diff options
| author | Christian Sigg <csigg@google.com> | 2019-09-26 00:17:13 -0700 |
|---|---|---|
| committer | A. Unique TensorFlower <gardener@tensorflow.org> | 2019-09-26 00:17:50 -0700 |
| commit | 116dac00baa6870aec2a2b469b2d6f95c2fbb316 (patch) | |
| tree | 1ff87872c7a0db12f4fed9ff715327b05921792d /mlir/test/Dialect/GPU | |
| parent | 94298cea933991b29dcb7f340725bc25e78cebcf (diff) | |
| download | bcm5719-llvm-116dac00baa6870aec2a2b469b2d6f95c2fbb316.tar.gz bcm5719-llvm-116dac00baa6870aec2a2b469b2d6f95c2fbb316.zip | |
Add AllReduceOp to GPU dialect with lowering to NVVM.
The reduction operation is currently fixed to "add", and the scope is fixed to "workgroup".
The implementation is currently limited to sizes that are multiple 32 (warp size) and no larger than 1024.
PiperOrigin-RevId: 271290265
Diffstat (limited to 'mlir/test/Dialect/GPU')
| -rw-r--r-- | mlir/test/Dialect/GPU/ops.mlir | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/mlir/test/Dialect/GPU/ops.mlir b/mlir/test/Dialect/GPU/ops.mlir index b78da9543fc..7c8f682dcc8 100644 --- a/mlir/test/Dialect/GPU/ops.mlir +++ b/mlir/test/Dialect/GPU/ops.mlir @@ -76,6 +76,9 @@ func @kernel_1(%arg0 : f32, %arg1 : memref<?xf32, 1>) %gDimY = "gpu.grid_dim"() {dimension = "y"} : () -> (index) %gDimZ = "gpu.grid_dim"() {dimension = "z"} : () -> (index) + %one = constant 1.0 : f32 + %sum = "gpu.all_reduce"(%one) : (f32) -> (f32) + "some_op"(%bIdX, %tIdX) : (index, index) -> () %42 = load %arg1[%bIdX] : memref<?xf32, 1> return |
