1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_mem_shuffle
2 ; RUN: llc < %s -mtriple=i686-apple-darwin -mcpu=corei7-avx | FileCheck %s --check-prefixes=CHECK,X86
3 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx | FileCheck %s --check-prefixes=CHECK,X64
5 define <4 x i32> @blendvb_fallback_v4i32(<4 x i1> %mask, <4 x i32> %x, <4 x i32> %y) {
6 ; CHECK-LABEL: blendvb_fallback_v4i32:
8 ; CHECK-NEXT: vpslld $31, %xmm0, %xmm0
9 ; CHECK-NEXT: vblendvps %xmm0, %xmm1, %xmm2, %xmm0
10 ; CHECK-NEXT: ret{{[l|q]}}
11 %ret = select <4 x i1> %mask, <4 x i32> %x, <4 x i32> %y
15 define <8 x i32> @blendvb_fallback_v8i32(<8 x i1> %mask, <8 x i32> %x, <8 x i32> %y) {
16 ; CHECK-LABEL: blendvb_fallback_v8i32:
18 ; CHECK-NEXT: vpmovzxwd {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
19 ; CHECK-NEXT: vpslld $31, %xmm3, %xmm3
20 ; CHECK-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
21 ; CHECK-NEXT: vpslld $31, %xmm0, %xmm0
22 ; CHECK-NEXT: vinsertf128 $1, %xmm0, %ymm3, %ymm0
23 ; CHECK-NEXT: vblendvps %ymm0, %ymm1, %ymm2, %ymm0
24 ; CHECK-NEXT: ret{{[l|q]}}
25 %ret = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y
29 define <8 x float> @blendvb_fallback_v8f32(<8 x i1> %mask, <8 x float> %x, <8 x float> %y) {
30 ; CHECK-LABEL: blendvb_fallback_v8f32:
32 ; CHECK-NEXT: vpmovzxwd {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
33 ; CHECK-NEXT: vpslld $31, %xmm3, %xmm3
34 ; CHECK-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
35 ; CHECK-NEXT: vpslld $31, %xmm0, %xmm0
36 ; CHECK-NEXT: vinsertf128 $1, %xmm0, %ymm3, %ymm0
37 ; CHECK-NEXT: vblendvps %ymm0, %ymm1, %ymm2, %ymm0
38 ; CHECK-NEXT: ret{{[l|q]}}
39 %ret = select <8 x i1> %mask, <8 x float> %x, <8 x float> %y
43 declare <4 x float> @llvm.x86.sse41.insertps(<4 x float>, <4 x float>, i32) nounwind readnone
45 define <4 x float> @insertps_from_vector_load(<4 x float> %a, ptr nocapture readonly %pb) {
46 ; X86-LABEL: insertps_from_vector_load:
48 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
49 ; X86-NEXT: vinsertps $48, (%eax), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]
52 ; X64-LABEL: insertps_from_vector_load:
54 ; X64-NEXT: vinsertps $48, (%rdi), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]
56 %1 = load <4 x float>, ptr %pb, align 16
57 %2 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %1, i32 48)
61 ;; Use a non-zero CountS for insertps
62 define <4 x float> @insertps_from_vector_load_offset(<4 x float> %a, ptr nocapture readonly %pb) {
63 ; X86-LABEL: insertps_from_vector_load_offset:
65 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
66 ; X86-NEXT: vinsertps $32, 4(%eax), %xmm0, %xmm0 ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
69 ; X64-LABEL: insertps_from_vector_load_offset:
71 ; X64-NEXT: vinsertps $32, 4(%rdi), %xmm0, %xmm0 ## xmm0 = xmm0[0,1],mem[0],xmm0[3]
73 %1 = load <4 x float>, ptr %pb, align 16
74 %2 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %1, i32 96)
78 define <4 x float> @insertps_from_vector_load_offset_2(<4 x float> %a, ptr nocapture readonly %pb, i64 %index) {
79 ; X86-LABEL: insertps_from_vector_load_offset_2:
81 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
82 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
83 ; X86-NEXT: shll $4, %ecx
84 ; X86-NEXT: vinsertps $0, 12(%eax,%ecx), %xmm0, %xmm0 ## xmm0 = mem[0],xmm0[1,2,3]
87 ; X64-LABEL: insertps_from_vector_load_offset_2:
89 ; X64-NEXT: shlq $4, %rsi
90 ; X64-NEXT: vinsertps $0, 12(%rdi,%rsi), %xmm0, %xmm0 ## xmm0 = mem[0],xmm0[1,2,3]
92 %1 = getelementptr inbounds <4 x float>, ptr %pb, i64 %index
93 %2 = load <4 x float>, ptr %1, align 16
94 %3 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %2, i32 192)
