nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 20 Packages Projects Releases Wiki Activity

ggml-cpu: use runtime SVE width in FWHT (#24059)

Browse Source
This commit is contained in:
Charles Xu
2026-06-03 13:45:10 +03:00
committed by GitHub
parent f8f0a47a55
commit 3571fa5435
1 changed files with 10 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cpu/ops.cpp
+10 -1
View File
@@ -8955,7 +8955,12 @@ static void ggml_compute_forward_flash_attn_ext_f16(
k->type == v->type && k->type == v->type &&
neq1 >= Q_TILE_SZ); neq1 >= Q_TILE_SZ);
#ifdef GGML_SIMD #ifdef GGML_SIMD
use_tiled &= (DV % GGML_F32_EPR == 0); #if defined(__ARM_FEATURE_SVE)
const int64_t f32_epr = svcntw();
#else
const int64_t f32_epr = GGML_F32_EPR;
#endif
use_tiled &= (DV % f32_epr == 0);
#endif #endif
int current_chunk = ith; int current_chunk = ith;
@@ -11358,7 +11363,11 @@ static void ggml_compute_forward_fwht_f32(const ggml_compute_params * params, gg
// Scalar passes // Scalar passes
#if defined(GGML_SIMD) #if defined(GGML_SIMD)
#if defined(__ARM_FEATURE_SVE)
const int step = svcntw();
#else
const int step = GGML_F32_EPR; const int step = GGML_F32_EPR;
#endif
#else #else
const int step = n; const int step = n;
#endif #endif