CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-09 10:17:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml-zdnn: code cleanup

Browse Source 
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo
2025-07-30 16:57:14 +08:00
parent b7a77cf683
commit 1eb7c35e3a
1 changed files with 15 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
ggml/src/ggml-zdnn/ggml-zdnn-rewrite.cpp
View File

@@ -67,12 +67,20 @@ inline void ggml_zdnn_init_tensor(ggml_backend_zdnn_buffer * buffer, const ggml_
default: default:
{ {
// For 4D tensors, GGML uses NCHW layout. However, because zDNN
// automatically transforms everything to NHWC, we will use it
// directly to avoid the performance penalty changing the
// layout and reshaping the tensor.
zdnn_init_pre_transformed_desc( zdnn_init_pre_transformed_desc(
ZDNN_NHWC, ZDNN_NHWC,
ggml_zdnn_type_mapping(tensor->type), ggml_zdnn_type_mapping(tensor->type),
&buffer->pre_tfm_desc, &buffer->pre_tfm_desc,
tensor->ne[3], tensor->ne[2], tensor->ne[1], tensor->ne[0] tensor->ne[3], tensor->ne[2], tensor->ne[1], tensor->ne[0]
); );
// TODO: Consider adding a ggml check.
// TODO: If tensor = 4D, use ZDNN_NCHW by default.
// TODO: If tensor = 2D, use ZDNN_NHWC by default.
} break; } break;
} }
@@ -108,11 +116,8 @@ static void ggml_zdnn_mul_mat_op(ggml_backend_zdnn_context * ctx, const ggml_ten
ggml_backend_zdnn_buffer * inputs_extra = (ggml_backend_zdnn_buffer *)inputs->extra; ggml_backend_zdnn_buffer * inputs_extra = (ggml_backend_zdnn_buffer *)inputs->extra;
ggml_backend_zdnn_buffer * output_extra = (ggml_backend_zdnn_buffer *)output->extra; ggml_backend_zdnn_buffer * output_extra = (ggml_backend_zdnn_buffer *)output->extra;
zdnn_tensor_desc ptd_weights, td_weights;
zdnn_tensor_desc ptd_inputs, td_inputs;
zdnn_tensor_desc ptd_bias, td_bias; zdnn_tensor_desc ptd_bias, td_bias;
zdnn_tensor_desc ptd_output, td_output; zdnn_ztensor zt_bias;
zdnn_ztensor zt_weights, zt_inputs, zt_bias, zt_output;
const int64_t weights_rows = ne01; const int64_t weights_rows = ne01;
const int64_t weights_cols = ne00; const int64_t weights_cols = ne00;
@@ -130,7 +135,6 @@ static void ggml_zdnn_mul_mat_op(ggml_backend_zdnn_context * ctx, const ggml_ten
const int64_t output_dim[GGML_MAX_DIMS] = { 1, 1, output_cols, output_rows }; const int64_t output_dim[GGML_MAX_DIMS] = { 1, 1, output_cols, output_rows };
ggml_zdnn_create_tensor(ptd_bias, td_bias, zt_bias, output, bias_dim, ZDNN_1D); ggml_zdnn_create_tensor(ptd_bias, td_bias, zt_bias, output, bias_dim, ZDNN_1D);
// ggml_zdnn_create_tensor(ptd_output, td_output, zt_output, output, output_dim, ZDNN_2D);
void * bias_data = (void *)calloc(ne0, ggml_element_size(output)); void * bias_data = (void *)calloc(ne0, ggml_element_size(output));
if (weights_extra->ztensor.is_transformed == false) { if (weights_extra->ztensor.is_transformed == false) {
