_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpcompressb %xmm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x63,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i16 %mask to <16 x i1>
call void @llvm.masked.compressstore.v16i8(<16 x i8> %data, i8* %addr, <16 x i1> %1)
ret void
}
define void @test_compress_store_b_128(i8* %addr, <16 x i8> %data) {
; X86-LABEL: test_compress_store_b_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X86-NEXT: vpcompressb %xmm0, (%eax) {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x63,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_compress_store_b_128:
; X64: # %bb.0:
; X64-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X64-NEXT: vpcompressb %xmm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x63,0x07]
; X64-NEXT: retq # encoding: [0xc3]
call void @llvm.masked.compressstore.v16i8(<16 x i8> %data, i8* %addr, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret void
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i8> @test_mask_compress_b_128(<16 x i8> %data, <16 x i8> %passthru, 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_compress_b_128:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpcompressb %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x63,0xc1]
; X86-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_b_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressb %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0x63,0xc1]
; X64-NEXT: vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i8> @llvm.x86.avx512.mask.compress.v16i8(<16 x i8> %data, <16 x i8> %passthru, <16 x i1> %1)
ret <16 x i8> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i8> @test_maskz_compress_b_128(<16 x i8> %data, 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_maskz_compress_b_128:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpcompressb %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0x89,0x63,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_compress_b_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressb %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0x89,0x63,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i8> @llvm.x86.avx512.mask.compress.v16i8(<16 x i8> %data, <16 x i8> zeroinitializer, <16 x i1> %1)
ret <16 x i8> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i8> @test_compress_b_128(<16 x i8> %data) {
; CHECK-LABEL: test_compress_b_128:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <16 x i8> @llvm.x86.avx512.mask.compress.v16i8(<16 x i8> %data, <16 x i8> undef, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <16 x i8> %1
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define <16 x i16> @test_mask_expand_load_w_256(i8* %addr, <16 x i16> %data, i16 %mask) {
; X86-LABEL: test_mask_expand_load_w_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpexpandw (%eax), %ymm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_load_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandw (%rdi), %ymm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = bitcast i16 %mask to <16 x i1>
%3 = call <16 x i16> @llvm.masked.expandload.v16i16(i16* %1, <16 x i1> %2, <16 x i16> %data)
ret <16 x i16> %3
}
define <16 x i16> @test_maskz_expand_load_w_256(i8* %addr, i16 %mask) {
; X86-LABEL: test_maskz_expand_load_w_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpexpandw (%eax), %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xa9,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_load_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandw (%rdi), %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xa9,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = bitcast i16 %mask to <16 x i1>
%3 = call <16 x i16> @llvm.masked.expandload.v16i16(i16* %1, <16 x i1> %2, <16 x i16> zeroinitializer)
ret <16 x i16> %3
}
define <16 x i16> @test_expand_load_w_256(i8* %addr, <16 x i16> %data) {
; X86-LABEL: test_expand_load_w_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X86-NEXT: vpexpandw (%eax), %ymm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_expand_load_w_256:
; X64: # %bb.0:
; X64-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X64-NEXT: vpexpandw (%rdi), %ymm0 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = call <16 x i16> @llvm.masked.expandload.v16i16(i16* %1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i16> %data)
ret <16 x i16> %2
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i16> @test_expand_w_256(<16 x i16> %data) {
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
; CHECK-LABEL: test_expand_w_256:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <16 x i16> @llvm.x86.avx512.mask.expand.v16i16(<16 x i16> %data, <16 x i16> undef, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <16 x i16> %1
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <16 x i16> @test_mask_expand_w_256(<16 x i16> %data, <16 x i16> %passthru, 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_expand_w_256:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpexpandw %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x62,0xc8]
; X86-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandw %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x62,0xc8]
; X64-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i16> @llvm.x86.avx512.mask.expand.v16i16(<16 x i16> %data, <16 x i16> %passthru, <16 x i1> %1)
ret <16 x i16> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i16> @test_maskz_expand_w_256(<16 x i16> %data, 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_maskz_expand_w_256:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpexpandw %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xa9,0x62,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandw %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xa9,0x62,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i16> @llvm.x86.avx512.mask.expand.v16i16(<16 x i16> %data, <16 x i16> zeroinitializer, <16 x i1> %1)
ret <16 x i16> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define <32 x i8> @test_mask_expand_load_b_256(i8* %addr, <32 x i8> %data, i32 %mask) {
; X86-LABEL: test_mask_expand_load_b_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpexpandb (%eax), %ymm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_load_b_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandb (%rdi), %ymm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i32 %mask to <32 x i1>
%2 = call <32 x i8> @llvm.masked.expandload.v32i8(i8* %addr, <32 x i1> %1, <32 x i8> %data)
ret <32 x i8> %2
}
define <32 x i8> @test_maskz_expand_load_b_256(i8* %addr, i32 %mask) {
; X86-LABEL: test_maskz_expand_load_b_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpexpandb (%eax), %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xa9,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_load_b_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpexpandb (%rdi), %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xa9,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i32 %mask to <32 x i1>
%2 = call <32 x i8> @llvm.masked.expandload.v32i8(i8* %addr, <32 x i1> %1, <32 x i8> zeroinitializer)
ret <32 x i8> %2
}
define <32 x i8> @test_expand_load_b_256(i8* %addr, <32 x i8> %data) {
; X86-LABEL: test_expand_load_b_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnord %k0, %k0, %k1 # encoding: [0xc4,0xe1,0xfd,0x46,0xc8]
