diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/bitcast-and-setcc-128.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/bitcast-and-setcc-128.ll | 459 |
1 files changed, 317 insertions, 142 deletions
diff --git a/llvm/test/CodeGen/X86/bitcast-and-setcc-128.ll b/llvm/test/CodeGen/X86/bitcast-and-setcc-128.ll index 190f1c92c24..1d482b1b00e 100644 --- a/llvm/test/CodeGen/X86/bitcast-and-setcc-128.ll +++ b/llvm/test/CodeGen/X86/bitcast-and-setcc-128.ll @@ -3,7 +3,8 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSSE3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX12,AVX1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX12,AVX2 -; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512 +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512 --check-prefixes=AVX512F +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512 --check-prefixes=AVX512BW define i8 @v8i16(<8 x i16> %a, <8 x i16> %b, <8 x i16> %c, <8 x i16> %d) { ; SSE2-SSSE3-LABEL: v8i16: @@ -26,13 +27,28 @@ define i8 @v8i16(<8 x i16> %a, <8 x i16> %b, <8 x i16> %c, <8 x i16> %d) { ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v8i16: -; AVX512: # BB#0: -; AVX512-NEXT: vpcmpgtw %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtw %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v8i16: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpcmpgtw %xmm1, %xmm0, %xmm0 +; AVX512F-NEXT: vpmovsxwq %xmm0, %zmm0 +; AVX512F-NEXT: vpsllq $63, %zmm0, %zmm0 +; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1 +; AVX512F-NEXT: vpcmpgtw %xmm3, %xmm2, %xmm0 +; AVX512F-NEXT: vpmovsxwq %xmm0, %zmm0 +; AVX512F-NEXT: vpsllq $63, %zmm0, %zmm0 +; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v8i16: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpcmpgtw %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtw %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <8 x i16> %a, %b %x1 = icmp sgt <8 x i16> %c, %d %y = and <8 x i1> %x0, %x1 @@ -59,14 +75,23 @@ define i4 @v4i32(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, <4 x i32> %d) { ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v4i32: -; AVX512: # BB#0: -; AVX512-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v4i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 +; AVX512F-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v4i32: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <4 x i32> %a, %b %x1 = icmp sgt <4 x i32> %c, %d %y = and <4 x i1> %x0, %x1 @@ -93,14 +118,23 @@ define i4 @v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x float> %d) ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v4f32: -; AVX512: # BB#0: -; AVX512-NEXT: vcmpltps %xmm0, %xmm1, %k1 -; AVX512-NEXT: vcmpltps %xmm2, %xmm3, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v4f32: +; AVX512F: # BB#0: +; AVX512F-NEXT: vcmpltps %xmm0, %xmm1, %k1 +; AVX512F-NEXT: vcmpltps %xmm2, %xmm3, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v4f32: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vcmpltps %xmm0, %xmm1, %k1 +; AVX512BW-NEXT: vcmpltps %xmm2, %xmm3, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = fcmp ogt <4 x float> %a, %b %x1 = fcmp ogt <4 x float> %c, %d %y = and <4 x i1> %x0, %x1 @@ -127,13 +161,28 @@ define i16 @v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c, <16 x i8> %d) { ; AVX12-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v16i8: -; AVX512: # BB#0: -; AVX512-NEXT: vpcmpgtb %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtb %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill> -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v16i8: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpcmpgtb %xmm1, %xmm0, %xmm0 +; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0 +; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0 +; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k1 +; AVX512F-NEXT: vpcmpgtb %xmm3, %xmm2, %xmm0 +; AVX512F-NEXT: vpmovsxbd %xmm0, %zmm0 +; AVX512F-NEXT: vpslld $31, %zmm0, %zmm0 +; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill> +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v16i8: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpcmpgtb %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtb %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: # kill: %AX<def> %AX<kill> %EAX<kill> +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <16 x i8> %a, %b %x1 = icmp sgt <16 x i8> %c, %d %y = and <16 x i1> %x0, %x1 @@ -256,22 +305,39 @@ define i2 @v2i8(<2 x i8> %a, <2 x i8> %b, <2 x i8> %c, <2 x i8> %d) { ; AVX2-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX2-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v2i8: -; AVX512: # BB#0: -; AVX512-NEXT: vpsllq $56, %xmm3, %xmm3 -; AVX512-NEXT: vpsraq $56, %xmm3, %xmm3 -; AVX512-NEXT: vpsllq $56, %xmm2, %xmm2 -; AVX512-NEXT: vpsraq $56, %xmm2, %xmm2 -; AVX512-NEXT: vpsllq $56, %xmm1, %xmm1 -; AVX512-NEXT: vpsraq $56, %xmm1, %xmm1 -; AVX512-NEXT: vpsllq $56, %xmm0, %xmm0 -; AVX512-NEXT: vpsraq $56, %xmm0, %xmm0 -; AVX512-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v2i8: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpsllq $56, %xmm3, %xmm3 +; AVX512F-NEXT: vpsraq $56, %xmm3, %xmm3 +; AVX512F-NEXT: vpsllq $56, %xmm2, %xmm2 +; AVX512F-NEXT: vpsraq $56, %xmm2, %xmm2 +; AVX512F-NEXT: vpsllq $56, %xmm1, %xmm1 +; AVX512F-NEXT: vpsraq $56, %xmm1, %xmm1 +; AVX512F-NEXT: vpsllq $56, %xmm0, %xmm0 +; AVX512F-NEXT: vpsraq $56, %xmm0, %xmm0 +; AVX512F-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512F-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v2i8: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpsllq $56, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsraq $56, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsllq $56, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsraq $56, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsllq $56, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsraq $56, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsllq $56, %xmm0, %xmm0 +; AVX512BW-NEXT: vpsraq $56, %xmm0, %xmm0 +; AVX512BW-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <2 x i8> %a, %b %x1 = icmp sgt <2 x i8> %c, %d %y = and <2 x i1> %x0, %x1 @@ -394,22 +460,39 @@ define i2 @v2i16(<2 x i16> %a, <2 x i16> %b, <2 x i16> %c, <2 x i16> %d) { ; AVX2-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX2-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v2i16: -; AVX512: # BB#0: -; AVX512-NEXT: vpsllq $48, %xmm3, %xmm3 -; AVX512-NEXT: vpsraq $48, %xmm3, %xmm3 -; AVX512-NEXT: vpsllq $48, %xmm2, %xmm2 -; AVX512-NEXT: vpsraq $48, %xmm2, %xmm2 -; AVX512-NEXT: vpsllq $48, %xmm1, %xmm1 -; AVX512-NEXT: vpsraq $48, %xmm1, %xmm1 -; AVX512-NEXT: vpsllq $48, %xmm0, %xmm0 -; AVX512-NEXT: vpsraq $48, %xmm0, %xmm0 -; AVX512-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v2i16: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpsllq $48, %xmm3, %xmm3 +; AVX512F-NEXT: vpsraq $48, %xmm3, %xmm3 +; AVX512F-NEXT: vpsllq $48, %xmm2, %xmm2 +; AVX512F-NEXT: vpsraq $48, %xmm2, %xmm2 +; AVX512F-NEXT: vpsllq $48, %xmm1, %xmm1 +; AVX512F-NEXT: vpsraq $48, %xmm1, %xmm1 +; AVX512F-NEXT: vpsllq $48, %xmm0, %xmm0 +; AVX512F-NEXT: vpsraq $48, %xmm0, %xmm0 +; AVX512F-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512F-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v2i16: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpsllq $48, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsraq $48, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsllq $48, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsraq $48, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsllq $48, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsraq $48, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsllq $48, %xmm0, %xmm0 +; AVX512BW-NEXT: vpsraq $48, %xmm0, %xmm0 +; AVX512BW-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <2 x i16> %a, %b %x1 = icmp sgt <2 x i16> %c, %d %y = and <2 x i1> %x0, %x1 @@ -516,22 +599,39 @@ define i2 @v2i32(<2 x i32> %a, <2 x i32> %b, <2 x i32> %c, <2 x i32> %d) { ; AVX2-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX2-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v2i32: -; AVX512: # BB#0: -; AVX512-NEXT: vpsllq $32, %xmm3, %xmm3 -; AVX512-NEXT: vpsraq $32, %xmm3, %xmm3 -; AVX512-NEXT: vpsllq $32, %xmm2, %xmm2 -; AVX512-NEXT: vpsraq $32, %xmm2, %xmm2 -; AVX512-NEXT: vpsllq $32, %xmm1, %xmm1 -; AVX512-NEXT: vpsraq $32, %xmm1, %xmm1 -; AVX512-NEXT: vpsllq $32, %xmm0, %xmm0 -; AVX512-NEXT: vpsraq $32, %xmm0, %xmm0 -; AVX512-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v2i32: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpsllq $32, %xmm3, %xmm3 +; AVX512F-NEXT: vpsraq $32, %xmm3, %xmm3 +; AVX512F-NEXT: vpsllq $32, %xmm2, %xmm2 +; AVX512F-NEXT: vpsraq $32, %xmm2, %xmm2 +; AVX512F-NEXT: vpsllq $32, %xmm1, %xmm1 +; AVX512F-NEXT: vpsraq $32, %xmm1, %xmm1 +; AVX512F-NEXT: vpsllq $32, %xmm0, %xmm0 +; AVX512F-NEXT: vpsraq $32, %xmm0, %xmm0 +; AVX512F-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512F-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v2i32: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpsllq $32, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsraq $32, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsllq $32, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsraq $32, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsllq $32, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsraq $32, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsllq $32, %xmm0, %xmm0 +; AVX512BW-NEXT: vpsraq $32, %xmm0, %xmm0 +; AVX512BW-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <2 x i32> %a, %b %x1 = icmp sgt <2 x i32> %c, %d %y = and <2 x i1> %x0, %x1 @@ -577,14 +677,23 @@ define i2 @v2i64(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c, <2 x i64> %d) { ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v2i64: -; AVX512: # BB#0: -; AVX512-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v2i64: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512F-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v2i64: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpcmpgtq %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtq %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <2 x i64> %a, %b %x1 = icmp sgt <2 x i64> %c, %d %y = and <2 x