Merge branch 'master' into compilade/mamba2
This commit is contained in:
Francis Couture-Harpin
commit
8d8f065743
56 changed files with 15958 additions and 20295 deletions
|
|
@ -305,27 +305,10 @@ extern "C" {
|
|||
GGML_API void ggml_backend_tensor_alloc(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, void * addr);
|
||||
GGML_API void ggml_backend_view_init(struct ggml_tensor * tensor);
|
||||
|
||||
//
|
||||
// CPU backend
|
||||
//
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_cpu_init(void);
|
||||
|
||||
GGML_API bool ggml_backend_is_cpu (ggml_backend_t backend);
|
||||
GGML_API void ggml_backend_cpu_set_n_threads (ggml_backend_t backend_cpu, int n_threads);
|
||||
GGML_API void ggml_backend_cpu_set_threadpool (ggml_backend_t backend_cpu, ggml_threadpool_t threadpool);
|
||||
GGML_API void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
|
||||
|
||||
// Create a backend buffer from an existing pointer
|
||||
// CPU buffer types are always available
|
||||
GGML_API ggml_backend_buffer_t ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size);
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_cpu_reg(void);
|
||||
|
||||
#ifdef GGML_USE_CPU_HBM
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
|
||||
#endif
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
|
|
|||
38
ggml/include/ggml-cpp.h
Normal file
38
ggml/include/ggml-cpp.h
Normal file
|
|
@ -0,0 +1,38 @@
|
|||
#pragma once
|
||||
|
||||
#ifndef __cplusplus
|
||||
#error "This header is for C++ only"
|
||||
#endif
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-alloc.h"
|
||||
#include "ggml-backend.h"
|
||||
#include <memory>
|
||||
|
||||
// Smart pointers for ggml types
|
||||
|
||||
// ggml
|
||||
|
||||
struct ggml_context_deleter { void operator()(ggml_context * ctx) { ggml_free(ctx); } };
|
||||
struct gguf_context_deleter { void operator()(gguf_context * ctx) { gguf_free(ctx); } };
|
||||
|
||||
typedef std::unique_ptr<ggml_context, ggml_context_deleter> ggml_context_ptr;
|
||||
typedef std::unique_ptr<gguf_context, gguf_context_deleter> gguf_context_ptr;
|
||||
|
||||
// ggml-alloc
|
||||
|
||||
struct ggml_gallocr_deleter { void operator()(ggml_gallocr_t galloc) { ggml_gallocr_free(galloc); } };
|
||||
|
||||
typedef std::unique_ptr<ggml_gallocr_t, ggml_gallocr_deleter> ggml_gallocr_ptr;
|
||||
|
||||
// ggml-backend
|
||||
|
||||
struct ggml_backend_deleter { void operator()(ggml_backend_t backend) { ggml_backend_free(backend); } };
|
||||
struct ggml_backend_buffer_deleter { void operator()(ggml_backend_buffer_t buffer) { ggml_backend_buffer_free(buffer); } };
|
||||
struct ggml_backend_event_deleter { void operator()(ggml_backend_event_t event) { ggml_backend_event_free(event); } };
|
||||
struct ggml_backend_sched_deleter { void operator()(ggml_backend_sched_t sched) { ggml_backend_sched_free(sched); } };
|
||||
|
||||
typedef std::unique_ptr<ggml_backend, ggml_backend_deleter> ggml_backend_ptr;
|
||||
typedef std::unique_ptr<ggml_backend_buffer, ggml_backend_buffer_deleter> ggml_backend_buffer_ptr;
|
||||
typedef std::unique_ptr<ggml_backend_event, ggml_backend_event_deleter> ggml_backend_event_ptr;
|
||||
typedef std::unique_ptr<ggml_backend_sched, ggml_backend_sched_deleter> ggml_backend_sched_ptr;
|
||||
150
ggml/include/ggml-cpu.h
Normal file
150
ggml/include/ggml-cpu.h
Normal file
|
|
@ -0,0 +1,150 @@
|
|||
#pragma once
|
||||
|
||||
#include "ggml.h"
|
||||
#include "ggml-backend.h"
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
#endif
|
||||
|
||||
// Scheduling priorities
|
||||
enum ggml_sched_priority {
|
||||
GGML_SCHED_PRIO_NORMAL,
|
||||
GGML_SCHED_PRIO_MEDIUM,
|
||||
GGML_SCHED_PRIO_HIGH,
|
||||
GGML_SCHED_PRIO_REALTIME
|
||||
};
|
||||
|
||||
// Threadpool params
|
||||
// Use ggml_threadpool_params_default() or ggml_threadpool_params_init() to populate the defaults
|
||||
struct ggml_threadpool_params {
|
||||
bool cpumask[GGML_MAX_N_THREADS]; // mask of cpu cores (all-zeros means use default affinity settings)
|
||||
int n_threads; // number of threads
|
||||
enum ggml_sched_priority prio; // thread priority
|
||||
uint32_t poll; // polling level (0 - no polling, 100 - aggressive polling)
|
||||
bool strict_cpu; // strict cpu placement
|
||||
bool paused; // start in paused state
|
||||
};
|
||||
|
||||
struct ggml_threadpool; // forward declaration, see ggml.c
|
||||
|
||||
typedef struct ggml_threadpool * ggml_threadpool_t;
|
||||
|
||||
// the compute plan that needs to be prepared for ggml_graph_compute()
|
||||
// since https://github.com/ggerganov/ggml/issues/287
|
||||
struct ggml_cplan {
|
||||
size_t work_size; // size of work buffer, calculated by `ggml_graph_plan()`
|
||||
uint8_t * work_data; // work buffer, to be allocated by caller before calling to `ggml_graph_compute()`
|
||||
|
||||
int n_threads;
|
||||
struct ggml_threadpool * threadpool;
|
||||
|
||||
// abort ggml_graph_compute when true
|
||||
ggml_abort_callback abort_callback;
|
||||
void * abort_callback_data;
|
||||
};
|
||||
|
||||
// numa strategies
|
||||
enum ggml_numa_strategy {
|
||||
GGML_NUMA_STRATEGY_DISABLED = 0,
|
||||
GGML_NUMA_STRATEGY_DISTRIBUTE = 1,
|
||||
GGML_NUMA_STRATEGY_ISOLATE = 2,
|
||||
GGML_NUMA_STRATEGY_NUMACTL = 3,
|
||||
GGML_NUMA_STRATEGY_MIRROR = 4,
|
||||
GGML_NUMA_STRATEGY_COUNT
|
||||
};
|
||||
|
||||
GGML_API void ggml_numa_init(enum ggml_numa_strategy numa); // call once for better performance on NUMA systems
|
||||
GGML_API bool ggml_is_numa(void); // true if init detected that system has >1 NUMA node
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value);
|
||||
GGML_API struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value);
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_set_i32 (struct ggml_tensor * tensor, int32_t value);
|
||||
GGML_API struct ggml_tensor * ggml_set_f32 (struct ggml_tensor * tensor, float value);
|
||||
|
||||
GGML_API int32_t ggml_get_i32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_API void ggml_set_i32_1d(const struct ggml_tensor * tensor, int i, int32_t value);
|
||||
|
||||
GGML_API int32_t ggml_get_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_API void ggml_set_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, int32_t value);
|
||||
|
||||
GGML_API float ggml_get_f32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_API void ggml_set_f32_1d(const struct ggml_tensor * tensor, int i, float value);
|
||||
|
||||
GGML_API float ggml_get_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_API void ggml_set_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, float value);
|
||||
|
||||
GGML_API struct ggml_threadpool_params ggml_threadpool_params_default(int n_threads);
|
||||
GGML_API void ggml_threadpool_params_init (struct ggml_threadpool_params * p, int n_threads);
|
||||
GGML_API bool ggml_threadpool_params_match (const struct ggml_threadpool_params * p0, const struct ggml_threadpool_params * p1);
|
||||
GGML_API struct ggml_threadpool * ggml_threadpool_new (struct ggml_threadpool_params * params);
|
||||
GGML_API void ggml_threadpool_free (struct ggml_threadpool * threadpool);
|
||||
GGML_API int ggml_threadpool_get_n_threads(struct ggml_threadpool * threadpool);
|
||||
GGML_API void ggml_threadpool_pause (struct ggml_threadpool * threadpool);
|
||||
GGML_API void ggml_threadpool_resume (struct ggml_threadpool * threadpool);
|
||||
|
||||
// ggml_graph_plan() has to be called before ggml_graph_compute()
|
||||
// when plan.work_size > 0, caller must allocate memory for plan.work_data
|
||||
GGML_API struct ggml_cplan ggml_graph_plan(
|
||||
const struct ggml_cgraph * cgraph,
|
||||
int n_threads, /* = GGML_DEFAULT_N_THREADS */
|
||||
struct ggml_threadpool * threadpool /* = NULL */ );
|
||||
GGML_API enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan);
|
||||
|
||||
// same as ggml_graph_compute() but the work data is allocated as a part of the context
|
||||
// note: the drawback of this API is that you must have ensured that the context has enough memory for the work data
|
||||
GGML_API enum ggml_status ggml_graph_compute_with_ctx(struct ggml_context * ctx, struct ggml_cgraph * cgraph, int n_threads);
|
||||
|
||||
// TODO: move to backend interface
|
||||
GGML_API int ggml_cpu_has_neon (void);
|
||||
GGML_API int ggml_cpu_has_sve (void);
|
||||
GGML_API int ggml_cpu_has_matmul_int8(void);
|
||||
// get the sve vector length in bytes
|
||||
GGML_API int ggml_cpu_get_sve_cnt(void);
|
||||
|
||||
// Internal types and functions exposed for tests and benchmarks
|
||||
|
||||
typedef void (*ggml_from_float_to_mat_t)
|
||||
(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t nr, int64_t k, int64_t bs);
|
||||
typedef void (*ggml_vec_dot_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x, size_t bx,
|
||||
const void * GGML_RESTRICT y, size_t by, int nrc);
|
||||
