Skip to content

ggml: unify backend logging mechanism #9709

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 14 commits into from
Oct 3, 2024
Merged

ggml: unify backend logging mechanism #9709

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
90222ac
Add scaffolding for ggml logging macros
bandoti Oct 1, 2024
1f0c1a9
Metal backend now uses GGML logging
bandoti Oct 1, 2024
15817d0
Cuda backend now uses GGML logging
bandoti Oct 1, 2024
c6b6540
Cann backend now uses GGML logging
bandoti Oct 1, 2024
5e61442
Add enum tag to parameters
bandoti Oct 1, 2024
675b7ea
Use C memory allocation funcs
bandoti Oct 1, 2024
31c3b89
Fix compile error
bandoti Oct 1, 2024
a731280
Use GGML_LOG instead of GGML_PRINT
bandoti Oct 1, 2024
e2535a1
Rename llama_state to llama_logger_state
bandoti Oct 1, 2024
49d45c5
Prevent null format string
bandoti Oct 2, 2024
065a440
Fix whitespace
bandoti Oct 2, 2024
9d9a7ee
Merge remote-tracking branch 'origin/master' into ggml-logger-subsystem
bandoti Oct 3, 2024
19eb9ec
Remove log callbacks from ggml backends
bandoti Oct 3, 2024
0a35c3c
Remove cuda log statement
bandoti Oct 3, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 1 addition & 4 deletions ggml/include/ggml-backend.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -164,7 +164,7 @@ extern "C" {
GGML_API size_t ggml_backend_reg_dev_count(ggml_backend_reg_t reg);
GGML_API ggml_backend_dev_t ggml_backend_reg_dev_get(ggml_backend_reg_t reg, size_t index);
GGML_API void * ggml_backend_reg_get_proc_address(ggml_backend_reg_t reg, const char * name);
GGML_API void ggml_backend_reg_set_log_callback(ggml_backend_reg_t reg, ggml_log_callback log_callback, void * user_data);


// Functions that may be obtained using ggml_backend_reg_get_proc_address
typedef ggml_backend_buffer_type_t (*ggml_backend_split_buffer_type_t)(const float *);
Expand All @@ -184,9 +184,6 @@ extern "C" {
GGML_API ggml_backend_dev_t ggml_backend_dev_by_name(const char * name);
GGML_API ggml_backend_dev_t ggml_backend_dev_by_type(enum ggml_backend_dev_type type);

// Set the log callback for all registered backends
GGML_API void ggml_backend_set_log_callback(ggml_log_callback log_callback, void * user_data);

// Direct backend (stream) initialization
// = ggml_backend_dev_init(ggml_backend_dev_by_name(name), params)
GGML_API ggml_backend_t ggml_backend_init_by_name(const char * name, const char * params);
Expand Down
11 changes: 0 additions & 11 deletions ggml/include/ggml-cann.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,17 +116,6 @@ GGML_API void ggml_backend_cann_get_device_memory(int32_t device,
size_t* free,
size_t* total);

/**
* @brief Set the logging callback for GGML.
*
* This function sets the logging callback and user data for logging.
*
* @param log_callback The logging callback to set.
* @param user_data User data to pass to the logging callback.
*/
GGML_API void ggml_backend_cann_log_set_callback(ggml_log_callback log_callback,
void* user_data);

#ifdef __cplusplus
}
#endif
2 changes: 0 additions & 2 deletions ggml/include/ggml-cuda.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,8 +40,6 @@ GGML_API void ggml_backend_cuda_get_device_memory(int device, size_t * free, siz
GGML_API bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size);
GGML_API void ggml_backend_cuda_unregister_host_buffer(void * buffer);

GGML_API void ggml_backend_cuda_log_set_callback(ggml_log_callback log_callback, void * user_data);

GGML_API ggml_backend_reg_t ggml_backend_cuda_reg(void);

#ifdef __cplusplus
Expand Down
2 changes: 0 additions & 2 deletions ggml/include/ggml-metal.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,8 +39,6 @@ extern "C" {
// user-code should use only these functions
//

GGML_API void ggml_backend_metal_log_set_callback(ggml_log_callback log_callback, void * user_data);

GGML_API ggml_backend_t ggml_backend_metal_init(void);

GGML_API bool ggml_backend_is_metal(ggml_backend_t backend);
Expand Down
4 changes: 4 additions & 0 deletions ggml/include/ggml.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2167,6 +2167,10 @@ extern "C" {
typedef void (*ggml_opt_callback)(void * data, int accum_step, float * sched, bool * cancel);
typedef void (*ggml_log_callback)(enum ggml_log_level level, const char * text, void * user_data);

// Set callback for all future logging events.
// If this is not called, or NULL is supplied, everything is output on stderr.
GGML_API void ggml_log_set(ggml_log_callback log_callback, void * user_data);

// optimization parameters
//
// see ggml.c (ggml_opt_default_params) for default values
Expand Down
3 changes: 0 additions & 3 deletions ggml/src/ggml-backend-impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -215,9 +215,6 @@ extern "C" {
// (optional) get a pointer to a function in the backend
// backends can add custom functions that are not part of the standard ggml-backend interface
void * (*get_proc_address)(ggml_backend_reg_t reg, const char * name);

