zefie/nvfp4_gguf_quant
1
0 Fork 0
Code Actions Activity

include llama.cpp since there were some edits

Browse Source
This commit is contained in:
zefie committed 2026-11-01 16:32:35 -05:00
1 parent db7f28d087
commit 8c66990853
3918 files changed
+1640170 -1
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

llama.cpp/ggml/src/ggml-hexagon/htp/im2col-ops.h
+69
View File
@@ -0,0 +1,69 @@
#ifndef HTP_IM2COL_OPS_H
#define HTP_IM2COL_OPS_H
#include "hex-common.h"
#include "hex-fastdiv.h"
#include "htp-vtcm.h"
#include <stdint.h>
enum htp_im2col_kernel_type {
HTP_IM2COL_KERNEL_UNSUPPORTED = 0,
HTP_IM2COL_KERNEL_PATCHEMBED_DMA = 1,
HTP_IM2COL_KERNEL_BLOCKED_DMA = 2,
};
struct htp_im2col_vtcm_layout {
size_t total_bytes;
size_t off_src;
size_t off_dst;
uint32_t src_bytes_per_thread;
uint32_t dst_bytes_per_thread;
};
static inline void htp_im2col_vtcm_layout_build(
struct htp_im2col_vtcm_layout * L,
uint32_t src_row_bytes,
uint32_t dst_row_bytes,
uint32_t n_threads) {
L->src_bytes_per_thread = 2 * src_row_bytes + 256;
L->dst_bytes_per_thread = 2 * dst_row_bytes;
size_t off = 0;
VTCM_LAYOUT_ALLOC(off, off_src, (size_t) L->src_bytes_per_thread * n_threads);
VTCM_LAYOUT_ALLOC(off, off_dst, (size_t) L->dst_bytes_per_thread * n_threads);
L->total_bytes = off;
}
struct htp_im2col_kernel_params {
uint8_t kernel_type;
uint8_t n_threads;
uint32_t vtcm_size;
uint32_t src_row_bytes;
uint32_t dst_row_bytes;
uint32_t owb;
uint32_t wb;
uint32_t nbpr;
uint32_t N;
uint32_t IC;
uint32_t ID, IH, IW;
uint32_t KD, KH, KW;
uint32_t OD, OH, OW;
uint32_t patch_stride;
uint32_t total_rows;
uint32_t rows_per_chunk;
struct fastdiv_values div_OH;
struct fastdiv_values div_OD;
};
#if defined(__cplusplus)
static_assert(sizeof(struct htp_im2col_kernel_params) <= 128, "htp_im2col_kernel_params is too large for kernel_params blob");
#else
_Static_assert(sizeof(struct htp_im2col_kernel_params) <= 128, "htp_im2col_kernel_params is too large for kernel_params blob");
#endif
#endif // HTP_IM2COL_OPS_H