| 1 |
// Test array-copy-value pass (copy elision) with array assignment |
| 2 |
// involving Fortran pointers. Focus in only on wether copy ellision |
| 3 |
// is made or not. |
| 4 |
// RUN: fir-opt %s --array-value-copy -split-input-file | FileCheck --check-prefixes=ALL,NOOPT %s |
| 5 |
// RUN: fir-opt %s --array-value-copy="optimize-conflicts=true" -split-input-file | FileCheck --check-prefixes=ALL,OPT %s |
| 6 |
|
| 7 |
// Test `pointer(:) = array(:)` |
| 8 |
// ALL-LABEL: func @maybe_overlap |
| 9 |
// ALL: %[[ALLOC:.*]] = fir.allocmem !fir.array<100xf32> |
| 10 |
// ALL: fir.do_loop |
| 11 |
// ALL: fir.do_loop |
| 12 |
// ALL: fir.do_loop |
| 13 |
// ALL: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<100xf32>> |
| 14 |
func.func @maybe_overlap(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1 : !fir.ref<!fir.array<100xf32>> {fir.target}) { |
| 15 |
%c100 = arith.constant 100 : index |
| 16 |
%c99 = arith.constant 99 : index |
| 17 |
%c1 = arith.constant 1 : index |
| 18 |
%c0 = arith.constant 0 : index |
| 19 |
%0 = fir.alloca f32 |
| 20 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 21 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 22 |
%3 = fir.array_load %arg1(%1) : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 23 |
%4 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %2) -> (!fir.array<100xf32>) { |
| 24 |
%5 = fir.array_fetch %3, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 25 |
%6 = fir.array_update %arg3, %5, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 26 |
fir.result %6 : !fir.array<100xf32> |
| 27 |
} |
| 28 |
fir.array_merge_store %2, %4 to %arg0 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ptr<!fir.array<100xf32>> |
| 29 |
return |
| 30 |
} |
| 31 |
|
| 32 |
// ----- |
| 33 |
|
| 34 |
// Test `pointer(:) = array(:)` |
| 35 |
// ALL-LABEL: func @no_overlap1 |
| 36 |
// OPT-NOT: fir.allocmem |
| 37 |
// NOOPT: %[[ALLOC:.*]] = fir.allocmem !fir.array<100xf32> |
| 38 |
// NOOPT: fir.do_loop |
| 39 |
// NOOPT: fir.do_loop |
| 40 |
// NOOPT: fir.do_loop |
| 41 |
// NOOPT: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<100xf32>> |
| 42 |
func.func @no_overlap1(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1 : !fir.ref<!fir.array<100xf32>>) { |
| 43 |
%c100 = arith.constant 100 : index |
| 44 |
%c99 = arith.constant 99 : index |
| 45 |
%c1 = arith.constant 1 : index |
| 46 |
%c0 = arith.constant 0 : index |
| 47 |
%0 = fir.alloca f32 |
| 48 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 49 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 50 |
%3 = fir.array_load %arg1(%1) : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 51 |
%4 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %2) -> (!fir.array<100xf32>) { |
| 52 |
%5 = fir.array_fetch %3, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 53 |
%6 = fir.array_update %arg3, %5, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 54 |
fir.result %6 : !fir.array<100xf32> |
| 55 |
} |
| 56 |
fir.array_merge_store %2, %4 to %arg0 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ptr<!fir.array<100xf32>> |
| 57 |
return |
| 58 |
} |
| 59 |
|
| 60 |
// ----- |
| 61 |
|
| 62 |
// Test `pointer(:) = pointer(:)` |
| 63 |
// ALL-LABEL: func @no_overlap |
| 64 |
// ALL-NOT: fir.allocmem |
| 65 |
// ALL: fir.do_loop |
| 66 |
// ALL: fir.array_coor |
| 67 |
// ALL: fir.array_coor |
| 68 |
// ALL: fir.store |
| 69 |
func.func @no_overlap(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1: !fir.ref<!fir.array<100xf32>>) { |
| 70 |
%c100 = arith.constant 100 : index |
| 71 |
%c99 = arith.constant 99 : index |
| 72 |
%c1 = arith.constant 1 : index |
| 73 |
%c0 = arith.constant 0 : index |
| 74 |
%0 = fir.alloca f32 |
| 75 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 76 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 77 |
%3 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %2) -> (!fir.array<100xf32>) { |
| 78 |
