-
Notifications
You must be signed in to change notification settings - Fork 2.6k
[GPU] Fix accuracy issues for mvn and sdpa_micro #30698
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
Open
ahnyoung-paul
wants to merge
4
commits into
openvinotoolkit:master
Choose a base branch
from
ahnyoung-paul:fix_ltx_video_acc
base: master
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
+285
−2
Open
Changes from all commits
Commits
Show all changes
4 commits
Select commit
Hold shift + click to select a range
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 hidden or 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 hidden or 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 hidden or 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
185 changes: 185 additions & 0 deletions
185
src/plugins/intel_gpu/tests/unit/test_cases/sdpa_gpu_test.cpp
This file contains hidden or 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 +1,185 @@ | ||
| // Copyright (C) 2018-2025 Intel Corporation | ||
| // SPDX-License-Identifier: Apache-2.0 | ||
| // | ||
|
|
||
| #include "test_utils.h" | ||
| #include "random_generator.hpp" | ||
|
|
||
| #include <intel_gpu/primitives/input_layout.hpp> | ||
| #include <intel_gpu/primitives/reorder.hpp> | ||
| #include <intel_gpu/primitives/eltwise.hpp> | ||
| #include <intel_gpu/runtime/debug_configuration.hpp> | ||
|
|
||
| #include "openvino/util/file_util.hpp" | ||
| #include <iostream> | ||
| #include <vector> | ||
| #include <cmath> | ||
| #include <numeric> | ||
| #include <iostream> | ||
|
|
||
| #include <intel_gpu/primitives/input_layout.hpp> | ||
| #include <intel_gpu/primitives/scaled_dot_product_attention.hpp> | ||
|
|
||
| #include <cstddef> | ||
| #include <vector> | ||
|
|
||
| using namespace cldnn; | ||
| using namespace ::tests; | ||
|
|
||
| namespace { | ||
| // #ifdef ENABLE_ONEDNN_FOR_GPU | ||
| // Disable onednn test because onednn does not support format_tag::cbda, format_tag::badc. | ||
|
|
||
|
|
||
| struct spda_gpu_test { | ||
|
|
||
| // const std::string opt_data_path = "/home/ahnyoung/cldnn/cvs_164660/dumps/outs/gpu.fp16.sdpa.ref.raw/"; | ||
| const std::string opt_data_path = "/home/ahnyoung/cldnn/cvs_164660/dumps/outs/gpu.fp16.sdpa.micro.raw/"; | ||
|
|
||
| void load_input(cldnn::memory::ptr mem, size_t idx) { | ||
| std::vector<std::string> bin_names = { | ||
| "program1_network1_0_sdpa___module.transformer_blocks.0.attn2_aten__scaled_dot_product_attention_ScaledDotProductAttention_dst0__f16__2_32_990_64__bfyx.bin", | ||
| "program1_network1_0_sdpa___module.transformer_blocks.0.attn2_aten__scaled_dot_product_attention_ScaledDotProductAttention_src0__f16__2_990_32_64__bfyx.bin", | ||
| "program1_network1_0_sdpa___module.transformer_blocks.0.attn2_aten__scaled_dot_product_attention_ScaledDotProductAttention_src1__f16__2_128_32_64__bfyx.bin", | ||
| "program1_network1_0_sdpa___module.transformer_blocks.0.attn2_aten__scaled_dot_product_attention_ScaledDotProductAttention_src2__f16__2_128_32_64__bfyx.bin", | ||
| "program1_network1_0_sdpa___module.transformer_blocks.0.attn2_aten__scaled_dot_product_attention_ScaledDotProductAttention_src3__f16__2_32_1_128__bfyx.bin" | ||
| }; | ||
| std::string input_file_name = opt_data_path + bin_names[idx]; | ||
| load_data_from_bin(mem, input_file_name); | ||
| } | ||
|
|
||
| void load_data_from_bin(cldnn::memory::ptr mem, const std::string filepath) { | ||
| GPU_DEBUG_COUT << "Load data from " << filepath << std::endl; | ||
| std::vector<uint8_t> bin = ov::util::load_binary(filepath); | ||
| mem->copy_from(get_test_stream(), static_cast<void *>(&bin[0]), true); | ||
| } | ||
|
|
||
