CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-28 08:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

metal : support build for iOS/tvOS (#3089)

Browse Source
This commit is contained in:
Jhen-Jie Hong
2023-09-09 16:46:04 +08:00
committed by GitHub
parent ec2a24fedf
commit 4fd5477955
1 changed files with 12 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
ggml-metal.m
View File

@@ -117,14 +117,17 @@ static NSString * const msl_library_source = @"see metal.metal";
struct ggml_metal_context * ggml_metal_init(int n_cb) { struct ggml_metal_context * ggml_metal_init(int n_cb) {
metal_printf("%s: allocating\n", __func__); metal_printf("%s: allocating\n", __func__);
// Show all the Metal device instances in the system
NSArray * devices = MTLCopyAllDevices();
id <MTLDevice> device; id <MTLDevice> device;
NSString * s; NSString * s;
#if TARGET_OS_OSX
// Show all the Metal device instances in the system
NSArray * devices = MTLCopyAllDevices();
for (device in devices) { for (device in devices) {
s = [device name]; s = [device name];
metal_printf("%s: found device: %s\n", __func__, [s UTF8String]); metal_printf("%s: found device: %s\n", __func__, [s UTF8String]);
} }
#endif
// Pick and show default Metal device // Pick and show default Metal device
device = MTLCreateSystemDefaultDevice(); device = MTLCreateSystemDefaultDevice();
@@ -247,13 +250,15 @@ struct ggml_metal_context * ggml_metal_init(int n_cb) {
#undef GGML_METAL_ADD_KERNEL #undef GGML_METAL_ADD_KERNEL
} }
metal_printf("%s: recommendedMaxWorkingSetSize = %8.2f MB\n", __func__, ctx->device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
metal_printf("%s: hasUnifiedMemory = %s\n", __func__, ctx->device.hasUnifiedMemory ? "true" : "false"); metal_printf("%s: hasUnifiedMemory = %s\n", __func__, ctx->device.hasUnifiedMemory ? "true" : "false");
#if TARGET_OS_OSX
metal_printf("%s: recommendedMaxWorkingSetSize = %8.2f MB\n", __func__, ctx->device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
if (ctx->device.maxTransferRate != 0) { if (ctx->device.maxTransferRate != 0) {
metal_printf("%s: maxTransferRate = %8.2f MB/s\n", __func__, ctx->device.maxTransferRate / 1024.0 / 1024.0); metal_printf("%s: maxTransferRate = %8.2f MB/s\n", __func__, ctx->device.maxTransferRate / 1024.0 / 1024.0);
} else { } else {
metal_printf("%s: maxTransferRate = built-in GPU\n", __func__); metal_printf("%s: maxTransferRate = built-in GPU\n", __func__);
} }
#endif
return ctx; return ctx;
} }
@@ -454,6 +459,7 @@ bool ggml_metal_add_buffer(
} }
} }
#if TARGET_OS_OSX
metal_printf(", (%8.2f / %8.2f)", metal_printf(", (%8.2f / %8.2f)",
ctx->device.currentAllocatedSize / 1024.0 / 1024.0, ctx->device.currentAllocatedSize / 1024.0 / 1024.0,
ctx->device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0); ctx->device.recommendedMaxWorkingSetSize / 1024.0 / 1024.0);
@@ -463,6 +469,9 @@ bool ggml_metal_add_buffer(
} else { } else {
metal_printf("\n"); metal_printf("\n");
} }
#else
metal_printf(", (%8.2f)\n", ctx->device.currentAllocatedSize / 1024.0 / 1024.0);
#endif
} }
return true; return true;