Update 1_Utilities/README.md to redirect bandwidthTest to NVBandwidth (#371)

Update the vulkan headers include sequence and the transpose code format check

.clang-format

@@ -0,0 +1,49 @@
+---
+AccessModifierOffset: -4
+AlignAfterOpenBracket: Align
+AlignConsecutiveAssignments: Consecutive
+AlignConsecutiveDeclarations: Consecutive
+AlignConsecutiveMacros: Consecutive
+AlignEscapedNewlines: Left
+AlignOperands: AlignAfterOperator
+AlignTrailingComments: true
+AllowAllParametersOfDeclarationOnNextLine: false
+BinPackArguments: false
+BinPackParameters: false
+BraceWrapping:
+    AfterClass: true
+    AfterControlStatement: false
+    AfterExternBlock: true
+    AfterFunction: true
+    AfterStruct: true
+    AfterUnion: true
+    BeforeCatch: true
+    BeforeElse: true
+    IndentBraces: false
+BreakBeforeBraces: Custom
+BreakBeforeConceptDeclarations: true
+BreakBeforeBinaryOperators: NonAssignment
+BreakBeforeTernaryOperators: true
+BreakConstructorInitializers: BeforeComma
+BreakInheritanceList: BeforeComma
+ColumnLimit: 120
+DerivePointerAlignment: false
+FixNamespaceComments: true
+IncludeCategories:
+  - Regex:           '^<.*>'
+    Priority:        1
+  - Regex:           '^".*"'
+    Priority:        2
+SortIncludes: true
+IncludeBlocks: Regroup
+IndentWidth: 4
+MaxEmptyLinesToKeep: 2
+PointerAlignment: Right
+SortUsingDeclarations: true
+SpaceAfterCStyleCast: false
+SpaceBeforeAssignmentOperators: true
+SpaceBeforeParens: ControlStatements
+Standard: c++17
+TabWidth: 4
+UseTab: Never
+...

.gitignore

@@ -1,3 +1,6 @@
 build
 .vs
 .clangd
+test
+settings.json
+launch.json

.pre-commit-config.yaml

@@ -0,0 +1,106 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.
+ci:
+    autofix_commit_msg: |
+      [pre-commit.ci] auto code formatting
+    autofix_prs: false
+    autoupdate_branch: ''
+    autoupdate_commit_msg: '[pre-commit.ci] pre-commit autoupdate'
+    autoupdate_schedule: quarterly
+    skip: []
+    submodules: false
+
+repos:
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v5.0.0
+    hooks:
+      - id: end-of-file-fixer
+        exclude: |
+          (?x)^(
+            .*\.raw$|
+            .*\.bin$|
+            .*\.dat$|
+            .*\.nv12$|
+            data/.*|
+            Common/.*
+          )
+        files: |
+          (?x)^(
+            .*\.txt$|
+            .*\.md$|
+            .*\.cpp$|
+            .*\.cxx$|
+            .*\.hpp$|
+            .*\.h$|
+            .*\.cu$|
+            .*\.cuh$|
+            .*\.py$|
+            .*\.json$
+          )
+      - id: mixed-line-ending
+        exclude: |
+          (?x)^(
+            .*\.raw$|
+            .*\.bin$|
+            .*\.dat$|
+            .*\.nv12$|
+            data/.*|
+            Common/.*
+          )
+        files: |
+          (?x)^(
+            .*\.txt$|
+            .*\.md$|
+            .*\.cpp$|
+            .*\.cxx$|
+            .*\.hpp$|
+            .*\.h$|
+            .*\.cu$|
+            .*\.cuh$|
+            .*\.py$|
+            .*\.json$
+          )
+      - id: trailing-whitespace
+        exclude: |
+          (?x)^(
+            .*\.raw$|
+            .*\.bin$|
+            .*\.dat$|
+            .*\.nv12$|
+            data/.*|
+            Common/.*
+          )
+        files: |
+          (?x)^(
+            .*\.txt$|
+            .*\.md$|
+            .*\.cpp$|
+            .*\.cxx$|
+            .*\.hpp$|
+            .*\.h$|
+            .*\.cu$|
+            .*\.cuh$|
+            .*\.py$|
+            .*\.json$
+          )
+  - repo: https://github.com/pre-commit/mirrors-clang-format
+    rev: v19.1.6
+    hooks:
+      - id: clang-format
+        types_or: [file]
+        files: |
+          (?x)^(
+            ^.*\.c$|
+            ^.*\.cpp$|
+            ^.*\.cu$|
+            ^.*\.cuh$|
+            ^.*\.cxx$|
+            ^.*\.h$|
+            ^.*\.hpp$|
+            ^.*\.inl$|
+            ^.*\.mm$
+          )
+        exclude: |
+          (?x)^(
+            Common/.*
+          )
+        args: ["-fallback-style=none", "-style=file", "-i"]

CHANGELOG.md

@@ -1,5 +1,15 @@
 ## Changelog
 
+### CUDA 12.9
+* Updated toolchain for cross-compilation for Tegra Linux platforms.
+* Added `run_tests.py` utility to exercise all samples. See README.md for details
+* Repository has been updated with consistent code formatting across all samples
+* Many small code tweaks and bug fixes (see commit history for details)
+* Removed the following outdated samples:
+  * `1_Utilities`
+    * `bandwidthTest` - this sample was out of date and did not produce accurate results. For bandwidth
+    testing of NVIDIA GPU platforms, please refer to [NVBandwidth](https://github.com/NVIDIA/nvbandwidth)
+
 ### CUDA 12.8
 * Updated build system across the repository to CMake. Removed Visual Studio project files and Makefiles.
 * Removed the following outdated samples:
@@ -36,6 +46,7 @@
     * `cuDLALayerwiseStatsHybrid`
     * `cuDLALayerwiseStatsStandalone`
     * `cuDLAStandaloneMode`
+    * `cudaNvSciBufMultiplanar`
     * `cudaNvSciNvMedia`
     * `fluidsGLES`
     * `nbody_opengles`

CMakeLists.txt

@@ -16,8 +16,10 @@ set(CMAKE_CUDA_STANDARD_REQUIRED ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --extended-lambda")

CONTRIBUTING.md

@@ -0,0 +1,103 @@
+
+# Contributing to the CUDA Samples
+
+Thank you for your interest in contributing to the CUDA Samples!
+
+
+## Getting Started
+
+1. **Fork & Clone the Repository**:
+
+   Fork the reporistory and clone the fork. For more information, check [GitHub's documentation on forking](https://docs.github.com/en/github/getting-started-with-github/fork-a-repo) and [cloning a repository](https://docs.github.com/en/github/creating-cloning-and-archiving-repositories/cloning-a-repository).
+
+## Making Changes
+
+1. **Create a New Branch**:
+
+   ```bash
+   git checkout -b your-feature-branch
+   ```
+
+2. **Make Changes**.
+
+3. **Build and Test**:
+
+   Ensure changes don't break existing functionality by building and running tests.
+
+   For more details on building and testing, refer to the [Building and Testing](#building-and-testing) section below.
+
+4. **Commit Changes**:
+
+   ```bash
+   git commit -m "Brief description of the change"
+   ```
+
+## Building and Testing
+
+For information on building a running tests on the samples, please refer to the main [README](README.md)
+
+## Creating a Pull Request
+
+1. Push changes to your fork
+2. Create a pull request targeting the `master` branch of the original CUDA Samples repository. Refer to [GitHub's documentation](https://docs.github.com/en/github/collaborating-with-pull-requests/proposing-changes-to-your-work-with-pull-requests/about-pull-requests) for more information on creating a pull request.
+3. Describe the purpose and context of the changes in the pull request description.
+
+## Code Formatting (pre-commit hooks)
+
+The CUDA Samples repository uses [pre-commit](https://pre-commit.com/) to execute all code linters and formatters. These
+tools ensure a consistent coding style throughout the project. Using pre-commit ensures that linter
+versions and options are aligned for all developers. Additionally, there is a CI check in place to
+enforce that committed code follows our standards.
+
+The linters used by the CUDA Samples are listed in `.pre-commit-config.yaml`.
+For example, C++ and CUDA code is formatted with [`clang-format`](https://clang.llvm.org/docs/ClangFormat.html).
+
+To use `pre-commit`, install via `conda` or `pip`:
+
+```bash
+conda config --add channels conda-forge
+conda install pre-commit
+```
+
+```bash
+pip install pre-commit
+```
+
+Then run pre-commit hooks before committing code:
+
+```bash
+pre-commit run
+```
+
+By default, pre-commit runs on staged files (only changes and additions that will be committed).
+To run pre-commit checks on all files, execute:
+
+```bash
+pre-commit run --all-files
+```
+
+Optionally, you may set up the pre-commit hooks to run automatically when you make a git commit. This can be done by running:
+
+```bash
+pre-commit install
+```
+
+Now code linters and formatters will be run each time you commit changes.
+
+You can skip these checks with `git commit --no-verify` or with the short version `git commit -n`, althoguh please note
+that this may result in pull requests being rejected if subsequent checks fail.
+
+## Review Process
+
+Once submitted, maintainers will be automatically assigned to review the pull request. They might suggest changes or improvements. Constructive feedback is a part of the collaborative process, aimed at ensuring the highest quality code.
+
+For constructive feedback and effective communication during reviews, we recommend following [Conventional Comments](https://conventionalcomments.org/).
+
+Further recommended reading for successful PR reviews:
+
+- [How to Do Code Reviews Like a Human (Part One)](https://mtlynch.io/human-code-reviews-1/)
+- [How to Do Code Reviews Like a Human (Part Two)](https://mtlynch.io/human-code-reviews-2/)
+
+## Thank You
+
+Your contributions enhance the CUDA Samples for the entire community. We appreciate your effort and collaboration!

Common/helper_cuda_drvapi.h

@@ -241,7 +241,7 @@ inline int gpuGetMaxGflopsDeviceIdDRV() {
       }
 
       unsigned long long compute_perf =
-          (unsigned long long)(multiProcessorCount * sm_per_multiproc *
+          ((unsigned long long)multiProcessorCount * sm_per_multiproc *
                                clockRate);
 
       if (compute_perf > max_compute_perf) {

Common/nvMatrix.h

@@ -258,7 +258,7 @@ namespace nv
                 s[2] = &r3[0];
                 s[3] = &r4[0];
 
-                register int i,j,p,jj;
+                int i,j,p,jj;
 
                 for (i=0; i<4; i++)
                 {

README.md

@@ -1,6 +1,6 @@
 # CUDA Samples
 
-Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 12.6](https://developer.nvidia.com/cuda-downloads).
+Samples for CUDA Developers which demonstrates features in CUDA Toolkit. This version supports [CUDA Toolkit 12.9](https://developer.nvidia.com/cuda-downloads).
 
 ## Release Notes
 
@@ -14,7 +14,7 @@ This section describes the release notes for the CUDA Samples on GitHub only.
 
 ### Prerequisites
 
-Download and install the [CUDA Toolkit 12.8](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 For system requirements and installation instructions of cuda toolkit, please refer to the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/), and the [Windows Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html).
 
 ### Getting the CUDA Samples
@@ -72,6 +72,17 @@ Open the generated solution file CUDA_Samples.sln in Visual Studio. Build the sa
 
 Run the samples from the output directories specified in Visual Studio.
 
+### Enabling On-GPU Debugging
+
+NVIDIA GPUs support on-GPU debugging through cuda-gdb. Enabling this may significantly affect application performance as certain compiler optimizations are disabled
+in this configuration, hence it's not on by default. Enablement of on-device debugging is controlled via the `-G` switch to nvcc.
+
+To enable cuda-gdb for samples builds, define the `ENABLE_CUDA_DEBUG` flag on the CMake command line. For example:
+
+```
+cmake -DENABLE_CUDA_DEBUG=True ...
+```
+
 ### Platform-Specific Samples
 
