mirror of
https://github.com/likelovewant/ollama-for-amd.git
synced 2025-12-21 22:33:56 +00:00
Update GGML to b6646 (#12245)
Notable EOLs with this change: - MacOS v12 and v13 are no longer supported (v14+ required) - AMD gfx900 and gfx906 are no longer supported
This commit is contained in:
Daniel Hiltgen
@@ -6,12 +6,12 @@ Subject: [PATCH] graph memory reporting on failure
|
||||
---
|
||||
ggml/include/ggml-alloc.h | 1 +
|
||||
ggml/include/ggml-backend.h | 1 +
|
||||
ggml/src/ggml-alloc.c | 36 ++++++++++++++++++++++++++++++++----
|
||||
ggml/src/ggml-alloc.c | 34 +++++++++++++++++++++++++++++++---
|
||||
ggml/src/ggml-backend.cpp | 7 +++++++
|
||||
4 files changed, 41 insertions(+), 4 deletions(-)
|
||||
4 files changed, 40 insertions(+), 3 deletions(-)
|
||||
|
||||
diff --git a/ggml/include/ggml-alloc.h b/ggml/include/ggml-alloc.h
|
||||
index 2cb150fd2..7ab3f0192 100644
|
||||
index 2cb150fd..7ab3f019 100644
|
||||
--- a/ggml/include/ggml-alloc.h
|
||||
+++ b/ggml/include/ggml-alloc.h
|
||||
@@ -65,6 +65,7 @@ GGML_API bool ggml_gallocr_reserve_n(
|
||||
@@ -23,31 +23,31 @@ index 2cb150fd2..7ab3f0192 100644
|
||||
// Utils
|
||||
// Create a buffer and allocate all the tensors in a ggml_context
|
||||
diff --git a/ggml/include/ggml-backend.h b/ggml/include/ggml-backend.h
|
||||
index a2977ea2e..e8cf30841 100644
|
||||
index 62b6d65e..fe20dca3 100644
|
||||
--- a/ggml/include/ggml-backend.h
|
||||
+++ b/ggml/include/ggml-backend.h
|
||||
@@ -303,6 +303,7 @@ extern "C" {
|
||||
GGML_API int ggml_backend_sched_get_n_copies(ggml_backend_sched_t sched);
|
||||
@@ -316,6 +316,7 @@ extern "C" {
|
||||
|
||||
GGML_API size_t ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend);
|
||||
+ GGML_API size_t ggml_backend_sched_get_attempted_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend);
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_sched_get_buffer_type(ggml_backend_sched_t sched, ggml_backend_t backend);
|
||||
GGML_API size_t ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend);
|
||||
+ GGML_API size_t ggml_backend_sched_get_attempted_buffer_size(ggml_backend_sched_t sched, ggml_backend_t backend);
|
||||
|
||||
GGML_API void ggml_backend_sched_set_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend);
|
||||
GGML_API ggml_backend_t ggml_backend_sched_get_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node);
|
||||
diff --git a/ggml/src/ggml-alloc.c b/ggml/src/ggml-alloc.c
|
||||
index 8b6e60283..b58bd671d 100644
|
||||
index fa46f3b4..421ff7c7 100644
|
||||
--- a/ggml/src/ggml-alloc.c
|
||||
+++ b/ggml/src/ggml-alloc.c
|
||||
@@ -350,6 +350,7 @@ struct node_alloc {
|
||||
@@ -492,6 +492,7 @@ struct node_alloc {
|
||||
struct ggml_gallocr {
|
||||
ggml_backend_buffer_type_t * bufts; // [n_buffers]
|
||||
ggml_backend_buffer_t * buffers; // [n_buffers]
|
||||
struct vbuffer ** buffers; // [n_buffers]
|
||||
+ size_t *buffer_sizes; // [n_buffers]
|
||||
struct ggml_dyn_tallocr ** buf_tallocs; // [n_buffers]
|
||||
int n_buffers;
|
||||
|
||||
@@ -373,6 +374,9 @@ ggml_gallocr_t ggml_gallocr_new_n(ggml_backend_buffer_type_t * bufts, int n_bufs
|
||||
galloc->buffers = calloc(n_bufs, sizeof(ggml_backend_buffer_t));
|
||||
@@ -515,6 +516,9 @@ ggml_gallocr_t ggml_gallocr_new_n(ggml_backend_buffer_type_t * bufts, int n_bufs
|
||||
galloc->buffers = calloc(n_bufs, sizeof(struct vbuffer *));
|
||||
GGML_ASSERT(galloc->buffers != NULL);
|
||||
|
||||
+ galloc->buffer_sizes = calloc(n_bufs, sizeof(size_t));
|
||||
@@ -56,7 +56,7 @@ index 8b6e60283..b58bd671d 100644
|
||||
galloc->buf_tallocs = calloc(n_bufs, sizeof(struct ggml_dyn_tallocr *));
|
||||
GGML_ASSERT(galloc->buf_tallocs != NULL);
|
||||
|
||||
@@ -439,6 +443,7 @@ void ggml_gallocr_free(ggml_gallocr_t galloc) {