%4 = fir.array_fetch %2, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 79 |
%5 = fir.array_update %arg3, %4, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 80 |
fir.result %5 : !fir.array<100xf32> |
| 81 |
} |
| 82 |
fir.array_merge_store %2, %3 to %arg0 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ptr<!fir.array<100xf32>> |
| 83 |
return |
| 84 |
} |
| 85 |
|
| 86 |
// ----- |
| 87 |
|
| 88 |
// Test `array(:) = pointer(:)` |
| 89 |
// ALL-LABEL: func @maybe_overlap_2 |
| 90 |
// ALL: %[[ALLOC:.*]] = fir.allocmem !fir.array<100xf32> |
| 91 |
// ALL: fir.do_loop |
| 92 |
// ALL: fir.do_loop |
| 93 |
// ALL: fir.do_loop |
| 94 |
// ALL: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<100xf32>> |
| 95 |
func.func @maybe_overlap_2(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1: !fir.ref<!fir.array<100xf32>> {fir.target}) { |
| 96 |
%c100 = arith.constant 100 : index |
| 97 |
%c99 = arith.constant 99 : index |
| 98 |
%c1 = arith.constant 1 : index |
| 99 |
%c0 = arith.constant 0 : index |
| 100 |
%0 = fir.alloca f32 |
| 101 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 102 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 103 |
%3 = fir.array_load %arg1(%1) : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 104 |
%4 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %3) -> (!fir.array<100xf32>) { |
| 105 |
%5 = fir.array_fetch %2, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 106 |
%6 = fir.array_update %arg3, %5, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 107 |
fir.result %6 : !fir.array<100xf32> |
| 108 |
} |
| 109 |
fir.array_merge_store %3, %4 to %arg1 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ref<!fir.array<100xf32>> |
| 110 |
return |
| 111 |
} |
| 112 |
|
| 113 |
// ----- |
| 114 |
|
| 115 |
// Test `array(:) = pointer(:)` |
| 116 |
// ALL-LABEL: func @no_overlap_2 |
| 117 |
// OPT-NOT: fir.allocmem |
| 118 |
// NOOPT: %[[ALLOC:.*]] = fir.allocmem !fir.array<100xf32> |
| 119 |
// NOOPT: fir.do_loop |
| 120 |
// NOOPT: fir.do_loop |
| 121 |
// NOOPT: fir.do_loop |
| 122 |
// NOOPT: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<100xf32>> |
| 123 |
func.func @no_overlap_2(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1: !fir.ref<!fir.array<100xf32>>) { |
| 124 |
%c100 = arith.constant 100 : index |
| 125 |
%c99 = arith.constant 99 : index |
| 126 |
%c1 = arith.constant 1 : index |
| 127 |
%c0 = arith.constant 0 : index |
| 128 |
%0 = fir.alloca f32 |
| 129 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 130 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 131 |
%3 = fir.array_load %arg1(%1) : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 132 |
%4 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %3) -> (!fir.array<100xf32>) { |
| 133 |
%5 = fir.array_fetch %2, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 134 |
%6 = fir.array_update %arg3, %5, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 135 |
fir.result %6 : !fir.array<100xf32> |
| 136 |
} |
| 137 |
fir.array_merge_store %3, %4 to %arg1 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ref<!fir.array<100xf32>> |
| 138 |
return |
| 139 |
} |
| 140 |
|
| 141 |
// ----- |
| 142 |
|
| 143 |
// Test `pointer1(:) = pointer2(:)` |
| 144 |
// ALL-LABEL: func @maybe_overlap_3 |
| 145 |
// ALL: %[[ALLOC:.*]] = fir.allocmem !fir.array<100xf32> |
| 146 |
// ALL: fir.do_loop |
| 147 |
// ALL: fir.do_loop |
| 148 |
// ALL: fir.do_loop |
| 149 |
// ALL: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<100xf32>> |
| 150 |
func.func @maybe_overlap_3(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1: !fir.ptr<!fir.array<100xf32>>) { |
| 151 |
%c100 = arith.constant 100 : index |
| 152 |
%c99 = arith.constant 99 : index |
| 153 |
%c1 = arith.constant 1 : index |
| 154 |
%c0 = arith.constant 0 : index |
| 155 |
%0 = fir.alloca f32 |
| 156 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 157 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 158 |