; X86-NEXT: vpexpandb (%eax), %ymm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x62,0x00]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_expand_load_b_256:
; X64: # %bb.0:
; X64-NEXT: kxnord %k0, %k0, %k1 # encoding: [0xc4,0xe1,0xfd,0x46,0xc8]
; X64-NEXT: vpexpandb (%rdi), %ymm0 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x62,0x07]
; X64-NEXT: retq # encoding: [0xc3]
%1 = call <32 x i8> @llvm.masked.expandload.v32i8(i8* %addr, <32 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <32 x i8> %data)
ret <32 x i8> %1
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <32 x i8> @test_expand_b_256(<32 x i8> %data) {
; CHECK-LABEL: test_expand_b_256:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <32 x i8> @llvm.x86.avx512.mask.expand.v32i8(<32 x i8> %data, <32 x i8> undef, <32 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <32 x i8> %1
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <32 x i8> @test_mask_expand_b_256(<32 x i8> %data, <32 x i8> %passthru, i32 %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_expand_b_256:
; X86: # %bb.0:
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpexpandb %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x62,0xc8]
; X86-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_expand_b_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandb %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x62,0xc8]
; X64-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i32 %mask to <32 x i1>
%2 = call <32 x i8> @llvm.x86.avx512.mask.expand.v32i8(<32 x i8> %data, <32 x i8> %passthru, <32 x i1> %1)
ret <32 x i8> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <32 x i8> @test_maskz_expand_b_256(<32 x i8> %data, i32 %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_maskz_expand_b_256:
; X86: # %bb.0:
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpexpandb %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xa9,0x62,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_expand_b_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpexpandb %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xa9,0x62,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i32 %mask to <32 x i1>
%2 = call <32 x i8> @llvm.x86.avx512.mask.expand.v32i8(<32 x i8> %data, <32 x i8> zeroinitializer, <32 x i1> %1)
ret <32 x i8> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define void @test_mask_compress_store_w_256(i8* %addr, <16 x i16> %data, i16 %mask) {
; X86-LABEL: test_mask_compress_store_w_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpcompressw %ymm0, (%eax) {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x63,0x00]
; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_store_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpcompressw %ymm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x63,0x07]
; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
%2 = bitcast i16 %mask to <16 x i1>
call void @llvm.masked.compressstore.v16i16(<16 x i16> %data, i16* %1, <16 x i1> %2)
ret void
}
define void @test_compress_store_w_256(i8* %addr, <16 x i16> %data) {
; X86-LABEL: test_compress_store_w_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X86-NEXT: vpcompressw %ymm0, (%eax) {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x63,0x00]
; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_compress_store_w_256:
; X64: # %bb.0:
; X64-NEXT: kxnorw %k0, %k0, %k1 # encoding: [0xc5,0xfc,0x46,0xc8]
; X64-NEXT: vpcompressw %ymm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x63,0x07]
; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i8* %addr to i16*
call void @llvm.masked.compressstore.v16i16(<16 x i16> %data, i16* %1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret void
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i16> @test_mask_compress_w_256(<16 x i16> %data, <16 x i16> %passthru, 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_compress_w_256:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpcompressw %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x63,0xc1]
; X86-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressw %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0x63,0xc1]
; X64-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i16> @llvm.x86.avx512.mask.compress.v16i16(<16 x i16> %data, <16 x i16> %passthru, <16 x i1> %1)
ret <16 x i16> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i16> @test_maskz_compress_w_256(<16 x i16> %data, 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_maskz_compress_w_256:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpcompressw %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xa9,0x63,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_compress_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressw %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xa9,0x63,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i16 %mask to <16 x i1>
%2 = call <16 x i16> @llvm.x86.avx512.mask.compress.v16i16(<16 x i16> %data, <16 x i16> zeroinitializer, <16 x i1> %1)
ret <16 x i16> %2
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <16 x i16> @test_compress_w_256(<16 x i16> %data) {
; CHECK-LABEL: test_compress_w_256:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <16 x i16> @llvm.x86.avx512.mask.compress.v16i16(<16 x i16> %data, <16 x i16> undef, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <16 x i16> %1
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
define void @test_mask_compress_store_b_256(i8* %addr, <32 x i8> %data, i32 %mask) {
; X86-LABEL: test_mask_compress_store_b_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT: vpcompressb %ymm0, (%eax) {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x63,0x00]
; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_store_b_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vpcompressb %ymm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x63,0x07]
; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT: retq # encoding: [0xc3]
%1 = bitcast i32 %mask to <32 x i1>
call void @llvm.masked.compressstore.v32i8(<32 x i8> %data, i8* %addr, <32 x i1> %1)
ret void
}
define void @test_compress_store_b_256(i8* %addr, <32 x i8> %data) {
; X86-LABEL: test_compress_store_b_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kxnord %k0, %k0, %k1 # encoding: [0xc4,0xe1,0xfd,0x46,0xc8]
; X86-NEXT: vpcompressb %ymm0, (%eax) {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x63,0x00]
; X86-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_compress_store_b_256:
; X64: # %bb.0:
; X64-NEXT: kxnord %k0, %k0, %k1 # encoding: [0xc4,0xe1,0xfd,0x46,0xc8]
; X64-NEXT: vpcompressb %ymm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x63,0x07]
; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT: retq # encoding: [0xc3]
call void @llvm.masked.compressstore.v32i8(<32 x i8> %data, i8* %addr, <32 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret void
}
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
define <32 x i8> @test_mask_compress_b_256(<32 x i8> %data, <32 x i8> %passthru, i32 %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_compress_b_256:
; X86: # %bb.0:
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpcompressb %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x63,0xc1]
