root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-24 18:34:36 +00:00
Code Issues Actions 7 Packages Projects Releases Wiki Activity

metal : handle zero-sized allocs (#9466)

Browse Source
This commit is contained in:
Georgi Gerganov 2024-09-16 09:05:56 +03:00 committed by GitHub
parent 90a2fff0e7
commit c4965a64f7
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 29 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

48
ggml/src/ggml-metal.m
View File

@ -3225,15 +3225,19 @@ GGML_CALL static ggml_backend_buffer_t ggml_backend_metal_buffer_type_alloc_buff
ctx->n_buffers = 1; ctx->n_buffers = 1;
if (ctx->all_data != NULL) { if (ctx->all_data != NULL) {
ctx->buffers[0].data = ctx->all_data; ctx->buffers[0].data = ctx->all_data;
ctx->buffers[0].size = size; ctx->buffers[0].size = size;
ctx->buffers[0].metal = [device newBufferWithBytesNoCopy:ctx->all_data ctx->buffers[0].metal = nil;
length:size_aligned
options:MTLResourceStorageModeShared if (size_aligned > 0) {
deallocator:nil]; ctx->buffers[0].metal = [device newBufferWithBytesNoCopy:ctx->all_data
length:size_aligned
options:MTLResourceStorageModeShared
deallocator:nil];
}
} }
if (ctx->all_data == NULL || ctx->buffers[0].metal == nil) { if (size_aligned > 0 && (ctx->all_data == NULL || ctx->buffers[0].metal == nil)) {
GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_aligned / 1024.0 / 1024.0); GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_aligned / 1024.0 / 1024.0);
free(ctx); free(ctx);
ggml_backend_metal_free_device(); ggml_backend_metal_free_device();
@ -3310,14 +3314,17 @@ GGML_CALL ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data,
// the buffer fits into the max buffer size allowed by the device // the buffer fits into the max buffer size allowed by the device
if (size_aligned <= device.maxBufferLength) { if (size_aligned <= device.maxBufferLength) {
ctx->buffers[ctx->n_buffers].data = data; ctx->buffers[ctx->n_buffers].data = data;
ctx->buffers[ctx->n_buffers].size = size; ctx->buffers[ctx->n_buffers].size = size;
ctx->buffers[ctx->n_buffers].metal = nil;
ctx->buffers[ctx->n_buffers].metal = [device newBufferWithBytesNoCopy:data length:size_aligned options:MTLResourceStorageModeShared deallocator:nil]; if (size_aligned > 0) {
ctx->buffers[ctx->n_buffers].metal = [device newBufferWithBytesNoCopy:data length:size_aligned options:MTLResourceStorageModeShared deallocator:nil];
if (ctx->buffers[ctx->n_buffers].metal == nil) { if (ctx->buffers[ctx->n_buffers].metal == nil) {
GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_aligned / 1024.0 / 1024.0); GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_aligned / 1024.0 / 1024.0);
return false; return false;
}
} }
ggml_backend_metal_log_allocated_size(device, size_aligned); ggml_backend_metal_log_allocated_size(device, size_aligned);
@ -3333,14 +3340,17 @@ GGML_CALL ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data,
for (size_t i = 0; i < size; i += size_step) { for (size_t i = 0; i < size; i += size_step) {
const size_t size_step_aligned = (i + size_view <= size) ? size_view : (size_aligned - i); const size_t size_step_aligned = (i + size_view <= size) ? size_view : (size_aligned - i);
ctx->buffers[ctx->n_buffers].data = (void *) ((uint8_t *) data + i); ctx->buffers[ctx->n_buffers].data = (void *) ((uint8_t *) data + i);
ctx->buffers[ctx->n_buffers].size = size_step_aligned; ctx->buffers[ctx->n_buffers].size = size_step_aligned;
ctx->buffers[ctx->n_buffers].metal = nil;
ctx->buffers[ctx->n_buffers].metal = [device newBufferWithBytesNoCopy:(void *) ((uint8_t *) data + i) length:size_step_aligned options:MTLResourceStorageModeShared deallocator:nil]; if (size_step_aligned > 0) {
ctx->buffers[ctx->n_buffers].metal = [device newBufferWithBytesNoCopy:(void *) ((uint8_t *) data + i) length:size_step_aligned options:MTLResourceStorageModeShared deallocator:nil];
if (ctx->buffers[ctx->n_buffers].metal == nil) { if (ctx->buffers[ctx->n_buffers].metal == nil) {
GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_step_aligned / 1024.0 / 1024.0); GGML_METAL_LOG_ERROR("%s: error: failed to allocate buffer, size = %8.2f MiB\n", __func__, size_step_aligned / 1024.0 / 1024.0);
return false; return false;
}
} }
ggml_backend_metal_log_allocated_size(device, size_step_aligned); ggml_backend_metal_log_allocated_size(device, size_step_aligned);