%3 = fir.array_load %arg1(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 159 |
%4 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %3) -> (!fir.array<100xf32>) { |
| 160 |
%5 = fir.array_fetch %2, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 161 |
%6 = fir.array_update %arg3, %5, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 162 |
fir.result %6 : !fir.array<100xf32> |
| 163 |
} |
| 164 |
fir.array_merge_store %3, %4 to %arg1 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ptr<!fir.array<100xf32>> |
| 165 |
return |
| 166 |
} |
| 167 |
|
| 168 |
// ----- |
| 169 |
|
| 170 |
// Test derived_target(:)%i = integer_pointer(:) |
| 171 |
// The integer pointer may be aliasing the derived target component. |
| 172 |
// ALL-LABEL: func @derived_whose_component_may_be_aliased |
| 173 |
// ALL: %[[ALLOC:.*]] = fir.allocmem !fir.array<4x!fir.type<some_type{i:i32}>> |
| 174 |
// ALL-COUNT-3: fir.do_loop |
| 175 |
// ALL: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<4x!fir.type<some_type{i:i32}>>> |
| 176 |
func.func @derived_whose_component_may_be_aliased(%arg0: !fir.box<!fir.array<4x!fir.type<some_type{i:i32}>>> {fir.target}, %arg1: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>) { |
| 177 |
%c4 = arith.constant 4 : index |
| 178 |
%0 = fir.field_index i, !fir.type<some_type{i:i32}> |
| 179 |
%c1 = arith.constant 1 : index |
| 180 |
%1 = fir.slice %c1, %c4, %c1 path %0 : (index, index, index, !fir.field) -> !fir.slice<1> |
| 181 |
%2 = fir.array_load %arg0 [%1] : (!fir.box<!fir.array<4x!fir.type<some_type{i:i32}>>>, !fir.slice<1>) -> !fir.array<4xi32> |
| 182 |
%3 = fir.load %arg1 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>> |
| 183 |
%c0 = arith.constant 0 : index |
| 184 |
%4:3 = fir.box_dims %3, %c0 : (!fir.box<!fir.ptr<!fir.array<?xi32>>>, index) -> (index, index, index) |
| 185 |
%5 = fir.shift %4#0 : (index) -> !fir.shift<1> |
| 186 |
%6 = fir.array_load %3(%5) : (!fir.box<!fir.ptr<!fir.array<?xi32>>>, !fir.shift<1>) -> !fir.array<?xi32> |
| 187 |
%7 = arith.subi %c4, %c1 : index |
| 188 |
%8 = fir.do_loop %arg2 = %c0 to %7 step %c1 unordered iter_args(%arg3 = %2) -> (!fir.array<4xi32>) { |
| 189 |
%9 = fir.array_fetch %6, %arg2 : (!fir.array<?xi32>, index) -> i32 |
| 190 |
%10 = fir.array_update %arg3, %9, %arg2 : (!fir.array<4xi32>, i32, index) -> !fir.array<4xi32> |
| 191 |
fir.result %10 : !fir.array<4xi32> |
| 192 |
} |
| 193 |
fir.array_merge_store %2, %8 to %arg0[%1] : !fir.array<4xi32>, !fir.array<4xi32>, !fir.box<!fir.array<4x!fir.type<some_type{i:i32}>>>, !fir.slice<1> |
| 194 |
return |
| 195 |
} |
| 196 |
|
| 197 |
// ----- |
| 198 |
|
| 199 |
// Test real_target = complex_target(:)%re |
| 200 |
// The real pointer may be aliasing the complex real part. |
| 201 |
// ALL-LABEL: func @complex_real_aliasing |
| 202 |
// ALL: %[[ALLOC:.*]] = fir.allocmem !fir.array<?xf32> |
| 203 |
// ALL-COUNT-3: fir.do_loop |
| 204 |
// ALL: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<?xf32>> |
| 205 |
func.func @complex_real_aliasing(%arg0: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>, %arg1: !fir.ref<!fir.array<4xcomplex<f32>>> {fir.target}) { |
| 206 |
%c4 = arith.constant 4 : index |
| 207 |
%0 = fir.load %arg0 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>> |
| 208 |
%c0 = arith.constant 0 : index |
| 209 |
%1:3 = fir.box_dims %0, %c0 : (!fir.box<!fir.ptr<!fir.array<?xf32>>>, index) -> (index, index, index) |
| 210 |
%2 = fir.shift %1#0 : (index) -> !fir.shift<1> |
| 211 |
%3 = fir.array_load %0(%2) : (!fir.box<!fir.ptr<!fir.array<?xf32>>>, !fir.shift<1>) -> !fir.array<?xf32> |
| 212 |
%c0_i32 = arith.constant 0 : i32 |
| 213 |
%4 = fir.shape %c4 : (index) -> !fir.shape<1> |
| 214 |
%c1 = arith.constant 1 : index |
| 215 |
%5 = fir.slice %c1, %c4, %c1 path %c0_i32 : (index, index, index, i32) -> !fir.slice<1> |
| 216 |
%6 = fir.array_load %arg1(%4) [%5] : (!fir.ref<!fir.array<4xcomplex<f32>>>, !fir.shape<1>, !fir.slice<1>) -> !fir.array<4xf32> |
