| /llvm-project-15.0.7/mlir/test/Dialect/NVGPU/ |
| H A D | invalid.mlir | 6 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf16, … 14 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf16, … 21 // expected-error @+1 {{expected vector register shape[0] and numTiles to match}} 22 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf16, … 30 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf32, … 38 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = true, numTiles = 4 : i32} : memref<128x128xf32, 3… 46 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf32, …
|
| H A D | roundtrip.mlir | 6 // CHECK-SAME: {numTiles = 4 : i32, transpose = false} : memref<?x?xf16, 3> -> vector<4x2xf16> 7 %l = nvgpu.ldmatrix %arg0[%x, %y] {numTiles = 4 : i32, transpose = false} :
|
| H A D | optimize-shared-memory.mlir | 31 %mat = nvgpu.ldmatrix %shm[%fragRow, %fragCol] {numTiles = 4 : i32, transpose = false} 51 %matB = nvgpu.ldmatrix %shmB[%fragRow, %fragCol] {numTiles = 4 : i32, transpose = false} 88 %mat = nvgpu.ldmatrix %shm[%fragRow, %fragCol] {numTiles = 4 : i32, transpose = false} 144 %matB = nvgpu.ldmatrix %shmB[%fragRow, %fragCol] {numTiles = 4 : i32, transpose = false} 210 %mat = nvgpu.ldmatrix %shm[%fragRow, %fragCol] {numTiles = 1 : i32, transpose = false} 236 %mat = nvgpu.ldmatrix %shmView[%fragRow, %fragCol] {numTiles = 1 : i32, transpose = false}
|
| /llvm-project-15.0.7/mlir/test/Conversion/VectorToGPU/ |
| H A D | vector-to-mma-ops-mma-sync.mlir | 47 …// CHECK: nvgpu.ldmatrix %arg0[[[row]], [[col]]] {numTiles = 4 : i32, transpose = false} : memref<… 191 // CHECK: nvgpu.ldmatrix %arg0[[[row]], [[col]]] {numTiles = 4 : i32, transpose = false} 195 // CHECK: nvgpu.ldmatrix %arg1[[[row]], [[col]]] {numTiles = 2 : i32, transpose = true} 269 // CHECK: nvgpu.ldmatrix %arg0[[[row]], [[col]]] {numTiles = 4 : i32 274 // CHECK: nvgpu.ldmatrix %arg1[[[row]], [[col]]] {numTiles = 2 : i32 279 // CHECK: nvgpu.ldmatrix %arg2[[[row]], [[col]]] {numTiles = 2 : i32 322 …// CHECK: [[a_frag:%.+]] = nvgpu.ldmatrix %arg0[[[row]], [[col]]] {numTiles = 2 : i32, transpose =… 380 …// CHECK: [[a_frag:%.+]] = nvgpu.ldmatrix %arg0[[[row]], [[col]]] {numTiles = 4 : i32, transpose =… 439 …// CHECK: nvgpu.ldmatrix %arg0[[[row]], [[col]]] {numTiles = 4 : i32, transpose = false} : memref<… 444 …// CHECK: nvgpu.ldmatrix %arg1[[[row]], [[col]]] {numTiles = 2 : i32, transpose = false} : memref<… [all …]
|
| /llvm-project-15.0.7/mlir/lib/Dialect/NVGPU/IR/ |
| H A D | NVGPUDialect.cpp | 213 int64_t numTiles = getNumTiles(); in verify() local 237 if (!(resShape[0] == numTiles)) in verify()
|
| /llvm-project-15.0.7/mlir/lib/Conversion/VectorToGPU/ |
| H A D | NvGpuSupport.cpp | 224 params.numTiles = (shape[0] / kNumRowsPerTile) * in getLdMatrixParams() 227 params.numTiles = (shape[1] / kNumRowsPerTile) * in getLdMatrixParams() 231 if (params.numTiles == 0) in getLdMatrixParams()
|
| H A D | NvGpuSupport.h | 73 int64_t numTiles; member
|
| H A D | VectorToGPU.cpp | 507 !op.getPermutationMap().isMinorIdentity(), params->numTiles); in creatLdMatrixCompatibleLoads()
|
| /llvm-project-15.0.7/mlir/lib/Dialect/Linalg/Transforms/ |
| H A D | Tiling.cpp | 358 OpFoldResult numTiles = std::get<0>(it); in tileToForeachThreadOpUsingTileSizes() local 359 if (!isConstantIntValue(numTiles, 0)) in tileToForeachThreadOpUsingTileSizes() 360 numTiles = makeComposedFoldedAffineApply( in tileToForeachThreadOpUsingTileSizes() 362 numThreads.push_back(numTiles); in tileToForeachThreadOpUsingTileSizes()
|
| /llvm-project-15.0.7/mlir/include/mlir/Dialect/NVGPU/IR/ |
| H A D | NVGPU.td | 72 %0 = nvgpu.ldmatrix %sm[%c0, %c0] {numTiles = 4 : i32, transpose = false} : 79 I32Attr:$numTiles);
|
| /llvm-project-15.0.7/mlir/test/Conversion/NVGPUToNVVM/ |
| H A D | nvgpu-to-nvvm.mlir | 177 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 4 : i32} : memref<128x128xf16, … 199 …%a = nvgpu.ldmatrix %arg0[%c0, %c0] {transpose = false, numTiles = 1 : i32} : memref<128x128xf16, …
|