mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-10-29 22:20:15 +01:00
[SYCL] remove global variables (#7710)
* separate DPCT helpers outside * replace global variables with context * remove useless extra * update mul_mat condition * remove duplicate buft initialization * remove duplicate extra and global work group size * remove useless backend check * remove duplicated extras * use macro for group_size and remove cuda-related
This commit is contained in:
parent
f8ec8877b7
commit
7b2f4a7d19
@ -684,7 +684,8 @@ if (LLAMA_SYCL)
|
||||
endif()
|
||||
|
||||
set(GGML_HEADERS_SYCL ggml-sycl.h)
|
||||
set(GGML_SOURCES_SYCL ggml-sycl.cpp)
|
||||
file(GLOB GGML_SOURCES_SYCL "ggml-sycl/*.cpp")
|
||||
list(APPEND GGML_SOURCES_SYCL "ggml-sycl.cpp")
|
||||
|
||||
if (WIN32)
|
||||
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} -fsycl sycl7 OpenCL mkl_sycl_blas_dll.lib mkl_intel_ilp64_dll.lib mkl_sequential_dll.lib mkl_core_dll.lib)
|
||||
|
5532
ggml-sycl.cpp
5532
ggml-sycl.cpp
File diff suppressed because it is too large
Load Diff
11
ggml-sycl.h
11
ggml-sycl.h
@ -8,14 +8,12 @@
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
#include "ggml-sycl/presets.hpp"
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
#define GGML_SYCL_MAX_DEVICES 48
|
||||
#define GGML_SYCL_NAME "SYCL"
|
||||
|
||||
// backend API
|
||||
GGML_API ggml_backend_t ggml_backend_sycl_init(int device);
|
||||
|
||||
@ -33,13 +31,6 @@ GGML_API GGML_CALL void ggml_sycl_get_gpu_list(int *id_list, int max_len);
|
||||
GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
|
||||
GGML_API GGML_CALL int ggml_backend_sycl_get_device_count();
|
||||
GGML_API GGML_CALL void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
|
||||
GGML_API GGML_CALL int ggml_backend_sycl_get_device_index(int device_id);
|
||||
|
||||
// TODO: these are temporary
|
||||
// ref: https://github.com/ggerganov/llama.cpp/pull/6022#issuecomment-1992615670
|
||||
GGML_API GGML_CALL int ggml_backend_sycl_get_device_id(int device_index);
|
||||
GGML_API GGML_CALL void ggml_backend_sycl_set_single_device_mode(int main_gpu_id);
|
||||
GGML_API GGML_CALL void ggml_backend_sycl_set_mul_device_mode();
|
||||
|
||||
// SYCL doesn't support registering host memory, keep here for reference
|
||||
// GGML_API GGML_CALL bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size);
|
||||
|
18
ggml-sycl/backend.hpp
Normal file
18
ggml-sycl/backend.hpp
Normal file
@ -0,0 +1,18 @@
|
||||
//
|
||||
// MIT license
|
||||
// Copyright (C) 2024 Intel Corporation
|
||||
// SPDX-License-Identifier: MIT
|
||||
//
|
||||
|
||||
//
|
||||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
||||
// See https://llvm.org/LICENSE.txt for license information.
|
||||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
||||
//
|
||||
|
||||
#ifndef GGML_SYCL_BACKEND_HPP
|
||||
#define GGML_SYCL_BACKEND_HPP
|
||||
|
||||
#include "common.hpp"
|
||||
|
||||
#endif // GGML_SYCL_BACKEND_HPP
|
53
ggml-sycl/common.cpp
Normal file
53
ggml-sycl/common.cpp
Normal file
@ -0,0 +1,53 @@
|
||||
//
|
||||
// MIT license
|
||||
// Copyright (C) 2024 Intel Corporation
|
||||
// SPDX-License-Identifier: MIT
|
||||
//
|
||||
|
||||
//
|
||||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
||||
// See https://llvm.org/LICENSE.txt for license information.
