| 1 | //===-- XeVMDialect.cpp - XeVM dialect registration -------------*- C++ -*-===// |
| 2 | // |
| 3 | // This file is licensed under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | #include "mlir/Dialect/LLVMIR/XeVMDialect.h" |
| 9 | #include "mlir/Dialect/GPU/IR/CompilationInterfaces.h" |
| 10 | #include "mlir/Dialect/LLVMIR/LLVMDialect.h" |
| 11 | #include "mlir/Dialect/Utils/StaticValueUtils.h" |
| 12 | #include "mlir/IR/DialectImplementation.h" |
| 13 | #include "llvm/ADT/TypeSwitch.h" |
| 14 | #include "llvm/Support/FileSystem.h" |
| 15 | #include "llvm/Support/MathExtras.h" |
| 16 | |
| 17 | using namespace mlir; |
| 18 | using namespace mlir::xevm; |
| 19 | |
| 20 | #include "mlir/Dialect/LLVMIR/XeVMOpsDialect.cpp.inc" |
| 21 | #include "mlir/Dialect/LLVMIR/XeVMOpsEnums.cpp.inc" |
| 22 | |
| 23 | namespace { |
| 24 | static constexpr uint32_t subgroupSize = 16; |
| 25 | |
| 26 | template <typename Op> |
| 27 | LogicalResult verifyMatrixInput(Op op) { |
| 28 | static_assert(llvm::is_one_of<Op, BlockLoad2dOp, BlockStore2dOp, |
| 29 | BlockPrefetch2dOp>::value, |
| 30 | "Unexpected template parameter" ); |
| 31 | |
| 32 | std::optional<int64_t> width = getConstantIntValue(op.getBaseWidth()); |
| 33 | std::optional<int64_t> pitch = getConstantIntValue(op.getBasePitch()); |
| 34 | if (pitch && width && *pitch < *width) |
| 35 | return op->emitOpError( |
| 36 | "4th operand (base pitch) should be >= 2nd operand (base width)" ); |
| 37 | |
| 38 | uint32_t elemSize = op.getElemSizeInBits(); |
| 39 | if (elemSize < 8 || !llvm::isPowerOf2_32(Value: elemSize) || elemSize > 32) |
| 40 | return op->emitOpError("expecting 'elem_size_in_bits' to be 8, 16, or 32" ); |
| 41 | |
| 42 | uint32_t tileHeight = op.getTileHeight(); |
| 43 | if (tileHeight > 32 || !llvm::isPowerOf2_32(Value: tileHeight)) |
| 44 | return op->emitOpError("expecting tile_height to be 1, 2, 4, 8, 16, or 32" ); |
| 45 | |
| 46 | uint32_t vBlocks = op.getVBlocks(); |
| 47 | if (vBlocks > 8 || !llvm::isPowerOf2_32(Value: vBlocks)) |
| 48 | return op->emitOpError("expecting v_blocks to be 1, 2, 4, or 8" ); |
| 49 | |
| 50 | return success(); |
| 51 | } |
| 52 | |
| 53 | LogicalResult verify2DBlockLoadRestriction(BlockLoad2dOp op) { |
| 54 | VectorType resTy = op.getRes().getType(); |
| 55 | if (!resTy.getElementType().isIntOrFloat()) |
| 56 | return op.emitOpError() |
| 57 | << "expecting result element type to be int or float" ; |
| 58 | unsigned resElemTySize = resTy.getElementType().getIntOrFloatBitWidth(); |
| 59 | unsigned resSize = resTy.getNumElements() * resElemTySize; |
| 60 | unsigned expectedSize = op.getElemSizeInBits() * op.getTileHeight() * |
| 61 | op.getTileWidth() * op.getVBlocks() / subgroupSize; |
| 62 | if (resSize != expectedSize) |
| 63 | return op.emitOpError() << "result size of " << resSize |
| 64 | << " bits does not match the expected size of " |
| 65 | << expectedSize << " bits" ; |
| 66 | |
| 67 | if (op.getTranspose() && op.getPackRegister()) |
| 68 | return op.emitOpError(message: "transpose and pack_register are mutually exclusive" ); |
| 69 | |
| 70 | if (!op.getTranspose() && !op.getPackRegister()) { |
| 71 | uint32_t tileHeight = op.getTileHeight(); |
| 72 | if (tileHeight < 1 || tileHeight > 32) |
| 73 | return op.emitOpError(message: "expecting tile_height to be between 1 and 32" ); |
| 74 | |
| 75 | uint32_t tileWidth = op.getTileWidth(); |
| 76 | uint32_t vBlocks = op.getVBlocks(); |
| 77 | switch (op.getElemSizeInBits()) { |
| 78 | case 8: |
| 79 | if (tileWidth < 4 || tileWidth > 64) |
| 80 | return op.emitOpError(message: "expecting tile_width to be between 4 and 64" ); |
| 81 | if (vBlocks != 1 && vBlocks != 2 && vBlocks != 4) |
