xref: /llvm-project/llvm/test/CodeGen/X86/vector-shuffle-mmx.ll (revision b7e4fba6e5dcae5ff51f8eced21470a1b3ccd895)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-darwin -mattr=+mmx,+sse2 | FileCheck --check-prefix=X86 %s
3; RUN: llc < %s -mtriple=x86_64-darwin -mattr=+mmx,+sse2 | FileCheck --check-prefix=X64 %s
4
5; If there is no explicit MMX type usage, always promote to XMM.
6
7define void @test0(ptr %x) nounwind {
8; X86-LABEL: test0:
9; X86:       ## %bb.0: ## %entry
10; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
11; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
12; X86-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
13; X86-NEXT:    movlps %xmm0, (%eax)
14; X86-NEXT:    retl
15;
16; X64-LABEL: test0:
17; X64:       ## %bb.0: ## %entry
18; X64-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
19; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
20; X64-NEXT:    movq %xmm0, (%rdi)
21; X64-NEXT:    retq
22entry:
23  %tmp2 = load <1 x i64>, ptr %x
24  %tmp6 = bitcast <1 x i64> %tmp2 to <2 x i32>
25  %tmp9 = shufflevector <2 x i32> %tmp6, <2 x i32> undef, <2 x i32> < i32 1, i32 1 >
26  %tmp10 = bitcast <2 x i32> %tmp9 to <1 x i64>
27  store <1 x i64> %tmp10, ptr %x
28  ret void
29}
30
31define void @test1() nounwind {
32; X86-LABEL: test1:
33; X86:       ## %bb.0: ## %entry
34; X86-NEXT:    pushl %edi
35; X86-NEXT:    pxor %mm0, %mm0
36; X86-NEXT:    movq {{\.?LCPI[0-9]+_[0-9]+}}, %mm1 ## mm1 = 0x7070606040400000
37; X86-NEXT:    xorl %edi, %edi
38; X86-NEXT:    maskmovq %mm1, %mm0
39; X86-NEXT:    popl %edi
40; X86-NEXT:    retl
41;
42; X64-LABEL: test1:
43; X64:       ## %bb.0: ## %entry
44; X64-NEXT:    pxor %mm0, %mm0
45; X64-NEXT:    movq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %mm1 ## mm1 = 0x7070606040400000
46; X64-NEXT:    xorl %edi, %edi
47; X64-NEXT:    maskmovq %mm1, %mm0
48; X64-NEXT:    retq
49entry:
50  %tmp528 = bitcast <8 x i8> zeroinitializer to <2 x i32>
51  %tmp529 = and <2 x i32> %tmp528, bitcast (<4 x i16> < i16 -32640, i16 16448, i16 8224, i16 4112 > to <2 x i32>)
52  %tmp542 = bitcast <2 x i32> %tmp529 to <4 x i16>
53  %tmp543 = add <4 x i16> %tmp542, < i16 0, i16 16448, i16 24672, i16 28784 >
54  %tmp555 = bitcast <4 x i16> %tmp543 to <8 x i8>
55  %tmp556 = bitcast <8 x i8> %tmp555 to <1 x i64>
56  %tmp557 = bitcast <8 x i8> zeroinitializer to <1 x i64>
57  tail call void @llvm.x86.mmx.maskmovq( <1 x i64> %tmp557, <1 x i64> %tmp556, ptr null)
58  ret void
59}
60
61@tmp_V2i = common global <2 x i32> zeroinitializer
62
63define void @test2() nounwind {
64; X86-LABEL: test2:
65; X86:       ## %bb.0: ## %entry
66; X86-NEXT:    movl L_tmp_V2i$non_lazy_ptr, %eax
67; X86-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
68; X86-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0,0,1,1]
69; X86-NEXT:    movlps %xmm0, (%eax)
70; X86-NEXT:    retl
71;
72; X64-LABEL: test2:
73; X64:       ## %bb.0: ## %entry
74; X64-NEXT:    movq _tmp_V2i@GOTPCREL(%rip), %rax
