|
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py |
| 2 | +; RUN: opt -passes=slp-vectorizer -S -slp-revec -slp-max-reg-size=1024 -slp-threshold=-1000 %s | FileCheck %s |
| 3 | + |
| 4 | +; The 4 stores can be re-vectorised, make sure the poison sources |
| 5 | +; are safely handled when trying to vectorise [ %0, poison, poison, %1 ] |
| 6 | +define 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 0 |
| 10 | +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL1:%.*]], i32 0 |
| 11 | +; CHECK-NEXT: [[GETELEMENTPTR0:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 0 |
| 12 | +; CHECK-NEXT: store <4 x i32> [[TMP0]], ptr [[GETELEMENTPTR0]], align 4 |
| 13 | +; CHECK-NEXT: [[GETELEMENTPTR1:%.*]] = getelementptr i32, ptr [[PTR]], i64 4 |
| 14 | +; CHECK-NEXT: store <8 x i32> poison, ptr [[GETELEMENTPTR1]], align 4 |
| 15 | +; CHECK-NEXT: [[GETELEMENTPTR3:%.*]] = getelementptr i32, ptr [[PTR]], i64 12 |
| 16 | +; CHECK-NEXT: store <4 x i32> [[TMP1]], ptr [[GETELEMENTPTR3]], align 4 |
| 17 | +; CHECK-NEXT: ret void |
| 18 | +; |
| 19 | +entry: |
| 20 | + %0 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val0, i32 0 |
| 21 | + %1 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val1, i32 0 |
| 22 | + |
| 23 | + %getelementptr0 = getelementptr i32, ptr %ptr, i64 0 |
| 24 | + store <4 x i32> %0, ptr %getelementptr0, align 4 |
| 25 | + %getelementptr1 = getelementptr i32, ptr %ptr, i64 4 |
| 26 | + store <4 x i32> poison, ptr %getelementptr1, align 4 |
| 27 | + %getelementptr2 = getelementptr i32, ptr %ptr, i64 8 |
| 28 | + store <4 x i32> poison, ptr %getelementptr2, align 4 |
| 29 | + %getelementptr3 = getelementptr i32, ptr %ptr, i64 12 |
| 30 | + store <4 x i32> %1, ptr %getelementptr3, align 4 |
| 31 | + |
| 32 | + ret void |
| 33 | +} |
| 34 | + |
| 35 | +; The 4 stores can be re-vectorised, make sure the poison sources |
| 36 | +; are safely handled when trying to vectorise [ %0, poison, %1, poison ] |
| 37 | +define 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 0 |
| 41 | +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL1:%.*]], i32 0 |
| 42 | +; CHECK-NEXT: [[GETELEMENTPTR0:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 0 |
| 43 | +; CHECK-NEXT: store <4 x i32> [[TMP0]], ptr [[GETELEMENTPTR0]], align 4 |
| 44 | +; CHECK-NEXT: [[GETELEMENTPTR1:%.*]] = getelementptr i32, ptr [[PTR]], i64 4 |
| 45 | +; CHECK-NEXT: [[TMP2:%.*]] = call <8 x i32> @llvm.vector.insert.v8i32.v4i32(<8 x i32> poison, <4 x i32> [[TMP1]], i64 4) |
| 46 | +; CHECK-NEXT: store <8 x i32> [[TMP2]], ptr [[GETELEMENTPTR1]], align 4 |
| 47 | +; CHECK-NEXT: [[GETELEMENTPTR3:%.*]] = getelementptr i32, ptr [[PTR]], i64 12 |
| 48 | +; CHECK-NEXT: store <4 x i32> poison, ptr [[GETELEMENTPTR3]], align 4 |
| 49 | +; CHECK-NEXT: ret void |
| 50 | +; |
| 51 | +entry: |
| 52 | + %0 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val0, i32 0 |
| 53 | + %1 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val1, i32 0 |
| 54 | + |
| 55 | + %getelementptr0 = getelementptr i32, ptr %ptr, i64 0 |
| 56 | + store <4 x i32> %0, ptr %getelementptr0, align 4 |
| 57 | + %getelementptr1 = getelementptr i32, ptr %ptr, i64 4 |
| 58 | + store <4 x i32> poison, ptr %getelementptr1, align 4 |
| 59 | + %getelementptr2 = getelementptr i32, ptr %ptr, i64 8 |
| 60 | + store <4 x i32> %1, ptr %getelementptr2, align 4 |
| 61 | + %getelementptr3 = getelementptr i32, ptr %ptr, i64 12 |
| 62 | + store <4 x i32> poison, ptr %getelementptr3, align 4 |
| 63 | + |
| 64 | + ret void |
| 65 | +} |
| 66 | + |
| 67 | +; This could be re-vectorised to use a store <8 x i32> instruction. |
| 68 | +define 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 0 |
| 72 | +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 [[VAL1:%.*]], i32 0 |
| 73 | +; CHECK-NEXT: [[GETELEMENTPTR0:%.*]] = getelementptr i32, ptr [[PTR:%.*]], i64 0 |
| 74 | +; CHECK-NEXT: store <4 x i32> [[TMP0]], ptr [[GETELEMENTPTR0]], align 4 |
| 75 | +; CHECK-NEXT: [[GETELEMENTPTR1:%.*]] = getelementptr i32, ptr [[PTR]], i64 4 |
| 76 | +; CHECK-NEXT: store <4 x i32> [[TMP1]], ptr [[GETELEMENTPTR1]], align 4 |
| 77 | +; CHECK-NEXT: ret void |
| 78 | +; |
| 79 | +entry: |
| 80 | + %0 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val0, i32 0 |
| 81 | + %1 = insertelement <4 x i32> <i32 poison, i32 0, i32 0, i32 0>, i32 %val1, i32 0 |
| 82 | + |
| 83 | + %getelementptr0 = getelementptr i32, ptr %ptr, i64 0 |
| 84 | + store <4 x i32> %0, ptr %getelementptr0, align 4 |
| 85 | + %getelementptr1 = getelementptr i32, ptr %ptr, i64 4 |
| 86 | + store <4 x i32> %1, ptr %getelementptr1, align 4 |
| 87 | + ret void |
| 88 | +} |
0 commit comments