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

FEAT: do PERMUTE eagerly

This commit is contained in:
Yu, Zijun 2025-04-21 15:14:43 +08:00 committed by Mustafa Cavus
parent 8b408869ae
commit 6ed44a3dff
1 changed files with 5 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
ggml/src/ggml-openvino/ggml-decoder.cpp
View File

@ -43,12 +43,8 @@ void GgmlOvDecoder::set_input_output(ggml_tensor* node, std::map<std::string, gg
switch (node->op) { switch (node->op) {
case GGML_OP_CONT: { case GGML_OP_CONT: {
if (ggml_is_contiguous(node->src[0]) && ggml_is_contiguous(node) && // Currently only two cases, either the input comes from a VIEW which is subtensor or from a PERMUTE
(node->src[0]->ne[0] * node->src[0]->nb[0] == node->src[0]->nb[1])) { m_continuous = ggml_nelements(node->src[0]) == ggml_nelements(node->src[0]->view_src);
m_continuous = true;
} else {
m_continuous = false;
}
break; break;
} }
case GGML_OP_CPY: { case GGML_OP_CPY: {
@ -183,9 +179,9 @@ GgmlOvDecoder::GgmlOvDecoder(struct ggml_tensor * node, struct ggml_cgraph * cgr
// Init model input and output // Init model input and output
set_input_output(cur_node, m_inputs, m_outputs); set_input_output(cur_node, m_inputs, m_outputs);
} }
#ifdef GGML_OPENVINO_DEBUG if (getenv("GGML_OPENVINO_DEBUG")) {
ggml_graph_op_print(m_cgraph); ggml_graph_op_print(m_cgraph);
#endif }
} }
} }