summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--llvm/test/CodeGen/X86/ptest.ll141
1 files changed, 111 insertions, 30 deletions
diff --git a/llvm/test/CodeGen/X86/ptest.ll b/llvm/test/CodeGen/X86/ptest.ll
index f298e78a1dc..8f2e10f45a3 100644
--- a/llvm/test/CodeGen/X86/ptest.ll
+++ b/llvm/test/CodeGen/X86/ptest.ll
@@ -1,6 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse4.1,-avx < %s | FileCheck %s --check-prefix=SSE41
-; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx,-avx2 < %s | FileCheck %s --check-prefix=AVX
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx,-avx2 < %s | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+avx512bw < %s | FileCheck %s --check-prefix=AVX --check-prefix=AVX512
define i32 @veccond128(<4 x i32> %input) {
; SSE41-LABEL: veccond128:
@@ -84,19 +85,49 @@ define i32 @veccond512(<16 x i32> %input) {
; SSE41-NEXT: movl $1, %eax
; SSE41-NEXT: retq
;
-; AVX-LABEL: veccond512:
-; AVX: # %bb.0: # %entry
-; AVX-NEXT: vorps %ymm1, %ymm0, %ymm0
-; AVX-NEXT: vptest %ymm0, %ymm0
-; AVX-NEXT: je .LBB2_2
-; AVX-NEXT: # %bb.1: # %if-true-block
-; AVX-NEXT: xorl %eax, %eax
-; AVX-NEXT: vzeroupper
-; AVX-NEXT: retq
-; AVX-NEXT: .LBB2_2: # %endif-block
-; AVX-NEXT: movl $1, %eax
-; AVX-NEXT: vzeroupper
-; AVX-NEXT: retq
+; AVX1-LABEL: veccond512:
+; AVX1: # %bb.0: # %entry
+; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT: vptest %ymm0, %ymm0
+; AVX1-NEXT: je .LBB2_2
+; AVX1-NEXT: # %bb.1: # %if-true-block
+; AVX1-NEXT: xorl %eax, %eax
+; AVX1-NEXT: vzeroupper
+; AVX1-NEXT: retq
+; AVX1-NEXT: .LBB2_2: # %endif-block
+; AVX1-NEXT: movl $1, %eax
+; AVX1-NEXT: vzeroupper
+; AVX1-NEXT: retq
+;
+; AVX512-LABEL: veccond512:
+; AVX512: # %bb.0: # %entry
+; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm1
+; AVX512-NEXT: vmovq %xmm1, %rax
+; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX512-NEXT: vmovq %xmm2, %rcx
+; AVX512-NEXT: orq %rax, %rcx
+; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm3
+; AVX512-NEXT: vmovq %xmm3, %rax
+; AVX512-NEXT: orq %rcx, %rax
+; AVX512-NEXT: vmovq %xmm0, %rcx
+; AVX512-NEXT: orq %rax, %rcx
+; AVX512-NEXT: vpextrq $1, %xmm1, %rax
+; AVX512-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512-NEXT: orq %rax, %rdx
+; AVX512-NEXT: vpextrq $1, %xmm3, %rax
+; AVX512-NEXT: orq %rdx, %rax
+; AVX512-NEXT: vpextrq $1, %xmm0, %rdx
+; AVX512-NEXT: orq %rax, %rdx
+; AVX512-NEXT: orq %rcx, %rdx
+; AVX512-NEXT: je .LBB2_2
+; AVX512-NEXT: # %bb.1: # %if-true-block
+; AVX512-NEXT: xorl %eax, %eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
+; AVX512-NEXT: .LBB2_2: # %endif-block
+; AVX512-NEXT: movl $1, %eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
entry:
%0 = bitcast <16 x i32> %input to i512
%1 = icmp ne i512 %0, 0
@@ -160,14 +191,39 @@ define i32 @vectest512(<16 x i32> %input) {
; SSE41-NEXT: setne %al
; SSE41-NEXT: retq
;
