|
1 | 1 | ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 5 |
2 | 2 | ; RUN: opt -p loop-vectorize -mtriple riscv64 -mattr=+v -S < %s | FileCheck %s |
| 3 | +; RUN: opt -p loop-vectorize -mtriple riscv64 -mattr=+v -vectorizer-consider-reg-pressure=true -S < %s | FileCheck %s |
| 4 | +; RUN: opt -p loop-vectorize -mtriple riscv64 -mattr=+v -vectorizer-consider-reg-pressure=false -S < %s | FileCheck %s --check-prefix=NO-REG-PRESSURE-CHECK |
3 | 5 |
|
4 | 6 | define void @f(ptr noalias %p0, ptr noalias %p1, ptr noalias %p2) { |
5 | 7 | ; CHECK-LABEL: define void @f( |
@@ -94,6 +96,98 @@ define void @f(ptr noalias %p0, ptr noalias %p1, ptr noalias %p2) { |
94 | 96 | ; CHECK: [[EXIT]]: |
95 | 97 | ; CHECK-NEXT: ret void |
96 | 98 | ; |
| 99 | +; NO-REG-PRESSURE-CHECK-LABEL: define void @f( |
| 100 | +; NO-REG-PRESSURE-CHECK-SAME: ptr noalias [[P0:%.*]], ptr noalias [[P1:%.*]], ptr noalias [[P2:%.*]]) #[[ATTR0:[0-9]+]] { |
| 101 | +; NO-REG-PRESSURE-CHECK-NEXT: [[ENTRY:.*:]] |
| 102 | +; NO-REG-PRESSURE-CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]] |
| 103 | +; NO-REG-PRESSURE-CHECK: [[VECTOR_PH]]: |
| 104 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP0:%.*]] = call <vscale x 8 x i64> @llvm.stepvector.nxv8i64() |
| 105 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP1:%.*]] = mul <vscale x 8 x i64> [[TMP0]], splat (i64 2) |
| 106 | +; NO-REG-PRESSURE-CHECK-NEXT: [[INDUCTION:%.*]] = add <vscale x 8 x i64> zeroinitializer, [[TMP1]] |
| 107 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 8 x i64> @llvm.stepvector.nxv8i64() |
| 108 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP3:%.*]] = mul <vscale x 8 x i64> [[TMP2]], splat (i64 3) |
| 109 | +; NO-REG-PRESSURE-CHECK-NEXT: [[INDUCTION1:%.*]] = add <vscale x 8 x i64> zeroinitializer, [[TMP3]] |
| 110 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 8 x i64> @llvm.stepvector.nxv8i64() |
| 111 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP5:%.*]] = mul <vscale x 8 x i64> [[TMP4]], splat (i64 4) |
| 112 | +; NO-REG-PRESSURE-CHECK-NEXT: [[INDUCTION2:%.*]] = add <vscale x 8 x i64> zeroinitializer, [[TMP5]] |
| 113 | +; NO-REG-PRESSURE-CHECK-NEXT: br label %[[VECTOR_BODY:.*]] |
| 114 | +; NO-REG-PRESSURE-CHECK: [[VECTOR_BODY]]: |
| 115 | +; NO-REG-PRESSURE-CHECK-NEXT: [[EVL_BASED_IV:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_EVL_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| 116 | +; NO-REG-PRESSURE-CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 8 x i64> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| 117 | +; NO-REG-PRESSURE-CHECK-NEXT: [[VEC_IND3:%.*]] = phi <vscale x 8 x i64> [ [[INDUCTION1]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT11:%.*]], %[[VECTOR_BODY]] ] |
| 118 | +; NO-REG-PRESSURE-CHECK-NEXT: [[VEC_IND4:%.*]] = phi <vscale x 8 x i64> [ [[INDUCTION2]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT12:%.*]], %[[VECTOR_BODY]] ] |
| 119 | +; NO-REG-PRESSURE-CHECK-NEXT: [[AVL:%.*]] = phi i64 [ 1025, %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ] |
| 120 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 8, i1 true) |
| 121 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP7:%.*]] = zext i32 [[TMP6]] to i64 |
| 122 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP8:%.*]] = mul i64 4, [[TMP7]] |
| 123 | +; NO-REG-PRESSURE-CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[TMP8]], i64 0 |
| 124 | +; NO-REG-PRESSURE-CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 125 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP9:%.*]] = zext i32 [[TMP6]] to i64 |