// (optional) set the log callback for the backend
void (*set_log_callback)(ggml_backend_reg_t reg, ggml_log_callback log_callback, void * user_data);
};

struct ggml_backend_reg {
Expand Down
14 changes: 0 additions & 14 deletions ggml/src/ggml-backend.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -505,12 +505,6 @@ void * ggml_backend_reg_get_proc_address(ggml_backend_reg_t reg, const char * na
return reg->iface.get_proc_address(reg, name);
}

void ggml_backend_reg_set_log_callback(ggml_backend_reg_t reg, ggml_log_callback log_callback, void * user_data) {
if (reg->iface.set_log_callback) {
reg->iface.set_log_callback(reg, log_callback, user_data);
}
}

// Backend registry

#ifdef GGML_USE_CUDA
Expand Down Expand Up @@ -614,13 +608,6 @@ ggml_backend_dev_t ggml_backend_dev_by_type(enum ggml_backend_dev_type type) {
return NULL;
}

void ggml_backend_set_log_callback(ggml_log_callback log_callback, void * user_data) {
for (size_t i = 0; i < ggml_backend_reg_count(); i++) {
ggml_backend_reg_t reg = ggml_backend_reg_get(i);
ggml_backend_reg_set_log_callback(reg, log_callback, user_data);
}
}

// Convenience functions
ggml_backend_t ggml_backend_init_by_name(const char * name, const char * params) {
ggml_backend_dev_t dev = ggml_backend_dev_by_name(name);
Expand Down Expand Up @@ -1161,7 +1148,6 @@ static const struct ggml_backend_reg_i ggml_backend_cpu_reg_i = {
/* .get_device_count = */ ggml_backend_cpu_reg_get_device_count,
/* .get_device = */ ggml_backend_cpu_reg_get_device,
/* .get_proc_address = */ NULL,
/* .set_log_callback = */ NULL,
};

ggml_backend_reg_t ggml_backend_cpu_reg(void) {
Expand Down
89 changes: 13 additions & 76 deletions ggml/src/ggml-cann.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,69 +39,6 @@

#include "ggml-common.h"

/**
* @brief Default logging callback for GGML.
*
* This function is the default logging callback that logs messages to stderr.
*
* @param level The log level.
* @param msg The log message.
* @param user_data User data passed to the callback.
*/
static void ggml_cann_default_log_callback(enum ggml_log_level level,
const char* msg, void* user_data) {
GGML_UNUSED(level);
GGML_UNUSED(user_data);
fprintf(stderr, "%s", msg);
}

ggml_log_callback ggml_cann_log_callback = ggml_cann_default_log_callback;
void* ggml_cann_log_user_data = NULL;

GGML_API void ggml_backend_cann_log_set_callback(ggml_log_callback log_callback,
void* user_data) {
ggml_cann_log_callback = log_callback;
ggml_cann_log_user_data = user_data;
}

#define GGML_CANN_LOG_INFO(...) ggml_cann_log(GGML_LOG_LEVEL_INFO, __VA_ARGS__)
#define GGML_CANN_LOG_WARN(...) ggml_cann_log(GGML_LOG_LEVEL_WARN, __VA_ARGS__)
#define GGML_CANN_LOG_ERROR(...) \
ggml_cann_log(GGML_LOG_LEVEL_ERROR, __VA_ARGS__)

GGML_ATTRIBUTE_FORMAT(2, 3)

/**
* @brief Log a message using the current logging callback.
*
* This function formats a log message and passes it to the current logging
* callback.
*
* @param level The log level.
* @param format The format string for the log message.
* @param ... The arguments for the format string.
*/
static void ggml_cann_log(enum ggml_log_level level, const char* format, ...) {
if (ggml_cann_log_callback != NULL) {
va_list args;
va_start(args, format);
char buffer[128];
int len = vsnprintf(buffer, 128, format, args);
if (len < 128) {
ggml_cann_log_callback(level, buffer, ggml_cann_log_user_data);
} else {
// vsnprintf adds a null terminator
std::vector<char> buffer2(len + 1);
va_end(args);
va_start(args, format);
vsnprintf(&buffer2[0], buffer2.size(), format, args);
ggml_cann_log_callback(level, buffer2.data(),
ggml_cann_log_user_data);
}
va_end(args);
}
}

/**
* @brief Handles CANN errors by printing an error message and aborting.
*
Expand All @@ -116,10 +53,10 @@ static void ggml_cann_log(enum ggml_log_level level, const char* format, ...) {
int32_t id = -1;
aclrtGetDevice(&id);

