CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-08 10:07:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml-zdnn: move everything back to local declaration

Browse Source 
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo
2025-07-28 18:08:47 +08:00
parent eefa943b0a
commit 6f42570194
1 changed files with 23 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
ggml/src/ggml-zdnn/ggml-zdnn-rewrite.cpp
View File

@@ -47,12 +47,16 @@ static bool ggml_zdnn_op_mul_mat(struct ggml_backend_zdnn_context * ctx, const g
const int64_t output_rows = ne1;
const int64_t output_cols = ne0;
zdnn_tensor_desc pre_tfm_desc_inputs, tfm_desc_inputs;
zdnn_tensor_desc pre_tfm_desc_weights, tfm_desc_weights;
zdnn_tensor_desc pre_tfm_desc_bias, tfm_desc_bias;
zdnn_ztensor ztensor_weights, ztensor_bias;
zdnn_tensor_desc pre_tfm_desc_output, tfm_desc_output;
zdnn_ztensor ztensor_inputs, ztensor_weights, ztensor_bias, ztensor_output;
const int64_t inputs_dim [GGML_MAX_DIMS] = { 1, 1, inputs_cols, inputs_rows };
const int64_t weights_dim[GGML_MAX_DIMS] = { 1, 1, weights_cols, weights_rows };
const int64_t bias_dim [GGML_MAX_DIMS] = { 1, 1, 1, output_cols };
const int64_t output_dim[GGML_MAX_DIMS] = { 1, 1, output_cols, output_rows };
// have to do this because weights apparently do not go through set_tensor
zdnn_init_pre_transformed_desc(
@@ -66,16 +70,15 @@ static bool ggml_zdnn_op_mul_mat(struct ggml_backend_zdnn_context * ctx, const g
ZDNN_CHECK(zdnn_transform_ztensor(&ztensor_weights, weights->data));
// have to do this here because although it was transformed, the shape is wrong
if (&inputs_extra->ztensor.is_transformed) zdnn_reset_ztensor(&inputs_extra->ztensor);
zdnn_init_pre_transformed_desc(
ZDNN_2D,
FP32,
&inputs_extra->pre_tfm_desc,
inputs->ne[1], inputs->ne[0]
&pre_tfm_desc_inputs,
inputs_dim[3], inputs_dim[2], inputs_dim[1], inputs_dim[0]
);
ZDNN_CHECK(zdnn_generate_transformed_desc(&inputs_extra->pre_tfm_desc, &inputs_extra->tfm_desc));
ZDNN_CHECK(zdnn_init_ztensor_with_malloc(&inputs_extra->pre_tfm_desc, &inputs_extra->tfm_desc, &inputs_extra->ztensor));
ZDNN_CHECK(zdnn_transform_ztensor(&inputs_extra->ztensor, inputs->data));
ZDNN_CHECK(zdnn_generate_transformed_desc(&pre_tfm_desc_inputs, &tfm_desc_inputs));
ZDNN_CHECK(zdnn_init_ztensor_with_malloc(&pre_tfm_desc_inputs, &tfm_desc_inputs, &ztensor_inputs));
ZDNN_CHECK(zdnn_transform_ztensor(&ztensor_inputs, inputs->data));
// have to transform the bias ztensor here because only GGML_OP_NONE goes through set_tensor
zdnn_init_pre_transformed_desc(
@@ -88,10 +91,20 @@ static bool ggml_zdnn_op_mul_mat(struct ggml_backend_zdnn_context * ctx, const g
ZDNN_CHECK(zdnn_init_ztensor_with_malloc(&pre_tfm_desc_bias, &tfm_desc_bias, &ztensor_bias));
ZDNN_CHECK(zdnn_transform_ztensor(&ztensor_bias, bias_extra->data));
zdnn_init_pre_transformed_desc(
ZDNN_2D,
FP32,
&pre_tfm_desc_output,
output_dim[3], output_dim[2], output_dim[1], output_dim[0]
);
ZDNN_CHECK(zdnn_generate_transformed_desc(&pre_tfm_desc_output, &tfm_desc_output));
ZDNN_CHECK(zdnn_init_ztensor_with_malloc(&pre_tfm_desc_output, &tfm_desc_output, &ztensor_output));
ZDNN_CHECK(zdnn_transform_ztensor(&ztensor_output, output->data));
std::raise(SIGINT);
ZDNN_CHECK(zdnn_matmul_transpose_op(&inputs_extra->ztensor, &ztensor_weights, &ztensor_bias,
false, true, MATMUL_OP_ADDITION, &output_extra->ztensor));
ZDNN_CHECK(zdnn_transform_origtensor(&output_extra->ztensor, output->data));
ZDNN_CHECK(zdnn_matmul_transpose_op(&ztensor_inputs, &ztensor_weights, &ztensor_bias,
false, true, MATMUL_OP_ADDITION, &ztensor_output));
ZDNN_CHECK(zdnn_transform_origtensor(&ztensor_output, output->data));
}
static bool ggml_backend_zdnn_compute_forward(struct ggml_backend_zdnn_context * ctx, struct ggml_tensor * dst) {