@@ -141,7 +145,6 @@ static void ggml_zdnn_mul_mat_op(ggml_backend_zdnn_context * ctx, const ggml_ten
ggml_zdnn_load_tensor(inputs_extra->ztensor, inputs->data); ggml_zdnn_load_tensor(inputs_extra->ztensor, inputs->data);
} }
ggml_zdnn_load_tensor(zt_bias, bias_data); ggml_zdnn_load_tensor(zt_bias, bias_data);
// ggml_zdnn_load_tensor(output_extra->ztensor, output->data);
// GGML_LOG_INFO("%s: tensor '%s' tensor dimensions: [%ld, %ld, %ld, %ld] pre_tfm_desc dimensions: [%ld, %ld, %ld, %ld]\n", // GGML_LOG_INFO("%s: tensor '%s' tensor dimensions: [%ld, %ld, %ld, %ld] pre_tfm_desc dimensions: [%ld, %ld, %ld, %ld]\n",
// __func__, weights_extra->name, // __func__, weights_extra->name,
@@ -159,21 +162,17 @@ static void ggml_zdnn_mul_mat_op(ggml_backend_zdnn_context * ctx, const ggml_ten
// inputs_extra->pre_tfm_desc.dim3, // inputs_extra->pre_tfm_desc.dim3,
// inputs_extra->pre_tfm_desc.dim4); // inputs_extra->pre_tfm_desc.dim4);
// GGML_ASSERT(weights_extra->pre_tfm_desc.layout == ZDNN_2D && "weights_extra->pre_tfm_desc.layout must be ZDNN_2D");
// GGML_ASSERT(inputs_extra->pre_tfm_desc.layout == ZDNN_2D && "inputs_extra->pre_tfm_desc.layout must be ZDNN_2D");
GGML_ASSERT(weights_extra->pre_tfm_desc.dim1 == weights->ne[0] && "weights_extra->pre_tfm_desc.dim1 must match weights->ne[0]"); GGML_ASSERT(weights_extra->pre_tfm_desc.dim1 == weights->ne[0] && "weights_extra->pre_tfm_desc.dim1 must match weights->ne[0]");
GGML_ASSERT(weights_extra->pre_tfm_desc.dim2 == weights->ne[1] && "weights_extra->pre_tfm_desc.dim2 must match weights->ne[1]"); GGML_ASSERT(weights_extra->pre_tfm_desc.dim2 == weights->ne[1] && "weights_extra->pre_tfm_desc.dim2 must match weights->ne[1]");
GGML_ASSERT(inputs_extra->pre_tfm_desc.dim1 == inputs->ne[0] && "inputs_extra->pre_tfm_desc.dim1 must match inputs->ne[0]"); GGML_ASSERT(inputs_extra->pre_tfm_desc.dim1 == inputs->ne[0] && "inputs_extra->pre_tfm_desc.dim1 must match inputs->ne[0]");
GGML_ASSERT(inputs_extra->pre_tfm_desc.dim2 == inputs->ne[1] && "inputs_extra->pre_tfm_desc.dim2 must match inputs->ne[1]"); GGML_ASSERT(inputs_extra->pre_tfm_desc.dim2 == inputs->ne[1] && "inputs_extra->pre_tfm_desc.dim2 must match inputs->ne[1]");
std::raise(SIGINT);
ZDNN_CHECK(zdnn_matmul_transpose_op(&inputs_extra->ztensor, &weights_extra->ztensor, &zt_bias, ZDNN_CHECK(zdnn_matmul_transpose_op(&inputs_extra->ztensor, &weights_extra->ztensor, &zt_bias,
false, true, MATMUL_OP_ADDITION, &output_extra->ztensor)); false, true, MATMUL_OP_ADDITION, &output_extra->ztensor));
// TODO: Remove in the future as we are currently DLF16 -> FP32 then in the next op, FP32 -> DLF16 again. Inefficient.
ZDNN_CHECK(zdnn_transform_origtensor(&output_extra->ztensor, output->data)); ZDNN_CHECK(zdnn_transform_origtensor(&output_extra->ztensor, output->data));
ZDNN_CHECK(zdnn_free_ztensor_buffer(&zt_bias)); ZDNN_CHECK(zdnn_free_ztensor_buffer(&zt_bias));
free(bias_data); free(bias_data);
} }