; X86-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_compress_b_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressb %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0x63,0xc1]
; X64-NEXT: vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i32 %mask to <32 x i1>
%2 = call <32 x i8> @llvm.x86.avx512.mask.compress.v32i8(<32 x i8> %data, <32 x i8> %passthru, <32 x i1> %1)
ret <32 x i8> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <32 x i8> @test_maskz_compress_b_256(<32 x i8> %data, i32 %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_maskz_compress_b_256:
; X86: # %bb.0:
; X86-NEXT: kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT: vpcompressb %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xa9,0x63,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_maskz_compress_b_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpcompressb %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xa9,0x63,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = bitcast i32 %mask to <32 x i1>
%2 = call <32 x i8> @llvm.x86.avx512.mask.compress.v32i8(<32 x i8> %data, <32 x i8> zeroinitializer, <32 x i1> %1)
ret <32 x i8> %2
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
define <32 x i8> @test_compress_b_256(<32 x i8> %data) {
; CHECK-LABEL: test_compress_b_256:
[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: ret{{[l|q]}} # encoding: [0xc3]
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
%1 = call <32 x i8> @llvm.x86.avx512.mask.compress.v32i8(<32 x i8> %data, <32 x i8> undef, <32 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>)
ret <32 x i8> %1
[X86] Expand the testing of expand and compress intrinsics The avx512f intrinsic tests were in the avx512vl file. We were also missing some combinations of masking. This does show that we fail to use the zero masking form of expand loads when the passthru is zero. I'll try to get that fixed shortly. llvm-svn: 333795
2018-06-02 05:59:24 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i32> @test_int_x86_avx512_mask_vpshld_d_128(<4 x i32> %x0, <4 x i32> %x1,<4 x i32> %x3, i8 %x4) {
[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_vpshld_d_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshldd $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshldd $23, %xmm1, %xmm0, %xmm3 # encoding: [0x62,0xf3,0x7d,0x08,0x71,0xd9,0x17]
; X86-NEXT: vpshldd $24, %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0x89,0x71,0xc1,0x18]
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
; X86-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
[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: vpaddd %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfe,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshld_d_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshldd $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshldd $23, %xmm1, %xmm0, %xmm3 # encoding: [0x62,0xf3,0x7d,0x08,0x71,0xd9,0x17]
; X64-NEXT: vpshldd $24, %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0x89,0x71,0xc1,0x18]
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
; X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
[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: vpaddd %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfe,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> <i32 22, i32 22, i32 22, i32 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%extract1 = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract1, <4 x i32> %1, <4 x i32> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> <i32 23, i32 23, i32 23, i32 23>)
%5 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> <i32 24, i32 24, i32 24, i32 24>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%6 = bitcast i8 %x4 to <8 x i1>
%extract = shufflevector <8 x i1> %6, <8 x i1> %6, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%7 = select <4 x i1> %extract, <4 x i32> %5, <4 x i32> zeroinitializer
%res3 = add <4 x i32> %3, %4
%res4 = add <4 x i32> %res3, %7
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <4 x i32> %res4
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i32> @test_int_x86_avx512_mask_vpshld_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x3, i8 %x4) {
[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_vpshld_d_256:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshldd $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x29,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshldd $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0x7d,0x28,0x71,0xc1,0x17]
; X86-NEXT: vpaddd %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfe,0xc0]
[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_vpshld_d_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshldd $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x29,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshldd $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0x7d,0x28,0x71,0xc1,0x17]
; X64-NEXT: vpaddd %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfe,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> <i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> <i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <8 x i32> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <8 x i32> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <2 x i64> @test_int_x86_avx512_mask_vpshld_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x3, i8 %x4) {
[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_vpshld_q_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshldq $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshldq $23, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x71,0xc1,0x17]
; X86-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
[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_vpshld_q_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshldq $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshldq $23, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x71,0xc1,0x17]
; X64-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> <i64 22, i64 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
%3 = select <2 x i1> %extract, <2 x i64> %1, <2 x i64> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> <i64 23, i64 23>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <2 x i64> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <2 x i64> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i64> @test_int_x86_avx512_mask_vpshld_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x3, i8 %x4) {
[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_vpshld_q_256:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshldq $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshldq $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x71,0xc1,0x17]
; X86-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
[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_vpshld_q_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshldq $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x71,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshldq $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x71,0xc1,0x17]
