]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
ggml-hexagon: create generalized functions for cpu side op (llama/17500)
authornullname <redacted>
Tue, 23 Dec 2025 07:13:24 +0000 (15:13 +0800)
committerGeorgi Gerganov <redacted>
Wed, 31 Dec 2025 10:39:43 +0000 (12:39 +0200)
commit049f610a10f2303fbf492ee50584c80f2ea53d6e
tree81b045b7cc3a4d99984f6e03c64b1e03e137bf1d
parent85750fc74a66cdd059ae63449867fe05fdfc621e
ggml-hexagon: create generalized functions for cpu side op (llama/17500)

* refactor: replace ggml_hexagon_mul_mat with template-based binary operation for improved flexibility

* refactor: replace ggml_hexagon_mul_mat_id with template-based binary operation for improved flexibility

* refactor: initialize buffer types and streamline dspqueue_buffers_init calls for clarity

* add comment

* refactor: remove redundant buffer checks in hexagon supported operations

* wip

* add missing include to fix weak symbol warning

* add ggml_hexagon_op_generic

* refactor: simplify tensor operation initialization and buffer management in hexagon implementation

* refactor: streamline hexagon operation initialization and buffer management

* refactor: update function signatures and streamline request handling in hexagon operations

* wip

* ggml-hexagon: clean up code formatting and improve unary operation handling

* wip

* rename

* fix: add support for permuted F16 tensors and enhance quantization checks in matrix operations

* refactor: replace ggml_hexagon_mul_mat with template-based binary operation for improved flexibility

refactor: replace ggml_hexagon_mul_mat_id with template-based binary operation for improved flexibility

refactor: initialize buffer types and streamline dspqueue_buffers_init calls for clarity

refactor: remove redundant buffer checks in hexagon supported operations

add missing include to fix weak symbol warning

add ggml_hexagon_op_generic

refactor: simplify tensor operation initialization and buffer management in hexagon implementation

refactor: streamline hexagon operation initialization and buffer management

refactor: update function signatures and streamline request handling in hexagon operations

ggml-hexagon: clean up code formatting and improve unary operation handling

fix: add support for permuted F16 tensors and enhance quantization checks in matrix operations

# Conflicts:
# ggml/src/ggml-hexagon/ggml-hexagon.cpp

* hexagon: fix merge conflicts

* hexagon: minor cleanup for buffer support checks

* hexagon: factor out op_desc and the overal op logging

* hexagon: further simplify and cleanup op dispatch logic

* snapdragon: update adb scripts to use llama-cli and llama-completion

* fix pipeline failure

---------

Co-authored-by: Max Krasnyansky <redacted>
src/ggml-hexagon/ggml-hexagon.cpp
src/ggml-hexagon/htp-utils.h
src/ggml-hexagon/op-desc.h [new file with mode: 0644]