GGML_CANN_LOG_ERROR("CANN error: %s\n", msg);
GGML_CANN_LOG_ERROR(" current device: %d, in function %s at %s:%d\n", id, func,
GGML_LOG_ERROR("CANN error: %s\n", msg);
GGML_LOG_ERROR(" current device: %d, in function %s at %s:%d\n", id, func,
file, line);
GGML_CANN_LOG_ERROR(" %s\n", stmt);
GGML_LOG_ERROR(" %s\n", stmt);
// abort with GGML_ASSERT to get a stack trace
GGML_ABORT("CANN error");
}
Expand Down Expand Up @@ -165,7 +102,7 @@ static ggml_cann_device_info ggml_cann_init() {
aclError err = aclrtGetDeviceCount((uint32_t*)&info.device_count);

if (err != ACL_SUCCESS) {
GGML_CANN_LOG_ERROR("%s: failed to initialize CANN: %s\n",
GGML_LOG_ERROR("%s: failed to initialize CANN: %s\n",
__func__, aclGetRecentErrMsg());
return info;
}
Expand Down Expand Up @@ -315,7 +252,7 @@ struct ggml_cann_pool_leg : public ggml_cann_pool {
*actual_size = look_ahead_size;
pool_size += look_ahead_size;
#ifdef DEBUG_CANN_MALLOC
GGML_CANN_LOG_INFO(
GGML_LOG_INFO(
"%s[%d]: %d buffers, max_size = %u MB, pool_size = %u MB, "
"requested %u MB\n",
__func__, device, nnz, (uint32_t)(max_size / 1024 / 1024),
Expand Down Expand Up @@ -470,7 +407,7 @@ struct ggml_cann_pool_vmm : public ggml_cann_pool {
// add to the pool
pool_size += reserve_size;

// GGML_CANN_LOG_INFO("cann pool[%d]: size increased to %llu MB (
// GGML_LOG_INFO("cann pool[%d]: size increased to %llu MB (
// reserved %llu MB)\n",
// device, (unsigned long long) (pool_size/1024/1024),
// (unsigned long long) (reserve_size/1024/1024));
Expand All @@ -483,7 +420,7 @@ struct ggml_cann_pool_vmm : public ggml_cann_pool {
pool_used += size;

#ifdef DEBUG_CANN_MALLOC
GGML_CANN_LOG_INFO("cann pool[%d]: allocated %llu bytes at %llx\n", device,
GGML_LOG_INFO("cann pool[%d]: allocated %llu bytes at %llx\n", device,
(unsigned long long)size, (unsigned long long)ptr);
#endif
return ptr;
Expand All @@ -497,7 +434,7 @@ struct ggml_cann_pool_vmm : public ggml_cann_pool {
*/
void free(void* ptr, size_t size) override {
#ifdef DEBUG_CANN_MALLOC
GGML_CANN_LOG_INFO("cann pool[%d]: freed %llu bytes at %llx\n", device,
GGML_LOG_INFO("cann pool[%d]: freed %llu bytes at %llx\n", device,
(unsigned long long)size, (unsigned long long)ptr);
#endif

Expand Down Expand Up @@ -1095,7 +1032,7 @@ ggml_backend_cann_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft,
void* dev_ptr;
aclError err = aclrtMalloc(&dev_ptr, size, ACL_MEM_MALLOC_HUGE_FIRST);
if (err != ACL_SUCCESS) {
GGML_CANN_LOG_ERROR(
GGML_LOG_ERROR(
"%s: allocating %.2f MiB on device %d: aclrtMalloc failed: %s\n",
__func__, size / 1024.0 / 1024.0, buft_ctx->device,
aclGetRecentErrMsg());
Expand Down Expand Up @@ -1280,7 +1217,7 @@ static void * ggml_cann_host_malloc(size_t size) {
aclError err = aclrtMallocHost((void **) &hostPtr, size);
if (err != ACL_SUCCESS) {

GGML_CANN_LOG_WARN("%s: failed to allocate %.2f MiB of pinned memory: %s\n", __func__,
GGML_LOG_WARN("%s: failed to allocate %.2f MiB of pinned memory: %s\n", __func__,
size / 1024.0 / 1024.0, aclGetRecentErrMsg());
return nullptr;
}
Expand Down Expand Up @@ -1733,7 +1670,7 @@ static enum ggml_status ggml_backend_cann_graph_compute(
bool ok = ggml_cann_compute_forward(*cann_ctx, node);

if (!ok) {
GGML_CANN_LOG_ERROR("%s: error: op not supported %s (%s)\n", __func__,
GGML_LOG_ERROR("%s: error: op not supported %s (%s)\n", __func__,
node->name, ggml_op_name(node->op));
}
GGML_ASSERT(ok);
Expand Down Expand Up @@ -2043,13 +1980,13 @@ static ggml_guid_t ggml_backend_cann_guid() {
ggml_backend_t ggml_backend_cann_init(int32_t device) {
aclInit(nullptr);
if (device < 0 || device >= ggml_backend_cann_get_device_count()) {
GGML_CANN_LOG_ERROR("%s: error: invalid device %d\n", __func__, device);
GGML_LOG_ERROR("%s: error: invalid device %d\n", __func__, device);
return nullptr;
}

ggml_backend_cann_context* ctx = new ggml_backend_cann_context(device);
if (ctx == nullptr) {
GGML_CANN_LOG_ERROR("%s: error: failed to allocate context\n", __func__);
GGML_LOG_ERROR("%s: error: failed to allocate context\n", __func__);
return nullptr;
}
ggml_cann_set_device(ctx->device);
Expand Down
Loading
Loading