; X64-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> <i64 22, i64 22, i64 22, i64 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract, <4 x i64> %1, <4 x i64> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> <i64 23, i64 23, i64 23, i64 23>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <4 x i64> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <4 x i64> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i16> @test_int_x86_avx512_mask_vpshld_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x3, i8 %x4) {
[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_vpshld_w_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X86-NEXT: vpshldw $6, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x70,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshldw $7, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x70,0xc1,0x07]
; X86-NEXT: vpaddw %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfd,0xc0]
[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_vpshld_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X64-NEXT: vpshldw $6, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x70,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshldw $7, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x70,0xc1,0x07]
; X64-NEXT: vpaddw %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfd,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> <i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <8 x i16> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <8 x i16> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <16 x i16> @test_int_x86_avx512_mask_vpshld_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x3, i16 %x4) {
[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_vpshld_w_256:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X86-NEXT: vpshldw $6, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x70,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshldw $7, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x70,0xc1,0x07]
; X86-NEXT: vpaddw %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfd,0xc0]
[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_vpshld_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X64-NEXT: vpshldw $6, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x70,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshldw $7, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x70,0xc1,0x07]
; X64-NEXT: vpaddw %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfd,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> <i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i16 %x4 to <16 x i1>
%3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <16 x i16> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <16 x i16> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i32> @test_int_x86_avx512_mask_vpshrd_d_128(<4 x i32> %x0, <4 x i32> %x1,<4 x i32> %x3, i8 %x4) {
[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_vpshrd_d_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshrdd $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshrdd $23, %xmm1, %xmm0, %xmm3 # encoding: [0x62,0xf3,0x7d,0x08,0x73,0xd9,0x17]
; X86-NEXT: vpshrdd $24, %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0x89,0x73,0xc1,0x18]
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
; X86-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
[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: vpaddd %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfe,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshrd_d_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshrdd $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x09,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshrdd $23, %xmm1, %xmm0, %xmm3 # encoding: [0x62,0xf3,0x7d,0x08,0x73,0xd9,0x17]
; X64-NEXT: vpshrdd $24, %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0x89,0x73,0xc1,0x18]
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
; X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
[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: vpaddd %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfe,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x1, <4 x i32> %x0, <4 x i32> <i32 22, i32 22, i32 22, i32 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%extract1 = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract1, <4 x i32> %1, <4 x i32> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x1, <4 x i32> %x0, <4 x i32> <i32 23, i32 23, i32 23, i32 23>)
%5 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x1, <4 x i32> %x0, <4 x i32> <i32 24, i32 24, i32 24, i32 24>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%6 = bitcast i8 %x4 to <8 x i1>
%extract = shufflevector <8 x i1> %6, <8 x i1> %6, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%7 = select <4 x i1> %extract, <4 x i32> %5, <4 x i32> zeroinitializer
%res3 = add <4 x i32> %3, %4
%res4 = add <4 x i32> %res3, %7
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <4 x i32> %res4
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i32> @test_int_x86_avx512_mask_vpshrd_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x3, i8 %x4) {
[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_vpshrd_d_256:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshrdd $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x29,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshrdd $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0x7d,0x28,0x73,0xc1,0x17]
; X86-NEXT: vpaddd %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfe,0xc0]
[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_vpshrd_d_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshrdd $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x29,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshrdd $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0x7d,0x28,0x73,0xc1,0x17]
; X64-NEXT: vpaddd %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfe,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x1, <8 x i32> %x0, <8 x i32> <i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x1, <8 x i32> %x0, <8 x i32> <i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <8 x i32> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <8 x i32> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <2 x i64> @test_int_x86_avx512_mask_vpshrd_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x3, i8 %x4) {
[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_vpshrd_q_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshrdq $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshrdq $23, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x73,0xc1,0x17]
; X86-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
[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_vpshrd_q_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshrdq $22, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshrdq $23, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x73,0xc1,0x17]
; X64-NEXT: vpaddq %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xd4,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x1, <2 x i64> %x0, <2 x i64> <i64 22, i64 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
