summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/horizontal-reduce-umax.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/horizontal-reduce-umax.ll')
-rw-r--r--llvm/test/CodeGen/X86/horizontal-reduce-umax.ll110
1 files changed, 48 insertions, 62 deletions
diff --git a/llvm/test/CodeGen/X86/horizontal-reduce-umax.ll b/llvm/test/CodeGen/X86/horizontal-reduce-umax.ll
index a4888e1cd3f..88f6b01131a 100644
--- a/llvm/test/CodeGen/X86/horizontal-reduce-umax.ll
+++ b/llvm/test/CodeGen/X86/horizontal-reduce-umax.ll
@@ -230,15 +230,14 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
; X86-SSE2-NEXT: pxor %xmm2, %xmm0
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
-; X86-SSE2-NEXT: pxor %xmm2, %xmm1
-; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; X86-SSE2-NEXT: pxor %xmm2, %xmm1
+; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
; X86-SSE2-NEXT: pxor %xmm2, %xmm0
-; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; X86-SSE2-NEXT: pxor %xmm2, %xmm0
+; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT: pxor %xmm2, %xmm1
; X86-SSE2-NEXT: psrld $16, %xmm1
-; X86-SSE2-NEXT: pxor %xmm2, %xmm0
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
@@ -273,15 +272,14 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
; X64-SSE2-NEXT: pxor %xmm2, %xmm0
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
-; X64-SSE2-NEXT: pxor %xmm2, %xmm1
-; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
-; X64-SSE2-NEXT: pxor %xmm2, %xmm1
+; X64-SSE2-NEXT: movdqa %xmm1, %xmm0
; X64-SSE2-NEXT: pxor %xmm2, %xmm0
-; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; X64-SSE2-NEXT: pxor %xmm2, %xmm0
+; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT: pxor %xmm2, %xmm1
; X64-SSE2-NEXT: psrld $16, %xmm1
-; X64-SSE2-NEXT: pxor %xmm2, %xmm0
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
@@ -832,20 +830,19 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
; X86-SSE2-NEXT: pxor %xmm2, %xmm0
; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
-; X86-SSE2-NEXT: pxor %xmm2, %xmm0
-; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X86-SSE2-NEXT: pxor %xmm2, %xmm0
-; X86-SSE2-NEXT: pxor %xmm2, %xmm1
-; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
-; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
+; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
; X86-SSE2-NEXT: pxor %xmm2, %xmm0
-; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; X86-SSE2-NEXT: pxor %xmm2, %xmm0
+; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT: pxor %xmm2, %xmm1
; X86-SSE2-NEXT: psrld $16, %xmm1
-; X86-SSE2-NEXT: pxor %xmm2, %xmm0
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
; X86-SSE2-NEXT: pxor %xmm2, %xmm1
@@ -896,20 +893,19 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
; X64-SSE2-NEXT: pxor %xmm2, %xmm0
; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
-; X64-SSE2-NEXT: pxor %xmm2, %xmm0
-; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; X64-SSE2-NEXT: pxor %xmm2, %xmm0
-; X64-SSE2-NEXT: pxor %xmm2, %xmm1
-; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
-; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
+; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT: movdqa %xmm1, %xmm0
; X64-SSE2-NEXT: pxor %xmm2, %xmm0
-; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; X64-SSE2-NEXT: pxor %xmm2, %xmm0
+; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT: pxor %xmm2, %xmm1
; X64-SSE2-NEXT: psrld $16, %xmm1
-; X64-SSE2-NEXT: pxor %xmm2, %xmm0
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
; X64-SSE2-NEXT: pxor %xmm2, %xmm1
@@ -1670,35 +1666,30 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
