1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-linux-gnu -mcpu=corei7 | FileCheck %s --check-prefix=X32
3 ; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 | FileCheck %s --check-prefix=X64
5 ; Make sure that we don't crash when legalizing vselect and vsetcc and that
6 ; we are able to generate vector blend instructions.
8 define void @simple_widen(<2 x float> %a, <2 x float> %b) {
9 ; X32-LABEL: simple_widen:
10 ; X32: # %bb.0: # %entry
11 ; X32-NEXT: movlps %xmm1, (%eax)
14 ; X64-LABEL: simple_widen:
15 ; X64: # %bb.0: # %entry
16 ; X64-NEXT: movlps %xmm1, (%rax)
19 %0 = select <2 x i1> undef, <2 x float> %a, <2 x float> %b
20 store <2 x float> %0, <2 x float>* undef
24 define void @complex_inreg_work(<2 x float> %a, <2 x float> %b, <2 x float> %c) {
25 ; X32-LABEL: complex_inreg_work:
26 ; X32: # %bb.0: # %entry
27 ; X32-NEXT: movaps %xmm0, %xmm3
28 ; X32-NEXT: cmpordps %xmm2, %xmm2
29 ; X32-NEXT: movaps %xmm2, %xmm0
30 ; X32-NEXT: blendvps %xmm0, %xmm3, %xmm1
31 ; X32-NEXT: movlps %xmm1, (%eax)
34 ; X64-LABEL: complex_inreg_work:
35 ; X64: # %bb.0: # %entry
36 ; X64-NEXT: movaps %xmm0, %xmm3
37 ; X64-NEXT: cmpordps %xmm2, %xmm2
38 ; X64-NEXT: movaps %xmm2, %xmm0
39 ; X64-NEXT: blendvps %xmm0, %xmm3, %xmm1
40 ; X64-NEXT: movlps %xmm1, (%rax)
43 %0 = fcmp oeq <2 x float> %c, %c
44 %1 = select <2 x i1> %0, <2 x float> %a, <2 x float> %b
45 store <2 x float> %1, <2 x float>* undef
49 define void @zero_test() {
50 ; X32-LABEL: zero_test:
51 ; X32: # %bb.0: # %entry
52 ; X32-NEXT: xorps %xmm0, %xmm0
53 ; X32-NEXT: movlps %xmm0, (%eax)
56 ; X64-LABEL: zero_test:
57 ; X64: # %bb.0: # %entry
58 ; X64-NEXT: xorps %xmm0, %xmm0
59 ; X64-NEXT: movlps %xmm0, (%rax)
62 %0 = select <2 x i1> undef, <2 x float> undef, <2 x float> zeroinitializer
63 store <2 x float> %0, <2 x float>* undef
67 define void @full_test() {
68 ; X32-LABEL: full_test:
69 ; X32: # %bb.0: # %entry
70 ; X32-NEXT: subl $60, %esp
71 ; X32-NEXT: .cfi_def_cfa_offset 64
72 ; X32-NEXT: movsd {{.*#+}} xmm2 = mem[0],zero
73 ; X32-NEXT: cvttps2dq %xmm2, %xmm0
74 ; X32-NEXT: cvtdq2ps %xmm0, %xmm1
75 ; X32-NEXT: xorps %xmm0, %xmm0
76 ; X32-NEXT: cmpltps %xmm2, %xmm0
77 ; X32-NEXT: movaps {{.*#+}} xmm3 = <1.0E+0,1.0E+0,u,u>
78 ; X32-NEXT: addps %xmm1, %xmm3
79 ; X32-NEXT: movaps %xmm1, %xmm4
80 ; X32-NEXT: blendvps %xmm0, %xmm3, %xmm4
81 ; X32-NEXT: cmpeqps %xmm2, %xmm1
82 ; X32-NEXT: movaps %xmm1, %xmm0
83 ; X32-NEXT: blendvps %xmm0, %xmm2, %xmm4
84 ; X32-NEXT: movlps %xmm4, {{[0-9]+}}(%esp)
85 ; X32-NEXT: movlps %xmm4, {{[0-9]+}}(%esp)
86 ; X32-NEXT: addl $60, %esp
87 ; X32-NEXT: .cfi_def_cfa_offset 4
90 ; X64-LABEL: full_test:
91 ; X64: # %bb.0: # %entry
92 ; X64-NEXT: movsd {{.*#+}} xmm2 = mem[0],zero
93 ; X64-NEXT: cvttps2dq %xmm2, %xmm0
94 ; X64-NEXT: cvtdq2ps %xmm0, %xmm1
95 ; X64-NEXT: xorps %xmm0, %xmm0
96 ; X64-NEXT: cmpltps %xmm2, %xmm0
97 ; X64-NEXT: movaps {{.*#+}} xmm3 = <1.0E+0,1.0E+0,u,u>
98 ; X64-NEXT: addps %xmm1, %xmm3
99 ; X64-NEXT: movaps %xmm1, %xmm4
100 ; X64-NEXT: blendvps %xmm0, %xmm3, %xmm4
101 ; X64-NEXT: cmpeqps %xmm2, %xmm1
102 ; X64-NEXT: movaps %xmm1, %xmm0
103 ; X64-NEXT: blendvps %xmm0, %xmm2, %xmm4
104 ; X64-NEXT: movlps %xmm4, -{{[0-9]+}}(%rsp)
105 ; X64-NEXT: movlps %xmm4, -{{[0-9]+}}(%rsp)
108 %Cy300 = alloca <4 x float>
109 %Cy11a = alloca <2 x float>
110 %Cy118 = alloca <2 x float>
111 %Cy119 = alloca <2 x float>
115 %0 = load <2 x float>, <2 x float>* %Cy119
116 %1 = fptosi <2 x float> %0 to <2 x i32>
117 %2 = sitofp <2 x i32> %1 to <2 x float>
118 %3 = fcmp ogt <2 x float> %0, zeroinitializer
119 %4 = fadd <2 x float> %2, <float 1.000000e+00, float 1.000000e+00>
120 %5 = select <2 x i1> %3, <2 x float> %4, <2 x float> %2
121 %6 = fcmp oeq <2 x float> %2, %0
122 %7 = select <2 x i1> %6, <2 x float> %0, <2 x float> %5
123 store <2 x float> %7, <2 x float>* %Cy118
124 %8 = load <2 x float>, <2 x float>* %Cy118
125 store <2 x float> %8, <2 x float>* %Cy11a