%3 = select <2 x i1> %extract, <2 x i64> %1, <2 x i64> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x1, <2 x i64> %x0, <2 x i64> <i64 23, i64 23>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <2 x i64> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <2 x i64> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i64> @test_int_x86_avx512_mask_vpshrd_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x3, i8 %x4) {
[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_vpshrd_q_256:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
; X86-NEXT: vpshrdq $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshrdq $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x73,0xc1,0x17]
; X86-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
[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_vpshrd_q_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT: vpshrdq $22, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x73,0xd1,0x16]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshrdq $23, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x73,0xc1,0x17]
; X64-NEXT: vpaddq %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xd4,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x1, <4 x i64> %x0, <4 x i64> <i64 22, i64 22, i64 22, i64 22>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract, <4 x i64> %1, <4 x i64> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x1, <4 x i64> %x0, <4 x i64> <i64 23, i64 23, i64 23, i64 23>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <4 x i64> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <4 x i64> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i16> @test_int_x86_avx512_mask_vpshrd_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x3, i8 %x4) {
[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_vpshrd_w_128:
; X86: # %bb.0:
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
; X86-NEXT: kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X86-NEXT: vpshrdw $6, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x72,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshrdw $7, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x72,0xc1,0x07]
; X86-NEXT: vpaddw %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfd,0xc0]
[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_vpshrd_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X64-NEXT: vpshrdw $6, %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x09,0x72,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshrdw $7, %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf3,0xfd,0x08,0x72,0xc1,0x07]
; X64-NEXT: vpaddw %xmm0, %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe9,0xfd,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x1, <8 x i16> %x0, <8 x i16> <i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i8 %x4 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x1, <8 x i16> %x0, <8 x i16> <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <8 x i16> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <8 x i16> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <16 x i16> @test_int_x86_avx512_mask_vpshrd_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x3, i16 %x4) {
[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_vpshrd_w_256:
; X86: # %bb.0:
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X86-NEXT: vpshrdw $6, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x72,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshrdw $7, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x72,0xc1,0x07]
; X86-NEXT: vpaddw %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfd,0xc0]
[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_vpshrd_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
[X86] Update VBMI2 vshld/vshrd tests to use an immediate that doesn't require a modulo. Planning to replace these with funnel shift intrinsics which would mask out the extra bits. This will help minimize test diffs. llvm-svn: 350504
2019-01-07 13:58:53 +08:00
; X64-NEXT: vpshrdw $6, %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf3,0xfd,0x29,0x72,0xd1,0x06]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshrdw $7, %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf3,0xfd,0x28,0x72,0xc1,0x07]
; X64-NEXT: vpaddw %ymm0, %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xed,0xfd,0xc0]
[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]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x1, <16 x i16> %x0, <16 x i16> <i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6, i16 6>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%2 = bitcast i16 %x4 to <16 x i1>
%3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %x3
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%4 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x1, <16 x i16> %x0, <16 x i16> <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>)
[X86] Remove masking from avx512vbmi2 concat and shift by immediate intrinsics. Use select in IR instead. llvm-svn: 334576
2018-06-13 15:19:21 +08:00
%res2 = add <16 x i16> %3, %4
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
ret <16 x i16> %res2
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i32> @test_int_x86_avx512_mask_vpshrdv_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32>* %x2p, <8 x i32> %x4, 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_vpshrdv_d_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X86-NEXT: vpshrdvd (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x73,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshrdvd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x73,0xc2]
[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: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshrdv_d_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X64-NEXT: vpshrdvd (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x73,0x1f]
; X64-NEXT: vpshrdvd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x73,0xc2]
; X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <8 x i32>, <8 x i32>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x1, <8 x i32> %x0, <8 x i32> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0
%4 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x1, <8 x i32> %x0, <8 x i32> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer
%res3 = add <8 x i32> %3, %6
ret <8 x i32> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i32> @test_int_x86_avx512_mask_vpshrdv_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32>* %x2p, <4 x i32> %x4, 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_vpshrdv_d_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X86-NEXT: vpshrdvd (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x73,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshrdvd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x73,0xc2]
[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: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshrdv_d_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X64-NEXT: vpshrdvd (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x73,0x1f]
; X64-NEXT: vpshrdvd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x73,0xc2]
; X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <4 x i32>, <4 x i32>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x1, <4 x i32> %x0, <4 x i32> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x0
%4 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x1, <4 x i32> %x0, <4 x i32> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer
