Add metal support (#80)

Author: mudler

.github/workflows/test.yaml

@@ -35,3 +35,22 @@ jobs:
       - name: Test
         run: |
           make test
+
+  macOS-metal-latest:
+    runs-on: macOS-latest
+
+    steps:
+      - name: Clone
+        uses: actions/checkout@v3
+        with: 
+          submodules: true
+
+      - name: Dependencies
+        run: |
+          brew update
+          brew install sdl2
+
+      - name: Test
+        run: |
+          make BUILD_TYPE=metal test
+          CGO_LDFLAGS="-framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders" LIBRARY_PATH=$PWD C_INCLUDE_PATH=$PWD go build -o testbuild ./examples

Makefile

@@ -143,6 +143,12 @@ ifeq ($(BUILD_TYPE),clblas)
 	EXTRA_TARGETS+=llama.cpp/ggml-opencl.o
 endif
 
+ifeq ($(BUILD_TYPE),metal)
+	EXTRA_LIBS=
+	CMAKE_ARGS+=-DLLAMA_METAL=ON
+	EXTRA_TARGETS+=llama.cpp/ggml-metal.o
+endif
+
 ifdef CLBLAST_DIR
 	CMAKE_ARGS+=-DCLBlast_dir=$(CLBLAST_DIR)
 endif
@@ -178,6 +184,9 @@ llama.cpp/ggml-cuda.o: llama.cpp/ggml.o
 llama.cpp/ggml-opencl.o: llama.cpp/ggml.o
 	cd build && cp -rf CMakeFiles/ggml.dir/ggml-opencl.cpp.o ../llama.cpp/ggml-opencl.o
 
+llama.cpp/ggml-metal.o: llama.cpp/ggml.o
+	cd build && cp -rf CMakeFiles/ggml.dir/ggml-metal.m.o ../llama.cpp/ggml-metal.o
+
 llama.cpp/llama.o:
 	$(MAKE) -C llama.cpp llama.o
 

README.md

@@ -58,6 +58,7 @@ BUILD_TYPE=clblas CLBLAS_DIR=... make libbinding.a
 CGO_LDFLAGS="-lOpenCL -lclblast -L/usr/local/lib64/" LIBRARY_PATH=$PWD C_INCLUDE_PATH=$PWD go run ./examples -m "/model/path/here" -t 14
 ```
 
+
 You should see something like this from the output when using the GPU:
 
 ```
@@ -66,6 +67,15 @@ ggml_opencl: selecting device: 'Intel(R) Graphics [0x46a6]'
 ggml_opencl: device FP16 support: true  
 ```
 
+## GPU offloading
+
+### Metal (Apple Silicon)
+
+```
+BUILD_TYPE=metal make libbinding.a
+CGO_LDFLAGS="-framework Foundation -framework Metal -framework MetalKit -framework MetalPerformanceShaders" LIBRARY_PATH=$PWD C_INCLUDE_PATH=$PWD go run ./examples -m "/model/path/here" -t 14
+```
+
 Enjoy!
 
 The documentation is available [here](https://pkg.go.dev/github.com/go-skynet/go-llama.cpp) and the full example code is [here](https://github.com/go-skynet/go-llama.cpp/blob/master/examples/main.go).