-
Notifications
You must be signed in to change notification settings - Fork 685
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[CodeGen]Support matmul codegen and runtime in mlir jit (#10283)
本pr支持在mlir jit中运行matmul op，该pr补全了在gpu dialect之后的codegen中健壮性的一部分问题
- Loading branch information
Showing
7 changed files
with
272 additions
and
58 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,77 @@ | ||
| """ | ||
| Copyright 2020 The OneFlow Authors. All rights reserved. | ||
| Licensed under the Apache License, Version 2.0 (the "License"); | ||
| you may not use this file except in compliance with the License. | ||
| You may obtain a copy of the License at | ||
| http://www.apache.org/licenses/LICENSE-2.0 | ||
| Unless required by applicable law or agreed to in writing, software | ||
| distributed under the License is distributed on an "AS IS" BASIS, | ||
| WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| See the License for the specific language governing permissions and | ||
| limitations under the License. | ||
| """ | ||
| # RUN: python3 -m oneflow.test_utils.throttle --with-cuda=%with_cuda python3 %s | FileCheck %s | ||
| # CHECK: jit | ||
|
|
||
| import unittest | ||
| import numpy as np | ||
|
|
||
| import os | ||
|
|
||
|
|
||
| import oneflow as flow | ||
| import oneflow.unittest | ||
|
|
||
|
|
||
| class MatMulModule(flow.nn.Module): | ||
| def __init__(self): | ||
| super().__init__() | ||
| self.w = flow.nn.Parameter(flow.Tensor(5, 10)) | ||
| self.b = flow.nn.Parameter(flow.Tensor(10)) | ||
|
|
||
| def forward(self, x): | ||
| return flow.matmul(x, self.w) + self.b | ||
|
|
||
|
|
||
| def do_matmul_graph(test_case, with_cuda=False): | ||
| x = flow.randn(2, 5) | ||
| module_to_run = MatMulModule() | ||
| if with_cuda: | ||
| x = x.cuda() | ||
| module_to_run = module_to_run.to("cuda") | ||
| y_eager = module_to_run(x) | ||
|
|
||
| class GraphToRun(flow.nn.Graph): | ||
| def __init__(self): | ||
| super().__init__() | ||
| self.fw = module_to_run | ||
|
|
||
| def build(self, x): | ||
| return self.fw(x) | ||
|
|
||
| graph_to_run = GraphToRun() | ||
| y_lazy = graph_to_run(x) | ||
| test_case.assertTrue(np.array_equal(y_eager.numpy(), y_lazy.numpy())) | ||
|
|
||
|
|
||
| @flow.unittest.skip_unless_1n1d() | ||
| class TestFuseCastScale(oneflow.unittest.MLIRTestCase): | ||
| def setUp(self): | ||
| os.environ["ONEFLOW_MLIR_ENABLE_ROUND_TRIP"] = "1" | ||
| os.environ["ONEFLOW_MLIR_ENABLE_CODEGEN_FUSERS"] = "1" | ||
| os.environ["ONEFLOW_MLIR_FUSE_FORWARD_OPS"] = "1" | ||
|
|
||
| def test_relu_graph(test_case): | ||
| import oneflow.sysconfig | ||
|
|
||
| if oneflow.sysconfig.with_cuda(): | ||
| do_matmul_graph(test_case, True) | ||
|
|
||
| do_matmul_graph(test_case) | ||
|
|
||
|
|
||
| if __name__ == "__main__": | ||
| unittest.main() |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,2 @@ | ||
| if not config.WITH_MLIR_CUDA_CODEGEN: | ||
| config.unsupported = True |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,36 @@ | ||
| // RUN: oneflow-opt %s --insert-ofmempool --convert-linalg-to-loops --convert-scf-to-cf --canonicalize --cse --memref-expand --gpu-kernel-outlining \ | ||
| // RUN: | oneflow-opt --pass-pipeline='builtin.module(gpu.module(expand-strided-metadata,lower-affine,strip-debuginfo,convert-gpu-to-nvvm,nvvm-to-cubin))' | ||
|
|
||
| module { | ||
| func.func @JITOpGenerated0(%arg0: memref<5x10xf32, strided<[?, ?], offset: ?>>, %arg1: memref<2x5xf32, strided<[?, ?], offset: ?>>, %arg2: memref<2x10xf32>) attributes {llvm.emit_c_interface} { | ||
| %alloc = memref.alloc() : memref<512xi8> | ||
| %c0 = arith.constant 0 : index | ||
| %view = memref.view %alloc[%c0][] : memref<512xi8> to memref<1x2x10xf32> | ||
| %c10 = arith.constant 10 : index | ||
| %c2 = arith.constant 2 : index | ||
| %c1 = arith.constant 1 : index | ||
| %c0_0 = arith.constant 0 : index | ||
| %c5 = arith.constant 5 : index | ||
| %cst = arith.constant 0.000000e+00 : f32 | ||
| %expand_shape = memref.expand_shape %arg0 [[0, 1], [2]] : memref<5x10xf32, strided<[?, ?], offset: ?>> into memref<1x5x10xf32, strided<[?, ?, ?], offset: ?>> | ||
| %expand_shape_1 = memref.expand_shape %arg1 [[0, 1], [2]] : memref<2x5xf32, strided<[?, ?], offset: ?>> into memref<1x2x5xf32, strided<[?, ?, ?], offset: ?>> | ||
| gpu.launch blocks(%arg3, %arg4, %arg5) in (%arg9 = %c1, %arg10 = %c2, %arg11 = %c10) threads(%arg6, %arg7, %arg8) in (%arg12 = %c1, %arg13 = %c1, %arg14 = %c1) { | ||
| memref.store %cst, %view[%c0_0, %arg4, %arg5] : memref<1x2x10xf32> | ||
| gpu.terminator | ||
| } {SCFToGPU_visited} | ||
| gpu.launch blocks(%arg3, %arg4, %arg5) in (%arg9 = %c1, %arg10 = %c2, %arg11 = %c10) threads(%arg6, %arg7, %arg8) in (%arg12 = %c1, %arg13 = %c1, %arg14 = %c1) { | ||
| scf.for %arg15 = %c0_0 to %c5 step %c1 { | ||
| %0 = memref.load %expand_shape_1[%c0_0, %arg4, %arg15] : memref<1x2x5xf32, strided<[?, ?, ?], offset: ?>> | ||
| %1 = memref.load %expand_shape[%c0_0, %arg15, %arg5] : memref<1x5x10xf32, strided<[?, ?, ?], offset: ?>> | ||
| %2 = memref.load %view[%c0_0, %arg4, %arg5] : memref<1x2x10xf32> | ||
| %3 = arith.mulf %0, %1 : f32 | ||
| %4 = arith.addf %2, %3 : f32 | ||
| memref.store %4, %view[%c0_0, %arg4, %arg5] : memref<1x2x10xf32> | ||
| } | ||
| gpu.terminator | ||
| } {SCFToGPU_visited} | ||
| %collapse_shape = memref.collapse_shape %view [[0, 1], [2]] : memref<1x2x10xf32> into memref<2x10xf32> | ||
| memref.copy %collapse_shape, %arg2 : memref<2x10xf32> to memref<2x10xf32> | ||
| return | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.