| cldnn::memory::ptr run_network(bool is_caching_test, bool use_micro_sdpa = false) { | ||
| auto& engine = get_test_engine(); | ||
| cldnn::layout input0_dyn_layout({-1, -1, 32, 64}, data_types::f16, format::bfyx); | ||
| cldnn::layout input1_dyn_layout({-1, -1, 32, 64}, data_types::f16, format::bfyx); | ||
| cldnn::layout input2_dyn_layout({-1, -1, 32, 64}, data_types::f16, format::bfyx); | ||
| cldnn::layout input3_dyn_layout({-1, 32, -1, -1}, data_types::f16, format::bfyx); | ||
|
|
||
| cldnn::layout input0_static_layout({2, 990, 32, 64}, data_types::f16, format::bfyx); | ||
| cldnn::layout input1_static_layout({2, 128, 32, 64}, data_types::f16, format::bfyx); | ||
| cldnn::layout input2_static_layout({2, 128, 32, 64}, data_types::f16, format::bfyx); | ||
| cldnn::layout input3_static_layout({2, 32, 1, 128}, data_types::f16, format::bfyx); | ||
|
|
||
| auto input0 = engine.allocate_memory(input0_static_layout); | ||
| auto input1 = engine.allocate_memory(input1_static_layout); | ||
| auto input2 = engine.allocate_memory(input2_static_layout); | ||
| auto input3 = engine.allocate_memory(input3_static_layout); | ||
|
|
||
| load_input(input0, 0); | ||
| load_input(input1, 1); | ||
| load_input(input2, 2); | ||
| load_input(input3, 3); | ||
|
|
||
| GPU_DEBUG_COUT << "Topology: SDPA kernel test " << std::endl; | ||
| GPU_DEBUG_COUT << "* use micro_sdpa : " << (use_micro_sdpa ? "Yes" : "No") << std::endl; | ||
| GPU_DEBUG_COUT << "* input0 : " << input0_static_layout.to_short_string() << ", " << input0_static_layout.count() << std::endl; | ||
| GPU_DEBUG_COUT << "* input1 : " << input1_static_layout.to_short_string() << ", " << input1_static_layout.count() << std::endl; | ||
| GPU_DEBUG_COUT << "* input2 : " << input2_static_layout.to_short_string() << ", " << input2_static_layout.count() << std::endl; | ||
| GPU_DEBUG_COUT << "* input3 : " << input3_static_layout.to_short_string() << ", " << input3_static_layout.count() << std::endl; | ||
|
|
||
| topology topo; | ||
| topo.add(input_layout("input0", input0_dyn_layout)); | ||
| topo.add(input_layout("input1", input1_dyn_layout)); | ||
| topo.add(input_layout("input2", input2_dyn_layout)); | ||
| topo.add(input_layout("input3", input3_dyn_layout)); | ||
| topo.add(scaled_dot_product_attention("sdpa", {input_info("input0"), input_info("input1"), input_info("input2"), input_info("input3")}, | ||
| false, -1, {0,2,1,3}, {0,2,1,3}, {0,2,1,3}, {0,1,2,3}, {}, false)); | ||
| topo.add(reorder("result",input_info("sdpa"), format::bfyx, data_types::f16)); | ||
|
|
||
| ExecutionConfig config = get_test_default_config(engine); | ||
| config.set_property(ov::intel_gpu::allow_new_shape_infer(true)); | ||
|
|
||
| if (use_micro_sdpa) { | ||
| config.set_property(ov::intel_gpu::force_implementations(ov::intel_gpu::ImplForcingMap{ {"sdpa", {format::type::bfyx, "sdpa_micro"}} })); | ||
| config.set_property(ov::intel_gpu::dump_iterations(std::set<int64_t>{0, 1})); | ||
| config.set_property(ov::intel_gpu::dump_tensors("all")); | ||
| config.set_property(ov::intel_gpu::dump_tensors_path("/home/ahnyoung/cldnn/cvs_164660/dumps/outs/units/")); | ||
| } else { | ||
| config.set_property(ov::intel_gpu::force_implementations(ov::intel_gpu::ImplForcingMap{ {"sdpa", {format::type::bfyx, "sdpa_ref"}} })); | ||
| } | ||
|
|
||
| cldnn::network::ptr net = get_network(engine, topo, config, get_test_stream_ptr(), is_caching_test); | ||
|
|
||
| net->set_input_data("input0", input0); | ||
| net->set_input_data("input1", input1); | ||
| net->set_input_data("input2", input2); | ||
| net->set_input_data("input3", input3); | ||
|
|
||
| auto outputs = net->execute(); | ||
| auto output = outputs.at("result").get_memory(); | ||
