379bdeb18c
* add GEMV implementation for matrix multiplication in hexagon * refactor: optimize GEMV implementation for matrix multiplication in hexagon * wip * refactor: enhance caching mechanism in GEMV implementation for matrix multiplication * wip * refactor: streamline caching logic in GEMV implementation for matrix multiplication * wip * wip * fix broadcase in flash_attn * format * refactor: optimize memory fetching in matrix multiplication implementations * wip * fix aligned gemv * rename * refactor: remove unused memory cache functions and initialize VTCM cache * wip * feat: add vector math functions for IEEE float and half float operations * feat: add vec_silu_f32 and vec_silu_f16 functions for SiLU activation * feat: implement GLU operation support in tensor processing * feat: add GLU operation support and related enhancements in tensor processing * wip * wip * wip * feat: add qhmath_hvx_div_vf functions for f32 vector operations * feat: add qhmath_hvx_div_vhf functions for f16 vector operations * fix: reorder parameters in vector operation functions for consistency * wip * feat: enhance vector operations with parameterized transformations and improved GLU implementations * wip * fix: increase default stack size and correct thread parameter indexing in thread pool * fix f16 div * fix f32 div * fix: update GLU vector operations to use explicit denominator calculation * wip * wip * Refactor cacheability check for matrix multiplication to handle multiple source tensors * Revert "fix: increase default stack size and correct thread parameter indexing in thread pool" This reverts commit 40e3f0974dbb04051aa30b397a9a171c6dd32678. * wip * fix comments * replace copy with memcpy |
||
|---|---|---|
| .. | ||
| cmake | ||
| include | ||
| src | ||
| .gitignore | ||
| CMakeLists.txt | ||