 Some CUDA samples are specific to certain platforms, and require passing flags into CMake to enable. In particular, we define the following platform-specific flags:
@@ -94,9 +105,9 @@ Navigate to the root of the cloned repository and create a build directory:
 ```
 mkdir build && cd build
 ```
-Configure the project with CMake, specifying the Tegra toolchain file:
+Configure the project with CMake, specifying the Tegra toolchain file. And you can use -DTARGET_FS to point to the target file system root path for necessary include and library files:
 ```
-cmake .. -DCMAKE_TOOLCHAIN_FILE=/path/to/tegra/toolchain.cmake
+cmake .. -DCMAKE_TOOLCHAIN_FILE=../cmake/toolchains/toolchain-aarch64-linux.cmake -DTARGET_FS=/path/to/target/system/file/system
 ```
 Build the samples:
 ```
@@ -111,7 +122,7 @@ Instead of being in the default location, `/usr/local/cuda/include` or `/usr/loc
 
 `/usr/local/cuda/<ARCH>/targets/aarch64-linux/lib`
 and
-`/usr/local/cuda-12.8/<ARCH>/include`
+`/usr/local/cuda/<ARCH>/include`
 
 An example build might look like this:
 
@@ -128,6 +139,168 @@ Note that in the current branch sample cross-compilation for QNX is not fully va
 near future with QNX cross-compilation instructions. In the meantime, if you want to cross-compile for QNX please check out one
 of the previous tags prior to the CMake build system transition in 12.8.
 
+## Running All Samples as Tests
+
+It's important to note that the CUDA samples are _not_ intended as a validation suite for CUDA. They do not cover corner cases, they do not completely cover the
+runtime and driver APIs, are not intended for performance benchmarking, etc. That said, it can sometimes be useful to run all of the samples as a quick sanity check and
+we provide a script to do so, `run_tests.py`.
+
+This Python3 script finds all executables in a subdirectory you choose, matching application names with command line arguments specified in `test_args.json`. It accepts
+the following command line arguments:
+
+| Switch     | Purpose                                                                                                        | Example                 |
+| ---------- | -------------------------------------------------------------------------------------------------------------- | ----------------------- |
+| --dir      | Specify the root directory to search for executables (recursively)                                             | --dir ./build/Samples   |
+| --config   | JSON configuration file for executable arguments                                                               | --config test_args.json |
+| --output   | Output directory for test results (stdout saved to .txt files - directory will be created if it doesn't exist) | --output ./test         |
+| --args     | Global arguments to pass to all executables (not currently used)                                               | --args arg_1 arg_2 ...  |
+| --parallel | Number of applications to execute in parallel.                                                                 | --parallel 8            |
+
+
+Application configurations are loaded from `test_args.json` and matched against executable names (discarding the `.exe` extension on Windows).
+
+The script returns 0 on success, or the first non-zero error code encountered during testing on failure. It will also print a condensed list of samples that failed, if any.
+
+There are three primary modes of configuration:
+
+**Skip**
+
+An executable configured with "skip" will not be executed. These generally rely on having attached graphical displays and are not suited to this kind of automation.
+
+Configuration example:
+```json
+"fluidsGL": {
+    "skip": true
+}
+```
+
+You will see:
+```
+Skipping fluidsGL (marked as skip in config)
+```
+
+**Single Run**
+
+For executables to run one time only with arguments, specify each argument as a list entry. Each entry in the JSON file will be appended to the command line, separated
+by a space.
+
+All applications execute from their current directory, so all paths are relative to the application's location.
+
+Note that if an application needs no arguments, this entry is optional. An executable found without a matching entry in the JSON will just run as `./application` from its
+current directory.
+
+Configuration example:
+```json
+"ptxgen": {
+    "args": [
+        "test.ll",
+        "-arch=compute_75"
+    ]
+}
+```
+
+You will see:
+```
+Running ptxgen
+    Command: ./ptxgen test.ll -arch=compute_75
+    Test completed with return code 0
+```
+
+**Multiple Runs**
+
+For executables to run multiple times with different command line arguments, specify any number of sets of args within a "runs" list.
+
+As with single runs, all applications execute from their current directory, so all paths are relative to the application's location.
+
+Configuration example:
+```json
+"recursiveGaussian": {
+    "runs": [
+        {
+            "args": [
+                "-sigma=10",
+                "-file=data/ref_10.ppm"
+            ]
+        },
+        {
+            "args": [
+                "-sigma=14",
+                "-file=data/ref_14.ppm"
+            ]
+        },
+        {
+            "args": [
+                "-sigma=18",
+                "-file=data/ref_18.ppm"
+            ]
+        },
+        {
+            "args": [
+                "-sigma=22",
+                "-file=data/ref_22.ppm"
+            ]
+        }
+    ]
+}
+```
+
+You will see:
+```
+Running recursiveGaussian (run 1/4)
+    Command: ./recursiveGaussian -sigma=10 -file=data/ref_10.ppm
+    Test completed with return code 0
+Running recursiveGaussian (run 2/4)
+    Command: ./recursiveGaussian -sigma=14 -file=data/ref_14.ppm
+    Test completed with return code 0
+Running recursiveGaussian (run 3/4)
+    Command: ./recursiveGaussian -sigma=18 -file=data/ref_18.ppm
+    Test completed with return code 0
+Running recursiveGaussian (run 4/4)
+    Command: ./recursiveGaussian -sigma=22 -file=data/ref_22.ppm
+    Test completed with return code 0
+```
+
+### Example Usage
+
+Here is an example set of commands to build and test all of the samples.
+
+First, build:
+```bash
+mkdir build
+cd build
+cmake ..
+make -j$(nproc)
+```
+
+Now, return to the samples root directory and run the test script:
+```bash
+cd ..
+python3 run_tests.py --output ./test --dir ./build/Samples --config test_args.json
+```
+
+If all applications run successfully, you will see something similar to this (the specific number of samples will depend on your build type
+and system configuration):
+
+```
+Test Summary:
+Ran 199 test runs for 180 executables.
+All test runs passed!
+```
+
+If some samples fail, you will see something like this:
+
+```
+Test Summary:
+Ran 199 test runs for 180 executables.
+Failed runs (2):
+  bicubicTexture (run 1/5): Failed (code 1)
+  Mandelbrot (run 1/2): Failed (code 1)
+```
+
+You can inspect the stdout logs in the output directory (generally `APM_<application_name>.txt` or `APM_<application_name>.run<n>.txt`) to help
+determine what may have gone wrong from the output logs. Please file issues against the samples repository if you believe a sample is failing
+incorrectly on your system.
+
 ## Samples list
 
 ### [0. Introduction](./Samples/0_Introduction/README.md)
@@ -170,7 +343,7 @@ These third-party dependencies are required by some CUDA samples. If available,
 
 FreeImage is an open source imaging library. FreeImage can usually be installed on Linux using your distribution's package manager system. FreeImage can also be downloaded from the FreeImage website.
 
-To set up FreeImage on a Windows system, extract the FreeImage DLL distribution into the folder `../../../Common/FreeImage/Dist/x64` such that it contains the .h and .lib files. Copy the .dll file to the Release/ Debug/ execution folder or pass the FreeImage folder when cmake configuring with the `-DFREEIMAGE_INCLUDE_DIR` and `-DFREEIMAGE_LIBRARY` options.
+To set up FreeImage on a Windows system, extract the FreeImage DLL distribution into the folder `./Common/FreeImage/Dist/x64` such that it contains the .h and .lib files. Copy the .dll file to the Release/ Debug/ execution folder or pass the FreeImage folder when cmake configuring with the `-DFreeImage_INCLUDE_DIR` and `-DFreeImage_LIBRARY` options.
 
 #### Message Passing Interface
 
@@ -203,11 +376,11 @@ Vulkan is a low-overhead, cross-platform 3D graphics and compute API. Vulkan tar
 #### GLFW
 GLFW is a lightweight, open-source library designed for managing OpenGL, OpenGL ES, and Vulkan contexts. It simplifies the process of creating and managing windows, handling user input (keyboard, mouse, and joystick), and working with multiple monitors in a cross-platform manner.
 
-To set up GLFW on a Windows system, Download the pre-built binaries from [GLFW website](https://www.glfw.org/download.html) and extract the zip file into the folder, pass the GLFW include header as `-DGLFW_INCLUDE_DIR` for cmake configuring and follow the Build_instructions.txt in the sample folder to set up the t.
+To set up GLFW on a Windows system, Download the pre-built binaries from [GLFW website](https://www.glfw.org/download.html) and extract the zip file into the folder, pass the GLFW include header folder as `-DGLFW_INCLUDE_DIR` and lib folder as `-DGLFW_LIB_DIR` for cmake configuring.
 
 #### OpenMP
 
-OpenMP is an API for multiprocessing programming. OpenMP can be installed using your Linux distribution's package manager system. It usually comes preinstalled with GCC. It can also be found at the [OpenMP website](http://openmp.org/).
+OpenMP is an API for multiprocessing programming. OpenMP can be installed using your Linux distribution's package manager system. It usually comes preinstalled with GCC. It can also be found at the [OpenMP website](http://openmp.org/). For compilers such as clang, `libomp.so` and other components for LLVM must be installed separated. You will also need to set additional flags in your CMake configuration files, such as: `-DOpenMP_CXX_FLAGS="-fopenmp=libomp" -DOpenMP_CXX_LIB_NAMES="omp" -DOpenMP_omp_LIBRARY="/path/to/libomp.so"`.
 
 #### Screen
 

Samples/0_Introduction/CMakeLists.txt

@@ -1,20 +1,3 @@
-cmake_minimum_required(VERSION 3.20)
-
-list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Modules")
-
-project(simpleCallback LANGUAGES C CXX CUDA)
-
-find_package(CUDAToolkit REQUIRED)
-
-set(CMAKE_POSITION_INDEPENDENT_CODE ON)
-
-set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
-set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
-endif()
-
-
 add_subdirectory(UnifiedMemoryStreams)
 add_subdirectory(asyncAPI)
 add_subdirectory(clock)
@@ -55,6 +38,7 @@ add_subdirectory(simpleTexture3D)
 add_subdirectory(simpleTextureDrv)
 add_subdirectory(simpleVoteIntrinsics)
 add_subdirectory(simpleZeroCopy)
+add_subdirectory(template)
 add_subdirectory(systemWideAtomics)
 add_subdirectory(vectorAdd)
 add_subdirectory(vectorAddDrv)

Samples/0_Introduction/UnifiedMemoryStreams/CMakeLists.txt

@@ -10,15 +10,21 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries
 include_directories(../../../Common)
 
 # Source file
+if(CMAKE_GENERATOR MATCHES "Visual Studio")
+    find_package(OpenMP REQUIRED C CXX)
+else()
     find_package(OpenMP REQUIRED)
+endif()
 
 if(${OpenMP_FOUND})
     # Add target for UnifiedMemoryStreams

Samples/0_Introduction/UnifiedMemoryStreams/README.md

@@ -28,7 +28,7 @@ cudaStreamDestroy, cudaFree, cudaMallocManaged, cudaStreamAttachMemAsync, cudaSe
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/UnifiedMemoryStreams/UnifiedMemoryStreams.cu

@@ -31,10 +31,10 @@
  */
 
 // system includes
+#include <algorithm>
 #include <cstdio>
 #include <ctime>
 #include <vector>
-#include <algorithm>
 #ifdef USE_PTHREADS
 #include <pthread.h>
 #else
@@ -58,15 +58,25 @@ double drand48() { return double(rand()) / RAND_MAX; }
 const char *sSDKname = "UnifiedMemoryStreams";
 
 // simple task
-template <typename T>
+template <typename T> struct Task
-struct Task {
+{
     unsigned int size, id;
     T           *data;
     T           *result;
     T           *vector;
 
-  Task() : size(0), id(0), data(NULL), result(NULL), vector(NULL){};
+    Task()
-  Task(unsigned int s) : size(s), id(0), data(NULL), result(NULL) {
+        : size(0)
+        , id(0)
+        , data(NULL)
+        , result(NULL)
+        , vector(NULL) {};
+    Task(unsigned int s)
+        : size(s)
+        , id(0)
+        , data(NULL)
+        , result(NULL)
+    {
         // allocate unified memory -- the operation performed in this example will
         // be a DGEMV
         checkCudaErrors(cudaMallocManaged(&data, sizeof(T) * size * size));
@@ -75,7 +85,8 @@ struct Task {
         checkCudaErrors(cudaDeviceSynchronize());
     }
 
-  ~Task() {
+    ~Task()
+    {
         // ensure all memory is deallocated
         checkCudaErrors(cudaDeviceSynchronize());
         checkCudaErrors(cudaFree(data));
@@ -83,7 +94,8 @@ struct Task {
         checkCudaErrors(cudaFree(vector));
     }
 
-  void allocate(const unsigned int s, const unsigned int unique_id) {
+    void allocate(const unsigned int s, const unsigned int unique_id)
+    {
         // allocate unified memory outside of constructor
         id   = unique_id;
         size = s;
@@ -105,7 +117,8 @@ struct Task {
 };
 
 #ifdef USE_PTHREADS
-struct threadData_t {
+struct threadData_t
+{
     int             tid;
     Task<double>   *TaskListPtr;
     cudaStream_t   *streams;
@@ -117,8 +130,8 @@ typedef struct threadData_t threadData;
 #endif
 
 // simple host dgemv: assume data is in row-major format and square
-template <typename T>
+template <typename T> void gemv(int m, int n, T alpha, T *A, T *x, T beta, T *result)
-void gemv(int m, int n, T alpha, T *A, T *x, T beta, T *result) {
+{
     // rows
     for (int i = 0; i < n; i++) {
         result[i] *= beta;
@@ -131,7 +144,8 @@ void gemv(int m, int n, T alpha, T *A, T *x, T beta, T *result) {
 
 // execute a single task on either host or device depending on size
 #ifdef USE_PTHREADS
-void *execute(void *inpArgs) {
+void *execute(void *inpArgs)
+{
     threadData     *dataPtr = (threadData *)inpArgs;
     cudaStream_t   *stream  = dataPtr->streams;
     cublasHandle_t *handle  = dataPtr->handles;
@@ -142,92 +156,75 @@ void *execute(void *inpArgs) {
 
         if (t.size < 100) {
             // perform on host
-      printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid,
+            printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid, t.size);
-             t.size);
 
             // attach managed memory to a (dummy) stream to allow host access while
             // the device is running
-      checkCudaErrors(
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
-          cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-      checkCudaErrors(
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
-          cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-      checkCudaErrors(
-          cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
             // necessary to ensure Async cudaStreamAttachMemAsync calls have finished
             checkCudaErrors(cudaStreamSynchronize(stream[0]));
             // call the host operation
             gemv(t.size, t.size, 1.0, t.data, t.vector, 0.0, t.result);
-    } else {
+        }
+        else {
             // perform on device
-      printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid,
+            printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid, t.size);
-             t.size);
             double one  = 1.0;
             double zero = 0.0;
 
             // attach managed memory to my stream
             checkCudaErrors(cublasSetStream(handle[tid + 1], stream[tid + 1]));
-      checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0,
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0, cudaMemAttachSingle));
-                                               cudaMemAttachSingle));
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0, cudaMemAttachSingle));
-      checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0,
+            checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0, cudaMemAttachSingle));
-                                               cudaMemAttachSingle));
-      checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0,
-                                               cudaMemAttachSingle));
             // call the device operation
-      checkCudaErrors(cublasDgemv(handle[tid + 1], CUBLAS_OP_N, t.size, t.size,
+            checkCudaErrors(cublasDgemv(
-                                  &one, t.data, t.size, t.vector, 1, &zero,
+                handle[tid + 1], CUBLAS_OP_N, t.size, t.size, &one, t.data, t.size, t.vector, 1, &zero, t.result, 1));
-                                  t.result, 1));
         }
     }
 
     pthread_exit(NULL);
 }
 #else
-template <typename T>
+template <typename T> void execute(Task<T> &t, cublasHandle_t *handle, cudaStream_t *stream, int tid)
-void execute(Task<T> &t, cublasHandle_t *handle, cudaStream_t *stream,
+{
-             int tid) {
     if (t.size < 100) {
         // perform on host
-    printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid,
+        printf("Task [%d], thread [%d] executing on host (%d)\n", t.id, tid, t.size);
-           t.size);
 
         // attach managed memory to a (dummy) stream to allow host access while the
         // device is running
-    checkCudaErrors(
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
-        cudaStreamAttachMemAsync(stream[0], t.data, 0, cudaMemAttachHost));
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-    checkCudaErrors(
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
-        cudaStreamAttachMemAsync(stream[0], t.vector, 0, cudaMemAttachHost));
-    checkCudaErrors(
-        cudaStreamAttachMemAsync(stream[0], t.result, 0, cudaMemAttachHost));
         // necessary to ensure Async cudaStreamAttachMemAsync calls have finished
         checkCudaErrors(cudaStreamSynchronize(stream[0]));
         // call the host operation
         gemv(t.size, t.size, 1.0, t.data, t.vector, 0.0, t.result);
-  } else {
+    }
+    else {
         // perform on device
-    printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid,
+        printf("Task [%d], thread [%d] executing on device (%d)\n", t.id, tid, t.size);
-           t.size);
         double one  = 1.0;
         double zero = 0.0;
 
         // attach managed memory to my stream
         checkCudaErrors(cublasSetStream(handle[tid + 1], stream[tid + 1]));
-    checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0,
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.data, 0, cudaMemAttachSingle));
-                                             cudaMemAttachSingle));
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0, cudaMemAttachSingle));
-    checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.vector, 0,
+        checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0, cudaMemAttachSingle));
-                                             cudaMemAttachSingle));
-    checkCudaErrors(cudaStreamAttachMemAsync(stream[tid + 1], t.result, 0,
-                                             cudaMemAttachSingle));
         // call the device operation
-    checkCudaErrors(cublasDgemv(handle[tid + 1], CUBLAS_OP_N, t.size, t.size,
+        checkCudaErrors(cublasDgemv(
-                                &one, t.data, t.size, t.vector, 1, &zero,
+            handle[tid + 1], CUBLAS_OP_N, t.size, t.size, &one, t.data, t.size, t.vector, 1, &zero, t.result, 1));
-                                t.result, 1));
     }
 }
 #endif
 
 // populate a list of tasks with random sizes
-template <typename T>
+template <typename T> void initialise_tasks(std::vector<Task<T>> &TaskList)
-void initialise_tasks(std::vector<Task<T> > &TaskList) {
+{
     for (unsigned int i = 0; i < TaskList.size(); i++) {
         // generate random size
         int size;
@@ -236,7 +233,8 @@ void initialise_tasks(std::vector<Task<T> > &TaskList) {
     }
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     // set device
     cudaDeviceProp device_prop;
     int            dev_id = findCudaDevice(argc, (const char **)argv);
@@ -294,19 +292,17 @@ int main(int argc, char **argv) {
 
         if ((TaskList.size() / nthreads) == 0) {
             InputToThreads[i].taskSize    = (TaskList.size() / nthreads);
-      InputToThreads[i].TaskListPtr =
+            InputToThreads[i].TaskListPtr = &TaskList[i * (TaskList.size() / nthreads)];
-          &TaskList[i * (TaskList.size() / nthreads)];
+        }
-    } else {
+        else {
             if (i == nthreads - 1) {
-        InputToThreads[i].taskSize =
+                InputToThreads[i].taskSize = (TaskList.size() / nthreads) + (TaskList.size() % nthreads);
-            (TaskList.size() / nthreads) + (TaskList.size() % nthreads);
                 InputToThreads[i].TaskListPtr =
-            &TaskList[i * (TaskList.size() / nthreads) +
+                    &TaskList[i * (TaskList.size() / nthreads) + (TaskList.size() % nthreads)];
-                      (TaskList.size() % nthreads)];
+            }
-      } else {
+            else {
                 InputToThreads[i].taskSize    = (TaskList.size() / nthreads);
-        InputToThreads[i].TaskListPtr =
+                InputToThreads[i].TaskListPtr = &TaskList[i * (TaskList.size() / nthreads)];
-            &TaskList[i * (TaskList.size() / nthreads)];
             }
         }
 

Samples/0_Introduction/asyncAPI/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/asyncAPI/README.md

@@ -27,6 +27,6 @@ cudaProfilerStop, cudaMalloc, cudaMemcpyAsync, cudaFree, cudaMallocHost, cudaPro
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/asyncAPI/asyncAPI.cu

@@ -38,19 +38,21 @@
 #include <stdio.h>
 
 // includes CUDA Runtime
-#include <cuda_runtime.h>
 #include <cuda_profiler_api.h>
+#include <cuda_runtime.h>
 
 // includes, project
 #include <helper_cuda.h>
 #include <helper_functions.h> // helper utility functions
 
-__global__ void increment_kernel(int *g_data, int inc_value) {
+__global__ void increment_kernel(int *g_data, int inc_value)
+{
     int idx     = blockIdx.x * blockDim.x + threadIdx.x;
     g_data[idx] = g_data[idx] + inc_value;
 }
 
-bool correct_output(int *data, const int n, const int x) {
+bool correct_output(int *data, const int n, const int x)
+{
     for (int i = 0; i < n; i++)
         if (data[i] != x) {
             printf("Error! data[%d] = %d, ref = %d\n", i, data[i], x);
@@ -60,7 +62,8 @@ bool correct_output(int *data, const int n, const int x) {
     return true;
 }
 
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     int            devID;
     cudaDeviceProp deviceProps;
 
@@ -126,8 +129,7 @@ int main(int argc, char *argv[]) {
     // print the cpu and gpu times
     printf("time spent executing by the GPU: %.2f\n", gpu_time);
     printf("time spent by CPU in CUDA calls: %.2f\n", sdkGetTimerValue(&timer));
-  printf("CPU executed %lu iterations while waiting for GPU to finish\n",
+    printf("CPU executed %lu iterations while waiting for GPU to finish\n", counter);
-         counter);
 
     // check the output for correctness
     bool bFinalResults = correct_output(a, n, value);

Samples/0_Introduction/clock/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/clock/README.md

@@ -27,6 +27,6 @@ cudaMalloc, cudaMemcpy, cudaFree
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/clock/clock.cu

@@ -48,15 +48,16 @@
 // This kernel computes a standard parallel reduction and evaluates the
 // time it takes to do that for each block. The timing results are stored
 // in device memory.
-__global__ static void timedReduction(const float *input, float *output,
+__global__ static void timedReduction(const float *input, float *output, clock_t *timer)
-                                      clock_t *timer) {
+{
     // __shared__ float shared[2 * blockDim.x];
     extern __shared__ float shared[];
 
     const int tid = threadIdx.x;
     const int bid = blockIdx.x;
 
-  if (tid == 0) timer[bid] = clock();
+    if (tid == 0)
+        timer[bid] = clock();
 
     // Copy input.
     shared[tid]              = input[tid];
@@ -77,11 +78,13 @@ __global__ static void timedReduction(const float *input, float *output,
     }
 
     // Write result.
-  if (tid == 0) output[bid] = shared[0];
+    if (tid == 0)
+        output[bid] = shared[0];
 
     __syncthreads();
 
-  if (tid == 0) timer[bid + gridDim.x] = clock();
+    if (tid == 0)
+        timer[bid + gridDim.x] = clock();
 }
 
 #define NUM_BLOCKS  64
@@ -104,7 +107,8 @@ __global__ static void timedReduction(const float *input, float *output,
 // the memory. With more than 32 the speed scales linearly.
 
 // Start the main CUDA Sample here
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("CUDA Clock sample\n");
 
     // This will pick the best possible CUDA capable device
@@ -121,20 +125,15 @@ int main(int argc, char **argv) {
         input[i] = (float)i;
     }
 
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&dinput, sizeof(float) * NUM_THREADS * 2));
-      cudaMalloc((void **)&dinput, sizeof(float) * NUM_THREADS * 2));
     checkCudaErrors(cudaMalloc((void **)&doutput, sizeof(float) * NUM_BLOCKS));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
-      cudaMalloc((void **)&dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
 
-  checkCudaErrors(cudaMemcpy(dinput, input, sizeof(float) * NUM_THREADS * 2,
+    checkCudaErrors(cudaMemcpy(dinput, input, sizeof(float) * NUM_THREADS * 2, cudaMemcpyHostToDevice));
-                             cudaMemcpyHostToDevice));
 
-  timedReduction<<<NUM_BLOCKS, NUM_THREADS, sizeof(float) * 2 * NUM_THREADS>>>(
+    timedReduction<<<NUM_BLOCKS, NUM_THREADS, sizeof(float) * 2 * NUM_THREADS>>>(dinput, doutput, dtimer);
-      dinput, doutput, dtimer);
 
-  checkCudaErrors(cudaMemcpy(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2,
+    checkCudaErrors(cudaMemcpy(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2, cudaMemcpyDeviceToHost));
-                             cudaMemcpyDeviceToHost));
 
     checkCudaErrors(cudaFree(dinput));
     checkCudaErrors(cudaFree(doutput));

Samples/0_Introduction/clock_nvrtc/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/clock_nvrtc/README.md

@@ -33,7 +33,7 @@ cudaBlockSize, cudaGridSize
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/clock_nvrtc/clock.cpp

@@ -34,12 +34,11 @@
  */
 
 // System includes
-#include <stdio.h>
-#include <stdint.h>
 #include <assert.h>
-
 #include <cuda_runtime.h>
 #include <nvrtc_helper.h>
+#include <stdint.h>
+#include <stdio.h>
 
 // helper functions and utilities to work with CUDA
 #include <helper_functions.h>
@@ -71,7 +70,8 @@
 
 // Start the main CUDA Sample here
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("CUDA Clock sample\n");
 
     typedef long clock_t;
@@ -106,17 +106,20 @@ int main(int argc, char **argv) {
 
     void *arr[] = {(void *)&dinput, (void *)&doutput, (void *)&dtimer};
 
-  checkCudaErrors(cuLaunchKernel(
+    checkCudaErrors(cuLaunchKernel(kernel_addr,
-      kernel_addr, cudaGridSize.x, cudaGridSize.y,
+                                   cudaGridSize.x,
+                                   cudaGridSize.y,
                                    cudaGridSize.z, /* grid dim */
-      cudaBlockSize.x, cudaBlockSize.y, cudaBlockSize.z, /* block dim */
+                                   cudaBlockSize.x,
-      sizeof(float) * 2 * NUM_THREADS, 0, /* shared mem, stream */
+                                   cudaBlockSize.y,
+                                   cudaBlockSize.z, /* block dim */
+                                   sizeof(float) * 2 * NUM_THREADS,
+                                   0,       /* shared mem, stream */
                                    &arr[0], /* arguments */
                                    0));
 
     checkCudaErrors(cuCtxSynchronize());
-  checkCudaErrors(
+    checkCudaErrors(cuMemcpyDtoH(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
-      cuMemcpyDtoH(timer, dtimer, sizeof(clock_t) * NUM_BLOCKS * 2));
     checkCudaErrors(cuMemFree(dinput));
     checkCudaErrors(cuMemFree(doutput));
     checkCudaErrors(cuMemFree(dtimer));

Samples/0_Introduction/clock_nvrtc/clock_kernel.cu

@@ -37,15 +37,16 @@
 // time it takes to do that for each block. The timing results are stored
 // in device memory.
 
-extern "C" __global__ void timedReduction(const float *input, float *output,
+extern "C" __global__ void timedReduction(const float *input, float *output, clock_t *timer)
-                                          clock_t *timer) {
+{
     // __shared__ float shared[2 * blockDim.x];
     extern __shared__ float shared[];
 
     const int tid = threadIdx.x;
     const int bid = blockIdx.x;
 
-  if (tid == 0) timer[bid] = clock();
+    if (tid == 0)
+        timer[bid] = clock();
 
     // Copy input.
     shared[tid]              = input[tid];
@@ -66,9 +67,11 @@ extern "C" __global__ void timedReduction(const float *input, float *output,
     }
 
     // Write result.
-  if (tid == 0) output[bid] = shared[0];
+    if (tid == 0)
+        output[bid] = shared[0];
 
     __syncthreads();
 
-  if (tid == 0) timer[bid + gridDim.x] = clock();
+    if (tid == 0)
+        timer[bid + gridDim.x] = clock();
 }

Samples/0_Introduction/cudaOpenMP/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/cudaOpenMP/README.md

@@ -30,7 +30,7 @@ cudaMemcpy, cudaGetErrorString, cudaFree, cudaGetLastError, cudaSetDevice, cudaG
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/cudaOpenMP/cudaOpenMP.cu

@@ -37,20 +37,24 @@
 using namespace std;
 
 // a simple kernel that simply increments each array element by b
-__global__ void kernelAddConstant(int *g_a, const int b) {
+__global__ void kernelAddConstant(int *g_a, const int b)
+{
     int idx = blockIdx.x * blockDim.x + threadIdx.x;