|
||||
@@ -582,6 +586,7 @@ void ggml_gallocr_free(ggml_gallocr_t galloc) {
|
||||
ggml_hash_set_free(&galloc->hash_set);
|
||||
free(galloc->hash_values);
|
||||
free(galloc->bufts);
|
||||
@@ -64,7 +64,7 @@ index 8b6e60283..b58bd671d 100644
|
||||
free(galloc->buffers);
|
||||
free(galloc->buf_tallocs);
|
||||
free(galloc->node_allocs);
|
||||
@@ -734,6 +739,8 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
|
||||
@@ -875,6 +880,8 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
|
||||
}
|
||||
}
|
||||
|
||||
@@ -73,23 +73,21 @@ index 8b6e60283..b58bd671d 100644
|
||||
// reallocate buffers if needed
|
||||
for (int i = 0; i < galloc->n_buffers; i++) {
|
||||
// if the buffer type is used multiple times, we reuse the same buffer
|
||||
@@ -755,15 +762,20 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
|
||||
@@ -896,14 +903,19 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
|
||||
|
||||
ggml_backend_buffer_free(galloc->buffers[i]);
|
||||
galloc->buffers[i] = ggml_backend_buft_alloc_buffer(galloc->bufts[i], new_size);
|
||||
ggml_vbuffer_free(galloc->buffers[i]);
|
||||
galloc->buffers[i] = ggml_vbuffer_alloc(galloc->bufts[i], galloc->buf_tallocs[i], GGML_BACKEND_BUFFER_USAGE_COMPUTE);
|
||||
- if (galloc->buffers[i] == NULL) {
|
||||
+ if (galloc->buffers[i]) {
|
||||
+ galloc->buffer_sizes[i] = ggml_backend_buffer_get_size(galloc->buffers[i]);
|
||||
+ ggml_backend_buffer_set_usage(galloc->buffers[i], GGML_BACKEND_BUFFER_USAGE_COMPUTE);
|
||||
+ galloc->buffer_sizes[i] = ggml_vbuffer_size(galloc->buffers[i]);
|
||||
+ } else {
|
||||
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;
|
||||
+ galloc->buffer_sizes[i] = new_size;
|
||||
+ success = false;
|
||||
}
|
||||
- ggml_backend_buffer_set_usage(galloc->buffers[i], GGML_BACKEND_BUFFER_USAGE_COMPUTE);
|
||||
+ } else {
|
||||
+ galloc->buffer_sizes[i] = ggml_backend_buffer_get_size(galloc->buffers[i]);
|
||||
+ galloc->buffer_sizes[i] = ggml_vbuffer_size(galloc->buffers[i]);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -98,8 +96,8 @@ index 8b6e60283..b58bd671d 100644
|
||||
}
|
||||
|
||||
bool ggml_gallocr_reserve(ggml_gallocr_t galloc, struct ggml_cgraph *graph) {
|
||||
@@ -920,6 +932,22 @@ size_t ggml_gallocr_get_buffer_size(ggml_gallocr_t galloc, int buffer_id) {
|
||||
return ggml_backend_buffer_get_size(galloc->buffers[buffer_id]);
|
||||
@@ -1058,6 +1070,22 @@ size_t ggml_gallocr_get_buffer_size(ggml_gallocr_t galloc, int buffer_id) {
|
||||
return ggml_vbuffer_size(galloc->buffers[buffer_id]);
|
||||
}
|
||||
|
||||
+size_t ggml_gallocr_get_attempted_buffer_size(ggml_gallocr_t galloc, int buffer_id) {
|
||||
@@ -122,10 +120,10 @@ index 8b6e60283..b58bd671d 100644
|
||||
|
||||
static void free_buffers(ggml_backend_buffer_t ** buffers, const size_t * n_buffers) {
|
||||
diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
|
||||
index 97f47abd2..d02a40e60 100644
|
||||
index 8ba86f82..cb2b9956 100644
|
||||
--- a/ggml/src/ggml-backend.cpp
|
||||
+++ b/ggml/src/ggml-backend.cpp
|
||||
@@ -1631,6 +1631,13 @@ size_t ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backe
|
||||
@@ -1809,6 +1809,13 @@ size_t ggml_backend_sched_get_buffer_size(ggml_backend_sched_t sched, ggml_backe
|
||||
return ggml_gallocr_get_buffer_size(sched->galloc, backend_index);
|
||||
}
|
||||
|
||||
@@ -137,5 +135,5 @@ index 97f47abd2..d02a40e60 100644
|
||||
+}
|
||||
+
|
||||
void ggml_backend_sched_set_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend) {
|
||||
GGML_ASSERT(sched);
|
||||
int backend_index = ggml_backend_sched_backend_id(sched, backend);
|
||||
GGML_ASSERT(backend_index >= 0 && backend_index < sched->n_backends);
|
||||
|
||||
Reference in New Issue
Block a user