Fixes #606: Can now use contiguous containers as sources and destinat…

…ions of memory copying functions
eyalroz · Apr 29, 2024 · 1b10c66 · 1b10c66
1 parent 4d9870a
commit 1b10c66
Show file tree

Hide file tree

Showing 6 changed files with 26 additions and 20 deletions.
diff --git a/examples/modified_cuda_samples/vectorAdd/vectorAdd.cu b/examples/modified_cuda_samples/vectorAdd/vectorAdd.cu
@@ -46,8 +46,8 @@ int main()
 	auto d_B = cuda::memory::make_unique_span<float>(device, numElements);
 	auto d_C = cuda::memory::make_unique_span<float>(device, numElements);
 
-	cuda::memory::copy(d_A, h_A.data());
-	cuda::memory::copy(d_B, h_B.data());
+	cuda::memory::copy(d_A, h_A);
+	cuda::memory::copy(d_B, h_B);
 
 	auto launch_config = cuda::launch_config_builder()
 		.overall_size(numElements)
@@ -63,7 +63,7 @@ int main()
 		d_A.data(), d_B.data(), d_C.data(), numElements
 	);
 
-	cuda::memory::copy(h_C.data(), d_C);
+	cuda::memory::copy(h_C, d_C);
 
 	// Verify that the result vector is correct
 	for (int i = 0; i < numElements; ++i) {

diff --git a/examples/modified_cuda_samples/vectorAddMMAP/vectorAddMMAP.cpp b/examples/modified_cuda_samples/vectorAddMMAP/vectorAddMMAP.cpp
@@ -278,8 +278,8 @@ int main()
 	auto d_B_sp = d_B.as_requested().as_span<float>();
 	auto d_C_sp = d_C.as_requested().as_span<float>();
 
-	cuda::memory::copy(d_A_sp, h_A.data());
-	cuda::memory::copy(d_B_sp, h_B.data());
+	cuda::memory::copy(d_A_sp, h_A);
+	cuda::memory::copy(d_B_sp, h_B);
 
 	// Launch the Vector Add CUDA Kernel
 	auto launch_config = cuda::launch_config_builder()
@@ -295,7 +295,7 @@ int main()
 		d_A_sp.data(), d_B_sp.data(), d_C_sp.data(), num_elements
 	);
 
-	cuda::memory::copy(h_C.data(), d_C_sp);
+	cuda::memory::copy(h_C, d_C_sp);
 
 //	std::cout << "Checking results...\n\n";
 

diff --git a/examples/modified_cuda_samples/vectorAdd_nvrtc/vectorAdd_nvrtc.cpp b/examples/modified_cuda_samples/vectorAdd_nvrtc/vectorAdd_nvrtc.cpp
@@ -70,8 +70,8 @@ int main(void)
 	auto d_B = cuda::memory::make_unique_span<float>(device, numElements);
 	auto d_C = cuda::memory::make_unique_span<float>(device, numElements);
 
-	cuda::memory::copy(d_A, h_A.data());
-	cuda::memory::copy(d_B, h_B.data());
+	cuda::memory::copy(d_A, h_A);
+	cuda::memory::copy(d_B, h_B);
 
 	auto launch_config = cuda::launch_config_builder()
 		.overall_size(numElements)
@@ -87,7 +87,7 @@ int main(void)
 		d_A.get(), d_B.get(), d_C.get(), numElements
 	);
 
-	cuda::memory::copy(h_C.data(), d_C);
+	cuda::memory::copy(h_C, d_C);
 
 	// Verify that the result vector is correct
 	for (int i = 0; i < numElements; ++i) {

diff --git a/examples/modified_cuda_samples/vectorAdd_ptx/vectorAdd_ptx.cpp b/examples/modified_cuda_samples/vectorAdd_ptx/vectorAdd_ptx.cpp
@@ -110,8 +110,8 @@ int main(void)
 	auto d_B = cuda::memory::make_unique_span<float>(device, numElements);
 	auto d_C = cuda::memory::make_unique_span<float>(device, numElements);
 
-	cuda::memory::copy(d_A, h_A.data());
-	cuda::memory::copy(d_B, h_B.data());
+	cuda::memory::copy(d_A, h_A);
+	cuda::memory::copy(d_B, h_B);
 
 	auto launch_config = cuda::launch_config_builder()
 		.overall_size(numElements)
@@ -127,7 +127,7 @@ int main(void)
 		d_A.get(), d_B.get(), d_C.get(), numElements
 	);
 
-	cuda::memory::copy(h_C.data(), d_C);
+	cuda::memory::copy(h_C, d_C);
 
 	// Verify that the result vector is correct
 	for (int i = 0; i < numElements; ++i) {

diff --git a/examples/modified_cuda_samples/vectorAdd_unique_regions/vectorAdd_unique_regions.cu b/examples/modified_cuda_samples/vectorAdd_unique_regions/vectorAdd_unique_regions.cu
@@ -51,8 +51,8 @@ int main()
 	auto sp_B = d_B.as_span<float>();
 	auto sp_C = d_C.as_span<float>();
 
-	cuda::memory::copy(sp_A, h_A.data());
-	cuda::memory::copy(sp_B, h_B.data());
+	cuda::memory::copy(sp_A, h_A);
+	cuda::memory::copy(sp_B, h_B);
 
 	auto launch_config = cuda::launch_config_builder()
 		.overall_size(numElements)
@@ -68,7 +68,7 @@ int main()
 		sp_A.data(), sp_B.data(), sp_C.data(), numElements
 	);
 
-	cuda::memory::copy(h_C.data(), sp_C);
+	cuda::memory::copy(h_C, sp_C);
 
 	// Verify that the result vector is correct
 	for (int i = 0; i < numElements; ++i) {

diff --git a/src/cuda/api/detail/region.hpp b/src/cuda/api/detail/region.hpp
@@ -12,7 +12,7 @@
 #ifndef CUDA_API_WRAPPERS_REGION_HPP_
 #define CUDA_API_WRAPPERS_REGION_HPP_
 
-#include <type_traits>
+#include "type_traits.hpp"
 #include <stdexcept>
 
 #ifndef CPP14_CONSTEXPR
@@ -60,11 +60,17 @@ class base_region_t {
 	constexpr base_region_t(pointer start, size_type size_in_bytes) noexcept
 		: start_(start), size_in_bytes_(size_in_bytes) {}
 
-	template <typename U>
-	constexpr base_region_t(span<U> span) noexcept : start_(span.data()), size_in_bytes_(span.size() * sizeof(U))
+	/**
+	 * A constructor from types such as `::std::span`'s or `::std::vector`'s, whose data is in
+	 * a contiguous region of memory
+	 */
+	template <typename ContiguousContainer, typename = cuda::detail_::enable_if_t<
+		cuda::detail_::is_kinda_like_contiguous_container<ContiguousContainer>::value, void>>
+	constexpr base_region_t(ContiguousContainer&& contiguous_container) noexcept
+	: start_(contiguous_container.data()), size_in_bytes_(contiguous_container.size() * sizeof(*(contiguous_container.data())))
 	{
-		static_assert(::std::is_const<T>::value or not ::std::is_const<U>::value,
-			"Attempt to construct a non-const memory region from a const span");
+		static_assert(::std::is_const<T>::value or not ::std::is_const<decltype(*(contiguous_container.data()))>::value,
+			"Attempt to construct a non-const memory region from a container of const data");
 	}
 
 	template <typename U>