| return output; | ||
| } | ||
|
|
||
| void execute(bool is_caching_test = false) { | ||
| GPU_DEBUG_COUT << "********************************************************************************" << std::endl; | ||
| GPU_DEBUG_COUT << "********************************************************************************" << std::endl; | ||
| auto mem_ref_ptr = run_network(is_caching_test, false); | ||
| GPU_DEBUG_COUT << "********************************************************************************" << std::endl; | ||
| GPU_DEBUG_COUT << "********************************************************************************" << std::endl; | ||
| auto mem_opt_ptr = run_network(is_caching_test, true); | ||
| GPU_DEBUG_COUT << "********************************************************************************" << std::endl; | ||
| GPU_DEBUG_COUT << "********************************************************************************" << std::endl; | ||
| cldnn::mem_lock<ov::float16, mem_lock_type::read> ref_data(mem_ref_ptr, get_test_stream()); | ||
| cldnn::mem_lock<ov::float16, mem_lock_type::read> opt_data(mem_opt_ptr, get_test_stream()); | ||
| // if (ret < 0.9f) { | ||
| { | ||
| std::vector<std::pair<size_t, ov::float16>> differences; | ||
| for (size_t idx = 0; idx < ref_data.size(); idx++) { | ||
| if (std::isnan(opt_data[idx])) { | ||
| GPU_DEBUG_COUT << "opt_data has nan " << opt_data[idx] << std::endl; | ||
| } | ||
| if (std::isnan(ref_data[idx])) { | ||
| GPU_DEBUG_COUT << "ref_data has nan " << ref_data[idx] << std::endl; | ||
| } | ||
| ASSERT_FALSE(std::isnan(opt_data[idx])); | ||
| float diff = std::abs(ref_data[idx] - opt_data[idx]); | ||
| differences.push_back({idx, diff}); | ||
| } | ||
| // auto ret = cosineSimilarity(ref_data, opt_data); | ||
| // GPU_DEBUG_COUT << "Cosine Similarity : " << ret << std::endl; | ||
| std::sort(differences.begin(), differences.end(), [](std::pair<size_t, ov::float16> a, std::pair<size_t, ov::float16> b){ | ||
| return a.second > b.second; | ||
| }); | ||
| GPU_DEBUG_COUT << "Compare data] ref_data : act_data" << std::endl; | ||
| for (size_t i = 0; i < 10 && i < differences.size(); i++) { | ||
| size_t idx = differences[i].first; | ||
| GPU_DEBUG_COUT << std::setw(8) << std::fixed << idx << "] " << std::setw(12) << ref_data[idx] << " : " | ||
| << std::setw(12) << opt_data[idx] | ||
| << " (Difference: " << differences[i].second << ")" << std::endl; | ||
| } | ||
| // ASSERT_GE(ret, 0.9f); | ||
| } | ||
| } | ||
|
|
||
| // float cosineSimilarity(cldnn::mem_lock<ov::float16, mem_lock_type::read>& vec1, cldnn::mem_lock<ov::float16, mem_lock_type::read>& memLockVec2) { | ||
| // if (vec1.size() != memLockVec2.size()) { | ||
| // std::cerr << "Vectors must be of the same size." << std::endl; | ||
| // return -1.0f; | ||
| // } | ||
|
|
||
| // float dotProduct = std::inner_product(vec1.begin(), vec1.end(), memLockVec2.begin(), 0.0f); | ||
|
|
||
| // float magnitude1 = std::sqrt(std::inner_product(vec1.begin(), vec1.end(), vec1.begin(), 0.0f)); | ||
| // float magnitude2 = std::sqrt(std::inner_product(memLockVec2.begin(), memLockVec2.end(), memLockVec2.begin(), 0.0f)); | ||
|
|
||
| // if (magnitude1 == 0.0f || magnitude2 == 0.0f) { | ||
| // std::cerr << "One of the vectors is zero vector." << std::endl; | ||
| // return -1.0f; | ||
| // } | ||
|
|
||
| // return dotProduct / (magnitude1 * magnitude2); | ||
| // } | ||
| }; | ||
|
|
||
| TEST(sdpa_gpu_test, basic) { | ||
| spda_gpu_test test; | ||
| test.execute(); | ||
| } | ||
| // #endif | ||
| } // namespace |
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.
why num_heads? masks are usually [q, seqlen]..