Skip to content

Commit

Permalink
[GPU] rope optimization (#27907)
Browse files Browse the repository at this point in the history 
### Details:
 - Optimize rope opencl kernel to improve its performance
- Test result shows it can improve RoPE performance about 50% in
average.

<html xmlns:o="urn:schemas-microsoft-com:office:office"
xmlns:dt="uuid:C2F41010-65B3-11d1-A29F-00AA00C14882"
xmlns="http://www.w3.org/TR/REC-html40">

<head>

<meta name=ProgId content=OneNote.File>
<meta name=Generator content="Microsoft OneNote 15">
</head>

<body lang=en-US style='font-family:Calibri;font-size:11.0pt'>
<!--StartFragment-->

<div style='direction:ltr'>


batch=128, seq_length = 7 | base latency(ns) | optimized latency(ns) |
latency decreased |   |  
-- | -- | -- | -- | -- | --
rope_ref_5266667119713786613_0_0__sa, | 921352 | 872395 | 5.31% |
RoPETestQwen7b | f32
rope_ref_2672092794364911740_0_0__sa, | 1724374 | 514790 | 70.15% |
RoPETestChatGLM | f32
rope_ref_8061762790816124098_0_0__sa, | 633019 | 127186 | 79.91% |
RoPETestQwen7b | f16
rope_ref_4392014836945391706_0_0__sa, | 629791 | 518749 | 17.63% |
RoPETestLlama2 | f32
rope_ref_13829176589243505378_0_0__sa, | 870312 | 259583 | 70.17% |
RoPETestChatGLM | f32
rope_ref_6813544162411765619_0_0__sa, | 749895 | 421875 | 43.74% |
RoPETestChatGLM | f16
rope_ref_15054358246334082928_0_0__sa, | 637708 | 45208 | 92.91% |
RoPETestFlux | f32
rope_ref_3898891400599565440_0_0__sa, | 378333 | 335937 | 11.21% |
RoPETestRotateHalfWithoutTranspose | f32
rope_ref_18119704851383556529_0_0__sa, | 371250 | 208645 | 43.80% |
RoPETestChatGLM | f16
rope_ref_17460680473512025171_0_0__sa, | 299166 | 98958 | 66.92% |
RoPETestFlux | f16



</div>

<!--EndFragment-->
</body>

</html>


![image](https://github.com/user-attachments/assets/4328b1a7-18ec-485f-abd0-b0fe16785854)


### Tickets:
 - *CVS-157438*
  • Loading branch information
@riverlijunjie
riverlijunjie authored Dec 16, 2024
1 parent 357eb54 commit 9651768
Show file tree
Hide file tree
Showing 13 changed files with 760 additions and 379 deletions.
24 changes: 19 additions & 5 deletions ...ugins/intel_cpu/tests/functional/shared_tests_instances/subgraph_tests/rotary_pos_emb.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,51 +9,65 @@ namespace test {

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestLlama2StridedSlice,
RoPETestLlama2StridedSlice,
::testing::Values(ov::test::utils::DEVICE_CPU),
::testing::Combine(
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestLlama2StridedSlice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestChatGLMStridedSlice,
RoPETestChatGLMStridedSlice,
::testing::Values(ov::test::utils::DEVICE_CPU),
::testing::Combine(
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestChatGLMStridedSlice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestQwen7bStridedSlice,
RoPETestQwen7bStridedSlice,
::testing::Combine(::testing::Values(true, false),
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestQwen7bStridedSlice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestGPTJStridedSlice,
RoPETestGPTJStridedSlice,
::testing::Combine(::testing::Values(true, false),
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestGPTJStridedSlice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestLlama2Slice,
RoPETestLlama2Slice,
::testing::Values(ov::test::utils::DEVICE_CPU),
::testing::Combine(
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestLlama2Slice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestChatGLMSlice,
RoPETestChatGLMSlice,
::testing::Values(ov::test::utils::DEVICE_CPU),
::testing::Combine(
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestChatGLMSlice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestQwen7bSlice,
RoPETestQwen7bSlice,
::testing::Combine(::testing::Values(true, false),
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestQwen7bSlice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestGPTJSlice,
RoPETestGPTJSlice,
::testing::Combine(::testing::Values(true, false),
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestGPTJSlice::getTestCaseName);

INSTANTIATE_TEST_SUITE_P(smoke_RoPETestChatGLM,
RoPETestChatGLM2DRoPEStridedSlice,
::testing::Values(ov::test::utils::DEVICE_CPU),
::testing::Combine(
::testing::Values(ov::element::f32),
::testing::Values(ov::test::utils::DEVICE_CPU)),
RoPETestChatGLM2DRoPEStridedSlice::getTestCaseName);

} // namespace test
Expand Down
2 changes: 1 addition & 1 deletion src/plugins/intel_gpu/src/graph/impls/ocl/rope.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@

#include "rope_inst.h"
#include "rope/rope_kernel_selector.h"
#include "rope/rope_kernel_ref.h"
#include "rope/rope_kernel_opt.h"

namespace cldnn {
namespace ocl {
Expand Down
Loading

0 comments on commit 9651768

Please sign in to comment.