%res3 = add <4 x i32> %3, %6
ret <4 x i32> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i64> @test_int_x86_avx512_mask_vpshrdv_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64>* %x2p, <4 x i64> %x4, 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_vpshrdv_q_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X86-NEXT: vpshrdvq (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x73,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshrdvq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0x73,0xc2]
[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: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshrdv_q_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X64-NEXT: vpshrdvq (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x73,0x1f]
; X64-NEXT: vpshrdvq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0x73,0xc2]
; X64-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <4 x i64>, <4 x i64>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x1, <4 x i64> %x0, <4 x i64> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract, <4 x i64> %1, <4 x i64> %x0
%4 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x1, <4 x i64> %x0, <4 x i64> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%6 = select <4 x i1> %extract1, <4 x i64> %4, <4 x i64> zeroinitializer
%res3 = add <4 x i64> %3, %6
ret <4 x i64> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <2 x i64> @test_int_x86_avx512_mask_vpshrdv_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64>* %x2p, <2 x i64> %x4, 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_vpshrdv_q_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X86-NEXT: vpshrdvq (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x73,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshrdvq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0x73,0xc2]
[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: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshrdv_q_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X64-NEXT: vpshrdvq (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x73,0x1f]
; X64-NEXT: vpshrdvq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0x73,0xc2]
; X64-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <2 x i64>, <2 x i64>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x1, <2 x i64> %x0, <2 x i64> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
%3 = select <2 x i1> %extract, <2 x i64> %1, <2 x i64> %x0
%4 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x1, <2 x i64> %x0, <2 x i64> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
%6 = select <2 x i1> %extract1, <2 x i64> %4, <2 x i64> zeroinitializer
%res3 = add <2 x i64> %3, %6
ret <2 x i64> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <16 x i16> @test_int_x86_avx512_mask_vpshrdv_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16>* %x2p, <16 x i16> %x4, 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_vpshrdv_w_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X86-NEXT: vpshrdvw (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x72,0x18]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshrdvw %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0xf5,0x28,0x72,0xc2]
[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: vpaddw %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfd,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshrdv_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X64-NEXT: vpshrdvw (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x72,0x1f]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshrdvw %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0xf5,0x28,0x72,0xc2]
[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: vpaddw %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfd,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <16 x i16>, <16 x i16>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x1, <16 x i16> %x0, <16 x i16> %x2)
%2 = bitcast i16 %x3 to <16 x i1>
%3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %x0
%4 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x1, <16 x i16> %x0, <16 x i16> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i16 %x3 to <16 x i1>
%6 = select <16 x i1> %5, <16 x i16> %4, <16 x i16> zeroinitializer
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%res3 = add <16 x i16> %3, %4
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
ret <16 x i16> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i16> @test_int_x86_avx512_mask_vpshrdv_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16>* %x2p, <8 x i16> %x4, 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_vpshrdv_w_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X86-NEXT: vpshrdvw (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x72,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshrdvw %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0x72,0xc2]
[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: vpaddw %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfd,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshrdv_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X64-NEXT: vpshrdvw (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x72,0x1f]
; X64-NEXT: vpshrdvw %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0x72,0xc2]
; X64-NEXT: vpaddw %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfd,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <8 x i16>, <8 x i16>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x1, <8 x i16> %x0, <8 x i16> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %x0
%4 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x1, <8 x i16> %x0, <8 x i16> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%6 = select <8 x i1> %5, <8 x i16> %4, <8 x i16> zeroinitializer
%res3 = add <8 x i16> %3, %6
ret <8 x i16> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i32> @test_int_x86_avx512_mask_vpshldv_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32>* %x2p, <8 x i32> %x4, 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_vpshldv_d_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X86-NEXT: vpshldvd (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x71,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshldvd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x71,0xc2]
[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: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshldv_d_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X64-NEXT: vpshldvd (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0x75,0x29,0x71,0x1f]
; X64-NEXT: vpshldvd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0x71,0xc2]
; X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <8 x i32>, <8 x i32>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i32> %1, <8 x i32> %x0
%4 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%6 = select <8 x i1> %5, <8 x i32> %4, <8 x i32> zeroinitializer
%res3 = add <8 x i32> %3, %6
ret <8 x i32> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i32> @test_int_x86_avx512_mask_vpshldv_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32>* %x2p, <4 x i32> %x4, 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_vpshldv_d_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X86-NEXT: vpshldvd (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x71,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshldvd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x71,0xc2]