typedef void (*ggml_gemv_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
|
||||
const void * GGML_RESTRICT y, int nr, int nc);
|
||||
typedef void (*ggml_gemm_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
|
||||
const void * GGML_RESTRICT y, int nr, int nc);
|
||||
|
||||
struct ggml_type_traits_cpu {
|
||||
ggml_from_float_to_mat_t from_float_to_mat;
|
||||
ggml_vec_dot_t vec_dot;
|
||||
enum ggml_type vec_dot_type;
|
||||
int64_t nrows; // number of rows to process simultaneously
|
||||
int64_t ncols; // number of columns to process simultaneously
|
||||
ggml_gemv_t gemv;
|
||||
ggml_gemm_t gemm;
|
||||
};
|
||||
|
||||
GGML_API const struct ggml_type_traits_cpu * ggml_get_type_traits_cpu(enum ggml_type type);
|
||||
|
||||
GGML_API void ggml_cpu_init(void);
|
||||
|
||||
//
|
||||
// CPU backend
|
||||
//
|
||||
|
||||
GGML_API ggml_backend_t ggml_backend_cpu_init(void);
|
||||
|
||||
GGML_API bool ggml_backend_is_cpu (ggml_backend_t backend);
|
||||
GGML_API void ggml_backend_cpu_set_n_threads (ggml_backend_t backend_cpu, int n_threads);
|
||||
GGML_API void ggml_backend_cpu_set_threadpool (ggml_backend_t backend_cpu, ggml_threadpool_t threadpool);
|
||||
GGML_API void ggml_backend_cpu_set_abort_callback(ggml_backend_t backend_cpu, ggml_abort_callback abort_callback, void * abort_callback_data);
|
||||
|
||||
GGML_API ggml_backend_reg_t ggml_backend_cpu_reg(void);
|
||||
|
||||
#ifdef GGML_USE_CPU_HBM
|
||||
GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void);
|
||||
#endif
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
|
@ -558,10 +558,10 @@ extern "C" {
|
|||
|
||||
enum ggml_log_level {
|
||||
GGML_LOG_LEVEL_NONE = 0,
|
||||
GGML_LOG_LEVEL_INFO = 1,
|
||||
GGML_LOG_LEVEL_WARN = 2,
|
||||
GGML_LOG_LEVEL_ERROR = 3,
|
||||
GGML_LOG_LEVEL_DEBUG = 4,
|
||||
GGML_LOG_LEVEL_DEBUG = 1,
|
||||
GGML_LOG_LEVEL_INFO = 2,
|
||||
GGML_LOG_LEVEL_WARN = 3,
|
||||
GGML_LOG_LEVEL_ERROR = 4,
|
||||
GGML_LOG_LEVEL_CONT = 5, // continue previous log
|
||||
};
|
||||
|
||||
|
|
@ -573,6 +573,13 @@ extern "C" {
|
|||
GGML_TENSOR_FLAG_LOSS = 8, // ...defines loss for numerical optimization (multiple loss tensors add up)
|
||||
};
|
||||
|
||||
struct ggml_init_params {
|
||||
// memory pool
|
||||
size_t mem_size; // bytes
|
||||
void * mem_buffer; // if NULL, memory will be allocated internally
|
||||
bool no_alloc; // don't allocate memory for the tensor data
|
||||
};
|
||||
|
||||
// n-dimensional tensor
|
||||
struct ggml_tensor {
|
||||
enum ggml_type type;
|
||||
|
|
@ -618,59 +625,6 @@ extern "C" {
|
|||
// If it returns true, the computation is aborted
|
||||
typedef bool (*ggml_abort_callback)(void * data);
|
||||
|
||||
// Scheduling priorities
|
||||
enum ggml_sched_priority {
|
||||
GGML_SCHED_PRIO_NORMAL,
|
||||
GGML_SCHED_PRIO_MEDIUM,
|
||||
GGML_SCHED_PRIO_HIGH,
|
||||
GGML_SCHED_PRIO_REALTIME
|
||||
};
|
||||
|
||||
// Threadpool params
|
||||
// Use ggml_threadpool_params_default() or ggml_threadpool_params_init() to populate the defaults
|
||||
struct ggml_threadpool_params {
|
||||
bool cpumask[GGML_MAX_N_THREADS]; // mask of cpu cores (all-zeros means use default affinity settings)
|
||||
int n_threads; // number of threads
|
||||
enum ggml_sched_priority prio; // thread priority
|
||||
uint32_t poll; // polling level (0 - no polling, 100 - aggressive polling)
|
||||
bool strict_cpu; // strict cpu placement
|
||||
bool paused; // start in paused state
|
||||
};
|
||||
|
||||
struct ggml_threadpool; // forward declaration, see ggml.c
|
||||
|
||||
typedef struct ggml_threadpool * ggml_threadpool_t;
|
||||
|
||||
// the compute plan that needs to be prepared for ggml_graph_compute()
|
||||
// since https://github.com/ggerganov/ggml/issues/287
|
||||
struct ggml_cplan {
|
||||
size_t work_size; // size of work buffer, calculated by `ggml_graph_plan()`
|
||||
uint8_t * work_data; // work buffer, to be allocated by caller before calling to `ggml_graph_compute()`
|
||||
|
||||
int n_threads;
|
||||
struct ggml_threadpool * threadpool;
|
||||
|
||||
// abort ggml_graph_compute when true
|
||||
ggml_abort_callback abort_callback;
|
||||
void * abort_callback_data;
|
||||
};
|
||||
|
||||
struct ggml_init_params {
|
||||
// memory pool
|
||||
size_t mem_size; // bytes
|
||||
void * mem_buffer; // if NULL, memory will be allocated internally
|
||||
bool no_alloc; // don't allocate memory for the tensor data
|
||||
};
|
||||
|
||||
// numa strategies
|
||||
enum ggml_numa_strategy {
|
||||
GGML_NUMA_STRATEGY_DISABLED = 0,
|
||||
GGML_NUMA_STRATEGY_DISTRIBUTE = 1,
|
||||
GGML_NUMA_STRATEGY_ISOLATE = 2,
|
||||
GGML_NUMA_STRATEGY_NUMACTL = 3,
|
||||
GGML_NUMA_STRATEGY_MIRROR = 4,
|
||||
GGML_NUMA_STRATEGY_COUNT
|
||||
};
|
||||
|
||||
//
|
||||
// GUID
|
||||
|
|
@ -693,9 +647,6 @@ extern "C" {
|
|||
// accepts a UTF-8 path, even on Windows
|
||||
GGML_API FILE * ggml_fopen(const char * fname, const char * mode);
|
||||
|
||||
GGML_API void ggml_numa_init(enum ggml_numa_strategy numa); // call once for better performance on NUMA systems
|
||||
GGML_API bool ggml_is_numa(void); // true if init detected that system has >1 NUMA node
|
||||
|
||||
GGML_API void ggml_print_object (const struct ggml_object * obj);
|
||||
GGML_API void ggml_print_objects(const struct ggml_context * ctx);
|
||||
|
||||
|
|
@ -797,8 +748,7 @@ extern "C" {
|
|||
int64_t ne2,
|
||||
int64_t ne3);
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value);
|
||||
GGML_API struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value);
|
||||
GGML_API void * ggml_new_buffer(struct ggml_context * ctx, size_t nbytes);
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_dup_tensor (struct ggml_context * ctx, const struct ggml_tensor * src);
|
||||
GGML_API struct ggml_tensor * ggml_view_tensor(struct ggml_context * ctx, struct ggml_tensor * src);
|
||||
|
|
@ -808,35 +758,25 @@ extern "C" {
|
|||
GGML_API struct ggml_tensor * ggml_get_next_tensor (const struct ggml_context * ctx, struct ggml_tensor * tensor);
|
||||
GGML_API struct ggml_tensor * ggml_get_tensor(struct ggml_context * ctx, const char * name);
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_set_zero(struct ggml_tensor * tensor);
|
||||
GGML_API struct ggml_tensor * ggml_set_i32 (struct ggml_tensor * tensor, int32_t value);
|
||||
GGML_API struct ggml_tensor * ggml_set_f32 (struct ggml_tensor * tensor, float value);
|
||||
|
||||
// Converts a flat index into coordinates
|
||||
GGML_API void ggml_unravel_index(const struct ggml_tensor * tensor, int64_t i, int64_t * i0, int64_t * i1, int64_t * i2, int64_t * i3);
|
||||
GGML_API void ggml_unravel_index(const struct ggml_tensor * tensor, int64_t i, int64_t * i0, int64_t * i1, int64_t * i2, int64_t * i3);
|
||||
|
||||
GGML_API int32_t ggml_get_i32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_API void ggml_set_i32_1d(const struct ggml_tensor * tensor, int i, int32_t value);
|
||||
|
||||
GGML_API int32_t ggml_get_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_API void ggml_set_i32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, int32_t value);
|
||||
|
||||
GGML_API float ggml_get_f32_1d(const struct ggml_tensor * tensor, int i);
|
||||
GGML_API void ggml_set_f32_1d(const struct ggml_tensor * tensor, int i, float value);
|
||||
|
||||
GGML_API float ggml_get_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3);
|
||||
GGML_API void ggml_set_f32_nd(const struct ggml_tensor * tensor, int i0, int i1, int i2, int i3, float value);
|
||||
GGML_API enum ggml_unary_op ggml_get_unary_op(const struct ggml_tensor * tensor);
|
||||
|
||||
GGML_API void * ggml_get_data (const struct ggml_tensor * tensor);
|
||||
GGML_API float * ggml_get_data_f32(const struct ggml_tensor * tensor);
|
||||
|
||||
GGML_API enum ggml_unary_op ggml_get_unary_op(const struct ggml_tensor * tensor);
|
||||
|
||||
GGML_API const char * ggml_get_name (const struct ggml_tensor * tensor);
|
||||
GGML_API struct ggml_tensor * ggml_set_name ( struct ggml_tensor * tensor, const char * name);
|
||||
GGML_ATTRIBUTE_FORMAT(2, 3)
|
||||
GGML_API struct ggml_tensor * ggml_format_name( struct ggml_tensor * tensor, const char * fmt, ...);
|
||||
|
||||
// Tensor flags
|
||||
GGML_API void ggml_set_input(struct ggml_tensor * tensor);
|
||||
GGML_API void ggml_set_output(struct ggml_tensor * tensor);
|
||||
GGML_API void ggml_set_param(struct ggml_context * ctx, struct ggml_tensor * tensor);
|
||||
GGML_API void ggml_set_loss(struct ggml_tensor * tensor);
|
||||
|
||||
//
|
||||
// operations on tensors with backpropagation
|
||||
//
|
||||
|
|
@ -2053,9 +1993,6 @@ extern "C" {
|
|||
// automatic differentiation
|
||||
//
|
||||
|
||||
GGML_API void ggml_set_param(struct ggml_context * ctx, struct ggml_tensor * tensor);
|
||||
GGML_API void ggml_set_loss(struct ggml_tensor * tensor);
|
||||
|
||||
GGML_API void ggml_build_forward_expand (struct ggml_cgraph * cgraph, struct ggml_tensor * tensor);