     g_a[idx] += b;
 }
 
 // a predicate that checks whether each array element is set to its index plus b
-int correctResult(int *data, const int n, const int b) {
+int correctResult(int *data, const int n, const int b)
+{
     for (int i = 0; i < n; i++)
-    if (data[i] != i + b) return 0;
+        if (data[i] != i + b)
+            return 0;
 
     return 1;
 }
 
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     int num_gpus = 0; // number of CUDA GPUs
 
     printf("%s Starting...\n\n", argv[0]);
@@ -93,7 +97,8 @@ int main(int argc, char *argv[]) {
         return 1;
     }
 
-  for (unsigned int i = 0; i < n; i++) a[i] = i;
+    for (unsigned int i = 0; i < n; i++)
+        a[i] = i;
 
     ////////////////////////////////////////////////////////////////
     // run as many CPU threads as there are CUDA devices
@@ -105,8 +110,7 @@ int main(int argc, char *argv[]) {
     //   Recall that all variables declared inside an "omp parallel" scope are
     //   local to each CPU thread
     //
-  omp_set_num_threads(
+    omp_set_num_threads(num_gpus); // create as many CPU threads as there are CUDA devices
-      num_gpus);  // create as many CPU threads as there are CUDA devices
 // omp_set_num_threads(2*num_gpus);// create twice as many CPU threads as there
 // are CUDA devices
 #pragma omp parallel
@@ -116,31 +120,23 @@ int main(int argc, char *argv[]) {
 
         // set and check the CUDA device for this CPU thread
         int gpu_id = -1;
-    checkCudaErrors(cudaSetDevice(
+        checkCudaErrors(
-        cpu_thread_id %
+            cudaSetDevice(cpu_thread_id % num_gpus)); // "% num_gpus" allows more CPU threads than GPU devices
-        num_gpus));  // "% num_gpus" allows more CPU threads than GPU devices
         checkCudaErrors(cudaGetDevice(&gpu_id));
-    printf("CPU thread %d (of %d) uses CUDA device %d\n", cpu_thread_id,
+        printf("CPU thread %d (of %d) uses CUDA device %d\n", cpu_thread_id, num_cpu_threads, gpu_id);
-           num_cpu_threads, gpu_id);
 
-    int *d_a =
+        int         *d_a   = 0; // pointer to memory on the device associated with this CPU thread
-        0;  // pointer to memory on the device associated with this CPU thread
+        int         *sub_a = a + cpu_thread_id * n / num_cpu_threads; // pointer to this CPU thread's portion of data
-    int *sub_a =
-        a +
-        cpu_thread_id * n /
-            num_cpu_threads;  // pointer to this CPU thread's portion of data
         unsigned int nbytes_per_kernel = nbytes / num_cpu_threads;
         dim3         gpu_threads(128); // 128 threads per block
         dim3         gpu_blocks(n / (gpu_threads.x * num_cpu_threads));
 
         checkCudaErrors(cudaMalloc((void **)&d_a, nbytes_per_kernel));
         checkCudaErrors(cudaMemset(d_a, 0, nbytes_per_kernel));
-    checkCudaErrors(
+        checkCudaErrors(cudaMemcpy(d_a, sub_a, nbytes_per_kernel, cudaMemcpyHostToDevice));
-        cudaMemcpy(d_a, sub_a, nbytes_per_kernel, cudaMemcpyHostToDevice));
         kernelAddConstant<<<gpu_blocks, gpu_threads>>>(d_a, b);
 
-    checkCudaErrors(
+        checkCudaErrors(cudaMemcpy(sub_a, d_a, nbytes_per_kernel, cudaMemcpyDeviceToHost));
-        cudaMemcpy(sub_a, d_a, nbytes_per_kernel, cudaMemcpyDeviceToHost));
         checkCudaErrors(cudaFree(d_a));
     }
     printf("---------------------------\n");
@@ -153,7 +149,8 @@ int main(int argc, char *argv[]) {
     //
     bool bResult = correctResult(a, n, b);
 
-  if (a) free(a);  // free CPU memory
+    if (a)
+        free(a); // free CPU memory
 
     exit(bResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }

Samples/0_Introduction/fp16ScalarProduct/CMakeLists.txt

@@ -9,8 +9,10 @@ find_package(CUDAToolkit REQUIRED)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 60 61 70 72 75 80 86 87 89 90 100 101 120)
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/fp16ScalarProduct/README.md

@@ -30,7 +30,7 @@ cudaMemcpy, cudaFree, cudaMallocHost, cudaFreeHost, cudaMalloc, cudaGetDevicePro
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/fp16ScalarProduct/fp16ScalarProduct.cu

@@ -25,17 +25,18 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include "cuda_fp16.h"
-#include "helper_cuda.h"
-
 #include <cstdio>
 #include <cstdlib>
 #include <ctime>
 
+#include "cuda_fp16.h"
+#include "helper_cuda.h"
+
 #define NUM_OF_BLOCKS  128
 #define NUM_OF_THREADS 128
 
-__forceinline__ __device__ void reduceInShared_intrinsics(half2 *const v) {
+__forceinline__ __device__ void reduceInShared_intrinsics(half2 *const v)
+{
     if (threadIdx.x < 64)
         v[threadIdx.x] = __hadd2(v[threadIdx.x], v[threadIdx.x + 64]);
     __syncthreads();
@@ -59,27 +60,34 @@ __forceinline__ __device__ void reduceInShared_intrinsics(half2 *const v) {
     __syncthreads();
 }
 
-__forceinline__ __device__ void reduceInShared_native(half2 *const v) {
+__forceinline__ __device__ void reduceInShared_native(half2 *const v)
-  if (threadIdx.x < 64) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 64];
+{
+    if (threadIdx.x < 64)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 64];
     __syncthreads();
-  if (threadIdx.x < 32) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 32];
+    if (threadIdx.x < 32)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 32];
     __syncthreads();
-  if (threadIdx.x < 16) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 16];
+    if (threadIdx.x < 16)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 16];
     __syncthreads();
-  if (threadIdx.x < 8) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 8];
+    if (threadIdx.x < 8)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 8];
     __syncthreads();
-  if (threadIdx.x < 4) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 4];
+    if (threadIdx.x < 4)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 4];
     __syncthreads();
-  if (threadIdx.x < 2) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 2];
+    if (threadIdx.x < 2)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 2];
     __syncthreads();
-  if (threadIdx.x < 1) v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 1];
+    if (threadIdx.x < 1)
+        v[threadIdx.x] = v[threadIdx.x] + v[threadIdx.x + 1];
     __syncthreads();
 }
 
-__global__ void scalarProductKernel_intrinsics(half2 const *const a,
+__global__ void
-                                               half2 const *const b,
+scalarProductKernel_intrinsics(half2 const *const a, half2 const *const b, float *const results, size_t const size)
-                                               float *const results,
+{
-                                               size_t const size) {
     const int        stride = gridDim.x * blockDim.x;
     __shared__ half2 shArray[NUM_OF_THREADS];
 
@@ -101,10 +109,9 @@ __global__ void scalarProductKernel_intrinsics(half2 const *const a,
     }
 }
 
-__global__ void scalarProductKernel_native(half2 const *const a,
+__global__ void
-                                           half2 const *const b,
+scalarProductKernel_native(half2 const *const a, half2 const *const b, float *const results, size_t const size)
-                                           float *const results,
+{
-                                           size_t const size) {
     const int        stride = gridDim.x * blockDim.x;
     __shared__ half2 shArray[NUM_OF_THREADS];
 
@@ -126,7 +133,8 @@ __global__ void scalarProductKernel_native(half2 const *const a,
     }
 }
 
-void generateInput(half2 *a, size_t size) {
+void generateInput(half2 *a, size_t size)
+{
     for (size_t i = 0; i < size; ++i) {
         half2 temp;
         temp.x = static_cast<float>(rand() % 4);
@@ -135,7 +143,8 @@ void generateInput(half2 *a, size_t size) {
     }
 }
 
-int main(int argc, char *argv[]) {
+int main(int argc, char *argv[])
+{
     srand((unsigned int)time(NULL));
     size_t size = NUM_OF_BLOCKS * NUM_OF_THREADS * 16;
 
@@ -151,8 +160,7 @@ int main(int argc, char *argv[]) {
     checkCudaErrors(cudaGetDeviceProperties(&devProp, devID));
 
     if (devProp.major < 5 || (devProp.major == 5 && devProp.minor < 3)) {
-    printf(
+        printf("ERROR: fp16ScalarProduct requires GPU devices with compute SM 5.3 or "
-        "ERROR: fp16ScalarProduct requires GPU devices with compute SM 5.3 or "
                "higher.\n");
         return EXIT_WAIVED;
     }
@@ -162,23 +170,17 @@ int main(int argc, char *argv[]) {
         checkCudaErrors(cudaMalloc((void **)&devVec[i], size * sizeof *devVec[i]));
     }
 
-  checkCudaErrors(
+    checkCudaErrors(cudaMallocHost((void **)&results, NUM_OF_BLOCKS * sizeof *results));
-      cudaMallocHost((void **)&results, NUM_OF_BLOCKS * sizeof *results));
+    checkCudaErrors(cudaMalloc((void **)&devResults, NUM_OF_BLOCKS * sizeof *devResults));
-  checkCudaErrors(
-      cudaMalloc((void **)&devResults, NUM_OF_BLOCKS * sizeof *devResults));
 
     for (int i = 0; i < 2; ++i) {
         generateInput(vec[i], size);
-    checkCudaErrors(cudaMemcpy(devVec[i], vec[i], size * sizeof *vec[i],
+        checkCudaErrors(cudaMemcpy(devVec[i], vec[i], size * sizeof *vec[i], cudaMemcpyHostToDevice));
-                               cudaMemcpyHostToDevice));
     }
 
-  scalarProductKernel_native<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(
+    scalarProductKernel_native<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(devVec[0], devVec[1], devResults, size);
-      devVec[0], devVec[1], devResults, size);
 
-  checkCudaErrors(cudaMemcpy(results, devResults,
+    checkCudaErrors(cudaMemcpy(results, devResults, NUM_OF_BLOCKS * sizeof *results, cudaMemcpyDeviceToHost));
-                             NUM_OF_BLOCKS * sizeof *results,
-                             cudaMemcpyDeviceToHost));
 
     float result_native = 0;
     for (int i = 0; i < NUM_OF_BLOCKS; ++i) {
@@ -186,12 +188,9 @@ int main(int argc, char *argv[]) {
     }
     printf("Result native operators\t: %f \n", result_native);
 
-  scalarProductKernel_intrinsics<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(
+    scalarProductKernel_intrinsics<<<NUM_OF_BLOCKS, NUM_OF_THREADS>>>(devVec[0], devVec[1], devResults, size);
-      devVec[0], devVec[1], devResults, size);
 
-  checkCudaErrors(cudaMemcpy(results, devResults,
+    checkCudaErrors(cudaMemcpy(results, devResults, NUM_OF_BLOCKS * sizeof *results, cudaMemcpyDeviceToHost));
-                             NUM_OF_BLOCKS * sizeof *results,
-                             cudaMemcpyDeviceToHost));
 
     float result_intrinsics = 0;
     for (int i = 0; i < NUM_OF_BLOCKS; ++i) {
@@ -199,9 +198,7 @@ int main(int argc, char *argv[]) {
     }
     printf("Result intrinsics\t: %f \n", result_intrinsics);
 
-  printf("&&&& fp16ScalarProduct %s\n",
+    printf("&&&& fp16ScalarProduct %s\n", (fabs(result_intrinsics - result_native) < 0.00001) ? "PASSED" : "FAILED");
-         (fabs(result_intrinsics - result_native) < 0.00001) ? "PASSED"
-                                                             : "FAILED");
 
     for (int i = 0; i < 2; ++i) {
         checkCudaErrors(cudaFree(devVec[i]));

Samples/0_Introduction/matrixMul/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/matrixMul/README.md

@@ -2,7 +2,7 @@
 
 ## Description
 
-This sample implements matrix multiplication and is exactly the same as Chapter 6 of the programming guide. It has been written for clarity of exposition to illustrate various CUDA programming principles, not with the goal of providing the most performant generic kernel for matrix multiplication.  To illustrate GPU performance for matrix multiply, this sample also shows how to use the new CUDA 4.0 interface for CUBLAS to demonstrate high-performance performance for matrix multiplication.
+This sample implements matrix multiplication and is exactly the same as the second example of the [Shared Memory](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory) section of the programming guide. It has been written for clarity of exposition to illustrate various CUDA programming principles, not with the goal of providing the most performant generic kernel for matrix multiplication.  To illustrate GPU performance for matrix multiply, this sample also shows how to use the CUDA 4.0+ interface for cuBLAS to demonstrate high-performance performance for matrix multiplication.
 
 ## Key Concepts
 
@@ -27,6 +27,6 @@ cudaStreamCreateWithFlags, cudaProfilerStop, cudaMalloc, cudaFree, cudaMallocHos
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/matrixMul/matrixMul.cu

@@ -40,24 +40,23 @@
  */
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
+#include <stdio.h>
 
 // CUDA runtime
-#include <cuda_runtime.h>
 #include <cuda_profiler_api.h>
+#include <cuda_runtime.h>
 
 // Helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 /**
  * Matrix multiplication (CUDA Kernel) on the device: C = A * B
  * wA is A's width and wB is B's width
  */
-template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A,
+template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A, float *B, int wA, int wB)
-    float *B, int wA,
+{
-    int wB) {
     // Block index
     int bx = blockIdx.x;
     int by = blockIdx.y;
@@ -87,9 +86,7 @@ template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A,
 
     // Loop over all the sub-matrices of A and B
     // required to compute the block sub-matrix
-  for (int a = aBegin, b = bBegin;
+    for (int a = aBegin, b = bBegin; a <= aEnd; a += aStep, b += bStep) {
-       a <= aEnd;
-       a += aStep, b += bStep) {
         // Declaration of the shared memory array As used to
         // store the sub-matrix of A
         __shared__ float As[BLOCK_SIZE][BLOCK_SIZE];
@@ -128,7 +125,8 @@ template <int BLOCK_SIZE> __global__ void MatrixMulCUDA(float *C, float *A,
     C[c + wB * ty + tx] = Csub;
 }
 
-void ConstantInit(float *data, int size, float val) {
+void ConstantInit(float *data, int size, float val)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = val;
     }
@@ -137,9 +135,8 @@ void ConstantInit(float *data, int size, float val) {
 /**
  * Run a simple test of matrix multiplication using CUDA
  */
-int MatrixMultiply(int argc, char **argv,
+int MatrixMultiply(int argc, char **argv, int block_size, const dim3 &dimsA, const dim3 &dimsB)
-                   int block_size, const dim3 &dimsA,
+{
-                   const dim3 &dimsB) {
     // Allocate host memory for matrices A and B
     unsigned int size_A     = dimsA.x * dimsA.y;
     unsigned int mem_size_A = sizeof(float) * size_A;
@@ -181,10 +178,8 @@ int MatrixMultiply(int argc, char **argv,
     checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
 
     // copy host memory to device
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(d_A, h_A, mem_size_A, cudaMemcpyHostToDevice, stream));
-      cudaMemcpyAsync(d_A, h_A, mem_size_A, cudaMemcpyHostToDevice, stream));
+    checkCudaErrors(cudaMemcpyAsync(d_B, h_B, mem_size_B, cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(
-      cudaMemcpyAsync(d_B, h_B, mem_size_B, cudaMemcpyHostToDevice, stream));
 
     // Setup execution parameters
     dim3 threads(block_size, block_size);
@@ -195,11 +190,10 @@ int MatrixMultiply(int argc, char **argv,
 
     // Performs warmup operation using matrixMul CUDA kernel
     if (block_size == 16) {
-    MatrixMulCUDA<16>
+        MatrixMulCUDA<16><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-        <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
+    }
-  } else {
+    else {
-    MatrixMulCUDA<32>
+        MatrixMulCUDA<32><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-        <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
     }
 
     printf("done\n");
@@ -213,11 +207,10 @@ int MatrixMultiply(int argc, char **argv,
 
     for (int j = 0; j < nIter; j++) {
         if (block_size == 16) {
-      MatrixMulCUDA<16>
+            MatrixMulCUDA<16><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-          <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
+        }
-    } else {
+        else {
-      MatrixMulCUDA<32>
+            MatrixMulCUDA<32><<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
-          <<<grid, threads, 0, stream>>>(d_C, d_A, d_B, dimsA.x, dimsB.x);
         }
     }
 
@@ -232,19 +225,18 @@ int MatrixMultiply(int argc, char **argv,
 
     // Compute and print the performance
     float  msecPerMatrixMul = msecTotal / nIter;
-  double flopsPerMatrixMul = 2.0 * static_cast<double>(dimsA.x) *
+    double flopsPerMatrixMul =
-                             static_cast<double>(dimsA.y) *
+        2.0 * static_cast<double>(dimsA.x) * static_cast<double>(dimsA.y) * static_cast<double>(dimsB.x);
-                             static_cast<double>(dimsB.x);
+    double gigaFlops = (flopsPerMatrixMul * 1.0e-9f) / (msecPerMatrixMul / 1000.0f);
-  double gigaFlops =
+    printf("Performance= %.2f GFlop/s, Time= %.3f msec, Size= %.0f Ops,"
-      (flopsPerMatrixMul * 1.0e-9f) / (msecPerMatrixMul / 1000.0f);
-  printf(
-      "Performance= %.2f GFlop/s, Time= %.3f msec, Size= %.0f Ops,"
            " WorkgroupSize= %u threads/block\n",
-      gigaFlops, msecPerMatrixMul, flopsPerMatrixMul, threads.x * threads.y);
+           gigaFlops,
+           msecPerMatrixMul,
+           flopsPerMatrixMul,
+           threads.x * threads.y);
 
     // Copy result from device to host
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(h_C, d_C, mem_size_C, cudaMemcpyDeviceToHost, stream));
-      cudaMemcpyAsync(h_C, d_C, mem_size_C, cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
 
     printf("Checking computed result for correctness: ");
@@ -261,8 +253,7 @@ int MatrixMultiply(int argc, char **argv,
         double rel_err    = abs_err / abs_val / dot_length;
 
         if (rel_err > eps) {
-      printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n",
+            printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i, h_C[i], dimsA.x * valB, eps);
-             i, h_C[i], dimsA.x * valB, eps);
             correct = false;
         }
     }
@@ -278,13 +269,13 @@ int MatrixMultiply(int argc, char **argv,
     checkCudaErrors(cudaFree(d_C));
     checkCudaErrors(cudaEventDestroy(start));
     checkCudaErrors(cudaEventDestroy(stop));
-  printf(
+    printf("\nNOTE: The CUDA Samples are not meant for performance "
-      "\nNOTE: The CUDA Samples are not meant for performance "
            "measurements. Results may vary when GPU Boost is enabled.\n");
 
     if (correct) {
         return EXIT_SUCCESS;
-  } else {
+    }
+    else {
         return EXIT_FAILURE;
     }
 }
@@ -293,15 +284,15 @@ int MatrixMultiply(int argc, char **argv,
 /**
  * Program main
  */
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[Matrix Multiply Using CUDA] - Starting...\n");
 
-  if (checkCmdLineFlag(argc, (const char **)argv, "help") ||
+    if (checkCmdLineFlag(argc, (const char **)argv, "help") || checkCmdLineFlag(argc, (const char **)argv, "?")) {
-      checkCmdLineFlag(argc, (const char **)argv, "?")) {
         printf("Usage -device=n (n >= 0 for deviceID)\n");
         printf("      -wA=WidthA -hA=HeightA (Width x Height of Matrix A)\n");
         printf("      -wB=WidthB -hB=HeightB (Width x Height of Matrix B)\n");
-    printf("  Note: Outer matrix dimensions of A & B matrices" \
+        printf("  Note: Outer matrix dimensions of A & B matrices"
                " must be equal.\n");
 
         exit(EXIT_SUCCESS);
@@ -337,13 +328,11 @@ int main(int argc, char **argv) {
     }
 
     if (dimsA.x != dimsB.y) {
-    printf("Error: outer matrix dimensions must be equal. (%d != %d)\n",
+        printf("Error: outer matrix dimensions must be equal. (%d != %d)\n", dimsA.x, dimsB.y);
-           dimsA.x, dimsB.y);
         exit(EXIT_FAILURE);
     }
 
-  printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y,
+    printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x, dimsB.y);
-         dimsB.x, dimsB.y);
 
     checkCudaErrors(cudaProfilerStart());
     int matrix_result = MatrixMultiply(argc, argv, block_size, dimsA, dimsB);

Samples/0_Introduction/matrixMulDrv/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries
@@ -38,6 +40,12 @@ target_link_libraries(matrixMulDrv PUBLIC
 set(CUDA_FATBIN_FILE "${CMAKE_CURRENT_BINARY_DIR}/matrixMul_kernel64.fatbin")
 set(CUDA_KERNEL_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/matrixMul_kernel.cu")
 
+# Construct GENCODE_FLAGS explicitly from CUDA architectures
+set(GENCODE_FLAGS "")
+foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
+    list(APPEND GENCODE_FLAGS "-gencode=arch=compute_${arch},code=sm_${arch}")
+endforeach()
+
 add_custom_command(
     OUTPUT ${CUDA_FATBIN_FILE}
     COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets  ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}

Samples/0_Introduction/matrixMulDrv/README.md

@@ -27,6 +27,6 @@ cuMemcpyDtoH, cuLaunchKernel, cuMemcpyHtoD, cuDeviceGetName, cuDeviceTotalMem, c
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/matrixMulDrv/matrixMulDrv.cpp

@@ -46,23 +46,23 @@
 
 // includes, system
 #include <builtin_types.h>
-#include <math.h>
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
-#include <iostream>
 #include <cstring>
+#include <iostream>
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 // includes, project, CUDA
+#include <cstring>
 #include <cuda.h>
 #include <helper_cuda_drvapi.h>
 #include <helper_image.h>
 #include <helper_string.h>
 #include <helper_timer.h>
-
-#include <cstring>
 #include <iostream>
 #include <string>
+
 #include "matrixMul.h"
 
 
@@ -71,11 +71,9 @@
 void runTest(int argc, char **argv);
 void randomInit(float *, int);
 
-extern "C" void computeGold(float *, const float *, const float *, unsigned int,
+extern "C" void computeGold(float *, const float *, const float *, unsigned int, unsigned int, unsigned int);
-                            unsigned int, unsigned int);
 
-static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
+static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *blk_size);
-                    int *blk_size);
 
 #ifndef FATBIN_FILE
 #define FATBIN_FILE "matrixMul_kernel64.fatbin"
@@ -91,7 +89,8 @@ size_t totalGlobalMem;
 
 const char *sSDKsample = "matrixMulDrv (Driver API)";
 
-void constantInit(float *data, int size, float val) {
+void constantInit(float *data, int size, float val)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = val;
     }
@@ -100,7 +99,8 @@ void constantInit(float *data, int size, float val) {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[ %s ]\n", sSDKsample);
 
     runTest(argc, argv);
@@ -109,7 +109,8 @@ int main(int argc, char **argv) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     // initialize CUDA
     CUfunction matrixMul  = NULL;
     int        block_size = 0;
@@ -172,10 +173,19 @@ void runTest(int argc, char **argv) {
         size_t Matrix_Width_B = (size_t)WB;
         void  *args[5]        = {&d_C, &d_A, &d_B, &Matrix_Width_A, &Matrix_Width_B};
         // new CUDA 4.0 Driver API Kernel launch call
-    checkCudaErrors(cuLaunchKernel(
+        checkCudaErrors(cuLaunchKernel(matrixMul,
-        matrixMul, grid.x, grid.y, grid.z, block.x, block.y, block.z,
+                                       grid.x,
-        2 * block_size * block_size * sizeof(float), NULL, args, NULL));
+                                       grid.y,
-  } else {
+                                       grid.z,
+                                       block.x,
+                                       block.y,
+                                       block.z,
+                                       2 * block_size * block_size * sizeof(float),
+                                       NULL,
+                                       args,
+                                       NULL));
+    }
+    else {
         // This is the new CUDA 4.0 API for Kernel Parameter passing and Kernel
         // Launching (advanced method)
         int  offset = 0;
@@ -198,14 +208,20 @@ void runTest(int argc, char **argv) {
         *(reinterpret_cast<CUdeviceptr *>(&argBuffer[offset])) = Matrix_Width_B;
         offset += sizeof(Matrix_Width_B);
 
-    void *kernel_launch_config[5] = {CU_LAUNCH_PARAM_BUFFER_POINTER, argBuffer,
+        void *kernel_launch_config[5] = {
-                                     CU_LAUNCH_PARAM_BUFFER_SIZE, &offset,
+            CU_LAUNCH_PARAM_BUFFER_POINTER, argBuffer, CU_LAUNCH_PARAM_BUFFER_SIZE, &offset, CU_LAUNCH_PARAM_END};
-                                     CU_LAUNCH_PARAM_END};
 
         // new CUDA 4.0 Driver API Kernel launch call
-    checkCudaErrors(cuLaunchKernel(
+        checkCudaErrors(cuLaunchKernel(matrixMul,
-        matrixMul, grid.x, grid.y, grid.z, block.x, block.y, block.z,
+                                       grid.x,
-        2 * block_size * block_size * sizeof(float), NULL, NULL,
+                                       grid.y,
+                                       grid.z,
+                                       block.x,
+                                       block.y,
+                                       block.z,
+                                       2 * block_size * block_size * sizeof(float),
+                                       NULL,
+                                       NULL,
                                        reinterpret_cast<void **>(&kernel_launch_config)));
     }
 
@@ -222,8 +238,7 @@ void runTest(int argc, char **argv) {
 
     for (int i = 0; i < static_cast<int>(WC * HC); i++) {
         if (fabs(h_C[i] - (WA * valB)) > 1e-5) {
-      printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > 1e-5\n", i,
+            printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > 1e-5\n", i, h_C[i], WA * valB);
-             h_C[i], WA * valB);
             correct = false;
         }
     }
@@ -244,14 +259,15 @@ void runTest(int argc, char **argv) {
 }
 
 // Allocates a matrix with random float entries.
-void randomInit(float *data, int size) {
+void randomInit(float *data, int size)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = rand() / static_cast<float>(RAND_MAX);
     }
 }
 
-static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
+static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *blk_size)
-                    int *blk_size) {
+{
     CUfunction cuFunction = 0;
     int        major = 0, minor = 0;
     char       deviceName[100];
@@ -259,16 +275,13 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
     cuDevice = findCudaDeviceDRV(argc, (const char **)argv);
 
     // get compute capabilities and the devicename
-  checkCudaErrors(cuDeviceGetAttribute(
+    checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
-      &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
+    checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
-  checkCudaErrors(cuDeviceGetAttribute(
-      &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
     checkCudaErrors(cuDeviceGetName(deviceName, sizeof(deviceName), cuDevice));
     printf("> GPU Device has SM %d.%d compute capability\n", major, minor);
 
     checkCudaErrors(cuDeviceTotalMem(&totalGlobalMem, cuDevice));
-  printf("  Total amount of global memory:     %llu bytes\n",
+    printf("  Total amount of global memory:     %llu bytes\n", (long long unsigned int)totalGlobalMem);
-         (long long unsigned int)totalGlobalMem);
 
     checkCudaErrors(cuCtxCreate(&cuContext, 0, cuDevice));
 
@@ -278,7 +291,8 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
 
     if (!findFatbinPath(FATBIN_FILE, module_path, argv, fatbin)) {
         exit(EXIT_FAILURE);
-  } else {
+    }
+    else {
         printf("> initCUDA loading module: <%s>\n", module_path.c_str());
     }
 
@@ -291,8 +305,7 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
     checkCudaErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));
 
     // select the suitable kernel function
-  const char *kernels[] = {"matrixMul_bs32_64bit", "matrixMul_bs16_64bit",
+    const char *kernels[] = {"matrixMul_bs32_64bit", "matrixMul_bs16_64bit", "matrixMul_bs8_64bit"};
-                           "matrixMul_bs8_64bit"};
 
     int idx        = 0;
     int block_size = 32;
@@ -302,12 +315,12 @@ static int initCUDA(int argc, char **argv, CUfunction *pMatrixMul,
 
         checkCudaErrors(cuModuleGetFunction(&cuFunction, cuModule, kernels[idx]));
         checkCudaErrors(cuOccupancyMaxPotentialBlockSize(
-        &blocksPerGrid, &threadsPerBlock, cuFunction, 0,
+            &blocksPerGrid, &threadsPerBlock, cuFunction, 0, 2 * block_size * block_size * sizeof(float), 0));
-        2 * block_size * block_size * sizeof(float), 0));
         if (block_size * block_size <= threadsPerBlock) {
             printf("> %d block size selected\n", block_size);
             break;
-    } else {
+        }
+        else {
             block_size /= 2;
         }
         idx++;

Samples/0_Introduction/matrixMulDrv/matrixMul_kernel.cu

@@ -42,8 +42,8 @@
 //! wA is A's width and wB is B's width
 ////////////////////////////////////////////////////////////////////////////////
 template <int block_size, typename size_type>
-__device__ void matrixMul(float *C, float *A, float *B, size_type wA,
+__device__ void matrixMul(float *C, float *A, float *B, size_type wA, size_type wB)
-                          size_type wB) {
+{
     // Block index
     size_type bx = blockIdx.x;
     size_type by = blockIdx.y;
@@ -96,7 +96,8 @@ __device__ void matrixMul(float *C, float *A, float *B, size_type wA,
         // of the block sub-matrix
 #pragma unroll
 
-    for (size_type k = 0; k < block_size; ++k) Csub += AS(ty, k) * BS(k, tx);
+        for (size_type k = 0; k < block_size; ++k)
+            Csub += AS(ty, k) * BS(k, tx);
 
         // Synchronize to make sure that the preceding
         // computation is done before loading two new
@@ -111,16 +112,16 @@ __device__ void matrixMul(float *C, float *A, float *B, size_type wA,
