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-gnu -mattr=+v < %s | FileCheck %s
3 @c = global [12 x i64] zeroinitializer
5 ; FIXME: after minbitwidth analysis and i32 conv.., 65535 is transformed to
6 ; and <4 x i16> , -1, which must be dropped.
7 ; FIXME: need to adjust the cost of the final transformation, since the user is
8 ; just a trunc to i16 (it must be free).
10 ; CHECK-LABEL: define i16 @test(
11 ; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
13 ; CHECK-NEXT: [[TMP0:%.*]] = call <4 x i64> @llvm.experimental.vp.strided.load.v4i64.p0.i64(ptr align 8 @c, i64 24, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, i32 4)
14 ; CHECK-NEXT: [[TMP1:%.*]] = trunc <4 x i64> [[TMP0]] to <4 x i16>
15 ; CHECK-NEXT: [[TMP3:%.*]] = xor <4 x i16> [[TMP1]], <i16 -1, i16 -1, i16 -1, i16 -1>
16 ; CHECK-NEXT: [[TMP4:%.*]] = call i16 @llvm.vector.reduce.umax.v4i16(<4 x i16> [[TMP3]])
17 ; CHECK-NEXT: [[TMP5:%.*]] = zext i16 [[TMP4]] to i32
18 ; CHECK-NEXT: [[T:%.*]] = trunc i32 [[TMP5]] to i16
19 ; CHECK-NEXT: ret i16 [[T]]
22 %0 = load i64, ptr @c, align 8
23 %conv = trunc i64 %0 to i32
24 %conv3 = and i32 %conv, 65535
25 %conv4 = xor i32 %conv3, 65535
26 %1 = load i64, ptr getelementptr inbounds ([12 x i64], ptr @c, i64 0, i64 3), align 8
27 %conv.1 = trunc i64 %1 to i32
28 %conv3.1 = and i32 %conv.1, 65535
29 %conv4.1 = xor i32 %conv3.1, 65535
30 %.conv4.1 = tail call i32 @llvm.umax.i32(i32 %conv4, i32 %conv4.1)
31 %2 = load i64, ptr getelementptr inbounds ([12 x i64], ptr @c, i64 0, i64 6), align 8
32 %conv.2 = trunc i64 %2 to i32
33 %conv3.2 = and i32 %conv.2, 65535
34 %conv4.2 = xor i32 %conv3.2, 65535
35 %.conv4.2 = tail call i32 @llvm.umax.i32(i32 %.conv4.1, i32 %conv4.2)
36 %3 = load i64, ptr getelementptr inbounds ([12 x i64], ptr @c, i64 0, i64 9), align 8
37 %conv.3 = trunc i64 %3 to i32
38 %conv3.3 = and i32 %conv.3, 65535
39 %conv4.3 = xor i32 %conv3.3, 65535
40 %.conv4.3 = tail call i32 @llvm.umax.i32(i32 %.conv4.2, i32 %conv4.3)
41 %t = trunc i32 %.conv4.3 to i16