[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: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshldv_d_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X64-NEXT: vpshldvd (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0x71,0x1f]
; X64-NEXT: vpshldvd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0x71,0xc2]
; X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <4 x i32>, <4 x i32>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract, <4 x i32> %1, <4 x i32> %x0
%4 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%6 = select <4 x i1> %extract1, <4 x i32> %4, <4 x i32> zeroinitializer
%res3 = add <4 x i32> %3, %6
ret <4 x i32> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <4 x i64> @test_int_x86_avx512_mask_vpshldv_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64>* %x2p, <4 x i64> %x4, 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_vpshldv_q_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X86-NEXT: vpshldvq (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x71,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshldvq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0x71,0xc2]
[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: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshldv_q_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X64-NEXT: vpshldvq (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x71,0x1f]
; X64-NEXT: vpshldvq %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0x71,0xc2]
; X64-NEXT: vpaddq %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xd4,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <4 x i64>, <4 x i64>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%3 = select <4 x i1> %extract, <4 x i64> %1, <4 x i64> %x0
%4 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
%6 = select <4 x i1> %extract1, <4 x i64> %4, <4 x i64> zeroinitializer
%res3 = add <4 x i64> %3, %6
ret <4 x i64> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <2 x i64> @test_int_x86_avx512_mask_vpshldv_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64>* %x2p, <2 x i64> %x4, 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_vpshldv_q_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X86-NEXT: vpshldvq (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x71,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshldvq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0x71,0xc2]
[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: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshldv_q_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X64-NEXT: vpshldvq (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x71,0x1f]
; X64-NEXT: vpshldvq %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0x71,0xc2]
; X64-NEXT: vpaddq %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xd4,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <2 x i64>, <2 x i64>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
%3 = select <2 x i1> %extract, <2 x i64> %1, <2 x i64> %x0
%4 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%extract1 = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
%6 = select <2 x i1> %extract1, <2 x i64> %4, <2 x i64> zeroinitializer
%res3 = add <2 x i64> %3, %6
ret <2 x i64> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <16 x i16> @test_int_x86_avx512_mask_vpshldv_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16>* %x2p, <16 x i16> %x4, 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_vpshldv_w_256:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X86-NEXT: vpshldvw (%eax), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x70,0x18]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: vpshldvw %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0x70,0xc2]
[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: vpaddw %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfd,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshldv_w_256:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %ymm0, %ymm3 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xd8]
; X64-NEXT: vpshldvw (%rdi), %ymm1, %ymm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0x70,0x1f]
; X64-NEXT: vpshldvw %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0x70,0xc2]
; X64-NEXT: vpaddw %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfd,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <16 x i16>, <16 x i16>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2)
%2 = bitcast i16 %x3 to <16 x i1>
%3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %x0
%4 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i16 %x3 to <16 x i1>
%6 = select <16 x i1> %5, <16 x i16> %4, <16 x i16> zeroinitializer
%res3 = add <16 x i16> %3, %6
ret <16 x i16> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
define <8 x i16> @test_int_x86_avx512_mask_vpshldv_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16>* %x2p, <8 x i16> %x4, 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_vpshldv_w_128:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
; X86-NEXT: kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
[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: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X86-NEXT: vpshldvw (%eax), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x70,0x18]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X86-NEXT: vpshldvw %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0xf5,0x08,0x70,0xc2]
[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: vpaddw %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfd,0xc0]
; X86-NEXT: retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpshldv_w_128:
; X64: # %bb.0:
; X64-NEXT: kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT: vmovdqa %xmm0, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd8]
; X64-NEXT: vpshldvw (%rdi), %xmm1, %xmm3 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0x70,0x1f]
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
; X64-NEXT: vpshldvw %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0xf5,0x08,0x70,0xc2]
[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: vpaddw %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfd,0xc0]
; X64-NEXT: retq # encoding: [0xc3]
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
%x2 = load <8 x i16>, <8 x i16>* %x2p
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%1 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2)
%2 = bitcast i8 %x3 to <8 x i1>
%3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %x0
%4 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x4)
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
%5 = bitcast i8 %x3 to <8 x i1>
%6 = select <8 x i1> %5, <8 x i16> %4, <8 x i16> zeroinitializer
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
%res3 = add <8 x i16> %3, %4
[X86] Split more masked instruction tests to enable D60940. More mechanical splitting of tests so we can add a one use check to the isel patterns for forming masked instructions. In a few cases I changed immediates of instructions in order to avoid needing to split.