 }
 
 // C wrappers around our template kernel
-extern "C" __global__ void matrixMul_bs8_64bit(float *C, float *A, float *B,
+extern "C" __global__ void matrixMul_bs8_64bit(float *C, float *A, float *B, size_t wA, size_t wB)
-                                               size_t wA, size_t wB) {
+{
     matrixMul<8, size_t>(C, A, B, wA, wB);
 }
-extern "C" __global__ void matrixMul_bs16_64bit(float *C, float *A, float *B,
+extern "C" __global__ void matrixMul_bs16_64bit(float *C, float *A, float *B, size_t wA, size_t wB)
-                                                size_t wA, size_t wB) {
+{
     matrixMul<16, size_t>(C, A, B, wA, wB);
 }
-extern "C" __global__ void matrixMul_bs32_64bit(float *C, float *A, float *B,
+extern "C" __global__ void matrixMul_bs32_64bit(float *C, float *A, float *B, size_t wA, size_t wB)
-                                                size_t wA, size_t wB) {
+{
     matrixMul<32, size_t>(C, A, B, wA, wB);
 }
 

Samples/0_Introduction/matrixMulDynlinkJIT/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/matrixMulDynlinkJIT/README.md

@@ -27,6 +27,6 @@ cuMemcpyDtoH, cuDeviceGetName, cuParamSeti, cuModuleLoadDataEx, cuModuleGetFunct
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/matrixMulDynlinkJIT/cuda_drvapi_dynlink.c

@@ -20,9 +20,10 @@
 // #define CUDA_INIT_D3D11
 // #define CUDA_INIT_OPENGL
 
-#include <stdio.h>
 #include "cuda_drvapi_dynlink.h"
 
+#include <stdio.h>
+
 tcuInit                    *_cuInit;
 tcuDriverGetVersion        *cuDriverGetVersion;
 tcuDeviceGet               *cuDeviceGet;
@@ -239,8 +240,7 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 {
     *pInstance = LoadLibrary(__CudaLibName);
 
-    if (*pInstance == NULL)
+    if (*pInstance == NULL) {
-    {
         printf("LoadLibrary \"%s\" failed!\n", __CudaLibName);
         return CUDA_ERROR_UNKNOWN;
     }
@@ -251,24 +251,21 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 #define GET_PROC_EX(name, alias, required)                                               \
     alias = (t##name *)GetProcAddress(CudaDrvLib, #name);                                \
     if (alias == NULL && required) {                                                     \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", #name, __CudaLibName); \
-               #name, __CudaLibName);                                  \
         return CUDA_ERROR_UNKNOWN;                                                       \
     }
 
 #define GET_PROC_EX_V2(name, alias, required)                                                           \
     alias = (t##name *)GetProcAddress(CudaDrvLib, STRINGIFY(name##_v2));                                \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v2), __CudaLibName); \
-               STRINGIFY(name##_v2), __CudaLibName);                       \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
 #define GET_PROC_EX_V3(name, alias, required)                                                           \
     alias = (t##name *)GetProcAddress(CudaDrvLib, STRINGIFY(name##_v3));                                \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v3), __CudaLibName); \
-               STRINGIFY(name##_v3), __CudaLibName);                       \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
@@ -294,8 +291,7 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 {
     *pInstance = dlopen(__CudaLibName, RTLD_NOW);
 
-    if (*pInstance == NULL)
+    if (*pInstance == NULL) {
-    {
         printf("dlopen \"%s\" failed!\n", __CudaLibName);
         return CUDA_ERROR_UNKNOWN;
     }
@@ -306,24 +302,21 @@ static CUresult LOAD_LIBRARY(CUDADRIVER *pInstance)
 #define GET_PROC_EX(name, alias, required)                                               \
     alias = (t##name *)dlsym(CudaDrvLib, #name);                                         \
     if (alias == NULL && required) {                                                     \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", #name, __CudaLibName); \
-               #name, __CudaLibName);                                  \
         return CUDA_ERROR_UNKNOWN;                                                       \
     }
 
 #define GET_PROC_EX_V2(name, alias, required)                                                           \
     alias = (t##name *)dlsym(CudaDrvLib, STRINGIFY(name##_v2));                                         \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v2), __CudaLibName); \
-               STRINGIFY(name##_v2), __CudaLibName);                    \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
 #define GET_PROC_EX_V3(name, alias, required)                                                           \
     alias = (t##name *)dlsym(CudaDrvLib, STRINGIFY(name##_v3));                                         \
     if (alias == NULL && required) {                                                                    \
-        printf("Failed to find required function \"%s\" in %s\n",       \
+        printf("Failed to find required function \"%s\" in %s\n", STRINGIFY(name##_v3), __CudaLibName); \
-               STRINGIFY(name##_v3), __CudaLibName);                    \
         return CUDA_ERROR_UNKNOWN;                                                                      \
     }
 
@@ -359,8 +352,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     // available since 2.2. if not present, version 1.0 is assumed
     GET_PROC_OPTIONAL(cuDriverGetVersion);
 
-    if (cuDriverGetVersion)
+    if (cuDriverGetVersion) {
-    {
         CHECKED_CALL(cuDriverGetVersion(&driverVer));
     }
 
@@ -428,24 +420,21 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     GET_PROC(cuStreamDestroy);
 
     // These are CUDA 5.0 new functions
-    if (driverVer >= 5000)
+    if (driverVer >= 5000) {
-    {
         GET_PROC(cuMipmappedArrayCreate);
         GET_PROC(cuMipmappedArrayDestroy);
         GET_PROC(cuMipmappedArrayGetLevel);
     }
 
     // These are CUDA 4.2 new functions
-    if (driverVer >= 4020)
+    if (driverVer >= 4020) {
-    {
         GET_PROC(cuFuncSetSharedMemConfig);
         GET_PROC(cuCtxGetSharedMemConfig);
         GET_PROC(cuCtxSetSharedMemConfig);
     }
 
     // These are CUDA 4.1 new functions
-    if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010)
+    if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010) {
-    {
         GET_PROC(cuDeviceGetByPCIBusId);
         GET_PROC(cuDeviceGetPCIBusId);
         GET_PROC(cuIpcGetEventHandle);
@@ -456,8 +445,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     }
 
     // These could be _v2 interfaces
-    if (cudaVersion >= 4000 && __CUDA_API_VERSION >= 4000)
+    if (cudaVersion >= 4000 && __CUDA_API_VERSION >= 4000) {
-    {
         GET_PROC_V2(cuCtxDestroy);
         GET_PROC_V2(cuCtxPopCurrent);
         GET_PROC_V2(cuCtxPushCurrent);
@@ -465,8 +453,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC_V2(cuEventDestroy);
     }
 
-    if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020)
+    if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020) {
-    {
         GET_PROC_V2(cuDeviceTotalMem);
         GET_PROC_V2(cuCtxCreate);
         GET_PROC_V2(cuModuleGetGlobal);
@@ -507,17 +494,14 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC_V2(cuTexRefSetAddress);
         GET_PROC_V2(cuTexRefGetAddress);
 
-        if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010)
+        if (cudaVersion >= 4010 && __CUDA_API_VERSION >= 4010) {
-        {
             GET_PROC_V3(cuTexRefSetAddress2D);
         }
-        else
+        else {
-        {
             GET_PROC_V2(cuTexRefSetAddress2D);
         }
     }
-    else
+    else {
-    {
         // versions earlier than 3020
         GET_PROC(cuDeviceTotalMem);
         GET_PROC(cuCtxCreate);
@@ -562,8 +546,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
     }
 
     // The following functions are specific to CUDA versions
-    if (driverVer >= 4000)
+    if (driverVer >= 4000) {
-    {
         GET_PROC(cuCtxSetCurrent);
         GET_PROC(cuCtxGetCurrent);
         GET_PROC(cuMemHostRegister);
@@ -574,8 +557,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuProfilerStop);
     }
 
-    if (driverVer >= 3010)
+    if (driverVer >= 3010) {
-    {
         GET_PROC(cuModuleGetSurfRef);
         GET_PROC(cuSurfRefSetArray);
         GET_PROC(cuSurfRefGetArray);
@@ -583,8 +565,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuCtxGetLimit);
     }
 
-    if (driverVer >= 3000)
+    if (driverVer >= 3000) {
-    {
         GET_PROC(cuMemcpyDtoDAsync);
         GET_PROC(cuFuncSetCacheConfig);
 #ifdef CUDA_INIT_D3D11
@@ -595,12 +576,10 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuGraphicsUnregisterResource);
         GET_PROC(cuGraphicsSubResourceGetMappedArray);
 
-        if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020)
+        if (cudaVersion >= 3020 && __CUDA_API_VERSION >= 3020) {
-        {
             GET_PROC_V2(cuGraphicsResourceGetMappedPointer);
         }
-        else
+        else {
-        {
             GET_PROC(cuGraphicsResourceGetMappedPointer);
         }
 
@@ -610,8 +589,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
         GET_PROC(cuGetExportTable);
     }
 
-    if (driverVer >= 2030)
+    if (driverVer >= 2030) {
-    {
         GET_PROC(cuMemHostGetFlags);
 #ifdef CUDA_INIT_D3D10
         GET_PROC(cuD3D10GetDevice);
@@ -624,8 +602,7 @@ CUresult CUDAAPI cuInit(unsigned int Flags, int cudaVersion)
 #endif
     }
 
-    if (driverVer >= 2010)
+    if (driverVer >= 2010) {
-    {
         GET_PROC(cuModuleLoadDataEx);
         GET_PROC(cuModuleLoadFatBinary);
 #ifdef CUDA_INIT_OPENGL

Samples/0_Introduction/matrixMulDynlinkJIT/cuda_drvapi_dynlink_cuda.h

@@ -43,7 +43,8 @@
 #define CUDA_VERSION 3020 /* 3.2 */
 
 #ifdef __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
 /**
@@ -81,8 +82,7 @@ typedef struct CUuuid_st                                  /**< CUDA definition o
     /**
      * Context creation flags
      */
-typedef enum CUctx_flags_enum
+    typedef enum CUctx_flags_enum {
-{
         CU_CTX_SCHED_AUTO          = 0x00, /**< Automatic scheduling */
         CU_CTX_SCHED_SPIN          = 0x01, /**< Set spin as default scheduling */
         CU_CTX_SCHED_YIELD         = 0x02, /**< Set yield as default scheduling */
@@ -103,8 +103,7 @@ typedef enum CUctx_flags_enum
     /**
      * Event creation flags
      */
-typedef enum CUevent_flags_enum
+    typedef enum CUevent_flags_enum {
-{
         CU_EVENT_DEFAULT        = 0, /**< Default event flag */
         CU_EVENT_BLOCKING_SYNC  = 1, /**< Event uses blocking synchronization */
         CU_EVENT_DISABLE_TIMING = 2  /**< Event will not record timing data */
@@ -113,8 +112,7 @@ typedef enum CUevent_flags_enum
     /**
      * Array formats
      */
-typedef enum CUarray_format_enum
+    typedef enum CUarray_format_enum {
-{
         CU_AD_FORMAT_UNSIGNED_INT8  = 0x01, /**< Unsigned 8-bit integers */
         CU_AD_FORMAT_UNSIGNED_INT16 = 0x02, /**< Unsigned 16-bit integers */
         CU_AD_FORMAT_UNSIGNED_INT32 = 0x03, /**< Unsigned 32-bit integers */
@@ -128,8 +126,7 @@ typedef enum CUarray_format_enum
     /**
      * Texture reference addressing modes
      */
-typedef enum CUaddress_mode_enum
+    typedef enum CUaddress_mode_enum {
-{
         CU_TR_ADDRESS_MODE_WRAP   = 0, /**< Wrapping address mode */
         CU_TR_ADDRESS_MODE_CLAMP  = 1, /**< Clamp to edge address mode */
         CU_TR_ADDRESS_MODE_MIRROR = 2, /**< Mirror address mode */
@@ -139,8 +136,7 @@ typedef enum CUaddress_mode_enum
     /**
      * Texture reference filtering modes
      */
-typedef enum CUfilter_mode_enum
+    typedef enum CUfilter_mode_enum {
-{
         CU_TR_FILTER_MODE_POINT  = 0, /**< Point filter mode */
         CU_TR_FILTER_MODE_LINEAR = 1  /**< Linear filter mode */
     } CUfilter_mode;
@@ -148,8 +144,7 @@ typedef enum CUfilter_mode_enum
     /**
      * Device properties
      */
-typedef enum CUdevice_attribute_enum
+    typedef enum CUdevice_attribute_enum {
-{
         CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK       = 1, /**< Maximum number of threads per block */
         CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X             = 2, /**< Maximum block dimension X */
         CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y             = 3, /**< Maximum block dimension Y */
@@ -158,12 +153,15 @@ typedef enum CUdevice_attribute_enum
         CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y              = 6, /**< Maximum grid dimension Y */
         CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z              = 7, /**< Maximum grid dimension Z */
         CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK = 8, /**< Maximum shared memory available per block in bytes */
-    CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK = 8,            /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK */
+        CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK =
-    CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY = 9,              /**< Memory available on device for __constant__ variables in a CUDA C kernel in bytes */
+            8, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK */
+        CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY =
+            9, /**< Memory available on device for __constant__ variables in a CUDA C kernel in bytes */
         CU_DEVICE_ATTRIBUTE_WARP_SIZE               = 10, /**< Warp size in threads */
         CU_DEVICE_ATTRIBUTE_MAX_PITCH               = 11, /**< Maximum pitch in bytes allowed by memory copies */
         CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK = 12, /**< Maximum number of 32-bit registers available per block */
-    CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK = 12,               /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK */
+        CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK =
+            12,                                     /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK */
         CU_DEVICE_ATTRIBUTE_CLOCK_RATE        = 13, /**< Peak clock frequency in kilohertz */
         CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT = 14, /**< Alignment requirement for textures */
         CU_DEVICE_ATTRIBUTE_GPU_OVERLAP = 15, /**< Device can possibly copy memory and execute a kernel concurrently */
@@ -190,7 +188,8 @@ typedef enum CUdevice_attribute_enum
         CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = 75, /**< Major compute capability version number */
         CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR = 76  /**< Minor compute capability version number */
 #if __CUDA_API_VERSION >= 4000
-                                     , CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE = 36,                 /**< Peak memory clock frequency in kilohertz */
+        ,
+        CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE              = 36, /**< Peak memory clock frequency in kilohertz */
         CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH        = 37, /**< Global memory bus width in bits */
         CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE                  = 38, /**< Size of L2 cache in bytes */
         CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR = 39, /**< Maximum resident threads per multiprocessor */
@@ -221,8 +220,7 @@ typedef struct CUdevprop_st
     /**
      * Function properties
      */
-typedef enum CUfunction_attribute_enum
+    typedef enum CUfunction_attribute_enum {
-{
         /**
          * The maximum number of threads per block, beyond which a launch of the
          * function would fail. This number depends on both the function and the
@@ -277,8 +275,7 @@ typedef enum CUfunction_attribute_enum
     /**
      * Function cache configurations
      */
-typedef enum CUfunc_cache_enum
+    typedef enum CUfunc_cache_enum {
-{
         CU_FUNC_CACHE_PREFER_NONE   = 0x00, /**< no preference for shared memory or L1 (default) */
         CU_FUNC_CACHE_PREFER_SHARED = 0x01, /**< prefer larger shared memory and smaller L1 cache */
         CU_FUNC_CACHE_PREFER_L1     = 0x02  /**< prefer larger L1 cache and smaller shared memory */
@@ -287,8 +284,7 @@ typedef enum CUfunc_cache_enum
     /**
      * Shared memory configurations
      */
-typedef enum CUsharedconfig_enum
+    typedef enum CUsharedconfig_enum {
-{
         CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE    = 0x00, /**< set default shared memory bank size */
         CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE  = 0x01, /**< set shared memory bank width to four bytes */
         CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE = 0x02  /**< set shared memory bank width to eight bytes */
@@ -297,33 +293,34 @@ typedef enum CUsharedconfig_enum
     /**
      * Memory types
      */
-typedef enum CUmemorytype_enum
+    typedef enum CUmemorytype_enum {
-{
         CU_MEMORYTYPE_HOST   = 0x01, /**< Host memory */
         CU_MEMORYTYPE_DEVICE = 0x02, /**< Device memory */
         CU_MEMORYTYPE_ARRAY  = 0x03  /**< Array memory */
 #if __CUDA_API_VERSION >= 4000
-                            , CU_MEMORYTYPE_UNIFIED = 0x04     /**< Unified device or host memory */
+        ,
+        CU_MEMORYTYPE_UNIFIED = 0x04 /**< Unified device or host memory */
 #endif
     } CUmemorytype;
 
     /**
      * Compute Modes
      */
-typedef enum CUcomputemode_enum
+    typedef enum CUcomputemode_enum {
-{
         CU_COMPUTEMODE_DEFAULT = 0, /**< Default compute mode (Multiple contexts allowed per device) */
-    CU_COMPUTEMODE_PROHIBITED        = 2  /**< Compute-prohibited mode (No contexts can be created on this device at this time) */
+        CU_COMPUTEMODE_PROHIBITED =
+            2 /**< Compute-prohibited mode (No contexts can be created on this device at this time) */
 #if __CUDA_API_VERSION >= 4000
-                                       , CU_COMPUTEMODE_EXCLUSIVE_PROCESS = 3  /**< Compute-exclusive-process mode (Only one context used by a single process can be present on this device at a time) */
+        ,
+        CU_COMPUTEMODE_EXCLUSIVE_PROCESS = 3 /**< Compute-exclusive-process mode (Only one context used by a single
+                                                process can be present on this device at a time) */
 #endif
     } CUcomputemode;
 
     /**
      * Online compiler options
      */
-typedef enum CUjit_option_enum
+    typedef enum CUjit_option_enum {
-{
         /**
          * Max number of registers that a thread may use.\n
          * Option type: unsigned int
@@ -414,8 +411,7 @@ typedef enum CUjit_option_enum
     /**
      * Online compilation targets
      */
-typedef enum CUjit_target_enum
+    typedef enum CUjit_target_enum {
-{
         CU_TARGET_COMPUTE_20 = 20, /**< Compute device class 2.0 */
         CU_TARGET_COMPUTE_21 = 21, /**< Compute device class 2.1 */
         CU_TARGET_COMPUTE_30 = 30, /**< Compute device class 3.0 */
@@ -434,8 +430,7 @@ typedef enum CUjit_target_enum
     /**
      * Cubin matching fallback strategies
      */
-typedef enum CUjit_fallback_enum
+    typedef enum CUjit_fallback_enum {
-{
         CU_PREFER_PTX = 0, /**< Prefer to compile ptx */
         CU_PREFER_BINARY   /**< Prefer to fall back to compatible binary code */
     } CUjit_fallback;
@@ -443,8 +438,7 @@ typedef enum CUjit_fallback_enum
     /**
      * Flags to register a graphics resource
      */
-typedef enum CUgraphicsRegisterFlags_enum
+    typedef enum CUgraphicsRegisterFlags_enum {
-{
         CU_GRAPHICS_REGISTER_FLAGS_NONE          = 0x00,
         CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY     = 0x01,
         CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD = 0x02,
@@ -454,8 +448,7 @@ typedef enum CUgraphicsRegisterFlags_enum
     /**
      * Flags for mapping and unmapping interop resources
      */
-typedef enum CUgraphicsMapResourceFlags_enum
+    typedef enum CUgraphicsMapResourceFlags_enum {
-{
         CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE          = 0x00,
         CU_GRAPHICS_MAP_RESOURCE_FLAGS_READ_ONLY     = 0x01,
         CU_GRAPHICS_MAP_RESOURCE_FLAGS_WRITE_DISCARD = 0x02
@@ -464,8 +457,7 @@ typedef enum CUgraphicsMapResourceFlags_enum
     /**
      * Array indices for cube faces
      */
-typedef enum CUarray_cubemap_face_enum
+    typedef enum CUarray_cubemap_face_enum {
-{
         CU_CUBEMAP_FACE_POSITIVE_X = 0x00, /**< Positive X face of cubemap */
         CU_CUBEMAP_FACE_NEGATIVE_X = 0x01, /**< Negative X face of cubemap */
         CU_CUBEMAP_FACE_POSITIVE_Y = 0x02, /**< Positive Y face of cubemap */
@@ -477,8 +469,7 @@ typedef enum CUarray_cubemap_face_enum
     /**
      * Limits
      */
-typedef enum CUlimit_enum
+    typedef enum CUlimit_enum {
-{
         CU_LIMIT_STACK_SIZE       = 0x00, /**< GPU thread stack size */
         CU_LIMIT_PRINTF_FIFO_SIZE = 0x01, /**< GPU printf FIFO size */
         CU_LIMIT_MALLOC_HEAP_SIZE = 0x02  /**< GPU malloc heap size */
@@ -487,8 +478,7 @@ typedef enum CUlimit_enum
     /**
      * Resource types
      */
-typedef enum CUresourcetype_enum
+    typedef enum CUresourcetype_enum {
-{
         CU_RESOURCE_TYPE_ARRAY           = 0x00, /**< Array resoure */
         CU_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01, /**< Mipmapped array resource */
         CU_RESOURCE_TYPE_LINEAR          = 0x02, /**< Linear resource */
@@ -498,8 +488,7 @@ typedef enum CUresourcetype_enum
     /**
      * Error codes
      */
-typedef enum cudaError_enum
+    typedef enum cudaError_enum {
-{
         /**
          * The API call returned with no errors. In the case of query calls, this
          * can also mean that the operation being queried is complete (see
@@ -1064,8 +1053,7 @@ typedef struct CUDA_TEXTURE_DESC_st
     /**
      * Resource view format
      */
-typedef enum CUresourceViewFormat_enum
+    typedef enum CUresourceViewFormat_enum {
-{
         CU_RES_VIEW_FORMAT_NONE          = 0x00, /**< No resource view format (use underlying resource format) */
         CU_RES_VIEW_FORMAT_UINT_1X8      = 0x01, /**< 1 channel unsigned 8-bit integers */
         CU_RES_VIEW_FORMAT_UINT_2X8      = 0x02, /**< 2 channel unsigned 8-bit integers */
@@ -1130,7 +1118,6 @@ typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st
 #endif
 
 
-
 /**
  * If set, the CUDA array is a collection of layers, where each layer is either a 1D
  * or a 2D array and the Depth member of CUDA_ARRAY3D_DESCRIPTOR specifies the number
@@ -1420,7 +1407,11 @@ typedef CUresult  CUDAAPI tcuCtxSynchronize(void);
 
     typedef CUresult CUDAAPI tcuModuleLoad(CUmodule *module, const char *fname);
     typedef CUresult CUDAAPI tcuModuleLoadData(CUmodule *module, const void *image);
-typedef CUresult  CUDAAPI tcuModuleLoadDataEx(CUmodule *module, const void *image, unsigned int numOptions, CUjit_option *options, void **optionValues);
+    typedef CUresult CUDAAPI tcuModuleLoadDataEx(CUmodule     *module,
+                                                 const void   *image,
+                                                 unsigned int  numOptions,
+                                                 CUjit_option *options,
+                                                 void        **optionValues);
     typedef CUresult CUDAAPI tcuModuleLoadFatBinary(CUmodule *module, const void *fatCubin);
     typedef CUresult CUDAAPI tcuModuleUnload(CUmodule hmod);
     typedef CUresult CUDAAPI tcuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, const char *name);
@@ -1449,8 +1440,7 @@ typedef CUresult CUDAAPI tcuMemAllocPitch(CUdeviceptr *dptr,
                                               size_t       Height,
                                               // size of biggest r/w to be performed by kernels on this memory
                                               // 4, 8 or 16 bytes
-                                          unsigned int ElementSizeBytes
+                                              unsigned int ElementSizeBytes);
-                                         );
 #else
 typedef CUresult CUDAAPI tcuMemGetInfo(unsigned int *free, unsigned int *total);
 typedef CUresult CUDAAPI tcuMemAlloc(CUdeviceptr *dptr, unsigned int bytesize);
@@ -1461,8 +1451,7 @@ typedef CUresult CUDAAPI tcuMemAllocPitch(CUdeviceptr *dptr,
                                           unsigned int  Height,
                                           // size of biggest r/w to be performed by kernels on this memory
                                           // 4, 8 or 16 bytes
-                                          unsigned int ElementSizeBytes
+                                          unsigned int ElementSizeBytes);
-                                         );
 #endif
 
     typedef CUresult CUDAAPI tcuMemFree(CUdeviceptr dptr);
@@ -1495,9 +1484,9 @@ typedef struct CUipcMemHandle_st
         char reserved[CU_IPC_HANDLE_SIZE];
     } CUipcMemHandle;
 
-typedef enum CUipcMem_flags_enum
+    typedef enum CUipcMem_flags_enum {
-{
+        CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS =
-    CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS = 0x1 /**< Automatically enable peer access between remote devices as needed */
+            0x1 /**< Automatically enable peer access between remote devices as needed */
     } CUipcMem_flags;
 
     typedef CUresult CUDAAPI tcuDeviceGetByPCIBusId(CUdevice *dev, char *pciBusId);
@@ -1510,9 +1499,14 @@ typedef CUresult CUDAAPI tcuIpcCloseMemHandle(CUdeviceptr dptr);
 #endif
 
     typedef CUresult CUDAAPI tcuMemHostRegister(void *p, size_t bytesize, unsigned int Flags);
-typedef CUresult CUDAAPI tcuMemHostUnregister(void *p);;
+    typedef CUresult CUDAAPI tcuMemHostUnregister(void *p);
+    ;
     typedef CUresult CUDAAPI tcuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount);
-typedef CUresult CUDAAPI tcuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount);
+    typedef CUresult CUDAAPI tcuMemcpyPeer(CUdeviceptr dstDevice,
+                                           CUcontext   dstContext,
+                                           CUdeviceptr srcDevice,
+                                           CUcontext   srcContext,
+                                           size_t      ByteCount);
 
 /************************************
  **
@@ -1541,7 +1535,8 @@ typedef CUresult  CUDAAPI tcuMemcpyHtoA(CUarray dstArray, size_t dstOffset, cons
     typedef CUresult CUDAAPI tcuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount);
 
     // array <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount);
+    typedef CUresult CUDAAPI
+    tcuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount);
 #else
 // system <-> device memory
 typedef CUresult CUDAAPI tcuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, unsigned int ByteCount);
@@ -1551,15 +1546,28 @@ typedef CUresult  CUDAAPI tcuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, un
 typedef CUresult CUDAAPI tcuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, unsigned int ByteCount);
 
 // device <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyDtoA(CUarray dstArray, unsigned int dstOffset, CUdeviceptr srcDevice, unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyDtoA(CUarray      dstArray,
-typedef CUresult  CUDAAPI tcuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount);
+                                       unsigned int dstOffset,
+                                       CUdeviceptr  srcDevice,
+                                       unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyAtoD(CUdeviceptr  dstDevice,
+                                       CUarray      srcArray,
+                                       unsigned int srcOffset,
+                                       unsigned int ByteCount);
 
 // system <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyHtoA(CUarray dstArray, unsigned int dstOffset, const void *srcHost, unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyHtoA(CUarray      dstArray,
+                                       unsigned int dstOffset,
+                                       const void  *srcHost,
+                                       unsigned int ByteCount);
 typedef CUresult CUDAAPI tcuMemcpyAtoH(void *dstHost, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount);
 
 // array <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyAtoA(CUarray dstArray, unsigned int dstOffset, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount);
+typedef CUresult CUDAAPI tcuMemcpyAtoA(CUarray      dstArray,
+                                       unsigned int dstOffset,
+                                       CUarray      srcArray,
+                                       unsigned int srcOffset,
+                                       unsigned int ByteCount);
 #endif
 
     // 2D memcpy
@@ -1586,36 +1594,51 @@ typedef CUresult  CUDAAPI tcuMemcpy3D(const CUDA_MEMCPY3D *pCopy);
 #if __CUDA_API_VERSION >= 3020
     // system <-> device memory
     typedef CUresult CUDAAPI tcuMemcpyHtoDAsync(CUdeviceptr dstDevice,
-                                             const void *srcHost, size_t ByteCount, CUstream hStream);
+                                                const void *srcHost,
+                                                size_t      ByteCount,
+                                                CUstream    hStream);
     typedef CUresult CUDAAPI tcuMemcpyDtoHAsync(void       *dstHost,
-                                             CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
+                                                CUdeviceptr srcDevice,
+                                                size_t      ByteCount,
+                                                CUstream    hStream);
 
     // device <-> device memory
