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: add more loggers

Browse Source 
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo
2025-07-21 21:09:21 +08:00
parent 1989fc9bf4
commit b9756b6dd4
1 changed files with 39 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

63
ggml/src/ggml-zdnn/ggml-zdnn.cpp
View File

@@ -14,6 +14,10 @@ struct zdnn_extra {
zdnn_ztensor ztensor; zdnn_ztensor ztensor;
struct zdnn_extra * extra; // for bias, etc. struct zdnn_extra * extra; // for bias, etc.
// Constructor
zdnn_extra()
: extra(nullptr) {}
}; };
struct ggml_backend_zdnn_context { struct ggml_backend_zdnn_context {
@@ -135,6 +139,19 @@ static void ggml_backend_zdnn_mul_mat(ggml_backend_zdnn_context * ctx, const ggm
GGML_LOG_INFO("%s: output->ztensor buffer_size=%zu is_transformed=%d rec_scale=%f\n", GGML_LOG_INFO("%s: output->ztensor buffer_size=%zu is_transformed=%d rec_scale=%f\n",
__func__, output_extra->ztensor.buffer_size, output_extra->ztensor.is_transformed, __func__, output_extra->ztensor.buffer_size, output_extra->ztensor.is_transformed,
output_extra->ztensor.rec_scale); output_extra->ztensor.rec_scale);
if (output_extra->extra) {
zdnn_extra * output_bias_extra = (zdnn_extra *)output_extra->extra;
GGML_LOG_INFO("%s: output_bias->pre_tfm_desc shape=[%ld, %ld, %ld, %ld] layout=%s type=%d\n",
__func__, output_bias_extra->pre_tfm_desc.dim1, output_bias_extra->pre_tfm_desc.dim2, output_bias_extra->pre_tfm_desc.dim3, output_bias_extra->pre_tfm_desc.dim4,
zdnn_layouts[output_bias_extra->pre_tfm_desc.layout], output_bias_extra->pre_tfm_desc.type);
GGML_LOG_INFO("%s: output_bias->tfm_desc shape=[%ld, %ld, %ld, %ld] layout=%s type=%d\n",
__func__, output_bias_extra->tfm_desc.dim1, output_bias_extra->tfm_desc.dim2, output_bias_extra->tfm_desc.dim3, output_bias_extra->tfm_desc.dim4,
zdnn_layouts[output_bias_extra->tfm_desc.layout], output_bias_extra->tfm_desc.type);
GGML_LOG_INFO("%s: output_bias->ztensor buffer_size=%zu is_transformed=%d rec_scale=%f\n",
__func__, output_bias_extra->ztensor.buffer_size, output_bias_extra->ztensor.is_transformed,
output_bias_extra->ztensor.rec_scale);
}
} }
zdnn_tensor_desc pre_tfm_desc_weights, tfm_desc_weights; zdnn_tensor_desc pre_tfm_desc_weights, tfm_desc_weights;
@@ -186,28 +203,18 @@ static void ggml_backend_zdnn_mul_mat(ggml_backend_zdnn_context * ctx, const ggm
ZDNN_CHECK(zdnn_transform_origtensor(&output_extra->ztensor, check_output_buffer)); ZDNN_CHECK(zdnn_transform_origtensor(&output_extra->ztensor, check_output_buffer));
ZDNN_CHECK(zdnn_transform_origtensor(&ztensor_output, output->data)); ZDNN_CHECK(zdnn_transform_origtensor(&ztensor_output, output->data));
// Compare the first 10 elements of the two buffers // Silently compare buffers and only log if there's a difference
GGML_LOG_INFO("%s: Comparing output buffers:\n", __func__);
GGML_LOG_INFO("Index | output->data | check_output_buffer\n");
GGML_LOG_INFO("------|--------------|--------------------\n");
for (int i = 0; i < 10 && i < output->ne[0] * output->ne[1]; i++) {
GGML_LOG_INFO("%5d | %12.6f | %18.6f\n",
i,
((float *)output->data)[i],
((float *)check_output_buffer)[i]);
}
GGML_LOG_INFO("... (snip) ...\n");
GGML_LOG_INFO("Index | output->data | check_output_buffer\n");
GGML_LOG_INFO("------|--------------|--------------------\n");
const int64_t num_elements = output->ne[0] * output->ne[1]; const int64_t num_elements = output->ne[0] * output->ne[1];
for (int64_t i = (num_elements > 10 ? num_elements - 10 : 0); i < num_elements; i++) { for (int64_t i = 0; i < num_elements; i++) {
GGML_LOG_INFO("%5lld | %12.6f | %18.6f\n", float output_val = ((float *)output->data)[i];
(long long) i, float check_val = ((float *)check_output_buffer)[i];
((float *)output->data)[i],
((float *)check_output_buffer)[i]);
}
if (output_val != check_val) {
GGML_LOG_INFO("%s: Difference found at index %lld: output->data = %12.6f, check_output_buffer = %12.6f\n",
__func__, (long long)i, output_val, check_val);
break;
}
}
std::raise(SIGINT); std::raise(SIGINT);
@@ -476,12 +483,20 @@ static void ggml_backend_zdnn_buffer_memset_tensor(ggml_backend_buffer_t buffer,
static void ggml_backend_zdnn_buffer_set_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor, const void * data, size_t offset, size_t size) { static void ggml_backend_zdnn_buffer_set_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
zdnn_extra * extra = (zdnn_extra *)tensor->extra; zdnn_extra * extra = (zdnn_extra *)tensor->extra;
ZDNN_CHECK(zdnn_transform_ztensor(&extra->ztensor, (void *)((char *)data + offset))); ZDNN_CHECK(zdnn_transform_ztensor(&extra->ztensor, (void *)((char *)data + offset)));
// Log operation type and extra->extra status
GGML_LOG_INFO("%s: tensor op = %s, has extra->extra = %s\n",
__func__,
ggml_op_name(tensor->op),
(extra->extra != nullptr) ? "true" : "false");
if (extra->extra != nullptr) { if (tensor->op == GGML_OP_MUL_MAT && extra->extra == nullptr) {
zdnn_extra * bias_extra = (zdnn_extra *)extra->extra; GGML_LOG_WARN("%s: MUL_MAT operation detected but extra->extra is nullptr\n", __func__);
void * bias_data = (void *)calloc(tensor->ne[0], ggml_element_size(tensor));
ZDNN_CHECK(zdnn_transform_ztensor(&bias_extra->ztensor, bias_data));
} }
// if (extra->extra != nullptr) {
// zdnn_extra * bias_extra = (zdnn_extra *)extra->extra;
// void * bias_data = (void *)calloc(tensor->ne[0], ggml_element_size(tensor));
// ZDNN_CHECK(zdnn_transform_ztensor(&bias_extra->ztensor, bias_data));
// }
memcpy((char *)tensor->data + offset, data, size); memcpy((char *)tensor->data + offset, data, size);