Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

hexagon refactor all Ops to use local context struct (#19819)

Browse Source 
* hexagon: refactor set/get/sum-rows ops to use local context

* hexagon: refactor ROPE and Softmax Ops to use local context

Improves performance a bit by precomputing things and saving in the context.

* hexagon: refactor activation ops to use local context struct

* hexagon: refactor unary ops to use local context struct and DMA/VTCM

* hexagon: use aligned hvx_scale function

* hexagon: remove unused fields from op_context

* hexagon: rewrite ROPE to use DMA and VTCM scratchpad

* hex-rope: keep N rows in scratchpad (instead of just two)

* hex-rope: introduce rowidx cache

* hex-rope: remove unused fields

* hex-rope: rewrite dma prefetch logic to allow for multi-row fetch/compute

also removes the need for fastdiv.

* hex-rope: minor formatting

* hex-rope: use indices and unroll the loops

* hex-rope: more updates to cleanup rope-block handling

* hexagon: cleanup supported type/dims checks

* hexagon: all reduce funcs replicated across lanes

There is no need to explicitly replicate the first value.

* snapdragon: update adb and windows scripts to use ubatch-size 256

Updated Ops support handles larger ubatches.
This commit is contained in:
Max Krasnyansky
2026-02-23 16:32:14 -08:00
committed by GitHub
parent 5eb0ea32f0
commit 39fb81f875
15 changed files with 977 additions and 1000 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-hexagon/htp/hex-dma.h
+28 -2
View File
@@ -102,7 +102,7 @@ static inline bool dma_queue_push(dma_queue * q,
dmlink(q->tail, desc);
q->tail = desc;
// FARF(ERROR, "dma-push: i %u len %u dst %p src %p\n", q->push_idx, len, dst, src);
// FARF(ERROR, "dma-push: i %u width %u nrows %d dst %p src %p\n", q->push_idx, width, nrows, dptr.dst, dptr.src);
q->push_idx = (q->push_idx + 1) & q->idx_mask;
return true;
}
@@ -144,11 +144,37 @@ static inline dma_ptr dma_queue_pop(dma_queue * q) {
dptr = q->dptr[q->pop_idx];
// FARF(ERROR, "dma-pop: i %u dst %p\n", q->pop_idx, dst);
// FARF(ERROR, "dma-pop: i %u dst %p src %p\n", q->pop_idx, dptr.dst, dptr.src);
q->pop_idx = (q->pop_idx + 1) & q->idx_mask;
return dptr;
}
static inline dma_ptr dma_queue_pop_nowait(dma_queue * q) {
dma_ptr dptr = { NULL };
if (q->push_idx == q->pop_idx) {
return dptr;
}
dptr = q->dptr[q->pop_idx];
// FARF(ERROR, "dma-pop-nowait: i %u dst %p src %p\n", q->pop_idx, dptr.dst, dptr.src);
q->pop_idx = (q->pop_idx + 1) & q->idx_mask;
return dptr;
}
static inline bool dma_queue_empty(dma_queue * q) {
return q->push_idx == q->pop_idx;
}
static inline uint32_t dma_queue_depth(dma_queue * q) {
return (q->push_idx - q->pop_idx) & q->idx_mask;
}
static inline uint32_t dma_queue_capacity(dma_queue * q) {
return q->capacity;
}
#ifdef __cplusplus
} // extern "C"
#endif