     typedef CUresult CUDAAPI tcuMemcpyDtoDAsync(CUdeviceptr dstDevice,
-                                             CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream);
+                                                CUdeviceptr srcDevice,
+                                                size_t      ByteCount,
+                                                CUstream    hStream);
 
     // system <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset,
+    typedef CUresult CUDAAPI
-                                             const void *srcHost, size_t ByteCount, CUstream hStream);
+    tcuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream);
-typedef CUresult  CUDAAPI tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOffset,
+    typedef CUresult CUDAAPI
-                                             size_t ByteCount, CUstream hStream);
+    tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream);
 
 #else
 // system <-> device memory
 typedef CUresult CUDAAPI tcuMemcpyHtoDAsync(CUdeviceptr  dstDevice,
-                                             const void *srcHost, unsigned int ByteCount, CUstream hStream);
+                                            const void  *srcHost,
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
 typedef CUresult CUDAAPI tcuMemcpyDtoHAsync(void        *dstHost,
-                                             CUdeviceptr srcDevice, unsigned int ByteCount, CUstream hStream);
+                                            CUdeviceptr  srcDevice,
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
 
 // device <-> device memory
 typedef CUresult CUDAAPI tcuMemcpyDtoDAsync(CUdeviceptr  dstDevice,
-                                             CUdeviceptr srcDevice, unsigned int ByteCount, CUstream hStream);
+                                            CUdeviceptr  srcDevice,
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
 
 // system <-> array memory
-typedef CUresult  CUDAAPI tcuMemcpyHtoAAsync(CUarray dstArray, unsigned int dstOffset,
+typedef CUresult CUDAAPI tcuMemcpyHtoAAsync(CUarray      dstArray,
-                                             const void *srcHost, unsigned int ByteCount, CUstream hStream);
+                                            unsigned int dstOffset,
-typedef CUresult  CUDAAPI tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, unsigned int srcOffset,
+                                            const void  *srcHost,
-                                             unsigned int ByteCount, CUstream hStream);
+                                            unsigned int ByteCount,
+                                            CUstream     hStream);
+typedef CUresult CUDAAPI
+tcuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount, CUstream hStream);
 #endif
 
     // 2D memcpy
@@ -1634,13 +1657,22 @@ typedef CUresult  CUDAAPI tcuMemsetD16(CUdeviceptr dstDevice, unsigned short us,
     typedef CUresult CUDAAPI tcuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, unsigned int N);
 
 #if __CUDA_API_VERSION >= 3020
-typedef CUresult  CUDAAPI tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, size_t Width, size_t Height);
+    typedef CUresult CUDAAPI
-typedef CUresult  CUDAAPI tcuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, size_t Width, size_t Height);
+    tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, size_t Width, size_t Height);
-typedef CUresult  CUDAAPI tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, size_t Width, size_t Height);
+    typedef CUresult CUDAAPI
+    tcuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, size_t Width, size_t Height);
+    typedef CUresult CUDAAPI
+    tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, size_t Width, size_t Height);
 #else
-typedef CUresult  CUDAAPI tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, unsigned int Width, unsigned int Height);
+typedef CUresult CUDAAPI
-typedef CUresult  CUDAAPI tcuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, unsigned int Width, unsigned int Height);
+tcuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, unsigned int Width, unsigned int Height);
-typedef CUresult  CUDAAPI tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, unsigned int Width, unsigned int Height);
+typedef CUresult CUDAAPI tcuMemsetD2D16(CUdeviceptr    dstDevice,
+                                        unsigned int   dstPitch,
+                                        unsigned short us,
+                                        unsigned int   Width,
+                                        unsigned int   Height);
+typedef CUresult CUDAAPI
+tcuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, unsigned int Width, unsigned int Height);
 #endif
 
     /************************************
@@ -1657,10 +1689,16 @@ typedef CUresult CUDAAPI tcuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache co
     typedef CUresult CUDAAPI tcuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig config);
 
     typedef CUresult CUDAAPI tcuLaunchKernel(CUfunction   f,
-                                         unsigned int gridDimX,  unsigned int gridDimY,  unsigned int gridDimZ,
+                                             unsigned int gridDimX,
-                                         unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ,
+                                             unsigned int gridDimY,
+                                             unsigned int gridDimZ,
+                                             unsigned int blockDimX,
+                                             unsigned int blockDimY,
+                                             unsigned int blockDimZ,
                                              unsigned int sharedMemBytes,
-                                         CUstream hStream, void **kernelParams, void **extra);
+                                             CUstream     hStream,
+                                             void       **kernelParams,
+                                             void       **extra);
 
     /************************************
      **
@@ -1676,8 +1714,12 @@ typedef CUresult  CUDAAPI tcuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_
     typedef CUresult CUDAAPI tcuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray);
 
 #if __CUDA_API_VERSION >= 5000
-typedef CUresult CUDAAPI tcuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc, unsigned int numMipmapLevels);
+    typedef CUresult CUDAAPI tcuMipmappedArrayCreate(CUmipmappedArray              *pHandle,
-typedef CUresult CUDAAPI tcuMipmappedArrayGetLevel(CUarray *pLevelArray, CUmipmappedArray hMipmappedArray, unsigned int level);
+                                                     const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc,
+                                                     unsigned int                   numMipmapLevels);
+    typedef CUresult CUDAAPI tcuMipmappedArrayGetLevel(CUarray         *pLevelArray,
+                                                       CUmipmappedArray hMipmappedArray,
+                                                       unsigned int     level);
     typedef CUresult CUDAAPI tcuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray);
 #endif
 
@@ -1694,10 +1736,19 @@ typedef CUresult  CUDAAPI tcuTexRefSetArray(CUtexref hTexRef, CUarray hArray, un
 
 #if __CUDA_API_VERSION >= 3020
     typedef CUresult CUDAAPI tcuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, size_t bytes);
-typedef CUresult  CUDAAPI tcuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, size_t Pitch);
+    typedef CUresult CUDAAPI tcuTexRefSetAddress2D(CUtexref                     hTexRef,
+                                                   const CUDA_ARRAY_DESCRIPTOR *desc,
+                                                   CUdeviceptr                  dptr,
+                                                   size_t                       Pitch);
 #else
-typedef CUresult  CUDAAPI tcuTexRefSetAddress(unsigned int *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, unsigned int bytes);
+typedef CUresult CUDAAPI tcuTexRefSetAddress(unsigned int *ByteOffset,
-typedef CUresult  CUDAAPI tcuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, unsigned int Pitch);
+                                             CUtexref      hTexRef,
+                                             CUdeviceptr   dptr,
+                                             unsigned int  bytes);
+typedef CUresult CUDAAPI tcuTexRefSetAddress2D(CUtexref                     hTexRef,
+                                               const CUDA_ARRAY_DESCRIPTOR *desc,
+                                               CUdeviceptr                  dptr,
+                                               unsigned int                 Pitch);
 #endif
 
     typedef CUresult CUDAAPI tcuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int NumPackedComponents);
@@ -1763,7 +1814,10 @@ typedef CUresult CUDAAPI tcuEventElapsedTime(float *pMilliseconds, CUevent hStar
      ***********************************/
     typedef CUresult CUDAAPI tcuStreamCreate(CUstream *phStream, unsigned int Flags);
     typedef CUresult CUDAAPI tcuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags);
-typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags);
+    typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream         hStream,
+                                                  CUstreamCallback callback,
+                                                  void            *userData,
+                                                  unsigned int     flags);
 
     typedef CUresult CUDAAPI tcuStreamQuery(CUstream hStream);
     typedef CUresult CUDAAPI tcuStreamSynchronize(CUstream hStream);
@@ -1775,17 +1829,28 @@ typedef CUresult CUDAAPI tcuStreamDestroy(CUstream hStream);
      **
      ***********************************/
     typedef CUresult CUDAAPI tcuGraphicsUnregisterResource(CUgraphicsResource resource);
-typedef CUresult CUDAAPI tcuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex, unsigned int mipLevel);
+    typedef CUresult CUDAAPI tcuGraphicsSubResourceGetMappedArray(CUarray           *pArray,
+                                                                  CUgraphicsResource resource,
+                                                                  unsigned int       arrayIndex,
+                                                                  unsigned int       mipLevel);
 
 #if __CUDA_API_VERSION >= 3020
-typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource);
+    typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr       *pDevPtr,
+                                                                 size_t            *pSize,
+                                                                 CUgraphicsResource resource);
 #else
-typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, unsigned int *pSize, CUgraphicsResource resource);
+typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer(CUdeviceptr       *pDevPtr,
+                                                             unsigned int      *pSize,
+                                                             CUgraphicsResource resource);
 #endif
 
     typedef CUresult CUDAAPI tcuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsigned int flags);
-typedef CUresult CUDAAPI tcuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream);
+    typedef CUresult CUDAAPI tcuGraphicsMapResources(unsigned int        count,
-typedef CUresult CUDAAPI tcuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream);
+                                                     CUgraphicsResource *resources,
+                                                     CUstream            hStream);
+    typedef CUresult CUDAAPI tcuGraphicsUnmapResources(unsigned int        count,
+                                                       CUgraphicsResource *resources,
+                                                       CUstream            hStream);
 
     /************************************
      **

Samples/0_Introduction/matrixMulDynlinkJIT/helper_cuda_drvapi.h

@@ -14,21 +14,17 @@
 #ifndef HELPER_CUDA_DRVAPI_H
 #define HELPER_CUDA_DRVAPI_H
 
+#include <helper_string.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
-#include <helper_string.h>
-
 #ifndef MAX
 #define MAX(a, b) (a > b ? a : b)
 #endif
 
 #ifndef HELPER_CUDA_DRVAPI_H
-inline int ftoi(float value) {
+inline int ftoi(float value) { return (value >= 0 ? static_cast<int>(value + 0.5) : static_cast<int>(value - 0.5)); }
-  return (value >= 0 ? static_cast<int>(value + 0.5)
-                     : static_cast<int>(value - 0.5));
-}
 #endif
 
 #ifndef EXIT_WAIVED
@@ -47,39 +43,43 @@ inline int ftoi(float value) {
 #define checkCudaErrors(err) __checkCudaErrors(err, __FILE__, __LINE__)
 
 // These are the inline versions for all of the SDK helper functions
-inline void __checkCudaErrors(CUresult err, const char *file, const int line) {
+inline void __checkCudaErrors(CUresult err, const char *file, const int line)
+{
     if (CUDA_SUCCESS != err) {
         const char *errorStr = NULL;
         cuGetErrorString(err, &errorStr);
         fprintf(stderr,
                 "checkCudaErrors() Driver API error = %04d \"%s\" from file <%s>, "
                 "line %i.\n",
-            err, errorStr, file, line);
+                err,
+                errorStr,
+                file,
+                line);
         exit(EXIT_FAILURE);
     }
 }
 #endif
 
 // This function wraps the CUDA Driver API into a template function
-template <class T>
+template <class T> inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute, int device)
-inline void getCudaAttribute(T *attribute, CUdevice_attribute device_attribute,
+{
-                             int device) {
     checkCudaErrors(cuDeviceGetAttribute(attribute, device_attribute, device));
 }
 #endif
 
 // Beginning of GPU Architecture definitions
-inline int _ConvertSMVer2CoresDRV(int major, int minor) {
+inline int _ConvertSMVer2CoresDRV(int major, int minor)
+{
     // Defines for GPU Architecture types (using the SM version to determine the #
     // of cores per SM
-  typedef struct {
+    typedef struct
+    {
         int SM; // 0xMm (hexidecimal notation), M = SM Major version, and m = SM
                 // minor version
         int Cores;
     } sSMtoCores;
 
-  sSMtoCores nGpuArchCoresPerSM[] = {
+    sSMtoCores nGpuArchCoresPerSM[] = {{0x30, 192},
-      {0x30, 192},
                                        {0x32, 192},
                                        {0x35, 192},
                                        {0x37, 192},
@@ -110,16 +110,18 @@ inline int _ConvertSMVer2CoresDRV(int major, int minor) {
 
     // If we don't find the values, we default use the previous one to run
     // properly
-  printf(
+    printf("MapSMtoCores for SM %d.%d is undefined.  Default to use %d Cores/SM\n",
-      "MapSMtoCores for SM %d.%d is undefined.  Default to use %d Cores/SM\n",
+           major,
-      major, minor, nGpuArchCoresPerSM[index - 1].Cores);
+           minor,
+           nGpuArchCoresPerSM[index - 1].Cores);
     return nGpuArchCoresPerSM[index - 1].Cores;
 }
 // end of GPU Architecture definitions
 
 #ifdef __cuda_cuda_h__
 // General GPU Device CUDA Initialization
-inline int gpuDeviceInitDRV(int ARGC, const char **ARGV) {
+inline int gpuDeviceInitDRV(int ARGC, const char **ARGV)
+{
     int cuDevice    = 0;
     int deviceCount = 0;
     checkCudaErrors(cuInit(0, __CUDA_API_VERSION));
@@ -140,11 +142,8 @@ inline int gpuDeviceInitDRV(int ARGC, const char **ARGV) {
 
     if (dev > deviceCount - 1) {
         fprintf(stderr, "\n");
-    fprintf(stderr, ">> %d CUDA capable GPU device(s) detected. <<\n",
+        fprintf(stderr, ">> %d CUDA capable GPU device(s) detected. <<\n", deviceCount);
-            deviceCount);
+        fprintf(stderr, ">> cudaDeviceInit (-device=%d) is not a valid GPU device. <<\n", dev);
-    fprintf(stderr,
-            ">> cudaDeviceInit (-device=%d) is not a valid GPU device. <<\n",
-            dev);
         fprintf(stderr, "\n");
         return -dev;
     }
@@ -171,7 +170,8 @@ inline int gpuDeviceInitDRV(int ARGC, const char **ARGV) {
 }
 
 // This function returns the best GPU based on performance
-inline int gpuGetMaxGflopsDeviceIdDRV() {
+inline int gpuGetMaxGflopsDeviceIdDRV()
+{
     CUdevice           current_device   = 0;
     CUdevice           max_perf_device  = 0;
     int                device_count     = 0;
@@ -187,8 +187,7 @@ inline int gpuGetMaxGflopsDeviceIdDRV() {
     checkCudaErrors(cuDeviceGetCount(&device_count));
 
     if (device_count == 0) {
-    fprintf(stderr,
+        fprintf(stderr, "gpuGetMaxGflopsDeviceIdDRV error: no devices supporting CUDA\n");
-            "gpuGetMaxGflopsDeviceIdDRV error: no devices supporting CUDA\n");
         exit(EXIT_FAILURE);
     }
 
@@ -196,36 +195,31 @@ inline int gpuGetMaxGflopsDeviceIdDRV() {
     current_device = 0;
 
     while (current_device < device_count) {
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(
-        &multiProcessorCount, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
+            cuDeviceGetAttribute(&multiProcessorCount, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT, current_device));
-        current_device));
+        checkCudaErrors(cuDeviceGetAttribute(&clockRate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, current_device));
-        &clockRate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, current_device));
+        checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
-        &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
-        &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, current_device));
 
         int computeMode;
-    getCudaAttribute<int>(&computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE,
+        getCudaAttribute<int>(&computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, current_device);
-                          current_device);
 
         if (computeMode != CU_COMPUTEMODE_PROHIBITED) {
             if (major == 9999 && minor == 9999) {
                 sm_per_multiproc = 1;
-      } else {
+            }
+            else {
                 sm_per_multiproc = _ConvertSMVer2CoresDRV(major, minor);
             }
 
-      unsigned long long compute_perf =
+            unsigned long long compute_perf = (unsigned long long)(multiProcessorCount * sm_per_multiproc * clockRate);
-          (unsigned long long)(multiProcessorCount * sm_per_multiproc *
-                               clockRate);
 
             if (compute_perf > max_compute_perf) {
                 max_compute_perf = compute_perf;
                 max_perf_device  = current_device;
             }
-    } else {
+        }
+        else {
             devices_prohibited++;
         }
 
@@ -243,7 +237,8 @@ inline int gpuGetMaxGflopsDeviceIdDRV() {
 }
 
 // General initialization call to pick the best CUDA Device
-inline CUdevice findCudaDeviceDRV(int argc, const char **argv) {
+inline CUdevice findCudaDeviceDRV(int argc, const char **argv)
+{
     CUdevice cuDevice;
     int      devID = 0;
 
@@ -255,7 +250,8 @@ inline CUdevice findCudaDeviceDRV(int argc, const char **argv) {
             printf("exiting...\n");
             exit(EXIT_SUCCESS);
         }
-  } else {
+    }
+    else {
         // Otherwise pick the device with highest Gflops/s
         char name[100];
         devID = gpuGetMaxGflopsDeviceIdDRV();
@@ -269,7 +265,8 @@ inline CUdevice findCudaDeviceDRV(int argc, const char **argv) {
     return cuDevice;
 }
 
-inline CUdevice findIntegratedGPUDrv() {
+inline CUdevice findIntegratedGPUDrv()
+{
     CUdevice current_device     = 0;
     int      device_count       = 0;
     int      devices_prohibited = 0;
@@ -286,28 +283,22 @@ inline CUdevice findIntegratedGPUDrv() {
     // Find the integrated GPU which is compute capable
     while (current_device < device_count) {
         int computeMode = -1;
-    checkCudaErrors(cuDeviceGetAttribute(
+        checkCudaErrors(cuDeviceGetAttribute(&isIntegrated, CU_DEVICE_ATTRIBUTE_INTEGRATED, current_device));
-        &isIntegrated, CU_DEVICE_ATTRIBUTE_INTEGRATED, current_device));
+        checkCudaErrors(cuDeviceGetAttribute(&computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, current_device));
-    checkCudaErrors(cuDeviceGetAttribute(
-        &computeMode, CU_DEVICE_ATTRIBUTE_COMPUTE_MODE, current_device));
 
         // If GPU is integrated and is not running on Compute Mode prohibited use
         // that
         if (isIntegrated && (computeMode != CU_COMPUTEMODE_PROHIBITED)) {
             int  major = 0, minor = 0;
             char deviceName[256];
-      checkCudaErrors(cuDeviceGetAttribute(
+            checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, current_device));
-          &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR,
+            checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, current_device));
-          current_device));
-      checkCudaErrors(cuDeviceGetAttribute(
-          &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR,
-          current_device));
             checkCudaErrors(cuDeviceGetName(deviceName, 256, current_device));
-      printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n",
+            printf("GPU Device %d: \"%s\" with compute capability %d.%d\n\n", current_device, deviceName, major, minor);
-             current_device, deviceName, major, minor);
 
             return current_device;
-    } else {
+        }
+        else {
             devices_prohibited++;
         }
 
@@ -323,29 +314,26 @@ inline CUdevice findIntegratedGPUDrv() {
 }
 
 // General check for CUDA GPU SM Capabilities
-inline bool checkCudaCapabilitiesDRV(int major_version, int minor_version,
+inline bool checkCudaCapabilitiesDRV(int major_version, int minor_version, int devID)
-                                     int devID) {
+{
     CUdevice cuDevice;
     char     name[256];
     int      major = 0, minor = 0;
 
     checkCudaErrors(cuDeviceGet(&cuDevice, devID));
     checkCudaErrors(cuDeviceGetName(name, 100, cuDevice));
-  checkCudaErrors(cuDeviceGetAttribute(
+    checkCudaErrors(cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
-      &major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice));
+    checkCudaErrors(cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
-  checkCudaErrors(cuDeviceGetAttribute(
-      &minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice));
 
-  if ((major > major_version) ||
+    if ((major > major_version) || (major == major_version && minor >= minor_version)) {
-      (major == major_version && minor >= minor_version)) {
+        printf("> Device %d: <%16s >, Compute SM %d.%d detected\n", devID, name, major, minor);
-    printf("> Device %d: <%16s >, Compute SM %d.%d detected\n", devID, name,
-           major, minor);
         return true;
-  } else {
+    }
-    printf(
+    else {
-        "No GPU device was found that can support CUDA compute capability "
+        printf("No GPU device was found that can support CUDA compute capability "
                "%d.%d.\n",
-        major_version, minor_version);
+               major_version,
+               minor_version);
         return false;
     }
 }
@@ -354,4 +342,3 @@ inline bool checkCudaCapabilitiesDRV(int major_version, int minor_version,
 // end of CUDA Helper Functions
 
 #endif // HELPER_CUDA_DRVAPI_H
-

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMulDynlinkJIT.cpp

@@ -43,10 +43,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
 #include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 // includes, CUDA
 #include "cuda_drvapi_dynlink.h"
@@ -78,8 +78,7 @@ static const char *sSDKsample = "matrixMulDynlinkJIT (CUDA dynamic linking)";
 ////////////////////////////////////////////////////////////////////////////////
 void randomInit(float *data, size_t size)
 {
-    for (size_t i = 0; i < size; ++i)
+    for (size_t i = 0; i < size; ++i) {
-    {
         data[i] = rand() / (float)RAND_MAX;
     }
 }
@@ -100,18 +99,14 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
     checkCudaErrors(cuInit(0, __CUDA_API_VERSION));
 
     // This assumes that the user is attempting to specify a explicit device -device=n
-    if (argc > 1)
+    if (argc > 1) {
-    {
         bool bFound = false;
 
-        for (int param=0; param < argc; param++)
+        for (int param = 0; param < argc; param++) {
-        {
+            if (!strncmp(argv[param], "-device", 7)) {
-            if (!strncmp(argv[param], "-device", 7))
-            {
                 int i = (int)strlen(argv[1]);
 
-                while (argv[1][i] != '=')
+                while (argv[1][i] != '=') {
-                {
                     i--;
                 }
 
@@ -128,16 +123,15 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
     int deviceCount = 0;
     checkCudaErrors(cuDeviceGetCount(&deviceCount));
 
-    if (deviceCount == 0)
+    if (deviceCount == 0) {
-    {
         fprintf(stderr, "No devices supporting CUDA detected, exiting...\n");
         exit(EXIT_SUCCESS);
     }
 
-    if (devID < 0) devID = 0;
+    if (devID < 0)
+        devID = 0;
 
-    if (devID > deviceCount -1)
+    if (devID > deviceCount - 1) {
-    {
         fprintf(stderr, "initCUDA (Device=%d) invalid GPU device.  %d GPU device(s) detected.\n\n", devID, deviceCount);
         status = CUDA_ERROR_NOT_FOUND;
 
@@ -159,8 +153,7 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
     // create context for picked device
     status = cuCtxCreate(&g_cuContext, 0, cuDevice);
 
-    if (CUDA_SUCCESS != status)
+    if (CUDA_SUCCESS != status) {
-    {
         cuCtxDestroy(g_cuContext);
         exit(EXIT_SUCCESS);
     }
@@ -191,9 +184,11 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
         printf("> Compiling CUDA module\n");
 
 #if defined(_WIN64) || defined(__LP64__)
-        status = cuModuleLoadDataEx(&cuModule, matrixMul_kernel_64_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
+        status =
+            cuModuleLoadDataEx(&cuModule, matrixMul_kernel_64_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
 #else
-        status = cuModuleLoadDataEx(&cuModule, matrixMul_kernel_32_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
+        status =
+            cuModuleLoadDataEx(&cuModule, matrixMul_kernel_32_ptxdump, jitNumOptions, jitOptions, (void **)jitOptVals);
 #endif
 
         printf("> PTX JIT log:\n%s\n", jitLogBuffer);
@@ -203,19 +198,17 @@ CUresult initCUDA(int argc, char **argv, CUfunction *pMatrixMul, int *block_size
         delete[] jitLogBuffer;
     }
 
-    if (CUDA_SUCCESS != status)
+    if (CUDA_SUCCESS != status) {
-    {
         printf("Error while compiling PTX\n");
         cuCtxDestroy(g_cuContext);
         exit(EXIT_FAILURE);
     }
 
     // retrieve CUDA function from the compiled module
-    status = cuModuleGetFunction(&cuFunction, cuModule,
+    status = cuModuleGetFunction(
-                                 (block_size == 16) ? "matrixMul_bs16_32bit" : "matrixMul_bs32_32bit");
+        &cuFunction, cuModule, (block_size == 16) ? "matrixMul_bs16_32bit" : "matrixMul_bs32_32bit");
 
-    if (CUDA_SUCCESS != status)
+    if (CUDA_SUCCESS != status) {
-    {
         cuCtxDestroy(g_cuContext);
         exit(EXIT_FAILURE);
     }
@@ -280,10 +273,8 @@ int main(int argc, char **argv)
         int   Matrix_Width_B = WB;
         void *args[5]        = {&d_C, &d_A, &d_B, &Matrix_Width_A, &Matrix_Width_B};
 
-        checkCudaErrors(cuLaunchKernel(matrixMul, (WC/block_size), (HC/block_size), 1,
+        checkCudaErrors(cuLaunchKernel(
-                                       block_size     , block_size     , 1,
+            matrixMul, (WC / block_size), (HC / block_size), 1, block_size, block_size, 1, 0, NULL, args, NULL));
-                                       0,
-                                       NULL, args, NULL));
     }
 #else // __CUDA_API_VERSION <= 3020
     {
@@ -331,8 +322,7 @@ int main(int argc, char **argv)
     // check result
     float diff = 0.0f;
 
-    for (unsigned int i=0; i<size_C; i++)
+    for (unsigned int i = 0; i < size_C; i++) {
-    {
         float tmp = reference[i] - h_C[i];
         diff += tmp * tmp;
     }

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMul_gold.cpp

@@ -28,8 +28,7 @@
 
 ////////////////////////////////////////////////////////////////////////////////
 // export C interface
-extern "C"
+extern "C" void computeGold(float *, const float *, const float *, unsigned int, unsigned int, unsigned int);
-void computeGold(float *, const float *, const float *, unsigned int, unsigned int, unsigned int);
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Compute reference data set
@@ -40,16 +39,13 @@ void computeGold(float *, const float *, const float *, unsigned int, unsigned i
 //! @param hA         height of matrix A
 //! @param wB         width of matrix B
 ////////////////////////////////////////////////////////////////////////////////
-void
+void computeGold(float *C, const float *A, const float *B, unsigned int hA, unsigned int wA, unsigned int wB)
-computeGold(float *C, const float *A, const float *B, unsigned int hA, unsigned int wA, unsigned int wB)
 {
     for (unsigned int i = 0; i < hA; ++i)
-        for (unsigned int j = 0; j < wB; ++j)
+        for (unsigned int j = 0; j < wB; ++j) {
-        {
             double sum = 0;
 
-            for (unsigned int k = 0; k < wA; ++k)
+            for (unsigned int k = 0; k < wA; ++k) {
-            {
                 double a = A[i * wA + k];
                 double b = B[k * wB + j];
                 sum += a * b;

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMul_kernel_32_ptxdump.h

@@ -32,7 +32,8 @@
 #define __matrixMul_kernel_32_ptxdump_h__
 