; X86-SSE2-LABEL: test_reduce_v32i16:
; X86-SSE2: ## %bb.0:
; X86-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-SSE2-NEXT: pxor %xmm4, %xmm2
-; X86-SSE2-NEXT: pxor %xmm4, %xmm0
-; X86-SSE2-NEXT: pmaxsw %xmm2, %xmm0
; X86-SSE2-NEXT: pxor %xmm4, %xmm3
; X86-SSE2-NEXT: pxor %xmm4, %xmm1
; X86-SSE2-NEXT: pmaxsw %xmm3, %xmm1
-; X86-SSE2-NEXT: movdqa %xmm4, %xmm2
-; X86-SSE2-NEXT: pxor %xmm4, %xmm2
-; X86-SSE2-NEXT: pxor %xmm2, %xmm1
-; X86-SSE2-NEXT: pxor %xmm0, %xmm2
-; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm2
-; X86-SSE2-NEXT: pxor %xmm4, %xmm2
-; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
; X86-SSE2-NEXT: pxor %xmm4, %xmm2
; X86-SSE2-NEXT: pxor %xmm4, %xmm0
; X86-SSE2-NEXT: pmaxsw %xmm2, %xmm0
-; X86-SSE2-NEXT: pxor %xmm4, %xmm0
-; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; X86-SSE2-NEXT: pxor %xmm4, %xmm0
+; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
; X86-SSE2-NEXT: pxor %xmm4, %xmm1
-; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; X86-SSE2-NEXT: pxor %xmm4, %xmm1
+; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
; X86-SSE2-NEXT: movdqa %xmm1, %xmm0
-; X86-SSE2-NEXT: psrld $16, %xmm0
-; X86-SSE2-NEXT: pxor %xmm4, %xmm1
; X86-SSE2-NEXT: pxor %xmm4, %xmm0
-; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; X86-SSE2-NEXT: pxor %xmm4, %xmm0
-; X86-SSE2-NEXT: movd %xmm0, %eax
+; X86-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X86-SSE2-NEXT: movdqa %xmm0, %xmm1
+; X86-SSE2-NEXT: pxor %xmm4, %xmm1
+; X86-SSE2-NEXT: psrld $16, %xmm1
+; X86-SSE2-NEXT: pxor %xmm4, %xmm1
+; X86-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X86-SSE2-NEXT: pxor %xmm4, %xmm1
+; X86-SSE2-NEXT: movd %xmm1, %eax
; X86-SSE2-NEXT: ## kill: def $ax killed $ax killed $eax
; X86-SSE2-NEXT: retl
;
@@ -1748,35 +1739,30 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
; X64-SSE2-LABEL: test_reduce_v32i16:
; X64-SSE2: ## %bb.0:
; X64-SSE2-NEXT: movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-SSE2-NEXT: pxor %xmm4, %xmm2
-; X64-SSE2-NEXT: pxor %xmm4, %xmm0
-; X64-SSE2-NEXT: pmaxsw %xmm2, %xmm0
; X64-SSE2-NEXT: pxor %xmm4, %xmm3
; X64-SSE2-NEXT: pxor %xmm4, %xmm1
; X64-SSE2-NEXT: pmaxsw %xmm3, %xmm1
-; X64-SSE2-NEXT: movdqa %xmm4, %xmm2
-; X64-SSE2-NEXT: pxor %xmm4, %xmm2
-; X64-SSE2-NEXT: pxor %xmm2, %xmm1
-; X64-SSE2-NEXT: pxor %xmm0, %xmm2
-; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm2
-; X64-SSE2-NEXT: pxor %xmm4, %xmm2
-; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
; X64-SSE2-NEXT: pxor %xmm4, %xmm2
; X64-SSE2-NEXT: pxor %xmm4, %xmm0
; X64-SSE2-NEXT: pmaxsw %xmm2, %xmm0
-; X64-SSE2-NEXT: pxor %xmm4, %xmm0
-; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; X64-SSE2-NEXT: pxor %xmm4, %xmm0
+; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
; X64-SSE2-NEXT: pxor %xmm4, %xmm1
-; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
; X64-SSE2-NEXT: pxor %xmm4, %xmm1
+; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
; X64-SSE2-NEXT: movdqa %xmm1, %xmm0
-; X64-SSE2-NEXT: psrld $16, %xmm0
-; X64-SSE2-NEXT: pxor %xmm4, %xmm1
; X64-SSE2-NEXT: pxor %xmm4, %xmm0
-; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
; X64-SSE2-NEXT: pxor %xmm4, %xmm0
-; X64-SSE2-NEXT: movd %xmm0, %eax
+; X64-SSE2-NEXT: pmaxsw %xmm1, %xmm0
+; X64-SSE2-NEXT: movdqa %xmm0, %xmm1
+; X64-SSE2-NEXT: pxor %xmm4, %xmm1
+; X64-SSE2-NEXT: psrld $16, %xmm1
+; X64-SSE2-NEXT: pxor %xmm4, %xmm1
+; X64-SSE2-NEXT: pmaxsw %xmm0, %xmm1
+; X64-SSE2-NEXT: pxor %xmm4, %xmm1
+; X64-SSE2-NEXT: movd %xmm1, %eax
; X64-SSE2-NEXT: ## kill: def $ax killed $ax killed $eax
; X64-SSE2-NEXT: retq
;
OpenPOWER on IntegriCloud