| 126 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP10:%.*]] = mul i64 3, [[TMP9]] |
| 127 | +; NO-REG-PRESSURE-CHECK-NEXT: [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[TMP10]], i64 0 |
| 128 | +; NO-REG-PRESSURE-CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 8 x i64> [[BROADCAST_SPLATINSERT5]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 129 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP11:%.*]] = zext i32 [[TMP6]] to i64 |
| 130 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP12:%.*]] = mul i64 2, [[TMP11]] |
| 131 | +; NO-REG-PRESSURE-CHECK-NEXT: [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[TMP12]], i64 0 |
| 132 | +; NO-REG-PRESSURE-CHECK-NEXT: [[BROADCAST_SPLAT8:%.*]] = shufflevector <vscale x 8 x i64> [[BROADCAST_SPLATINSERT7]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 133 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP13:%.*]] = sub <vscale x 8 x i64> [[VEC_IND]], splat (i64 1) |
| 134 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[P0]], <vscale x 8 x i64> [[TMP13]] |
| 135 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 8 x i8> @llvm.vp.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> align 1 [[TMP14]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]]) |
| 136 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP15:%.*]] = sub <vscale x 8 x i64> [[VEC_IND3]], splat (i64 1) |
| 137 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[P0]], <vscale x 8 x i64> [[TMP15]] |
| 138 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_MASKED_GATHER9:%.*]] = call <vscale x 8 x i8> @llvm.vp.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> align 1 [[TMP16]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]]) |
| 139 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP17:%.*]] = sub <vscale x 8 x i64> [[VEC_IND4]], splat (i64 1) |
| 140 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP18:%.*]] = getelementptr i8, ptr [[P0]], <vscale x 8 x i64> [[TMP17]] |
| 141 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_MASKED_GATHER10:%.*]] = call <vscale x 8 x i8> @llvm.vp.gather.nxv8i8.nxv8p0(<vscale x 8 x ptr> align 1 [[TMP18]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP6]]) |
| 142 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP19:%.*]] = mul i64 [[EVL_BASED_IV]], 3 |
| 143 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[P1]], i64 [[TMP19]] |
| 144 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP21:%.*]] = getelementptr i8, ptr [[TMP20]], i8 0 |
| 145 | +; NO-REG-PRESSURE-CHECK-NEXT: [[INTERLEAVE_EVL:%.*]] = mul nuw nsw i32 [[TMP6]], 3 |
| 146 | +; NO-REG-PRESSURE-CHECK-NEXT: [[INTERLEAVED_VEC:%.*]] = call <vscale x 24 x i8> @llvm.vector.interleave3.nxv24i8(<vscale x 8 x i8> [[WIDE_MASKED_GATHER]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER9]], <vscale x 8 x i8> [[WIDE_MASKED_GATHER10]]) |
| 147 | +; NO-REG-PRESSURE-CHECK-NEXT: call void @llvm.vp.store.nxv24i8.p0(<vscale x 24 x i8> [[INTERLEAVED_VEC]], ptr align 1 [[TMP21]], <vscale x 24 x i1> splat (i1 true), i32 [[INTERLEAVE_EVL]]) |
| 148 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP22:%.*]] = zext i32 [[TMP6]] to i64 |
| 149 | +; NO-REG-PRESSURE-CHECK-NEXT: [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP22]], [[EVL_BASED_IV]] |
| 150 | +; NO-REG-PRESSURE-CHECK-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP22]] |
| 151 | +; NO-REG-PRESSURE-CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 8 x i64> [[VEC_IND]], [[BROADCAST_SPLAT8]] |
