diff options
| author | Onur Gündüz <ogunduz@gmail.com> | 2026-06-25 17:58:46 +0200 |
|---|---|---|
| committer | Onur Gündüz <ogunduz@gmail.com> | 2026-06-25 17:58:46 +0200 |
| commit | f7ef995ff54da671f9d7665b3d92ff25934c1e8b (patch) | |
| tree | 1401d04c0f464b32bffd63f787a63a82c6959e97 /libffwd/backend_cuda.c | |
| parent | 417f861f3661c2253258b5683a627d961e81fc68 (diff) | |
| download | ffwd-f7ef995ff54da671f9d7665b3d92ff25934c1e8b.tar.gz ffwd-f7ef995ff54da671f9d7665b3d92ff25934c1e8b.zip | |
chore: run clang-format
Diffstat (limited to 'libffwd/backend_cuda.c')
| -rw-r--r-- | libffwd/backend_cuda.c | 21 |
1 files changed, 7 insertions, 14 deletions
diff --git a/libffwd/backend_cuda.c b/libffwd/backend_cuda.c index dd05c16..a925404 100644 --- a/libffwd/backend_cuda.c +++ b/libffwd/backend_cuda.c @@ -58,8 +58,8 @@ int ffwd_init(const ffwd_options_t *opts, char *err, size_t errlen) { return 0; } -ffwd_t *ffwd_open( - const char *model_dir, int is_late, const ffwd_options_t *opts, char *err, size_t errlen) { +ffwd_t * +ffwd_open(const char *model_dir, int is_late, const ffwd_options_t *opts, char *err, size_t errlen) { (void)opts; ffwd_t *b = (ffwd_t *)calloc(1, sizeof(*b)); if (!b || !(b->dir = strdup(model_dir))) { @@ -108,9 +108,7 @@ const ffwd_config_t *ffwd_config(const ffwd_t *b) { return b->is_late ? ffwd_cuda_late_config(b->late_ctx) : ffwd_cuda_config(b->ctx); } -int ffwd_token_dim(const ffwd_t *b) { - return b->is_late ? ffwd_cuda_late_token_dim(b->late_ctx) : 0; -} +int ffwd_token_dim(const ffwd_t *b) { return b->is_late ? ffwd_cuda_late_token_dim(b->late_ctx) : 0; } int ffwd_uses_dense_batches(const ffwd_t *b) { (void)b; @@ -121,10 +119,7 @@ int ffwd_encode_batch(ffwd_t *b, const ffwd_input_t *inputs, int batch, float *o return ffwd_cuda_encode_batch(b->ctx, inputs, batch, out); } -int ffwd_encode_spans_batch(ffwd_t *b, - const ffwd_context_input_t *inputs, - int batch, - float *out) { +int ffwd_encode_spans_batch(ffwd_t *b, const ffwd_context_input_t *inputs, int batch, float *out) { return ffwd_cuda_encode_spans_batch(b->ctx, inputs, batch, out); } @@ -142,14 +137,12 @@ int ffwd_rerank(ffwd_t *b, const ffwd_rerank_input_t *in, float *scores) { /* GPU encodes the query and every candidate in one packed forward; MaxSim * runs on the host, where the grouped-GEMM scorer beats a device graph. */ if (rc == 0) - rc = - ffwd_cuda_late_encode_tokens(b->late_ctx, in->query_ids, in->query_n_tokens, 1, query); + rc = ffwd_cuda_late_encode_tokens(b->late_ctx, in->query_ids, in->query_n_tokens, 1, query); if (rc == 0) rc = ffwd_cuda_late_encode_docs(b->late_ctx, in->doc_ids, in->doc_n_tokens, in->doc_keep, - in->doc_n_keep, in->n_docs, 1, docs, offsets); + in->doc_n_keep, in->n_docs, 1, docs, offsets); if (rc == 0) - rc = ffwd_late_maxsim_batch(query, in->query_n_keep, docs, offsets, in->n_docs, dim, - scores); + rc = ffwd_late_maxsim_batch(query, in->query_n_keep, docs, offsets, in->n_docs, dim, scores); free(offsets); free(docs); free(query); |