|
||||
GGML_API void ggml_build_backward_expand(struct ggml_context * ctx, struct ggml_cgraph * gf, struct ggml_cgraph * gb, bool accumulate);
|
||||
|
||||
|
|
@ -2087,27 +2024,6 @@ extern "C" {
|
|||
GGML_API size_t ggml_graph_overhead(void);
|
||||
GGML_API size_t ggml_graph_overhead_custom(size_t size, bool grads);
|
||||
|
||||
GGML_API struct ggml_threadpool_params ggml_threadpool_params_default(int n_threads);
|
||||
GGML_API void ggml_threadpool_params_init (struct ggml_threadpool_params * p, int n_threads);
|
||||
GGML_API bool ggml_threadpool_params_match (const struct ggml_threadpool_params * p0, const struct ggml_threadpool_params * p1);
|
||||
GGML_API struct ggml_threadpool * ggml_threadpool_new (struct ggml_threadpool_params * params);
|
||||
GGML_API void ggml_threadpool_free (struct ggml_threadpool * threadpool);
|
||||
GGML_API int ggml_threadpool_get_n_threads(struct ggml_threadpool * threadpool);
|
||||
GGML_API void ggml_threadpool_pause (struct ggml_threadpool * threadpool);
|
||||
GGML_API void ggml_threadpool_resume (struct ggml_threadpool * threadpool);
|
||||
|
||||
// ggml_graph_plan() has to be called before ggml_graph_compute()
|
||||
// when plan.work_size > 0, caller must allocate memory for plan.work_data
|
||||
GGML_API struct ggml_cplan ggml_graph_plan(
|
||||
const struct ggml_cgraph * cgraph,
|
||||
int n_threads, /* = GGML_DEFAULT_N_THREADS */
|
||||
struct ggml_threadpool * threadpool /* = NULL */ );
|
||||
GGML_API enum ggml_status ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan);
|
||||
|
||||
// same as ggml_graph_compute() but the work data is allocated as a part of the context
|
||||
// note: the drawback of this API is that you must have ensured that the context has enough memory for the work data
|
||||
GGML_API enum ggml_status ggml_graph_compute_with_ctx(struct ggml_context * ctx, struct ggml_cgraph * cgraph, int n_threads);
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_graph_get_tensor(struct ggml_cgraph * cgraph, const char * name);
|
||||
|
||||
GGML_API void ggml_graph_export(const struct ggml_cgraph * cgraph, const char * fname);
|
||||
|
|
@ -2278,6 +2194,8 @@ extern "C" {
|
|||
} lbfgs;
|
||||
};
|
||||
|
||||
GGML_API struct ggml_tensor * ggml_set_zero(struct ggml_tensor * tensor);
|
||||
|
||||
GGML_API struct ggml_opt_params ggml_opt_default_params(enum ggml_opt_type type);
|
||||
|
||||
// optimize the function defined by the tensor f
|
||||
|
|
@ -2309,12 +2227,6 @@ extern "C" {
|
|||
ggml_opt_callback callback,
|
||||
void * callback_data);
|
||||
|
||||
//
|
||||
// tensor flags
|
||||
//
|
||||
GGML_API void ggml_set_input(struct ggml_tensor * tensor);
|
||||
GGML_API void ggml_set_output(struct ggml_tensor * tensor);
|
||||
|
||||
//
|
||||
// quantization
|
||||
//
|
||||
|
|
@ -2483,8 +2395,6 @@ extern "C" {
|
|||
GGML_API int ggml_cpu_has_avx512_bf16(void);
|
||||
GGML_API int ggml_cpu_has_amx_int8 (void);
|
||||
GGML_API int ggml_cpu_has_fma (void);
|
||||
GGML_API int ggml_cpu_has_neon (void);
|
||||
GGML_API int ggml_cpu_has_sve (void);
|
||||
GGML_API int ggml_cpu_has_arm_fma (void);
|
||||
GGML_API int ggml_cpu_has_metal (void);
|
||||
GGML_API int ggml_cpu_has_f16c (void);
|
||||
|
|
@ -2501,17 +2411,9 @@ extern "C" {
|
|||
GGML_API int ggml_cpu_has_sycl (void);
|
||||
GGML_API int ggml_cpu_has_rpc (void);
|
||||
GGML_API int ggml_cpu_has_vsx (void);
|
||||
GGML_API int ggml_cpu_has_matmul_int8(void);
|
||||
GGML_API int ggml_cpu_has_cann (void);
|
||||
GGML_API int ggml_cpu_has_llamafile (void);
|
||||
|
||||
// get the sve vector length in bytes
|
||||
GGML_API int ggml_cpu_get_sve_cnt(void);
|
||||
|
||||
//
|
||||
// Internal types and functions exposed for tests and benchmarks
|
||||
//
|
||||
|
||||
#ifdef __cplusplus
|
||||
// restrict not standard in C++
|
||||
#define GGML_RESTRICT
|
||||
|
|
@ -2520,14 +2422,6 @@ extern "C" {
|
|||
#endif
|
||||
typedef void (*ggml_to_float_t) (const void * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
|
||||
typedef void (*ggml_from_float_t)(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
|
||||
typedef void (*ggml_from_float_to_mat_t)
|
||||
(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t nr, int64_t k, int64_t bs);
|
||||
typedef void (*ggml_vec_dot_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x, size_t bx,
|
||||
const void * GGML_RESTRICT y, size_t by, int nrc);
|
||||
typedef void (*ggml_gemv_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
|
||||
const void * GGML_RESTRICT y, int nr, int nc);
|
||||
typedef void (*ggml_gemm_t) (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT x,
|
||||
const void * GGML_RESTRICT y, int nr, int nc);
|
||||
|
||||
struct ggml_type_traits {
|
||||
const char * type_name;
|
||||
|
|
@ -2538,13 +2432,6 @@ extern "C" {
|
|||
ggml_to_float_t to_float;
|
||||
ggml_from_float_t from_float;
|
||||
ggml_from_float_t from_float_ref;
|
||||
ggml_from_float_to_mat_t from_float_to_mat;
|
||||
ggml_vec_dot_t vec_dot;
|
||||
enum ggml_type vec_dot_type;
|
||||
int64_t nrows; // number of rows to process simultaneously
|
||||
int64_t ncols; // number of columns to process simultaneously
|
||||
ggml_gemv_t gemv;
|
||||
ggml_gemm_t gemm;
|
||||
};
|
||||
|
||||
GGML_API const struct ggml_type_traits * ggml_get_type_traits(enum ggml_type type);
|
||||
|
|
|
|||
|
|
@ -1366,9 +1366,12 @@ endif()
|
|||
|
||||
add_library(ggml
|
||||
../include/ggml.h
|
||||
../include/ggml-cpu.h
|
||||
../include/ggml-alloc.h
|
||||
../include/ggml-backend.h
|
||||
../include/ggml-cpp.h
|
||||
ggml.c
|
||||
ggml-cpu.c
|
||||
ggml-alloc.c
|
||||
ggml-backend.cpp
|
||||
ggml-quants.c
|
||||
|
|
@ -1393,7 +1396,7 @@ if (EMSCRIPTEN)
|
|||
endif()
|
||||
|
||||
target_compile_definitions(ggml PUBLIC ${GGML_CDEF_PUBLIC})
|
||||
target_include_directories(ggml PUBLIC ../include)
|
||||
target_include_directories(ggml PUBLIC $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/../include> $<INSTALL_INTERFACE:include>)
|
||||
target_include_directories(ggml PRIVATE . ${GGML_EXTRA_INCLUDES})
|
||||
target_link_directories (ggml PRIVATE ${GGML_EXTRA_LIBDIRS})
|
||||
target_compile_features (ggml PRIVATE c_std_11) # don't bump
|
||||
|
|
|
|||
|
|
@ -7,6 +7,7 @@
|
|||
|
||||
#include "ggml-quants.h"
|
||||
#include "ggml-impl.h"
|
||||
#include "ggml-cpu.h"
|
||||
#include "ggml-cpu-impl.h"
|
||||
|
||||
#include <math.h>
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load diff
|
|
@ -1227,7 +1227,6 @@ static ggml_backend_buffer_t ggml_backend_cann_host_buffer_type_alloc_buffer(ggm
|
|||
|
||||
ggml_backend_buffer_t buffer = ggml_backend_cpu_buffer_from_ptr(hostPtr, size);
|
||||
buffer->buft = buft;
|
||||
buffer->iface.get_name = ggml_backend_cann_host_buffer_name;
|
||||
buffer->iface.free_buffer = ggml_backend_cann_host_buffer_free;
|
||||
|
||||
return buffer;
|
||||
|
|
|
|||
13791
ggml/src/ggml-cpu.c
Normal file
13791
ggml/src/ggml-cpu.c
Normal file
File diff suppressed because it is too large
Load diff
|
|
@ -1297,11 +1297,17 @@ static void ggml_cuda_set_peer_access(const int n_tokens, int main_device) {
|
|||
cudaError_t err = cudaDeviceEnablePeerAccess(id_other, 0);
|
||||
if (err != cudaErrorPeerAccessAlreadyEnabled) {
|
||||
CUDA_CHECK(err);
|
||||
} else {
|
||||
// reset the error
|
||||
cudaGetLastError();
|
||||
}
|
||||
} else {
|
||||
cudaError_t err = cudaDeviceDisablePeerAccess(id_other);
|
||||
if (err != cudaErrorPeerAccessNotEnabled) {
|
||||
CUDA_CHECK(err);
|
||||
} else {
|
||||
// reset the error
|
||||
cudaGetLastError();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -8,6 +8,7 @@
|
|||
#include <stdlib.h> // load `stdlib.h` before other headers to work around MinGW bug: https://sourceforge.net/p/mingw-w64/bugs/192/
|
||||
#include <stdbool.h>
|
||||
#include <stdint.h>
|
||||
#include <string.h>
|
||||
|
||||
#ifdef __cplusplus
|
||||
extern "C" {
|
||||
|
|
@ -36,6 +37,20 @@ extern "C" {
|
|||
#endif
|
||||
#endif
|
||||
|
||||
static inline int ggml_up32(int n) {
|
||||
return (n + 31) & ~31;
|
||||
}
|
||||
|
||||
//static inline int ggml_up64(int n) {
|
||||
// return (n + 63) & ~63;
|
||||
//}
|
||||
|
||||
static inline int ggml_up(int n, int m) {
|
||||
// assert m is a power of 2
|
||||
GGML_ASSERT((m & (m - 1)) == 0);
|
||||
return (n + m - 1) & ~(m - 1);
|
||||
}
|
||||
|
||||
//
|
||||
// logging
|
||||
//
|
||||
|
|
@ -51,6 +66,74 @@ void ggml_log_callback_default(enum ggml_log_level level, const char * text, voi
|
|||
#define GGML_LOG_DEBUG(...) ggml_log_internal(GGML_LOG_LEVEL_DEBUG, __VA_ARGS__)
|
||||
#define GGML_LOG_CONT(...) ggml_log_internal(GGML_LOG_LEVEL_CONT , __VA_ARGS__)
|
||||
|
||||
#define GGML_DEBUG 0
|
||||
|
||||
#if (GGML_DEBUG >= 1)
|
||||
#define GGML_PRINT_DEBUG(...) GGML_LOG_DEBUG(__VA_ARGS__)
|
||||
#else
|
||||
#define GGML_PRINT_DEBUG(...)
