1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown -mattr=+mmx,+sse2 | FileCheck %s --check-prefix=X32
3 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+mmx,+sse2 | FileCheck %s --check-prefix=X64
5 define i32 @test0(<1 x i64>* %v4) nounwind {
7 ; X32: # %bb.0: # %entry
9 ; X32-NEXT: movl %esp, %ebp
10 ; X32-NEXT: andl $-8, %esp
11 ; X32-NEXT: subl $8, %esp
12 ; X32-NEXT: movl 8(%ebp), %eax
13 ; X32-NEXT: movl (%eax), %ecx
14 ; X32-NEXT: movl 4(%eax), %eax
15 ; X32-NEXT: movl %eax, {{[0-9]+}}(%esp)
16 ; X32-NEXT: movl %ecx, (%esp)
17 ; X32-NEXT: pshufw $238, (%esp), %mm0 # mm0 = mem[2,3,2,3]
18 ; X32-NEXT: movd %mm0, %eax
19 ; X32-NEXT: addl $32, %eax
20 ; X32-NEXT: movl %ebp, %esp
25 ; X64: # %bb.0: # %entry
26 ; X64-NEXT: pshufw $238, (%rdi), %mm0 # mm0 = mem[2,3,2,3]
27 ; X64-NEXT: movd %mm0, %eax
28 ; X64-NEXT: addl $32, %eax
31 %v5 = load <1 x i64>, <1 x i64>* %v4, align 8
32 %v12 = bitcast <1 x i64> %v5 to <4 x i16>
33 %v13 = bitcast <4 x i16> %v12 to x86_mmx
34 %v14 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %v13, i8 -18)
35 %v15 = bitcast x86_mmx %v14 to <4 x i16>
36 %v16 = bitcast <4 x i16> %v15 to <1 x i64>
37 %v17 = extractelement <1 x i64> %v16, i32 0
38 %v18 = bitcast i64 %v17 to <2 x i32>
39 %v19 = extractelement <2 x i32> %v18, i32 0
40 %v20 = add i32 %v19, 32
44 define i32 @test1(i32* nocapture readonly %ptr) nounwind {
46 ; X32: # %bb.0: # %entry
47 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
48 ; X32-NEXT: movd (%eax), %mm0
49 ; X32-NEXT: pshufw $232, %mm0, %mm0 # mm0 = mm0[0,2,2,3]
50 ; X32-NEXT: movd %mm0, %eax
55 ; X64: # %bb.0: # %entry
56 ; X64-NEXT: movd (%rdi), %mm0
57 ; X64-NEXT: pshufw $232, %mm0, %mm0 # mm0 = mm0[0,2,2,3]
58 ; X64-NEXT: movd %mm0, %eax
62 %0 = load i32, i32* %ptr, align 4
63 %1 = insertelement <2 x i32> undef, i32 %0, i32 0
64 %2 = insertelement <2 x i32> %1, i32 0, i32 1
65 %3 = bitcast <2 x i32> %2 to x86_mmx
66 %4 = bitcast x86_mmx %3 to i64
67 %5 = bitcast i64 %4 to <4 x i16>
68 %6 = bitcast <4 x i16> %5 to x86_mmx
69 %7 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %6, i8 -24)
70 %8 = bitcast x86_mmx %7 to <4 x i16>
71 %9 = bitcast <4 x i16> %8 to <1 x i64>
72 %10 = extractelement <1 x i64> %9, i32 0
73 %11 = bitcast i64 %10 to <2 x i32>
74 %12 = extractelement <2 x i32> %11, i32 0
75 tail call void @llvm.x86.mmx.emms()
79 define i32 @test2(i32* nocapture readonly %ptr) nounwind {
81 ; X32: # %bb.0: # %entry
82 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
83 ; X32-NEXT: pshufw $232, (%eax), %mm0 # mm0 = mem[0,2,2,3]
84 ; X32-NEXT: movd %mm0, %eax
89 ; X64: # %bb.0: # %entry
90 ; X64-NEXT: pshufw $232, (%rdi), %mm0 # mm0 = mem[0,2,2,3]
91 ; X64-NEXT: movd %mm0, %eax
95 %0 = bitcast i32* %ptr to x86_mmx*
96 %1 = load x86_mmx, x86_mmx* %0, align 8
97 %2 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %1, i8 -24)
98 %3 = bitcast x86_mmx %2 to <4 x i16>
99 %4 = bitcast <4 x i16> %3 to <1 x i64>
100 %5 = extractelement <1 x i64> %4, i32 0
101 %6 = bitcast i64 %5 to <2 x i32>
102 %7 = extractelement <2 x i32> %6, i32 0
103 tail call void @llvm.x86.mmx.emms()
107 define i32 @test3(x86_mmx %a) nounwind {
110 ; X32-NEXT: movd %mm0, %eax
115 ; X64-NEXT: movd %mm0, %eax
117 %tmp0 = bitcast x86_mmx %a to <2 x i32>
118 %tmp1 = extractelement <2 x i32> %tmp0, i32 0
122 ; Verify we don't muck with extractelts from the upper lane.
123 define i32 @test4(x86_mmx %a) nounwind {
126 ; X32-NEXT: pushl %ebp
127 ; X32-NEXT: movl %esp, %ebp
128 ; X32-NEXT: andl $-8, %esp
129 ; X32-NEXT: subl $8, %esp
130 ; X32-NEXT: movq %mm0, (%esp)
131 ; X32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
132 ; X32-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,0,1]
133 ; X32-NEXT: movd %xmm0, %eax
134 ; X32-NEXT: movl %ebp, %esp
135 ; X32-NEXT: popl %ebp
140 ; X64-NEXT: movq %mm0, -{{[0-9]+}}(%rsp)
141 ; X64-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
142 ; X64-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,3,0,1]
143 ; X64-NEXT: movd %xmm0, %eax
145 %tmp0 = bitcast x86_mmx %a to <2 x i32>
146 %tmp1 = extractelement <2 x i32> %tmp0, i32 1
150 declare x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx, i8)
151 declare void @llvm.x86.mmx.emms()