6 Commits
Author SHA1 Message Date
Aparna M PandGeorgi Gerganov def36a5831 hexagon: add VISION RoPE support (llama/25216)
* hexagon: add VISION RoPE support

* hexagon: support RoPE on strided half-dim views for all modes

* hex-rope: decouple src0 DMA copy size from row stride

* hex-rope: support non-contiguous dst for RoPE

* hex-rope: fix dst spad pitch for non-contiguous dst
2026-07-10 13:06:42 +03:00
Aparna M PandGeorgi Gerganov aca63e7638 hexagon: add MROPE and IMROPE support in HTP rope op (llama/23317) 2026-05-25 12:26:07 +03:00
Aparna M PandGeorgi Gerganov 459ff0707b hexagon: enable support for NORM op (llama/23319) 2026-05-25 12:26:07 +03:00
Aparna M PandGeorgi Gerganov f2ce24fa5c hexagon: enable non-contiguous row tensor support for unary ops (llama/22574) 2026-05-02 15:02:42 +03:00
d6a417408c hexagon: add support for FILL op (llama/22198)
Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
2026-04-30 11:29:15 +03:00
9a40dd9365 hexagon: improve RMS_NORM and DIV accuracy (llama/21251)
* hexagon-rms_norm: fix RMS_NORM for non-aligned tensor sizes

Co-authored-by: Krishna Sridhar <srsr@qti.qualcomm.com>

* hexagon-div: perform DIV in fp16 domain for lower dsp archs

---------

Co-authored-by: Krishna Sridhar <srsr@qti.qualcomm.com>
2026-04-30 11:28:59 +03:00