Skip to content

Commit 20e2b4f

Browse files
committed
[SLP][NFC]Add a test for non-consecutive inserts, NFC.
1 parent deb2b20 commit 20e2b4f

File tree

1 file changed

+55
-0
lines changed

1 file changed

+55
-0
lines changed

llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll

Lines changed: 55 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -38,6 +38,61 @@ define <4 x float> @simple_select(<4 x float> %a, <4 x float> %b, <4 x i32> %c)
3838
ret <4 x float> %rd
3939
}
4040

41+
define <8 x float> @simple_select2(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 {
42+
; CHECK-LABEL: @simple_select2(
43+
; CHECK-NEXT: [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
44+
; CHECK-NEXT: [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
45+
; CHECK-NEXT: [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
46+
; CHECK-NEXT: [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
47+
; CHECK-NEXT: [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
48+
; CHECK-NEXT: [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
49+
; CHECK-NEXT: [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2
50+
; CHECK-NEXT: [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3
51+
; CHECK-NEXT: [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0
52+
; CHECK-NEXT: [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1
53+
; CHECK-NEXT: [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2
54+
; CHECK-NEXT: [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3
55+
; CHECK-NEXT: [[CMP0:%.*]] = icmp ne i32 [[C0]], 0
56+
; CHECK-NEXT: [[CMP1:%.*]] = icmp ne i32 [[C1]], 0
57+
; CHECK-NEXT: [[CMP2:%.*]] = icmp ne i32 [[C2]], 0
58+
; CHECK-NEXT: [[CMP3:%.*]] = icmp ne i32 [[C3]], 0
59+
; CHECK-NEXT: [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]]
60+
; CHECK-NEXT: [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]]
61+
; CHECK-NEXT: [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
62+
; CHECK-NEXT: [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
63+
; CHECK-NEXT: [[RA:%.*]] = insertelement <8 x float> undef, float [[S0]], i32 0
64+
; CHECK-NEXT: [[RB:%.*]] = insertelement <8 x float> [[RA]], float [[S1]], i32 2
65+
; CHECK-NEXT: [[RC:%.*]] = insertelement <8 x float> [[RB]], float [[S2]], i32 4
66+
; CHECK-NEXT: [[RD:%.*]] = insertelement <8 x float> [[RC]], float [[S3]], i32 7
67+
; CHECK-NEXT: ret <8 x float> [[RD]]
68+
;
69+
%c0 = extractelement <4 x i32> %c, i32 0
70+
%c1 = extractelement <4 x i32> %c, i32 1
71+
%c2 = extractelement <4 x i32> %c, i32 2
72+
%c3 = extractelement <4 x i32> %c, i32 3
73+
%a0 = extractelement <4 x float> %a, i32 0
74+
%a1 = extractelement <4 x float> %a, i32 1
75+
%a2 = extractelement <4 x float> %a, i32 2
76+
%a3 = extractelement <4 x float> %a, i32 3
77+
%b0 = extractelement <4 x float> %b, i32 0
78+
%b1 = extractelement <4 x float> %b, i32 1
79+
%b2 = extractelement <4 x float> %b, i32 2
80+
%b3 = extractelement <4 x float> %b, i32 3
81+
%cmp0 = icmp ne i32 %c0, 0
82+
%cmp1 = icmp ne i32 %c1, 0
83+
%cmp2 = icmp ne i32 %c2, 0
84+
%cmp3 = icmp ne i32 %c3, 0
85+
%s0 = select i1 %cmp0, float %a0, float %b0
86+
%s1 = select i1 %cmp1, float %a1, float %b1
87+
%s2 = select i1 %cmp2, float %a2, float %b2
88+
%s3 = select i1 %cmp3, float %a3, float %b3
89+
%ra = insertelement <8 x float> undef, float %s0, i32 0
90+
%rb = insertelement <8 x float> %ra, float %s1, i32 2
91+
%rc = insertelement <8 x float> %rb, float %s2, i32 4
92+
%rd = insertelement <8 x float> %rc, float %s3, i32 7
93+
ret <8 x float> %rd
94+
}
95+
4196
declare void @llvm.assume(i1) nounwind
4297

4398
; This entire tree is ephemeral, don't vectorize any of it.

0 commit comments

Comments
 (0)