| 82 | return op.emitOpError(message: "expecting v_blocks to be 1, 2, or 4" ); |
| 83 | if (tileWidth * vBlocks > 64) |
| 84 | return op.emitOpError( |
| 85 | message: "tile_width * v_blocks should be less than or equal " |
| 86 | "to 64 for 8 bit elements" ); |
| 87 | break; |
| 88 | case 16: |
| 89 | if (tileWidth < 2 || tileWidth > 32) |
| 90 | return op.emitOpError(message: "expecting tile_width to be between 2 and 32" ); |
| 91 | if (vBlocks != 1 && vBlocks != 2 && vBlocks != 4) |
| 92 | return op.emitOpError(message: "expecting v_blocks to be 1, 2, or 4" ); |
| 93 | if (tileWidth * vBlocks > 32) |
| 94 | return op.emitOpError( |
| 95 | message: "tile_width * v_blocks should be less than or equal " |
| 96 | "to 32 for 16 bit elements" ); |
| 97 | break; |
| 98 | case 32: |
| 99 | if (tileWidth < 1 || tileWidth > 16) |
| 100 | return op.emitOpError(message: "expecting tile_width to be between 1 and 16" ); |
| 101 | if (vBlocks != 1 && vBlocks != 2) |
| 102 | return op.emitOpError(message: "expecting v_blocks to be 1 or 2" ); |
| 103 | if (tileWidth * vBlocks > 16) |
| 104 | return op.emitOpError( |
| 105 | message: "tile_width * v_blocks should be less than or equal " |
| 106 | "to 16 for 32 bit elements" ); |
| 107 | break; |
| 108 | case 64: |
| 109 | if (tileWidth < 1 || tileWidth > 8) |
| 110 | return op.emitOpError(message: "expecting tile_width to be between 1 and 8" ); |
| 111 | if (vBlocks != 1) |
| 112 | return op.emitOpError(message: "expecting v_blocks to be 1" ); |
| 113 | break; |
| 114 | default: |
| 115 | return op.emitOpError( |
| 116 | message: "expecting elem_size_in_bits to be 8, 16, 32, or 64" ); |
| 117 | } |
| 118 | |
| 119 | return success(); |
| 120 | } |
| 121 | |
| 122 | if (op.getTranspose()) { |
| 123 | assert(!op.getPackRegister() && "Expecting pack_register should be false" ); |
| 124 | |
| 125 | uint32_t vBlocks = op.getVBlocks(); |
| 126 | if (vBlocks != 1) |
| 127 | return op.emitOpError(message: "expecting v_blocks to be 1" ); |
| 128 | |
| 129 | uint32_t tileHeight = op.getTileHeight(); |
| 130 | uint32_t tileWidth = op.getTileWidth(); |
| 131 | switch (op.getElemSizeInBits()) { |
| 132 | case 32: |
| 133 | if (tileHeight < 1 || tileHeight > 32) |
| 134 | return op.emitOpError(message: "expecting tile_height to be between 1 and 32" ); |
| 135 | if (tileWidth < 1 || tileWidth > 8) |
| 136 | return op.emitOpError(message: "expecting tile_width to be between 1 and 8" ); |
| 137 | break; |
| 138 | case 64: |
| 139 | if (tileHeight != 8) |
| 140 | return op.emitOpError( |
| 141 | message: "expecting tile_height to be 8 for 64 bit elements" ); |
| 142 | if (tileWidth != 1 && tileWidth != 2 && tileWidth != 4) |
| 143 | return op.emitOpError(message: "expecting tile_width to be 1, 2, or 4" ); |
| 144 | break; |
| 145 | default: |
| 146 | return op.emitOpError(message: "transpose is only supported for 32 and 64 bit " |
| 147 | "elements" ); |
| 148 | } |
| 149 | |
| 150 | return success(); |
| 151 | } |
| 152 | |
| 153 | assert(op.getPackRegister() && !op.getTranspose() && |
| 154 | "Expecting pack_register should be true and transpose should be " |
| 155 | "false" ); |
| 156 | |
| 157 | uint32_t vBlocks = op.getVBlocks(); |
| 158 | if (vBlocks != 1 && vBlocks != 2 && vBlocks != 4) |
| 159 | return op.emitOpError(message: "expecting v_blocks to be 1, 2, or 4" ); |
| 160 | |
| 161 | uint32_t tileHeight = op.getTileHeight(); |
| 162 | uint32_t tileWidth = op.getTileWidth(); |
| 163 | switch (op.getElemSizeInBits()) { |
| 164 | case 8: |
| 165 | if (tileHeight < 4 || tileHeight > 32) |
| 166 | return op.emitOpError(message: "expecting tile_height to be between 4 and 32" ); |
| 167 | if (tileWidth < 4 || tileWidth > 16) |
