mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2025-09-14 10:59:41 +00:00
Merge branch 'master' into concedo_experimental
# Conflicts: # .clang-tidy # .github/workflows/build.yml # CMakeLists.txt # Makefile # README.md # flake.nix # tests/test-quantize-perf.cpp
This commit is contained in:
Concedo
commit
ece4fda9c6
31 changed files with 1125 additions and 1009 deletions
40
ggml.c
40
ggml.c
|
|
@ -1,4 +1,3 @@
|
|||
#define _GNU_SOURCE // Defines CLOCK_MONOTONIC on Linux
|
||||
#define _CRT_SECURE_NO_DEPRECATE // Disables ridiculous "unsafe" warnigns on Windows
|
||||
|
||||
#include "ggml.h"
|
||||
|
|
@ -47,6 +46,10 @@
|
|||
// disable "possible loss of data" to avoid hundreds of casts
|
||||
// we should just be careful :)
|
||||
#pragma warning(disable: 4244 4267)
|
||||
|
||||
// disable POSIX deprecation warnigns
|
||||
// these functions are never going away, anyway
|
||||
#pragma warning(disable: 4996)
|
||||
#endif
|
||||
|
||||
#if defined(_WIN32)
|
||||
|
|
@ -103,6 +106,9 @@ typedef void * thread_ret_t;
|
|||
#include <sys/stat.h>
|
||||
#include <unistd.h>
|
||||
|
||||
#endif
|
||||
#ifdef GGML_USE_CPU_HBM
|
||||
#include <hbwmalloc.h>
|
||||
#endif
|
||||
|
||||
// __FMA__ and __F16C__ are not defined in MSVC, however they are implied with AVX2/AVX512
|
||||
|
|
@ -192,9 +198,15 @@ typedef void * thread_ret_t;
|
|||
#define GGML_ALIGNED_FREE(ptr) _aligned_free(ptr)
|
||||
#else
|
||||
inline static void * ggml_aligned_malloc(size_t size) {
|
||||
if (size == 0) {
|
||||
GGML_PRINT("WARNING: Behavior may be unexpected when allocating 0 bytes for ggml_aligned_malloc!\n");
|
||||
return NULL;
|
||||
}
|
||||
void * aligned_memory = NULL;
|
||||
#ifdef GGML_USE_METAL
|
||||
int result = posix_memalign(&aligned_memory, getpagesize(), size);
|
||||
#ifdef GGML_USE_CPU_HBM
|
||||
int result = hbw_posix_memalign(&aligned_memory, 16, size);
|
||||
#elif GGML_USE_METAL
|
||||
int result = posix_memalign(&aligned_memory, sysconf(_SC_PAGESIZE), size);
|
||||
#else
|
||||
int result = posix_memalign(&aligned_memory, GGML_MEM_ALIGN, size);
|
||||
#endif
|
||||
|
|
@ -215,8 +227,12 @@ inline static void * ggml_aligned_malloc(size_t size) {
|
|||
return aligned_memory;
|
||||
}
|
||||
#define GGML_ALIGNED_MALLOC(size) ggml_aligned_malloc(size)
|
||||
#ifdef GGML_USE_CPU_HBM
|
||||
#define GGML_ALIGNED_FREE(ptr) if(NULL != ptr) hbw_free(ptr)
|
||||
#else
|
||||
#define GGML_ALIGNED_FREE(ptr) free(ptr)
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#define UNUSED GGML_UNUSED
|
||||
#define SWAP(x, y, T) do { T SWAP = x; x = y; y = SWAP; } while (0)
|
||||
|
|
@ -294,12 +310,14 @@ typedef double ggml_float;
|
|||
#if defined(_MSC_VER) || defined(__MINGW32__)
|
||||
#include <intrin.h>
|
||||
#else
|
||||
#if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__) || defined(__SSE3__)
|
||||
#if !defined(__riscv)
|
||||
#include <immintrin.h>
|
||||
#endif
|
||||
#endif
|
||||
#endif
|
||||
#endif
|
||||
#endif
|
||||
|
||||
#ifdef __riscv_v_intrinsic
|
||||
#include <riscv_vector.h>
|
||||
|
|
@ -4567,6 +4585,11 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
|
|||
return NULL;
|
||||
}
|
||||
|
||||
// allow to call ggml_init with 0 size
|
||||
if (params.mem_size == 0) {
|
||||
params.mem_size = GGML_MEM_ALIGN;
|
||||
}
|
||||
|
||||
const size_t mem_size = params.mem_buffer ? params.mem_size : GGML_PAD(params.mem_size, GGML_MEM_ALIGN);
|
||||
|
||||
*ctx = (struct ggml_context) {
|
||||
|
|
@ -4769,7 +4792,7 @@ static struct ggml_tensor * ggml_new_tensor_impl(
|
|||
|
||||
size_t obj_alloc_size = 0;
|
||||
|
||||
if (view_src == NULL && ctx->no_alloc == false) {
|
||||
if (view_src == NULL && !ctx->no_alloc) {
|
||||
if (ctx->scratch.data != NULL) {
|
||||
// allocate tensor data in the scratch buffer
|
||||
if (ctx->scratch.offs + data_size > ctx->scratch.size) {
|
||||
|
|
@ -5470,7 +5493,7 @@ static struct ggml_tensor * ggml_mul_impl(
|
|||
}
|
||||
|
||||
if (inplace) {
|
||||
GGML_ASSERT(is_node == false);
|
||||
GGML_ASSERT(!is_node);
|
||||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
|
|
@ -5513,7 +5536,7 @@ static struct ggml_tensor * ggml_div_impl(
|
|||
}
|
||||
|
||||
if (inplace) {
|
||||
GGML_ASSERT(is_node == false);
|
||||
GGML_ASSERT(!is_node);
|
||||
}
|
||||
|
||||
struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
|
||||
|
|
@ -18854,7 +18877,6 @@ static enum ggml_opt_result linesearch_backtracking(
|
|||
// strong Wolfe condition (GGML_LINESEARCH_BACKTRACKING_STRONG_WOLFE)
|
||||
return count;
|
||||
}
|
||||
return count;
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -19957,7 +19979,7 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
|||
|
||||
struct ggml_tensor * data = NULL;
|
||||
|
||||
if (params.no_alloc == false) {
|
||||
if (!params.no_alloc) {
|
||||
data = ggml_new_tensor_1d(ctx_data, GGML_TYPE_I8, ctx->size);
|
||||
|
||||
ok = ok && data != NULL;
|
||||
|
|
@ -19998,7 +20020,7 @@ struct gguf_context * gguf_init_from_file(const char * fname, struct gguf_init_p
|
|||
}
|
||||
|
||||
// point the data member to the appropriate location in the binary blob using the tensor infos
|
||||
if (params.no_alloc == false) {
|
||||
if (!params.no_alloc) {
|
||||
//cur->data = (char *) data->data + ctx->infos[i].offset - ctx->offset; // offset from start of file
|
||||
cur->data = (char *) data->data + ctx->infos[i].offset; // offset from data
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue