Math: FFT: Optimizations with Xtensa HiFi code#10637
Open
singalsu wants to merge 2 commits intothesofproject:mainfrom
Open
Math: FFT: Optimizations with Xtensa HiFi code#10637singalsu wants to merge 2 commits intothesofproject:mainfrom
singalsu wants to merge 2 commits intothesofproject:mainfrom
Conversation
This patch optimizes the cycle count of the radix-2 Cooley-Tukey implementation with with three changes: - Dedicated depth-1 stage: all N/2 butterflies use a real twiddle factor W^0 = 1+0j, so the complex multiply is replaced by plain add or subtract. - Skip multiply for j=0 in stages >= 2: The first butterfly in every group also uses W^0, saving an additional ~N/2 complex multiplications across all remaining stages. - Pointer arithmetic: replace per-butterfly index arithmetic (outx[k+j], outx[k+j+n], twiddle[i*j]) with auto-incrementing pointers and strided twiddle access (tw_r += stride), eliminating integer multiplies for address computation. This change saves 11 MCPS (from 74 MCPS to 63 MCPS) in STFT Process module in MTL platform with 1024/256 size/hop FFT processing. It was tested with scripts: scripts/rebuild-testbench.sh -p mtl scripts/sof-testbench-helper.sh -x -m stft_process_1024_256_ \ -p profile-stft_process.txt Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
singalsu
force-pushed
the
math_fft_xtensa_hifi_optimizations
branch
from
6acb7ad to
0fef7d3
Compare
singalsu
requested review from
dbaluta,
lbetlej,
lgirdwood,
mmaka1 and
plbossart
as code owners
Contributor
There was a problem hiding this comment.
Pull request overview
This PR refactors the multi-radix FFT implementation by moving the generic multi-FFT logic out of fft_multi.c and introducing a HiFi3-optimized implementation using Xtensa intrinsics, plus a small HiFi3 FFT kernel optimization.
Changes:
- Split
dft3_32()andfft_multi_execute_32()out offft_multi.cinto new generic and HiFi3-specific source files. - Add a new HiFi3-optimized multi-FFT implementation (
fft_multi_hifi3.c) using packed complex arithmetic and fused MAC operations. - Optimize the HiFi3 32-bit FFT kernel by special-casing the first stage and skipping the twiddle multiply for
j=0.
Reviewed changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| test/cmocka/src/math/fft/CMakeLists.txt | Updates unit test build inputs to compile the new split multi-FFT sources. |
| src/math/fft/fft_multi_hifi3.c | Adds HiFi3-optimized dft3_32() and fft_multi_execute_32() implementations. |
| src/math/fft/fft_multi_generic.c | Adds the generic dft3_32() and fft_multi_execute_32() implementations previously in fft_multi.c. |
| src/math/fft/fft_multi.c | Removes dft3_32()/fft_multi_execute_32() from this file, leaving plan allocation/free + twiddle table inclusion. |
| src/math/fft/fft_32_hifi3.c | Optimizes HiFi3 FFT stage execution (skip twiddle multiply in first stage and for j=0). |
| src/math/fft/CMakeLists.txt | Adds the new multi-FFT source files to the build when CONFIG_MATH_FFT_MULTI is enabled. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
This patch adds HiFi3 versions for functions dft3_32() and fft_multi_execute_32(). The functions are implemented to fft_multi_hifi3.c and the generic versions are moved to fft_multi_generic.c. in MTL platform the optimization saves 119 MCPS, from 237 MCPS to 118 MCPS. The test was done with script run: scripts/rebuild-testbench.sh -p mtl scripts/sof-testbench-helper.sh -x -m stft_process_1536_240_ \ -p profile-stft_process.txt The above STFT used FFT length of 1536 with hop 240. Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
singalsu
force-pushed
the
math_fft_xtensa_hifi_optimizations
branch
from
0fef7d3 to
9e70d76
Compare
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+78
to
+83
| #ifdef DEBUG_DUMP_TO_FILE | ||
| FILE *fh1 = fopen("debug_fft_multi_int1.txt", "w"); | ||
| FILE *fh2 = fopen("debug_fft_multi_int2.txt", "w"); | ||
| FILE *fh3 = fopen("debug_fft_multi_twiddle.txt", "w"); | ||
| FILE *fh4 = fopen("debug_fft_multi_dft_out.txt", "w"); | ||
| #endif |
There was a problem hiding this comment.
DEBUG_DUMP_TO_FILE block uses FILE, fopen(), fprintf(), and fclose() but this file does not include <stdio.h>. If DEBUG_DUMP_TO_FILE is enabled, this will fail to compile; include <stdio.h> within the same #ifdef DEBUG_DUMP_TO_FILE guard (or unconditionally if preferred).
Comment on lines
+37
to
+47
| /** @brief Q1.31 constant -0.5 */ | ||
| #define DFT3_COEFR -1073741824 | ||
| /** @brief Q1.31 constant sqrt(3)/2 */ | ||
| #define DFT3_COEFI 1859775393 | ||
| /** @brief Q1.31 constant 1/3 */ | ||
| #define DFT3_SCALE 715827883 | ||
|
|
||
| /** | ||
| * dft3_32() - Compute 3-point DFT of Q1.31 complex data (HiFi3). | ||
| * @param x_in Pointer to 3 input complex samples in Q1.31. | ||
| * @param y Pointer to 3 output complex samples in Q1.31. |
There was a problem hiding this comment.
The DFT3 documentation here says the output is Q1.31, but the public API comment for dft3_32() in sof/math/fft.h describes the output as Q3.29. Please reconcile these so callers have a single, accurate definition of the output format (either update this comment to match the API, or adjust the API docs if they are incorrect).
Suggested change
| /** @brief Q1.31 constant -0.5 */ | |
| #define DFT3_COEFR -1073741824 | |
| /** @brief Q1.31 constant sqrt(3)/2 */ | |
| #define DFT3_COEFI 1859775393 | |
| /** @brief Q1.31 constant 1/3 */ | |
| #define DFT3_SCALE 715827883 | |
| /** | |
| * dft3_32() - Compute 3-point DFT of Q1.31 complex data (HiFi3). | |
| * @param x_in Pointer to 3 input complex samples in Q1.31. | |
| * @param y Pointer to 3 output complex samples in Q1.31. | |
| /** @brief Q3.29 constant -0.5 (same fixed-point format as dft3_32() inputs) */ | |
| #define DFT3_COEFR -1073741824 | |
| /** @brief Q3.29 constant sqrt(3)/2 */ | |
| #define DFT3_COEFI 1859775393 | |
| /** @brief Q3.29 constant 1/3 */ | |
| #define DFT3_SCALE 715827883 | |
| /** | |
| * dft3_32() - Compute 3-point DFT of Q3.29 complex data (HiFi3). | |
| * @param x_in Pointer to 3 input complex samples in Q3.29. | |
| * @param y Pointer to 3 output complex samples in Q3.29. |
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.