summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/subvector-broadcast.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/subvector-broadcast.ll')
-rw-r--r--llvm/test/CodeGen/X86/subvector-broadcast.ll48
1 files changed, 32 insertions, 16 deletions
diff --git a/llvm/test/CodeGen/X86/subvector-broadcast.ll b/llvm/test/CodeGen/X86/subvector-broadcast.ll
index a4477b2375b..bcb7d14f953 100644
--- a/llvm/test/CodeGen/X86/subvector-broadcast.ll
+++ b/llvm/test/CodeGen/X86/subvector-broadcast.ll
@@ -751,64 +751,72 @@ define <8 x i32> @test_broadcast_4i32_8i32_chain(<4 x i32>* %p0, <4 x float>* %p
; X32-AVX: # %bb.0:
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX-NEXT: vmovaps (%ecx), %xmm0
; X32-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X32-AVX-NEXT: vmovaps %xmm1, (%eax)
+; X32-AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X32-AVX-NEXT: retl
;
; X32-AVX512F-LABEL: test_broadcast_4i32_8i32_chain:
; X32-AVX512F: # %bb.0:
; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX512F-NEXT: vmovaps (%ecx), %xmm0
; X32-AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X32-AVX512F-NEXT: vmovdqa %xmm1, (%eax)
+; X32-AVX512F-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X32-AVX512F-NEXT: retl
;
; X32-AVX512BW-LABEL: test_broadcast_4i32_8i32_chain:
; X32-AVX512BW: # %bb.0:
; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX512BW-NEXT: vmovaps (%ecx), %xmm0
; X32-AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX512BW-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X32-AVX512BW-NEXT: vmovdqa %xmm1, (%eax)
+; X32-AVX512BW-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X32-AVX512BW-NEXT: retl
;
; X32-AVX512DQ-LABEL: test_broadcast_4i32_8i32_chain:
; X32-AVX512DQ: # %bb.0:
; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX512DQ-NEXT: vmovaps (%ecx), %xmm0
; X32-AVX512DQ-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X32-AVX512DQ-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X32-AVX512DQ-NEXT: vmovaps %xmm1, (%eax)
+; X32-AVX512DQ-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X32-AVX512DQ-NEXT: retl
;
; X64-AVX-LABEL: test_broadcast_4i32_8i32_chain:
; X64-AVX: # %bb.0:
+; X64-AVX-NEXT: vmovaps (%rdi), %xmm0
; X64-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X64-AVX-NEXT: vmovaps %xmm1, (%rsi)
+; X64-AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X64-AVX-NEXT: retq
;
; X64-AVX512F-LABEL: test_broadcast_4i32_8i32_chain:
; X64-AVX512F: # %bb.0:
+; X64-AVX512F-NEXT: vmovaps (%rdi), %xmm0
; X64-AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X64-AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X64-AVX512F-NEXT: vmovdqa %xmm1, (%rsi)
+; X64-AVX512F-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X64-AVX512F-NEXT: retq
;
; X64-AVX512BW-LABEL: test_broadcast_4i32_8i32_chain:
; X64-AVX512BW: # %bb.0:
+; X64-AVX512BW-NEXT: vmovaps (%rdi), %xmm0
; X64-AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X64-AVX512BW-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X64-AVX512BW-NEXT: vmovdqa %xmm1, (%rsi)
+; X64-AVX512BW-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X64-AVX512BW-NEXT: retq
;
; X64-AVX512DQ-LABEL: test_broadcast_4i32_8i32_chain:
; X64-AVX512DQ: # %bb.0:
+; X64-AVX512DQ-NEXT: vmovaps (%rdi), %xmm0
; X64-AVX512DQ-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64-AVX512DQ-NEXT: vbroadcasti128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X64-AVX512DQ-NEXT: vmovaps %xmm1, (%rsi)
+; X64-AVX512DQ-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X64-AVX512DQ-NEXT: retq
%1 = load <4 x i32>, <4 x i32>* %p0
