commute-blend-sse41.ll
3.04 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown -mattr=+sse4.1,-slow-unaligned-mem-16 | FileCheck %s
define <8 x i16> @commute_fold_pblendw(<8 x i16> %a, <8 x i16>* %b) {
; CHECK-LABEL: commute_fold_pblendw:
; CHECK: # %bb.0:
; CHECK-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],mem[1,2,3],xmm0[4],mem[5,6,7]
; CHECK-NEXT: retq
%1 = load <8 x i16>, <8 x i16>* %b
%2 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %1, <8 x i16> %a, i8 17)
ret <8 x i16> %2
}
declare <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16>, <8 x i16>, i8) nounwind readnone
define <4 x float> @commute_fold_blendps(<4 x float> %a, <4 x float>* %b) {
; CHECK-LABEL: commute_fold_blendps:
; CHECK: # %bb.0:
; CHECK-NEXT: blendps {{.*#+}} xmm0 = xmm0[0],mem[1],xmm0[2],mem[3]
; CHECK-NEXT: retq
%1 = load <4 x float>, <4 x float>* %b
%2 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %1, <4 x float> %a, i8 5)
ret <4 x float> %2
}
declare <4 x float> @llvm.x86.sse41.blendps(<4 x float>, <4 x float>, i8) nounwind readnone
define <2 x double> @commute_fold_blendpd(<2 x double> %a, <2 x double>* %b) {
; CHECK-LABEL: commute_fold_blendpd:
; CHECK: # %bb.0:
; CHECK-NEXT: blendps {{.*#+}} xmm0 = xmm0[0,1],mem[2,3]
; CHECK-NEXT: retq
%1 = load <2 x double>, <2 x double>* %b
%2 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %1, <2 x double> %a, i8 1)
ret <2 x double> %2
}
declare <2 x double> @llvm.x86.sse41.blendpd(<2 x double>, <2 x double>, i8) nounwind readnone
define <4 x i32> @commute_fold_blend_v4i32(<4 x i32>* %a, <4 x i32> %b) {
; CHECK-LABEL: commute_fold_blend_v4i32:
; CHECK: # %bb.0:
; CHECK-NEXT: paddd %xmm0, %xmm0
; CHECK-NEXT: pblendw {{.*#+}} xmm0 = mem[0,1,2,3,4,5],xmm0[6,7]
; CHECK-NEXT: retq
%1 = load <4 x i32>, <4 x i32>* %a
%2 = add <4 x i32> %b, %b ; force integer domain
%3 = shufflevector <4 x i32> %1, <4 x i32> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
ret <4 x i32> %3
}
; Test case for a crash that occurred due to blendi being commuted to
; movsd during two address instruction pass. The change in number of operands
; caused a bad call to getOperand. This caused the revert in r354713.
%struct.spam = type { i64, i64 }
define void @baz(<2 x i64>* %arg, %struct.spam* %arg1) optsize {
; CHECK-LABEL: baz:
; CHECK: # %bb.0: # %bb
; CHECK-NEXT: movaps (%rdi), %xmm0
; CHECK-NEXT: movaps {{.*#+}} xmm1 = [3,3]
; CHECK-NEXT: andps %xmm0, %xmm1
; CHECK-NEXT: blendps {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3]
; CHECK-NEXT: movups %xmm1, (%rsi)
; CHECK-NEXT: retq
bb:
%tmp = load <2 x i64>, <2 x i64>* %arg, align 16
%tmp2 = and <2 x i64> %tmp, <i64 3, i64 3>
%tmp3 = getelementptr inbounds %struct.spam, %struct.spam* %arg1, i64 0, i32 0
%tmp4 = extractelement <2 x i64> %tmp, i32 0
store i64 %tmp4, i64* %tmp3, align 8
%tmp5 = getelementptr inbounds %struct.spam, %struct.spam* %arg1, i64 0, i32 1
%tmp6 = extractelement <2 x i64> %tmp2, i32 1
store i64 %tmp6, i64* %tmp5, align 8
ret void
}