|
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4 |
| 2 | +; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=sse2 | FileCheck %s --check-prefixes=CHECK,SSE |
| 3 | +; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=avx2 | FileCheck %s --check-prefixes=CHECK,AVX |
| 4 | + |
| 5 | +; Fold "shuffle (binop (shuffle, shuffle)), undef" --> "binop (shuffle), (shuffle)" |
| 6 | + |
| 7 | +declare void @use_v4f64(<4 x double>) |
| 8 | + |
| 9 | +define <4 x double> @fadd_v4f64(<4 x double> %a, <4 x double> %b) { |
| 10 | +; CHECK-LABEL: define <4 x double> @fadd_v4f64( |
| 11 | +; CHECK-SAME: <4 x double> [[A:%.*]], <4 x double> [[B:%.*]]) #[[ATTR0:[0-9]+]] { |
| 12 | +; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 13 | +; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 14 | +; CHECK-NEXT: [[POST:%.*]] = fadd <4 x double> [[TMP1]], [[TMP2]] |
| 15 | +; CHECK-NEXT: [[POST1:%.*]] = shufflevector <4 x double> [[POST]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 16 | +; CHECK-NEXT: ret <4 x double> [[POST1]] |
| 17 | +; |
| 18 | + %a1 = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 19 | + %b1 = shufflevector <4 x double> %b, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 20 | + %op = fadd <4 x double> %a1, %b1 |
| 21 | + %post = shufflevector <4 x double> %op, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 22 | + ret <4 x double> %post |
| 23 | +} |
| 24 | + |
| 25 | +define <4 x double> @fadd_v4f64_poison_idx(<4 x double> %a, <4 x double> %b) { |
| 26 | +; CHECK-LABEL: define <4 x double> @fadd_v4f64_poison_idx( |
| 27 | +; CHECK-SAME: <4 x double> [[A:%.*]], <4 x double> [[B:%.*]]) #[[ATTR0]] { |
| 28 | +; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 29 | +; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 30 | +; CHECK-NEXT: [[POST:%.*]] = fadd <4 x double> [[TMP1]], [[TMP2]] |
| 31 | +; CHECK-NEXT: [[POST1:%.*]] = shufflevector <4 x double> [[POST]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 4> |
| 32 | +; CHECK-NEXT: ret <4 x double> [[POST1]] |
| 33 | +; |
| 34 | + %a1 = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 35 | + %b1 = shufflevector <4 x double> %b, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 36 | + %op = fadd <4 x double> %a1, %b1 |
| 37 | + %post = shufflevector <4 x double> %op, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 4> |
| 38 | + ret <4 x double> %post |
| 39 | +} |
| 40 | + |
| 41 | +define <4 x double> @fadd_mixed_types(<4 x double> %a, <2 x double> %b) { |
| 42 | +; CHECK-LABEL: define <4 x double> @fadd_mixed_types( |
| 43 | +; CHECK-SAME: <4 x double> [[A:%.*]], <2 x double> [[B:%.*]]) #[[ATTR0]] { |
| 44 | +; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 45 | +; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x double> [[B]], <2 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 46 | +; CHECK-NEXT: [[POST:%.*]] = fadd <4 x double> [[TMP1]], [[TMP2]] |
| 47 | +; CHECK-NEXT: [[POST1:%.*]] = shufflevector <4 x double> [[POST]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 48 | +; CHECK-NEXT: ret <4 x double> [[POST1]] |
| 49 | +; |
| 50 | + %a1 = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 51 | + %b1 = shufflevector <2 x double> %b, <2 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 52 | + %op = fadd <4 x double> %a1, %b1 |
| 53 | + %post = shufflevector <4 x double> %op, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 54 | + ret <4 x double> %post |
| 55 | +} |
| 56 | + |
| 57 | +; Negative test - multiple use of fadd |
| 58 | +define <4 x double> @fadd_v4f64_multiuse_op(<4 x double> %a, <4 x double> %b) { |
| 59 | +; CHECK-LABEL: define <4 x double> @fadd_v4f64_multiuse_op( |
| 60 | +; CHECK-SAME: <4 x double> [[A:%.*]], <4 x double> [[B:%.*]]) #[[ATTR0]] { |
| 61 | +; CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 62 | +; CHECK-NEXT: [[B1:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 63 | +; CHECK-NEXT: [[OP:%.*]] = fadd <4 x double> [[A1]], [[B1]] |
| 64 | +; CHECK-NEXT: [[POST:%.*]] = shufflevector <4 x double> [[OP]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 65 | +; CHECK-NEXT: call void @use_v4f64(<4 x double> [[OP]]) |
| 66 | +; CHECK-NEXT: ret <4 x double> [[POST]] |
| 67 | +; |
| 68 | + %a1 = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 69 | + %b1 = shufflevector <4 x double> %b, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 70 | + %op = fadd <4 x double> %a1, %b1 |
| 71 | + %post = shufflevector <4 x double> %op, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 72 | + call void @use_v4f64(<4 x double> %op) |
