Skip to content

Commit

Permalink
ggml : move more prints to the ggml log system (ggerganov#9839)
Browse files Browse the repository at this point in the history 
* ggml : move more prints to the ggml log system

* show BLAS OpenMP warnings in all builds using debug print
  • Loading branch information
@slaren
slaren authored Oct 11, 2024
1 parent 7eee341 commit 9677640
Show file tree
Hide file tree
Showing 4 changed files with 48 additions and 48 deletions.
34 changes: 17 additions & 17 deletions ggml/src/ggml-alloc.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@

//#define GGML_ALLOCATOR_DEBUG

//#define AT_PRINTF(...) fprintf(stderr, __VA_ARGS__)
//#define AT_PRINTF(...) GGML_LOG_DEBUG(__VA_ARGS__)
#define AT_PRINTF(...)


Expand Down Expand Up @@ -89,7 +89,7 @@ void ggml_tallocr_alloc(struct ggml_tallocr * talloc, struct ggml_tensor * tenso
size = GGML_PAD(size, talloc->alignment);

if (talloc->offset + size > ggml_backend_buffer_get_size(talloc->buffer)) {
fprintf(stderr, "%s: not enough space in the buffer to allocate %s (needed %zu, available %zu)\n",
GGML_LOG_ERROR("%s: not enough space in the buffer to allocate %s (needed %zu, available %zu)\n",
__func__, tensor->name, size, ggml_backend_buffer_get_size(talloc->buffer) - talloc->offset);
GGML_ABORT("not enough space in the buffer");
}
Expand Down Expand Up @@ -172,7 +172,7 @@ static size_t ggml_dyn_tallocr_alloc(struct ggml_dyn_tallocr * alloc, size_t siz
best_fit_block = alloc->n_free_blocks - 1;
} else {
// this should never happen
fprintf(stderr, "%s: not enough space in the buffer to allocate %zu bytes, largest block available %zu bytes\n",
GGML_LOG_ERROR("%s: not enough space in the buffer to allocate %zu bytes, largest block available %zu bytes\n",
__func__, size, max_avail);
GGML_ABORT("not enough space in the buffer");
}
Expand Down Expand Up @@ -209,16 +209,16 @@ static size_t ggml_dyn_tallocr_alloc(struct ggml_dyn_tallocr * alloc, size_t siz
}
}
}
fprintf(stderr, "max_size = %.2f MB: tensors: ", cur_max / 1024.0 / 1024.0);
GGML_LOG_DEBUG("max_size = %.2f MB: tensors: ", cur_max / 1024.0 / 1024.0);
for (int i = 0; i < 1024; i++) {
if (alloc->allocated_tensors[i].tensor) {
fprintf(stderr, "%s [%zx-%zx] (%.2f MB) ", alloc->allocated_tensors[i].tensor->name,
GGML_LOG_DEBUG("%s [%zx-%zx] (%.2f MB) ", alloc->allocated_tensors[i].tensor->name,
alloc->allocated_tensors[i].offset,
alloc->allocated_tensors[i].offset + ggml_nbytes(alloc->allocated_tensors[i].tensor),
ggml_nbytes(alloc->allocated_tensors[i].tensor) / 1024.0 / 1024.0);
}
}
fprintf(stderr, "\n");
GGML_LOG_DEBUG("\n");
}
#endif

Expand Down Expand Up @@ -768,13 +768,13 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
// even if there are no tensors allocated in this buffer, we still need to allocate it to initialize views
if (new_size > cur_size || galloc->buffers[i] == NULL) {
#ifndef NDEBUG
fprintf(stderr, "%s: reallocating %s buffer from size %.02f MiB to %.02f MiB\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), cur_size / 1024.0 / 1024.0, new_size / 1024.0 / 1024.0);
GGML_LOG_DEBUG("%s: reallocating %s buffer from size %.02f MiB to %.02f MiB\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), cur_size / 1024.0 / 1024.0, new_size / 1024.0 / 1024.0);
#endif

ggml_backend_buffer_free(galloc->buffers[i]);
galloc->buffers[i] = ggml_backend_buft_alloc_buffer(galloc->bufts[i], new_size);
if (galloc->buffers[i] == NULL) {
fprintf(stderr, "%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), new_size);
GGML_LOG_ERROR("%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), new_size);
return false;
}
ggml_backend_buffer_set_usage(galloc->buffers[i], GGML_BACKEND_BUFFER_USAGE_COMPUTE);
Expand Down Expand Up @@ -825,14 +825,14 @@ static bool ggml_gallocr_node_needs_realloc(ggml_gallocr_t galloc, struct ggml_t
static bool ggml_gallocr_needs_realloc(ggml_gallocr_t galloc, struct ggml_cgraph * graph) {
if (galloc->n_nodes != graph->n_nodes) {
#ifndef NDEBUG
fprintf(stderr, "%s: graph has different number of nodes\n", __func__);
GGML_LOG_DEBUG("%s: graph has different number of nodes\n", __func__);
#endif
return true;
}

