Skip to content

Add checks for buffer size with Metal #1706

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 6, 2023
Merged

Add checks for buffer size with Metal #1706

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions ggml-metal.m
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -204,6 +204,11 @@ bool ggml_metal_add_buffer(
ctx->buffers[ctx->n_buffers].name = name;
ctx->buffers[ctx->n_buffers].data = data;
ctx->buffers[ctx->n_buffers].size = size;

if (ctx->device.maxBufferLength < aligned_size) {
fprintf(stderr, "%s: buffer '%s' size %zu is larger than buffer maximum of %zu\n", __func__, name, aligned_size, ctx->device.maxBufferLength);
return false;
}
ctx->buffers[ctx->n_buffers].metal = [ctx->device newBufferWithBytesNoCopy:data length:aligned_size options:MTLResourceStorageModeShared deallocator:nil];

if (ctx->buffers[ctx->n_buffers].metal == nil) {
Expand Down
27 changes: 20 additions & 7 deletions llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2405,17 +2405,30 @@ struct llama_context * llama_init_from_file(
// this allocates all Metal resources and memory buffers
ctx->ctx_metal = ggml_metal_init();

void *data_ptr = NULL;
size_t data_size = 0;
if (params.use_mmap) {
ggml_metal_add_buffer(ctx->ctx_metal, "data", ctx->model.mapping->addr, ctx->model.mapping->size);
ggml_metal_add_buffer(ctx->ctx_metal, "eval", ctx->buf_compute.addr, ctx->buf_compute.size);
data_ptr = ctx->model.mapping->addr;
data_size= ctx->model.mapping->size;
} else {
ggml_metal_add_buffer(ctx->ctx_metal, "data", ggml_get_mem_buffer(ctx->model.ctx), ggml_get_mem_size(ctx->model.ctx));
ggml_metal_add_buffer(ctx->ctx_metal, "eval", ctx->buf_compute.addr, ctx->buf_compute.size);
data_ptr = ggml_get_mem_buffer(ctx->model.ctx);
data_size= ggml_get_mem_size(ctx->model.ctx);
}

ggml_metal_add_buffer(ctx->ctx_metal, "kv", ctx->model.kv_self.buf.addr, ctx->model.kv_self.buf.size);
ggml_metal_add_buffer(ctx->ctx_metal, "scr0", ctx->buf_scratch[0].addr, ctx->buf_scratch[0].size);
ggml_metal_add_buffer(ctx->ctx_metal, "scr1", ctx->buf_scratch[1].addr, ctx->buf_scratch[1].size);
#define LLAMA_METAL_CHECK_BUF(result) \
if (!(result)) { \
fprintf(stderr, "%s: failed to add buffer\n", __func__); \
llama_free(ctx); \
return NULL; \
}

LLAMA_METAL_CHECK_BUF(ggml_metal_add_buffer(ctx->ctx_metal, "data", data_ptr, data_size));
LLAMA_METAL_CHECK_BUF(ggml_metal_add_buffer(ctx->ctx_metal, "eval", ctx->buf_compute.addr, ctx->buf_compute.size));

LLAMA_METAL_CHECK_BUF(ggml_metal_add_buffer(ctx->ctx_metal, "kv", ctx->model.kv_self.buf.addr, ctx->model.kv_self.buf.size));
LLAMA_METAL_CHECK_BUF(ggml_metal_add_buffer(ctx->ctx_metal, "scr0", ctx->buf_scratch[0].addr, ctx->buf_scratch[0].size));
LLAMA_METAL_CHECK_BUF(ggml_metal_add_buffer(ctx->ctx_metal, "scr1", ctx->buf_scratch[1].addr, ctx->buf_scratch[1].size));
#undef LLAMA_METAL_CHECK_BUF
}
#endif

Expand Down