riscv optimize convolution packed#6731
Merged
Merged
Conversation
Member
|
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #6731 +/- ##
==========================================
- Coverage 95.95% 95.95% -0.01%
==========================================
Files 970 960 -10
Lines 403476 403285 -191
==========================================
- Hits 387159 386967 -192
- Misses 16317 16318 +1 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Contributor
There was a problem hiding this comment.
Pull request overview
This PR consolidates RISC-V (RVV/Zvfh) convolution and deconvolution implementations by replacing multiple pack-specific kernels (packn/pack1ton/packnto1/pack1) with unified “packed” kernel transform + execution paths, and removes the older per-pack headers.
Changes:
- Introduces unified packed kernel transform/execution helpers for convolution and deconvolution (fp32 + fp16s/fp16sa paths).
- Updates RISC-V convolution/deconvolution pipeline/forward code to call the new packed entrypoints.
- Removes the legacy pack-specific header implementations that are now superseded.
Reviewed changes
Copilot reviewed 22 out of 22 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| src/layer/riscv/deconvolution_riscv.cpp | Switches fp32 deconvolution to use deconvolution_packed.* transform + forward path. |
| src/layer/riscv/deconvolution_riscv_zfh.cpp | Switches fp16 deconvolution to use deconvolution_packed_fp16s.* transform + forward paths. |
| src/layer/riscv/deconvolution_packnto1.h | Removed legacy RVV packnto1 deconvolution kernel. |
| src/layer/riscv/deconvolution_packnto1_fp16s.h | Removed legacy RVV fp16s/fp16sa packnto1 deconvolution kernels. |
| src/layer/riscv/deconvolution_packn.h | Removed legacy RVV packn deconvolution kernel. |
| src/layer/riscv/deconvolution_packn_fp16s.h | Removed legacy RVV fp16s/fp16sa packn deconvolution kernels. |
| src/layer/riscv/deconvolution_pack1ton.h | Removed legacy RVV pack1ton deconvolution kernel. |
| src/layer/riscv/deconvolution_pack1ton_fp16s.h | Removed legacy RVV fp16s/fp16sa pack1ton deconvolution kernels. |
| src/layer/riscv/deconvolution_fp16s.h | Removed legacy fp16s deconvolution fallback implementation. |
| src/layer/riscv/deconvolution_packed.h | Adds unified fp32 packed deconvolution kernel transform + forward implementation. |
| src/layer/riscv/deconvolution_packed_fp16s.h | Adds unified fp16s/fp16sa packed deconvolution kernel transform + forward implementations. |
| src/layer/riscv/convolution_riscv.cpp | Switches fp32 convolution to use convolution_packed.* transform + forward path. |
| src/layer/riscv/convolution_riscv_zfh.cpp | Switches fp16 convolution to use convolution_packed_fp16s.* transform + forward paths. |
| src/layer/riscv/convolution_packnto1.h | Removed legacy RVV packnto1 convolution kernel. |
| src/layer/riscv/convolution_packnto1_fp16s.h | Removed legacy RVV fp16s/fp16sa packnto1 convolution kernels. |
| src/layer/riscv/convolution_packn.h | Removed legacy RVV packn convolution kernel. |
| src/layer/riscv/convolution_packn_fp16s.h | Removed legacy RVV fp16s/fp16sa packn convolution kernels. |
| src/layer/riscv/convolution_pack1ton.h | Removed legacy RVV pack1ton convolution kernel. |
| src/layer/riscv/convolution_pack1ton_fp16s.h | Removed legacy RVV fp16s/fp16sa pack1ton convolution kernels. |
| src/layer/riscv/convolution_fp16s.h | Removed legacy fp16s convolution fallback implementation. |
| src/layer/riscv/convolution_packed.h | Adds unified fp32 packed convolution kernel transform + forward implementation. |
| src/layer/riscv/convolution_packed_fp16s.h | Adds unified fp16s/fp16sa packed convolution kernel transform + forward implementations. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+50
to
+56
| #if __riscv_vector | ||
| if (num_output >= packn) | ||
| { | ||
| if (num_input >= packn) | ||
| weight_data_tm.create(packn * packn * maxk, num_input / packn + num_input % packn, num_output / packn + num_output % packn); | ||
| else | ||
| weight_data_tm.create(packn * maxk, num_input, num_output / packn + num_output % packn); |
Comment on lines
+54
to
+61
| weight_data_tm.create(packn * packn * maxk, num_input / packn + num_input % packn, num_output / packn + num_output % packn, (size_t)2u); | ||
| else | ||
| weight_data_tm.create(packn * maxk, num_input, num_output / packn + num_output % packn, (size_t)2u); | ||
| } | ||
| else | ||
| { | ||
| if (num_input >= packn) | ||
| weight_data_tm.create(packn * maxk, num_input / packn + num_input % packn, num_output, (size_t)2u); |
Comment on lines
+34
to
+44
| if (num_output >= packn) | ||
| { | ||
| if (num_input >= packn) | ||
| weight_data_tm.create(packn * packn * maxk, num_input / packn + num_input % packn, num_output / packn + num_output % packn); | ||
| else | ||
| weight_data_tm.create(packn * maxk, num_input, num_output / packn + num_output % packn); | ||
| } | ||
| else | ||
| { | ||
| if (num_input >= packn) | ||
| weight_data_tm.create(packn * maxk, num_input / packn + num_input % packn, num_output); |
Comment on lines
+37
to
+44
| weight_data_tm.create(packn * packn * maxk, num_input / packn + num_input % packn, num_output / packn + num_output % packn, (size_t)2u); | ||
| else | ||
| weight_data_tm.create(packn * maxk, num_input, num_output / packn + num_output % packn, (size_t)2u); | ||
| } | ||
| else | ||
| { | ||
| if (num_input >= packn) | ||
| weight_data_tm.create(packn * maxk, num_input / packn + num_input % packn, num_output, (size_t)2u); |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
No description provided.