if (galloc->n_leafs != graph->n_leafs) {
#ifndef NDEBUG
fprintf(stderr, "%s: graph has different number of leafs\n", __func__);
GGML_LOG_DEBUG("%s: graph has different number of leafs\n", __func__);
#endif
return true;
}
Expand All @@ -843,7 +843,7 @@ static bool ggml_gallocr_needs_realloc(ggml_gallocr_t galloc, struct ggml_cgraph

if (!ggml_gallocr_node_needs_realloc(galloc, node, &node_alloc->dst)) {
#ifndef NDEBUG
fprintf(stderr, "%s: node %s is not valid\n", __func__, node->name);
GGML_LOG_DEBUG("%s: node %s is not valid\n", __func__, node->name);
#endif
return true;
}
Expand All @@ -855,7 +855,7 @@ static bool ggml_gallocr_needs_realloc(ggml_gallocr_t galloc, struct ggml_cgraph
}
if (!ggml_gallocr_node_needs_realloc(galloc, src, &node_alloc->src[j])) {
#ifndef NDEBUG
fprintf(stderr, "%s: src %d (%s) of node %s is not valid\n", __func__, j, src->name, node->name);
GGML_LOG_DEBUG("%s: src %d (%s) of node %s is not valid\n", __func__, j, src->name, node->name);
#endif
return true;
}
Expand All @@ -869,14 +869,14 @@ bool ggml_gallocr_alloc_graph(ggml_gallocr_t galloc, struct ggml_cgraph * graph)
if (ggml_gallocr_needs_realloc(galloc, graph)) {
if (galloc->n_buffers == 1) {
#ifndef NDEBUG
fprintf(stderr, "%s: reallocating buffers automatically\n", __func__);
GGML_LOG_DEBUG("%s: reallocating buffers automatically\n", __func__);
#endif
if (!ggml_gallocr_reserve(galloc, graph)) {
return false;
}
} else {
#ifndef NDEBUG
fprintf(stderr, "%s: cannot reallocate multi buffer graph automatically, call reserve\n", __func__);
GGML_LOG_DEBUG("%s: cannot reallocate multi buffer graph automatically, call reserve\n", __func__);
#endif
return false;
}
Expand Down Expand Up @@ -940,7 +940,7 @@ static bool alloc_tensor_range(struct ggml_context * ctx,
ggml_backend_buffer_t buffer = ggml_backend_buft_alloc_buffer(buft, size);
if (buffer == NULL) {
#ifndef NDEBUG
fprintf(stderr, "%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(buft), size);
GGML_LOG_DEBUG("%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(buft), size);
#endif
for (size_t i = 0; i < *n_buffers; i++) {
ggml_backend_buffer_free((*buffers)[i]);
Expand Down Expand Up @@ -990,7 +990,7 @@ ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_conte
}

if (this_size > max_size) {
fprintf(stderr, "%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n",
GGML_LOG_ERROR("%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n",
__func__, t->name,
ggml_backend_buft_name(buft),
this_size, max_size);
Expand Down Expand Up @@ -1022,7 +1022,7 @@ ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_conte

if (n_buffers == 0) {
#ifndef NDEBUG
fprintf(stderr, "%s: all tensors in the context are already allocated\n", __func__);
GGML_LOG_DEBUG("%s: all tensors in the context are already allocated\n", __func__);
#endif
return NULL;
}
Expand Down
32 changes: 16 additions & 16 deletions ggml/src/ggml-backend.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -379,7 +379,7 @@ void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst
ggml_backend_tensor_get(src, dst->data, 0, ggml_nbytes(src));
} else if (!ggml_backend_buffer_copy_tensor(src, dst)) {
#ifndef NDEBUG
fprintf(stderr, "%s: warning: slow copy from %s to %s\n", __func__, ggml_backend_buffer_name(src->buffer), ggml_backend_buffer_name(dst->buffer));
GGML_LOG_DEBUG("%s: warning: slow copy from %s to %s\n", __func__, ggml_backend_buffer_name(src->buffer), ggml_backend_buffer_name(dst->buffer));
#endif
size_t nbytes = ggml_nbytes(src);
void * data = malloc(nbytes);
Expand Down Expand Up @@ -571,7 +571,7 @@ struct ggml_backend_registry {

