summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/widen_conv-4.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/widen_conv-4.ll')
-rw-r--r--llvm/test/CodeGen/X86/widen_conv-4.ll96
1 files changed, 53 insertions, 43 deletions
diff --git a/llvm/test/CodeGen/X86/widen_conv-4.ll b/llvm/test/CodeGen/X86/widen_conv-4.ll
index 9c567fc3841..e807d6f4043 100644
--- a/llvm/test/CodeGen/X86/widen_conv-4.ll
+++ b/llvm/test/CodeGen/X86/widen_conv-4.ll
@@ -28,15 +28,15 @@ define void @convert_v7i16_v7f32(<7 x float>* %dst.addr, <7 x i16> %src) nounwin
; X86-SSE42-LABEL: convert_v7i16_v7f32:
; X86-SSE42: # %bb.0: # %entry
; X86-SSE42-NEXT: movl {{[0-9]+}}(%esp), %eax
-; X86-SSE42-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X86-SSE42-NEXT: pmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; X86-SSE42-NEXT: cvtdq2ps %xmm1, %xmm1
-; X86-SSE42-NEXT: pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; X86-SSE42-NEXT: pxor %xmm1, %xmm1
+; X86-SSE42-NEXT: pmovzxwd {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; X86-SSE42-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
; X86-SSE42-NEXT: cvtdq2ps %xmm0, %xmm0
-; X86-SSE42-NEXT: movups %xmm0, (%eax)
-; X86-SSE42-NEXT: extractps $2, %xmm1, 24(%eax)
-; X86-SSE42-NEXT: extractps $1, %xmm1, 20(%eax)
-; X86-SSE42-NEXT: movss %xmm1, 16(%eax)
+; X86-SSE42-NEXT: cvtdq2ps %xmm2, %xmm1
+; X86-SSE42-NEXT: extractps $2, %xmm0, 24(%eax)
+; X86-SSE42-NEXT: extractps $1, %xmm0, 20(%eax)
+; X86-SSE42-NEXT: movups %xmm1, (%eax)
+; X86-SSE42-NEXT: movss %xmm0, 16(%eax)
; X86-SSE42-NEXT: retl
;
; X64-SSE2-LABEL: convert_v7i16_v7f32:
@@ -55,14 +55,14 @@ define void @convert_v7i16_v7f32(<7 x float>* %dst.addr, <7 x i16> %src) nounwin
;
; X64-SSE42-LABEL: convert_v7i16_v7f32:
; X64-SSE42: # %bb.0: # %entry
-; X64-SSE42-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X64-SSE42-NEXT: pmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; X64-SSE42-NEXT: cvtdq2ps %xmm1, %xmm1
-; X64-SSE42-NEXT: pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; X64-SSE42-NEXT: pxor %xmm1, %xmm1
+; X64-SSE42-NEXT: pmovzxwd {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; X64-SSE42-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
; X64-SSE42-NEXT: cvtdq2ps %xmm0, %xmm0
-; X64-SSE42-NEXT: movups %xmm0, (%rdi)
-; X64-SSE42-NEXT: extractps $2, %xmm1, 24(%rdi)
-; X64-SSE42-NEXT: movlps %xmm1, 16(%rdi)
+; X64-SSE42-NEXT: cvtdq2ps %xmm2, %xmm1
+; X64-SSE42-NEXT: extractps $2, %xmm0, 24(%rdi)
+; X64-SSE42-NEXT: movlps %xmm0, 16(%rdi)
+; X64-SSE42-NEXT: movups %xmm1, (%rdi)
; X64-SSE42-NEXT: retq
entry:
%val = uitofp <7 x i16> %src to <7 x float>
@@ -75,58 +75,66 @@ entry:
define void @convert_v3i8_to_v3f32(<3 x float>* %dst.addr, <3 x i8>* %src.addr) nounwind {
; X86-SSE2-LABEL: convert_v3i8_to_v3f32:
; X86-SSE2: # %bb.0: # %entry
-; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
-; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; X86-SSE2-NEXT: pushl %ebp
+; X86-SSE2-NEXT: movl %esp, %ebp
+; X86-SSE2-NEXT: andl $-16, %esp
+; X86-SSE2-NEXT: subl $32, %esp
+; X86-SSE2-NEXT: movl 8(%ebp), %eax
+; X86-SSE2-NEXT: movl 12(%ebp), %ecx
; X86-SSE2-NEXT: movzwl (%ecx), %edx
; X86-SSE2-NEXT: movd %edx, %xmm0
-; X86-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; X86-SSE2-NEXT: pand %xmm1, %xmm0
+; X86-SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X86-SSE2-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
; X86-SSE2-NEXT: movzbl 2(%ecx), %ecx
-; X86-SSE2-NEXT: movd %ecx, %xmm2
-; X86-SSE2-NEXT: pslld $16, %xmm2
-; X86-SSE2-NEXT: pandn %xmm2, %xmm1
-; X86-SSE2-NEXT: por %xmm0, %xmm1
