-
Notifications
You must be signed in to change notification settings - Fork 160
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[transform] e2e UT for packed gemm #892
Merged
Merged
Changes from all commits
Commits
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
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
10 changes: 10 additions & 0 deletions
10
tao_compiler/mlir/disc/tests/disc-transform/data/packed_matmul_nn_p_512x1024_f32.mlir
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 +1,10 @@ | ||
module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 0 : i32}} { | ||
func.func @main(%arg0: tensor<?x?xf32>) -> (tensor<?x?xf32>) attributes {tf.entry_function = {inputs = "{{INPUTS}}", outputs = "{{OUTPUTS}}", input_placements="{{INPUT_PLACEMENTS}}", output_placements="{{OUTPUT_PLACEMENTS}}"}} { | ||
%graph = tf_executor.graph { | ||
%0:2 = tf_executor.island wraps "tf.Const"() {value = dense<-0.8> : tensor<512x1024xf32>} : () -> tensor<512x1024xf32> | ||
%1:2 = tf_executor.island wraps "tf.MatMul"(%arg0, %0) {transpose_a = false, transpose_b = false} : (tensor<?x?xf32>, tensor<512x1024xf32>) -> (tensor<?x?xf32>) | ||
tf_executor.fetch %1 : tensor<?x?xf32> | ||
} | ||
return %graph : tensor<?x?xf32> | ||
} | ||
} |
73 changes: 73 additions & 0 deletions
73
tao_compiler/mlir/disc/tests/disc-transform/data/packed_matmul_nn_p_f32_large_schedule.mlir
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 +1,73 @@ | ||
transform.structured.canonicalized_sequence failures(propagate) { | ||
^bb1(%arg1: !pdl.operation): | ||
%fill = transform.structured.match ops{["linalg.fill"]} in %arg1 | ||
%matmul = transform.structured.match ops{["linalg.matmul"]} in %arg1 | ||
|
||
%0:2 = transform.structured.tile_to_foreach_thread_op %matmul num_threads [1, 1] | ||
transform.structured.fuse_into_containing_op %fill into %0#0 | ||
|
||
// first level tile and fuse matmul and fill op. | ||
%1:3 = transform.structured.fuse %0#1 {tile_sizes = [288, 48, 0], tile_interchange = [0, 1, 2]} | ||
// second level tile and fuse matmul and fill op. | ||
%2:3 = transform.structured.fuse %1#0 {tile_sizes = [6, 16, 0], tile_interchange = [0, 1, 2]} | ||
|
||
// gemm reduction axis tiling | ||
%3:2 = transform.structured.tile %2#0 [0, 0, 1] {interchange=[0, 1, 2]} | ||
|
||
// clean up | ||
%func0 = transform.structured.match ops{["func.func"]} in %arg1 | ||
transform.disc.apply_patterns %func0 {canonicalization} | ||
// fold two extract_slice ops generated by two-level tiling. It's needed to enable following | ||
// pad and hosit schedule. | ||
%weight_inner_slice = get_producer_of_operand %3#0[1] : (!pdl.operation) -> !pdl.operation | ||
transform.disc.fold_producer_extract_slice %weight_inner_slice {max_repeat_num = 2} | ||
|
||
// pad to match the requirement of hardware vector/tensor instruction. | ||
%4 = transform.structured.pad %3#0 {padding_values=[0.0 : f32, 0.0 : f32, 0.0 : f32], padding_dimensions=[0, 1, 2], pack_paddings=[1, 1, 0], hoist_paddings=[0, 0, 0], transpose_paddings=[[1, 0], [0, 1], [0, 1]]} | ||
|
||
%pad_for_input = get_producer_of_operand %4[0] : (!pdl.operation) -> !pdl.operation | ||
%pad_for_weight = get_producer_of_operand %4[1] : (!pdl.operation) -> !pdl.operation | ||
%foreach_op = transform.structured.match ops{["scf.foreach_thread"]} in %arg1 | ||
%loop_for_outter_most_n = transform.loop.get_parent_for %4 { num_loops = 4} : (!pdl.operation) -> !pdl.operation | ||
transform.disc.cache_read {padded} %pad_for_input at %loop_for_outter_most_n with tile_levels = [1, 1] tile_sizes = [6, 1] permutation = [0, 2, 3, 1] | ||