|
||||
#endif
|
||||
|
||||
#if (GGML_DEBUG >= 5)
|
||||
#define GGML_PRINT_DEBUG_5(...) GGML_LOG_DEBUG(__VA_ARGS__)
|
||||
#else
|
||||
#define GGML_PRINT_DEBUG_5(...)
|
||||
#endif
|
||||
|
||||
#if (GGML_DEBUG >= 10)
|
||||
#define GGML_PRINT_DEBUG_10(...) GGML_LOG_DEBUG(__VA_ARGS__)
|
||||
#else
|
||||
#define GGML_PRINT_DEBUG_10(...)
|
||||
#endif
|
||||
|
||||
// tensor params
|
||||
|
||||
static void ggml_set_op_params(struct ggml_tensor * tensor, const void * params, size_t params_size) {
|
||||
GGML_ASSERT(tensor != NULL); // silence -Warray-bounds warnings
|
||||
assert(params_size <= GGML_MAX_OP_PARAMS);
|
||||
memcpy(tensor->op_params, params, params_size);
|
||||
}
|
||||
|
||||
static int32_t ggml_get_op_params_i32(const struct ggml_tensor * tensor, uint32_t i) {
|
||||
assert(i < GGML_MAX_OP_PARAMS / sizeof(int32_t));
|
||||
return ((const int32_t *)(tensor->op_params))[i];
|
||||
}
|
||||
|
||||
static float ggml_get_op_params_f32(const struct ggml_tensor * tensor, uint32_t i) {
|
||||
assert(i < GGML_MAX_OP_PARAMS / sizeof(float));
|
||||
return ((const float *)(tensor->op_params))[i];
|
||||
}
|
||||
|
||||
static void ggml_set_op_params_i32(struct ggml_tensor * tensor, uint32_t i, int32_t value) {
|
||||
assert(i < GGML_MAX_OP_PARAMS / sizeof(int32_t));
|
||||
((int32_t *)(tensor->op_params))[i] = value;
|
||||
}
|
||||
|
||||
static void ggml_set_op_params_f32(struct ggml_tensor * tensor, uint32_t i, float value) {
|
||||
assert(i < GGML_MAX_OP_PARAMS / sizeof(float));
|
||||
((float *)(tensor->op_params))[i] = value;
|
||||
}
|
||||
|
||||
struct ggml_map_custom1_op_params {
|
||||
ggml_custom1_op_t fun;
|
||||
int n_tasks;
|
||||
void * userdata;
|
||||
};
|
||||
|
||||
|
||||
struct ggml_map_custom2_op_params {
|
||||
ggml_custom2_op_t fun;
|
||||
int n_tasks;
|
||||
void * userdata;
|
||||
};
|
||||
|
||||
|
||||
struct ggml_map_custom3_op_params {
|
||||
ggml_custom3_op_t fun;
|
||||
int n_tasks;
|
||||
void * userdata;
|
||||
};
|
||||
|
||||
// bitset
|
||||
|
||||
typedef uint32_t ggml_bitset_t;
|
||||
|
|
@ -204,6 +287,10 @@ struct ggml_cgraph ggml_graph_view(struct ggml_cgraph * cgraph, int i0, int i1);
|
|||
void * ggml_aligned_malloc(size_t size);
|
||||
void ggml_aligned_free(void * ptr, size_t size);
|
||||
|
||||
// TODO: move to threading file
|
||||
void ggml_critical_section_start(void);
|
||||
void ggml_critical_section_end(void);
|
||||
|
||||
#ifdef __cplusplus
|
||||
}
|
||||
#endif
|
||||
|
|
|
|||
|
|
@ -451,7 +451,14 @@ static struct ggml_backend_metal_context * ggml_metal_init(ggml_backend_dev_t de
|
|||
GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
|
||||
return NULL;
|
||||
}
|
||||
|
||||
#if !__has_feature(objc_arc)
|
||||
[options release];
|
||||
#endif
|
||||
}
|
||||
#if GGML_METAL_EMBED_LIBRARY
|
||||
[src release];
|
||||
#endif // GGML_METAL_EMBED_LIBRARY
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -12,6 +12,436 @@ using namespace metal;
|
|||
|
||||
#define N_SIMDWIDTH 32 // assuming SIMD group size is 32
|
||||
|
||||
constexpr constant static float kvalues_iq4nl_f[16] = {
|
||||
-127.f, -104.f, -83.f, -65.f, -49.f, -35.f, -22.f, -10.f, 1.f, 13.f, 25.f, 38.f, 53.f, 69.f, 89.f, 113.f
|
||||
};
|
||||
|
||||
// NOTE: this is not dequantizing - we are simply fitting the template
|
||||
template <typename type4x4>
|
||||
void dequantize_f32(device const float4x4 * src, short il, thread type4x4 & reg) {
|
||||
reg = (type4x4)(*src);
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_f16(device const half4x4 * src, short il, thread type4x4 & reg) {
|
||||
reg = (type4x4)(*src);
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q4_0(device const block_q4_0 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 1);
|
||||
const float d1 = il ? (xb->d / 16.h) : xb->d;
|
||||
const float d2 = d1 / 256.f;
|
||||
const float md = -8.h * xb->d;
|
||||
const ushort mask0 = il ? 0x00F0 : 0x000F;
|
||||
const ushort mask1 = mask0 << 8;
|
||||
|
||||
for (int i=0;i<8;i++) {
|
||||
reg[i/2][2*(i%2)+0] = d1 * (qs[i] & mask0) + md;
|
||||
reg[i/2][2*(i%2)+1] = d2 * (qs[i] & mask1) + md;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q4_1(device const block_q4_1 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 2);
|
||||
const float d1 = il ? (xb->d / 16.h) : xb->d;
|
||||
const float d2 = d1 / 256.f;
|
||||
const float m = xb->m;
|
||||
const ushort mask0 = il ? 0x00F0 : 0x000F;
|
||||
const ushort mask1 = mask0 << 8;
|
||||
|
||||
for (int i=0;i<8;i++) {
|
||||
reg[i/2][2*(i%2)+0] = ((qs[i] & mask0) * d1) + m;
|
||||
reg[i/2][2*(i%2)+1] = ((qs[i] & mask1) * d2) + m;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q5_0(device const block_q5_0 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 3);
|
||||
const float d = xb->d;
|
||||
const float md = -16.h * xb->d;
|
||||
const ushort mask = il ? 0x00F0 : 0x000F;
|
||||
|
||||
const uint32_t qh = *((device const uint32_t *)xb->qh);
|
||||
|
||||
const int x_mv = il ? 4 : 0;
|
||||
|
||||
const int gh_mv = il ? 12 : 0;
|
||||
const int gh_bk = il ? 0 : 4;
|
||||
|
||||
for (int i = 0; i < 8; i++) {
|
||||
// extract the 5-th bits for x0 and x1
|
||||
const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
|
||||
const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
|
||||
|
||||
// combine the 4-bits from qs with the 5th bit
|
||||
const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
|
||||
const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
|
||||
|
||||
reg[i/2][2*(i%2)+0] = d * x0 + md;
|
||||
reg[i/2][2*(i%2)+1] = d * x1 + md;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q5_1(device const block_q5_1 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 4);
|
||||
const float d = xb->d;
|
||||
const float m = xb->m;
|
||||
const ushort mask = il ? 0x00F0 : 0x000F;
|
||||
|
||||
const uint32_t qh = *((device const uint32_t *)xb->qh);
|
||||
|
||||
const int x_mv = il ? 4 : 0;
|
||||
|
||||
const int gh_mv = il ? 12 : 0;
|
||||
const int gh_bk = il ? 0 : 4;
|
||||
|
||||
for (int i = 0; i < 8; i++) {
|
||||
// extract the 5-th bits for x0 and x1
|
||||
const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
|
||||
const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
|
||||
|
||||
// combine the 4-bits from qs with the 5th bit
|
||||
const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
|
||||
const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
|
||||
|
||||
reg[i/2][2*(i%2)+0] = d * x0 + m;
|
||||
reg[i/2][2*(i%2)+1] = d * x1 + m;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg) {
|
||||
device const int8_t * qs = ((device const int8_t *)xb->qs);
|
||||
const half d = xb->d;
|
||||
|
||||
for (int i = 0; i < 16; i++) {
|
||||
reg[i/4][i%4] = (qs[i + 16*il] * d);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q2_K(device const block_q2_K *xb, short il, thread type4x4 & reg) {
|
||||
const float d = xb->d;
|
||||
const float min = xb->dmin;
|
||||
device const uint8_t * q = (device const uint8_t *)xb->qs;
|
||||
float dl, ml;
|
||||
uint8_t sc = xb->scales[il];
|
||||
|
||||
q = q + 32*(il/8) + 16*(il&1);
|
||||
il = (il/2)%4;
|
||||
|
||||
half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
|
||||
uchar mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
|
||||
dl = d * (sc & 0xF) * coef, ml = min * (sc >> 4);