| 152 | +; NO-REG-PRESSURE-CHECK-NEXT: [[VEC_IND_NEXT11]] = add <vscale x 8 x i64> [[VEC_IND3]], [[BROADCAST_SPLAT6]] |
| 153 | +; NO-REG-PRESSURE-CHECK-NEXT: [[VEC_IND_NEXT12]] = add <vscale x 8 x i64> [[VEC_IND4]], [[BROADCAST_SPLAT]] |
| 154 | +; NO-REG-PRESSURE-CHECK-NEXT: [[TMP23:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 |
| 155 | +; NO-REG-PRESSURE-CHECK-NEXT: br i1 [[TMP23]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| 156 | +; NO-REG-PRESSURE-CHECK: [[MIDDLE_BLOCK]]: |
| 157 | +; NO-REG-PRESSURE-CHECK-NEXT: br label %[[EXIT:.*]] |
| 158 | +; NO-REG-PRESSURE-CHECK: [[SCALAR_PH]]: |
| 159 | +; NO-REG-PRESSURE-CHECK-NEXT: br label %[[LOOP:.*]] |
| 160 | +; NO-REG-PRESSURE-CHECK: [[LOOP]]: |
| 161 | +; NO-REG-PRESSURE-CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] |
| 162 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_0:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[WIDE_IV_0_NEXT:%.*]], %[[LOOP]] ] |
| 163 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_1:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[WIDE_IV_1_NEXT:%.*]], %[[LOOP]] ] |
| 164 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_2:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[WIDE_IV_2_NEXT:%.*]], %[[LOOP]] ] |
| 165 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_0_SUB:%.*]] = sub i64 [[WIDE_IV_0]], 1 |
| 166 | +; NO-REG-PRESSURE-CHECK-NEXT: [[A_GEP0:%.*]] = getelementptr i8, ptr [[P0]], i64 [[WIDE_IV_0_SUB]] |
| 167 | +; NO-REG-PRESSURE-CHECK-NEXT: [[A:%.*]] = load i8, ptr [[A_GEP0]], align 1 |
| 168 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_1_SUB:%.*]] = sub i64 [[WIDE_IV_1]], 1 |
| 169 | +; NO-REG-PRESSURE-CHECK-NEXT: [[B_GEP0:%.*]] = getelementptr i8, ptr [[P0]], i64 [[WIDE_IV_1_SUB]] |
| 170 | +; NO-REG-PRESSURE-CHECK-NEXT: [[B:%.*]] = load i8, ptr [[B_GEP0]], align 1 |
| 171 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_2_SUB:%.*]] = sub i64 [[WIDE_IV_2]], 1 |
| 172 | +; NO-REG-PRESSURE-CHECK-NEXT: [[C_GEP0:%.*]] = getelementptr i8, ptr [[P0]], i64 [[WIDE_IV_2_SUB]] |
| 173 | +; NO-REG-PRESSURE-CHECK-NEXT: [[C:%.*]] = load i8, ptr [[C_GEP0]], align 1 |
| 174 | +; NO-REG-PRESSURE-CHECK-NEXT: [[IV_MUL:%.*]] = mul i64 [[IV]], 3 |
| 175 | +; NO-REG-PRESSURE-CHECK-NEXT: [[BASE:%.*]] = getelementptr i8, ptr [[P1]], i64 [[IV_MUL]] |
| 176 | +; NO-REG-PRESSURE-CHECK-NEXT: [[A_GEP1:%.*]] = getelementptr i8, ptr [[BASE]], i8 0 |
| 177 | +; NO-REG-PRESSURE-CHECK-NEXT: store i8 [[A]], ptr [[A_GEP1]], align 1 |
| 178 | +; NO-REG-PRESSURE-CHECK-NEXT: [[B_GEP1:%.*]] = getelementptr i8, ptr [[BASE]], i8 1 |
| 179 | +; NO-REG-PRESSURE-CHECK-NEXT: store i8 [[B]], ptr [[B_GEP1]], align 1 |
| 180 | +; NO-REG-PRESSURE-CHECK-NEXT: [[C_GEP1:%.*]] = getelementptr i8, ptr [[BASE]], i8 2 |
| 181 | +; NO-REG-PRESSURE-CHECK-NEXT: store i8 [[C]], ptr [[C_GEP1]], align 1 |
| 182 | +; NO-REG-PRESSURE-CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1 |
| 183 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_0_NEXT]] = add i64 [[WIDE_IV_0]], 2 |
| 184 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_1_NEXT]] = add i64 [[WIDE_IV_1]], 3 |
| 185 | +; NO-REG-PRESSURE-CHECK-NEXT: [[WIDE_IV_2_NEXT]] = add i64 [[WIDE_IV_2]], 4 |
| 186 | +; NO-REG-PRESSURE-CHECK-NEXT: [[DONE:%.*]] = icmp eq i64 [[IV]], 1024 |
| 187 | +; NO-REG-PRESSURE-CHECK-NEXT: br i1 [[DONE]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP4:![0-9]+]] |
| 188 | +; NO-REG-PRESSURE-CHECK: [[EXIT]]: |
| 189 | +; NO-REG-PRESSURE-CHECK-NEXT: ret void |
| 190 | +; |
97 | 191 | entry: |
98 | 192 | br label %loop |
99 | 193 |
|
|
0 commit comments