|
||||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
||||
//
|
||||
|
||||
#include "common.hpp"
|
||||
|
||||
int get_current_device_id() {
|
||||
return dpct::dev_mgr::instance().current_device_id();
|
||||
}
|
||||
|
||||
void* ggml_sycl_host_malloc(size_t size) try {
|
||||
if (getenv("GGML_SYCL_NO_PINNED") != nullptr) {
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
void* ptr = nullptr;
|
||||
// allow to use dpct::get_in_order_queue() for host malloc
|
||||
dpct::err0 err = CHECK_TRY_ERROR(
|
||||
ptr = (void*)sycl::malloc_host(size, dpct::get_in_order_queue()));
|
||||
|
||||
if (err != 0) {
|
||||
// clear the error
|
||||
fprintf(
|
||||
stderr,
|
||||
"WARNING: failed to allocate %.2f MB of pinned memory: %s\n",
|
||||
size / 1024.0 / 1024.0,
|
||||
"syclGetErrorString is not supported");
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
return ptr;
|
||||
} catch (sycl::exception const& exc) {
|
||||
std::cerr << exc.what() << "Exception caught at file:" << __FILE__
|
||||
<< ", line:" << __LINE__ << std::endl;
|
||||
std::exit(1);
|
||||
}
|
||||
|
||||
void ggml_sycl_host_free(void* ptr) try {
|
||||
// allow to use dpct::get_in_order_queue() for host malloc
|
||||
SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(ptr, dpct::get_in_order_queue())));
|
||||
} catch (sycl::exception const& exc) {
|
||||
std::cerr << exc.what() << "Exception caught at file:" << __FILE__
|
||||
<< ", line:" << __LINE__ << std::endl;
|
||||
std::exit(1);
|
||||
}
|
298
ggml-sycl/common.hpp
Normal file
298
ggml-sycl/common.hpp
Normal file
@ -0,0 +1,298 @@
|
||||
//
|
||||
// MIT license
|
||||
// Copyright (C) 2024 Intel Corporation
|
||||
// SPDX-License-Identifier: MIT
|
||||
//
|
||||
|
||||
//
|
||||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
||||
// See https://llvm.org/LICENSE.txt for license information.
|
||||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
||||
//
|
||||
|
||||
#ifndef GGML_SYCL_COMMON_HPP
|
||||
#define GGML_SYCL_COMMON_HPP
|
||||
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
|
||||
#include "dpct/helper.hpp"
|
||||
#include "presets.hpp"
|
||||
|
||||
#define GGML_COMMON_DECL_SYCL
|
||||
#define GGML_COMMON_IMPL_SYCL
|
||||
#include "ggml-common.h"
|
||||
|
||||
void* ggml_sycl_host_malloc(size_t size);
|
||||
void ggml_sycl_host_free(void* ptr);
|
||||
|
||||
static int g_ggml_sycl_debug = 0;
|
||||
#define GGML_SYCL_DEBUG(...) \
|
||||
do { \
|
||||
if (g_ggml_sycl_debug) \
|
||||
fprintf(stderr, __VA_ARGS__); \
|
||||
} while (0)
|
||||
|
||||
#define CHECK_TRY_ERROR(expr) \
|
||||
[&]() { \
|
||||
try { \
|
||||
expr; \
|
||||
return dpct::success; \
|
||||
} catch (std::exception const& e) { \
|
||||
std::cerr << e.what() << "\nException caught at file:" << __FILE__ \
|
||||
<< ", line:" << __LINE__ << ", func:" << __func__ \
|
||||
<< std::endl; \
|
||||
return dpct::default_error; \
|
||||
} \
|
||||
}()
|
||||
|
||||
// #define DEBUG_SYCL_MALLOC
|
||||
|
||||
static int g_work_group_size = 0;
|
||||
// typedef sycl::half ggml_fp16_t;
|
||||
|
||||
#define __SYCL_ARCH__ DPCT_COMPATIBILITY_TEMP
|
||||
#define VER_4VEC 610 // todo for hardward optimize.
|
||||
#define VER_GEN9 700 // todo for hardward optimize.
|
||||
#define VER_GEN12 1000000 // todo for hardward optimize.
|
||||
#define VER_GEN13 (VER_GEN12 + 1030) // todo for hardward optimize.
