| 1 |
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py |
| 2 |
; RUN: opt -passes=vector-combine -S -mtriple=aarch64 %s | FileCheck %s |
| 3 |
|
| 4 |
; This file tests the foldPermuteOfIntrinsic optimization which transforms: |
| 5 |
; shuffle(intrinsic(args), poison) -> intrinsic(shuffle(args)) |
| 6 |
; when the shuffle is a permute (operates on single vector) and cost model |
| 7 |
; determines the transformation is beneficial. |
| 8 |
|
| 9 |
;; ============================================================================ |
| 10 |
;; Positive Tests - Should Optimize |
| 11 |
;; ============================================================================ |
| 12 |
|
| 13 |
define <4 x i32> @extract_lower_sadd_sat(<8 x i32> %v1, <8 x i32> %v2) { |
| 14 |
; CHECK-LABEL: @extract_lower_sadd_sat( |
| 15 |
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[V1:%.*]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 16 |
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[V2:%.*]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 17 |
; CHECK-NEXT: [[RESULT:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[TMP1]], <4 x i32> [[TMP2]]) |
| 18 |
; CHECK-NEXT: ret <4 x i32> [[RESULT]] |
| 19 |
; |
| 20 |
%sat = call <8 x i32> @llvm.sadd.sat.v8i32(<8 x i32> %v1, <8 x i32> %v2) |
| 21 |
%result = shufflevector <8 x i32> %sat, <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 22 |
ret <4 x i32> %result |
| 23 |
} |
| 24 |
|
| 25 |
define <4 x i32> @extract_lower_uadd_sat(<8 x i32> %v1, <8 x i32> %v2) { |
| 26 |
; CHECK-LABEL: @extract_lower_uadd_sat( |
| 27 |
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[V1:%.*]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 28 |
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[V2:%.*]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 29 |
; CHECK-NEXT: [[RESULT:%.*]] = call <4 x i32> @llvm.uadd.sat.v4i32(<4 x i32> [[TMP1]], <4 x i32> [[TMP2]]) |
| 30 |
; CHECK-NEXT: ret <4 x i32> [[RESULT]] |
| 31 |
; |
| 32 |
%sat = call <8 x i32> @llvm.uadd.sat.v8i32(<8 x i32> %v1, <8 x i32> %v2) |
| 33 |
%result = shufflevector <8 x i32> %sat, <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 34 |
ret <4 x i32> %result |
| 35 |
} |
| 36 |
|
| 37 |
define <4 x float> @extract_lower_fma(<8 x float> %a, <8 x float> %b, <8 x float> %c) { |
| 38 |
; CHECK-LABEL: @extract_lower_fma( |
| 39 |
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <8 x float> [[A:%.*]], <8 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 40 |
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <8 x float> [[B:%.*]], <8 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 41 |
; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <8 x float> [[C:%.*]], <8 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 42 |
; CHECK-NEXT: [[RESULT:%.*]] = call <4 x float> @llvm.fma.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x float> [[TMP3]]) |
| 43 |
; CHECK-NEXT: ret <4 x float> [[RESULT]] |
| 44 |
; |
| 45 |
%fma = call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) |
| 46 |
%result = shufflevector <8 x float> %fma, <8 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 47 |
ret <4 x float> %result |
| 48 |
} |
| 49 |
|
| 50 |
define <4 x i32> @extract_lower_abs_should_not_shuffle_scalar(<8 x i32> %v) { |
| 51 |
; CHECK-LABEL: @extract_lower_abs_should_not_shuffle_scalar( |
| 52 |
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[V:%.*]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 53 |
; CHECK-NEXT: [[RESULT:%.*]] = call <4 x i32> @llvm.abs.v4i32(<4 x i32> [[TMP1]], i1 false) |
| 54 |
; CHECK-NEXT: ret <4 x i32> [[RESULT]] |
| 55 |
; |
| 56 |
%abs = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %v, i1 false) |
| 57 |
%result = shufflevector <8 x i32> %abs, <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 58 |
ret <4 x i32> %result |
| 59 |
} |
| 60 |
|
| 61 |
define <2 x i64> @extract_lower_i64(<4 x i64> %v1, <4 x i64> %v2) { |
| 62 |
; CHECK-LABEL: @extract_lower_i64( |
| 63 |
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i64> [[V1:%.*]], <4 x i64> poison, <2 x i32> <i32 0, i32 1> |
| 64 |
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i64> [[V2:%.*]], <4 x i64> poison, <2 x i32> <i32 0, i32 1> |
| 65 |
; CHECK-NEXT: [[RESULT:%.*]] = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> [[TMP1]], <2 x i64> [[TMP2]]) |