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * (q[i] & mask) - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q3_K(device const block_q3_K *xb, short il, thread type4x4 & reg) {
|
||||
const half d_all = xb->d;
|
||||
device const uint8_t * q = (device const uint8_t *)xb->qs;
|
||||
device const uint8_t * h = (device const uint8_t *)xb->hmask;
|
||||
device const int8_t * scales = (device const int8_t *)xb->scales;
|
||||
|
||||
q = q + 32 * (il/8) + 16 * (il&1);
|
||||
h = h + 16 * (il&1);
|
||||
uint8_t m = 1 << (il/2);
|
||||
uint16_t kmask1 = (il/4)>1 ? ((il/4)>2 ? 192 : 48) : \
|
||||
((il/4)>0 ? 12 : 3);
|
||||
uint16_t kmask2 = il/8 ? 0xF0 : 0x0F;
|
||||
uint16_t scale_2 = scales[il%8], scale_1 = scales[8 + il%4];
|
||||
int16_t dl_int = (il/4)&1 ? (scale_2&kmask2) | ((scale_1&kmask1) << 2)
|
||||
: (scale_2&kmask2) | ((scale_1&kmask1) << 4);
|
||||
float dl = il<8 ? d_all * (dl_int - 32.f) : d_all * (dl_int / 16.f - 32.f);
|
||||
const float ml = 4.f * dl;
|
||||
|
||||
il = (il/2) & 3;
|
||||
const half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
|
||||
const uint8_t mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
|
||||
dl *= coef;
|
||||
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * (q[i] & mask) - (h[i] & m ? 0 : ml);
|
||||
}
|
||||
}
|
||||
|
||||
static inline uchar2 get_scale_min_k4_just2(int j, int k, device const uchar * q) {
|
||||
return j < 4 ? uchar2{uchar(q[j+0+k] & 63), uchar(q[j+4+k] & 63)}
|
||||
: uchar2{uchar((q[j+4+k] & 0xF) | ((q[j-4+k] & 0xc0) >> 2)), uchar((q[j+4+k] >> 4) | ((q[j-0+k] & 0xc0) >> 2))};
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q4_K(device const block_q4_K *xb, short il, thread type4x4 & reg) {
|
||||
device const uchar * q = xb->qs;
|
||||
|
||||
short is = (il/4) * 2;
|
||||
q = q + (il/4) * 32 + 16 * (il&1);
|
||||
il = il & 3;
|
||||
const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
|
||||
const float d = il < 2 ? xb->d : xb->d / 16.h;
|
||||
const float min = xb->dmin;
|
||||
const float dl = d * sc[0];
|
||||
const float ml = min * sc[1];
|
||||
|
||||
const ushort mask = il<2 ? 0x0F : 0xF0;
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * (q[i] & mask) - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q5_K(device const block_q5_K *xb, short il, thread type4x4 & reg) {
|
||||
device const uint8_t * q = xb->qs;
|
||||
device const uint8_t * qh = xb->qh;
|
||||
|
||||
short is = (il/4) * 2;
|
||||
q = q + 32 * (il/4) + 16 * (il&1);
|
||||
qh = qh + 16 * (il&1);
|
||||
uint8_t ul = 1 << (il/2);
|
||||
il = il & 3;
|
||||
const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
|
||||
const float d = il < 2 ? xb->d : xb->d / 16.f;
|
||||
const float min = xb->dmin;
|
||||
const float dl = d * sc[0];
|
||||
const float ml = min * sc[1];
|
||||
|
||||
const ushort mask = il<2 ? 0x0F : 0xF0;
|
||||
const float qh_val = il<2 ? 16.f : 256.f;
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * ((q[i] & mask) + (qh[i] & ul ? qh_val : 0)) - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q6_K(device const block_q6_K *xb, short il, thread type4x4 & reg) {
|
||||
const half d_all = xb->d;
|
||||
device const uint8_t * ql = (device const uint8_t *)xb->ql;
|
||||
device const uint8_t * qh = (device const uint8_t *)xb->qh;
|
||||
device const int8_t * scales = (device const int8_t *)xb->scales;
|
||||
|
||||
ql = ql + 64*(il/8) + 32*((il/2)&1) + 16*(il&1);
|
||||
qh = qh + 32*(il/8) + 16*(il&1);
|
||||
float sc = scales[(il%2) + 2 * ((il/2))];
|
||||
il = (il/2) & 3;
|
||||
|
||||
const uint16_t kmask1 = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
|
||||
const uint16_t kmask2 = il>1 ? 0xF0 : 0x0F;
|
||||
const float coef = il>1 ? 1.f/16.f : 1.f;
|
||||
const float ml = d_all * sc * 32.f;
|
||||
const float dl = d_all * sc * coef;
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
const half q = il&1 ? ((ql[i] & kmask2) | ((qh[i] & kmask1) << 2))
|
||||
: ((ql[i] & kmask2) | ((qh[i] & kmask1) << 4));
|
||||
reg[i/4][i%4] = dl * q - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq2_xxs(device const block_iq2_xxs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
// each block of 32 needs 2 uint32_t's for the quants & scale, so 4 uint16_t's.
|
||||
device const uint16_t * q2 = xb->qs + 4*ib32;
|
||||
const uint32_t aux32_g = q2[0] | (q2[1] << 16);
|
||||
const uint32_t aux32_s = q2[2] | (q2[3] << 16);
|
||||
thread const uint8_t * aux8 = (thread const uint8_t *)&aux32_g;
|
||||
const float dl = d * (0.5f + (aux32_s >> 28)) * 0.25f;
|
||||
constant uint8_t * grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+0]);
|
||||
uint8_t signs = ksigns_iq2xs[(aux32_s >> 14*il) & 127];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+1]);
|
||||
signs = ksigns_iq2xs[(aux32_s >> (14*il+7)) & 127];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq2_xs(device const block_iq2_xs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint16_t * q2 = xb->qs + 4*ib32;
|
||||
const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
|
||||
constant uint8_t * grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+0] & 511));
|
||||
uint8_t signs = ksigns_iq2xs[q2[2*il+0] >> 9];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+1] & 511));
|
||||
signs = ksigns_iq2xs[q2[2*il+1] >> 9];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq3_xxs(device const block_iq3_xxs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint8_t * q3 = xb->qs + 8*ib32;
|
||||
device const uint16_t * gas = (device const uint16_t *)(xb->qs + QK_K/4) + 2*ib32;
|
||||
const uint32_t aux32 = gas[0] | (gas[1] << 16);
|
||||
const float dl = d * (0.5f + (aux32 >> 28)) * 0.5f;
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+0]);
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+1]);
|
||||
uint8_t signs = ksigns_iq2xs[(aux32 >> 14*il) & 127];
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
|
||||
reg[1][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
|
||||
}
|
||||
grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+2]);
|
||||
grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+3]);
|
||||
signs = ksigns_iq2xs[(aux32 >> (14*il+7)) & 127];
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[2][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
|
||||
reg[3][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq3_s(device const block_iq3_s * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint8_t * qs = xb->qs + 8*ib32;
|
||||
device const uint8_t * signs = xb->signs + 4*ib32 + 2*il;
|
||||
const uint8_t qh = xb->qh[ib32] >> 4*il;
|
||||
const float dl = d * (1 + 2*((xb->scales[ib32/2] >> 4*(ib32%2)) & 0xf));
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+0] | ((qh << 8) & 256)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+1] | ((qh << 7) & 256)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i+0]);
|
||||
reg[1][i] = dl * grid2[i] * select(1, -1, signs[0] & kmask_iq2xs[i+4]);
|
||||
}
|
||||
grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+2] | ((qh << 6) & 256)));
|
||||
grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+3] | ((qh << 5) & 256)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[2][i] = dl * grid1[i] * select(1, -1, signs[1] & kmask_iq2xs[i+0]);
|
||||
reg[3][i] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i+4]);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq2_s(device const block_iq2_s * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
|
||||
device const uint8_t * signs = qs + QK_K/8;
|
