holm/whisper.cpp
1
1
Fork 0
forked from extern/whisper.cpp
Code Issues Pull Requests Packages Projects Releases Wiki Activity

tmp : demonstrate how to measure time of ggml ops

Browse Source
This commit is contained in:
Georgi Gerganov 2023-03-09 09:23:45 +02:00
parent 09e9068007
commit 4f074fb7a8
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
2 changed files with 7 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml.c
View File

@ -79,7 +79,7 @@ typedef void* thread_ret_t;
#define static_assert(cond, msg) _Static_assert(cond, msg)
#endif
/*#define GGML_PERF*/
#define GGML_PERF
#define GGML_DEBUG 0
#define GGML_GELU_FP16

6
whisper.cpp
View File

@ -2160,6 +2160,12 @@ static bool whisper_decode_internal(
ggml_graph_compute (ctx0, &gf);
}
// print the time for computing the last ggml_mul_mat that computes logits
// also print the total decoder time
// these need to be called after ggml_graph_compute()
printf("logits t = %7.3f ms (%2d runs, N = %3d, ggml_mul_mat: [%d x %d] * [%d x %d])\n", 1e-3*double(logits->perf_time_us)/logits->perf_runs, logits->perf_runs, N, logits->ne[0], logits->ne[1], cur->ne[1], cur->ne[0]);
printf("total t = %7.3f ms (%2d runs)\n", 1e-3*double(gf.perf_time_us)/gf.perf_runs, gf.perf_runs);
// extract logits for all N tokens
//logits_out.resize(N*n_vocab);
//memcpy(logits_out.data(), ggml_get_data(logits), sizeof(float)*N*n_vocab);