|
||||
|
||||
#define GGML_SYCL_MAX_NODES 8192 // TODO: adapt to hardwares
|
||||
|
||||
// define for XMX in Intel GPU
|
||||
// TODO: currently, it's not used for XMX really.
|
||||
#if !defined(GGML_SYCL_FORCE_MMQ)
|
||||
#define SYCL_USE_XMX
|
||||
#endif
|
||||
|
||||
// max batch size to use MMQ kernels when tensor cores are available
|
||||
#define MMQ_MAX_BATCH_SIZE 32
|
||||
|
||||
#if defined(_MSC_VER)
|
||||
#pragma warning(disable : 4244 4267) // possible loss of data
|
||||
#endif
|
||||
|
||||
// dmmv = dequantize_mul_mat_vec
|
||||
#ifndef GGML_SYCL_DMMV_X
|
||||
#define GGML_SYCL_DMMV_X 32
|
||||
#endif
|
||||
#ifndef GGML_SYCL_MMV_Y
|
||||
#define GGML_SYCL_MMV_Y 1
|
||||
#endif
|
||||
|
||||
typedef sycl::queue *queue_ptr;
|
||||
|
||||
enum ggml_sycl_backend_gpu_mode {
|
||||
SYCL_UNSET_GPU_MODE = -1,
|
||||
SYCL_SINGLE_GPU_MODE = 0,
|
||||
SYCL_MUL_GPU_MODE
|
||||
};
|
||||
|
||||
static_assert(sizeof(sycl::half) == sizeof(ggml_fp16_t), "wrong fp16 size");
|
||||
|
||||
static void crash() {
|
||||
int* ptr = NULL;
|
||||
*ptr = 0;
|
||||
}
|
||||
|
||||
[[noreturn]] static void ggml_sycl_error(
|
||||
const char* stmt,
|
||||
const char* func,
|
||||
const char* file,
|
||||
const int line,
|
||||
const char* msg) {
|
||||
fprintf(stderr, "SYCL error: %s: %s\n", stmt, msg);
|
||||
fprintf(stderr, " in function %s at %s:%d\n", func, file, line);
|
||||
GGML_ASSERT(!"SYCL error");
|
||||
}
|
||||
|
||||
#define SYCL_CHECK(err) \
|
||||
do { \
|
||||
auto err_ = (err); \
|
||||
if (err_ != 0) \
|
||||
ggml_sycl_error( \
|
||||
#err, \
|
||||
__func__, \
|
||||
__FILE__, \
|
||||
__LINE__, \
|
||||
"Meet error in this line code!"); \
|
||||
} while (0)
|
||||
|
||||
#if DPCT_COMPAT_RT_VERSION >= 11100
|
||||
#define GGML_SYCL_ASSUME(x) __builtin_assume(x)
|
||||
#else
|
||||
#define GGML_SYCL_ASSUME(x)
|
||||
#endif // DPCT_COMPAT_RT_VERSION >= 11100
|
||||
|
||||
#ifdef GGML_SYCL_F16
|
||||
typedef sycl::half dfloat; // dequantize float
|
||||
typedef sycl::half2 dfloat2;
|
||||
#else
|
||||
typedef float dfloat; // dequantize float
|
||||
typedef sycl::float2 dfloat2;
|
||||
#endif // GGML_SYCL_F16
|
||||
|
||||
#define MMVQ_MAX_BATCH_SIZE 8
|
||||
|
||||
static const int8_t kvalues_iq4nl[16]={-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};
|
||||
|
||||
static int g_all_sycl_device_count = -1;
|
||||
static bool g_ggml_backend_sycl_buffer_type_initialized = false;
|
||||
|
||||
static ggml_sycl_backend_gpu_mode g_ggml_sycl_backend_gpu_mode =
|
||||
SYCL_UNSET_GPU_MODE;
|
||||
|
||||
static void* g_scratch_buffer = nullptr;
|
||||
static size_t g_scratch_size = 0; // disabled by default
|
||||
static size_t g_scratch_offset = 0;
|
||||
|
||||
[[noreturn]] static inline void bad_arch(const sycl::stream& stream_ct1) {
|
||||
stream_ct1 << "ERROR: ggml-sycl was compiled without support for the "
|
||||
"current GPU architecture.\n";
|
||||
// __trap();
|
||||
std::exit(1);
|
||||
|
||||
(void)bad_arch; // suppress unused function warning
|
||||
}
|
||||
|
||||
int get_current_device_id();
|
||||
|
||||
inline dpct::err0 ggml_sycl_set_device(const int device) try {
|
||||
|
||||
int current_device_id;
|
||||
SYCL_CHECK(CHECK_TRY_ERROR(current_device_id = get_current_device_id()));
|
||||
|
||||
// GGML_SYCL_DEBUG("ggml_sycl_set_device device_id=%d,
|
||||
// current_device_id=%d\n", device, current_device);
|
||||
if (device == current_device_id) {
|
||||
return 0;
|
||||
}
|
||||
|
||||
return CHECK_TRY_ERROR(dpct::select_device(device));
|
||||
} catch (sycl::exception const& exc) {
|
||||
std::cerr << exc.what() << "Exception caught at file:" << __FILE__
|
||||
<< ", line:" << __LINE__ << std::endl;
|
||||
crash();
|
||||
std::exit(1);
|
||||
}
|
||||
|
||||
//////////////////////
|
||||
|
||||
struct ggml_sycl_device_info {
|
||||
int device_count;
|
||||
|
||||
struct sycl_device_info {
|
||||
int cc; // compute capability
|
||||
// int nsm; // number of streaming multiprocessors
|
||||
// size_t smpb; // max. shared memory per block
|
||||
bool vmm; // virtual memory support
|
||||
size_t total_vram;
|
||||
};
|
||||
|
||||
sycl_device_info devices[GGML_SYCL_MAX_DEVICES] = {};
|
||||
|
||||
std::array<float, GGML_SYCL_MAX_DEVICES> default_tensor_split = {};
|
||||
};
|
||||
|
||||
const ggml_sycl_device_info & ggml_sycl_info();
|
||||
|
||||
struct ggml_sycl_pool {
|
||||
virtual ~ggml_sycl_pool() = default;
|
||||
|
||||
virtual void * alloc(size_t size, size_t * actual_size) = 0;
|
||||
virtual void free(void * ptr, size_t size) = 0;
|
||||
};
|
||||
|
||||
template<typename T>
|
||||
struct ggml_sycl_pool_alloc {
|
||||
ggml_sycl_pool * pool = nullptr;
|
||||
T * ptr = nullptr;
|
||||
size_t actual_size = 0;
|
||||
|
||||
explicit ggml_sycl_pool_alloc(ggml_sycl_pool & pool) : pool(&pool) {
|
||||
}
|
||||
|
||||
ggml_sycl_pool_alloc(ggml_sycl_pool & pool, size_t size) : pool(&pool) {
|
||||
alloc(size);
|
||||
}
|
||||
|
||||
~ggml_sycl_pool_alloc() {
|
||||
if (ptr != nullptr) {
|
||||
pool->free(ptr, actual_size);
|
||||
}
|
||||
}
|
||||
|
||||
// size is in number of elements
|
||||
T * alloc(size_t size) {
|
||||
GGML_ASSERT(pool != nullptr);
|
||||
GGML_ASSERT(ptr == nullptr);
|
||||
ptr = (T *) pool->alloc(size * sizeof(T), &this->actual_size);
|
||||
return ptr;
|
||||
}
|
||||
|
||||
T * alloc(ggml_sycl_pool & pool, size_t size) {
|
||||
this->pool = &pool;
|
||||
return alloc(size);
|
||||
}
|
||||
|
||||
T * get() {
|
||||
return ptr;
|
||||
}
|
||||
|
||||
ggml_sycl_pool_alloc() = default;
|
||||
ggml_sycl_pool_alloc(const ggml_sycl_pool_alloc &) = delete;
|
||||
ggml_sycl_pool_alloc(ggml_sycl_pool_alloc &&) = delete;
|
||||
ggml_sycl_pool_alloc& operator=(const ggml_sycl_pool_alloc &) = delete;
|
||||
ggml_sycl_pool_alloc& operator=(ggml_sycl_pool_alloc &&) = delete;
|
||||
};
|
||||
|
||||
// backend interface
|
||||
|
||||
struct ggml_tensor_extra_gpu {
|
||||
void* data_device[GGML_SYCL_MAX_DEVICES]; // 1 pointer for each device for split
|
||||
// tensors
|
||||
dpct::event_ptr events[GGML_SYCL_MAX_DEVICES]
|
||||
[GGML_SYCL_MAX_STREAMS]; // events for synchronizing multiple GPUs
|
||||
};
|
||||
|
||||
struct ggml_backend_sycl_context {
|
||||
int device;
|
||||
std::string name;
|
||||
|
||||
queue_ptr qptrs[GGML_SYCL_MAX_DEVICES][GGML_SYCL_MAX_STREAMS] = { { nullptr } };
|
||||
|
||||
explicit ggml_backend_sycl_context(int device) :
|
||||
device(device),
|
||||
name(GGML_SYCL_NAME + std::to_string(device)) {
|
||||
}
|
||||
|
||||
queue_ptr stream(int device, int stream) {
|
||||
if (qptrs[device][stream] == nullptr) {
|
||||
qptrs[device][stream] = &(dpct::get_current_device().default_queue());
|
||||
}
|
||||
return qptrs[device][stream];
|
||||
}
|
||||
|
||||
queue_ptr stream() {
|
||||
return stream(device, 0);
|
||||
}
|
||||
|
||||
// pool
|
||||
std::unique_ptr<ggml_sycl_pool> pools[GGML_SYCL_MAX_DEVICES];
|
||||
|
||||
static std::unique_ptr<ggml_sycl_pool> new_pool_for_device(queue_ptr qptr, int device);
|
||||
|
||||
ggml_sycl_pool & pool(int device) {
|
||||
if (pools[device] == nullptr) {
|
||||
pools[device] = new_pool_for_device(stream(device,0), device);
|
||||
}
|
||||
return *pools[device];
|
||||
}
|
||||
|
||||
ggml_sycl_pool & pool() {
|
||||
return pool(device);
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
#endif // GGML_SYCL_COMMON_HPP
|
2980
ggml-sycl/dpct/helper.hpp
Normal file
2980
ggml-sycl/dpct/helper.hpp
Normal file
File diff suppressed because it is too large
Load Diff
69
ggml-sycl/presets.hpp
Normal file
69
ggml-sycl/presets.hpp
Normal file
@ -0,0 +1,69 @@
|
||||
//
|
||||
// MIT license
|
||||
// Copyright (C) 2024 Intel Corporation
|
||||
// SPDX-License-Identifier: MIT
|
||||
//
|
||||
|
||||
//
|
||||
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
||||
// See https://llvm.org/LICENSE.txt for license information.
|
||||
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
||||
//
|
||||
|
||||
#ifndef GGML_SYCL_PRESETS_HPP
|
||||
#define GGML_SYCL_PRESETS_HPP
|
||||
|
||||
#define GGML_SYCL_MAX_STREAMS 8
|
||||
#define GGML_SYCL_MAX_BUFFERS 256
|
||||
#define GGML_SYCL_MAX_DEVICES 48
|
||||
#define GGML_SYCL_NAME "SYCL"
|
||||
|
||||
// FIXME: 1024 from cuda
|
||||
#define GROUP_SIZE 1024
|
||||
#define WARP_SIZE 32
|
||||
#define MATRIX_ROW_PADDING 512 // last row of quant. matrices is a multiple of this to avoid out-of-bounds memory accesses
|
||||
|
||||
#define SYCL_GELU_BLOCK_SIZE 256
|
||||
#define SYCL_SILU_BLOCK_SIZE 256
|
||||
#define SYCL_TANH_BLOCK_SIZE 256
|
||||
#define SYCL_RELU_BLOCK_SIZE 256
|
||||
#define SYCL_HARDSIGMOID_BLOCK_SIZE 256
|
||||
