xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/RISCV/select-profitability.ll (revision 38fffa630ee80163dc65e759392ad29798905679)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
2; RUN: opt -S --passes=slp-vectorizer -mtriple=riscv64-unknown-linux -mattr=+v < %s | FileCheck %s
3
4define i32 @pow2_zero_constant_shift(i16 zeroext %a, i16 zeroext %b, i16 zeroext %c, i16 zeroext %d) {
5; CHECK-LABEL: define i32 @pow2_zero_constant_shift(
6; CHECK-SAME: i16 zeroext [[A:%.*]], i16 zeroext [[B:%.*]], i16 zeroext [[C:%.*]], i16 zeroext [[D:%.*]]) #[[ATTR0:[0-9]+]] {
7; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x i16> poison, i16 [[A]], i32 0
8; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x i16> [[TMP1]], i16 [[B]], i32 1
9; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x i16> [[TMP2]], i16 [[C]], i32 2
10; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x i16> [[TMP3]], i16 [[D]], i32 3
11; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq <4 x i16> [[TMP4]], splat (i16 1)
12; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[TMP5]], <4 x i32> splat (i32 65536), <4 x i32> zeroinitializer
13; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[TMP6]])
14; CHECK-NEXT:    ret i32 [[TMP7]]
15;
16  %t39.i0 = icmp eq i16 %a, 1
17  %t39.i1 = icmp eq i16 %b, 1
18  %t39.i2 = icmp eq i16 %c, 1
19  %t39.i3 = icmp eq i16 %d, 1
20  %t40.i0 = select i1 %t39.i0, i32 65536, i32 0
21  %t40.i1 = select i1 %t39.i1, i32 65536, i32 0
22  %t40.i2 = select i1 %t39.i2, i32 65536, i32 0
23  %t40.i3 = select i1 %t39.i3, i32 65536, i32 0
24  %or.rdx0 = or i32 %t40.i0, %t40.i1
25  %or.rdx1 = or i32 %t40.i2, %t40.i3
26  %or.rdx2 = or i32 %or.rdx0, %or.rdx1
27  ret i32 %or.rdx2
28}
29
30; TODO: This case is unprofitable, and we should not be vectorizing this.
31define i32 @pow2_zero_variable_shift(i16 zeroext %a, i16 zeroext %b, i16 zeroext %c, i16 zeroext %d) {
32; CHECK-LABEL: define i32 @pow2_zero_variable_shift(
33; CHECK-SAME: i16 zeroext [[A:%.*]], i16 zeroext [[B:%.*]], i16 zeroext [[C:%.*]], i16 zeroext [[D:%.*]]) #[[ATTR0]] {
34; CHECK-NEXT:    [[T39_I0:%.*]] = icmp eq i16 [[A]], 1
35; CHECK-NEXT:    [[T39_I1:%.*]] = icmp eq i16 [[B]], 1
36; CHECK-NEXT:    [[T39_I2:%.*]] = icmp eq i16 [[C]], 1
37; CHECK-NEXT:    [[T39_I3:%.*]] = icmp eq i16 [[D]], 1
38; CHECK-NEXT:    [[T40_I0:%.*]] = select i1 [[T39_I0]], i32 524288, i32 0
39; CHECK-NEXT:    [[T40_I1:%.*]] = select i1 [[T39_I1]], i32 262144, i32 0
40; CHECK-NEXT:    [[T40_I2:%.*]] = select i1 [[T39_I2]], i32 131072, i32 0
41; CHECK-NEXT:    [[T40_I3:%.*]] = select i1 [[T39_I3]], i32 65536, i32 0
42; CHECK-NEXT:    [[OR_RDX0:%.*]] = or i32 [[T40_I0]], [[T40_I1]]
43; CHECK-NEXT:    [[OR_RDX1:%.*]] = or i32 [[T40_I2]], [[T40_I3]]
44; CHECK-NEXT:    [[OR_RDX2:%.*]] = or i32 [[OR_RDX0]], [[OR_RDX1]]
45; CHECK-NEXT:    ret i32 [[OR_RDX2]]
46;
47  %t39.i0 = icmp eq i16 %a, 1
48  %t39.i1 = icmp eq i16 %b, 1
49  %t39.i2 = icmp eq i16 %c, 1
50  %t39.i3 = icmp eq i16 %d, 1
51  %t40.i0 = select i1 %t39.i0, i32 524288, i32 0
52  %t40.i1 = select i1 %t39.i1, i32 262144, i32 0
53  %t40.i2 = select i1 %t39.i2, i32 131072, i32 0
54  %t40.i3 = select i1 %t39.i3, i32 65536, i32 0
55  %or.rdx0 = or i32 %t40.i0, %t40.i1
56  %or.rdx1 = or i32 %t40.i2, %t40.i3
57  %or.rdx2 = or i32 %or.rdx0, %or.rdx1
58  ret i32 %or.rdx2
59}
60