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

hexagon: Flash Attention optimizations (dma, mpyacc, multi-row) and MatMul updates (#20118)

Browse Source 
* ggml-hexagon: enhance hvx_dot_f16_f16_aa_rx4 for improved performance by expanding vector handling and optimizing accumulation

# Conflicts:
#	ggml/src/ggml-hexagon/htp/flash-attn-ops.c

* ggml-hexagon: optimize hvx_dot_f16_f16_aa_rx4 and enhance hvx_vec_reduce_sum_f32x4 for improved performance and reduced complexity

* ggml-hexagon: add hvx_dot_f16_f16_aa_rx32 for enhanced vector processing in flash attention

# Conflicts:
#	ggml/src/ggml-hexagon/htp/flash-attn-ops.c

* optimize hvx_dot_f16_f16_aa_rx4 and hvx_dot_f16_f16_aa_rx32 by removing unused scale parameter and improving vector accumulation

# Conflicts:
#	ggml/src/ggml-hexagon/htp/flash-attn-ops.c

* ggml-hexagon: refactor hvx_dot_f16_f16_aa_rx4 for improved readability and return HVX_Vector for better integration

# Conflicts:
#	ggml/src/ggml-hexagon/htp/flash-attn-ops.c

* ggml-hexagon: initialize sums variable in hvx_dot_f16_f16_aa_rx32 for clarity

* ggml-hexagon: fix compiling error

* fix hvx_dot_f16_f16_aa_rx4 to handle leftover elements correctly using masking

* refactor hvx_dot_f16_f16_aa_rx4 to accept vector and leftover element counts as parameters for improved clarity and flexibility

* wip

* fa: instrumentation and dma reordering

* hex-fa: use block-size 64 to improve DMA pipelining

* hex-fa: optimize vec-dot for v79 and above

* hex-fa: use block size 64

* hex-fa: avoid scalar fp32->fp16 conversions

* hex-fa: simplify dot_f16 functions using optimized vec_mpyacc

* hex-fa: rewrite mad_f32_f16 using hvx_vec_mpyacc

* hex-mm: use mpyacc in matmul dot functions

---------

Co-authored-by: chraac <chraac@gmail.com>
This commit is contained in:
Max Krasnyansky
2026-03-04 21:55:29 -08:00
committed by GitHub
co-authored by chraac
parent 69fd345335
commit 7a99dc85e2
5 changed files with 293 additions and 221 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-hexagon/htp/hvx-base.h
+20 -1
View File
@@ -38,7 +38,7 @@ static inline HVX_Vector hvx_vec_splat_f32(float v) {
return Q6_V_vsplat_R(u.i);
}
static inline HVX_Vector hvx_vec_splat_f16(float v) {
static inline HVX_Vector hvx_vec_splat_f16(_Float16 v) {
union { __fp16 f; uint16_t i; } u = { .f = v };
return Q6_Vh_vsplat_R(u.i);
}
@@ -170,4 +170,23 @@ static inline HVX_Vector hvx_vec_i16_from_hf_rnd_sat(HVX_Vector vin) {
return Q6_Vh_vround_VwVw_sat(vsf_1, vsf_0);
}
#if __HVX_ARCH__ < 79
static inline HVX_VectorPair hvx_vec_mpyacc_f32_f16(HVX_VectorPair acc, HVX_Vector x, HVX_Vector y)
{
HVX_VectorPair m = Q6_Wqf32_vmpy_VhfVhf(x, y);
HVX_Vector a0 = Q6_Vsf_equals_Vqf32(Q6_Vqf32_vadd_Vqf32Vsf(Q6_V_lo_W(m), Q6_V_lo_W(acc)));
HVX_Vector a1 = Q6_Vsf_equals_Vqf32(Q6_Vqf32_vadd_Vqf32Vsf(Q6_V_hi_W(m), Q6_V_hi_W(acc)));
return Q6_W_vcombine_VV(a1, a0);
}
#else
static inline HVX_VectorPair hvx_vec_mpyacc_f32_f16(HVX_VectorPair acc, HVX_Vector x, HVX_Vector y)
{
return Q6_Wsf_vmpyacc_WsfVhfVhf(acc, x, y);
}
#endif
#endif /* HVX_BASE_H */