Reorganize driver host tests, fix bugs around pointer host code (#492)

Cargo.lock

@@ -447,21 +447,6 @@ dependencies = [
  "syn 2.0.89",
 ]
 
-[[package]]
-name = "cuda_tests"
-version = "0.0.0"
-dependencies = [
- "cuda_macros",
- "cuda_types",
- "dtor 0.0.6",
- "lazy_static",
- "lz4-sys",
- "num_enum",
- "paste",
- "rustc-hash 1.1.0",
- "tempfile",
-]
-
 [[package]]
 name = "cuda_types"
 version = "0.0.0"
@@ -3729,13 +3714,13 @@ dependencies = [
  "blake3",
  "comgr",
  "cuda_macros",
- "cuda_tests",
  "cuda_types",
  "dark_api",
  "dtor 0.0.7",
  "hip_runtime-sys",
  "lazy_static",
  "libc",
+ "libloading",
  "lz4-sys",
  "num_enum",
  "paste",

Cargo.toml

@@ -6,7 +6,6 @@ members = [
     "comgr",
     "cuda_macros",
     "cuda_types",
-    "cuda_tests",
     "dark_api",
     "detours-sys",
     "ext/amd_comgr-sys",

cuda_macros/src/lib.rs

@@ -2,7 +2,7 @@ extern crate proc_macro;
 
 use proc_macro::TokenStream;
 use proc_macro2::Span;
-use quote::{quote, ToTokens};
+use quote::{format_ident, quote, ToTokens};
 use rustc_hash::FxHashMap;
 use std::iter;
 use syn::parse::{Parse, ParseStream};
@@ -10,7 +10,7 @@ use syn::punctuated::Punctuated;
 use syn::visit_mut::VisitMut;
 use syn::{
     bracketed, parse_macro_input, File, ForeignItem, ForeignItemFn, Ident, Item, Path, Signature,
-    Token, token
+    Token
 };
 
 const CUDA_RS: &'static str = include_str! {"cuda.rs"};
@@ -309,39 +309,22 @@ fn join(
     
 }
 
-#[proc_macro]
-pub fn generate_api_macro(input: TokenStream) -> TokenStream {
-    let input = parse_macro_input!(input as ApiMacroInput);
-    let ApiMacroInput(_, trait_name, _, type_name, _, macro_name) = input;
-    let expanded = quote! {
-        struct #type_name;
-        macro_rules! #macro_name {
-            ($($abi:literal fn $fn_name:ident( $( $arg_id:ident : $arg_type:ty ),* ) -> $ret_type:ty;)*) => {
-                impl #trait_name for #type_name {
-                    fn new() -> Self { Self }
-                    $(
-                        #[inline(always)]
-                        fn $fn_name(&self, $( $arg_id : $arg_type ),* ) -> $ret_type {
-                            unsafe { super::$fn_name( $( $arg_id ),* ) }
-                        }
-                    )*
-                }
-            };
+#[proc_macro_attribute]
+pub fn test_cuda(_attr: TokenStream, item: TokenStream) -> TokenStream {
+    let fn_ = parse_macro_input!(item as syn::ItemFn);
+    let cuda_fn = format_ident!("{}{}", fn_.sig.ident, "_nvidia");
+    let zluda_fn = format_ident!("{}{}", fn_.sig.ident, "_amdgpu");
+    let fn_name = fn_.sig.ident.clone();
+    quote! { 
+        #[test]
+        fn #cuda_fn() {
+            unsafe { #fn_name(<crate::tests::Cuda>::new()) }
         }
-    };
-    TokenStream::from(expanded)
-}
-#[allow(dead_code)]
-struct ApiMacroInput(token::Impl, Path, token::For, Path, token::Use, Ident);
-impl syn::parse::Parse for ApiMacroInput {
-    fn parse(input: syn::parse::ParseStream) -> syn::Result<Self> {
-        Ok(ApiMacroInput(
-            input.parse()?, // impl 
-            input.parse()?, // trait
-            input.parse()?, // for
-            input.parse()?, // type
-            input.parse()?, // using
-            input.parse()?  // macro
-        ))
-    }
-}
+        #[test]
+        fn #zluda_fn() {
+            unsafe { #fn_name(<crate::tests::Zluda>::new()) }
+        }
+        
+        #fn_
+    }.into()
+}

