168 add cuda 13 to main branch #169

albertandaluz · 2025-12-11T11:31:21Z

added cuda 13.x compiler to CI
enforce standards compliant preprocessor in msvc
skipping 13.x for jetson

Copilot

Pull request overview

Key changes attempted:

Addition of CUDA 13.x to CI matrices for Windows and Linux AMD64 workflows
Enforcement of standards-compliant preprocessor flag (/Zc:preprocessor) for MSVC across test and benchmark targets
Conditional logic for CUDA 13+ to use "Hybrid" instead of "Shared" runtime library

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
`.github/workflows/cmake-windows-amd64.yml`	Adds CUDA 13.x matrix entries for cl and llvm compilers (version does not exist)
`.github/workflows/cmake-linux-amd64.yml`	Adds CUDA 13.x matrix entries for g++ and clang++ compilers (version does not exist)
`cmake/libs/cuda/target_generation.cmake`	Adds conditional logic for CUDA 13+ to use Hybrid runtime library and passes /Zc:preprocessor to CUDA compiler
`cmake/libs/cuda/deploy.cmake`	Adds CUDA 13 specific library version mappings and x64 subdirectory path handling
`cmake/tests/add_shared_test_libs.cmake`	Adds /Zc:preprocessor flag for MSVC to test libraries
`cmake/tests/add_generated_test.cmake`	Adds /Zc:preprocessor flag for MSVC to generated tests
`benchmarks/CMakeLists.txt`	Adds /Zc:preprocessor flag for MSVC to benchmarks

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

cmake/libs/cuda/target_generation.cmake

cmake/libs/cuda/deploy.cmake

Copilot · 2025-12-11T11:34:53Z

cmake/libs/cuda/deploy.cmake

+	elseif(CUDA_VERSION_MAJOR EQUAL 13)		
+		# cufft is at version 11 for CUDA 12
+		list(TRANSFORM TARGET_DEPENDENCIES_COMMON REPLACE "cufft64_${CUDA_VERSION_MAJOR}" "cufft64_12")
+		list(TRANSFORM TARGET_DEPENDENCIES_COMMON REPLACE "curand64_${CUDA_VERSION_MAJOR}" "curand64_10")	
 	endif()
    list(TRANSFORM TARGET_DEPENDENCIES_COMMON APPEND ".dll")
-    list(TRANSFORM TARGET_DEPENDENCIES_COMMON PREPEND "${CUDAToolkit_BIN_DIR}/")
+	if(CUDA_VERSION_MAJOR EQUAL 13)		
+	#todo: winarm64
+		list(TRANSFORM TARGET_DEPENDENCIES_COMMON PREPEND "${CUDAToolkit_BIN_DIR}/x64/")
+	else()
+		list(TRANSFORM TARGET_DEPENDENCIES_COMMON PREPEND "${CUDAToolkit_BIN_DIR}/")
+	endif()


Inconsistent indentation: These lines use tab characters while the rest of the file uses spaces. Please use consistent spacing (4 spaces based on surrounding code) to maintain code style consistency.

Copilot · 2025-12-11T11:47:09Z

@albertandaluz I've opened a new pull request, #170, to work on those changes. Once the pull request is ready, I'll request review from you.

added ci configurations for cuda 13.1 on windows and linux (x64) pass :/Zc:preprocessor to supress warnings about legacy preprocessor on msvc add compiler preprocvessor guards to skip compilation of test_tuple.h with cuda 12.4 added a reproducer for a nvcc bug found with cuda 12.4

Copilot

Pull request overview

Copilot reviewed 13 out of 13 changed files in this pull request and generated 8 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-27T12:30:46Z

tests/cudabug/test_nvcc131_BugReproducer.h

+#define NVCC_VERSION_CALCULATED (__CUDACC_VER_MAJOR__ * 10000 + __CUDACC_VER_MINOR__ * 100 + __CUDACC_VER_BUILD__)
+#define NVCC_VERSION_12_4_99 120499
+
+// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.93


The comment mentions 'nvcc versions lower than 12.4.93', but the actual comparison on line 32 checks against NVCC_VERSION_12_4_99 (which represents version 12.4.99). This creates a mismatch between the comment and the code logic. Consider updating the comment to accurately reflect the version being checked (12.4.99) or adjusting the version constant if 12.4.93 was the intended threshold.