store <4 x float> zeroinitializer, <4 x float>* %p1
@@ -821,9 +829,10 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>*
; X32-AVX: # %bb.0:
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX-NEXT: vmovaps (%ecx), %xmm0
; X32-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X32-AVX-NEXT: vmovaps %xmm1, (%eax)
+; X32-AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X32-AVX-NEXT: vmovaps %ymm0, %ymm1
; X32-AVX-NEXT: retl
;
@@ -831,56 +840,63 @@ define <16 x i32> @test_broadcast_4i32_16i32_chain(<4 x i32>* %p0, <4 x float>*
; X32-AVX512F: # %bb.0:
; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX512F-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX512F-NEXT: vmovdqa (%ecx), %xmm0
; X32-AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX512F-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X32-AVX512F-NEXT: vmovdqa %xmm1, (%eax)
+; X32-AVX512F-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X32-AVX512F-NEXT: retl
;
; X32-AVX512BW-LABEL: test_broadcast_4i32_16i32_chain:
; X32-AVX512BW: # %bb.0:
; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX512BW-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX512BW-NEXT: vmovdqa (%ecx), %xmm0
; X32-AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X32-AVX512BW-NEXT: vmovdqa %xmm1, (%eax)
+; X32-AVX512BW-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X32-AVX512BW-NEXT: retl
;
; X32-AVX512DQ-LABEL: test_broadcast_4i32_16i32_chain:
; X32-AVX512DQ: # %bb.0:
; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-AVX512DQ-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X32-AVX512DQ-NEXT: vmovdqa (%ecx), %xmm0
; X32-AVX512DQ-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X32-AVX512DQ-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X32-AVX512DQ-NEXT: vmovaps %xmm1, (%eax)
+; X32-AVX512DQ-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X32-AVX512DQ-NEXT: retl
;
; X64-AVX-LABEL: test_broadcast_4i32_16i32_chain:
; X64-AVX: # %bb.0:
+; X64-AVX-NEXT: vmovaps (%rdi), %xmm0
; X64-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64-AVX-NEXT: vbroadcastf128 {{.*#+}} ymm0 = mem[0,1,0,1]
; X64-AVX-NEXT: vmovaps %xmm1, (%rsi)
+; X64-AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0
; X64-AVX-NEXT: vmovaps %ymm0, %ymm1
; X64-AVX-NEXT: retq
;
; X64-AVX512F-LABEL: test_broadcast_4i32_16i32_chain:
; X64-AVX512F: # %bb.0:
+; X64-AVX512F-NEXT: vmovdqa (%rdi), %xmm0
; X64-AVX512F-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X64-AVX512F-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X64-AVX512F-NEXT: vmovdqa %xmm1, (%rsi)
+; X64-AVX512F-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X64-AVX512F-NEXT: retq
;
; X64-AVX512BW-LABEL: test_broadcast_4i32_16i32_chain:
; X64-AVX512BW: # %bb.0:
+; X64-AVX512BW-NEXT: vmovdqa (%rdi), %xmm0
; X64-AVX512BW-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; X64-AVX512BW-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X64-AVX512BW-NEXT: vmovdqa %xmm1, (%rsi)
+; X64-AVX512BW-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X64-AVX512BW-NEXT: retq
;
; X64-AVX512DQ-LABEL: test_broadcast_4i32_16i32_chain:
; X64-AVX512DQ: # %bb.0:
+; X64-AVX512DQ-NEXT: vmovdqa (%rdi), %xmm0
; X64-AVX512DQ-NEXT: vxorps %xmm1, %xmm1, %xmm1
-; X64-AVX512DQ-NEXT: vbroadcasti32x4 {{.*#+}} zmm0 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X64-AVX512DQ-NEXT: vmovaps %xmm1, (%rsi)
+; X64-AVX512DQ-NEXT: vshufi32x4 {{.*#+}} zmm0 = zmm0[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
; X64-AVX512DQ-NEXT: retq
%1 = load <4 x i32>, <4 x i32>* %p0
store <4 x float> zeroinitializer, <4 x float>* %p1
OpenPOWER on IntegriCloud