||||
const uint8_t qh = xb->qh[ib32] >> 4*il;
|
||||
const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq2s_grid + (qs[0] | ((qh << 8) & 0x300)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq2s_grid + (qs[1] | ((qh << 6) & 0x300)));
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[i/4+0][i%4] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i]);
|
||||
reg[i/4+2][i%4] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i]);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq1_s(device const block_iq1_s * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
const float d = xb->d;
|
||||
device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
|
||||
device const uint16_t * qh = xb->qh;
|
||||
const float dl = d * (2*((qh[ib32] >> 12) & 7) + 1);
|
||||
const float ml = dl * (qh[ib32] & 0x8000 ? -1 - IQ1S_DELTA : -1 + IQ1S_DELTA);
|
||||
const uint16_t h = qh[ib32] >> 6*il;
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((h << 8) & 0x700)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((h << 5) & 0x700)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * (grid1[i] & 0xf) + ml;
|
||||
reg[1][i] = dl * (grid1[i] >> 4) + ml;
|
||||
reg[2][i] = dl * (grid2[i] & 0xf) + ml;
|
||||
reg[3][i] = dl * (grid2[i] >> 4) + ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq1_m(device const block_iq1_m * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
device const uint16_t * sc = (device const uint16_t *)xb->scales;
|
||||
|
||||
iq1m_scale_t scale;
|
||||
scale.u16 = (sc[0] >> 12) | ((sc[1] >> 8) & 0x00f0) | ((sc[2] >> 4) & 0x0f00) | (sc[3] & 0xf000);
|
||||
const float d = scale.f16;
|
||||
|
||||
device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
|
||||
device const uint8_t * qh = xb->qh + 2*ib32 + il;
|
||||
|
||||
const float dl = d * (2*((sc[ib32/2] >> (6*(ib32%2)+3*il)) & 7) + 1);
|
||||
const float ml1 = dl * (qh[0] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
|
||||
const float ml2 = dl * (qh[0] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 4) & 0x700)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * (grid1[i] & 0xf) + ml1;
|
||||
reg[1][i] = dl * (grid1[i] >> 4) + ml1;
|
||||
reg[2][i] = dl * (grid2[i] & 0xf) + ml2;
|
||||
reg[3][i] = dl * (grid2[i] >> 4) + ml2;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq4_nl(device const block_iq4_nl * xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * q4 = (device const uint16_t *)xb->qs;
|
||||
const float d = xb->d;
|
||||
uint32_t aux32;
|
||||
thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
aux32 = ((q4[2*i] | (q4[2*i+1] << 16)) >> 4*il) & 0x0f0f0f0f;
|
||||
reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
|
||||
reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
|
||||
reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
|
||||
reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq4_xs(device const block_iq4_xs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint32_t * q4 = (device const uint32_t *)xb->qs + 4*ib32;
|
||||
const int ls = ((xb->scales_l[ib32/2] >> 4*(ib32%2)) & 0xf) | (((xb->scales_h >> 2*ib32) & 3) << 4);
|
||||
const float d = (float)xb->d * (ls - 32);
|
||||
uint32_t aux32;
|
||||
thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
aux32 = (q4[i] >> 4*il) & 0x0f0f0f0f;
|
||||
reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
|
||||
reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
|
||||
reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
|
||||
reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
|
||||
}
|
||||
}
|
||||
|
||||
enum ggml_sort_order {
|
||||
GGML_SORT_ORDER_ASC,
|
||||
GGML_SORT_ORDER_DESC,
|
||||
|
|
@ -2868,11 +3298,11 @@ kernel void kernel_flash_attn_ext_vec_f16(
|
|||
const short iv3 = iq3 / rv3;
|
||||
|
||||
// load the queries from shared memory into local memory
|
||||
float4 mq[D4];
|
||||
float4 mq[D4/NW];
|
||||
|
||||
for (short ii = 0; ii < D4; ii += NW) {
|
||||
short i = ii + tiisg;
|
||||
mq[i] = (float4) sq4[i];
|
||||
mq[ii/NW] = (float4) sq4[i];
|
||||
}
|
||||
|
||||
// pointer to the mask
|
||||
|
|
@ -2904,7 +3334,7 @@ kernel void kernel_flash_attn_ext_vec_f16(
|
|||
mk[2] = (float4) pk4[i + 2*(nb11/8)];
|
||||
mk[3] = (float4) pk4[i + 3*(nb11/8)];
|
||||
|
||||
mqk += (float4) (mq[i] * mk);
|
||||
mqk += (float4) (mq[ii/NW] * mk);
|
||||
}
|
||||
|
||||
// reduce the results from the threads in the simdgroup
|
||||
|
|
@ -2949,8 +3379,7 @@ kernel void kernel_flash_attn_ext_vec_f16(
|
|||
// O = diag(ms)*O
|
||||
#pragma unroll
|
||||
for (short ii = 0; ii < D4; ii += NW) {
|
||||
const short i = ii + tiisg;
|
||||
lo[i/NW] *= ms;
|
||||
lo[ii/NW] *= ms;
|
||||
}
|
||||
}
|
||||
|
||||
|
|
@ -2964,10 +3393,10 @@ kernel void kernel_flash_attn_ext_vec_f16(
|
|||
for (short ii = 0; ii < D4; ii += NW) {
|
||||
const short i = ii + tiisg;
|
||||
|
||||
lo[i/NW] += pv4[i + 0*(nb21/8)] * ss[4*cc + 0];
|
||||
lo[i/NW] += pv4[i + 1*(nb21/8)] * ss[4*cc + 1];
|
||||
lo[i/NW] += pv4[i + 2*(nb21/8)] * ss[4*cc + 2];
|
||||
lo[i/NW] += pv4[i + 3*(nb21/8)] * ss[4*cc + 3];
|
||||
lo[ii/NW] += pv4[i + 0*(nb21/8)] * ss[4*cc + 0];
|
||||
lo[ii/NW] += pv4[i + 1*(nb21/8)] * ss[4*cc + 1];
|
||||
lo[ii/NW] += pv4[i + 2*(nb21/8)] * ss[4*cc + 2];
|
||||
lo[ii/NW] += pv4[i + 3*(nb21/8)] * ss[4*cc + 3];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -3432,10 +3861,6 @@ static inline int best_index_int8(int n, constant float * val, float x) {
|
|||
return x - val[mu-1] < val[mu] - x ? mu-1 : mu;
|
||||
}
|
||||
|
||||
constexpr constant static float kvalues_iq4nl_f[16] = {
|
||||
-127.f, -104.f, -83.f, -65.f, -49.f, -35.f, -22.f, -10.f, 1.f, 13.f, 25.f, 38.f, 53.f, 69.f, 89.f, 113.f
|
||||
};
|
||||
|
||||
kernel void kernel_cpy_f32_iq4_nl(
|
||||
device const float * src0,
|
||||
device void * dst,
|
||||
|
|
@ -5550,440 +5975,6 @@ kernel void kernel_mul_mv_iq4_xs_f32(
|
|||
kernel_mul_mv_iq4_xs_f32_impl(src0, src1, dst, ne00, ne01, ne02, nb01, nb02, nb03, ne10, ne12, nb11, nb12, nb13, ne0, ne1, r2, r3, shared_values, tgpig, tiisg, sgitg);
|
||||
}
|
||||
|
||||
//============================= templates and their specializations =============================
|
||||
|
||||
// NOTE: this is not dequantizing - we are simply fitting the template
|
||||
template <typename type4x4>
|
||||
void dequantize_f32(device const float4x4 * src, short il, thread type4x4 & reg) {
|
||||
float4x4 temp = *(((device float4x4 *)src));
|
||||
for (int i = 0; i < 16; i++){
|
||||
reg[i/4][i%4] = temp[i/4][i%4];
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_f16(device const half4x4 * src, short il, thread type4x4 & reg) {
|
||||
half4x4 temp = *(((device half4x4 *)src));
|
||||
for (int i = 0; i < 16; i++){
|
||||
reg[i/4][i%4] = temp[i/4][i%4];
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q4_0(device const block_q4_0 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 1);
|
||||
const float d1 = il ? (xb->d / 16.h) : xb->d;
|
||||
const float d2 = d1 / 256.f;
|
||||
const float md = -8.h * xb->d;
|
||||
const ushort mask0 = il ? 0x00F0 : 0x000F;
|
||||
const ushort mask1 = mask0 << 8;
|
||||
|
||||
for (int i=0;i<8;i++) {
|
||||
reg[i/2][2*(i%2)+0] = d1 * (qs[i] & mask0) + md;
|
||||
reg[i/2][2*(i%2)+1] = d2 * (qs[i] & mask1) + md;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q4_1(device const block_q4_1 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 2);
|
||||