2020-03-26 00:23:07 +08:00
ret <8 x i16> %res3
[x86][icelake]vbmi2 introducing vbmi2, consisting of vpcompress{b,w} vpexpand{b,w} vpsh{l,r}d{w,d,q} vpsh{l,r}dv{w,d,q} Differential Revision: https://reviews.llvm.org/D40206 llvm-svn: 318745
2017-11-21 17:48:44 +08:00
}
Recommit r350554 "[X86] Remove AVX512VBMI2 concat and shift intrinsics. Replace with target independent funnel shift intrinsics." The MSVC limit we hit on AutoUpgrade.cpp has been worked around for now. llvm-svn: 350567
2019-01-08 05:00:32 +08:00
declare <4 x i32> @llvm.fshl.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
declare <8 x i32> @llvm.fshl.v8i32(<8 x i32>, <8 x i32>, <8 x i32>)
declare <2 x i64> @llvm.fshl.v2i64(<2 x i64>, <2 x i64>, <2 x i64>)
declare <4 x i64> @llvm.fshl.v4i64(<4 x i64>, <4 x i64>, <4 x i64>)
declare <8 x i16> @llvm.fshl.v8i16(<8 x i16>, <8 x i16>, <8 x i16>)
declare <16 x i16> @llvm.fshl.v16i16(<16 x i16>, <16 x i16>, <16 x i16>)
declare <4 x i32> @llvm.fshr.v4i32(<4 x i32>, <4 x i32>, <4 x i32>)
declare <8 x i32> @llvm.fshr.v8i32(<8 x i32>, <8 x i32>, <8 x i32>)
declare <2 x i64> @llvm.fshr.v2i64(<2 x i64>, <2 x i64>, <2 x i64>)
declare <4 x i64> @llvm.fshr.v4i64(<4 x i64>, <4 x i64>, <4 x i64>)
declare <8 x i16> @llvm.fshr.v8i16(<8 x i16>, <8 x i16>, <8 x i16>)
declare <16 x i16> @llvm.fshr.v16i16(<16 x i16>, <16 x i16>, <16 x i16>)
[X86] Add vbmi2 compressstore and expandload tests that aren't fast-isel tests. These got removed when we autoupgraded to target independent intrinsics, but we didn't have coverage anywhere else. The avx512f/avx512vl versions do have coverage. Also move some tests back from the upgrade file that aren't really upgraded. llvm-svn: 352342
2019-01-28 13:42:39 +08:00
declare <8 x i16> @llvm.masked.expandload.v8i16(i16*, <8 x i1>, <8 x i16>)
declare <16 x i8> @llvm.masked.expandload.v16i8(i8*, <16 x i1>, <16 x i8>)
declare void @llvm.masked.compressstore.v8i16(<8 x i16>, i16*, <8 x i1>)
declare void @llvm.masked.compressstore.v16i8(<16 x i8>, i8*, <16 x i1>)
declare <16 x i16> @llvm.masked.expandload.v16i16(i16*, <16 x i1>, <16 x i16>)
declare <32 x i8> @llvm.masked.expandload.v32i8(i8*, <32 x i1>, <32 x i8>)
declare void @llvm.masked.compressstore.v16i16(<16 x i16>, i16*, <16 x i1>)
declare void @llvm.masked.compressstore.v32i8(<32 x i8>, i8*, <32 x i1>)
[X86] Add new variadic avx512 compress/expand intrinsics that use vXi1 types for the mask argument. Remove and autoupgrade the old intrinsics llvm-svn: 352343
2019-01-28 15:03:03 +08:00
declare <8 x i16> @llvm.x86.avx512.mask.expand.v8i16(<8 x i16>, <8 x i16>, <8 x i1>)
declare <16 x i8> @llvm.x86.avx512.mask.expand.v16i8(<16 x i8>, <16 x i8>, <16 x i1>)
declare <8 x i16> @llvm.x86.avx512.mask.compress.v8i16(<8 x i16>, <8 x i16>, <8 x i1>)
declare <16 x i8> @llvm.x86.avx512.mask.compress.v16i8(<16 x i8>, <16 x i8>, <16 x i1>)
declare <16 x i16> @llvm.x86.avx512.mask.expand.v16i16(<16 x i16>, <16 x i16>, <16 x i1>)
declare <32 x i8> @llvm.x86.avx512.mask.expand.v32i8(<32 x i8>, <32 x i8>, <32 x i1>)
declare <16 x i16> @llvm.x86.avx512.mask.compress.v16i16(<16 x i16>, <16 x i16>, <16 x i1>)
declare <32 x i8> @llvm.x86.avx512.mask.compress.v32i8(<32 x i8>, <32 x i8>, <32 x i1>)