cuda_tests/Cargo.toml

@@ -1,16 +0,0 @@
-[package]
-name = "cuda_tests"
-version = "0.0.0"
-authors = ["Andrzej Janik <vosen@vosen.pl>"]
-edition = "2021"
-
-[dependencies]
-cuda_types = { path = "../cuda_types" }
-cuda_macros = { path = "../cuda_macros" }
-lazy_static = "1.4"
-num_enum = "0.4"
-lz4-sys = "1.9"
-tempfile = "3"
-paste = "1.0"
-rustc-hash = "1.1"
-dtor = "0.0.6"

cuda_tests/src/cuda/mod.rs

@@ -1,8 +0,0 @@
-use crate::api_trait;
-use cuda_types::cuda::*;
-
-cuda_macros::cuda_function_declarations!(api_trait);
-
-pub unsafe fn init_check<T: Api>(api: T) {
-    assert_eq!(api.cuInit(0), CUresult::SUCCESS);
-}

cuda_tests/src/lib.rs

@@ -1,24 +0,0 @@
-#[macro_export]
-macro_rules! api_trait {
-    ($($abi:literal fn $fn_name:ident( $($arg_id:ident : $arg_type:ty),* ) -> $ret_type:ty;)*) => {
-        pub trait Api {
-            fn new() -> Self;
-            $(fn $fn_name(&self, $( $arg_id : $arg_type ),* ) -> $ret_type;)*
-        }
-    };
-}
-
-#[macro_export]
-macro_rules! api_test {
-    ($func:ident, $type:ty) => {
-        paste::paste! {
-            #[test]
-            #[allow(non_snake_case)]
-            fn [<$func _test>]() {
-                unsafe { $func::<$type>(<$type>::new()) }
-            }
-        }
-    };
-}
-
-pub mod cuda;

zluda/Cargo.toml

@@ -27,7 +27,6 @@ zluda_common = { path = "../zluda_common" }
 blake3 = "1.8.2"
 serde = "1.0.219"
 serde_json = "1.0.142"
-cuda_tests = { path = "../cuda_tests" }
 
 [target.'cfg(windows)'.dependencies]
 winapi = { version = "0.3", features = ["heapapi", "std"] }
@@ -39,6 +38,9 @@ dtor = "0.0.7"
 [build-dependencies]
 vergen-gix = "1.0.9"
 
+[dev-dependencies]
+libloading = "0.8"
+
 [package.metadata.zluda]
 linux_symlinks = [
     "libcuda.so",

zluda/src/impl/driver.rs

@@ -528,8 +528,15 @@ pub(crate) unsafe fn launch_kernel_ex(
 #[cfg(test)]
 mod tests {
     use crate::r#impl::driver::AllocationInfo;
+    use crate::tests::CudaApi;
+    use cuda_macros::test_cuda;
     use cuda_types::cuda::CUcontext;
 
+    #[test_cuda]
+    fn init(api: impl CudaApi) {
+        api.cuInit(0);
+    }
+
     #[test]
     fn get_allocation() {
         let ctx1 = CUcontext(0x1234 as _);

zluda/src/impl/pointer.rs

@@ -27,12 +27,8 @@ pub(crate) unsafe fn get_attribute(
     }
     match attribute {
         hipPointer_attribute::HIP_POINTER_ATTRIBUTE_CONTEXT => {
-            let globals = driver::global_state()?;
-            let allocations = globals.allocations.lock().map_err(|_| CUerror::UNKNOWN)?;
-            let (_, alloc) = allocations
-                .get_offset_and_info(ptr.0 as usize)
-                .ok_or(CUerror::INVALID_VALUE)?;
-            unsafe { *(data.cast()) = alloc.context };
+            let context = get_context(ptr)?;
+            unsafe { *(data.cast()) = context };
             Ok(())
         }
         hipPointer_attribute::HIP_POINTER_ATTRIBUTE_MEMORY_TYPE => {
@@ -48,6 +44,15 @@ pub(crate) unsafe fn get_attribute(
     }
 }
 
