vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Use VLAs instead of alloca, if possible.

Browse source
This commit is contained in:
Olaf Seibert 2023-04-12 20:37:41 +02:00
parent 315a95a4d3
commit 2299b0a5f5
1 changed files with 10 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

10
ggml.c
View file

@ -3,11 +3,13 @@
#include "ggml.h"
#if __STDC_NO_VLA__
#if defined(_MSC_VER) || defined(__MINGW32__)
#include <malloc.h> // using malloc.h with MSC/MINGW
#elif !defined(__FreeBSD__) && !defined(__NetBSD__) && !defined(__OpenBSD__)
#include <alloca.h>
#endif
#endif /* __STDC_NO_VLA__ */
#include <assert.h>
#include <errno.h>
@ -10362,7 +10364,11 @@ void ggml_graph_compute(struct ggml_context * ctx, struct ggml_cgraph * cgraph)
/*.has_work =*/ false,
/*.stop =*/ false,
};
#if __STDC_NO_VLA__
struct ggml_compute_state * workers = n_threads > 1 ? alloca(sizeof(struct ggml_compute_state)*(n_threads - 1)) : NULL;
#else
struct ggml_compute_state workers[n_threads];
#endif
// create thread pool
if (n_threads > 1) {
@ -11374,7 +11380,11 @@ static enum ggml_opt_result ggml_opt_lbfgs(
ggml_opt_get_params(np, ps, x);
// the L-BFGS memory
#if __STDC_NO_VLA__
struct ggml_lbfgs_iteration_data * lm = alloca(sizeof(struct ggml_lbfgs_iteration_data)*m);
#else
struct ggml_lbfgs_iteration_data lm[m];
#endif
for (int i = 0; i < m; ++i) {
lm[i].alpha = 0.0f;