-; AVX-LABEL: vectest512:
-; AVX: # %bb.0:
-; AVX-NEXT: vorps %ymm1, %ymm0, %ymm0
-; AVX-NEXT: xorl %eax, %eax
-; AVX-NEXT: vptest %ymm0, %ymm0
-; AVX-NEXT: setne %al
-; AVX-NEXT: vzeroupper
-; AVX-NEXT: retq
+; AVX1-LABEL: vectest512:
+; AVX1: # %bb.0:
+; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT: xorl %eax, %eax
+; AVX1-NEXT: vptest %ymm0, %ymm0
+; AVX1-NEXT: setne %al
+; AVX1-NEXT: vzeroupper
+; AVX1-NEXT: retq
+;
+; AVX512-LABEL: vectest512:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm1
+; AVX512-NEXT: vmovq %xmm1, %rax
+; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX512-NEXT: vmovq %xmm2, %rcx
+; AVX512-NEXT: orq %rax, %rcx
+; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm3
+; AVX512-NEXT: vmovq %xmm3, %rax
+; AVX512-NEXT: orq %rcx, %rax
+; AVX512-NEXT: vmovq %xmm0, %rcx
+; AVX512-NEXT: orq %rax, %rcx
+; AVX512-NEXT: vpextrq $1, %xmm1, %rax
+; AVX512-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512-NEXT: orq %rax, %rdx
+; AVX512-NEXT: vpextrq $1, %xmm3, %rax
+; AVX512-NEXT: orq %rdx, %rax
+; AVX512-NEXT: vpextrq $1, %xmm0, %rdx
+; AVX512-NEXT: orq %rax, %rdx
+; AVX512-NEXT: xorl %eax, %eax
+; AVX512-NEXT: orq %rcx, %rdx
+; AVX512-NEXT: setne %al
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%t0 = bitcast <16 x i32> %input to i512
%t1 = icmp ne i512 %t0, 0
%t2 = zext i1 %t1 to i32
@@ -227,14 +283,39 @@ define i32 @vecsel512(<16 x i32> %input, i32 %a, i32 %b) {
; SSE41-NEXT: movl %edi, %eax
; SSE41-NEXT: retq
;
-; AVX-LABEL: vecsel512:
-; AVX: # %bb.0:
-; AVX-NEXT: vorps %ymm1, %ymm0, %ymm0
-; AVX-NEXT: vptest %ymm0, %ymm0
-; AVX-NEXT: cmovel %esi, %edi
-; AVX-NEXT: movl %edi, %eax
-; AVX-NEXT: vzeroupper
-; AVX-NEXT: retq
+; AVX1-LABEL: vecsel512:
+; AVX1: # %bb.0:
+; AVX1-NEXT: vorps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT: vptest %ymm0, %ymm0
+; AVX1-NEXT: cmovel %esi, %edi
+; AVX1-NEXT: movl %edi, %eax
+; AVX1-NEXT: vzeroupper
+; AVX1-NEXT: retq
+;
+; AVX512-LABEL: vecsel512:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm1
+; AVX512-NEXT: vmovq %xmm1, %rax
+; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX512-NEXT: vmovq %xmm2, %rcx
+; AVX512-NEXT: orq %rax, %rcx
+; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm3
+; AVX512-NEXT: vmovq %xmm3, %rax
+; AVX512-NEXT: orq %rcx, %rax
+; AVX512-NEXT: vmovq %xmm0, %rcx
+; AVX512-NEXT: orq %rax, %rcx
+; AVX512-NEXT: vpextrq $1, %xmm1, %rax
+; AVX512-NEXT: vpextrq $1, %xmm2, %rdx
+; AVX512-NEXT: orq %rax, %rdx
+; AVX512-NEXT: vpextrq $1, %xmm3, %rax
+; AVX512-NEXT: orq %rdx, %rax
+; AVX512-NEXT: vpextrq $1, %xmm0, %rdx
+; AVX512-NEXT: orq %rax, %rdx
+; AVX512-NEXT: orq %rcx, %rdx
+; AVX512-NEXT: cmovel %esi, %edi
+; AVX512-NEXT: movl %edi, %eax
+; AVX512-NEXT: vzeroupper
+; AVX512-NEXT: retq
%t0 = bitcast <16 x i32> %input to i512
%t1 = icmp ne i512 %t0, 0
%t2 = select i1 %t1, i32 %a, i32 %b
OpenPOWER on IntegriCloud