i1> %x0, %x1 @@ -611,14 +720,23 @@ define i2 @v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c, <2 x double> ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v2f64: -; AVX512: # BB#0: -; AVX512-NEXT: vcmpltpd %xmm0, %xmm1, %k1 -; AVX512-NEXT: vcmpltpd %xmm2, %xmm3, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v2f64: +; AVX512F: # BB#0: +; AVX512F-NEXT: vcmpltpd %xmm0, %xmm1, %k1 +; AVX512F-NEXT: vcmpltpd %xmm2, %xmm3, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v2f64: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vcmpltpd %xmm0, %xmm1, %k1 +; AVX512BW-NEXT: vcmpltpd %xmm2, %xmm3, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = fcmp ogt <2 x double> %a, %b %x1 = fcmp ogt <2 x double> %c, %d %y = and <2 x i1> %x0, %x1 @@ -661,22 +779,39 @@ define i4 @v4i8(<4 x i8> %a, <4 x i8> %b, <4 x i8> %c, <4 x i8> %d) { ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v4i8: -; AVX512: # BB#0: -; AVX512-NEXT: vpslld $24, %xmm3, %xmm3 -; AVX512-NEXT: vpsrad $24, %xmm3, %xmm3 -; AVX512-NEXT: vpslld $24, %xmm2, %xmm2 -; AVX512-NEXT: vpsrad $24, %xmm2, %xmm2 -; AVX512-NEXT: vpslld $24, %xmm1, %xmm1 -; AVX512-NEXT: vpsrad $24, %xmm1, %xmm1 -; AVX512-NEXT: vpslld $24, %xmm0, %xmm0 -; AVX512-NEXT: vpsrad $24, %xmm0, %xmm0 -; AVX512-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v4i8: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpslld $24, %xmm3, %xmm3 +; AVX512F-NEXT: vpsrad $24, %xmm3, %xmm3 +; AVX512F-NEXT: vpslld $24, %xmm2, %xmm2 +; AVX512F-NEXT: vpsrad $24, %xmm2, %xmm2 +; AVX512F-NEXT: vpslld $24, %xmm1, %xmm1 +; AVX512F-NEXT: vpsrad $24, %xmm1, %xmm1 +; AVX512F-NEXT: vpslld $24, %xmm0, %xmm0 +; AVX512F-NEXT: vpsrad $24, %xmm0, %xmm0 +; AVX512F-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 +; AVX512F-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v4i8: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpslld $24, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsrad $24, %xmm3, %xmm3 +; AVX512BW-NEXT: vpslld $24, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsrad $24, %xmm2, %xmm2 +; AVX512BW-NEXT: vpslld $24, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsrad $24, %xmm1, %xmm1 +; AVX512BW-NEXT: vpslld $24, %xmm0, %xmm0 +; AVX512BW-NEXT: vpsrad $24, %xmm0, %xmm0 +; AVX512BW-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <4 x i8> %a, %b %x1 = icmp sgt <4 x i8> %c, %d %y = and <4 x i1> %x0, %x1 @@ -719,22 +854,39 @@ define i4 @v4i16(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c, <4 x i16> %d) { ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v4i16: -; AVX512: # BB#0: -; AVX512-NEXT: vpslld $16, %xmm3, %xmm3 -; AVX512-NEXT: vpsrad $16, %xmm3, %xmm3 -; AVX512-NEXT: vpslld $16, %xmm2, %xmm2 -; AVX512-NEXT: vpsrad $16, %xmm2, %xmm2 -; AVX512-NEXT: vpslld $16, %xmm1, %xmm1 -; AVX512-NEXT: vpsrad $16, %xmm1, %xmm1 -; AVX512-NEXT: vpslld $16, %xmm0, %xmm0 -; AVX512-NEXT: vpsrad $16, %xmm0, %xmm0 -; AVX512-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: movb %al, -{{[0-9]+}}(%rsp) -; AVX512-NEXT: movb -{{[0-9]+}}(%rsp), %al -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v4i16: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpslld $16, %xmm3, %xmm3 +; AVX512F-NEXT: vpsrad $16, %xmm3, %xmm3 +; AVX512F-NEXT: vpslld $16, %xmm2, %xmm2 +; AVX512F-NEXT: vpsrad $16, %xmm2, %xmm2 +; AVX512F-NEXT: vpslld $16, %xmm1, %xmm1 +; AVX512F-NEXT: vpsrad $16, %xmm1, %xmm1 +; AVX512F-NEXT: vpslld $16, %xmm0, %xmm0 +; AVX512F-NEXT: vpsrad $16, %xmm0, %xmm0 +; AVX512F-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 +; AVX512F-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512F-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v4i16: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpslld $16, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsrad $16, %xmm3, %xmm3 +; AVX512BW-NEXT: vpslld $16, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsrad $16, %xmm2, %xmm2 +; AVX512BW-NEXT: vpslld $16, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsrad $16, %xmm1, %xmm1 +; AVX512BW-NEXT: vpslld $16, %xmm0, %xmm0 +; AVX512BW-NEXT: vpsrad $16, %xmm0, %xmm0 +; AVX512BW-NEXT: vpcmpgtd %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtd %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: movb %al, -{{[0-9]+}}(%rsp) +; AVX512BW-NEXT: movb -{{[0-9]+}}(%rsp), %al +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <4 x i16> %a, %b %x1 = icmp sgt <4 x i16> %c, %d %y = and <4 x i1> %x0, %x1 @@ -779,21 +931,44 @@ define i8 @v8i8(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %d) { ; AVX12-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> ; AVX12-NEXT: ret{{[l|q]}} ; -; AVX512-LABEL: v8i8: -; AVX512: # BB#0: -; AVX512-NEXT: vpsllw $8, %xmm3, %xmm3 -; AVX512-NEXT: vpsraw $8, %xmm3, %xmm3 -; AVX512-NEXT: vpsllw $8, %xmm2, %xmm2 -; AVX512-NEXT: vpsraw $8, %xmm2, %xmm2 -; AVX512-NEXT: vpsllw $8, %xmm1, %xmm1 -; AVX512-NEXT: vpsraw $8, %xmm1, %xmm1 -; AVX512-NEXT: vpsllw $8, %xmm0, %xmm0 -; AVX512-NEXT: vpsraw $8, %xmm0, %xmm0 -; AVX512-NEXT: vpcmpgtw %xmm1, %xmm0, %k1 -; AVX512-NEXT: vpcmpgtw %xmm3, %xmm2, %k0 {%k1} -; AVX512-NEXT: kmovd %k0, %eax -; AVX512-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> -; AVX512-NEXT: ret{{[l|q]}} +; AVX512F-LABEL: v8i8: +; AVX512F: # BB#0: +; AVX512F-NEXT: vpsllw $8, %xmm3, %xmm3 +; AVX512F-NEXT: vpsraw $8, %xmm3, %xmm3 +; AVX512F-NEXT: vpsllw $8, %xmm2, %xmm2 +; AVX512F-NEXT: vpsraw $8, %xmm2, %xmm2 +; AVX512F-NEXT: vpcmpgtw %xmm3, %xmm2, %xmm2 +; AVX512F-NEXT: vpsllw $8, %xmm1, %xmm1 +; AVX512F-NEXT: vpsraw $8, %xmm1, %xmm1 +; AVX512F-NEXT: vpsllw $8, %xmm0, %xmm0 +; AVX512F-NEXT: vpsraw $8, %xmm0, %xmm0 +; AVX512F-NEXT: vpcmpgtw %xmm1, %xmm0, %xmm0 +; AVX512F-NEXT: vpmovsxwq %xmm0, %zmm0 +; AVX512F-NEXT: vpsllq $63, %zmm0, %zmm0 +; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k1 +; AVX512F-NEXT: vpmovsxwq %xmm2, %zmm0 +; AVX512F-NEXT: vpsllq $63, %zmm0, %zmm0 +; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0 {%k1} +; AVX512F-NEXT: kmovw %k0, %eax +; AVX512F-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> +; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: ret{{[l|q]}} +; +; AVX512BW-LABEL: v8i8: +; AVX512BW: # BB#0: +; AVX512BW-NEXT: vpsllw $8, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsraw $8, %xmm3, %xmm3 +; AVX512BW-NEXT: vpsllw $8, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsraw $8, %xmm2, %xmm2 +; AVX512BW-NEXT: vpsllw $8, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsraw $8, %xmm1, %xmm1 +; AVX512BW-NEXT: vpsllw $8, %xmm0, %xmm0 +; AVX512BW-NEXT: vpsraw $8, %xmm0, %xmm0 +; AVX512BW-NEXT: vpcmpgtw %xmm1, %xmm0, %k1 +; AVX512BW-NEXT: vpcmpgtw %xmm3, %xmm2, %k0 {%k1} +; AVX512BW-NEXT: kmovd %k0, %eax +; AVX512BW-NEXT: # kill: %AL<def> %AL<kill> %EAX<kill> +; AVX512BW-NEXT: ret{{[l|q]}} %x0 = icmp sgt <8 x i8> %a, %b %x1 = icmp sgt <8 x i8> %c, %d %y = and <8 x i1> %x0, %x1 |