transform.disc.cache_read {padded} %pad_for_weight at %foreach_op with tile_levels = [1, 1] tile_sizes = [1, 16] permutation = [2, 0, 1, 3] | ||
|
||
%func1 = transform.structured.match ops{["func.func"]} in %arg1 | ||
transform.disc.apply_patterns %func1 {canonicalization} | ||
|
||
%pack_op = transform.structured.match ops{["disc_linalg_ext.multi_level_pack"]} in %arg1 | ||
transform.disc.lower_multi_level_pack_to_loop %pack_op | ||
|
||
%func2 = transform.structured.match ops{["func.func"]} in %arg1 | ||
transform.disc.apply_patterns %func2 {canonicalization} | ||
|
||
%func3 = transform.structured.match ops{["func.func"]} in %arg1 | ||
transform.structured.vectorize %func3 {vectorize_padding} | ||
|
||
%func4 = transform.structured.match ops{["func.func"]} in %arg1 | ||
transform.disc.apply_patterns %func4 {canonicalization} | ||
|
||
transform.disc.bufferize %arg1 | ||
|
||
transform.lower_vectors { | ||
contraction_lowering = "outerproduct", | ||
multireduction_lowering = "innerparallel", | ||
split_transfers = "linalg-copy", | ||
// stages = [0, 1, 2, 3, 4, 5, 6, 7], | ||
stages = [0, 1, 2, 3], | ||
transpose_avx2_lowering = false, | ||
transpose_lowering = "eltwise", | ||
unroll_vector_transfers = true | ||
} | ||
|
||
transform.lower_vectors { | ||
contraction_lowering = "outerproduct", | ||
multireduction_lowering = "innerparallel", | ||
split_transfers = "linalg-copy", | ||
// stages = [0, 1, 2, 3, 4, 5, 6, 7], | ||
stages = [5, 6, 7], | ||
transpose_avx2_lowering = false, | ||
transpose_lowering = "eltwise", | ||
unroll_vector_transfers = true | ||
} | ||
} |
50 changes: 50 additions & 0 deletions
50
tao_compiler/mlir/disc/tests/disc-transform/packed_matmul.cc
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 +1,50 @@ | ||
/* Copyright 2022 The BladeDISC 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. | ||
==============================================================================*/ | ||
|
||
#include "tensorflow/compiler/mlir/disc/tests/mlir_feature_test.h" | ||
#include "tensorflow/compiler/mlir/disc/tests/mlir_test.h" | ||
#include "tensorflow/core/platform/test.h" | ||
|
||
namespace mlir_test { | ||
|
||
const std::string c_ft_path = | ||
"tensorflow/compiler/mlir/disc/tests/disc-transform/data/"; | ||
|
||
static bool init_threads = []() { | ||
setenv("OMP_NUM_THREADS", "1", 1); | ||
return true; | ||
}(); | ||
|
||
TEST(PackedMatmul, F32_304x1024x512) { | ||
EnvSetting setting = { | ||
{"DISC_TRANSFORM_SCHEDULE_FILE", | ||
{c_ft_path + "packed_matmul_nn_p_f32_large_schedule.mlir", false}}, | ||
{"DISC_ENABLE_TRANSFORM_SCHEDULE", {"1", false}}, | ||
{"DISC_ENABLE_SHAPE_CONSTRAINT_IR", {"1", false}}, | ||
{"DISC_MEM_INTENSIVE_OPT_EXPERIMENTAL", {"0", false}}}; | ||
EnvSettingContext ctx(setting); | ||
EXPECT_TRUE(feature_test_main( | ||
/*mlir_file_path*/ c_ft_path + "packed_matmul_nn_p_512x1024_f32.mlir", | ||
/*backend_types*/ {BackendType::kAArch64}, | ||
/*num_inputs*/ 1, | ||
/*num_outputs*/ 1, | ||
/*input_descriptors*/ {"304x512xf32_X"}, | ||
/*output_descriptors*/ {"f32_X"}, | ||
/*input_vals*/ {}, | ||
/*expected_output_vals*/ {}, | ||
/*profiling*/ true)); | ||
} | ||
|
||
} // namespace mlir_test |
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
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
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
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.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It looks pretty readable except for things like
%2#0
, one has to memorize meaning of each output.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'll try to give more meaningful names to these return results in the later schedule.