+fn get_context(ptr: hipDeviceptr_t) -> Result<CUcontext, CUerror> {
+    let globals = driver::global_state()?;
+    let allocations = globals.allocations.lock().map_err(|_| CUerror::UNKNOWN)?;
+    let (_, alloc) = allocations
+        .get_offset_and_info(ptr.0 as usize)
+        .ok_or(CUerror::INVALID_VALUE)?;
+    Ok(alloc.context)
+}
+
 pub(crate) unsafe fn get_attributes(
     num_attributes: ::core::ffi::c_uint,
     attributes: &mut hipPointer_attribute,
@@ -59,15 +64,102 @@ pub(crate) unsafe fn get_attributes(
     let data = std::slice::from_raw_parts_mut(data, num_attributes as usize);
     for (attr, data_ptr) in attributes.iter().copied().zip(data.iter().copied()) {
         match attr {
+            hipPointer_attribute::HIP_POINTER_ATTRIBUTE_HOST_POINTER => {
+                if (*(data_ptr.cast::<hipDeviceptr_t>())).0.is_null() {
+                    *(data_ptr.cast::<hipDeviceptr_t>()) = ptr;
+                }
+            }
             hipPointer_attribute::HIP_POINTER_ATTRIBUTE_CONTEXT => {
-                get_attribute(data_ptr, attr, ptr).ok();
+                *(data_ptr.cast::<CUcontext>()) =
+                    get_context(ptr).unwrap_or(CUcontext(ptr::null_mut()));
             }
             hipPointer_attribute::HIP_POINTER_ATTRIBUTE_MEMORY_TYPE => {
                 *(data_ptr.cast::<CUmemorytype>()) =
-                    to_cu_memory_type(*data_ptr.cast::<hipMemoryType>())?;
+                    to_cu_memory_type(*data_ptr.cast::<hipMemoryType>()).unwrap_or(CUmemorytype(0));
             }
             _ => {}
         }
     }
     Ok(())
 }
+
+#[cfg(test)]
+mod tests {
+    use crate::tests::CudaApi;
+    use cuda_macros::test_cuda;
+    use cuda_types::cuda::*;
+    use std::{ffi::c_void, mem, ptr};
+
+    #[test_cuda]
+    pub unsafe fn unknown_ptr_attribute(api: impl CudaApi) {
+        api.cuInit(0);
+        api.cuCtxCreate_v2(&mut mem::zeroed(), 0, 0);
+        let mut ctx = mem::zeroed::<CUcontext>();
+        assert_eq!(
+            CUresult::ERROR_INVALID_VALUE,
+            api.cuPointerGetAttribute_unchecked(
+                std::ptr::from_mut(&mut ctx).cast(),
+                CUpointer_attribute::CU_POINTER_ATTRIBUTE_CONTEXT,
+                CUdeviceptr_v2(0xDEAD as _)
+            )
+        );
+    }
+
+    #[test_cuda]
+    pub unsafe fn unknown_ptr_attributes(api: impl CudaApi) {
+        api.cuInit(0);
+        api.cuCtxCreate_v2(&mut mem::zeroed(), 0, 0);
+        let mut mem_type = mem::zeroed::<CUmemorytype>();
+        let mut host_ptr = mem::zeroed::<*mut c_void>();
+        let mut dev_ptr = mem::zeroed::<*mut c_void>();
+        let mut is_managed = mem::zeroed::<bool>();
+        let mut ordinal = mem::zeroed::<i32>();
+        let mut attrs = [
+            CUpointer_attribute::CU_POINTER_ATTRIBUTE_MEMORY_TYPE,
+            CUpointer_attribute::CU_POINTER_ATTRIBUTE_DEVICE_POINTER,
+            CUpointer_attribute::CU_POINTER_ATTRIBUTE_HOST_POINTER,
+            CUpointer_attribute::CU_POINTER_ATTRIBUTE_IS_MANAGED,
+            CUpointer_attribute::CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL,
+        ];
+        let mut values = [
+            std::ptr::from_mut(&mut mem_type).cast::<c_void>(),
+            std::ptr::from_mut(&mut dev_ptr).cast(),
+            std::ptr::from_mut(&mut host_ptr).cast(),
+            std::ptr::from_mut(&mut is_managed).cast(),
+            std::ptr::from_mut(&mut ordinal).cast(),
+        ];
+        assert_eq!(
+            CUresult::SUCCESS,
+            api.cuPointerGetAttributes_unchecked(
+                attrs.len() as u32,
+                attrs.as_mut_ptr(),
+                values.as_mut_ptr(),
+                CUdeviceptr_v2(0xDEAD as _)
+            )
+        );
+        assert_eq!(mem_type, CUmemorytype(0));
+        assert_eq!(host_ptr, 0xDEAD as _);
+        assert_eq!(dev_ptr, ptr::null_mut());
+        assert_eq!(is_managed, false);
+        assert_eq!(ordinal, -2);
+    }
+
+    #[test_cuda]
+    pub unsafe fn unknown_ptr_attributes_no_context(api: impl CudaApi) {
+        api.cuInit(0);
+        api.cuCtxCreate_v2(&mut mem::zeroed(), 0, 0);
+        let mut context = CUcontext(1 as _);
+        let mut attrs = [CUpointer_attribute::CU_POINTER_ATTRIBUTE_CONTEXT];
+        let mut values = [std::ptr::from_mut(&mut context).cast::<c_void>()];
+        assert_eq!(
+            CUresult::SUCCESS,
+            api.cuPointerGetAttributes_unchecked(
+                attrs.len() as u32,
+                attrs.as_mut_ptr(),
+                values.as_mut_ptr(),
+                CUdeviceptr_v2(0xDEAD as _)
+            )
+        );
+        assert_eq!(context, CUcontext(ptr::null_mut()));
+    }
+}

