Working build.rs for apple metal

Author: SilasMarvin

llama-cpp-sys-2/build.rs

@@ -12,15 +12,13 @@ fn main() {
     }
 
     let mut ggml = cc::Build::new();
-    let mut ggml_cuda = if cublas_enabled { Some(cc::Build::new()) } else { None };
-    let mut ggml_metal= if cfg!(target_os = "macos") { Some(cc::Build::new()) } else { None };
     let mut llama_cpp = cc::Build::new();
 
     ggml.cpp(false);
     llama_cpp.cpp(true);
 
     // https://github.com/ggerganov/llama.cpp/blob/a836c8f534ab789b02da149fbdaf7735500bff74/Makefile#L364-L368
-    if let Some(ggml_cuda) = &mut ggml_cuda {
+    if cublas_enabled {
         for lib in [
             "cuda", "cublas", "culibos", "cudart", "cublasLt", "pthread", "dl", "rt",
         ] {
@@ -30,32 +28,34 @@ fn main() {
         println!("cargo:rustc-link-search=native=/usr/local/cuda/lib64");
 
         if cfg!(target_arch = "aarch64") {
-            ggml_cuda
-                .flag_if_supported("-mfp16-format=ieee")
+            ggml.flag_if_supported("-mfp16-format=ieee")
                 .flag_if_supported("-mno-unaligned-access");
             llama_cpp
                 .flag_if_supported("-mfp16-format=ieee")
                 .flag_if_supported("-mno-unaligned-access");
-            ggml_cuda
-                .flag_if_supported("-mfp16-format=ieee")
+            ggml.flag_if_supported("-mfp16-format=ieee")
                 .flag_if_supported("-mno-unaligned-access");
         }
 
-        ggml_cuda
-            .cuda(true)
+        ggml.cuda(true)
             .std("c++17")
             .flag("-arch=all")
             .file("llama.cpp/ggml-cuda.cu");
 
         ggml.define("GGML_USE_CUBLAS", None);
-        ggml_cuda.define("GGML_USE_CUBLAS", None);
+        ggml.define("GGML_USE_CUBLAS", None);
         llama_cpp.define("GGML_USE_CUBLAS", None);
     }
 
     // https://github.com/ggerganov/llama.cpp/blob/191221178f51b6e81122c5bda0fd79620e547d07/Makefile#L133-L141
     if cfg!(target_os = "macos") {
         assert!(!cublas_enabled, "CUBLAS is not supported on macOS");
 
+        println!("cargo:rustc-link-lib=framework=Metal");
+        println!("cargo:rustc-link-lib=framework=Foundation");
+        println!("cargo:rustc-link-lib=framework=MetalPerformanceShaders");
+        println!("cargo:rustc-link-lib=framework=MetalKit");
+
         llama_cpp.define("_DARWIN_C_SOURCE", None);
 
         // https://github.com/ggerganov/llama.cpp/blob/3c0d25c4756742ebf15ad44700fabc0700c638bd/Makefile#L340-L343
@@ -70,35 +70,21 @@ fn main() {
         println!("cargo:rustc-link-arg=framework=Foundation");
         println!("cargo:rustc-link-arg=framework=Metal");
         println!("cargo:rustc-link-arg=framework=MetalKit");
-    }
 
-    if let Some(ggml_metal) = &mut ggml_metal {
-        metal_hack(ggml_metal);
-        ggml_metal
-            .file("llama.cpp/ggml-metal")
-            .include("llama.cpp");
+        metal_hack(&mut ggml);
+        ggml.include("./llama.cpp/ggml-metal.h");
     }
 
     if cfg!(target_os = "dragonfly") {
         llama_cpp.define("__BSD_VISIBLE", None);
     }
 
-    if let Some(ggml_cuda) = ggml_cuda {
-        println!("compiling ggml-cuda");
-        ggml_cuda.compile("ggml-cuda");
-    }
-
-
-    if let Some(ggml_metal) = ggml_metal {
-        println!("compiling ggml-metal");
-        ggml_metal.compile("ggml-metal")
-    }
-
     if cfg!(target_os = "linux") {
         ggml.define("_GNU_SOURCE", None);
     }
 
     ggml.std("c17")
+        .include("./llama.cpp")
         .file("llama.cpp/ggml.c")
         .file("llama.cpp/ggml-alloc.c")
         .file("llama.cpp/ggml-backend.c")
@@ -136,7 +122,6 @@ fn main() {
         .expect("failed to write bindings to file");
 }
 
-
 // courtesy of https://github.com/rustformers/llm
 fn metal_hack(build: &mut cc::Build) {
     const GGML_METAL_METAL_PATH: &str = "llama.cpp/ggml-metal.metal";
@@ -174,4 +159,4 @@ fn metal_hack(build: &mut cc::Build) {
     };
 
     build.file(ggml_metal_path);
-}
+}