 #if defined __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
     extern unsigned char matrixMul_kernel_32_ptxdump[25784];

Samples/0_Introduction/matrixMulDynlinkJIT/matrixMul_kernel_64_ptxdump.h

@@ -32,7 +32,8 @@
 #define __matrixMul_kernel_64_ptxdump_h__
 
 #if defined __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
     extern unsigned char matrixMul_kernel_64_ptxdump[26489];

Samples/0_Introduction/matrixMul_nvrtc/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/matrixMul_nvrtc/README.md

@@ -2,7 +2,7 @@
 
 ## Description
 
-This sample implements matrix multiplication and is exactly the same as Chapter 6 of the programming guide. It has been written for clarity of exposition to illustrate various CUDA programming principles, not with the goal of providing the most performant generic kernel for matrix multiplication.  To illustrate GPU performance for matrix multiply, this sample also shows how to use the new CUDA 4.0 interface for CUBLAS to demonstrate high-performance performance for matrix multiplication.
+This sample implements matrix multiplication and is exactly the same as the second example of the [Shared Memory](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory) section of the programming guide. It has been written for clarity of exposition to illustrate various CUDA programming principles, not with the goal of providing the most performant generic kernel for matrix multiplication.  To illustrate GPU performance for matrix multiply, this sample also shows how to use the CUDA 4.0+ interface for cuBLAS to demonstrate high-performance performance for matrix multiplication.
 
 ## Key Concepts
 
@@ -30,7 +30,7 @@ cuMemcpyDtoH, cuLaunchKernel, cuMemcpyHtoD, cuCtxSynchronize, cuMemAlloc, cuMemF
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/matrixMul_nvrtc/matrixMul.cpp

@@ -42,17 +42,19 @@
  */
 
 // System includes
-#include <stdio.h>
 #include <assert.h>
+#include <stdio.h>
 
 // CUDA runtime
 #include <cuda_runtime.h>
+
 #include "nvrtc_helper.h"
 
 // Helper functions and utilities to work with CUDA
 #include <helper_functions.h>
 
-void constantInit(float *data, int size, float val) {
+void constantInit(float *data, int size, float val)
+{
     for (int i = 0; i < size; ++i) {
         data[i] = val;
     }
@@ -61,8 +63,8 @@ void constantInit(float *data, int size, float val) {
 /**
  * Run a simple test of matrix multiplication using CUDA
  */
-int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
+int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA, dim3 &dimsB)
-                   dim3 &dimsB) {
+{
     // Allocate host memory for matrices A and B
     unsigned int size_A     = dimsA.x * dimsA.y;
     unsigned int mem_size_A = sizeof(float) * size_A;
@@ -114,24 +116,27 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
 
     CUfunction kernel_addr;
     if (block_size == 16) {
-    checkCudaErrors(
+        checkCudaErrors(cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block16"));
-        cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block16"));
+    }
-  } else {
+    else {
-    checkCudaErrors(
+        checkCudaErrors(cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block32"));
-        cuModuleGetFunction(&kernel_addr, module, "matrixMulCUDA_block32"));
     }
 
-  void *arr[] = {(void *)&d_C, (void *)&d_A, (void *)&d_B, (void *)&dimsA.x,
+    void *arr[] = {(void *)&d_C, (void *)&d_A, (void *)&d_B, (void *)&dimsA.x, (void *)&dimsB.x};
-                 (void *)&dimsB.x};
 
     // Execute the kernel
     int nIter = 300;
 
     for (int j = 0; j < nIter; j++) {
-    checkCudaErrors(
+        checkCudaErrors(cuLaunchKernel(kernel_addr,
-        cuLaunchKernel(kernel_addr, grid.x, grid.y, grid.z, /* grid dim */
+                                       grid.x,
-                       threads.x, threads.y, threads.z,     /* block dim */
+                                       grid.y,
-                       0, 0,    /* shared mem, stream */
+                                       grid.z, /* grid dim */
+                                       threads.x,
+                                       threads.y,
+                                       threads.z, /* block dim */
+                                       0,
+                                       0,       /* shared mem, stream */
                                        &arr[0], /* arguments */
                                        0));
 
@@ -157,16 +162,14 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
         double rel_err    = abs_err / abs_val / dot_length;
 
         if (rel_err > eps) {
-      printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i,
+            printf("Error! Matrix[%05d]=%.8f, ref=%.8f error term is > %E\n", i, h_C[i], dimsA.x * valB, eps);
-             h_C[i], dimsA.x * valB, eps);
             correct = false;
         }
     }
 
     printf("%s\n", correct ? "Result = PASS" : "Result = FAIL");
 
-  printf(
+    printf("\nNOTE: The CUDA Samples are not meant for performance measurements. "
-      "\nNOTE: The CUDA Samples are not meant for performance measurements. "
            "Results may vary when GPU Boost is enabled.\n");
 
     // Clean up memory
@@ -180,7 +183,8 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
 
     if (correct) {
         return EXIT_SUCCESS;
-  } else {
+    }
+    else {
         return EXIT_FAILURE;
     }
 }
@@ -189,16 +193,15 @@ int matrixMultiply(int argc, char **argv, int block_size, dim3 &dimsA,
  * Program main
  */
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("[Matrix Multiply Using CUDA] - Starting...\n");
 
-  if (checkCmdLineFlag(argc, (const char **)argv, "help") ||
+    if (checkCmdLineFlag(argc, (const char **)argv, "help") || checkCmdLineFlag(argc, (const char **)argv, "?")) {
-      checkCmdLineFlag(argc, (const char **)argv, "?")) {
         printf("Usage -device=n (n >= 0 for deviceID)\n");
         printf("      -wA=WidthA -hA=HeightA (Width x Height of Matrix A)\n");
         printf("      -wB=WidthB -hB=HeightB (Width x Height of Matrix B)\n");
-    printf(
+        printf("  Note: Outer matrix dimensions of A & B matrices must be equal.\n");
-        "  Note: Outer matrix dimensions of A & B matrices must be equal.\n");
 
         exit(EXIT_SUCCESS);
     }
@@ -234,13 +237,11 @@ int main(int argc, char **argv) {
     }
 
     if (dimsA.x != dimsB.y) {
-    printf("Error: outer matrix dimensions must be equal. (%d != %d)\n",
+        printf("Error: outer matrix dimensions must be equal. (%d != %d)\n", dimsA.x, dimsB.y);
-           dimsA.x, dimsB.y);
         exit(EXIT_FAILURE);
     }
 
-  printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x,
+    printf("MatrixA(%d,%d), MatrixB(%d,%d)\n", dimsA.x, dimsA.y, dimsB.x, dimsB.y);
-         dimsB.y);
 
     int matrix_result = matrixMultiply(argc, argv, block_size, dimsA, dimsB);
 

Samples/0_Introduction/matrixMul_nvrtc/matrixMul_kernel.cu

@@ -48,11 +48,10 @@
 
 #include <cooperative_groups.h>
 
-template <int BLOCK_SIZE>
+template <int BLOCK_SIZE> __device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB)
-__device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB) {
+{
     // Handle to thread block group
-  cooperative_groups::thread_block cta =
+    cooperative_groups::thread_block cta = cooperative_groups::this_thread_block();
-      cooperative_groups::this_thread_block();
     // Block index
     int bx = blockIdx.x;
     int by = blockIdx.y;
@@ -120,12 +119,12 @@ __device__ void matrixMulCUDA(float *C, float *A, float *B, int wA, int wB) {
     C[c + wB * ty + tx] = Csub;
 }
 
-extern "C" __global__ void matrixMulCUDA_block16(float *C, float *A, float *B,
+extern "C" __global__ void matrixMulCUDA_block16(float *C, float *A, float *B, int wA, int wB)
-                                                 int wA, int wB) {
+{
     matrixMulCUDA<16>(C, A, B, wA, wB);
 }
 
-extern "C" __global__ void matrixMulCUDA_block32(float *C, float *A, float *B,
+extern "C" __global__ void matrixMulCUDA_block32(float *C, float *A, float *B, int wA, int wB)
-                                                 int wA, int wB) {
+{
     matrixMulCUDA<32>(C, A, B, wA, wB);
 }

Samples/0_Introduction/mergeSort/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/mergeSort/README.md

@@ -27,6 +27,6 @@ cudaMalloc, cudaDeviceSynchronize, cudaMemcpy, cudaFree
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/mergeSort/bitonic.cu

@@ -28,12 +28,13 @@
 #include <cooperative_groups.h>
 
 namespace cg = cooperative_groups;
-#include <helper_cuda.h>
 #include <assert.h>
+#include <helper_cuda.h>
+
 #include "mergeSort_common.h"
 
-inline __device__ void Comparator(uint &keyA, uint &valA, uint &keyB,
+inline __device__ void Comparator(uint &keyA, uint &valA, uint &keyB, uint &valB, uint arrowDir)
-                                  uint &valB, uint arrowDir) {
+{
     uint t;
 
     if ((keyA > keyB) == arrowDir) {
@@ -46,9 +47,9 @@ inline __device__ void Comparator(uint &keyA, uint &valA, uint &keyB,
     }
 }
 
-__global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
+__global__ void
-                                        uint *d_SrcKey, uint *d_SrcVal,
+bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey, uint *d_SrcVal, uint arrayLength, uint sortDir)
-                                        uint arrayLength, uint sortDir) {
+{
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     // Shared memory storage for one or more short vectors
@@ -62,10 +63,8 @@ __global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
     d_DstVal += blockIdx.x * SHARED_SIZE_LIMIT + threadIdx.x;
     s_key[threadIdx.x + 0]                       = d_SrcKey[0];
     s_val[threadIdx.x + 0]                       = d_SrcVal[0];
-  s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
+    s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
-      d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
+    s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
-  s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
-      d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
 
     for (uint size = 2; size < arrayLength; size <<= 1) {
         // Bitonic merge
@@ -74,8 +73,7 @@ __global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
         for (uint stride = size / 2; stride > 0; stride >>= 1) {
             cg::sync(cta);
             uint pos = 2 * threadIdx.x - (threadIdx.x & (stride - 1));
-      Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride],
+            Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride], s_val[pos + stride], dir);
-                 s_val[pos + stride], dir);
         }
     }
 
@@ -84,26 +82,25 @@ __global__ void bitonicSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
         for (uint stride = arrayLength / 2; stride > 0; stride >>= 1) {
             cg::sync(cta);
             uint pos = 2 * threadIdx.x - (threadIdx.x & (stride - 1));
-      Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride],
+            Comparator(s_key[pos + 0], s_val[pos + 0], s_key[pos + stride], s_val[pos + stride], sortDir);
-                 s_val[pos + stride], sortDir);
         }
     }
 
     cg::sync(cta);
     d_DstKey[0]                       = s_key[threadIdx.x + 0];
     d_DstVal[0]                       = s_val[threadIdx.x + 0];
-  d_DstKey[(SHARED_SIZE_LIMIT / 2)] =
+    d_DstKey[(SHARED_SIZE_LIMIT / 2)] = s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-      s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
+    d_DstVal[(SHARED_SIZE_LIMIT / 2)] = s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-  d_DstVal[(SHARED_SIZE_LIMIT / 2)] =
-      s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
 }
 
 // Helper function (also used by odd-even merge sort)
-extern "C" uint factorRadix2(uint *log2L, uint L) {
+extern "C" uint factorRadix2(uint *log2L, uint L)
+{
     if (!L) {
         *log2L = 0;
         return 0;
-  } else {
+    }
+    else {
         for (*log2L = 0; (L & 1) == 0; L >>= 1, *log2L++)
             ;
 
@@ -111,10 +108,14 @@ extern "C" uint factorRadix2(uint *log2L, uint L) {
     }
 }
 
-extern "C" void bitonicSortShared(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicSortShared(uint *d_DstKey,
-                                  uint *d_SrcKey, uint *d_SrcVal,
+                                  uint *d_DstVal,
-                                  uint batchSize, uint arrayLength,
+                                  uint *d_SrcKey,
-                                  uint sortDir) {
+                                  uint *d_SrcVal,
+                                  uint  batchSize,
+                                  uint  arrayLength,
+                                  uint  sortDir)
+{
     // Nothing to sort
     if (arrayLength < 2) {
         return;
@@ -131,32 +132,25 @@ extern "C" void bitonicSortShared(uint *d_DstKey, uint *d_DstVal,
     assert(arrayLength <= SHARED_SIZE_LIMIT);
     assert((batchSize * arrayLength) % SHARED_SIZE_LIMIT == 0);
 
-  bitonicSortSharedKernel<<<blockCount, threadCount>>>(
+    bitonicSortSharedKernel<<<blockCount, threadCount>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength, sortDir);
-      d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength, sortDir);
     getLastCudaError("bitonicSortSharedKernel<<<>>> failed!\n");
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 3: merge elementary intervals
 ////////////////////////////////////////////////////////////////////////////////
-static inline __host__ __device__ uint iDivUp(uint a, uint b) {
+static inline __host__ __device__ uint iDivUp(uint a, uint b) { return ((a % b) == 0) ? (a / b) : (a / b + 1); }
-  return ((a % b) == 0) ? (a / b) : (a / b + 1);
-}
 
-static inline __host__ __device__ uint getSampleCount(uint dividend) {
+static inline __host__ __device__ uint getSampleCount(uint dividend) { return iDivUp(dividend, SAMPLE_STRIDE); }
-  return iDivUp(dividend, SAMPLE_STRIDE);
-}
 
 template <uint sortDir>
-static inline __device__ void ComparatorExtended(uint &keyA, uint &valA,
+static inline __device__ void
-                                                 uint &flagA, uint &keyB,
+ComparatorExtended(uint &keyA, uint &valA, uint &flagA, uint &keyB, uint &valB, uint &flagB, uint arrowDir)
-                                                 uint &valB, uint &flagB,
+{
-                                                 uint arrowDir) {
     uint t;
 
-  if ((!(flagA || flagB) && ((keyA > keyB) == arrowDir)) ||
+    if ((!(flagA || flagB) && ((keyA > keyB) == arrowDir)) || ((arrowDir == sortDir) && (flagA == 1))
-      ((arrowDir == sortDir) && (flagA == 1)) ||
+        || ((arrowDir != sortDir) && (flagB == 1))) {
-      ((arrowDir != sortDir) && (flagB == 1))) {
         t     = keyA;
         keyA  = keyB;
         keyB  = t;
@@ -170,9 +164,15 @@ static inline __device__ void ComparatorExtended(uint &keyA, uint &valA,
 }
 
 template <uint sortDir>
-__global__ void bitonicMergeElementaryIntervalsKernel(
+__global__ void bitonicMergeElementaryIntervalsKernel(uint *d_DstKey,
-    uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey, uint *d_SrcVal,
+                                                      uint *d_DstVal,
-    uint *d_LimitsA, uint *d_LimitsB, uint stride, uint N) {
+                                                      uint *d_SrcKey,
+                                                      uint *d_SrcVal,
+                                                      uint *d_LimitsA,
+                                                      uint *d_LimitsB,
+                                                      uint  stride,
+                                                      uint  N)
+{
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     __shared__ uint  s_key[2 * SAMPLE_STRIDE];
@@ -200,10 +200,8 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
         startSrcB = d_LimitsB[blockIdx.x];
         startDst  = startSrcA + startSrcB;
 
-    uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1]
+        uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1] : segmentElementsA;
-                                                    : segmentElementsA;
+        uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1] : segmentElementsB;
-    uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1]
-                                                    : segmentElementsB;
         lenSrcA      = endSrcA - startSrcA;
         lenSrcB      = endSrcB - startSrcB;
     }
@@ -222,10 +220,8 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
 
     // Prepare for bitonic merge by inversing the ordering
     if (threadIdx.x < lenSrcB) {
-    s_key[2 * SAMPLE_STRIDE - 1 - threadIdx.x] =
+        s_key[2 * SAMPLE_STRIDE - 1 - threadIdx.x] = d_SrcKey[stride + startSrcB + threadIdx.x];
-        d_SrcKey[stride + startSrcB + threadIdx.x];
+        s_val[2 * SAMPLE_STRIDE - 1 - threadIdx.x] = d_SrcVal[stride + startSrcB + threadIdx.x];
-    s_val[2 * SAMPLE_STRIDE - 1 - threadIdx.x] =
-        d_SrcVal[stride + startSrcB + threadIdx.x];
         s_inf[2 * SAMPLE_STRIDE - 1 - threadIdx.x] = 0;
     }
 
@@ -233,9 +229,13 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
     for (uint stride = SAMPLE_STRIDE; stride > 0; stride >>= 1) {
         cg::sync(cta);
         uint pos = 2 * threadIdx.x - (threadIdx.x & (stride - 1));
-    ComparatorExtended<sortDir>(s_key[pos + 0], s_val[pos + 0], s_inf[pos + 0],
+        ComparatorExtended<sortDir>(s_key[pos + 0],
-                                s_key[pos + stride], s_val[pos + stride],
+                                    s_val[pos + 0],
-                                s_inf[pos + stride], sortDir);
+                                    s_inf[pos + 0],
+                                    s_key[pos + stride],
+                                    s_val[pos + stride],
+                                    s_inf[pos + stride],
+                                    sortDir);
     }
 
     // Store sorted data
@@ -254,26 +254,28 @@ __global__ void bitonicMergeElementaryIntervalsKernel(
     }
 }
 
-extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey,
-                                                uint *d_SrcKey, uint *d_SrcVal,
+                                                uint *d_DstVal,
+                                                uint *d_SrcKey,
+                                                uint *d_SrcVal,
                                                 uint *d_LimitsA,
-                                                uint *d_LimitsB, uint stride,
+                                                uint *d_LimitsB,
-                                                uint N, uint sortDir) {
+                                                uint  stride,
+                                                uint  N,
+                                                uint  sortDir)
+{
     uint lastSegmentElements = N % (2 * stride);
 
-  uint mergePairs = (lastSegmentElements > stride)
+    uint mergePairs = (lastSegmentElements > stride) ? getSampleCount(N) : (N - lastSegmentElements) / SAMPLE_STRIDE;
-                        ? getSampleCount(N)
-                        : (N - lastSegmentElements) / SAMPLE_STRIDE;
 
     if (sortDir) {
-    bitonicMergeElementaryIntervalsKernel<1U><<<mergePairs, SAMPLE_STRIDE>>>(
+        bitonicMergeElementaryIntervalsKernel<1U>
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
-        N);
         getLastCudaError("mergeElementaryIntervalsKernel<1> failed\n");
-  } else {
+    }
-    bitonicMergeElementaryIntervalsKernel<0U><<<mergePairs, SAMPLE_STRIDE>>>(
+    else {
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+        bitonicMergeElementaryIntervalsKernel<0U>
-        N);
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
         getLastCudaError("mergeElementaryIntervalsKernel<0> failed\n");
     }
 }

Samples/0_Introduction/mergeSort/main.cpp

@@ -26,17 +26,19 @@
  */
 
 #include <assert.h>
+#include <cuda_runtime.h>
+#include <helper_cuda.h>
+#include <helper_functions.h>
 #include <stdio.h>
 #include <stdlib.h>
-#include <cuda_runtime.h>
+
-#include <helper_functions.h>
-#include <helper_cuda.h>
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Test driver
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     uint               *h_SrcKey, *h_SrcVal, *h_DstKey, *h_DstVal;
     uint               *d_SrcKey, *d_SrcVal, *d_BufKey, *d_BufVal, *d_DstKey, *d_DstVal;
     StopWatchInterface *hTimer = NULL;
@@ -75,10 +77,8 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&d_BufVal, N * sizeof(uint)));
     checkCudaErrors(cudaMalloc((void **)&d_SrcKey, N * sizeof(uint)));
     checkCudaErrors(cudaMalloc((void **)&d_SrcVal, N * sizeof(uint)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(d_SrcKey, h_SrcKey, N * sizeof(uint), cudaMemcpyHostToDevice));
-      cudaMemcpy(d_SrcKey, h_SrcKey, N * sizeof(uint), cudaMemcpyHostToDevice));
+    checkCudaErrors(cudaMemcpy(d_SrcVal, h_SrcVal, N * sizeof(uint), cudaMemcpyHostToDevice));
-  checkCudaErrors(
-      cudaMemcpy(d_SrcVal, h_SrcVal, N * sizeof(uint), cudaMemcpyHostToDevice));
 
     printf("Initializing GPU merge sort...\n");
     initMergeSort();
@@ -93,10 +93,8 @@ int main(int argc, char **argv) {
     printf("Time: %f ms\n", sdkGetTimerValue(&hTimer));
 
     printf("Reading back GPU merge sort results...\n");
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpy(h_DstKey, d_DstKey, N * sizeof(uint), cudaMemcpyDeviceToHost));
-      cudaMemcpy(h_DstKey, d_DstKey, N * sizeof(uint), cudaMemcpyDeviceToHost));
+    checkCudaErrors(cudaMemcpy(h_DstVal, d_DstVal, N * sizeof(uint), cudaMemcpyDeviceToHost));
-  checkCudaErrors(
-      cudaMemcpy(h_DstVal, d_DstVal, N * sizeof(uint), cudaMemcpyDeviceToHost));
 
     printf("Inspecting the results...\n");
     uint keysFlag = validateSortedKeys(h_DstKey, h_SrcKey, 1, N, numValues, DIR);

Samples/0_Introduction/mergeSort/mergeSort.cu

@@ -39,21 +39,19 @@
 namespace cg = cooperative_groups;
 
 #include <helper_cuda.h>
+
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Helper functions
 ////////////////////////////////////////////////////////////////////////////////
-static inline __host__ __device__ uint iDivUp(uint a, uint b) {
+static inline __host__ __device__ uint iDivUp(uint a, uint b) { return ((a % b) == 0) ? (a / b) : (a / b + 1); }
-  return ((a % b) == 0) ? (a / b) : (a / b + 1);
-}
 
-static inline __host__ __device__ uint getSampleCount(uint dividend) {
+static inline __host__ __device__ uint getSampleCount(uint dividend) { return iDivUp(dividend, SAMPLE_STRIDE); }
-  return iDivUp(dividend, SAMPLE_STRIDE);
-}
 
 #define W (sizeof(uint) * 8)
-static inline __device__ uint nextPowerOfTwo(uint x) {
+static inline __device__ uint nextPowerOfTwo(uint x)
+{
     /*
         --x;
         x |= x >> 1;
@@ -66,9 +64,8 @@ static inline __device__ uint nextPowerOfTwo(uint x) {
     return 1U << (W - __clz(x - 1));
 }
 
-template <uint sortDir>
+template <uint sortDir> static inline __device__ uint binarySearchInclusive(uint val, uint *data, uint L, uint stride)
-static inline __device__ uint binarySearchInclusive(uint val, uint *data,
+{
-                                                    uint L, uint stride) {
     if (L == 0) {
         return 0;
     }
@@ -78,8 +75,7 @@ static inline __device__ uint binarySearchInclusive(uint val, uint *data,
     for (; stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] <= val)) ||
+        if ((sortDir && (data[newPos - 1] <= val)) || (!sortDir && (data[newPos - 1] >= val))) {
-        (!sortDir && (data[newPos - 1] >= val))) {
             pos = newPos;
         }
     }
@@ -87,9 +83,8 @@ static inline __device__ uint binarySearchInclusive(uint val, uint *data,
     return pos;
 }
 
-template <uint sortDir>
+template <uint sortDir> static inline __device__ uint binarySearchExclusive(uint val, uint *data, uint L, uint stride)
-static inline __device__ uint binarySearchExclusive(uint val, uint *data,
+{
-                                                    uint L, uint stride) {
     if (L == 0) {
         return 0;
     }
@@ -99,8 +94,7 @@ static inline __device__ uint binarySearchExclusive(uint val, uint *data,
     for (; stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] < val)) ||
+        if ((sortDir && (data[newPos - 1] < val)) || (!sortDir && (data[newPos - 1] > val))) {
-        (!sortDir && (data[newPos - 1] > val))) {
             pos = newPos;
         }
     }
@@ -112,9 +106,8 @@ static inline __device__ uint binarySearchExclusive(uint val, uint *data,
 // Bottom-level merge sort (binary search-based)
 ////////////////////////////////////////////////////////////////////////////////
 template <uint sortDir>
-__global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
+__global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey, uint *d_SrcVal, uint arrayLength)
-                                      uint *d_SrcKey, uint *d_SrcVal,
+{
-                                      uint arrayLength) {
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     __shared__ uint  s_key[SHARED_SIZE_LIMIT];
@@ -126,10 +119,8 @@ __global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
     d_DstVal += blockIdx.x * SHARED_SIZE_LIMIT + threadIdx.x;
     s_key[threadIdx.x + 0]                       = d_SrcKey[0];
     s_val[threadIdx.x + 0]                       = d_SrcVal[0];
-  s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
+    s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
-      d_SrcKey[(SHARED_SIZE_LIMIT / 2)];
+    s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] = d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
-  s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)] =
-      d_SrcVal[(SHARED_SIZE_LIMIT / 2)];
 
     for (uint stride = 1; stride < arrayLength; stride <<= 1) {
         uint  lPos    = threadIdx.x & (stride - 1);
@@ -141,12 +132,8 @@ __global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
         uint valA = baseVal[lPos + 0];
         uint keyB = baseKey[lPos + stride];
         uint valB = baseVal[lPos + stride];
-    uint posA =
+        uint posA = binarySearchExclusive<sortDir>(keyA, baseKey + stride, stride, stride) + lPos;
-        binarySearchExclusive<sortDir>(keyA, baseKey + stride, stride, stride) +
+        uint posB = binarySearchInclusive<sortDir>(keyB, baseKey + 0, stride, stride) + lPos;
-        lPos;
-    uint posB =
-        binarySearchInclusive<sortDir>(keyB, baseKey + 0, stride, stride) +
-        lPos;
 
         cg::sync(cta);
         baseKey[posA] = keyA;
@@ -158,15 +145,18 @@ __global__ void mergeSortSharedKernel(uint *d_DstKey, uint *d_DstVal,
     cg::sync(cta);
     d_DstKey[0]                       = s_key[threadIdx.x + 0];
     d_DstVal[0]                       = s_val[threadIdx.x + 0];
-  d_DstKey[(SHARED_SIZE_LIMIT / 2)] =
+    d_DstKey[(SHARED_SIZE_LIMIT / 2)] = s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-      s_key[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
+    d_DstVal[(SHARED_SIZE_LIMIT / 2)] = s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
-  d_DstVal[(SHARED_SIZE_LIMIT / 2)] =
-      s_val[threadIdx.x + (SHARED_SIZE_LIMIT / 2)];
 }
 
-static void mergeSortShared(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey,
+static void mergeSortShared(uint *d_DstKey,
-                            uint *d_SrcVal, uint batchSize, uint arrayLength,
+                            uint *d_DstVal,
-                            uint sortDir) {
+                            uint *d_SrcKey,
+                            uint *d_SrcVal,
+                            uint  batchSize,
+                            uint  arrayLength,
+                            uint  sortDir)
+{
     if (arrayLength < 2) {
         return;
     }
@@ -177,12 +167,11 @@ static void mergeSortShared(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey,
     uint threadCount = SHARED_SIZE_LIMIT / 2;
 
     if (sortDir) {
-    mergeSortSharedKernel<1U><<<blockCount, threadCount>>>(
+        mergeSortSharedKernel<1U><<<blockCount, threadCount>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
         getLastCudaError("mergeSortShared<1><<<>>> failed\n");
-  } else {
+    }
-    mergeSortSharedKernel<0U><<<blockCount, threadCount>>>(
+    else {
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
+        mergeSortSharedKernel<0U><<<blockCount, threadCount>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, arrayLength);
         getLastCudaError("mergeSortShared<0><<<>>> failed\n");
     }
 }
@@ -191,9 +180,9 @@ static void mergeSortShared(uint *d_DstKey, uint *d_DstVal, uint *d_SrcKey,
 // Merge step 1: generate sample ranks
 ////////////////////////////////////////////////////////////////////////////////
 template <uint sortDir>
-__global__ void generateSampleRanksKernel(uint *d_RanksA, uint *d_RanksB,
+__global__ void
-                                          uint *d_SrcKey, uint stride, uint N,
+generateSampleRanksKernel(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey, uint stride, uint N, uint threadCount)
-                                          uint threadCount) {
+{
     uint pos = blockIdx.x * blockDim.x + threadIdx.x;
 
     if (pos >= threadCount) {
@@ -214,33 +203,30 @@ __global__ void generateSampleRanksKernel(uint *d_RanksA, uint *d_RanksB,
     if (i < segmentSamplesA) {
         d_RanksA[i] = i * SAMPLE_STRIDE;
         d_RanksB[i] = binarySearchExclusive<sortDir>(
-        d_SrcKey[i * SAMPLE_STRIDE], d_SrcKey + stride, segmentElementsB,
+            d_SrcKey[i * SAMPLE_STRIDE], d_SrcKey + stride, segmentElementsB, nextPowerOfTwo(segmentElementsB));
-        nextPowerOfTwo(segmentElementsB));
     }
 
     if (i < segmentSamplesB) {
         d_RanksB[(stride / SAMPLE_STRIDE) + i] = i * SAMPLE_STRIDE;