#define SYCL_HARDSWISH_BLOCK_SIZE 256
|
||||
#define SYCL_SQR_BLOCK_SIZE 256
|
||||
#define SYCL_CPY_BLOCK_SIZE 32
|
||||
#define SYCL_SCALE_BLOCK_SIZE 256
|
||||
#define SYCL_CLAMP_BLOCK_SIZE 256
|
||||
#define SYCL_ROPE_BLOCK_SIZE 256
|
||||
#define SYCL_ALIBI_BLOCK_SIZE 32
|
||||
#define SYCL_DIAG_MASK_INF_BLOCK_SIZE 32
|
||||
#define SYCL_QUANTIZE_BLOCK_SIZE 256
|
||||
#define SYCL_DEQUANTIZE_BLOCK_SIZE 256
|
||||
#define SYCL_GET_ROWS_BLOCK_SIZE 256
|
||||
#define SYCL_UPSCALE_BLOCK_SIZE 256
|
||||
#define SYCL_CONCAT_BLOCK_SIZE 256
|
||||
#define SYCL_PAD_BLOCK_SIZE 256
|
||||
#define SYCL_ACC_BLOCK_SIZE 256
|
||||
#define SYCL_IM2COL_BLOCK_SIZE 256
|
||||
#define SYCL_POOL2D_BLOCK_SIZE 256
|
||||
|
||||
// dmmv = dequantize_mul_mat_vec
|
||||
#ifndef GGML_SYCL_DMMV_X
|
||||
#define GGML_SYCL_DMMV_X 32
|
||||
#endif
|
||||
#ifndef GGML_SYCL_MMV_Y
|
||||
#define GGML_SYCL_MMV_Y 1
|
||||
#endif
|
||||
|
||||
#ifndef K_QUANTS_PER_ITERATION
|
||||
#define K_QUANTS_PER_ITERATION 2
|
||||
#else
|
||||
static_assert(K_QUANTS_PER_ITERATION == 1 || K_QUANTS_PER_ITERATION == 2, "K_QUANTS_PER_ITERATION must be 1 or 2");
|
||||
#endif
|
||||
|
||||
#ifndef GGML_SYCL_PEER_MAX_BATCH_SIZE
|
||||
#define GGML_SYCL_PEER_MAX_BATCH_SIZE 128
|
||||
#endif // GGML_SYCL_PEER_MAX_BATCH_SIZE
|
||||
|
||||
#define MUL_MAT_SRC1_COL_STRIDE 128
|
||||
|
||||
#endif // GGML_SYCL_PRESETS_HPP
|
13
llama.cpp
13
llama.cpp
@ -6625,16 +6625,6 @@ static int llama_model_load(const std::string & fname, llama_model & model, llam
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef GGML_USE_SYCL
|
||||
if (params.split_mode == LLAMA_SPLIT_MODE_NONE) {
|
||||
ggml_backend_sycl_set_single_device_mode(params.main_gpu);
|
||||
//SYCL use device index (0, 1, 2) directly, uer input device id, then convert to device index.
|
||||
params.main_gpu = ggml_backend_sycl_get_device_index(params.main_gpu);
|
||||
} else {
|
||||
ggml_backend_sycl_set_mul_device_mode();
|
||||
}
|
||||
#endif
|
||||
|
||||
if (!llm_load_tensors(
|
||||
ml, model, params.n_gpu_layers, params.split_mode, params.main_gpu, params.tensor_split, params.use_mlock,
|
||||
params.progress_callback, params.progress_callback_user_data
|
||||
@ -16241,8 +16231,7 @@ struct llama_context * llama_new_context_with_model(
|
||||
if (model->split_mode == LLAMA_SPLIT_MODE_NONE || model->split_mode == LLAMA_SPLIT_MODE_ROW) {
|
||||
ggml_backend_t backend = ggml_backend_sycl_init(model->main_gpu);
|
||||
if (backend == nullptr) {
|
||||
int main_gpu_id = ggml_backend_sycl_get_device_id(model->main_gpu);
|
||||
LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d (index %d) backend\n", __func__, main_gpu_id, model->main_gpu);
|
||||
LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d backend\n", __func__, model->main_gpu);
|
||||
llama_free(ctx);
|
||||
return nullptr;
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user