ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

ggml : upgrade init_tensor API to return a ggml_status (#11854)

Browse source 
* Upgrade init_tensor API to return a ggml_status

To prepare for an 'abort-free' ggml
(ggml not to abort on OOMs but return a OOM status),
as agreeed with Diego in the ggml repo,
upgrade the init_tensor() and view_init() APIs
to return a ggml_status.

* misc fixes

---------

Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
William Tambellini 2025-02-28 05:41:47 -08:00 • committed by GitHub
parent c43a3e7996
commit 70680c48e5
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
16 changed files with 136 additions and 97 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml/src/ggml-cuda/ggml-cuda.cu
View file

@ -540,12 +540,12 @@ static void * ggml_backend_cuda_buffer_get_base(ggml_backend_buffer_t buffer) {
return ctx->dev_ptr;
}
static void ggml_backend_cuda_buffer_init_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor) {
static enum ggml_status ggml_backend_cuda_buffer_init_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor) {
ggml_backend_cuda_buffer_context * ctx = (ggml_backend_cuda_buffer_context *)buffer->context;
if (tensor->view_src != NULL) {
assert(tensor->view_src->buffer->buft == buffer->buft);
return;
return GGML_STATUS_SUCCESS;
}
if (ggml_is_quantized(tensor->type) && tensor->view_src == nullptr && ggml_backend_buffer_get_usage(buffer) != GGML_BACKEND_BUFFER_USAGE_COMPUTE) {
@ -558,6 +558,7 @@ static void ggml_backend_cuda_buffer_init_tensor(ggml_backend_buffer_t buffer, g
CUDA_CHECK(cudaMemset((char *)tensor->data + original_size, 0, padded_size - original_size));
}
}
return GGML_STATUS_SUCCESS;
}
static void ggml_backend_cuda_buffer_memset_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
@ -792,7 +793,7 @@ static void * ggml_backend_cuda_split_buffer_get_base(ggml_backend_buffer_t buff
GGML_UNUSED(buffer);
}
static void ggml_backend_cuda_split_buffer_init_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor) {
static enum ggml_status ggml_backend_cuda_split_buffer_init_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor) {
GGML_ASSERT(tensor->view_src == nullptr); // views of split tensors are not supported
ggml_backend_cuda_split_buffer_context * ctx = (ggml_backend_cuda_split_buffer_context *)buffer->context;
@ -838,6 +839,7 @@ static void ggml_backend_cuda_split_buffer_init_tensor(ggml_backend_buffer_t buf
}
}
tensor->extra = extra;
return GGML_STATUS_SUCCESS;
}
static void ggml_backend_cuda_split_buffer_set_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor, const void * data, size_t offset, size_t size) {