const float d1 = il ? (xb->d / 16.h) : xb->d;
|
||||
const float d2 = d1 / 256.f;
|
||||
const float m = xb->m;
|
||||
const ushort mask0 = il ? 0x00F0 : 0x000F;
|
||||
const ushort mask1 = mask0 << 8;
|
||||
|
||||
for (int i=0;i<8;i++) {
|
||||
reg[i/2][2*(i%2)+0] = ((qs[i] & mask0) * d1) + m;
|
||||
reg[i/2][2*(i%2)+1] = ((qs[i] & mask1) * d2) + m;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q5_0(device const block_q5_0 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 3);
|
||||
const float d = xb->d;
|
||||
const float md = -16.h * xb->d;
|
||||
const ushort mask = il ? 0x00F0 : 0x000F;
|
||||
|
||||
const uint32_t qh = *((device const uint32_t *)xb->qh);
|
||||
|
||||
const int x_mv = il ? 4 : 0;
|
||||
|
||||
const int gh_mv = il ? 12 : 0;
|
||||
const int gh_bk = il ? 0 : 4;
|
||||
|
||||
for (int i = 0; i < 8; i++) {
|
||||
// extract the 5-th bits for x0 and x1
|
||||
const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
|
||||
const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
|
||||
|
||||
// combine the 4-bits from qs with the 5th bit
|
||||
const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
|
||||
const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
|
||||
|
||||
reg[i/2][2*(i%2)+0] = d * x0 + md;
|
||||
reg[i/2][2*(i%2)+1] = d * x1 + md;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q5_1(device const block_q5_1 *xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * qs = ((device const uint16_t *)xb + 4);
|
||||
const float d = xb->d;
|
||||
const float m = xb->m;
|
||||
const ushort mask = il ? 0x00F0 : 0x000F;
|
||||
|
||||
const uint32_t qh = *((device const uint32_t *)xb->qh);
|
||||
|
||||
const int x_mv = il ? 4 : 0;
|
||||
|
||||
const int gh_mv = il ? 12 : 0;
|
||||
const int gh_bk = il ? 0 : 4;
|
||||
|
||||
for (int i = 0; i < 8; i++) {
|
||||
// extract the 5-th bits for x0 and x1
|
||||
const uint8_t xh_0 = ((qh >> (gh_mv + 2*i )) << gh_bk) & 0x10;
|
||||
const uint8_t xh_1 = ((qh >> (gh_mv + 2*i+1)) << gh_bk) & 0x10;
|
||||
|
||||
// combine the 4-bits from qs with the 5th bit
|
||||
const int32_t x0 = ((((qs[i] ) & mask) >> x_mv) | xh_0);
|
||||
const int32_t x1 = ((((qs[i] >> 8) & mask) >> x_mv) | xh_1);
|
||||
|
||||
reg[i/2][2*(i%2)+0] = d * x0 + m;
|
||||
reg[i/2][2*(i%2)+1] = d * x1 + m;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q8_0(device const block_q8_0 *xb, short il, thread type4x4 & reg) {
|
||||
device const int8_t * qs = ((device const int8_t *)xb->qs);
|
||||
const half d = xb->d;
|
||||
|
||||
for (int i = 0; i < 16; i++) {
|
||||
reg[i/4][i%4] = (qs[i + 16*il] * d);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q2_K(device const block_q2_K *xb, short il, thread type4x4 & reg) {
|
||||
const float d = xb->d;
|
||||
const float min = xb->dmin;
|
||||
device const uint8_t * q = (device const uint8_t *)xb->qs;
|
||||
float dl, ml;
|
||||
uint8_t sc = xb->scales[il];
|
||||
|
||||
q = q + 32*(il/8) + 16*(il&1);
|
||||
il = (il/2)%4;
|
||||
|
||||
half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
|
||||
uchar mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
|
||||
dl = d * (sc & 0xF) * coef, ml = min * (sc >> 4);
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * (q[i] & mask) - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q3_K(device const block_q3_K *xb, short il, thread type4x4 & reg) {
|
||||
const half d_all = xb->d;
|
||||
device const uint8_t * q = (device const uint8_t *)xb->qs;
|
||||
device const uint8_t * h = (device const uint8_t *)xb->hmask;
|
||||
device const int8_t * scales = (device const int8_t *)xb->scales;
|
||||
|
||||
q = q + 32 * (il/8) + 16 * (il&1);
|
||||
h = h + 16 * (il&1);
|
||||
uint8_t m = 1 << (il/2);
|
||||
uint16_t kmask1 = (il/4)>1 ? ((il/4)>2 ? 192 : 48) : \
|
||||
((il/4)>0 ? 12 : 3);
|
||||
uint16_t kmask2 = il/8 ? 0xF0 : 0x0F;
|
||||
uint16_t scale_2 = scales[il%8], scale_1 = scales[8 + il%4];
|
||||
int16_t dl_int = (il/4)&1 ? (scale_2&kmask2) | ((scale_1&kmask1) << 2)
|
||||
: (scale_2&kmask2) | ((scale_1&kmask1) << 4);
|
||||
float dl = il<8 ? d_all * (dl_int - 32.f) : d_all * (dl_int / 16.f - 32.f);
|
||||
const float ml = 4.f * dl;
|
||||
|
||||
il = (il/2) & 3;
|
||||
const half coef = il>1 ? (il>2 ? 1/64.h : 1/16.h) : (il>0 ? 1/4.h : 1.h);
|
||||
const uint8_t mask = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
|
||||
dl *= coef;
|
||||
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * (q[i] & mask) - (h[i] & m ? 0 : ml);
|
||||
}
|
||||
}
|
||||
|
||||
static inline uchar2 get_scale_min_k4_just2(int j, int k, device const uchar * q) {
|
||||
return j < 4 ? uchar2{uchar(q[j+0+k] & 63), uchar(q[j+4+k] & 63)}
|
||||
: uchar2{uchar((q[j+4+k] & 0xF) | ((q[j-4+k] & 0xc0) >> 2)), uchar((q[j+4+k] >> 4) | ((q[j-0+k] & 0xc0) >> 2))};
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q4_K(device const block_q4_K *xb, short il, thread type4x4 & reg) {
|
||||
device const uchar * q = xb->qs;
|
||||
|
||||
short is = (il/4) * 2;
|
||||
q = q + (il/4) * 32 + 16 * (il&1);
|
||||
il = il & 3;
|
||||
const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
|
||||
const float d = il < 2 ? xb->d : xb->d / 16.h;
|
||||
const float min = xb->dmin;
|
||||
const float dl = d * sc[0];
|
||||
const float ml = min * sc[1];
|
||||
|
||||
const ushort mask = il<2 ? 0x0F : 0xF0;
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * (q[i] & mask) - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q5_K(device const block_q5_K *xb, short il, thread type4x4 & reg) {
|
||||
device const uint8_t * q = xb->qs;
|
||||
device const uint8_t * qh = xb->qh;
|
||||
|
||||
short is = (il/4) * 2;
|
||||
q = q + 32 * (il/4) + 16 * (il&1);
|
||||
qh = qh + 16 * (il&1);
|
||||
uint8_t ul = 1 << (il/2);
|
||||
il = il & 3;
|
||||
const uchar2 sc = get_scale_min_k4_just2(is, il/2, xb->scales);
|
||||
const float d = il < 2 ? xb->d : xb->d / 16.f;
|
||||
const float min = xb->dmin;
|
||||
const float dl = d * sc[0];
|
||||
const float ml = min * sc[1];
|
||||
|
||||
const ushort mask = il<2 ? 0x0F : 0xF0;
|
||||
const float qh_val = il<2 ? 16.f : 256.f;
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
reg[i/4][i%4] = dl * ((q[i] & mask) + (qh[i] & ul ? qh_val : 0)) - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_q6_K(device const block_q6_K *xb, short il, thread type4x4 & reg) {
|
||||
const half d_all = xb->d;
|
||||
device const uint8_t * ql = (device const uint8_t *)xb->ql;
|
||||
device const uint8_t * qh = (device const uint8_t *)xb->qh;
|
||||
device const int8_t * scales = (device const int8_t *)xb->scales;
|
||||
|
||||
ql = ql + 64*(il/8) + 32*((il/2)&1) + 16*(il&1);
|
||||
qh = qh + 32*(il/8) + 16*(il&1);
|
||||
float sc = scales[(il%2) + 2 * ((il/2))];
|
||||
il = (il/2) & 3;
|
||||
|
||||
const uint16_t kmask1 = il>1 ? (il>2 ? 192 : 48) : (il>0 ? 12 : 3);
|
||||
const uint16_t kmask2 = il>1 ? 0xF0 : 0x0F;
|
||||
const float coef = il>1 ? 1.f/16.f : 1.f;
|
||||
const float ml = d_all * sc * 32.f;
|
||||
const float dl = d_all * sc * coef;
|
||||
for (int i = 0; i < 16; ++i) {
|
||||
const half q = il&1 ? ((ql[i] & kmask2) | ((qh[i] & kmask1) << 2))
|
||||
: ((ql[i] & kmask2) | ((qh[i] & kmask1) << 4));
|
||||
reg[i/4][i%4] = dl * q - ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq2_xxs(device const block_iq2_xxs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
// each block of 32 needs 2 uint32_t's for the quants & scale, so 4 uint16_t's.