75; X64-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
76; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
77; X64-NEXT:    movq %xmm0, (%rax)
78; X64-NEXT:    retq
79entry:
80  %0 = load <2 x i32>, ptr @tmp_V2i, align 8
81  %1 = shufflevector <2 x i32> %0, <2 x i32> undef, <2 x i32> zeroinitializer
82  store <2 x i32> %1, ptr @tmp_V2i, align 8
83  ret void
84}
85
86define <4 x float> @pr35869() nounwind {
87; X86-LABEL: pr35869:
88; X86:       ## %bb.0:
89; X86-NEXT:    movl $64, %eax
90; X86-NEXT:    movd %eax, %mm0
91; X86-NEXT:    pxor %mm1, %mm1
92; X86-NEXT:    punpcklbw %mm1, %mm0 ## mm0 = mm0[0],mm1[0],mm0[1],mm1[1],mm0[2],mm1[2],mm0[3],mm1[3]
93; X86-NEXT:    pcmpgtw %mm0, %mm1
94; X86-NEXT:    movq %mm0, %mm2
95; X86-NEXT:    punpckhwd %mm1, %mm2 ## mm2 = mm2[2],mm1[2],mm2[3],mm1[3]
96; X86-NEXT:    xorps %xmm0, %xmm0
97; X86-NEXT:    cvtpi2ps %mm2, %xmm0
98; X86-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0,0]
99; X86-NEXT:    punpcklwd %mm1, %mm0 ## mm0 = mm0[0],mm1[0],mm0[1],mm1[1]
100; X86-NEXT:    cvtpi2ps %mm0, %xmm0
101; X86-NEXT:    retl
102;
103; X64-LABEL: pr35869:
104; X64:       ## %bb.0:
105; X64-NEXT:    movl $64, %eax
106; X64-NEXT:    movd %eax, %mm0
107; X64-NEXT:    pxor %mm1, %mm1
108; X64-NEXT:    punpcklbw %mm1, %mm0 ## mm0 = mm0[0],mm1[0],mm0[1],mm1[1],mm0[2],mm1[2],mm0[3],mm1[3]
109; X64-NEXT:    pcmpgtw %mm0, %mm1
110; X64-NEXT:    movq %mm0, %mm2
111; X64-NEXT:    punpckhwd %mm1, %mm2 ## mm2 = mm2[2],mm1[2],mm2[3],mm1[3]
112; X64-NEXT:    xorps %xmm0, %xmm0
113; X64-NEXT:    cvtpi2ps %mm2, %xmm0
114; X64-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0,0]
115; X64-NEXT:    punpcklwd %mm1, %mm0 ## mm0 = mm0[0],mm1[0],mm0[1],mm1[1]
116; X64-NEXT:    cvtpi2ps %mm0, %xmm0
117; X64-NEXT:    retq
118  %1 = tail call <1 x i64> @llvm.x86.mmx.punpcklbw(<1 x i64> bitcast (<8 x i8> <i8 64, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0> to <1 x i64>), <1 x i64> bitcast (<8 x i8> zeroinitializer to <1 x i64>))
119  %2 = tail call <1 x i64> @llvm.x86.mmx.pcmpgt.w(<1 x i64> bitcast (<4 x i16> zeroinitializer to <1 x i64>), <1 x i64> %1)
120  %3 = tail call <1 x i64> @llvm.x86.mmx.punpckhwd(<1 x i64> %1, <1 x i64> %2)
121  %4 = tail call <4 x float> @llvm.x86.sse.cvtpi2ps(<4 x float> zeroinitializer, <1 x i64> %3)
122  %5 = shufflevector <4 x float> %4, <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
123  %6 = tail call <1 x i64> @llvm.x86.mmx.punpcklwd(<1 x i64> %1, <1 x i64> %2)
124  %7 = tail call <4 x float> @llvm.x86.sse.cvtpi2ps(<4 x float> %5, <1 x i64> %6)
125  ret <4 x float> %7
126}
127
128declare void @llvm.x86.mmx.maskmovq(<1 x i64>, <1 x i64>, ptr)
129declare <1 x i64> @llvm.x86.mmx.pcmpgt.w(<1 x i64>, <1 x i64>)
130declare <1 x i64> @llvm.x86.mmx.punpcklbw(<1 x i64>, <1 x i64>)
131declare <1 x i64> @llvm.x86.mmx.punpcklwd(<1 x i64>, <1 x i64>)
132declare <1 x i64> @llvm.x86.mmx.punpckhwd(<1 x i64>, <1 x i64>)
133declare <4 x float> @llvm.x86.sse.cvtpi2ps(<4 x float>, <1 x i64>)
134