98 define <4 x float> @insertps_from_broadcast_loadf32(<4 x float> %a, ptr nocapture readonly %fb, i64 %index) {
99 ; X86-LABEL: insertps_from_broadcast_loadf32:
101 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
102 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
103 ; X86-NEXT: vinsertps $48, (%ecx,%eax,4), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]
106 ; X64-LABEL: insertps_from_broadcast_loadf32:
108 ; X64-NEXT: vinsertps $48, (%rdi,%rsi,4), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]
110 %1 = getelementptr inbounds float, ptr %fb, i64 %index
111 %2 = load float, ptr %1, align 4
112 %3 = insertelement <4 x float> undef, float %2, i32 0
113 %4 = insertelement <4 x float> %3, float %2, i32 1
114 %5 = insertelement <4 x float> %4, float %2, i32 2
115 %6 = insertelement <4 x float> %5, float %2, i32 3
116 %7 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %6, i32 48)
120 define <4 x float> @insertps_from_broadcast_loadv4f32(<4 x float> %a, ptr nocapture readonly %b) {
121 ; X86-LABEL: insertps_from_broadcast_loadv4f32:
123 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
124 ; X86-NEXT: vinsertps $48, (%eax), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]
127 ; X64-LABEL: insertps_from_broadcast_loadv4f32:
129 ; X64-NEXT: vinsertps $48, (%rdi), %xmm0, %xmm0 ## xmm0 = xmm0[0,1,2],mem[0]
131 %1 = load <4 x float>, ptr %b, align 4
132 %2 = extractelement <4 x float> %1, i32 0
133 %3 = insertelement <4 x float> undef, float %2, i32 0
134 %4 = insertelement <4 x float> %3, float %2, i32 1
135 %5 = insertelement <4 x float> %4, float %2, i32 2
136 %6 = insertelement <4 x float> %5, float %2, i32 3
137 %7 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %6, i32 48)
141 define <4 x float> @insertps_from_broadcast_multiple_use(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x float> %d, ptr nocapture readonly %fb, i64 %index) {
142 ; X86-LABEL: insertps_from_broadcast_multiple_use:
144 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
145 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
146 ; X86-NEXT: vbroadcastss (%ecx,%eax,4), %xmm4
147 ; X86-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm4[3]
148 ; X86-NEXT: vblendps {{.*#+}} xmm1 = xmm1[0,1,2],xmm4[3]
149 ; X86-NEXT: vaddps %xmm1, %xmm0, %xmm0
150 ; X86-NEXT: vblendps {{.*#+}} xmm1 = xmm2[0,1,2],xmm4[3]
151 ; X86-NEXT: vblendps {{.*#+}} xmm2 = xmm3[0,1,2],xmm4[3]
152 ; X86-NEXT: vaddps %xmm2, %xmm1, %xmm1
153 ; X86-NEXT: vaddps %xmm1, %xmm0, %xmm0
156 ; X64-LABEL: insertps_from_broadcast_multiple_use:
158 ; X64-NEXT: vbroadcastss (%rdi,%rsi,4), %xmm4
159 ; X64-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0,1,2],xmm4[3]
160 ; X64-NEXT: vblendps {{.*#+}} xmm1 = xmm1[0,1,2],xmm4[3]
161 ; X64-NEXT: vaddps %xmm1, %xmm0, %xmm0
162 ; X64-NEXT: vblendps {{.*#+}} xmm1 = xmm2[0,1,2],xmm4[3]
163 ; X64-NEXT: vblendps {{.*#+}} xmm2 = xmm3[0,1,2],xmm4[3]
164 ; X64-NEXT: vaddps %xmm2, %xmm1, %xmm1
165 ; X64-NEXT: vaddps %xmm1, %xmm0, %xmm0
167 %1 = getelementptr inbounds float, ptr %fb, i64 %index
168 %2 = load float, ptr %1, align 4
169 %3 = insertelement <4 x float> undef, float %2, i32 0
170 %4 = insertelement <4 x float> %3, float %2, i32 1
171 %5 = insertelement <4 x float> %4, float %2, i32 2
172 %6 = insertelement <4 x float> %5, float %2, i32 3
173 %7 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a, <4 x float> %6, i32 48)
174 %8 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %b, <4 x float> %6, i32 48)
175 %9 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %c, <4 x float> %6, i32 48)
176 %10 = tail call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %d, <4 x float> %6, i32 48)
177 %11 = fadd <4 x float> %7, %8
178 %12 = fadd <4 x float> %9, %10
179 %13 = fadd <4 x float> %11, %12