|
||||
device const uint16_t * q2 = xb->qs + 4*ib32;
|
||||
const uint32_t aux32_g = q2[0] | (q2[1] << 16);
|
||||
const uint32_t aux32_s = q2[2] | (q2[3] << 16);
|
||||
thread const uint8_t * aux8 = (thread const uint8_t *)&aux32_g;
|
||||
const float dl = d * (0.5f + (aux32_s >> 28)) * 0.25f;
|
||||
constant uint8_t * grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+0]);
|
||||
uint8_t signs = ksigns_iq2xs[(aux32_s >> 14*il) & 127];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
grid = (constant uint8_t *)(iq2xxs_grid + aux8[2*il+1]);
|
||||
signs = ksigns_iq2xs[(aux32_s >> (14*il+7)) & 127];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq2_xs(device const block_iq2_xs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint16_t * q2 = xb->qs + 4*ib32;
|
||||
const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
|
||||
constant uint8_t * grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+0] & 511));
|
||||
uint8_t signs = ksigns_iq2xs[q2[2*il+0] >> 9];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
grid = (constant uint8_t *)(iq2xs_grid + (q2[2*il+1] & 511));
|
||||
signs = ksigns_iq2xs[q2[2*il+1] >> 9];
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[2+i/4][i%4] = dl * grid[i] * (signs & kmask_iq2xs[i] ? -1.f : 1.f);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq3_xxs(device const block_iq3_xxs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint8_t * q3 = xb->qs + 8*ib32;
|
||||
device const uint16_t * gas = (device const uint16_t *)(xb->qs + QK_K/4) + 2*ib32;
|
||||
const uint32_t aux32 = gas[0] | (gas[1] << 16);
|
||||
const float dl = d * (0.5f + (aux32 >> 28)) * 0.5f;
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+0]);
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+1]);
|
||||
uint8_t signs = ksigns_iq2xs[(aux32 >> 14*il) & 127];
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
|
||||
reg[1][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
|
||||
}
|
||||
grid1 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+2]);
|
||||
grid2 = (constant uint8_t *)(iq3xxs_grid + q3[4*il+3]);
|
||||
signs = ksigns_iq2xs[(aux32 >> (14*il+7)) & 127];
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[2][i] = dl * grid1[i] * (signs & kmask_iq2xs[i+0] ? -1.f : 1.f);
|
||||
reg[3][i] = dl * grid2[i] * (signs & kmask_iq2xs[i+4] ? -1.f : 1.f);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq3_s(device const block_iq3_s * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint8_t * qs = xb->qs + 8*ib32;
|
||||
device const uint8_t * signs = xb->signs + 4*ib32 + 2*il;
|
||||
const uint8_t qh = xb->qh[ib32] >> 4*il;
|
||||
const float dl = d * (1 + 2*((xb->scales[ib32/2] >> 4*(ib32%2)) & 0xf));
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+0] | ((qh << 8) & 256)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+1] | ((qh << 7) & 256)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i+0]);
|
||||
reg[1][i] = dl * grid2[i] * select(1, -1, signs[0] & kmask_iq2xs[i+4]);
|
||||
}
|
||||
grid1 = (constant uint8_t *)(iq3s_grid + (qs[4*il+2] | ((qh << 6) & 256)));
|
||||
grid2 = (constant uint8_t *)(iq3s_grid + (qs[4*il+3] | ((qh << 5) & 256)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[2][i] = dl * grid1[i] * select(1, -1, signs[1] & kmask_iq2xs[i+0]);
|
||||
reg[3][i] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i+4]);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq2_s(device const block_iq2_s * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const float d = xb->d;
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
|
||||
device const uint8_t * signs = qs + QK_K/8;
|
||||
const uint8_t qh = xb->qh[ib32] >> 4*il;
|
||||
const float dl = d * (0.5f + ((xb->scales[ib32] >> 4*il) & 0xf)) * 0.25f;
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq2s_grid + (qs[0] | ((qh << 8) & 0x300)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq2s_grid + (qs[1] | ((qh << 6) & 0x300)));
|
||||
for (int i = 0; i < 8; ++i) {
|
||||
reg[i/4+0][i%4] = dl * grid1[i] * select(1, -1, signs[0] & kmask_iq2xs[i]);
|
||||
reg[i/4+2][i%4] = dl * grid2[i] * select(1, -1, signs[1] & kmask_iq2xs[i]);
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq1_s(device const block_iq1_s * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
const float d = xb->d;
|
||||
device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
|
||||
device const uint16_t * qh = xb->qh;
|
||||
const float dl = d * (2*((qh[ib32] >> 12) & 7) + 1);
|
||||
const float ml = dl * (qh[ib32] & 0x8000 ? -1 - IQ1S_DELTA : -1 + IQ1S_DELTA);
|
||||
const uint16_t h = qh[ib32] >> 6*il;
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((h << 8) & 0x700)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((h << 5) & 0x700)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * (grid1[i] & 0xf) + ml;
|
||||
reg[1][i] = dl * (grid1[i] >> 4) + ml;
|
||||
reg[2][i] = dl * (grid2[i] & 0xf) + ml;
|
||||
reg[3][i] = dl * (grid2[i] >> 4) + ml;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq1_m(device const block_iq1_m * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
device const uint16_t * sc = (device const uint16_t *)xb->scales;
|
||||
|
||||
iq1m_scale_t scale;
|
||||
scale.u16 = (sc[0] >> 12) | ((sc[1] >> 8) & 0x00f0) | ((sc[2] >> 4) & 0x0f00) | (sc[3] & 0xf000);
|
||||
const float d = scale.f16;
|
||||
|
||||
device const uint8_t * qs = xb->qs + 4*ib32 + 2*il;
|
||||
device const uint8_t * qh = xb->qh + 2*ib32 + il;
|
||||
|
||||
const float dl = d * (2*((sc[ib32/2] >> (6*(ib32%2)+3*il)) & 7) + 1);
|
||||
const float ml1 = dl * (qh[0] & 0x08 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
|
||||
const float ml2 = dl * (qh[0] & 0x80 ? -1 - IQ1M_DELTA : -1 + IQ1M_DELTA);
|
||||
constant uint8_t * grid1 = (constant uint8_t *)(iq1s_grid_gpu + (qs[0] | ((qh[0] << 8) & 0x700)));
|
||||
constant uint8_t * grid2 = (constant uint8_t *)(iq1s_grid_gpu + (qs[1] | ((qh[0] << 4) & 0x700)));
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
reg[0][i] = dl * (grid1[i] & 0xf) + ml1;
|
||||
reg[1][i] = dl * (grid1[i] >> 4) + ml1;
|
||||
reg[2][i] = dl * (grid2[i] & 0xf) + ml2;
|
||||
reg[3][i] = dl * (grid2[i] >> 4) + ml2;
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq4_nl(device const block_iq4_nl * xb, short il, thread type4x4 & reg) {
|
||||
device const uint16_t * q4 = (device const uint16_t *)xb->qs;
|
||||
const float d = xb->d;
|
||||
uint32_t aux32;
|
||||
thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
aux32 = ((q4[2*i] | (q4[2*i+1] << 16)) >> 4*il) & 0x0f0f0f0f;
|
||||
reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
|
||||
reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
|
||||
reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
|
||||
reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
|
||||
}
|
||||
}
|
||||
|
||||
template <typename type4x4>
|
||||
void dequantize_iq4_xs(device const block_iq4_xs * xb, short il, thread type4x4 & reg) {
|
||||
// il is 0...15 for QK_K = 256 => index of block of 32 is il/2
|
||||
const int ib32 = il/2;
|
||||
il = il%2;
|
||||
// il = 0 or 1. il = 0 processes the first 16 quants in a block of 32, il = 1 the second 16
|
||||
device const uint32_t * q4 = (device const uint32_t *)xb->qs + 4*ib32;
|
||||
const int ls = ((xb->scales_l[ib32/2] >> 4*(ib32%2)) & 0xf) | (((xb->scales_h >> 2*ib32) & 3) << 4);
|
||||
const float d = (float)xb->d * (ls - 32);
|
||||
uint32_t aux32;
|
||||
thread const uint8_t * q8 = (thread const uint8_t *)&aux32;
|
||||
for (int i = 0; i < 4; ++i) {
|
||||
aux32 = (q4[i] >> 4*il) & 0x0f0f0f0f;
|
||||
reg[i][0] = d * kvalues_iq4nl_f[q8[0]];
|
||||
reg[i][1] = d * kvalues_iq4nl_f[q8[1]];
|
||||
reg[i][2] = d * kvalues_iq4nl_f[q8[2]];
|
||||
reg[i][3] = d * kvalues_iq4nl_f[q8[3]];
|
||||
}
|
||||
}
|
||||
|
||||
template<typename block_q, short nl, void (*dequantize_func)(device const block_q *, short, thread float4x4 &)>
|
||||
kernel void kernel_get_rows_q(
|
||||
device const void * src0,
|
||||
|
|
|
|||
|
|
@ -4,7 +4,7 @@
|
|||
#include "ggml-quants.h"
|
||||
#include "ggml-impl.h"
|
||||
#include "ggml-cpu-impl.h"
|
||||
|
||||
#include "ggml-cpu.h"
|
||||
|
||||
#include <math.h>
|
||||
#include <string.h>
|
||||
|
|
|
|||
|
|
@ -1296,13 +1296,6 @@ static ggml_backend_buffer_type_t ggml_backend_rpc_device_get_buffer_type(ggml_b
|
|||
UNUSED(dev);
|
||||
}
|
||||
|
||||
static ggml_backend_buffer_t ggml_backend_rpc_device_buffer_from_ptr(ggml_backend_dev_t dev, void * ptr, size_t size, size_t max_tensor_size) {
|
||||
return ggml_backend_cpu_buffer_from_ptr(ptr, size);
|
||||
|
||||
UNUSED(dev);
|
||||
UNUSED(max_tensor_size);
|
||||
}
|
||||
|
||||
static bool ggml_backend_rpc_device_supports_op(ggml_backend_dev_t dev, const struct ggml_tensor * op) {
|
||||
UNUSED(dev);
|
||||
UNUSED(op);
|
||||
|
|
@ -1328,7 +1321,7 @@ static const struct ggml_backend_device_i ggml_backend_rpc_device_i = {
|
|||
/* .init_backend = */ ggml_backend_rpc_device_init,
|
||||
/* .get_buffer_type = */ ggml_backend_rpc_device_get_buffer_type,
|
||||
/* .get_host_buffer_type = */ NULL,
|
||||
/* .buffer_from_host_ptr = */ ggml_backend_rpc_device_buffer_from_ptr,
|
||||
/* .buffer_from_host_ptr = */ NULL,
|
||||
/* .supports_op = */ ggml_backend_rpc_device_supports_op,
|
||||
/* .supports_buft = */ ggml_backend_rpc_device_supports_buft,
|
||||
/* .offload_op = */ NULL,
|
||||
|
|
|
|||
|
|
@ -1047,7 +1047,6 @@ static vk_buffer ggml_vk_create_buffer(vk_device& device, size_t size, vk::Memor
|
|||
return buf;
|
||||
}
|
||||
|
||||
buf->size = size;
|
||||
vk::BufferCreateInfo buffer_create_info{
|
||||
vk::BufferCreateFlags(),
|
||||
size,
|
||||
|
|
@ -1075,7 +1074,6 @@ static vk_buffer ggml_vk_create_buffer(vk_device& device, size_t size, vk::Memor
|
|||
|
||||
if (memory_type_index == UINT32_MAX) {
|
||||
device->device.destroyBuffer(buf->buffer);
|
||||
buf->size = 0;
|
||||
throw vk::OutOfDeviceMemoryError("No suitable memory type found");
|
||||
}
|
||||
|
||||
|
|
@ -1092,13 +1090,11 @@ static vk_buffer ggml_vk_create_buffer(vk_device& device, size_t size, vk::Memor
|
|||
}
|
||||
catch (const vk::SystemError& e) {
|
||||
device->device.destroyBuffer(buf->buffer);
|
||||
buf->size = 0;
|
||||
throw e;
|
||||
}
|
||||
} else {
|
||||
// Out of Host/Device memory, clean up buffer
|
||||
device->device.destroyBuffer(buf->buffer);
|
||||
buf->size = 0;
|
||||
throw e;
|
||||
}
|
||||
}
|
||||
|
|
@ -1111,6 +1107,7 @@ static vk_buffer ggml_vk_create_buffer(vk_device& device, size_t size, vk::Memor
|
|||
device->device.bindBufferMemory(buf->buffer, buf->device_memory, 0);
|
||||
|
||||
buf->device = device;
|
||||
buf->size = size;
|
||||
|
||||
#ifdef GGML_VULKAN_MEMORY_DEBUG
|
||||
device->memory_logger->log_allocation(buf, size);
|
||||
|
|
|
|||
15350
ggml/src/ggml.c
15350
ggml/src/ggml.c
File diff suppressed because it is too large
Load diff
Loading…
Add table
Add a link
Reference in a new issue