89 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=slp-vectorizer -S -slp-revec -slp-max-reg-size=1024 -slp-threshold=-1000 %s | FileCheck %s3 4; The 4 stores can be re-vectorised, make sure the poison sources5; are safely handled when trying to vectorise [ %0, poison, poison, %1 ]6define void @test_missing_lanes_1_2(ptr %ptr, i32 %val0, i32 %val1) {7; CHECK-LABEL: @test_missing_lanes_1_2(8; CHECK-NEXT: entry:9; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL0:%.*]], i32 010; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL1:%.*]], i32 011; CHECK-NEXT: [[GETELEMENTPTR0:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 012; CHECK-NEXT: store <4 x i32> [[TMP0]], ptr [[GETELEMENTPTR0]], align 413; CHECK-NEXT: [[GETELEMENTPTR1:%.*]] = getelementptr i32, ptr [[PTR]], i64 414; CHECK-NEXT: store <8 x i32> poison, ptr [[GETELEMENTPTR1]], align 415; CHECK-NEXT: [[GETELEMENTPTR3:%.*]] = getelementptr i32, ptr [[PTR]], i64 1216; CHECK-NEXT: store <4 x i32> [[TMP1]], ptr [[GETELEMENTPTR3]], align 417; CHECK-NEXT: ret void18;19entry:20 %0 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val0, i32 021 %1 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val1, i32 022 23 %getelementptr0 = getelementptr i32, ptr %ptr, i64 024 store <4 x i32> %0, ptr %getelementptr0, align 425 %getelementptr1 = getelementptr i32, ptr %ptr, i64 426 store <4 x i32> poison, ptr %getelementptr1, align 427 %getelementptr2 = getelementptr i32, ptr %ptr, i64 828 store <4 x i32> poison, ptr %getelementptr2, align 429 %getelementptr3 = getelementptr i32, ptr %ptr, i64 1230 store <4 x i32> %1, ptr %getelementptr3, align 431 32 ret void33}34 35; The 4 stores can be re-vectorised, make sure the poison sources36; are safely handled when trying to vectorise [ %0, poison, %1, poison ]37define void @test_missing_lanes_1_3(ptr %ptr, i32 %val0, i32 %val1) {38; CHECK-LABEL: @test_missing_lanes_1_3(39; CHECK-NEXT: entry:40; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL0:%.*]], i32 041; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL1:%.*]], i32 042; CHECK-NEXT: [[GETELEMENTPTR0:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 043; CHECK-NEXT: store <4 x i32> [[TMP0]], ptr [[GETELEMENTPTR0]], align 444; CHECK-NEXT: [[GETELEMENTPTR1:%.*]] = getelementptr i32, ptr [[PTR]], i64 445; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 0, i32 1, i32 2, i32 3>46; CHECK-NEXT: store <8 x i32> [[TMP2]], ptr [[GETELEMENTPTR1]], align 447; CHECK-NEXT: [[GETELEMENTPTR3:%.*]] = getelementptr i32, ptr [[PTR]], i64 1248; CHECK-NEXT: store <4 x i32> poison, ptr [[GETELEMENTPTR3]], align 449; CHECK-NEXT: ret void50;51entry:52 %0 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val0, i32 053 %1 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val1, i32 054 55 %getelementptr0 = getelementptr i32, ptr %ptr, i64 056 store <4 x i32> %0, ptr %getelementptr0, align 457 %getelementptr1 = getelementptr i32, ptr %ptr, i64 458 store <4 x i32> poison, ptr %getelementptr1, align 459 %getelementptr2 = getelementptr i32, ptr %ptr, i64 860 store <4 x i32> %1, ptr %getelementptr2, align 461 %getelementptr3 = getelementptr i32, ptr %ptr, i64 1262 store <4 x i32> poison, ptr %getelementptr3, align 463 64 ret void65}66 67; This could be re-vectorised to use a store <8 x i32> instruction.68define void @test_valid_value_operands(ptr %ptr, i32 %val0, i32 %val1) {69; CHECK-LABEL: @test_valid_value_operands(70; CHECK-NEXT: entry:71; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL0:%.*]], i32 072; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL1:%.*]], i32 073; CHECK-NEXT: [[GETELEMENTPTR0:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 074; CHECK-NEXT: store <4 x i32> [[TMP0]], ptr [[GETELEMENTPTR0]], align 475; CHECK-NEXT: [[GETELEMENTPTR1:%.*]] = getelementptr i32, ptr [[PTR]], i64 476; CHECK-NEXT: store <4 x i32> [[TMP1]], ptr [[GETELEMENTPTR1]], align 477; CHECK-NEXT: ret void78;79entry:80 %0 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val0, i32 081 %1 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val1, i32 082 83 %getelementptr0 = getelementptr i32, ptr %ptr, i64 084 store <4 x i32> %0, ptr %getelementptr0, align 485 %getelementptr1 = getelementptr i32, ptr %ptr, i64 486 store <4 x i32> %1, ptr %getelementptr1, align 487 ret void88}89