From 2cf3769a7e7fa5aba6a43f3a91f121d6d447762b Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Tue, 20 Feb 2018 18:44:21 +0000
Subject: [PATCH] [X86][3DNow] Regenerate intrinsics tests

llvm-svn: 325609
---
 llvm/test/CodeGen/X86/3dnow-intrinsics.ll | 719 ++++++++++++++++++++++++++++--
 1 file changed, 693 insertions(+), 26 deletions(-)

diff --git a/llvm/test/CodeGen/X86/3dnow-intrinsics.ll b/llvm/test/CodeGen/X86/3dnow-intrinsics.ll
index 6817df4..98e21f3 100644
--- a/llvm/test/CodeGen/X86/3dnow-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/3dnow-intrinsics.ll
@@ -1,7 +1,22 @@
-; RUN: llc < %s -mtriple=i686-- -mattr=+3dnow | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+3dnow | FileCheck %s --check-prefixes=CHECK,X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+3dnow | FileCheck %s --check-prefixes=CHECK,X64
 
 define <8 x i8> @test_pavgusb(x86_mmx %a.coerce, x86_mmx %b.coerce) nounwind readnone {
-; CHECK: pavgusb
+; X86-LABEL: test_pavgusb:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    pavgusb %mm1, %mm0
+; X86-NEXT:    movq %mm0, (%eax)
+; X86-NEXT:    retl $4
+;
+; X64-LABEL: test_pavgusb:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    pavgusb %mm1, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast x86_mmx %a.coerce to <8 x i8>
   %1 = bitcast x86_mmx %b.coerce to <8 x i8>
@@ -15,7 +30,32 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pavgusb(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x i32> @test_pf2id(<2 x float> %a) nounwind readnone {
-; CHECK: pf2id
+; X86-LABEL: test_pf2id:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $16, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    pf2id {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    movl (%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pf2id:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm0, %mm0
+; X64-NEXT:    pf2id %mm0, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = tail call x86_mmx @llvm.x86.3dnow.pf2id(x86_mmx %0)
@@ -26,7 +66,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pf2id(x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfacc(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfacc
+; X86-LABEL: test_pfacc:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfacc {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfacc:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfacc %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -38,7 +108,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfacc(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfadd(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfadd
+; X86-LABEL: test_pfadd:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfadd {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfadd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfadd %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -50,7 +150,38 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfadd(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x i32> @test_pfcmpeq(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfcmpeq
+; X86-LABEL: test_pfcmpeq:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfcmpeq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    movl (%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfcmpeq:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfcmpeq %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -62,7 +193,38 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfcmpeq(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x i32> @test_pfcmpge(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfcmpge
+; X86-LABEL: test_pfcmpge:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfcmpge {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    movl (%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfcmpge:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfcmpge %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -74,7 +236,38 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfcmpge(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x i32> @test_pfcmpgt(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfcmpgt
+; X86-LABEL: test_pfcmpgt:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfcmpgt {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    movl (%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfcmpgt:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfcmpgt %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -86,7 +279,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfcmpgt(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfmax(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfmax
+; X86-LABEL: test_pfmax:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfmax {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfmax:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfmax %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -98,7 +321,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfmax(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfmin(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfmin
+; X86-LABEL: test_pfmin:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfmin {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfmin:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfmin %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -110,7 +363,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfmin(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfmul(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfmul
+; X86-LABEL: test_pfmul:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfmul {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfmul:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfmul %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -122,7 +405,31 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfmul(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfrcp(<2 x float> %a) nounwind readnone {
-; CHECK: pfrcp
+; X86-LABEL: test_pfrcp:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $16, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    pfrcp {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfrcp:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm0, %mm0
+; X64-NEXT:    pfrcp %mm0, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = tail call x86_mmx @llvm.x86.3dnow.pfrcp(x86_mmx %0)
@@ -133,7 +440,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfrcp(x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfrcpit1(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfrcpit1
+; X86-LABEL: test_pfrcpit1:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfrcpit1 {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfrcpit1:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfrcpit1 %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -145,7 +482,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfrcpit1(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfrcpit2(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfrcpit2
+; X86-LABEL: test_pfrcpit2:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfrcpit2 {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfrcpit2:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfrcpit2 %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -157,7 +524,31 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfrcpit2(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfrsqrt(<2 x float> %a) nounwind readnone {
-; CHECK: pfrsqrt
+; X86-LABEL: test_pfrsqrt:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $16, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    pfrsqrt {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfrsqrt:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm0, %mm0
+; X64-NEXT:    pfrsqrt %mm0, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = tail call x86_mmx @llvm.x86.3dnow.pfrsqrt(x86_mmx %0)
@@ -168,7 +559,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfrsqrt(x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfrsqit1(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfrsqit1
+; X86-LABEL: test_pfrsqit1:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfrsqit1 {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfrsqit1:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfrsqit1 %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -180,7 +601,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfrsqit1(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfsub(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfsub
+; X86-LABEL: test_pfsub:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfsub {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfsub:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfsub %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -192,7 +643,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfsub(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfsubr(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfsubr
+; X86-LABEL: test_pfsubr:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfsubr {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfsubr:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfsubr %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -204,7 +685,26 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pfsubr(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pi2fd(x86_mmx %a.coerce) nounwind readnone {
-; CHECK: pi2fd
+; X86-LABEL: test_pi2fd:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $8, %esp
+; X86-NEXT:    pi2fd %mm0, %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pi2fd:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    pi2fd %mm0, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast x86_mmx %a.coerce to <2 x i32>
   %1 = bitcast <2 x i32> %0 to x86_mmx
@@ -216,7 +716,20 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pi2fd(x86_mmx) nounwind readnone
 
