fix(ios): build metallib

.gitignore

@@ -70,3 +70,5 @@ android/keystores/debug.keystore
 lib/
 
 docs/API/.nojekyll
+
+*.metallib

cpp/ggml-metal.m

@@ -301,7 +301,7 @@ static void lm_ggml_metal_log(enum lm_ggml_log_level level, const char * format,
         const bool try_metallib = true;
 #endif
 
-        NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
+        NSString * path_lib = [bundle pathForResource:@"ggml-llama" ofType:@"metallib"];
         if (try_metallib && path_lib != nil) {
             // pre-compiled library found
             NSURL * libURL = [NSURL fileURLWithPath:path_lib];
@@ -329,9 +329,9 @@ static void lm_ggml_metal_log(enum lm_ggml_log_level level, const char * format,
             LM_GGML_METAL_LOG_INFO("%s: LM_GGML_METAL_PATH_RESOURCES = %s\n", __func__, path_resource ? [path_resource UTF8String] : "nil");
 
             if (path_resource) {
-                path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal-llama.metal"];
+                path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal.metal"];
             } else {
-                path_source = [bundle pathForResource:@"ggml-metal-llama" ofType:@"metal"];
+                path_source = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
             }
 
             if (path_source == nil) {

example/ios/.xcode.env.local

@@ -0,0 +1 @@
+export NODE_BINARY="/Users/jhen/.nvm/versions/node/v20.10.0/bin/node"

example/ios/Podfile.lock

@@ -1261,7 +1261,7 @@ SPEC CHECKSUMS:
   glog: 04b94705f318337d7ead9e6d17c019bd9b1f6b1b
   hermes-engine: 10fbd3f62405c41ea07e71973ea61e1878d07322
   libevent: 4049cae6c81cdb3654a443be001fb9bdceff7913
-  llama-rn: f4d2ce3b00ccb7b70760b2a07f7751bc366bc462
+  llama-rn: 1672eecf8df2623f19c4b7493c4642433213bc50
   RCT-Folly: 424b8c9a7a0b9ab2886ffe9c3b041ef628fd4fb1
   RCTRequired: a2faf4bad4e438ca37b2040cb8f7799baa065c18
   RCTTypeSafety: cb09f3e4747b6d18331a15eb05271de7441ca0b3

example/src/App.tsx

@@ -105,7 +105,7 @@ export default function App() {
     initLlama({
       model: file.uri,
       use_mlock: true,
-      n_gpu_layers: Platform.OS === 'ios' ? 0 : 0, // > 0: enable GPU
+      n_gpu_layers: Platform.OS === 'ios' ? 1 : 0, // > 0: enable GPU
       // embedding: true,
     })
       .then((ctx) => {

llama-rn.podspec

@@ -25,7 +25,7 @@ Pod::Spec.new do |s|
   s.source       = { :git => "https://github.com/mybigday/llama.rn.git", :tag => "#{s.version}" }
 
   s.source_files = "ios/**/*.{h,m,mm}", "cpp/**/*.{h,cpp,hpp,c,m,mm}"
-  s.resources = "cpp/**/*.{metal}"
+  s.resources = "cpp/**/*.{metallib}"
 
   s.dependency "React-Core"
 

scripts/bootstrap.sh

@@ -7,7 +7,6 @@ cp ./llama.cpp/ggml.h ./cpp/ggml.h
 cp ./llama.cpp/ggml.c ./cpp/ggml.c
 cp ./llama.cpp/ggml-metal.h ./cpp/ggml-metal.h
 cp ./llama.cpp/ggml-metal.m ./cpp/ggml-metal.m
-cp ./llama.cpp/ggml-metal.metal ./cpp/ggml-metal-llama.metal
 cp ./llama.cpp/ggml-alloc.h ./cpp/ggml-alloc.h
 cp ./llama.cpp/ggml-alloc.c ./cpp/ggml-alloc.c
 cp ./llama.cpp/ggml-backend.h ./cpp/ggml-backend.h
@@ -79,3 +78,12 @@ patch -p0 -d ./cpp < ./scripts/common.cpp.patch
 patch -p0 -d ./cpp < ./scripts/log.h.patch
 patch -p0 -d ./cpp < ./scripts/llama.cpp.patch
 patch -p0 -d ./cpp < ./scripts/ggml-metal.m.patch
+
+# Build metallib (~1.4MB)
+if [ "$OS" = "Darwin" ]; then
+  cd llama.cpp
+  xcrun --sdk iphoneos metal -c ggml-metal.metal -o ggml-metal.air
+  xcrun --sdk iphoneos metallib ggml-metal.air   -o ggml-llama.metallib
+  rm ggml-metal.air
+  cp ./ggml-llama.metallib ../cpp/ggml-llama.metallib
+fi

scripts/ggml-metal.m.patch

@@ -1,14 +1,11 @@
---- ggml-metal.m.orig	2024-03-20 12:34:08
-+++ ggml-metal.m	2024-03-20 12:35:42
-@@ -329,9 +329,9 @@
-             LM_GGML_METAL_LOG_INFO("%s: LM_GGML_METAL_PATH_RESOURCES = %s\n", __func__, path_resource ? [path_resource UTF8String] : "nil");
+--- ggml-metal.m.orig	2024-03-20 13:48:14
++++ ggml-metal.m	2024-03-20 13:46:42
+@@ -301,7 +301,7 @@
+         const bool try_metallib = true;
+ #endif
 
-             if (path_resource) {
--                path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal.metal"];
-+                path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal-llama.metal"];
-             } else {
--                path_source = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
-+                path_source = [bundle pathForResource:@"ggml-metal-llama" ofType:@"metal"];
-             }
-
-             if (path_source == nil) {
+-        NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
++        NSString * path_lib = [bundle pathForResource:@"ggml-llama" ofType:@"metallib"];
+         if (try_metallib && path_lib != nil) {
+             // pre-compiled library found
+             NSURL * libURL = [NSURL fileURLWithPath:path_lib];