void register_backend(ggml_backend_reg_t reg) {
#ifndef NDEBUG
fprintf(stderr, "%s: registered backend %s (%zu devices)\n",
GGML_LOG_DEBUG("%s: registered backend %s (%zu devices)\n",
__func__, ggml_backend_reg_name(reg), ggml_backend_reg_dev_count(reg));
#endif
backends.push_back(reg);
Expand All @@ -582,7 +582,7 @@ struct ggml_backend_registry {

void register_device(ggml_backend_dev_t device) {
#ifndef NDEBUG
fprintf(stderr, "%s: registered device %s (%s)\n", __func__, ggml_backend_dev_name(device), ggml_backend_dev_description(device));
GGML_LOG_DEBUG("%s: registered device %s (%s)\n", __func__, ggml_backend_dev_name(device), ggml_backend_dev_description(device));
#endif
devices.push_back(device);
}
Expand Down Expand Up @@ -773,7 +773,7 @@ static ggml_backend_buffer_t ggml_backend_cpu_buffer_type_alloc_buffer(ggml_back
size += TENSOR_ALIGNMENT; // malloc may return an address that is not aligned
void * data = malloc(size); // TODO: use GGML_ALIGNED_MALLOC (move to ggml-impl.h)
if (data == NULL) {
fprintf(stderr, "%s: failed to allocate buffer of size %zu\n", __func__, size);
GGML_LOG_ERROR("%s: failed to allocate buffer of size %zu\n", __func__, size);
return NULL;
}

Expand Down Expand Up @@ -836,7 +836,7 @@ static ggml_backend_buffer_t ggml_backend_cpu_hbm_buffer_type_alloc_buffer(ggml_
void * ptr;
int result = hbw_posix_memalign(&ptr, ggml_backend_cpu_buffer_type_get_alignment(buft), size);
if (result != 0) {
fprintf(stderr, "failed to allocate HBM buffer of size %zu\n", size);
GGML_LOG_ERROR("failed to allocate HBM buffer of size %zu\n", size);
return NULL;
}

Expand Down Expand Up @@ -1459,7 +1459,7 @@ static int ggml_backend_sched_backend_from_buffer(ggml_backend_sched_t sched, co
}

#ifndef NDEBUG
fprintf(stderr, "%s: warning: no backend supports op %s with a weight with buffer type %s used in tensor %s, the weight will need to be copied\n",
GGML_LOG_DEBUG("%s: warning: no backend supports op %s with a weight with buffer type %s used in tensor %s, the weight will need to be copied\n",
__func__, ggml_op_desc(tensor), ggml_backend_buffer_name(buffer), tensor->name);
#endif

Expand Down Expand Up @@ -1548,32 +1548,32 @@ static void ggml_backend_sched_print_assignments(ggml_backend_sched_t sched, str
for (int i = 0; i < graph->n_nodes; i++) {
if (cur_split < sched->n_splits && i == sched->splits[cur_split].i_start) {
ggml_backend_t split_backend = sched->backends[sched->splits[cur_split].backend_id];
fprintf(stderr, "\n## SPLIT #%d: %s # %d inputs: ", cur_split, ggml_backend_name(split_backend),
GGML_LOG_DEBUG("\n## SPLIT #%d: %s # %d inputs: ", cur_split, ggml_backend_name(split_backend),
sched->splits[cur_split].n_inputs);
for (int j = 0; j < sched->splits[cur_split].n_inputs; j++) {
fprintf(stderr, "[%s (%5.5s)] ", sched->splits[cur_split].inputs[j]->name,
GGML_LOG_DEBUG("[%s (%5.5s)] ", sched->splits[cur_split].inputs[j]->name,
fmt_size(ggml_nbytes(sched->splits[cur_split].inputs[j])));
}
fprintf(stderr, "\n");
GGML_LOG_DEBUG("\n");
cur_split++;
}
struct ggml_tensor * node = graph->nodes[i];
if (ggml_is_view_op(node->op)) {
continue;
}
ggml_backend_t tensor_backend = ggml_backend_sched_get_tensor_backend(sched, node);
fprintf(stderr, "node #%3d (%10.10s): %20.20s (%5.5s) [%5.5s %8.8s]:", i, ggml_op_name(node->op), node->name,
GGML_LOG_DEBUG("node #%3d (%10.10s): %20.20s (%5.5s) [%5.5s %8.8s]:", i, ggml_op_name(node->op), node->name,
fmt_size(ggml_nbytes(node)), tensor_backend ? ggml_backend_name(tensor_backend) : "NULL", GET_CAUSE(node));
for (int j = 0; j < GGML_MAX_SRC; j++) {
struct ggml_tensor * src = node->src[j];
if (src == NULL) {
continue;
}
ggml_backend_t src_backend = ggml_backend_sched_get_tensor_backend(sched, src);
fprintf(stderr, " %20.20s (%5.5s) [%5.5s %8.8s]", src->name,
GGML_LOG_DEBUG(" %20.20s (%5.5s) [%5.5s %8.8s]", src->name,
fmt_size(ggml_nbytes(src)), src_backend ? ggml_backend_name(src_backend) : "NULL", GET_CAUSE(src));
}
fprintf(stderr, "\n");
GGML_LOG_DEBUG("\n");
}
}

