Conversation
The CGEMM N packing step was still using the generic C kernel/generic/zgemm_ncopy_8.c, whose address arithmetic is scalar: for every complex element of an 8-column panel it builds one base+offset address per store, which perf annotate shows as eight add/addi per iteration. Add kernel/riscv64/zgemm_ncopy_8_rvv.c and point CGEMMONCOPY at it. Each of the eight source rows is loaded as one strided segment-2 (real / imaginary) load, both halves are merged with vget/vset, and the panel is written with two strided segment-8 stores whose byte stride is 16*sizeof(FLOAT). The eight per-store addresses collapse into a single stride. The n&4 / n&2 / n&1 tails keep the contiguous segment-8 / segment-4 / segment-2 layout, so the packed byte sequence is unchanged - it was checked against the generic packer over 27200 m x n cases with no mismatch. The file follows the usual oncopy convention of being written for both precisions behind a single #if !defined(DOUBLE) switch, so it can also serve ZGEMM. Only the CGEMM side is wired up here, matching the C prefix the benchmark covers; the -DDOUBLE build was verified to compile and to match the generic packer, but ZGEMM still uses ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c. Measured on a SpaceMiT X100 (2.2 GHz, single thread, ZVL256B), median of three runs: cblas_cgemm -0.5% at 256, +4.0% at 512, +1.1% at 1024. Tests: make tests returns 0; 125/125 utest and 1473/1473 extension tests pass; the CBLAS L1/L2/L3 suites show no new failures. Co-authored-by: Yuansheng <yuansheng@isrc.iscas.ac.cn> Co-authored-by: Ning Tian <tianning24@iscas.ac.cn> Signed-off-by: jiakai xu <xujiakai2025@iscas.ac.cn>
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
2 participants
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.
The CGEMM N packing step was still using the generic C
kernel/generic/zgemm_ncopy_8.c, whose address arithmetic is scalar: for every complex element of an 8-column panel it builds one base+offset address per store, whichperf annotateshows as eightadd/addiper iteration.Add
kernel/riscv64/zgemm_ncopy_8_rvv.cand pointCGEMMONCOPYat it. Each of the eight source rows is loaded as one strided segment-2 (real / imaginary) load, both halves are merged withvget/vset, and the panel is written with two strided segment-8 stores whose byte stride is16*sizeof(FLOAT). The eight per-store addresses collapse into a single stride. Then&4/n&2/n&1tails keep the contiguous segment-8 / segment-4 / segment-2 layout, so the packed byte sequence is unchanged — it was checked against the generic packer over 27200 m x n cases with no mismatch.The file follows the usual oncopy convention of being written for both precisions behind a single
#if !defined(DOUBLE)switch, so it can also serve ZGEMM. Only the CGEMM side is wired up here, matching the C prefix the benchmark covers; the-DDOUBLEbuild was verified to compile and to match the generic packer, butZGEMMstill uses../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c.Performance
Measured on a SpaceMiT X100 (2.2 GHz, single thread, ZVL256B), median of three runs:
cblas_cgemm-0.5% at 256, +4.0% at 512, +1.1% at 1024.Testing
make testsreturns 0; 125/125 utest and 1473/1473 extension tests pass; the CBLAS L1/L2/L3 suites show no new failures.