Repository navigation
perf: VSX kernels on ppc64le (sum, dot, sqrt, elementwise, 8x8 GEMM) - #20
Merged
Merged
Conversation
Go's ppc64 assembler has the VSX loads, stores and permutes but no vector double arithmetic, so ppc64le ran the scalar code. go-asmgen v0.13.0 now encodes xvadddp/xvmaddadp/... as WORDs (pinned against GNU as, run bit-identical to Go's arithmetic and math.FMA on POWER8 and POWER9); this uses them for sum, dot, sqrt, add/sub/mul/div and an 8x8 GEMM micro-kernel (sum_ppc64le.s, generated). max/min stay scalar: xvmaxdp's NaN rule is not NumPy's. The GEMM kernel follows OpenBLAS's POWER9 dgemm scheme: no splats; A pairs and their swaps multiply B column pairs into diagonal/anti-diagonal accumulators, rebuilt into rows with xxpermdi at the store. Measured in one binary on a POWER9 core: 13.3 GFLOP/s with lxvdsx splats, 14.8 with xxpermdi splats, 15.4 unrolled, 16.9 with the diagonal scheme (shipped). POWER9 (cfarm29), against the scalar code, best of 3-4 interleaved runs: micro-kernel 8.3 -> 16.2 GFLOP/s; serial GEMM 512^2 7.9 -> 15.8 MatMul 1024^2 (32 threads) 39.2 -> 23.6 ms; 128^2 x1.8; 512^2 x1.5 MatVec x3.1; Sum x3.2-7.3 below 4 Mi; AddInto x2.7-4.1; Sqrt x1.05-1.18 Against OpenBLAS 0.3.34's POWER9 kernel on the same host: serial 0.69x, 32 threads 0.56x. Tests: the whole suite passes on a real POWER8 and POWER9; replacing xvsubdp's operand order or the store's xxpermdi selector makes it fail there. CI regenerates sum_ppc64le.s and fails if it is stale. Docs and comments that said ppc64le has no SIMD kernel are updated. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tannevaled
added a commit
to go-ndarray/go-ndarray.github.io
that referenced
this pull request
Oct 4, 2026
ndarray v0.2.3 (go-ndarray/ndarray#20) adds VSX SIMD kernels on ppc64le. The SIMD wording now names amd64, arm64 and ppc64le, ppc64le leaves the list of targets running the pure-Go reference code, the GEMM blurb lists the VSX 8x8 micro-kernel next to NEON and AVX2/FMA, and the library card shows v0.2.3. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
tannevaled
added a commit
to go-ndarray/docs
that referenced
this pull request
Oct 4, 2026
ndarray v0.2.3 (go-ndarray/ndarray#20) adds VSX SIMD kernels on ppc64le. - performance.md: re-synced verbatim from docs/perf.md at v0.2.3 (new section 'ppc64le: VSX kernels (2026-10-04)' and the SIMD coverage update); the summary is now 'Where it stands at v0.2.3', names amd64, arm64 and ppc64le and gives the POWER9 GEMM figures from that section. - index.md: SIMD on amd64, arm64 and ppc64le, worded as the ndarray README at v0.2.3 (max/min SIMD on amd64 only, VSX 8x8 micro-kernel). - roadmap.md: the 'Today (v0.2.1)' sentence and the 'other four targets' clause no longer leave ppc64le on scalar code. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
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
1 participant
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.
Go's ppc64 assembler has the VSX loads, stores and permutes but no vector
double arithmetic, so ppc64le ran the scalar code. go-asmgen v0.13.0 now
encodes xvadddp/xvmaddadp/... as WORDs (pinned against GNU as, run
bit-identical to Go's arithmetic and math.FMA on POWER8 and POWER9); this
uses them for sum, dot, sqrt, add/sub/mul/div and an 8x8 GEMM micro-kernel
(sum_ppc64le.s, generated). max/min stay scalar: xvmaxdp's NaN rule is not
NumPy's.
The GEMM kernel follows OpenBLAS's POWER9 dgemm scheme: no splats; A pairs
and their swaps multiply B column pairs into diagonal/anti-diagonal
accumulators, rebuilt into rows with xxpermdi at the store. Measured in one
binary on a POWER9 core: 13.3 GFLOP/s with lxvdsx splats, 14.8 with
xxpermdi splats, 15.4 unrolled, 16.9 with the diagonal scheme (shipped).
POWER9 (cfarm29), against the scalar code, best of 3-4 interleaved runs:
micro-kernel 8.3 -> 16.2 GFLOP/s; serial GEMM 512^2 7.9 -> 15.8
MatMul 1024^2 (32 threads) 39.2 -> 23.6 ms; 128^2 x1.8; 512^2 x1.5
MatVec x3.1; Sum x3.2-7.3 below 4 Mi; AddInto x2.7-4.1; Sqrt x1.05-1.18
Against OpenBLAS 0.3.34's POWER9 kernel on the same host: serial 0.69x,
32 threads 0.56x.
Tests: the whole suite passes on a real POWER8 and POWER9; replacing
xvsubdp's operand order or the store's xxpermdi selector makes it fail
there. CI regenerates sum_ppc64le.s and fails if it is stale. Docs and
comments that said ppc64le has no SIMD kernel are updated.
🤖 Generated with Claude Code