Expand Down Expand Up @@ -2087,11 +2087,11 @@ static bool ggml_backend_sched_alloc_splits(ggml_backend_sched_t sched) {
// the re-allocation may cause the split inputs to be moved to a different address
ggml_backend_sched_synchronize(sched);
#ifndef NDEBUG
fprintf(stderr, "%s: failed to allocate graph, reserving (backend_ids_changed = %d)\n", __func__, backend_ids_changed);
GGML_LOG_DEBUG("%s: failed to allocate graph, reserving (backend_ids_changed = %d)\n", __func__, backend_ids_changed);
#endif
ggml_gallocr_reserve_n(sched->galloc, &sched->graph, sched->node_backend_ids, sched->leaf_backend_ids);
if (!ggml_gallocr_alloc_graph(sched->galloc, &sched->graph)) {
fprintf(stderr, "%s: failed to allocate graph\n", __func__);
GGML_LOG_ERROR("%s: failed to allocate graph\n", __func__);
return false;
}
}
Expand Down Expand Up @@ -2485,7 +2485,7 @@ struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, s
struct ggml_context * ctx_unallocated = ggml_init(params);

if (ctx_allocated == NULL || ctx_unallocated == NULL) {
fprintf(stderr, "failed to allocate context for graph copy\n");
GGML_LOG_ERROR("%s: failed to allocate context for graph copy\n", __func__);
ggml_hash_set_free(&hash_set);
free(node_copies);
free(node_init);
Expand All @@ -2508,7 +2508,7 @@ struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, s
// allocate nodes
ggml_backend_buffer_t buffer = ggml_backend_alloc_ctx_tensors(ctx_allocated, backend);
if (buffer == NULL) {
fprintf(stderr, "failed to allocate buffer for graph copy\n");
GGML_LOG_ERROR("%s: failed to allocate buffer for graph copy\n", __func__);
ggml_hash_set_free(&hash_set);
free(node_copies);
free(node_init);
Expand Down
8 changes: 4 additions & 4 deletions ggml/src/ggml-blas.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -297,14 +297,14 @@ ggml_backend_t ggml_backend_blas_init(void) {
/* .context = */ ctx,
};

#if !defined(NDEBUG) && defined(OPENBLAS_VERSION) && defined(GGML_USE_OPENMP)
#if defined(OPENBLAS_VERSION) && defined(GGML_USE_OPENMP)
if (openblas_get_parallel() != OPENBLAS_OPENMP) {
fprintf(stderr, "%s: warning: ggml is using OpenMP, but OpenBLAS was compiled without OpenMP support\n", __func__);
GGML_LOG_DEBUG("%s: warning: ggml is using OpenMP, but OpenBLAS was compiled without OpenMP support\n", __func__);
}
#endif

#if !defined(NDEBUG) && defined(BLIS_ENABLE_CBLAS) && defined(GGML_USE_OPENMP) && !defined(BLIS_ENABLE_OPENMP)
fprintf(stderr, "%s: warning: ggml is using OpenMP, but BLIS was compiled without OpenMP support\n", __func__);
#if defined(BLIS_ENABLE_CBLAS) && defined(GGML_USE_OPENMP) && !defined(BLIS_ENABLE_OPENMP)
GGML_LOG_DEBUG("%s: warning: ggml is using OpenMP, but BLIS was compiled without OpenMP support\n", __func__);
#endif

return backend;
Expand Down
Loading

0 comments on commit 9677640

Please sign in to comment.