Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

hexagon: add fp16 support for binary ops: add,sub,mul,div (#20139)

Browse Source 
* hexagon: add fp16 support for binary ops: add,sub,mul,div

* hexagon: fix test-backend-ops failures for fp16 binary ops on older arches (<v79)

* hexagon: decide on n_threads (aka n_jobs) early to avoid overallocating scratchpad

* snapdragon: fix readme link

---------

Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
This commit is contained in:
YardenTal44
2026-03-05 18:29:13 -08:00
committed by GitHub
co-authored by Max Krasnyansky
parent a0ed91a442
commit 2b10b62677
16 changed files with 637 additions and 378 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-hexagon/htp/cpy-ops.c
+4 -3
View File
@@ -202,6 +202,8 @@ static void cpy_work_func(unsigned int n, unsigned int i, void *data) {
int op_cpy(struct htp_ops_context * octx) {
cpy_preamble;
const uint32_t n_threads = MIN(nr, octx->n_threads);
struct htp_copy_context ct;
ct.octx = octx;
@@ -227,8 +229,7 @@ int op_cpy(struct htp_ops_context * octx) {
const bool transposed = (nb00 > nb01) || (nb0 > nb1);
const bool sameshape = !transposed && (ne00 == ne0 && ne01 == ne1 && ne02 == ne2 && ne03 == ne3);
const uint32_t n_jobs = MIN(nr, octx->n_threads);
ct.src0_nrows_per_thread = (nr + n_jobs - 1) / n_jobs;
ct.src0_nrows_per_thread = (nr + n_threads - 1) / n_threads;
if (sametype && sameshape) {
ct.copy = cpy_thread_sametype_sameshape;
@@ -245,7 +246,7 @@ int op_cpy(struct htp_ops_context * octx) {
return HTP_STATUS_NO_SUPPORT;
}
worker_pool_run_func(octx->ctx->worker_pool, cpy_work_func, &ct, n_jobs);
worker_pool_run_func(octx->ctx->worker_pool, cpy_work_func, &ct, n_threads);
return HTP_STATUS_OK;
}