Suggested change

// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.93

// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.99

Copilot · 2025-12-27T12:30:46Z

tests/data/test_tuple.h

        std::cout << "test_tuple Failed!!" << std::endl;
        return -1;
    }
+ #else


Inconsistent indentation: the '#else' directive has a leading space while the surrounding '#ifdef' and '#endif' directives don't. All preprocessor directives should have consistent indentation.

Suggested change

#else

#else

Copilot · 2025-12-27T12:30:47Z

tests/cudabug/test_nvcc131_BugReproducer.h

+
+    // Using and_v2 here and keeping and_v in line 67
+    // or the other way arround, compiles
+    // Changin result1 to the literal true, compiles


Typo in comment: 'Changin' should be 'Changing'.

Suggested change

// Changin result1 to the literal true, compiles

// Changing result1 to the literal true, compiles

Copilot · 2025-12-27T12:30:47Z

tests/data/test_tuple.h

 #include <fused_kernel/core/execution_model/operation_model/operation_tuple.h>
 #include <fused_kernel/algorithms/basic_ops/vector_ops.h>
 #include <fused_kernel/core/execution_model/memory_operations.h>
+#ifdef __NVCC__


The macros NVCC_VERSION_CALCULATED and NVCC_VERSION_12_4_99 are used on line 24 but are never defined in this file. These macros need to be defined before use, similar to how they are defined in tests/cudabug/test_nvcc131_BugReproducer.h (lines 28-29).

Suggested change

#ifdef __NVCC__

#ifdef __NVCC__

// Helper macros to compute and compare the NVCC version.

// NVCC_VERSION_CALCULATED follows the pattern MAJOR * 10000 + MINOR * 100 + BUILD.

#ifndef NVCC_VERSION_CALCULATED

#if defined(__CUDACC_VER_MAJOR__) && defined(__CUDACC_VER_MINOR__) && defined(__CUDACC_VER_BUILD__)

#define NVCC_VERSION_CALCULATED (__CUDACC_VER_MAJOR__ * 10000 + __CUDACC_VER_MINOR__ * 100 + __CUDACC_VER_BUILD__)

#else

// Fallback: treat unknown NVCC version as 0 so that the comparison remains well-formed.

#define NVCC_VERSION_CALCULATED 0

#endif

#endif

#ifndef NVCC_VERSION_12_4_99

#define NVCC_VERSION_12_4_99 120499

#endif

Copilot · 2025-12-27T12:30:47Z

tests/cudabug/test_nvcc131_BugReproducer.h

+    constexpr bool result1 = myint.instance == 1;
+
+    // Using and_v2 here and keeping and_v in line 67
+    // or the other way arround, compiles


Typo in comment: 'arround' should be 'around'.

Suggested change

// or the other way arround, compiles

// or the other way around, compiles

Copilot · 2025-12-27T12:30:47Z

include/fused_kernel/core/execution_model/parallel_architectures.h

+#ifndef CLANG_HOST_DEVICE
+static_assert(false);
+#endif
+
+#if defined(__NVCC__) || (CLANG_HOST_DEVICE == 1)


This static_assert will always fail, causing compilation to fail unconditionally. The check should verify whether CLANG_HOST_DEVICE is defined, not trigger a failure when it is defined. Consider using '#ifndef' instead to ensure the macro is properly defined, or remove this assertion if it's intended for debugging.

Suggested change

#ifndef CLANG_HOST_DEVICE

static_assert(false);

#endif

#if defined(__NVCC__) || (CLANG_HOST_DEVICE == 1)

#if defined(__NVCC__) || (defined(CLANG_HOST_DEVICE) && (CLANG_HOST_DEVICE == 1))

Copilot · 2025-12-27T12:30:48Z

cmake/libs/cuda/deploy.cmake

+	elseif(CUDA_VERSION_MAJOR EQUAL 13)		
+		# cufft is at version 12 for CUDA 13
+		list(TRANSFORM TARGET_DEPENDENCIES_COMMON REPLACE "cufft64_${CUDA_VERSION_MAJOR}" "cufft64_12")
+		list(TRANSFORM TARGET_DEPENDENCIES_COMMON REPLACE "curand64_${CUDA_VERSION_MAJOR}" "curand64_10")	
 	endif()
    list(TRANSFORM TARGET_DEPENDENCIES_COMMON APPEND ".dll")
-    list(TRANSFORM TARGET_DEPENDENCIES_COMMON PREPEND "${CUDAToolkit_BIN_DIR}/")
+	if(CUDA_VERSION_MAJOR EQUAL 13)		


Inconsistent spacing: this line uses tabs after 'if' statements. The file appears to mix tabs and spaces. Consider using consistent spacing (spaces) throughout to match CMake conventions.

