Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 22:08:46 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : fixed runtime bugs and compile errors related to GGML_PERF and GGML_DEBUG (#2219)

Browse Source 
* fixed runtime bugs and compile errors related to GGML_PERF and GGML_DEBUG

* remove ifdef GGML_PERF; update fmt
This commit is contained in:
Qingyou Meng 2023-07-17 03:57:28 +08:00 committed by GitHub
parent 27ab66e437
commit 672dda10e4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 5 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
ggml.c
View File

@ -4412,8 +4412,8 @@ void ggml_free(struct ggml_context * ctx) {
if (&g_state.contexts[i].context == ctx) { if (&g_state.contexts[i].context == ctx) {
g_state.contexts[i].used = false; g_state.contexts[i].used = false;
GGML_PRINT_DEBUG("%s: context %d with %d objects has been freed. memory used = %zu\n", GGML_PRINT_DEBUG("%s: context %d has been freed. memory used = %zu\n",
__func__, i, ctx->n_objects, ctx->objects_end->offs + ctx->objects_end->size); __func__, i, ggml_used_mem(ctx));
if (ctx->mem_buffer_owned) { if (ctx->mem_buffer_owned) {
GGML_ALIGNED_FREE(ctx->mem_buffer); GGML_ALIGNED_FREE(ctx->mem_buffer);
@ -16317,8 +16317,8 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
if (GGML_OP_HAS_FINALIZE[node->op]) { if (GGML_OP_HAS_FINALIZE[node->op]) {
params.nth = n_tasks_arr[node_n]; params.nth = n_tasks_arr[node_n];
ggml_compute_forward(&params, node); ggml_compute_forward(&params, node);
ggml_graph_compute_perf_stats_node(node, state->shared);
} }
ggml_graph_compute_perf_stats_node(node, state->shared);
} }
// distribute new work or execute it direct if 1T // distribute new work or execute it direct if 1T
@ -16348,8 +16348,9 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
if (GGML_OP_HAS_FINALIZE[node->op]) { if (GGML_OP_HAS_FINALIZE[node->op]) {
params.type = GGML_TASK_FINALIZE; params.type = GGML_TASK_FINALIZE;
ggml_compute_forward(&params, node); ggml_compute_forward(&params, node);
ggml_graph_compute_perf_stats_node(node, state->shared);
} }
ggml_graph_compute_perf_stats_node(node, state->shared);
} else { } else {
break; break;
} }
@ -16891,9 +16892,6 @@ static void ggml_graph_export_node(const struct ggml_tensor * tensor, const char
} }
void ggml_graph_export(const struct ggml_cgraph * cgraph, const char * fname) { void ggml_graph_export(const struct ggml_cgraph * cgraph, const char * fname) {
//assert(cgraph->work == NULL);
//assert(cgraph->work_size == 0);
uint64_t size_eval = 0; uint64_t size_eval = 0;
// compute size of intermediate results // compute size of intermediate results
@ -17332,9 +17330,6 @@ void ggml_graph_print(const struct ggml_cgraph * cgraph) {
GGML_PRINT("=== GRAPH ===\n"); GGML_PRINT("=== GRAPH ===\n");
GGML_PRINT_DEBUG("n_threads = %d\n", cgraph->n_threads);
GGML_PRINT_DEBUG("total work size = %zu bytes\n", cgraph->work_size);
GGML_PRINT("n_nodes = %d\n", cgraph->n_nodes); GGML_PRINT("n_nodes = %d\n", cgraph->n_nodes);
for (int i = 0; i < cgraph->n_nodes; i++) { for (int i = 0; i < cgraph->n_nodes; i++) {
struct ggml_tensor * node = cgraph->nodes[i]; struct ggml_tensor * node = cgraph->nodes[i];