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

ggml : introduce ggml_status (ggml/750)

Browse source 
* using enum as an exit code instead of macros

* update return type from enum to unsigned int

* indentation fix

* compound update
ggml_compute_exit_code -> ggml_status
changed ggml_status from a bit-field type to simple codes
ggml_status to string cast

* ggml_status to string cast

* GGML_CALL was removed

Co-authored-by: slaren <slarengh@gmail.com>

---------

Co-authored-by: slaren <slarengh@gmail.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Michael Podvitskiy 2024-03-04 10:05:42 +01:00 • committed by Georgi Gerganov
parent fe52be11e3
commit 9fa2627347
11 changed files with 88 additions and 63 deletions
Download patch file Download diff file Expand all files Collapse all files

29
ggml.c
View file

@ -320,6 +320,16 @@ static ggml_fp16_t ggml_table_exp_f16[1 << 16];
// precomputed f32 table for f16 (256 KB) (ggml-impl.h)
float ggml_table_f32_f16[1 << 16];
const char * ggml_status_to_string(enum ggml_status status) {
switch (status) {
case GGML_STATUS_ALLOC_FAILED: return "GGML status: error (failed to allocate memory)";
case GGML_STATUS_FAILED: return "GGML status: error (operation failed)";
case GGML_STATUS_SUCCESS: return "GGML status: success";
case GGML_STATUS_ABORTED: return "GGML status: warning (operation aborted)";
default: GGML_ASSERT(false);
}
}
// note: do not use these inside ggml.c
// these are meant to be used via the ggml.h API
float ggml_fp16_to_fp32(ggml_fp16_t x) {
@ -17400,6 +17410,7 @@ struct ggml_compute_state {
ggml_thread_t thrd;
int ith;
struct ggml_compute_state_shared * shared;
enum ggml_status ec;
};
static void ggml_graph_compute_perf_stats_node(struct ggml_tensor * node, const struct ggml_compute_state_shared * st) {
@ -17693,7 +17704,8 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
while (true) {
if (cplan->abort_callback && cplan->abort_callback(cplan->abort_callback_data)) {
state->shared->node_n += 1;
return (thread_ret_t) GGML_EXIT_ABORTED;
state->ec = GGML_STATUS_ABORTED;
return 0;
}
if (atomic_fetch_sub(&state->shared->n_active, 1) == 1) {
@ -17815,7 +17827,7 @@ static thread_ret_t ggml_graph_compute_thread(void * data) {
}
}
return GGML_EXIT_SUCCESS;
return 0;
}
struct ggml_cplan ggml_graph_plan(const struct ggml_cgraph * cgraph, int n_threads) {
@ -18011,7 +18023,7 @@ struct ggml_cplan ggml_graph_plan(const struct ggml_cgraph * cgraph, int n_threa
return cplan;
}
int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
{
GGML_ASSERT(cplan);
GGML_ASSERT(cplan->n_threads > 0);
@ -18055,6 +18067,7 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
.thrd = 0,
.ith = j,
.shared = &state_shared,
.ec = GGML_STATUS_SUCCESS,
};
const int rc = ggml_thread_create(&workers[j].thrd, NULL, ggml_graph_compute_thread, &workers[j]);
@ -18065,12 +18078,14 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
workers[0].ith = 0;
workers[0].shared = &state_shared;
workers[0].ec = GGML_STATUS_SUCCESS;
const int64_t perf_start_cycles = ggml_perf_cycles();
const int64_t perf_start_time_us = ggml_perf_time_us();
// this is a work thread too
int compute_status = (size_t) ggml_graph_compute_thread(&workers[0]);
ggml_graph_compute_thread(&workers[0]);
enum ggml_status compute_status = workers[0].ec;
// don't leave affinity set on the main thread
clear_numa_thread_affinity();
@ -18080,6 +18095,8 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
for (int j = 1; j < n_threads; j++) {
const int rc = ggml_thread_join(workers[j].thrd, NULL);
GGML_ASSERT(rc == 0);
if (workers[j].ec != GGML_STATUS_SUCCESS)
compute_status = workers[j].ec;
}
}
@ -18107,14 +18124,14 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
return compute_status;
}
void ggml_graph_compute_with_ctx(struct ggml_context * ctx, struct ggml_cgraph * cgraph, int n_threads) {
enum ggml_status ggml_graph_compute_with_ctx(struct ggml_context * ctx, struct ggml_cgraph * cgraph, int n_threads) {
struct ggml_cplan cplan = ggml_graph_plan(cgraph, n_threads);
struct ggml_object * obj = ggml_new_object(ctx, GGML_OBJECT_TYPE_WORK_BUFFER, cplan.work_size);
cplan.work_data = (uint8_t *)ctx->mem_buffer + obj->offs;
ggml_graph_compute(cgraph, &cplan);
return ggml_graph_compute(cgraph, &cplan);
}
struct ggml_tensor * ggml_graph_get_tensor(struct ggml_cgraph * cgraph, const char * name) {