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:
co-authored by
Max Krasnyansky
parent
a0ed91a442
commit
2b10b62677
@@ -102,11 +102,9 @@ int op_sum_rows(struct htp_ops_context * octx) {
|
||||
return HTP_STATUS_OK;
|
||||
}
|
||||
|
||||
const int n_threads = octx->n_threads;
|
||||
const uint32_t src0_nrows = ne01 * ne02 * ne03;
|
||||
|
||||
uint32_t n_jobs = MIN(n_threads, src0_nrows);
|
||||
uint32_t rows_per_thread = (src0_nrows + n_jobs - 1) / n_jobs;
|
||||
const uint32_t n_threads = MIN(octx->n_threads, src0_nrows);
|
||||
const uint32_t rows_per_thread = (src0_nrows + n_threads - 1) / n_threads;
|
||||
|
||||
bool opt_path = false;
|
||||
if ((0 == hex_is_aligned((void *) src0->data, VLEN)) && !(nb01 & (VLEN - 1))) {
|
||||
@@ -124,7 +122,7 @@ int op_sum_rows(struct htp_ops_context * octx) {
|
||||
.opt_path = opt_path,
|
||||
};
|
||||
|
||||
worker_pool_run_func(octx->ctx->worker_pool, sum_rows_thread_f32, &smctx, n_jobs);
|
||||
worker_pool_run_func(octx->ctx->worker_pool, sum_rows_thread_f32, &smctx, n_threads);
|
||||
|
||||
return HTP_STATUS_OK;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user