| 168 | return op.emitOpError(message: "expecting tile_width to be between 4 and 16" ); |
| 169 | break; |
| 170 | case 16: |
| 171 | if (tileHeight < 2 || tileHeight > 32) |
| 172 | return op.emitOpError(message: "expecting tile_height to be between 2 and 32" ); |
| 173 | if (tileWidth < 2 || tileWidth > 16) |
| 174 | return op.emitOpError(message: "expecting tile_width to be between 2 and 16" ); |
| 175 | if (tileWidth * vBlocks > 32) |
| 176 | return op.emitOpError( |
| 177 | message: "tile_width * v_blocks should be less than or equal " |
| 178 | "to 32 for 16 bit elements" ); |
| 179 | break; |
| 180 | default: |
| 181 | return op.emitOpError(message: "pack_register is only supported for 8 and 16 bit " |
| 182 | "elements" ); |
| 183 | } |
| 184 | |
| 185 | return success(); |
| 186 | } |
| 187 | |
| 188 | static LogicalResult verify2DBlockStoreRestriction(BlockStore2dOp op) { |
| 189 | uint32_t tileHeight = op.getTileHeight(); |
| 190 | if (tileHeight < 1 || tileHeight > 8) |
| 191 | return op.emitOpError(message: "expecting tile_height to be between 1 and 8" ); |
| 192 | |
| 193 | uint32_t tileWidth = op.getTileWidth(); |
| 194 | switch (op.getElemSizeInBits()) { |
| 195 | case 8: |
| 196 | if (tileWidth < 4 || tileWidth > 64) |
| 197 | return op.emitOpError(message: "expecting tile_width to be between 4 and 64" ); |
| 198 | break; |
| 199 | case 16: |
| 200 | if (tileWidth < 2 || tileWidth > 32) |
| 201 | return op.emitOpError(message: "expecting tile_width to be between 2 and 32" ); |
| 202 | break; |
| 203 | case 32: |
| 204 | if (tileWidth < 1 || tileWidth > 16) |
| 205 | return op.emitOpError(message: "expecting tile_width to be between 1 and 16" ); |
| 206 | break; |
| 207 | case 64: |
| 208 | if (tileWidth < 1 || tileWidth > 8) |
| 209 | return op.emitOpError(message: "expecting tile_width to be between 1 and 8" ); |
| 210 | break; |
| 211 | default: |
| 212 | return op.emitOpError(message: "expecting elem_size_in_bits to be 8, 16, 32, or 64" ); |
| 213 | } |
| 214 | |
| 215 | uint32_t vBlocks = op.getVBlocks(); |
| 216 | if (vBlocks != 1) |
| 217 | return op.emitOpError(message: "expecting v_blocks to be 1" ); |
| 218 | return success(); |
| 219 | } |
| 220 | |
| 221 | } // namespace |
| 222 | |
| 223 | LogicalResult BlockLoad2dOp::verify() { |
| 224 | if (verify2DBlockLoadRestriction(op: *this).failed()) |
| 225 | return failure(); |
| 226 | |
| 227 | if (verifyMatrixInput(op: *this).failed()) |
| 228 | return failure(); |
| 229 | |
| 230 | VectorType resTy = getRes().getType(); |
| 231 | if (!resTy.getElementType().isIntOrFloat()) |
| 232 | return emitOpError() << "expecting result element type to be int of float" ; |
| 233 | unsigned resElemTySize = resTy.getElementType().getIntOrFloatBitWidth(); |
| 234 | if (getElemSizeInBits() == 32 || getPackRegister()) { |
| 235 | if (resElemTySize != 32) |
| 236 | return emitOpError() << "expecting result element type to be 32 bits" ; |
| 237 | } |
| 238 | |
| 239 | uint32_t tileWidth = getTileWidth(); |
| 240 | if (getPackRegister()) { |
| 241 | if (tileWidth != 16) |
| 242 | return emitOpError( |
| 243 | message: "tile_width when pack_register is true should be equal " |
| 244 | "to subgroup size (16 elements)" ); |
| 245 | return success(); |
| 246 | } |
| 247 | |
| 248 | return success(); |
| 249 | } |
| 250 | |
| 251 | LogicalResult BlockStore2dOp::verify() { |
| 252 | if (verify2DBlockStoreRestriction(op: *this).failed()) |
| 253 | return failure(); |
| 254 | |
| 255 | if (verifyMatrixInput(op: *this).failed()) |
| 256 | return failure(); |
| 257 | |
| 258 | uint32_t tileWidth = getTileWidth(); |
| 259 | switch (getElemSizeInBits()) { |
| 260 | case 8: |
