Skip to content

Commit aaaac34

Browse files
tannevaledclaude
andcommitted
docs: where the parallel GEMM loses on Zen 3, and a 2-D tiling measured and not shipped
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1 parent 80cdfdf commit aaaac34

1 file changed

Lines changed: 22 additions & 4 deletions

File tree

‎docs/perf.md‎

Lines changed: 22 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -401,10 +401,28 @@ makes anyway), so a transposed, sliced, reversed or broadcast operand costs no
401401
extra pass.
402402

403403
**Still behind OpenBLAS with 16 threads** (0.54–0.66× on large products, 0.3×
404-
below 256²): per-core throughput in the parallel run is about a third of the
405-
serial kernel's, so the next lever is the parallel schedule (a persistent
406-
worker pool instead of a goroutine launch per round, and the BLIS jr-loop
407-
split for short m), then an AVX-512 kernel for the hosts that have it.
404+
below 256²). Where the parallel loss is, measured on the Zen 3:
405+
406+
- It is not the hardware: 16 *independent* serial 512² GEMMs, one per core,
407+
each ran at 41 GFLOP/s (656 in total), while the parallel 1024² GEMM ran its
408+
kernel at 15–17 GFLOP/s per core. Per-core efficiency already drops from 40
409+
to ~30 GFLOP/s at 2 cores.
410+
- It is not scheduling overhead: 83% of the parallel profile is the
411+
micro-kernel itself, which stalls on the shared packed B panel. On this
412+
part the last-level cache is split between core complexes, so the panel is
413+
partly remote.
414+
- Taller row bands (more reuse of each B micro-panel) do not help: interleaved
415+
over 5 runs they were worse at 1024² and equal at 2048².
416+
- **A 2-D tiling of C with private packing** (no barrier, nothing shared) was
417+
built and measured, and is **not shipped**: ×1.24 at 1024² on the Zen 3, but
418+
×0.62 at 512² and Inner, and ×0.70–0.93 at every size on M4, whose cluster
419+
L2 holds the shared panel well. A tile repacks (rows + cols)·k elements for
420+
rows·cols·k multiply-adds, and a strided pack costs far more per element than
421+
an FMA, so tiles under ~256×256 lose.
422+
423+
The next lever is therefore a shared B panel per core complex (or per cache
424+
domain), not a different split of C, then an AVX-512 kernel for the hosts that
425+
have it.
408426

409427
Correctness was checked on real hardware for five of the six 64-bit targets:
410428
amd64 (Zen 3, FMA path), arm64 (cfarm185), ppc64le (cfarm120), riscv64 (cfarm94)

0 commit comments

Comments
 (0)