vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

metal: Log recommendedMaxWorkingSetSize on iOS 16+

Browse source
This commit is contained in:
Alex Azarov 2024-01-14 16:07:02 +01:00
parent 467a882fd2
commit e26d2242e0
1 changed files with 22 additions and 30 deletions
Download patch file Download diff file Expand all files Collapse all files

52
ggml-metal.m
View file

@ -369,8 +369,11 @@ static struct ggml_metal_context * ggml_metal_init(int n_cb) {
GGML_METAL_LOG_INFO("%s: simdgroup reduction support = %s\n", __func__, ctx->support_simdgroup_reduction ? "true" : "false");
GGML_METAL_LOG_INFO("%s: simdgroup matrix mul. support = %s\n", __func__, ctx->support_simdgroup_mm ? "true" : "false");
GGML_METAL_LOG_INFO("%s: hasUnifiedMemory = %s\n", __func__, ctx->device.hasUnifiedMemory ? "true" : "false");
if (@available(macOS 10.12, iOS 16.0, *)) {
GGML_METAL_LOG_INFO("%s: recommendedMaxWorkingSetSize = %8.2f MB\n", __func__, ctx->device.recommendedMaxWorkingSetSize / 1e6);
}
#if TARGET_OS_OSX
GGML_METAL_LOG_INFO("%s: recommendedMaxWorkingSetSize = %8.2f MB\n", __func__, ctx->device.recommendedMaxWorkingSetSize / 1e6);
if (ctx->device.maxTransferRate != 0) {
GGML_METAL_LOG_INFO("%s: maxTransferRate = %8.2f MB/s\n", __func__, ctx->device.maxTransferRate / 1e6);
} else {
@ -2369,6 +2372,22 @@ static const char * ggml_backend_metal_buffer_type_get_name(ggml_backend_buffer_
UNUSED(buft);
}
static void ggml_backend_metal_log_allocated_size(id<MTLDevice> device) {
if (@available(macOS 10.12, iOS 16.0, *)) {
GGML_METAL_LOG_INFO(", (%8.2f / %8.2f)",
device.currentAllocatedSize / 1024.0 / 1024.0,
device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
if (device.currentAllocatedSize > device.recommendedMaxWorkingSetSize) {
GGML_METAL_LOG_WARN("%s: warning: current allocated size is greater than the recommended max working set size\n", __func__);
} else {
GGML_METAL_LOG_INFO("\n");
}
} else {
GGML_METAL_LOG_INFO(", (%8.2f)\n", device.currentAllocatedSize / 1024.0 / 1024.0);
}
}
static ggml_backend_buffer_t ggml_backend_metal_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
struct ggml_backend_metal_buffer_context * ctx = malloc(sizeof(struct ggml_backend_metal_buffer_context));
@ -2401,22 +2420,7 @@ static ggml_backend_buffer_t ggml_backend_metal_buffer_type_alloc_buffer(ggml_ba
}
GGML_METAL_LOG_INFO("%s: allocated buffer, size = %8.2f MiB", __func__, size_aligned / 1024.0 / 1024.0);
#if TARGET_OS_OSX
GGML_METAL_LOG_INFO(", (%8.2f / %8.2f)",
device.currentAllocatedSize / 1024.0 / 1024.0,
device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
if (device.currentAllocatedSize > device.recommendedMaxWorkingSetSize) {
GGML_METAL_LOG_WARN("%s: warning: current allocated size is greater than the recommended max working set size\n", __func__);
} else {
GGML_METAL_LOG_INFO("\n");
}
#else
GGML_METAL_LOG_INFO(", (%8.2f)\n", device.currentAllocatedSize / 1024.0 / 1024.0);
#endif
ggml_backend_metal_log_allocated_size(device);
return ggml_backend_buffer_init(buft, ggml_backend_metal_buffer_i, ctx, size);
}
@ -2524,19 +2528,7 @@ ggml_backend_buffer_t ggml_backend_metal_buffer_from_ptr(void * data, size_t siz
}
}
#if TARGET_OS_OSX
GGML_METAL_LOG_INFO(", (%8.2f / %8.2f)",
device.currentAllocatedSize / 1024.0 / 1024.0,
device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
if (device.currentAllocatedSize > device.recommendedMaxWorkingSetSize) {
GGML_METAL_LOG_WARN("%s: warning: current allocated size is greater than the recommended max working set size\n", __func__);
} else {
GGML_METAL_LOG_INFO("\n");
}
#else
GGML_METAL_LOG_INFO(", (%8.2f)\n", device.currentAllocatedSize / 1024.0 / 1024.0);
#endif
ggml_backend_metal_log_allocated_size(device);
return ggml_backend_buffer_init(ggml_backend_metal_buffer_type(), ggml_backend_metal_buffer_i, ctx, size);
}