 define <4 x i16> @test_pmulhrw(x86_mmx %a.coerce, x86_mmx %b.coerce) nounwind readnone {
-; CHECK: pmulhrw
+; X86-LABEL: test_pmulhrw:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    pmulhrw %mm1, %mm0
+; X86-NEXT:    movq %mm0, (%eax)
+; X86-NEXT:    retl $4
+;
+; X64-LABEL: test_pmulhrw:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    pmulhrw %mm1, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast x86_mmx %a.coerce to <4 x i16>
   %1 = bitcast x86_mmx %b.coerce to <4 x i16>
@@ -230,7 +743,32 @@ entry:
 declare x86_mmx @llvm.x86.3dnow.pmulhrw(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x i32> @test_pf2iw(<2 x float> %a) nounwind readnone {
-; CHECK: pf2iw
+; X86-LABEL: test_pf2iw:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $16, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    pf2iw {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    movl (%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pf2iw:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm0, %mm0
+; X64-NEXT:    pf2iw %mm0, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = tail call x86_mmx @llvm.x86.3dnowa.pf2iw(x86_mmx %0)
@@ -241,7 +779,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnowa.pf2iw(x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfnacc(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfnacc
+; X86-LABEL: test_pfnacc:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfnacc {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfnacc:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfnacc %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -253,7 +821,37 @@ entry:
 declare x86_mmx @llvm.x86.3dnowa.pfnacc(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pfpnacc(<2 x float> %a, <2 x float> %b) nounwind readnone {
-; CHECK: pfpnacc
+; X86-LABEL: test_pfpnacc:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 20(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 16(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    movq {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    pfpnacc {{[0-9]+}}(%esp), %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pfpnacc:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm1, %mm0
+; X64-NEXT:    movdq2q %xmm0, %mm1
+; X64-NEXT:    pfpnacc %mm0, %mm1
+; X64-NEXT:    movq %mm1, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = bitcast <2 x float> %b to x86_mmx
@@ -265,7 +863,26 @@ entry:
 declare x86_mmx @llvm.x86.3dnowa.pfpnacc(x86_mmx, x86_mmx) nounwind readnone
 
 define <2 x float> @test_pi2fw(x86_mmx %a.coerce) nounwind readnone {
-; CHECK: pi2fw
+; X86-LABEL: test_pi2fw:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $8, %esp
+; X86-NEXT:    pi2fw %mm0, %mm0
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pi2fw:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    pi2fw %mm0, %mm0
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast x86_mmx %a.coerce to <2 x i32>
   %1 = bitcast <2 x i32> %0 to x86_mmx
@@ -277,7 +894,31 @@ entry:
 declare x86_mmx @llvm.x86.3dnowa.pi2fw(x86_mmx) nounwind readnone
 
 define <2 x float> @test_pswapdsf(<2 x float> %a) nounwind readnone {
-; CHECK: pswapd {{.*#+}} mm0 = mem[1,0]
+; X86-LABEL: test_pswapdsf:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $16, %esp
+; X86-NEXT:    flds 12(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    flds 8(%ebp)
+; X86-NEXT:    fstps {{[0-9]+}}(%esp)
+; X86-NEXT:    pswapd {{[0-9]+}}(%esp), %mm0 # mm0 = mem[1,0]
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    flds {{[0-9]+}}(%esp)
+; X86-NEXT:    flds (%esp)
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pswapdsf:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    movdq2q %xmm0, %mm0
+; X64-NEXT:    pswapd %mm0, %mm0 # mm0 = mm0[1,0]
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movaps -{{[0-9]+}}(%rsp), %xmm0
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x float> %a to x86_mmx
   %1 = tail call x86_mmx @llvm.x86.3dnowa.pswapd(x86_mmx %0)
@@ -286,7 +927,33 @@ entry:
 }
 
 define <2 x i32> @test_pswapdsi(<2 x i32> %a) nounwind readnone {
-; CHECK: pswapd {{.*#+}} mm0 = mem[1,0]
+; X86-LABEL: test_pswapdsi:
+; X86:       # %bb.0: # %entry
+; X86-NEXT:    pushl %ebp
+; X86-NEXT:    movl %esp, %ebp
+; X86-NEXT:    andl $-8, %esp
+; X86-NEXT:    subl $16, %esp
+; X86-NEXT:    movl 12(%ebp), %eax
+; X86-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT:    movl 8(%ebp), %eax
+; X86-NEXT:    movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT:    pswapd {{[0-9]+}}(%esp), %mm0 # mm0 = mem[1,0]
+; X86-NEXT:    movq %mm0, (%esp)
+; X86-NEXT:    movl (%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT:    movl %ebp, %esp
+; X86-NEXT:    popl %ebp
+; X86-NEXT:    retl
+;
+; X64-LABEL: test_pswapdsi:
+; X64:       # %bb.0: # %entry
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; X64-NEXT:    movq %xmm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    pswapd -{{[0-9]+}}(%rsp), %mm0 # mm0 = mem[1,0]
+; X64-NEXT:    movq %mm0, -{{[0-9]+}}(%rsp)
+; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; X64-NEXT:    retq
 entry:
   %0 = bitcast <2 x i32> %a to x86_mmx
   %1 = tail call x86_mmx @llvm.x86.3dnowa.pswapd(x86_mmx %0)
-- 
2.7.4