| 73 | + ret <4 x double> %post |
| 74 | +} |
| 75 | + |
| 76 | +; Negative test - multiple use of inner shuffle |
| 77 | +define <4 x double> @fadd_v4f64_multiuse_shuffle(<4 x double> %a, <4 x double> %b) { |
| 78 | +; CHECK-LABEL: define <4 x double> @fadd_v4f64_multiuse_shuffle( |
| 79 | +; CHECK-SAME: <4 x double> [[A:%.*]], <4 x double> [[B:%.*]]) #[[ATTR0]] { |
| 80 | +; CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 81 | +; CHECK-NEXT: [[B1:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 82 | +; CHECK-NEXT: [[OP:%.*]] = fadd <4 x double> [[A1]], [[B1]] |
| 83 | +; CHECK-NEXT: [[POST:%.*]] = shufflevector <4 x double> [[OP]], <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 84 | +; CHECK-NEXT: call void @use_v4f64(<4 x double> [[A1]]) |
| 85 | +; CHECK-NEXT: ret <4 x double> [[POST]] |
| 86 | +; |
| 87 | + %a1 = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 88 | + %b1 = shufflevector <4 x double> %b, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 89 | + %op = fadd <4 x double> %a1, %b1 |
| 90 | + %post = shufflevector <4 x double> %op, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2> |
| 91 | + call void @use_v4f64(<4 x double> %a1) |
| 92 | + ret <4 x double> %post |
| 93 | +} |
| 94 | + |
| 95 | +define <4 x i32> @sdiv_v4i32(<4 x i32> %a, <4 x i32> %b) { |
| 96 | +; CHECK-LABEL: define <4 x i32> @sdiv_v4i32( |
| 97 | +; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]]) #[[ATTR0]] { |
| 98 | +; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 99 | +; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 100 | +; CHECK-NEXT: [[POST:%.*]] = sdiv <4 x i32> [[TMP1]], [[TMP2]] |
| 101 | +; CHECK-NEXT: [[POST1:%.*]] = shufflevector <4 x i32> [[POST]], <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 0> |
| 102 | +; CHECK-NEXT: ret <4 x i32> [[POST1]] |
| 103 | +; |
| 104 | + %a1 = shufflevector <4 x i32> %a, <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 105 | + %b1 = shufflevector <4 x i32> %b, <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 106 | + %op = sdiv <4 x i32> %a1, %b1 |
| 107 | + %post = shufflevector <4 x i32> %op, <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 0> |
| 108 | + ret <4 x i32> %post |
| 109 | +} |
| 110 | + |
| 111 | +; Negative test - don't introduce poison element into div/rem instruction |
| 112 | +define <4 x i32> @sdiv_v4i32_poison(<4 x i32> %a, <4 x i32> %b) { |
| 113 | +; CHECK-LABEL: define <4 x i32> @sdiv_v4i32_poison( |
| 114 | +; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]]) #[[ATTR0]] { |
| 115 | +; CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 116 | +; CHECK-NEXT: [[B1:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 117 | +; CHECK-NEXT: [[OP:%.*]] = sdiv <4 x i32> [[A1]], [[B1]] |
| 118 | +; CHECK-NEXT: [[POST:%.*]] = shufflevector <4 x i32> [[OP]], <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 poison> |
| 119 | +; CHECK-NEXT: ret <4 x i32> [[POST]] |
| 120 | +; |
| 121 | + %a1 = shufflevector <4 x i32> %a, <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 122 | + %b1 = shufflevector <4 x i32> %b, <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 123 | + %op = sdiv <4 x i32> %a1, %b1 |
| 124 | + %post = shufflevector <4 x i32> %op, <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 poison> |
| 125 | + ret <4 x i32> %post |
| 126 | +} |
| 127 | + |
| 128 | +; Negative test - don't introduce poison element into div/rem instruction |
| 129 | +define <4 x i32> @sdiv_v4i32_poison_idx(<4 x i32> %a, <4 x i32> %b) { |
| 130 | +; CHECK-LABEL: define <4 x i32> @sdiv_v4i32_poison_idx( |
| 131 | +; CHECK-SAME: <4 x i32> [[A:%.*]], <4 x i32> [[B:%.*]]) #[[ATTR0]] { |
| 132 | +; CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 133 | +; CHECK-NEXT: [[B1:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 134 | +; CHECK-NEXT: [[OP:%.*]] = sdiv <4 x i32> [[A1]], [[B1]] |
| 135 | +; CHECK-NEXT: [[POST:%.*]] = shufflevector <4 x i32> [[OP]], <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 4> |
| 136 | +; CHECK-NEXT: ret <4 x i32> [[POST]] |
| 137 | +; |
| 138 | + %a1 = shufflevector <4 x i32> %a, <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> |
| 139 | + %b1 = shufflevector <4 x i32> %b, <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0> |
| 140 | + %op = sdiv <4 x i32> %a1, %b1 |
| 141 | + %post = shufflevector <4 x i32> %op, <4 x i32> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 4> |
| 142 | + ret <4 x i32> %post |
| 143 | +} |
| 144 | +;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: |
| 145 | +; AVX: {{.*}} |
| 146 | +; SSE: {{.*}} |
0 commit comments