zluda/src/tests.rs

@@ -0,0 +1,51 @@
+pub(crate) struct Zluda;
+
+pub(crate) struct Cuda(libloading::Library);
+
+impl Cuda {
+    #[cfg(not(windows))]
+    const CUDA_PATH: &'static str = "/usr/lib/x86_64-linux-gnu/libcuda.so.1";
+    #[cfg(windows)]
+    const CUDA_PATH: &'static str = "C:\\Windows\\System32\\nvcuda.dll";
+
+    fn load() -> Self {
+        unsafe { Self(libloading::Library::new(Self::CUDA_PATH).unwrap()) }
+    }
+}
+
+macro_rules! implemented_test {
+    ($($abi:literal fn $fn_name:ident( $( $arg_id:ident : $arg_type:ty ),* ) -> $ret_type:ty;)* ) => {
+        pub(crate) trait CudaApi {
+            fn new() -> Self;
+            $(
+                #[allow(non_snake_case, dead_code)]
+                fn $fn_name(&self, $( $arg_id : $arg_type ),* ) {
+                    paste::paste!{ self.[< $fn_name _unchecked >]( $( $arg_id ),* ) }.unwrap()
+                }
+                paste::paste!{ #[allow(non_snake_case, dead_code)] fn [< $fn_name _unchecked>](&self, $( $arg_id : $arg_type ),* ) -> $ret_type; }
+            )*
+        }
+
+
+
+        impl CudaApi for Cuda {
+            fn new() -> Self { Self::load() }
+            $(
+                paste::paste!{ fn [< $fn_name _unchecked >](&self, $( $arg_id : $arg_type ),* )  -> $ret_type {
+                    let func = unsafe { self.0.get::<unsafe extern $abi fn ( $( $arg_type ),* ) -> $ret_type>(concat!(stringify!($fn_name), "\0").as_bytes()) }.unwrap();
+                    unsafe { (func)( $( $arg_id ),* ) }
+                }}
+            )*
+        }
+
+        impl CudaApi for Zluda {
+            fn new() -> Self { Self }
+            $(
+                paste::paste!{ fn [< $fn_name _unchecked >](&self, $( $arg_id : $arg_type ),* )  -> $ret_type {
+                    unsafe { super::$fn_name( $( $arg_id ),* ) }
+                }}
+            )*
+        }
+    };
+}
+cuda_macros::cuda_function_declarations!(implemented_test);

zluda/src/tests/mod.rs

@@ -1,8 +0,0 @@
-use cuda_macros::generate_api_macro;
-use cuda_tests::api_test;
-use cuda_tests::cuda::*;
-
-generate_api_macro!(impl cuda_tests::cuda::Api for TestApi use implemented_test);
-cuda_macros::cuda_function_declarations!(implemented_test);
-
-api_test!(init_check, TestApi);