110 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 42; RUN: opt -S -passes=slp-vectorizer -mtriple=riscv64-unknown-linux-gnu -mattr=+v < %s | FileCheck %s3 4define i32 @test(ptr %p) {5; CHECK-LABEL: define i32 @test(6; CHECK-SAME: ptr [[P:%.*]]) #[[ATTR0:[0-9]+]] {7; CHECK-NEXT: entry:8; CHECK-NEXT: [[D_0:%.*]] = load i16, ptr [[P]], align 49; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i16> <i16 poison, i16 0, i16 0, i16 0>, i16 [[D_0]], i32 010; CHECK-NEXT: [[TMP1:%.*]] = or <4 x i16> [[TMP0]], zeroinitializer11; CHECK-NEXT: [[TMP2:%.*]] = and <4 x i16> [[TMP1]], zeroinitializer12; CHECK-NEXT: [[TMP3:%.*]] = zext <4 x i16> [[TMP0]] to <4 x i32>13; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, <4 x i32> <i32 0, i32 5, i32 6, i32 7>14; CHECK-NEXT: [[TMP5:%.*]] = icmp sgt <4 x i32> [[TMP4]], zeroinitializer15; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[TMP5]], <4 x i16> [[TMP2]], <4 x i16> <i16 2, i16 0, i16 0, i16 0>16; CHECK-NEXT: [[TMP7:%.*]] = call i16 @llvm.vector.reduce.umax.v4i16(<4 x i16> [[TMP6]])17; CHECK-NEXT: [[TMP8:%.*]] = zext i16 [[TMP7]] to i3218; CHECK-NEXT: [[TMP9:%.*]] = call i32 @llvm.umax.i32(i32 [[TMP8]], i32 1)19; CHECK-NEXT: ret i32 [[TMP9]]20;21entry:22 %d.0 = load i16, ptr %p, align 423 %zext.d.0 = zext i16 %d.0 to i3224 %zero.0 = zext i16 0 to i3225 %zero.1 = zext i16 0 to i3226 %zero.2 = zext i16 0 to i3227 28 %or.d.0 = or i32 %zext.d.0, 029 %or.zero.0 = or i32 %zero.0, 030 %or.zero.1 = or i32 %zero.1, 031 %or.zero.2 = or i32 %zero.2, 032 33 %zero.d.0 = and i32 %or.d.0, 034 %and.zero.0 = and i32 %or.zero.0, 035 %and.zero.1 = and i32 %or.zero.1, 036 %and.zero.2 = and i32 %or.zero.2, 037 38 %d.0.gt.0 = icmp sgt i32 %zext.d.0, 039 %false.0 = icmp sgt i32 0, 040 %false.1 = icmp sgt i32 0, 041 %false.2 = icmp sgt i32 0, 042 43 %select.0.2 = select i1 %d.0.gt.0, i32 %zero.d.0, i32 244 %select.1.0 = select i1 %false.0, i32 %and.zero.0, i32 045 %select.2.0 = select i1 %false.1, i32 %and.zero.1, i32 046 %select.3.0 = select i1 %false.2, i32 %and.zero.2, i32 047 48 %max.0 = call i32 @llvm.umax.i32(i32 %select.0.2, i32 %select.1.0)49 %max.1 = call i32 @llvm.umax.i32(i32 %max.0, i32 %select.2.0)50 %max.2 = call i32 @llvm.umax.i32(i32 %max.1, i32 %select.3.0)51 %max.3 = call i32 @llvm.umax.i32(i32 %max.2, i32 1)52 53 ret i32 %max.354}55 56define i32 @test1(ptr %p) {57; CHECK-LABEL: define i32 @test1(58; CHECK-SAME: ptr [[P:%.*]]) #[[ATTR0]] {59; CHECK-NEXT: entry:60; CHECK-NEXT: [[D_0:%.*]] = load i16, ptr [[P]], align 461; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i16> <i16 poison, i16 0, i16 0, i16 0>, i16 [[D_0]], i32 062; CHECK-NEXT: [[TMP3:%.*]] = zext <4 x i16> [[TMP0]] to <4 x i32>63; CHECK-NEXT: [[TMP2:%.*]] = or <4 x i32> [[TMP3]], zeroinitializer64; CHECK-NEXT: [[TMP6:%.*]] = and <4 x i32> [[TMP2]], <i32 -1, i32 -16383, i32 65535, i32 -1>65; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> <i32 -1, i32 -16383, i32 65535, i32 -1>, <4 x i32> <i32 0, i32 5, i32 6, i32 7>66; CHECK-NEXT: [[TMP5:%.*]] = icmp eq <4 x i32> [[TMP4]], <i32 65535, i32 -16383, i32 65535, i32 65535>67; CHECK-NEXT: [[TMP7:%.*]] = select <4 x i1> [[TMP5]], <4 x i32> [[TMP6]], <4 x i32> <i32 4, i32 3, i32 2, i32 1>68; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP7]])69; CHECK-NEXT: ret i32 [[TMP8]]70;71entry:72 %d.0 = load i16, ptr %p, align 473 %zext.d.0 = zext i16 %d.0 to i3274 %zero.0 = zext i16 0 to i3275 %zero.1 = zext i16 0 to i3276 %zero.2 = zext i16 0 to i3277 78 %or.d.0 = or i32 %zext.d.0, 079 %or.zero.0 = or i32 %zero.0, 080 %or.zero.1 = or i32 %zero.1, 081 %or.zero.2 = or i32 %zero.2, 082 83 %szero.00 = sext i16 65535 to i3284 %szero.0 = sext i16 -16383 to i3285 %uzero.1 = zext i16 65535 to i3286 %szero.2 = sext i16 65535 to i3287 88 %zero.d.0 = and i32 %or.d.0, %szero.0089 %and.zero.0 = and i32 %or.zero.0, %szero.090 %and.zero.1 = and i32 %or.zero.1, %uzero.191 %and.zero.2 = and i32 %or.zero.2, %szero.292 93 %d.0.gt.0 = icmp eq i32 %zext.d.0, 6553594 %false.0 = icmp eq i32 %szero.0, -1638395 %false.1 = icmp eq i32 %uzero.1, 6553596 %false.2 = icmp eq i32 %szero.2, 6553597 98 %select.0.2 = select i1 %d.0.gt.0, i32 %zero.d.0, i32 499 %select.1.0 = select i1 %false.0, i32 %and.zero.0, i32 3100 %select.2.0 = select i1 %false.1, i32 %and.zero.1, i32 2101 %select.3.0 = select i1 %false.2, i32 %and.zero.2, i32 1102 103 %max.0 = add i32 %select.0.2, %select.1.0104 %max.1 = add i32 %max.0, %select.2.0105 %max.2 = add i32 %max.1, %select.3.0106 107 ret i32 %max.2108}109 110