| 66 |
; CHECK-NEXT: ret <2 x i64> [[RESULT]] |
| 67 |
; |
| 68 |
%sat = call <4 x i64> @llvm.sadd.sat.v4i64(<4 x i64> %v1, <4 x i64> %v2) |
| 69 |
%result = shufflevector <4 x i64> %sat, <4 x i64> poison, <2 x i32> <i32 0, i32 1> |
| 70 |
ret <2 x i64> %result |
| 71 |
} |
| 72 |
|
| 73 |
define <8 x i16> @extract_lower_i16(<16 x i16> %v1, <16 x i16> %v2) { |
| 74 |
; CHECK-LABEL: @extract_lower_i16( |
| 75 |
; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <16 x i16> [[V1:%.*]], <16 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> |
| 76 |
; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[V2:%.*]], <16 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> |
| 77 |
; CHECK-NEXT: [[RESULT:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> [[TMP1]], <8 x i16> [[TMP2]]) |
| 78 |
; CHECK-NEXT: ret <8 x i16> [[RESULT]] |
| 79 |
; |
| 80 |
%sat = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %v1, <16 x i16> %v2) |
| 81 |
%result = shufflevector <16 x i16> %sat, <16 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> |
| 82 |
ret <8 x i16> %result |
| 83 |
} |
| 84 |
|
| 85 |
;; ============================================================================ |
| 86 |
;; Negative Tests - Should NOT Optimize |
| 87 |
;; ============================================================================ |
| 88 |
|
| 89 |
define <4 x i32> @same_size_permute(<4 x i32> %v1, <4 x i32> %v2) { |
| 90 |
; CHECK-LABEL: @same_size_permute( |
| 91 |
; CHECK-NEXT: [[SAT:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[V1:%.*]], <4 x i32> [[V2:%.*]]) |
| 92 |
; CHECK-NEXT: [[RESULT:%.*]] = shufflevector <4 x i32> [[SAT]], <4 x i32> poison, <4 x i32> <i32 2, i32 0, i32 3, i32 1> |
| 93 |
; CHECK-NEXT: ret <4 x i32> [[RESULT]] |
| 94 |
; |
| 95 |
%sat = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> %v1, <4 x i32> %v2) |
| 96 |
%result = shufflevector <4 x i32> %sat, <4 x i32> poison, <4 x i32> <i32 2, i32 0, i32 3, i32 1> |
| 97 |
ret <4 x i32> %result |
| 98 |
} |
| 99 |
|
| 100 |
define <4 x i32> @not_a_permute_uses_second_operand(<4 x i32> %v1, <4 x i32> %v2, <4 x i32> %other) { |
| 101 |
; CHECK-LABEL: @not_a_permute_uses_second_operand( |
| 102 |
; CHECK-NEXT: [[SAT:%.*]] = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> [[V1:%.*]], <4 x i32> [[V2:%.*]]) |
| 103 |
; CHECK-NEXT: [[RESULT:%.*]] = shufflevector <4 x i32> [[SAT]], <4 x i32> [[OTHER:%.*]], <4 x i32> <i32 0, i32 4, i32 1, i32 5> |
| 104 |
; CHECK-NEXT: ret <4 x i32> [[RESULT]] |
| 105 |
; |
| 106 |
%sat = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> %v1, <4 x i32> %v2) |
| 107 |
%result = shufflevector <4 x i32> %sat, <4 x i32> %other, <4 x i32> <i32 0, i32 4, i32 1, i32 5> |
| 108 |
ret <4 x i32> %result |
| 109 |
} |
| 110 |
|
| 111 |
define <4 x i32> @not_an_intrinsic(<8 x i32> %v1, <8 x i32> %v2) { |
| 112 |
; CHECK-LABEL: @not_an_intrinsic( |
| 113 |
; CHECK-NEXT: [[ADD:%.*]] = add <8 x i32> [[V1:%.*]], [[V2:%.*]] |
| 114 |
; CHECK-NEXT: [[RESULT:%.*]] = shufflevector <8 x i32> [[ADD]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 115 |
; CHECK-NEXT: ret <4 x i32> [[RESULT]] |
| 116 |
; |
| 117 |
%add = add <8 x i32> %v1, %v2 |
| 118 |
%result = shufflevector <8 x i32> %add, <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3> |
| 119 |
ret <4 x i32> %result |
| 120 |
} |
| 121 |
|
| 122 |
declare <8 x i32> @llvm.sadd.sat.v8i32(<8 x i32>, <8 x i32>) |
| 123 |
declare <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32>, <4 x i32>) |
| 124 |
declare <4 x i64> @llvm.sadd.sat.v4i64(<4 x i64>, <4 x i64>) |
| 125 |
declare <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64>, <2 x i64>) |
| 126 |
declare <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16>, <16 x i16>) |
| 127 |
declare <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16>, <8 x i16>) |
| 128 |
|
| 129 |
declare <8 x i32> @llvm.uadd.sat.v8i32(<8 x i32>, <8 x i32>) |
| 130 |
declare <4 x i32> @llvm.uadd.sat.v4i32(<4 x i32>, <4 x i32>) |
| 131 |
|
| 132 |
declare <8 x i32> @llvm.abs.v8i32(<8 x i32>, i1 immarg) |
| 133 |
declare <4 x i32> @llvm.abs.v4i32(<4 x i32>, i1 immarg) |
| 134 |
|
| 135 |
declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) |
| 136 |
declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) |