include more stuff and fix cuda

Author: MarcusDunn

llama-cpp-sys-2/Cargo.toml

@@ -39,6 +39,8 @@ include = [
     "/llama.cpp/ggml-common.h",
     "/llama.cpp/ggml-cuda",
     "/llama.cpp/sgemm.h",
+    "/llama.cpp/ggml-cuda/*",
+    "/llama.cpp/ggml-cuda/template_instances/*",
 ]
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html

llama-cpp-sys-2/build.rs

@@ -463,10 +463,17 @@ fn compile_cuda(cx: &mut Build, cxx: &mut Build, featless_cxx: Build) -> &'stati
         .map(|f| f.unwrap())
         .filter(|entry| entry.file_name().to_string_lossy().ends_with(".cu"))
         .map(|entry| entry.path());
+    
+    let template_instances = read_dir(cuda_path.join("template-instances"))
+        .unwrap()
+        .map(|f| f.unwrap())
+        .filter(|entry| entry.file_name().to_string_lossy().ends_with(".cu"))
+        .map(|entry| entry.path());
 
     nvcc.include(cuda_path.as_path())
         .include(LLAMA_PATH.as_path())
         .files(cuda_sources)
+        .files(template_instances)
         .file(LLAMA_PATH.join("ggml-cuda.cu"))
         .compile(lib_name);