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

opt mulmat base on official doc (#25) 

https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md
This commit is contained in:
nullname 2025-02-25 19:46:48 +08:00 committed by GitHub
parent 84328ffbda
commit ff033e1e23
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 11 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
ggml/src/ggml-qnn/op-config-impl.cpp
View File

@ -385,36 +385,26 @@ bool ggml_qnn_matmul_op_config::create_mat_mul_nodes(QNNBackend device, Qnn_Grap
* [5, 4], * [5, 4],
* ]) * ])
* # Perform matrix multiplication * # Perform matrix multiplication
* result = torch.matmul(A, B.T) * C = torch.matmul(A, B.T)
* print(result.T) * print(C.T)
* ``` * ```
* Here, the B.T is the transpose of B. * Here, the B.T is the transpose of B.
* So C.T = A * B.T which is equivalent to C = B * A.T.
* See: https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md
* *
* So here we need to create graph like: * So here we need to create graph like:
* ```mermaid * ```mermaid
* graph TD; * graph TD;
* i1>ggml_tensor_in0] --src0--> mat_mul0; * i1>ggml_tensor_in0] --src1--> mat_mul0;
* i2>ggml_tensor_in1] --src1--> mat_mul0; * i2>ggml_tensor_in1] --src0.T--> mat_mul0;
* mat_mul0 --dst_trans--> transpose_out; * mat_mul0 --dst0--> o1>ggml_tensor_out];
* transpose1 --dst0--> o1>ggml_tensor_out];
* ``` * ```
*/ */
// create src0_trans tensor // create src0_trans tensor
auto src1 = tensor_inputs.back();
static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS does not match the expected value"); static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS does not match the expected value");
GGML_ASSERT(tensor_inputs.size() == 2);
qnn_dimension_array_t dimensions = get_transposed_dimensions(src1->get_dimensions(), rank); GGML_ASSERT(tensor_outputs.size() == 1);
// create dst_trans tensor
auto dst = tensor_outputs.front();
dimensions = get_transposed_dimensions(dst->get_dimensions(), rank);
auto dst_trans = std::make_shared<ggml_qnn_tensor>(ggml_qnn_tensor::INTERMEDIATE, "dst_trans", dimensions,
dst->get_data_type(), rank, device, graph_handle, _qnn_instance);
// create transpose_out
auto transpose_out = std::make_shared<ggml_qnn_single_op_config>(_name + "_trans1", QNN_OP_PACKAGE_NAME_QTI_AISW,
QNN_OP_TRANSPOSE, _qnn_instance);
// create mat_mul // create mat_mul
auto mat_mul = auto mat_mul =
@ -425,24 +415,12 @@ bool ggml_qnn_matmul_op_config::create_mat_mul_nodes(QNNBackend device, Qnn_Grap
scalar.bool8Value = 1; scalar.bool8Value = 1;
mat_mul->add_scalar_param(QNN_OP_MAT_MUL_PARAM_TRANSPOSE_IN1, scalar); mat_mul->add_scalar_param(QNN_OP_MAT_MUL_PARAM_TRANSPOSE_IN1, scalar);
// set transpose_out parameters
auto *params_data = reinterpret_cast<const uint8_t *>(kTransposeParamData[rank - 1].data());
const qnn_dimension_array_t param_dims = {(uint32_t)rank, 1, 1, 1};
transpose_out->add_tensor_param(QNN_OP_TRANSPOSE_PARAM_PERM, param_dims, 1, params_data, QNN_DATATYPE_UINT_32,
device, graph_handle);
// set tensor to mat_mul // set tensor to mat_mul
std::swap(tensor_inputs[0], tensor_inputs[1]);
mat_mul->set_input_tensors(tensor_inputs); mat_mul->set_input_tensors(tensor_inputs);
qnn_tensor_array_t tensors = {dst_trans}; mat_mul->set_output_tensors(tensor_outputs);
mat_mul->set_output_tensors(tensors);
// set tensor to transpose_out
tensors = {dst_trans};
transpose_out->set_input_tensors(tensors);
transpose_out->set_output_tensors(tensor_outputs);
_operations.push_back(mat_mul); _operations.push_back(mat_mul);
_operations.push_back(transpose_out);
return true; return true;
} }