| 217 |
%7 = arith.subi %c4, %c1 : index |
| 218 |
%8 = fir.do_loop %arg2 = %c0 to %7 step %c1 unordered iter_args(%arg3 = %3) -> (!fir.array<?xf32>) { |
| 219 |
%9 = fir.array_fetch %6, %arg2 : (!fir.array<4xf32>, index) -> f32 |
| 220 |
%10 = fir.array_update %arg3, %9, %arg2 : (!fir.array<?xf32>, f32, index) -> !fir.array<?xf32> |
| 221 |
fir.result %10 : !fir.array<?xf32> |
| 222 |
} |
| 223 |
fir.array_merge_store %3, %8 to %0 : !fir.array<?xf32>, !fir.array<?xf32>, !fir.box<!fir.ptr<!fir.array<?xf32>>> |
| 224 |
return |
| 225 |
} |
| 226 |
|
| 227 |
// ----- |
| 228 |
|
| 229 |
// Test `array(:) = pointer(:)` |
| 230 |
// ALL-LABEL: func @maybe_overlap_3 |
| 231 |
// ALL: %[[ALLOC:.*]] = fir.allocmem !fir.array<100xf32> |
| 232 |
// ALL: fir.do_loop |
| 233 |
// ALL: fir.do_loop |
| 234 |
// ALL: fir.do_loop |
| 235 |
// ALL: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<100xf32>> |
| 236 |
fir.global @_QMdataEglob target : !fir.array<100xf32> { |
| 237 |
%0 = fir.undefined !fir.array<100xf32> |
| 238 |
fir.has_value %0 : !fir.array<100xf32> |
| 239 |
} |
| 240 |
|
| 241 |
func.func @maybe_overlap_3(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1: !fir.ref<!fir.array<100xf32>> {fir.target}) { |
| 242 |
%c100 = arith.constant 100 : index |
| 243 |
%c99 = arith.constant 99 : index |
| 244 |
%c1 = arith.constant 1 : index |
| 245 |
%c0 = arith.constant 0 : index |
| 246 |
%0 = fir.address_of(@_QMdataEglob) : !fir.ref<!fir.array<100xf32>> |
| 247 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 248 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 249 |
%3 = fir.array_load %0(%1) : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 250 |
%4 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %3) -> (!fir.array<100xf32>) { |
| 251 |
%5 = fir.array_fetch %2, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 252 |
%6 = fir.array_update %arg3, %5, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 253 |
fir.result %6 : !fir.array<100xf32> |
| 254 |
} |
| 255 |
fir.array_merge_store %3, %4 to %0 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ref<!fir.array<100xf32>> |
| 256 |
return |
| 257 |
} |
| 258 |
|
| 259 |
// ----- |
| 260 |
|
| 261 |
// Test `array(:) = pointer(:)` |
| 262 |
// ALL-LABEL: func @no_overlap_3 |
| 263 |
// OPT-NOT: fir.allocmem |
| 264 |
// NOOPT: %[[ALLOC:.*]] = fir.allocmem !fir.array<100xf32> |
| 265 |
// NOOPT: fir.do_loop |
| 266 |
// NOOPT: fir.do_loop |
| 267 |
// NOOPT: fir.do_loop |
| 268 |
// NOOPT: fir.freemem %[[ALLOC]] : !fir.heap<!fir.array<100xf32>> |
| 269 |
fir.global @_QMdataEglob : !fir.array<100xf32> { |
| 270 |
%0 = fir.undefined !fir.array<100xf32> |
| 271 |
fir.has_value %0 : !fir.array<100xf32> |
| 272 |
} |
| 273 |
|
| 274 |
func.func @no_overlap_3(%arg0: !fir.ptr<!fir.array<100xf32>>, %arg1: !fir.ref<!fir.array<100xf32>> {fir.target}) { |
| 275 |
%c100 = arith.constant 100 : index |
| 276 |
%c99 = arith.constant 99 : index |
| 277 |
%c1 = arith.constant 1 : index |
| 278 |
%c0 = arith.constant 0 : index |
| 279 |
%0 = fir.address_of(@_QMdataEglob) : !fir.ref<!fir.array<100xf32>> |
| 280 |
%1 = fir.shape %c100 : (index) -> !fir.shape<1> |
| 281 |
%2 = fir.array_load %arg0(%1) : (!fir.ptr<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 282 |
%3 = fir.array_load %0(%1) : (!fir.ref<!fir.array<100xf32>>, !fir.shape<1>) -> !fir.array<100xf32> |
| 283 |
%4 = fir.do_loop %arg2 = %c0 to %c99 step %c1 unordered iter_args(%arg3 = %3) -> (!fir.array<100xf32>) { |
| 284 |
%5 = fir.array_fetch %2, %arg2 : (!fir.array<100xf32>, index) -> f32 |
| 285 |
%6 = fir.array_update %arg3, %5, %arg2 : (!fir.array<100xf32>, f32, index) -> !fir.array<100xf32> |
| 286 |
fir.result %6 : !fir.array<100xf32> |
| 287 |
} |
| 288 |
fir.array_merge_store %3, %4 to %0 : !fir.array<100xf32>, !fir.array<100xf32>, !fir.ref<!fir.array<100xf32>> |
| 289 |
return |
| 290 |
} |