-; X86-SSE2-NEXT: pxor %xmm0, %xmm0
-; X86-SSE2-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; X86-SSE2-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X86-SSE2-NEXT: cvtdq2ps %xmm1, %xmm0
+; X86-SSE2-NEXT: movdqa %xmm0, (%esp)
+; X86-SSE2-NEXT: movzbl (%esp), %edx
+; X86-SSE2-NEXT: movd %edx, %xmm0
+; X86-SSE2-NEXT: movzbl {{[0-9]+}}(%esp), %edx
+; X86-SSE2-NEXT: pinsrw $2, %edx, %xmm0
+; X86-SSE2-NEXT: pinsrw $4, %ecx, %xmm0
+; X86-SSE2-NEXT: cvtdq2ps %xmm0, %xmm0
; X86-SSE2-NEXT: movss %xmm0, (%eax)
; X86-SSE2-NEXT: movaps %xmm0, %xmm1
; X86-SSE2-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
; X86-SSE2-NEXT: movss %xmm1, 8(%eax)
; X86-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,2,3]
; X86-SSE2-NEXT: movss %xmm0, 4(%eax)
+; X86-SSE2-NEXT: movl %ebp, %esp
+; X86-SSE2-NEXT: popl %ebp
; X86-SSE2-NEXT: retl
;
; X86-SSE42-LABEL: convert_v3i8_to_v3f32:
; X86-SSE42: # %bb.0: # %entry
+; X86-SSE42-NEXT: pushl %eax
; X86-SSE42-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-SSE42-NEXT: movl {{[0-9]+}}(%esp), %ecx
-; X86-SSE42-NEXT: movzwl (%ecx), %edx
-; X86-SSE42-NEXT: movd %edx, %xmm0
-; X86-SSE42-NEXT: pinsrb $2, 2(%ecx), %xmm0
+; X86-SSE42-NEXT: movzbl 2(%ecx), %edx
+; X86-SSE42-NEXT: movzwl (%ecx), %ecx
+; X86-SSE42-NEXT: movd %ecx, %xmm0
; X86-SSE42-NEXT: pmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; X86-SSE42-NEXT: pinsrd $2, %edx, %xmm0
+; X86-SSE42-NEXT: pand {{\.LCPI.*}}, %xmm0
; X86-SSE42-NEXT: cvtdq2ps %xmm0, %xmm0
; X86-SSE42-NEXT: extractps $2, %xmm0, 8(%eax)
; X86-SSE42-NEXT: extractps $1, %xmm0, 4(%eax)
; X86-SSE42-NEXT: movss %xmm0, (%eax)
+; X86-SSE42-NEXT: popl %eax
; X86-SSE42-NEXT: retl
;
; X64-SSE2-LABEL: convert_v3i8_to_v3f32:
; X64-SSE2: # %bb.0: # %entry
; X64-SSE2-NEXT: movzwl (%rsi), %eax
; X64-SSE2-NEXT: movd %eax, %xmm0
-; X64-SSE2-NEXT: movdqa {{.*#+}} xmm1 = [255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; X64-SSE2-NEXT: pand %xmm1, %xmm0
+; X64-SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X64-SSE2-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
; X64-SSE2-NEXT: movzbl 2(%rsi), %eax
-; X64-SSE2-NEXT: movd %eax, %xmm2
-; X64-SSE2-NEXT: pslld $16, %xmm2
-; X64-SSE2-NEXT: pandn %xmm2, %xmm1
-; X64-SSE2-NEXT: por %xmm0, %xmm1
-; X64-SSE2-NEXT: pxor %xmm0, %xmm0
-; X64-SSE2-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; X64-SSE2-NEXT: punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE2-NEXT: cvtdq2ps %xmm1, %xmm0
+; X64-SSE2-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp)
+; X64-SSE2-NEXT: movzbl -{{[0-9]+}}(%rsp), %ecx
+; X64-SSE2-NEXT: movd %ecx, %xmm0
+; X64-SSE2-NEXT: movzbl -{{[0-9]+}}(%rsp), %ecx
+; X64-SSE2-NEXT: pinsrw $2, %ecx, %xmm0
+; X64-SSE2-NEXT: pinsrw $4, %eax, %xmm0
+; X64-SSE2-NEXT: cvtdq2ps %xmm0, %xmm0
; X64-SSE2-NEXT: movlps %xmm0, (%rdi)
; X64-SSE2-NEXT: movhlps {{.*#+}} xmm0 = xmm0[1,1]
; X64-SSE2-NEXT: movss %xmm0, 8(%rdi)
@@ -134,10 +142,12 @@ define void @convert_v3i8_to_v3f32(<3 x float>* %dst.addr, <3 x i8>* %src.addr)
;
; X64-SSE42-LABEL: convert_v3i8_to_v3f32:
; X64-SSE42: # %bb.0: # %entry
-; X64-SSE42-NEXT: movzwl (%rsi), %eax
-; X64-SSE42-NEXT: movd %eax, %xmm0
-; X64-SSE42-NEXT: pinsrb $2, 2(%rsi), %xmm0
+; X64-SSE42-NEXT: movzbl 2(%rsi), %eax
+; X64-SSE42-NEXT: movzwl (%rsi), %ecx
+; X64-SSE42-NEXT: movd %ecx, %xmm0
; X64-SSE42-NEXT: pmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; X64-SSE42-NEXT: pinsrd $2, %eax, %xmm0
+; X64-SSE42-NEXT: pand {{.*}}(%rip), %xmm0
; X64-SSE42-NEXT: cvtdq2ps %xmm0, %xmm0
; X64-SSE42-NEXT: extractps $2, %xmm0, 8(%rdi)
; X64-SSE42-NEXT: movlps %xmm0, (%rdi)
OpenPOWER on IntegriCloud