happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

cann: move offload op check after device context declaration

This commit is contained in:
DocShotgun 2026-01-02 14:02:06 -08:00
parent a449358f83
commit 7a838e78fc
1 changed files with 20 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
ggml/src/ggml-cann/ggml-cann.cpp
View File

@ -2497,27 +2497,6 @@ static bool ggml_backend_buft_is_cann(ggml_backend_buffer_type_t buft) {
return buft->iface.get_name == ggml_backend_cann_buffer_type_name;
}
/**
* @brief Determines if a tensor operation should be offloaded to the CANN
* backend.
*
* This function checks if a given tensor operation should be offloaded to the
* CANN backend based on the operation type and the size of the tensor. It
* returns true if the second dimension (ne[1]) of the tensor is greater than or
* equal to the minimum batch size and the operation is not GGML_OP_GET_ROWS.
*
* @param backend Pointer to the CANN backend.
* @param op Pointer to the tensor operation to check.
* @return bool Returns true if the operation should be offloaded, otherwise
* false.
*/
struct ggml_backend_cann_device_context;
static bool ggml_backend_cann_offload_op(ggml_backend_dev_t dev, const ggml_tensor * op) {
ggml_backend_cann_device_context * dev_ctx = (ggml_backend_cann_device_context *)dev->context;
return op->ne[1] >= dev_ctx->op_offload_min_batch_size && op->op != GGML_OP_GET_ROWS;
}
/**
* @brief Records an event on the CANN backend stream.
*
@ -2670,6 +2649,26 @@ static ggml_backend_buffer_type_t ggml_backend_cann_device_get_host_buffer_type(
return ggml_backend_cann_host_buffer_type();
}
/**
* @brief Determines if a tensor operation should be offloaded to the CANN
* backend.
*
* This function checks if a given tensor operation should be offloaded to the
* CANN backend based on the operation type and the size of the tensor. It
* returns true if the second dimension (ne[1]) of the tensor is greater than or
* equal to the minimum batch size and the operation is not GGML_OP_GET_ROWS.
*
* @param backend Pointer to the CANN backend.
* @param op Pointer to the tensor operation to check.
* @return bool Returns true if the operation should be offloaded, otherwise
* false.
*/
static bool ggml_backend_cann_offload_op(ggml_backend_dev_t dev, const ggml_tensor * op) {
ggml_backend_cann_device_context * dev_ctx = (ggml_backend_cann_device_context *)dev->context;
return op->ne[1] >= dev_ctx->op_offload_min_batch_size && op->op != GGML_OP_GET_ROWS;
}
/**
* @brief Creates a new event for the CANN backend device.
*