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

Metal support for Swift

Browse source
This commit is contained in:
kchro3 2023-09-07 22:36:02 -07:00
parent ebc96086af
commit 89a96fdb2b
2 changed files with 19 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

8
Package.swift
View file

@ -4,6 +4,7 @@ import PackageDescription
let package = Package( let package = Package(
name: "llama", name: "llama",
platforms: [.macOS(.v11)],
products: [ products: [
.library(name: "llama", targets: ["llama"]), .library(name: "llama", targets: ["llama"]),
], ],
@ -11,16 +12,19 @@ let package = Package(
.target( .target(
name: "llama", name: "llama",
path: ".", path: ".",
exclude: ["ggml-metal.metal"],
sources: [ sources: [
"ggml.c", "ggml.c",
"llama.cpp", "llama.cpp",
"ggml-alloc.c", "ggml-alloc.c",
"k_quants.c" "k_quants.c",
"ggml-metal.m",
], ],
publicHeadersPath: "spm-headers", publicHeadersPath: "spm-headers",
cSettings: [ cSettings: [
.unsafeFlags(["-Wno-shorten-64-to-32"]), .unsafeFlags(["-Wno-shorten-64-to-32"]),
.unsafeFlags(["-fno-objc-arc"]),
.define("GGML_SWIFT"),
.define("GGML_USE_METAL"),
.define("GGML_USE_K_QUANTS"), .define("GGML_USE_K_QUANTS"),
.define("GGML_USE_ACCELERATE") .define("GGML_USE_ACCELERATE")
], ],

18
ggml-metal.m
View file

@ -141,12 +141,22 @@ struct ggml_metal_context * ggml_metal_init(int n_cb) {
ctx->d_queue = dispatch_queue_create("llama.cpp", DISPATCH_QUEUE_CONCURRENT); ctx->d_queue = dispatch_queue_create("llama.cpp", DISPATCH_QUEUE_CONCURRENT);
#if 0 MTLCompileOptions* options = [MTLCompileOptions new];
// compile from source string and show compile log options.preprocessorMacros = @{ @"QK_K" : @(64) };
#ifdef GGML_SWIFT
// load the default.metallib file
{ {
NSError * error = nil; NSError * error = nil;
ctx->library = [ctx->device newLibraryWithSource:msl_library_source options:nil error:&error]; NSBundle * bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
NSString * llamaBundlePath = [bundle pathForResource:@"llama_llama" ofType:@"bundle"];
NSBundle * llamaBundle = [NSBundle bundleWithPath:llamaBundlePath];
NSString * libPath = [llamaBundle pathForResource:@"default" ofType:@"metallib"];
// Load the metallib file into a Metal library
ctx->library = [ctx->device newLibraryWithFile:libPath error:&error];
if (error) { if (error) {
metal_printf("%s: error: %s\n", __func__, [[error description] UTF8String]); metal_printf("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL; return NULL;
@ -171,8 +181,6 @@ struct ggml_metal_context * ggml_metal_init(int n_cb) {
} }
#ifdef GGML_QKK_64 #ifdef GGML_QKK_64
MTLCompileOptions* options = [MTLCompileOptions new];
options.preprocessorMacros = @{ @"QK_K" : @(64) };
ctx->library = [ctx->device newLibraryWithSource:src options:options error:&error]; ctx->library = [ctx->device newLibraryWithSource:src options:options error:&error];
#else #else
ctx->library = [ctx->device newLibraryWithSource:src options:nil error:&error]; ctx->library = [ctx->device newLibraryWithSource:src options:nil error:&error];