From cfdaa69b9fb2fe5fab256e6d82f5d1f7786dfd3a Mon Sep 17 00:00:00 2001 From: hmeiland Date: Fri, 10 Jul 2026 05:44:15 +0200 Subject: [PATCH] U74 asm kernel: note full-memory HPL result in header Full-memory HPL N=27456 (asm hybrid + Q=256) measured at 5.99 GFLOPS (residual PASSED, ~50% of the 12 GF peak, +10.7% over the tuned C kernel, 1.81x the stock 2x2) - the best clean figure. Doc-only header update. --- kernel/generic/gemmkernel_4x4_u74.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/kernel/generic/gemmkernel_4x4_u74.c b/kernel/generic/gemmkernel_4x4_u74.c index 42fe108ce..73eed1a57 100644 --- a/kernel/generic/gemmkernel_4x4_u74.c +++ b/kernel/generic/gemmkernel_4x4_u74.c @@ -54,9 +54,10 @@ * double-buffering (P/Q ping-pong) and load-before-FMA issue ordering that * reaches the U74 FP-pipe peak (~16.5 cycles per 16 fmadd.d). All other * shapes, odd bk, and the TRMM builds fall back to the portable C 4x4 code - * below. Measured ~+20% single-core DGEMM and +4% four-core HPL over the C - * kernel; validated against the full BLAS Level-3 suite (DGEMM 17,496 calls, - * 0 failures) and HPL (residual PASSED). + * below. Measured ~+20% single-core DGEMM over the C kernel; four-core HPL vs + * the tuned C kernel is +4% at N=10000 and +10.7% at N=27456 (5.99 GFLOPS, + * ~50% of the 12 GF peak - the best clean figure). Validated against the full + * BLAS Level-3 suite (DGEMM 17,496 calls, 0 failures) and HPL (residual PASSED). */ #include "common.h"