• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=x86_64-- -mattr=sse4.1 < %s | FileCheck %s -check-prefix=SSE4
3; RUN: llc -mtriple=x86_64-- -mattr=avx < %s | FileCheck %s -check-prefix=AVX1
4; RUN: llc -mtriple=x86_64-- -mattr=avx2 < %s | FileCheck %s -check-prefix=AVX2
5
6define <16 x i16> @split16(<16 x i16> %a, <16 x i16> %b, <16 x i8> %__mask) {
7; SSE4-LABEL: split16:
8; SSE4:       # %bb.0:
9; SSE4-NEXT:    pminuw %xmm2, %xmm0
10; SSE4-NEXT:    pminuw %xmm3, %xmm1
11; SSE4-NEXT:    retq
12;
13; AVX1-LABEL: split16:
14; AVX1:       # %bb.0:
15; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
16; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
17; AVX1-NEXT:    vpminuw %xmm2, %xmm3, %xmm2
18; AVX1-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
19; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
20; AVX1-NEXT:    retq
21;
22; AVX2-LABEL: split16:
23; AVX2:       # %bb.0:
24; AVX2-NEXT:    vpminuw %ymm1, %ymm0, %ymm0
25; AVX2-NEXT:    retq
26  %1 = icmp ult <16 x i16> %a, %b
27  %2 = select <16 x i1> %1, <16 x i16> %a, <16 x i16> %b
28  ret <16 x i16> %2
29}
30
31define <32 x i16> @split32(<32 x i16> %a, <32 x i16> %b, <32 x i8> %__mask) {
32; SSE4-LABEL: split32:
33; SSE4:       # %bb.0:
34; SSE4-NEXT:    pminuw %xmm4, %xmm0
35; SSE4-NEXT:    pminuw %xmm5, %xmm1
36; SSE4-NEXT:    pminuw %xmm6, %xmm2
37; SSE4-NEXT:    pminuw %xmm7, %xmm3
38; SSE4-NEXT:    retq
39;
40; AVX1-LABEL: split32:
41; AVX1:       # %bb.0:
42; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
43; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
44; AVX1-NEXT:    vpminuw %xmm4, %xmm5, %xmm4
45; AVX1-NEXT:    vpminuw %xmm2, %xmm0, %xmm0
46; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
47; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm2
48; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
49; AVX1-NEXT:    vpminuw %xmm2, %xmm4, %xmm2
50; AVX1-NEXT:    vpminuw %xmm3, %xmm1, %xmm1
51; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
52; AVX1-NEXT:    retq
53;
54; AVX2-LABEL: split32:
55; AVX2:       # %bb.0:
56; AVX2-NEXT:    vpminuw %ymm2, %ymm0, %ymm0
57; AVX2-NEXT:    vpminuw %ymm3, %ymm1, %ymm1
58; AVX2-NEXT:    retq
59  %1 = icmp ult <32 x i16> %a, %b
60  %2 = select <32 x i1> %1, <32 x i16> %a, <32 x i16> %b
61  ret <32 x i16> %2
62}
63
64; PR19492
65define i128 @split128(<2 x i128> %a, <2 x i128> %b) {
66; SSE4-LABEL: split128:
67; SSE4:       # %bb.0:
68; SSE4-NEXT:    movq %rdx, %rax
69; SSE4-NEXT:    addq %r8, %rdi
70; SSE4-NEXT:    adcq %r9, %rsi
71; SSE4-NEXT:    addq {{[0-9]+}}(%rsp), %rax
72; SSE4-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx
73; SSE4-NEXT:    addq %rdi, %rax
74; SSE4-NEXT:    adcq %rsi, %rcx
75; SSE4-NEXT:    movq %rcx, %rdx
76; SSE4-NEXT:    retq
77;
78; AVX1-LABEL: split128:
79; AVX1:       # %bb.0:
80; AVX1-NEXT:    movq %rdx, %rax
81; AVX1-NEXT:    addq %r8, %rdi
82; AVX1-NEXT:    adcq %r9, %rsi
83; AVX1-NEXT:    addq {{[0-9]+}}(%rsp), %rax
84; AVX1-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx
85; AVX1-NEXT:    addq %rdi, %rax
86; AVX1-NEXT:    adcq %rsi, %rcx
87; AVX1-NEXT:    movq %rcx, %rdx
88; AVX1-NEXT:    retq
89;
90; AVX2-LABEL: split128:
91; AVX2:       # %bb.0:
92; AVX2-NEXT:    movq %rdx, %rax
93; AVX2-NEXT:    addq %r8, %rdi
94; AVX2-NEXT:    adcq %r9, %rsi
95; AVX2-NEXT:    addq {{[0-9]+}}(%rsp), %rax
96; AVX2-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx
97; AVX2-NEXT:    addq %rdi, %rax
98; AVX2-NEXT:    adcq %rsi, %rcx
99; AVX2-NEXT:    movq %rcx, %rdx
100; AVX2-NEXT:    retq
101  %add = add nsw <2 x i128> %a, %b
102  %rdx.shuf = shufflevector <2 x i128> %add, <2 x i128> undef, <2 x i32> <i32 undef, i32 0>
103  %bin.rdx = add <2 x i128> %add, %rdx.shuf
104  %e = extractelement <2 x i128> %bin.rdx, i32 1
105  ret i128 %e
106}
107