Copilot · 2025-12-27T12:30:48Z

tests/data/test_tuple.h

 #include <fused_kernel/algorithms/basic_ops/vector_ops.h>
 #include <fused_kernel/core/execution_model/memory_operations.h>
+#ifdef __NVCC__
+// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.93


The comment on line 23 mentions 'nvcc versions lower than 12.4.93', but the actual comparison on line 24 checks against NVCC_VERSION_12_4_99 (which represents version 12.4.99). This creates a mismatch between the comment and the code logic. Consider updating the comment to accurately reflect the version being checked (12.4.99) or adjusting the version constant if 12.4.93 was the intended threshold.

Suggested change

// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.93

// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.99

morousg

Review the copilot comments

albertandaluz requested a review from Copilot December 11, 2025 11:31

albertandaluz linked an issue Dec 11, 2025 that may be closed by this pull request

Add cuda 13.1 to main branch #168

Open

albertandaluz self-assigned this Dec 11, 2025

Copilot started reviewing on behalf of albertandaluz December 11, 2025 11:31 View session

Copilot AI reviewed Dec 11, 2025

View reviewed changes

Copilot AI mentioned this pull request Dec 11, 2025

[WIP] Update stacked PR to address feedback on CUDA 13.1 #170

Closed

albertandaluz force-pushed the 168-add-cuda-131-to-main-branch branch 12 times, most recently from 7051a99 to 5866303 Compare December 11, 2025 18:06

albertandaluz changed the title ~~168 add cuda 131 to main branch~~ 168 add cuda 13 to main branch Dec 12, 2025

albertandaluz added 4 commits December 27, 2025 13:22

downgrading cuda to 13.0

d454f77

downgrading cuda to 13.0 on linux x86-64 too

e7b5652

fixed compilation of cuda code with clang (hotfix from LTS)

67dc450

albertandaluz force-pushed the 168-add-cuda-131-to-main-branch branch from 06c7ca5 to 67dc450 Compare December 27, 2025 12:27

Copilot AI review requested due to automatic review settings December 27, 2025 12:27

Copilot started reviewing on behalf of albertandaluz December 27, 2025 12:27 View session

Copilot AI reviewed Dec 27, 2025

View reviewed changes

morousg requested changes Jan 5, 2026

View reviewed changes

	// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.93
	// Condition 1: we are compiling with MSVC + nvcc OR other compilers + nvcc versions lower than 12.4.99

	// Changin result1 to the literal true, compiles
	// Changing result1 to the literal true, compiles

-#ifdef __NVCC__
+#ifdef __NVCC__
+// Helper macros to compute and compare the NVCC version.
+// NVCC_VERSION_CALCULATED follows the pattern MAJOR * 10000 + MINOR * 100 + BUILD.
+#ifndef NVCC_VERSION_CALCULATED
+#if defined(__CUDACC_VER_MAJOR__) && defined(__CUDACC_VER_MINOR__) && defined(__CUDACC_VER_BUILD__)
+#define NVCC_VERSION_CALCULATED (__CUDACC_VER_MAJOR__ * 10000 + __CUDACC_VER_MINOR__ * 100 + __CUDACC_VER_BUILD__)
+#else
+// Fallback: treat unknown NVCC version as 0 so that the comparison remains well-formed.
+#define NVCC_VERSION_CALCULATED 0
+#endif
+#endif
+#ifndef NVCC_VERSION_12_4_99
+#define NVCC_VERSION_12_4_99 120499
+#endif

	// or the other way arround, compiles
	// or the other way around, compiles

Uh oh!

168 add cuda 13 to main branch #169

Are you sure you want to change the base?

168 add cuda 13 to main branch #169

Uh oh!

Conversation

albertandaluz commented Dec 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment • edited by albertandaluz Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Copilot AI Dec 11, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI commented Dec 11, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 27, 2025

Choose a reason for hiding this comment

Uh oh!

morousg left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

albertandaluz commented Dec 11, 2025 •

edited

Loading

Copilot AI left a comment •

edited by albertandaluz

Loading