         d_RanksA[(stride / SAMPLE_STRIDE) + i] = binarySearchInclusive<sortDir>(
-        d_SrcKey[stride + i * SAMPLE_STRIDE], d_SrcKey + 0, segmentElementsA,
+            d_SrcKey[stride + i * SAMPLE_STRIDE], d_SrcKey + 0, segmentElementsA, nextPowerOfTwo(segmentElementsA));
-        nextPowerOfTwo(segmentElementsA));
     }
 }
 
-static void generateSampleRanks(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey,
+static void generateSampleRanks(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey, uint stride, uint N, uint sortDir)
-                                uint stride, uint N, uint sortDir) {
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint threadCount =
+    uint threadCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
     if (sortDir) {
-    generateSampleRanksKernel<1U><<<iDivUp(threadCount, 256), 256>>>(
+        generateSampleRanksKernel<1U>
-        d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
+            <<<iDivUp(threadCount, 256), 256>>>(d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
         getLastCudaError("generateSampleRanksKernel<1U><<<>>> failed\n");
-  } else {
+    }
-    generateSampleRanksKernel<0U><<<iDivUp(threadCount, 256), 256>>>(
+    else {
-        d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
+        generateSampleRanksKernel<0U>
+            <<<iDivUp(threadCount, 256), 256>>>(d_RanksA, d_RanksB, d_SrcKey, stride, N, threadCount);
         getLastCudaError("generateSampleRanksKernel<0U><<<>>> failed\n");
     }
 }
@@ -248,9 +234,8 @@ static void generateSampleRanks(uint *d_RanksA, uint *d_RanksB, uint *d_SrcKey,
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 2: generate sample ranks and indices
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void mergeRanksAndIndicesKernel(uint *d_Limits, uint *d_Ranks,
+__global__ void mergeRanksAndIndicesKernel(uint *d_Limits, uint *d_Ranks, uint stride, uint N, uint threadCount)
-                                           uint stride, uint N,
+{
-                                           uint threadCount) {
     uint pos = blockIdx.x * blockDim.x + threadIdx.x;
 
     if (pos >= threadCount) {
@@ -269,36 +254,29 @@ __global__ void mergeRanksAndIndicesKernel(uint *d_Limits, uint *d_Ranks,
 
     if (i < segmentSamplesA) {
         uint dstPos = binarySearchExclusive<1U>(
-                      d_Ranks[i], d_Ranks + segmentSamplesA, segmentSamplesB,
+                          d_Ranks[i], d_Ranks + segmentSamplesA, segmentSamplesB, nextPowerOfTwo(segmentSamplesB))
-                      nextPowerOfTwo(segmentSamplesB)) +
+                    + i;
-                  i;
         d_Limits[dstPos] = d_Ranks[i];
     }
 
     if (i < segmentSamplesB) {
-    uint dstPos = binarySearchInclusive<1U>(d_Ranks[segmentSamplesA + i],
+        uint dstPos = binarySearchInclusive<1U>(
-                                            d_Ranks, segmentSamplesA,
+                          d_Ranks[segmentSamplesA + i], d_Ranks, segmentSamplesA, nextPowerOfTwo(segmentSamplesA))
-                                            nextPowerOfTwo(segmentSamplesA)) +
+                    + i;
-                  i;
         d_Limits[dstPos] = d_Ranks[segmentSamplesA + i];
     }
 }
 
-static void mergeRanksAndIndices(uint *d_LimitsA, uint *d_LimitsB,
+static void mergeRanksAndIndices(uint *d_LimitsA, uint *d_LimitsB, uint *d_RanksA, uint *d_RanksB, uint stride, uint N)
-                                 uint *d_RanksA, uint *d_RanksB, uint stride,
+{
-                                 uint N) {
     uint lastSegmentElements = N % (2 * stride);
-  uint threadCount =
+    uint threadCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
-  mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(
+    mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(d_LimitsA, d_RanksA, stride, N, threadCount);
-      d_LimitsA, d_RanksA, stride, N, threadCount);
     getLastCudaError("mergeRanksAndIndicesKernel(A)<<<>>> failed\n");
 
-  mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(
+    mergeRanksAndIndicesKernel<<<iDivUp(threadCount, 256), 256>>>(d_LimitsB, d_RanksB, stride, N, threadCount);
-      d_LimitsB, d_RanksB, stride, N, threadCount);
     getLastCudaError("mergeRanksAndIndicesKernel(B)<<<>>> failed\n");
 }
 
@@ -306,24 +284,30 @@ static void mergeRanksAndIndices(uint *d_LimitsA, uint *d_LimitsB,
 // Merge step 3: merge elementary intervals
 ////////////////////////////////////////////////////////////////////////////////
 template <uint sortDir>
-inline __device__ void merge(uint *dstKey, uint *dstVal, uint *srcAKey,
+inline __device__ void merge(uint            *dstKey,
-                             uint *srcAVal, uint *srcBKey, uint *srcBVal,
+                             uint            *dstVal,
-                             uint lenA, uint nPowTwoLenA, uint lenB,
+                             uint            *srcAKey,
-                             uint nPowTwoLenB, cg::thread_block cta) {
+                             uint            *srcAVal,
+                             uint            *srcBKey,
+                             uint            *srcBVal,
+                             uint             lenA,
+                             uint             nPowTwoLenA,
+                             uint             lenB,
+                             uint             nPowTwoLenB,
+                             cg::thread_block cta)
+{
     uint keyA, valA, keyB, valB, dstPosA, dstPosB;
 
     if (threadIdx.x < lenA) {
         keyA    = srcAKey[threadIdx.x];
         valA    = srcAVal[threadIdx.x];
-    dstPosA = binarySearchExclusive<sortDir>(keyA, srcBKey, lenB, nPowTwoLenB) +
+        dstPosA = binarySearchExclusive<sortDir>(keyA, srcBKey, lenB, nPowTwoLenB) + threadIdx.x;
-              threadIdx.x;
     }
 
     if (threadIdx.x < lenB) {
         keyB    = srcBKey[threadIdx.x];
         valB    = srcBVal[threadIdx.x];
-    dstPosB = binarySearchInclusive<sortDir>(keyB, srcAKey, lenA, nPowTwoLenA) +
+        dstPosB = binarySearchInclusive<sortDir>(keyB, srcAKey, lenA, nPowTwoLenA) + threadIdx.x;
-              threadIdx.x;
     }
 
     cg::sync(cta);
@@ -340,10 +324,15 @@ inline __device__ void merge(uint *dstKey, uint *dstVal, uint *srcAKey,
 }
 
 template <uint sortDir>
-__global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
+__global__ void mergeElementaryIntervalsKernel(uint *d_DstKey,
-                                               uint *d_SrcKey, uint *d_SrcVal,
+                                               uint *d_DstVal,
-                                               uint *d_LimitsA, uint *d_LimitsB,
+                                               uint *d_SrcKey,
-                                               uint stride, uint N) {
+                                               uint *d_SrcVal,
+                                               uint *d_LimitsA,
+                                               uint *d_LimitsB,
+                                               uint  stride,
+                                               uint  N)
+{
     // Handle to thread block group
     cg::thread_block cta = cg::this_thread_block();
     __shared__ uint  s_key[2 * SAMPLE_STRIDE];
@@ -368,10 +357,8 @@ __global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
 
         startSrcA    = d_LimitsA[blockIdx.x];
         startSrcB    = d_LimitsB[blockIdx.x];
-    uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1]
+        uint endSrcA = (intervalI + 1 < segmentSamples) ? d_LimitsA[blockIdx.x + 1] : segmentElementsA;
-                                                    : segmentElementsA;
+        uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1] : segmentElementsB;
-    uint endSrcB = (intervalI + 1 < segmentSamples) ? d_LimitsB[blockIdx.x + 1]
-                                                    : segmentElementsB;
         lenSrcA      = endSrcA - startSrcA;
         lenSrcB      = endSrcB - startSrcB;
         startDstA    = startSrcA + startSrcB;
@@ -387,17 +374,23 @@ __global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
     }
 
     if (threadIdx.x < lenSrcB) {
-    s_key[threadIdx.x + SAMPLE_STRIDE] =
+        s_key[threadIdx.x + SAMPLE_STRIDE] = d_SrcKey[stride + startSrcB + threadIdx.x];
-        d_SrcKey[stride + startSrcB + threadIdx.x];
+        s_val[threadIdx.x + SAMPLE_STRIDE] = d_SrcVal[stride + startSrcB + threadIdx.x];
-    s_val[threadIdx.x + SAMPLE_STRIDE] =
-        d_SrcVal[stride + startSrcB + threadIdx.x];
     }
 
     // Merge data in shared memory
     cg::sync(cta);
-  merge<sortDir>(s_key, s_val, s_key + 0, s_val + 0, s_key + SAMPLE_STRIDE,
+    merge<sortDir>(s_key,
-                 s_val + SAMPLE_STRIDE, lenSrcA, SAMPLE_STRIDE, lenSrcB,
+                   s_val,
-                 SAMPLE_STRIDE, cta);
+                   s_key + 0,
+                   s_val + 0,
+                   s_key + SAMPLE_STRIDE,
+                   s_val + SAMPLE_STRIDE,
+                   lenSrcA,
+                   SAMPLE_STRIDE,
+                   lenSrcB,
+                   SAMPLE_STRIDE,
+                   cta);
 
     // Store merged data
     cg::sync(cta);
@@ -413,63 +406,77 @@ __global__ void mergeElementaryIntervalsKernel(uint *d_DstKey, uint *d_DstVal,
     }
 }
 
-static void mergeElementaryIntervals(uint *d_DstKey, uint *d_DstVal,
+static void mergeElementaryIntervals(uint *d_DstKey,
-                                     uint *d_SrcKey, uint *d_SrcVal,
+                                     uint *d_DstVal,
-                                     uint *d_LimitsA, uint *d_LimitsB,
+                                     uint *d_SrcKey,
-                                     uint stride, uint N, uint sortDir) {
+                                     uint *d_SrcVal,
+                                     uint *d_LimitsA,
+                                     uint *d_LimitsB,
+                                     uint  stride,
+                                     uint  N,
+                                     uint  sortDir)
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint mergePairs = (lastSegmentElements > stride)
+    uint mergePairs = (lastSegmentElements > stride) ? getSampleCount(N) : (N - lastSegmentElements) / SAMPLE_STRIDE;
-                        ? getSampleCount(N)
-                        : (N - lastSegmentElements) / SAMPLE_STRIDE;
 
     if (sortDir) {
-    mergeElementaryIntervalsKernel<1U><<<mergePairs, SAMPLE_STRIDE>>>(
+        mergeElementaryIntervalsKernel<1U>
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
-        N);
         getLastCudaError("mergeElementaryIntervalsKernel<1> failed\n");
-  } else {
+    }
-    mergeElementaryIntervalsKernel<0U><<<mergePairs, SAMPLE_STRIDE>>>(
+    else {
-        d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride,
+        mergeElementaryIntervalsKernel<0U>
-        N);
+            <<<mergePairs, SAMPLE_STRIDE>>>(d_DstKey, d_DstVal, d_SrcKey, d_SrcVal, d_LimitsA, d_LimitsB, stride, N);
         getLastCudaError("mergeElementaryIntervalsKernel<0> failed\n");
     }
 }
 
-extern "C" void bitonicSortShared(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicSortShared(uint *d_DstKey,
-                                  uint *d_SrcKey, uint *d_SrcVal,
+                                  uint *d_DstVal,
-                                  uint batchSize, uint arrayLength,
+                                  uint *d_SrcKey,
+                                  uint *d_SrcVal,
+                                  uint  batchSize,
+                                  uint  arrayLength,
                                   uint  sortDir);
 
-extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey, uint *d_DstVal,
+extern "C" void bitonicMergeElementaryIntervals(uint *d_DstKey,
-                                                uint *d_SrcKey, uint *d_SrcVal,
+                                                uint *d_DstVal,
+                                                uint *d_SrcKey,
+                                                uint *d_SrcVal,
                                                 uint *d_LimitsA,
-                                                uint *d_LimitsB, uint stride,
+                                                uint *d_LimitsB,
-                                                uint N, uint sortDir);
+                                                uint  stride,
+                                                uint  N,
+                                                uint  sortDir);
 
 static uint      *d_RanksA, *d_RanksB, *d_LimitsA, *d_LimitsB;
 static const uint MAX_SAMPLE_COUNT = 32768;
 
-extern "C" void initMergeSort(void) {
+extern "C" void initMergeSort(void)
-  checkCudaErrors(
+{
-      cudaMalloc((void **)&d_RanksA, MAX_SAMPLE_COUNT * sizeof(uint)));
+    checkCudaErrors(cudaMalloc((void **)&d_RanksA, MAX_SAMPLE_COUNT * sizeof(uint)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&d_RanksB, MAX_SAMPLE_COUNT * sizeof(uint)));
-      cudaMalloc((void **)&d_RanksB, MAX_SAMPLE_COUNT * sizeof(uint)));
+    checkCudaErrors(cudaMalloc((void **)&d_LimitsA, MAX_SAMPLE_COUNT * sizeof(uint)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc((void **)&d_LimitsB, MAX_SAMPLE_COUNT * sizeof(uint)));
-      cudaMalloc((void **)&d_LimitsA, MAX_SAMPLE_COUNT * sizeof(uint)));
-  checkCudaErrors(
-      cudaMalloc((void **)&d_LimitsB, MAX_SAMPLE_COUNT * sizeof(uint)));
 }
 
-extern "C" void closeMergeSort(void) {
+extern "C" void closeMergeSort(void)
+{
     checkCudaErrors(cudaFree(d_RanksA));
     checkCudaErrors(cudaFree(d_RanksB));
     checkCudaErrors(cudaFree(d_LimitsB));
     checkCudaErrors(cudaFree(d_LimitsA));
 }
 
-extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
+extern "C" void mergeSort(uint *d_DstKey,
-                          uint *d_BufVal, uint *d_SrcKey, uint *d_SrcVal,
+                          uint *d_DstVal,
-                          uint N, uint sortDir) {
+                          uint *d_BufKey,
+                          uint *d_BufVal,
+                          uint *d_SrcKey,
+                          uint *d_SrcVal,
+                          uint  N,
+                          uint  sortDir)
+{
     uint stageCount = 0;
 
     for (uint stride = SHARED_SIZE_LIMIT; stride < N; stride <<= 1, stageCount++)
@@ -482,7 +489,8 @@ extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
         ival = d_BufVal;
         okey = d_DstKey;
         oval = d_DstVal;
-  } else {
+    }
+    else {
         ikey = d_DstKey;
         ival = d_DstVal;
         okey = d_BufKey;
@@ -491,8 +499,7 @@ extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
 
     assert(N <= (SAMPLE_STRIDE * MAX_SAMPLE_COUNT));
     assert(N % SHARED_SIZE_LIMIT == 0);
-  mergeSortShared(ikey, ival, d_SrcKey, d_SrcVal, N / SHARED_SIZE_LIMIT,
+    mergeSortShared(ikey, ival, d_SrcKey, d_SrcVal, N / SHARED_SIZE_LIMIT, SHARED_SIZE_LIMIT, sortDir);
-                  SHARED_SIZE_LIMIT, sortDir);
 
     for (uint stride = SHARED_SIZE_LIMIT; stride < N; stride <<= 1) {
         uint lastSegmentElements = N % (2 * stride);
@@ -504,18 +511,19 @@ extern "C" void mergeSort(uint *d_DstKey, uint *d_DstVal, uint *d_BufKey,
         mergeRanksAndIndices(d_LimitsA, d_LimitsB, d_RanksA, d_RanksB, stride, N);
 
         // Merge elementary intervals
-    mergeElementaryIntervals(okey, oval, ikey, ival, d_LimitsA, d_LimitsB,
+        mergeElementaryIntervals(okey, oval, ikey, ival, d_LimitsA, d_LimitsB, stride, N, sortDir);
-                             stride, N, sortDir);
 
         if (lastSegmentElements <= stride) {
             // Last merge segment consists of a single array which just needs to be
             // passed through
-      checkCudaErrors(cudaMemcpy(
+            checkCudaErrors(cudaMemcpy(okey + (N - lastSegmentElements),
-          okey + (N - lastSegmentElements), ikey + (N - lastSegmentElements),
+                                       ikey + (N - lastSegmentElements),
-          lastSegmentElements * sizeof(uint), cudaMemcpyDeviceToDevice));
+                                       lastSegmentElements * sizeof(uint),
-      checkCudaErrors(cudaMemcpy(
+                                       cudaMemcpyDeviceToDevice));
-          oval + (N - lastSegmentElements), ival + (N - lastSegmentElements),
+            checkCudaErrors(cudaMemcpy(oval + (N - lastSegmentElements),
-          lastSegmentElements * sizeof(uint), cudaMemcpyDeviceToDevice));
+                                       ival + (N - lastSegmentElements),
+                                       lastSegmentElements * sizeof(uint),
+                                       cudaMemcpyDeviceToDevice));
         }
 
         uint *t;

Samples/0_Introduction/mergeSort/mergeSort_common.h

@@ -36,14 +36,12 @@ typedef unsigned int uint;
 ////////////////////////////////////////////////////////////////////////////////
 // Extensive sort validation routine
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
+extern "C" uint
-                                   uint arrayLength, uint numValues,
+validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize, uint arrayLength, uint numValues, uint sortDir);
-                                   uint sortDir);
 
 extern "C" void fillValues(uint *val, uint N);
 
-extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey,
+extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey, uint batchSize, uint arrayLength);
-                                    uint batchSize, uint arrayLength);
 
 ////////////////////////////////////////////////////////////////////////////////
 // CUDA merge sort
@@ -52,13 +50,11 @@ extern "C" void initMergeSort(void);
 
 extern "C" void closeMergeSort(void);
 
-extern "C" void mergeSort(uint *dstKey, uint *dstVal, uint *bufKey,
+extern "C" void
-                          uint *bufVal, uint *srcKey, uint *srcVal, uint N,
+mergeSort(uint *dstKey, uint *dstVal, uint *bufKey, uint *bufVal, uint *srcKey, uint *srcVal, uint N, uint sortDir);
-                          uint sortDir);
 
 ////////////////////////////////////////////////////////////////////////////////
 // CPU "emulation"
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
+extern "C" void
-                              uint *bufVal, uint *srcKey, uint *srcVal, uint N,
+mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey, uint *bufVal, uint *srcKey, uint *srcVal, uint N, uint sortDir);
-                              uint sortDir);

Samples/0_Introduction/mergeSort/mergeSort_host.cpp

@@ -29,19 +29,20 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Helper functions
 ////////////////////////////////////////////////////////////////////////////////
-static void checkOrder(uint *data, uint N, uint sortDir) {
+static void checkOrder(uint *data, uint N, uint sortDir)
+{
     if (N <= 1) {
         return;
     }
 
     for (uint i = 0; i < N - 1; i++)
-    if ((sortDir && (data[i] > data[i + 1])) ||
+        if ((sortDir && (data[i] > data[i + 1])) || (!sortDir && (data[i] < data[i + 1]))) {
-        (!sortDir && (data[i] < data[i + 1]))) {
             fprintf(stderr, "checkOrder() failed!!!\n");
             exit(EXIT_FAILURE);
         }
@@ -49,12 +50,13 @@ static void checkOrder(uint *data, uint N, uint sortDir) {
 
 static uint umin(uint a, uint b) { return (a <= b) ? a : b; }
 
-static uint getSampleCount(uint dividend) {
+static uint getSampleCount(uint dividend)
-  return ((dividend % SAMPLE_STRIDE) != 0) ? (dividend / SAMPLE_STRIDE + 1)
+{
-                                           : (dividend / SAMPLE_STRIDE);
+    return ((dividend % SAMPLE_STRIDE) != 0) ? (dividend / SAMPLE_STRIDE + 1) : (dividend / SAMPLE_STRIDE);
 }
 
-static uint nextPowerOfTwo(uint x) {
+static uint nextPowerOfTwo(uint x)
+{
     --x;
     x |= x >> 1;
     x |= x >> 2;
@@ -64,7 +66,8 @@ static uint nextPowerOfTwo(uint x) {
     return ++x;
 }
 
-static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir) {
+static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir)
+{
     if (L == 0) {
         return 0;
     }
@@ -74,8 +77,7 @@ static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir) {
     for (uint stride = nextPowerOfTwo(L); stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] <= val)) ||
+        if ((sortDir && (data[newPos - 1] <= val)) || (!sortDir && (data[newPos - 1] >= val))) {
-        (!sortDir && (data[newPos - 1] >= val))) {
             pos = newPos;
         }
     }
@@ -83,7 +85,8 @@ static uint binarySearchInclusive(uint val, uint *data, uint L, uint sortDir) {
     return pos;
 }
 
-static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir) {
+static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir)
+{
     if (L == 0) {
         return 0;
     }
@@ -93,8 +96,7 @@ static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir) {
     for (uint stride = nextPowerOfTwo(L); stride > 0; stride >>= 1) {
         uint newPos = umin(pos + stride, L);
 
-    if ((sortDir && (data[newPos - 1] < val)) ||
+        if ((sortDir && (data[newPos - 1] < val)) || (!sortDir && (data[newPos - 1] > val))) {
-        (!sortDir && (data[newPos - 1] > val))) {
             pos = newPos;
         }
     }
@@ -105,12 +107,10 @@ static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir) {
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 1: find sample ranks in each segment
 ////////////////////////////////////////////////////////////////////////////////
-static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey,
+static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey, uint stride, uint N, uint sortDir)
-                                uint stride, uint N, uint sortDir) {
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint sampleCount =
+    uint sampleCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
     for (uint pos = 0; pos < sampleCount; pos++) {
@@ -124,17 +124,14 @@ static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey,
 
         if (i < nA) {
             ranksA[(segmentBase + 0) / SAMPLE_STRIDE + i] = i * SAMPLE_STRIDE;
-      ranksB[(segmentBase + 0) / SAMPLE_STRIDE + i] =
+            ranksB[(segmentBase + 0) / SAMPLE_STRIDE + i] = binarySearchExclusive(
-          binarySearchExclusive(srcKey[segmentBase + i * SAMPLE_STRIDE],
+                srcKey[segmentBase + i * SAMPLE_STRIDE], srcKey + segmentBase + stride, lenB, sortDir);
-                                srcKey + segmentBase + stride, lenB, sortDir);
         }
 
         if (i < nB) {
             ranksB[(segmentBase + stride) / SAMPLE_STRIDE + i] = i * SAMPLE_STRIDE;
-      ranksA[(segmentBase + stride) / SAMPLE_STRIDE + i] =
+            ranksA[(segmentBase + stride) / SAMPLE_STRIDE + i] = binarySearchInclusive(
-          binarySearchInclusive(
+                srcKey[segmentBase + stride + i * SAMPLE_STRIDE], srcKey + segmentBase, lenA, sortDir);
-              srcKey[segmentBase + stride + i * SAMPLE_STRIDE],
-              srcKey + segmentBase, lenA, sortDir);
         }
     }
 }
@@ -142,12 +139,10 @@ static void generateSampleRanks(uint *ranksA, uint *ranksB, uint *srcKey,
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 2: merge ranks and indices to derive elementary intervals
 ////////////////////////////////////////////////////////////////////////////////
-static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride,
+static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride, uint N)
-                                 uint N) {
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint sampleCount =
+    uint sampleCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
-      (lastSegmentElements > stride)
-          ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE)
                                                       : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
 
     for (uint pos = 0; pos < sampleCount; pos++) {
@@ -161,23 +156,20 @@ static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride,
 
         if (i < nA) {
             uint dstPosA =
-          binarySearchExclusive(ranks[(segmentBase + 0) / SAMPLE_STRIDE + i],
+                binarySearchExclusive(
-                                ranks + (segmentBase + stride) / SAMPLE_STRIDE,
+                    ranks[(segmentBase + 0) / SAMPLE_STRIDE + i], ranks + (segmentBase + stride) / SAMPLE_STRIDE, nB, 1)
-                                nB, 1) +
+                + i;
-          i;
             assert(dstPosA < nA + nB);
-      limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] =
+            limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] = ranks[(segmentBase + 0) / SAMPLE_STRIDE + i];
-          ranks[(segmentBase + 0) / SAMPLE_STRIDE + i];
         }
 
         if (i < nB) {
-      uint dstPosA = binarySearchInclusive(
+            uint dstPosA =
-                         ranks[(segmentBase + stride) / SAMPLE_STRIDE + i],
+                binarySearchInclusive(
-                         ranks + (segmentBase + 0) / SAMPLE_STRIDE, nA, 1) +
+                    ranks[(segmentBase + stride) / SAMPLE_STRIDE + i], ranks + (segmentBase + 0) / SAMPLE_STRIDE, nA, 1)
-                     i;
+                + i;
             assert(dstPosA < nA + nB);
-      limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] =
+            limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] = ranks[(segmentBase + stride) / SAMPLE_STRIDE + i];
-          ranks[(segmentBase + stride) / SAMPLE_STRIDE + i];
         }
     }
 }
@@ -185,9 +177,16 @@ static void mergeRanksAndIndices(uint *limits, uint *ranks, uint stride,
 ////////////////////////////////////////////////////////////////////////////////
 // Merge step 3: merge elementary intervals (each interval is <= SAMPLE_STRIDE)
 ////////////////////////////////////////////////////////////////////////////////
-static void merge(uint *dstKey, uint *dstVal, uint *srcAKey, uint *srcAVal,
+static void merge(uint *dstKey,
-                  uint *srcBKey, uint *srcBVal, uint lenA, uint lenB,
+                  uint *dstVal,
-                  uint sortDir) {
+                  uint *srcAKey,
+                  uint *srcAVal,
+                  uint *srcBKey,
+                  uint *srcBVal,
+                  uint  lenA,
+                  uint  lenB,
+                  uint  sortDir)
+{
     checkOrder(srcAKey, lenA, sortDir);
     checkOrder(srcBKey, lenB, sortDir);
 
@@ -206,13 +205,18 @@ static void merge(uint *dstKey, uint *dstVal, uint *srcAKey, uint *srcAVal,
     }
 }
 
-static void mergeElementaryIntervals(uint *dstKey, uint *dstVal, uint *srcKey,
+static void mergeElementaryIntervals(uint *dstKey,
-                                     uint *srcVal, uint *limitsA, uint *limitsB,
+                                     uint *dstVal,
-                                     uint stride, uint N, uint sortDir) {
+                                     uint *srcKey,
+                                     uint *srcVal,
+                                     uint *limitsA,
+                                     uint *limitsB,
+                                     uint  stride,
+                                     uint  N,
+                                     uint  sortDir)
+{
     uint lastSegmentElements = N % (2 * stride);
-  uint mergePairs = (lastSegmentElements > stride)
+    uint mergePairs = (lastSegmentElements > stride) ? getSampleCount(N) : (N - lastSegmentElements) / SAMPLE_STRIDE;
-                        ? getSampleCount(N)
-                        : (N - lastSegmentElements) / SAMPLE_STRIDE;
 
     for (uint pos = 0; pos < mergePairs; pos++) {
         uint i           = pos & ((2 * stride) / SAMPLE_STRIDE - 1);
@@ -240,15 +244,18 @@ static void mergeElementaryIntervals(uint *dstKey, uint *dstVal, uint *srcKey,
               (srcKey + segmentBase + 0) + startPosA,
               (srcVal + segmentBase + 0) + startPosA,
               (srcKey + segmentBase + stride) + startPosB,
-          (srcVal + segmentBase + stride) + startPosB, endPosA - startPosA,
+              (srcVal + segmentBase + stride) + startPosB,
-          endPosB - startPosB, sortDir);
+              endPosA - startPosA,
+              endPosB - startPosB,
+              sortDir);
     }
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Retarded bubble sort
 ////////////////////////////////////////////////////////////////////////////////
-static void bubbleSort(uint *key, uint *val, uint N, uint sortDir) {
+static void bubbleSort(uint *key, uint *val, uint N, uint sortDir)
+{
     if (N <= 1) {
         return;
     }
@@ -278,9 +285,9 @@ static void bubbleSort(uint *key, uint *val, uint N, uint sortDir) {
 ////////////////////////////////////////////////////////////////////////////////
 // Interface function
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
+extern "C" void
-                              uint *bufVal, uint *srcKey, uint *srcVal, uint N,
+mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey, uint *bufVal, uint *srcKey, uint *srcVal, uint N, uint sortDir)
-                              uint sortDir) {
+{
     uint *ikey, *ival, *okey, *oval;
     uint  stageCount = 0;
 
@@ -292,7 +299,8 @@ extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
         ival = bufVal;
         okey = dstKey;
         oval = dstVal;
-  } else {
+    }
+    else {
         ikey = dstKey;
         ival = dstVal;
         okey = bufKey;
@@ -304,8 +312,7 @@ extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
     memcpy(ival, srcVal, N * sizeof(uint));
 
     for (uint pos = 0; pos < N; pos += SHARED_SIZE_LIMIT) {
-    bubbleSort(ikey + pos, ival + pos, umin(SHARED_SIZE_LIMIT, N - pos),
+        bubbleSort(ikey + pos, ival + pos, umin(SHARED_SIZE_LIMIT, N - pos), sortDir);
-               sortDir);
     }
 
     printf("Merge...\n");
@@ -329,16 +336,15 @@ extern "C" void mergeSortHost(uint *dstKey, uint *dstVal, uint *bufKey,