| 261 | if (tileWidth != 16 && tileWidth != 32) |
| 262 | return emitOpError(message: "tile_width for 8 bit elements should be equal to " |
| 263 | "16 or 32" ); |
| 264 | break; |
| 265 | case 16: |
| 266 | if (tileWidth != 16) |
| 267 | return emitOpError(message: "tile_width for 16 bit elements should be equal " |
| 268 | "to 16" ); |
| 269 | break; |
| 270 | case 32: |
| 271 | if (tileWidth != 16) |
| 272 | return emitOpError(message: "tile_width for 32 bit elements should be equal " |
| 273 | "to 16" ); |
| 274 | break; |
| 275 | default: |
| 276 | llvm_unreachable("unexpected element size" ); |
| 277 | } |
| 278 | |
| 279 | return success(); |
| 280 | } |
| 281 | |
| 282 | LogicalResult BlockPrefetch2dOp::verify() { |
| 283 | if (verifyMatrixInput(op: *this).failed()) |
| 284 | return failure(); |
| 285 | |
| 286 | uint32_t tileWidth = getTileWidth(); |
| 287 | switch (getElemSizeInBits()) { |
| 288 | case 8: |
| 289 | if (tileWidth != 16 && tileWidth != 32) |
| 290 | return emitOpError(message: "tile_width for 8 bit elements should be equal to " |
| 291 | "16 or 32" ); |
| 292 | break; |
| 293 | case 16: |
| 294 | if (tileWidth != 16) |
| 295 | return emitOpError(message: "tile_width for 16 bit elements should be equal " |
| 296 | "to 16" ); |
| 297 | break; |
| 298 | case 32: |
| 299 | if (tileWidth != 8 && tileWidth != 16) |
| 300 | return emitOpError( |
| 301 | message: "tile_width for 32 bit elements should be equal to 8 or 16" ); |
| 302 | break; |
| 303 | default: |
| 304 | llvm_unreachable("unexpected element size" ); |
| 305 | } |
| 306 | |
| 307 | return success(); |
| 308 | } |
| 309 | |
| 310 | LogicalResult MMAOp::verify() { |
| 311 | if (getC()) { |
| 312 | if (getResult().getType() != getC().getType()) |
| 313 | return emitOpError(message: "type of C operand must match result type" ); |
| 314 | } |
| 315 | return success(); |
| 316 | } |
| 317 | |
| 318 | LogicalResult |
| 319 | XeVMTargetAttr::verify(function_ref<InFlightDiagnostic()> emitError, int O, |
| 320 | StringRef triple, StringRef chip, DictionaryAttr flags, |
| 321 | ArrayAttr linkFiles) { |
| 322 | if (O < 0 || O > 3) { |
| 323 | return emitError() |
| 324 | << "The optimization level must be a number between 0 and 3." ; |
| 325 | } |
| 326 | if (triple.empty()) { |
| 327 | return emitError() << "The target triple cannot be empty." ; |
| 328 | } |
| 329 | if (chip.empty()) { |
| 330 | return emitError() << "The target chip cannot be empty." ; |
| 331 | } |
| 332 | if (linkFiles) { |
| 333 | for (Attribute fileAttr : linkFiles) { |
| 334 | if (auto fileStrAttr = llvm::dyn_cast<StringAttr>(Val&: fileAttr)) { |
| 335 | StringRef filePath = fileStrAttr.getValue(); |
| 336 | if (filePath.empty()) { |
| 337 | return emitError() << "File paths in linkFiles cannot be empty." ; |
| 338 | } |
| 339 | if (!llvm::sys::fs::exists(Path: filePath)) { |
| 340 | return emitError() << "File '" << filePath << "' does not exist." ; |
| 341 | } |
| 342 | } |
| 343 | } |
| 344 | } |
| 345 | return success(); |
| 346 | } |
| 347 | |
| 348 | void XeVMDialect::initialize() { |
| 349 | addOperations< |
| 350 | #define GET_OP_LIST |
| 351 | #include "mlir/Dialect/LLVMIR/XeVMOps.cpp.inc" |
| 352 | >(); |
| 353 | |
| 354 | addAttributes< |
| 355 | #define GET_ATTRDEF_LIST |
| 356 | #include "mlir/Dialect/LLVMIR/XeVMOpsAttributes.cpp.inc" |
| 357 | >(); |
| 358 | declarePromisedInterface<mlir::gpu::TargetAttrInterface, |
| 359 | mlir::xevm::XeVMTargetAttr>(); |
| 360 | } |
| 361 | |
| 362 | #define GET_OP_CLASSES |
| 363 | #include "mlir/Dialect/LLVMIR/XeVMOps.cpp.inc" |
| 364 | |
| 365 | #define GET_ATTRDEF_CLASSES |
| 366 | #include "mlir/Dialect/LLVMIR/XeVMOpsAttributes.cpp.inc" |
| 367 | |