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

* 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 639 additions and 380 deletions
+18 -7
View File
@@ -1865,15 +1865,26 @@ static bool ggml_hexagon_supported_binary(const struct ggml_hexagon_session * se
const struct ggml_tensor * src1 = op->src[1];
const struct ggml_tensor * dst = op;
if (src0->type != GGML_TYPE_F32) {
return false;
}
if (src1->type != GGML_TYPE_F32) {
return false;
}
if (dst->type != GGML_TYPE_F32) {
if (src0->type == GGML_TYPE_F32) {
if (src1->type != GGML_TYPE_F32) {
return false;
}
if (dst->type != GGML_TYPE_F32) {
return false;
}
}
else if (src0->type == GGML_TYPE_F16) {
if (src1->type != GGML_TYPE_F16) {
return false;
}
if (dst->type != GGML_TYPE_F16) {
return false;
}
}
else {
return false;
}
if (!ggml_are_same_shape(src0, dst)) {
return false;
}