feat(ios): still enable metal backend on simulator & build metallib for simulator

cpp/ggml-backend-reg.cpp

@@ -35,14 +35,9 @@
 #endif
 
 #ifdef LM_GGML_USE_METAL
-#include <TargetConditionals.h>
-
-#if !TARGET_OS_SIMULATOR
 #include "ggml-metal.h"
 #endif
 
-#endif
-
 #ifdef LM_GGML_USE_SYCL
 #include "ggml-sycl.h"
 #endif
@@ -162,12 +157,8 @@ struct lm_ggml_backend_registry {
         register_backend(lm_ggml_backend_cuda_reg());
 #endif
 #ifdef LM_GGML_USE_METAL
-
-#if !TARGET_OS_SIMULATOR
         register_backend(lm_ggml_backend_metal_reg());
 #endif
-
-#endif
 #ifdef LM_GGML_USE_SYCL
         register_backend(lm_ggml_backend_sycl_reg());
 #endif

cpp/ggml-metal.m

@@ -530,7 +530,11 @@ @implementation LMGGMLMetalClass
         const bool try_metallib = true;
 #endif
 
+#if TARGET_OS_SIMULATOR
+        NSString * path_lib = [bundle pathForResource:@"ggml-llama-sim" ofType:@"metallib"];
+#else
         NSString * path_lib = [bundle pathForResource:@"ggml-llama" ofType:@"metallib"];
+#endif
         if (path_lib == nil) {
             // Try to find the resource in the directory where the current binary located.
             NSString * current_binary = [[NSProcessInfo processInfo] arguments][0];

ios/RNLlamaContext.mm

@@ -116,8 +116,14 @@ + (instancetype)initWithParams:(NSDictionary *)params onProgress:(void (^)(unsig
             if (pipeline == nil) {
                 reasonNoMetal = [error localizedDescription];
             } else {
+#if TARGET_OS_SIMULATOR
+                defaultParams.n_gpu_layers = 0;
+                isMetalEnabled = false;
+                reasonNoMetal = @"Metal is not supported on simulator";
+#else
                 defaultParams.n_gpu_layers = [params[@"n_gpu_layers"] intValue];
                 isMetalEnabled = true;
+#endif
             }
         }
         device = nil;

scripts/bootstrap.sh

@@ -229,7 +229,6 @@ patch -p0 -d ./cpp < ./scripts/patches/common.h.patch
 patch -p0 -d ./cpp < ./scripts/patches/common.cpp.patch
 patch -p0 -d ./cpp < ./scripts/patches/log.cpp.patch
 patch -p0 -d ./cpp < ./scripts/patches/ggml-metal.m.patch
-patch -p0 -d ./cpp < ./scripts/patches/ggml-backend-reg.cpp.patch
 patch -p0 -d ./cpp < ./scripts/patches/ggml.c.patch
 patch -p0 -d ./cpp < ./scripts/patches/ggml-quants.c.patch
 patch -p0 -d ./cpp < ./scripts/patches/llama-mmap.cpp.patch
@@ -246,6 +245,11 @@ if [ "$OS" = "Darwin" ]; then
   rm ggml-metal.air
   mv ./ggml-llama.metallib ../../../../cpp/ggml-llama.metallib
 
+  xcrun --sdk iphonesimulator metal -c ggml-metal.metal -o ggml-metal.air -DGGML_METAL_USE_BF16=1
+  xcrun --sdk iphonesimulator metallib ggml-metal.air   -o ggml-llama.metallib
+  rm ggml-metal.air
+  mv ./ggml-llama.metallib ../../../../cpp/ggml-llama-sim.metallib
+
   cd -
 
   # Generate .xcode.env.local in iOS example

scripts/build-ios.sh

@@ -43,7 +43,11 @@ function build_framework() {
   # Copy headers and metallib
   cp_headers $4
   # TODO: May need to re-build metallib for tvOS
-  cp ../cpp/ggml-llama.metallib ../ios/rnllama.xcframework/$4/rnllama.framework/ggml-llama.metallib
+  if [ "$4" == "ios-arm64_x86_64-simulator" ] || [ "$4" == "tvos-arm64_x86_64-simulator" ]; then
+    cp ../cpp/ggml-llama-sim.metallib ../ios/rnllama.xcframework/$4/rnllama.framework/ggml-llama-sim.metallib
+  else
+    cp ../cpp/ggml-llama.metallib ../ios/rnllama.xcframework/$4/rnllama.framework/ggml-llama.metallib
+  fi
 
   rm -rf ./*
   cd ..

scripts/patches/ggml-metal.m.patch

@@ -1,11 +1,15 @@
---- ggml-metal.m.orig	2024-12-20 13:36:22
-+++ ggml-metal.m	2024-12-20 13:37:17
-@@ -509,7 +509,7 @@
+--- ggml-metal.m.orig	2025-02-06 12:19:16
++++ ggml-metal.m	2025-02-06 12:18:32
+@@ -530,7 +530,11 @@
          const bool try_metallib = true;
  #endif
-
+ 
 -        NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
++#if TARGET_OS_SIMULATOR
++        NSString * path_lib = [bundle pathForResource:@"ggml-llama-sim" ofType:@"metallib"];
++#else
 +        NSString * path_lib = [bundle pathForResource:@"ggml-llama" ofType:@"metallib"];
++#endif
          if (path_lib == nil) {
              // Try to find the resource in the directory where the current binary located.
              NSString * current_binary = [[NSProcessInfo processInfo] arguments][0];