extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-08-23 07:01:24 +02:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: perf_logger improvements (llama/15246)

Browse Source 
* vulkan: perf_logger improvements

- Account for batch dimension in flops calculation.
- Fix how "_VEC" is detected for mat_mul_id.
- Fix "n" dimension for mat_mul_id (in case of broadcasting).
- Include a->type in name.

* use <=mul_mat_vec_max_cols rather than ==1
This commit is contained in:
Jeff Bolz
2025-08-14 08:38:10 -05:00
committed by Georgi Gerganov
parent 4a6cf896ad
commit 1d8b21caa0
1 changed files with 15 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@@ -1112,16 +1112,22 @@ class vk_perf_logger {
} }
if (node->op == GGML_OP_MUL_MAT || node->op == GGML_OP_MUL_MAT_ID) { if (node->op == GGML_OP_MUL_MAT || node->op == GGML_OP_MUL_MAT_ID) {
const uint64_t m = node->src[0]->ne[1]; const uint64_t m = node->src[0]->ne[1];
const uint64_t n = node->src[1]->ne[1]; const uint64_t n = node->ne[1];
const uint64_t k = node->src[1]->ne[0]; const uint64_t k = node->src[1]->ne[0];
const uint64_t batch = node->src[1]->ne[2] * node->src[1]->ne[3];
std::string name = ggml_op_name(node->op); std::string name = ggml_op_name(node->op);
if (n == 1) { if ((node->op == GGML_OP_MUL_MAT && n <= mul_mat_vec_max_cols) ||
name += "_VEC m=" + std::to_string(m) + " k=" + std::to_string(k); (node->op == GGML_OP_MUL_MAT_ID && node->src[2]->ne[1] == 1)) {
} else { name += "_VEC";
}
name += " ";
name += ggml_type_name(node->src[0]->type);
name += " m=" + std::to_string(m) + " n=" + std::to_string(n) + " k=" + std::to_string(k); name += " m=" + std::to_string(m) + " n=" + std::to_string(n) + " k=" + std::to_string(k);
if (batch > 1) {
name += " batch=" + std::to_string(batch);
} }
timings[name].push_back(time); timings[name].push_back(time);
flops[name].push_back(m * n * (k + (k - 1))); flops[name].push_back(m * n * (k + (k - 1)) * batch);
return; return;
} }
if (node->op == GGML_OP_CONV_2D) { if (node->op == GGML_OP_CONV_2D) {