         mergeRanksAndIndices(limitsB, ranksB, stride, N);
 
         // Merge elementary intervals
-    mergeElementaryIntervals(okey, oval, ikey, ival, limitsA, limitsB, stride,
+        mergeElementaryIntervals(okey, oval, ikey, ival, limitsA, limitsB, stride, N, sortDir);
-                             N, sortDir);
 
         if (lastSegmentElements <= stride) {
             // Last merge segment consists of a single array which just needs to be
             // passed through
-      memcpy(okey + (N - lastSegmentElements), ikey + (N - lastSegmentElements),
+            memcpy(
-             lastSegmentElements * sizeof(uint));
+                okey + (N - lastSegmentElements), ikey + (N - lastSegmentElements), lastSegmentElements * sizeof(uint));
-      memcpy(oval + (N - lastSegmentElements), ival + (N - lastSegmentElements),
+            memcpy(
-             lastSegmentElements * sizeof(uint));
+                oval + (N - lastSegmentElements), ival + (N - lastSegmentElements), lastSegmentElements * sizeof(uint));
         }
 
         uint *t;

Samples/0_Introduction/mergeSort/mergeSort_validate.cpp

@@ -29,14 +29,15 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+
 #include "mergeSort_common.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Validate sorted keys array (check for integrity and proper order)
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
+extern "C" uint
-                                   uint arrayLength, uint numValues,
+validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize, uint arrayLength, uint numValues, uint sortDir)
-                                   uint sortDir) {
+{
     uint *srcHist;
     uint *resHist;
 
@@ -51,8 +52,7 @@ extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
 
     int flag = 1;
 
-  for (uint j = 0; j < batchSize;
+    for (uint j = 0; j < batchSize; j++, srcKey += arrayLength, resKey += arrayLength) {
-       j++, srcKey += arrayLength, resKey += arrayLength) {
         // Build histograms for keys arrays
         memset(srcHist, 0, numValues * sizeof(uint));
         memset(resHist, 0, numValues * sizeof(uint));
@@ -61,11 +61,9 @@ extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
             if ((srcKey[i] < numValues) && (resKey[i] < numValues)) {
                 srcHist[srcKey[i]]++;
                 resHist[resKey[i]]++;
-      } else {
+            }
-        fprintf(
+            else {
-            stderr,
+                fprintf(stderr, "***Set %u source/result key arrays are not limited properly***\n", j);
-            "***Set %u source/result key arrays are not limited properly***\n",
-            j);
                 flag = 0;
                 goto brk;
             }
@@ -74,18 +72,15 @@ extern "C" uint validateSortedKeys(uint *resKey, uint *srcKey, uint batchSize,
         // Compare the histograms
         for (uint i = 0; i < numValues; i++)
             if (srcHist[i] != resHist[i]) {
-        fprintf(stderr,
+                fprintf(stderr, "***Set %u source/result keys histograms do not match***\n", j);
-                "***Set %u source/result keys histograms do not match***\n", j);
                 flag = 0;
                 goto brk;
             }
 
         // Finally check the ordering
         for (uint i = 0; i < arrayLength - 1; i++)
-      if ((sortDir && (resKey[i] > resKey[i + 1])) ||
+            if ((sortDir && (resKey[i] > resKey[i + 1])) || (!sortDir && (resKey[i] < resKey[i + 1]))) {
-          (!sortDir && (resKey[i] < resKey[i + 1]))) {
+                fprintf(stderr, "***Set %u result key array is not ordered properly***\n", j);
-        fprintf(stderr,
-                "***Set %u result key array is not ordered properly***\n", j);
                 flag = 0;
                 goto brk;
             }
@@ -95,7 +90,8 @@ brk:
     free(resHist);
     free(srcHist);
 
-  if (flag) printf("OK\n");
+    if (flag)
+        printf("OK\n");
 
     return flag;
 }
@@ -103,30 +99,30 @@ brk:
 ////////////////////////////////////////////////////////////////////////////////
 // Value validation / stability check routines
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void fillValues(uint *val, uint N) {
+extern "C" void fillValues(uint *val, uint N)
-  for (uint i = 0; i < N; i++) val[i] = i;
+{
+    for (uint i = 0; i < N; i++)
+        val[i] = i;
 }
 
-extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey,
+extern "C" int validateSortedValues(uint *resKey, uint *resVal, uint *srcKey, uint batchSize, uint arrayLength)
-                                    uint batchSize, uint arrayLength) {
+{
     int correctFlag = 1, stableFlag = 1;
 
     printf("...inspecting keys and values array: ");
 
-  for (uint i = 0; i < batchSize;
+    for (uint i = 0; i < batchSize; i++, resKey += arrayLength, resVal += arrayLength) {
-       i++, resKey += arrayLength, resVal += arrayLength) {
         for (uint j = 0; j < arrayLength; j++) {
-      if (resKey[j] != srcKey[resVal[j]]) correctFlag = 0;
+            if (resKey[j] != srcKey[resVal[j]])
+                correctFlag = 0;
 
-      if ((j < arrayLength - 1) && (resKey[j] == resKey[j + 1]) &&
+            if ((j < arrayLength - 1) && (resKey[j] == resKey[j + 1]) && (resVal[j] > resVal[j + 1]))
-          (resVal[j] > resVal[j + 1]))
                 stableFlag = 0;
         }
     }
 
     printf(correctFlag ? "OK\n" : "***corrupted!!!***\n");
-  printf(stableFlag ? "...stability property: stable!\n"
+    printf(stableFlag ? "...stability property: stable!\n" : "...stability property: NOT stable\n");
-                    : "...stability property: NOT stable\n");
 
     return correctFlag;
 }

Samples/0_Introduction/simpleAWBarrier/CMakeLists.txt

@@ -11,8 +11,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 set(CMAKE_CUDA_ARCHITECTURES 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
 
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleAWBarrier/README.md

@@ -30,7 +30,7 @@ cudaStreamCreateWithFlags, cudaFree, cudaDeviceGetAttribute, cudaMallocHost, cud
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/simpleAWBarrier/simpleAWBarrier.cu

@@ -29,9 +29,9 @@
 #include <stdio.h>
 
 // Includes CUDA
-#include <cuda_runtime.h>
-#include <cuda/barrier>
 #include <cooperative_groups.h>
+#include <cuda/barrier>
+#include <cuda_runtime.h>
 
 // Utilities and timing functions
 #include <helper_functions.h> // includes cuda.h and cuda_runtime_api.h
@@ -43,9 +43,11 @@ namespace cg = cooperative_groups;
 
 #if __CUDA_ARCH__ >= 700
 template <bool writeSquareRoot>
-__device__ void reduceBlockData(
+__device__ void reduceBlockData(cuda::barrier<cuda::thread_scope_block> &barrier,
-    cuda::barrier<cuda::thread_scope_block> &barrier,
+                                cg::thread_block_tile<32>               &tile32,
-    cg::thread_block_tile<32> &tile32, double &threadSum, double *result) {
+                                double                                  &threadSum,
+                                double                                  *result)
+{
     extern __shared__ double tmp[];
 
 #pragma unroll
@@ -62,9 +64,7 @@ __device__ void reduceBlockData(
 
     // The warp 0 will perform last round of reduction
     if (tile32.meta_group_rank() == 0) {
-    double beta = tile32.thread_rank() < tile32.meta_group_size()
+        double beta = tile32.thread_rank() < tile32.meta_group_size() ? tmp[tile32.thread_rank()] : 0.0;
-                      ? tmp[tile32.thread_rank()]
-                      : 0.0;
 
 #pragma unroll
         for (int offset = tile32.size() / 2; offset > 0; offset /= 2) {
@@ -81,8 +81,8 @@ __device__ void reduceBlockData(
 }
 #endif
 
-__global__ void normVecByDotProductAWBarrier(float *vecA, float *vecB,
+__global__ void normVecByDotProductAWBarrier(float *vecA, float *vecB, double *partialResults, int size)
-                                             double *partialResults, int size) {
+{
 #if __CUDA_ARCH__ >= 700
 #pragma diag_suppress static_var_with_dynamic_init
     cg::thread_block cta  = cg::this_thread_block();
@@ -105,8 +105,7 @@ __global__ void normVecByDotProductAWBarrier(float *vecA, float *vecB,
 
     // Each thread block performs reduction of partial dotProducts and writes to
     // global mem.
-  reduceBlockData<false>(barrier, tile32, threadSum,
+    reduceBlockData<false>(barrier, tile32, threadSum, &partialResults[blockIdx.x]);
-                         &partialResults[blockIdx.x]);
 
     cg::sync(grid);
 
@@ -137,15 +136,15 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", argv[0]);
 
     // This will pick the best possible CUDA capable device
     int dev = findCudaDevice(argc, (const char **)argv);
 
     int major = 0;
-  checkCudaErrors(
+    checkCudaErrors(cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, dev));
-      cudaDeviceGetAttribute(&major, cudaDevAttrComputeCapabilityMajor, dev));
 
     // Arrive-Wait Barrier require a GPU of Volta (SM7X) architecture or higher.
     if (major < 7) {
@@ -154,12 +153,10 @@ int main(int argc, char **argv) {
     }
 
     int supportsCooperativeLaunch = 0;
-  checkCudaErrors(cudaDeviceGetAttribute(&supportsCooperativeLaunch,
+    checkCudaErrors(cudaDeviceGetAttribute(&supportsCooperativeLaunch, cudaDevAttrCooperativeLaunch, dev));
-                                         cudaDevAttrCooperativeLaunch, dev));
 
     if (!supportsCooperativeLaunch) {
-    printf(
+        printf("\nSelected GPU (%d) does not support Cooperative Kernel Launch, "
-        "\nSelected GPU (%d) does not support Cooperative Kernel Launch, "
                "Waiving the run\n",
                dev);
         exit(EXIT_WAIVED);
@@ -171,7 +168,8 @@ int main(int argc, char **argv) {
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
+int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId)
+{
     float  *vecA, *d_vecA;
     float  *vecB, *d_vecB;
     double *d_partialResults;
@@ -191,16 +189,14 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
     cudaStream_t stream;
     checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
 
-  checkCudaErrors(cudaMemcpyAsync(d_vecA, vecA, sizeof(float) * size,
+    checkCudaErrors(cudaMemcpyAsync(d_vecA, vecA, sizeof(float) * size, cudaMemcpyHostToDevice, stream));
-                                  cudaMemcpyHostToDevice, stream));
+    checkCudaErrors(cudaMemcpyAsync(d_vecB, vecB, sizeof(float) * size, cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(cudaMemcpyAsync(d_vecB, vecB, sizeof(float) * size,
-                                  cudaMemcpyHostToDevice, stream));
 
     // Kernel configuration, where a one-dimensional
     // grid and one-dimensional blocks are configured.
     int minGridSize = 0, blockSize = 0;
-  checkCudaErrors(cudaOccupancyMaxPotentialBlockSize(
+    checkCudaErrors(
-      &minGridSize, &blockSize, (void *)normVecByDotProductAWBarrier, 0, size));
+        cudaOccupancyMaxPotentialBlockSize(&minGridSize, &blockSize, (void *)normVecByDotProductAWBarrier, 0, size));
 
     int smemSize = ((blockSize / 32) + 1) * sizeof(double);
 
@@ -209,28 +205,24 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
         &numBlocksPerSm, normVecByDotProductAWBarrier, blockSize, smemSize));
 
     int multiProcessorCount = 0;
-  checkCudaErrors(cudaDeviceGetAttribute(
+    checkCudaErrors(cudaDeviceGetAttribute(&multiProcessorCount, cudaDevAttrMultiProcessorCount, deviceId));
-      &multiProcessorCount, cudaDevAttrMultiProcessorCount, deviceId));
 
     minGridSize = multiProcessorCount * numBlocksPerSm;
     checkCudaErrors(cudaMalloc(&d_partialResults, minGridSize * sizeof(double)));
 
-  printf(
+    printf("Launching normVecByDotProductAWBarrier kernel with numBlocks = %d "
-      "Launching normVecByDotProductAWBarrier kernel with numBlocks = %d "
            "blockSize = %d\n",
-      minGridSize, blockSize);
+           minGridSize,
+           blockSize);
 
     dim3 dimGrid(minGridSize, 1, 1), dimBlock(blockSize, 1, 1);
 
-  void *kernelArgs[] = {(void *)&d_vecA, (void *)&d_vecB,
+    void *kernelArgs[] = {(void *)&d_vecA, (void *)&d_vecB, (void *)&d_partialResults, (void *)&size};
-                        (void *)&d_partialResults, (void *)&size};
 
-  checkCudaErrors(
+    checkCudaErrors(cudaLaunchCooperativeKernel(
-      cudaLaunchCooperativeKernel((void *)normVecByDotProductAWBarrier, dimGrid,
+        (void *)normVecByDotProductAWBarrier, dimGrid, dimBlock, kernelArgs, smemSize, stream));
-                                  dimBlock, kernelArgs, smemSize, stream));
 
-  checkCudaErrors(cudaMemcpyAsync(vecA, d_vecA, sizeof(float) * size,
+    checkCudaErrors(cudaMemcpyAsync(vecA, d_vecA, sizeof(float) * size, cudaMemcpyDeviceToHost, stream));
-                                  cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
 
     float        expectedResult = (baseVal / sqrt(size * baseVal * baseVal));
@@ -239,7 +231,8 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId) {
         if ((vecA[i] - expectedResult) > 0.00001) {
             printf("mismatch at i = %d\n", i);
             break;
-    } else {
+        }
+        else {
             matches++;
         }
     }

Samples/0_Introduction/simpleAssert/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Removes -DNDEBUG For Print specific logs in this sample.

Samples/0_Introduction/simpleAssert/README.md

@@ -27,6 +27,6 @@ cudaDeviceSynchronize, cudaGetErrorString
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/simpleAssert/simpleAssert.cu

@@ -34,8 +34,8 @@
 #endif
 
 // Includes, system
-#include <stdio.h>
 #include <cassert>
+#include <stdio.h>
 
 // Includes CUDA
 #include <cuda_runtime.h>
@@ -58,7 +58,8 @@ bool testResult = true;
 //! Tests assert function.
 //! Thread whose id > N will print assertion failed error message.
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void testKernel(int N) {
+__global__ void testKernel(int N)
+{
     int gtid = blockIdx.x * blockDim.x + threadIdx.x;
     assert(gtid < N);
 }
@@ -70,17 +71,18 @@ void runTest(int argc, char **argv);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     int         Nblocks  = 2;
     int         Nthreads = 32;
     cudaError_t error;
@@ -94,7 +96,8 @@ void runTest(int argc, char **argv) {
     if (!strcasecmp(OS_System_Type.sysname, "Darwin")) {
         printf("simpleAssert is not current supported on Mac OSX\n\n");
         exit(EXIT_SUCCESS);
-  } else {
+    }
+    else {
         printf("OS Info: <%s>\n\n", OS_System_Type.version);
     }
 
@@ -118,8 +121,7 @@ void runTest(int argc, char **argv) {
 
     // Check for errors and failed asserts in asynchronous kernel launch.
     if (error == cudaErrorAssert) {
-    printf(
+        printf("Device assert failed as expected, "
-        "Device assert failed as expected, "
                "CUDA error message is: %s\n\n",
                cudaGetErrorString(error));
     }

Samples/0_Introduction/simpleAssert_nvrtc/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleAssert_nvrtc/README.md

@@ -30,7 +30,7 @@ cuModuleGetFunction, cuLaunchKernel, cuCtxSynchronize
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/simpleAssert_nvrtc/simpleAssert.cpp

@@ -34,11 +34,12 @@
 #endif
 
 // Includes, system
-#include <stdio.h>
 #include <cassert>
+#include <stdio.h>
 
 // Includes CUDA
 #include <cuda_runtime.h>
+
 #include "nvrtc_helper.h"
 
 // Utilities and timing functions
@@ -58,7 +59,8 @@ void runTest(int argc, char **argv);
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
@@ -66,7 +68,8 @@ int main(int argc, char **argv) {
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     int Nblocks  = 2;
     int Nthreads = 32;
 
@@ -91,10 +94,15 @@ void runTest(int argc, char **argv) {
     int   count  = 60;
     void *args[] = {(void *)&count};
 
-  checkCudaErrors(cuLaunchKernel(
+    checkCudaErrors(cuLaunchKernel(kernel_addr,
-      kernel_addr, dimGrid.x, dimGrid.y, dimGrid.z, /* grid dim */
+                                   dimGrid.x,
-      dimBlock.x, dimBlock.y, dimBlock.z,           /* block dim */
+                                   dimGrid.y,
-      0, 0,                                         /* shared mem, stream */
+                                   dimGrid.z, /* grid dim */
+                                   dimBlock.x,
+                                   dimBlock.y,
+                                   dimBlock.z, /* block dim */
+                                   0,
+                                   0,        /* shared mem, stream */
                                    &args[0], /* arguments */
                                    0));
 

Samples/0_Introduction/simpleAssert_nvrtc/simpleAssert_kernel.cu

@@ -32,7 +32,8 @@
 //! Thread whose id > N will print assertion failed error message.
 ////////////////////////////////////////////////////////////////////////////////
 
-extern "C" __global__ void testKernel(int N) {
+extern "C" __global__ void testKernel(int N)
+{
     int gtid = blockIdx.x * blockDim.x + threadIdx.x;
     assert(gtid < N);
 }

Samples/0_Introduction/simpleAtomicIntrinsics/CMakeLists.txt

@@ -11,8 +11,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 set(CMAKE_CUDA_ARCHITECTURES  50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
 
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleAtomicIntrinsics/README.md

@@ -27,6 +27,6 @@ cudaStreamCreateWithFlags, cudaFree, cudaMallocHost, cudaFreeHost, cudaStreamSyn
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/simpleAtomicIntrinsics/simpleAtomicIntrinsics.cu

@@ -30,10 +30,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
 #include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 #ifdef _WIN32
 #define WINDOWS_LEAN_AND_MEAN
@@ -68,20 +68,21 @@ extern "C" bool computeGold(int *gpuData, const int len);
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     cudaStream_t stream;
     // This will pick the best possible CUDA capable device
     findCudaDevice(argc, (const char **)argv);
@@ -100,7 +101,8 @@ void runTest(int argc, char **argv) {
     checkCudaErrors(cudaMallocHost(&hOData, memSize));
 
     // initialize the memory
-  for (unsigned int i = 0; i < numData; i++) hOData[i] = 0;
+    for (unsigned int i = 0; i < numData; i++)
+        hOData[i] = 0;
 
     // To make the AND and XOR tests generate something other than 0...
     hOData[8] = hOData[10] = 0xff;
@@ -110,15 +112,13 @@ void runTest(int argc, char **argv) {
     int *dOData;
     checkCudaErrors(cudaMalloc((void **)&dOData, memSize));
     // copy host memory to device to initialize to zero
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(dOData, hOData, memSize, cudaMemcpyHostToDevice, stream));
-      cudaMemcpyAsync(dOData, hOData, memSize, cudaMemcpyHostToDevice, stream));
 
     // execute the kernel
     testKernel<<<numBlocks, numThreads, 0, stream>>>(dOData);
 
     // Copy result from device to host
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(hOData, dOData, memSize, cudaMemcpyDeviceToHost, stream));
-      cudaMemcpyAsync(hOData, dOData, memSize, cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
 
     sdkStopTimer(&timer);

Samples/0_Introduction/simpleAtomicIntrinsics/simpleAtomicIntrinsics_cpu.cpp

@@ -42,7 +42,8 @@ extern "C" int computeGold(int *gpuData, const int len);
 //! @param idata      input data as provided to device
 //! @param len        number of elements in reference / idata
 ////////////////////////////////////////////////////////////////////////////////
-int computeGold(int *gpuData, const int len) {
+int computeGold(int *gpuData, const int len)
+{
     int val = 0;
 
     for (int i = 0; i < len; ++i) {

Samples/0_Introduction/simpleAtomicIntrinsics/simpleAtomicIntrinsics_kernel.cuh

@@ -35,7 +35,8 @@
 //! @param g_idata  input data in global memory
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void testKernel(int *g_odata) {
+__global__ void testKernel(int *g_odata)
+{
     // access thread id
     const unsigned int tid = blockDim.x * blockIdx.x + threadIdx.x;
 

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/README.md

@@ -33,7 +33,7 @@ cudaBlockSize, cudaGridSize
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/simpleAtomicIntrinsics.cpp

@@ -30,10 +30,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
 #include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 #ifdef _WIN32
 #define WINDOWS_LEAN_AND_MEAN
@@ -64,13 +64,13 @@ extern "C" bool computeGold(int *gpuData, const int len);
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("%s starting...\n", sampleName);
 
     runTest(argc, argv);
 
-  printf("%s completed, returned %s\n", sampleName,
+    printf("%s completed, returned %s\n", sampleName, testResult ? "OK" : "ERROR!");
-         testResult ? "OK" : "ERROR!");
 
     exit(testResult ? EXIT_SUCCESS : EXIT_FAILURE);
 }
@@ -79,7 +79,8 @@ int main(int argc, char **argv) {
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
 
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     int dev = 0;
 
     char  *cubin, *kernel_file;
@@ -106,7 +107,8 @@ void runTest(int argc, char **argv) {
     int *hOData = (int *)malloc(memSize);
 
     // initialize the memory
-  for (unsigned int i = 0; i < numData; i++) hOData[i] = 0;
+    for (unsigned int i = 0; i < numData; i++)
+        hOData[i] = 0;
 
     // To make the AND and XOR tests generate something other than 0...
     hOData[8] = hOData[10] = 0xff;
@@ -121,11 +123,15 @@ void runTest(int argc, char **argv) {
     dim3 cudaGridSize(numBlocks, 1, 1);
 
     void *arr[] = {(void *)&dOData};
-  checkCudaErrors(cuLaunchKernel(kernel_addr, cudaGridSize.x, cudaGridSize.y,
+    checkCudaErrors(cuLaunchKernel(kernel_addr,
+                                   cudaGridSize.x,
+                                   cudaGridSize.y,
                                    cudaGridSize.z, /* grid dim */
-                                 cudaBlockSize.x, cudaBlockSize.y,
+                                   cudaBlockSize.x,
+                                   cudaBlockSize.y,
                                    cudaBlockSize.z, /* block dim */
-                                 0, 0,            /* shared mem, stream */
+                                   0,
+                                   0,       /* shared mem, stream */
                                    &arr[0], /* arguments */
                                    0));
 

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/simpleAtomicIntrinsics_cpu.cpp

@@ -43,7 +43,8 @@ extern "C" int computeGold(int *gpuData, const int len);
 //! @param len        number of elements in reference / idata
 ////////////////////////////////////////////////////////////////////////////////
 
-int computeGold(int *gpuData, const int len) {
+int computeGold(int *gpuData, const int len)
+{
     int val = 0;
 
     for (int i = 0; i < len; ++i) {

Samples/0_Introduction/simpleAtomicIntrinsics_nvrtc/simpleAtomicIntrinsics_kernel.cuh

@@ -36,7 +36,8 @@
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
 
-extern "C" __global__ void testKernel(int *g_odata) {
+extern "C" __global__ void testKernel(int *g_odata)
+{
     // access thread id
     const unsigned int tid = blockDim.x * blockIdx.x + threadIdx.x;
 

Samples/0_Introduction/simpleAttributes/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleAttributes/README.md

@@ -27,6 +27,6 @@ cudaFree, cudaMallocHost, cudaFreeHost, cudaStreamSynchronize, cudaStreamSetAttr
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/simpleAttributes/simpleAttributes.cu

@@ -26,10 +26,10 @@
  */
 
 // includes, system
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
 #include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 // includes CUDA
 #include <cuda_runtime.h>
@@ -42,7 +42,8 @@
 // declaration, forward
 void runTest(int argc, char **argv);
 
-cudaAccessPolicyWindow initAccessPolicyWindow(void) {
+cudaAccessPolicyWindow initAccessPolicyWindow(void)
+{
     cudaAccessPolicyWindow accessPolicyWindow = {0};
     accessPolicyWindow.base_ptr               = (void *)0;
     accessPolicyWindow.num_bytes              = 0;
@@ -60,8 +61,8 @@ cudaAccessPolicyWindow initAccessPolicyWindow(void) {
 //! @param bigDataSize  input bigData size
 //! @param hitcount how many data access are done within block
 ////////////////////////////////////////////////////////////////////////////////
-static __global__ void kernCacheSegmentTest(int *data, int dataSize, int *trash,
+static __global__ void kernCacheSegmentTest(int *data, int dataSize, int *trash, int bigDataSize, int hitCount)
-                                            int bigDataSize, int hitCount) {
+{
     __shared__ unsigned int hit;
     int                     row    = blockIdx.y * blockDim.y + threadIdx.y;
     int                     col    = blockIdx.x * blockDim.x + threadIdx.x;
@@ -82,9 +83,9 @@ static __global__ void kernCacheSegmentTest(int *data, int dataSize, int *trash,
 
         if ((tID % 2) == 0) {
             data[psRand % dataSize] = data[psRand % dataSize] + data[idx % dataSize];
-    } else {
+        }
-      trash[psRand % bigDataSize] =
+        else {
-          trash[psRand % bigDataSize] + trash[idx % bigDataSize];
+            trash[psRand % bigDataSize] = trash[psRand % bigDataSize] + trash[idx % bigDataSize];
         }
 
         atomicAdd(&hit, 1);
@@ -98,7 +99,8 @@ int main(int argc, char **argv) { runTest(argc, argv); }
 ////////////////////////////////////////////////////////////////////////////////
 //! Run a simple test for CUDA
 ////////////////////////////////////////////////////////////////////////////////
-void runTest(int argc, char **argv) {
+void runTest(int argc, char **argv)
+{
     bool                   bTestResult = true;
     cudaAccessPolicyWindow accessPolicyWindow;
     cudaDeviceProp         deviceProp;
@@ -127,8 +129,7 @@ void runTest(int argc, char **argv) {
 
     // Make sure device the l2 optimization
     if (deviceProp.persistingL2CacheMaxSize == 0) {
-    printf(
+        printf("Waiving execution as device %d does not support persisting L2 "
-        "Waiving execution as device %d does not support persisting L2 "
                "Caching\n",
                devID);
         exit(EXIT_WAIVED);
@@ -139,8 +140,7 @@ void runTest(int argc, char **argv) {
 
     // Set the amount of l2 cache that will be persisting to maximum the device
     // can support
-  checkCudaErrors(cudaDeviceSetLimit(cudaLimitPersistingL2CacheSize,
+    checkCudaErrors(cudaDeviceSetLimit(cudaLimitPersistingL2CacheSize, deviceProp.persistingL2CacheMaxSize));
-                                     deviceProp.persistingL2CacheMaxSize));
 
     // Stream attribute to set
     streamAttrID = cudaStreamAttributeAccessPolicyWindow;
@@ -155,8 +155,7 @@ void runTest(int argc, char **argv) {
 
     // Allocate data
     checkCudaErrors(cudaMallocHost(&dataHostPointer, dataSize * sizeof(int)));
-  checkCudaErrors(
+    checkCudaErrors(cudaMallocHost(&bigDataHostPointer, bigDataSize * sizeof(int)));
-      cudaMallocHost(&bigDataHostPointer, bigDataSize * sizeof(int)));
 
     for (int i = 0; i < bigDataSize; ++i) {
         if (i < dataSize) {
@@ -166,16 +165,12 @@ void runTest(int argc, char **argv) {
         bigDataHostPointer[bigDataSize - i - 1] = i;
     }
 
+    checkCudaErrors(cudaMalloc((void **)&dataDevicePointer, dataSize * sizeof(int)));
+    checkCudaErrors(cudaMalloc((void **)&bigDataDevicePointer, bigDataSize * sizeof(int)));
     checkCudaErrors(
-      cudaMalloc((void **)&dataDevicePointer, dataSize * sizeof(int)));
+        cudaMemcpyAsync(dataDevicePointer, dataHostPointer, dataSize * sizeof(int), cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(
-      cudaMalloc((void **)&bigDataDevicePointer, bigDataSize * sizeof(int)));
+        bigDataDevicePointer, bigDataHostPointer, bigDataSize * sizeof(int), cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(cudaMemcpyAsync(dataDevicePointer, dataHostPointer,
-                                  dataSize * sizeof(int),
-                                  cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(cudaMemcpyAsync(bigDataDevicePointer, bigDataHostPointer,
-                                  bigDataSize * sizeof(int),
-                                  cudaMemcpyHostToDevice, stream));
 
     // Make a window for the buffer of interest
     accessPolicyWindow.base_ptr        = (void *)dataDevicePointer;
@@ -186,8 +181,7 @@ void runTest(int argc, char **argv) {
     streamAttrValue.accessPolicyWindow = accessPolicyWindow;
 
     // Assign window to stream
-  checkCudaErrors(
+    checkCudaErrors(cudaStreamSetAttribute(stream, streamAttrID, &streamAttrValue));
-      cudaStreamSetAttribute(stream, streamAttrID, &streamAttrValue));
 
     // Demote any previous persisting lines
     checkCudaErrors(cudaCtxResetPersistingL2Cache());

Samples/0_Introduction/simpleCUDA2GL/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries
@@ -59,12 +61,16 @@ if(${OpenGL_FOUND})
 
                 add_custom_command(TARGET simpleCUDA2GL
                     POST_BUILD
-                    COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/../../../bin/win64/$<CONFIGURATION>/freeglut.dll ${CMAKE_CURRENT_BINARY_DIR}
+                    COMMAND ${CMAKE_COMMAND} -E
+                    copy ${CMAKE_CURRENT_SOURCE_DIR}/../../../bin/win64/$<CONFIGURATION>/freeglut.dll
+                    ${CMAKE_CURRENT_BINARY_DIR}/$<CONFIGURATION>
                 )
 
                 add_custom_command(TARGET simpleCUDA2GL
                     POST_BUILD
-                    COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/../../../bin/win64/$<CONFIGURATION>/glew64.dll ${CMAKE_CURRENT_BINARY_DIR}
+                    COMMAND ${CMAKE_COMMAND} -E
+                    copy ${CMAKE_CURRENT_SOURCE_DIR}/../../../bin/win64/$<CONFIGURATION>/glew64.dll
+                    ${CMAKE_CURRENT_BINARY_DIR}/$<CONFIGURATION>
                 )
             endif()
         endif()

Samples/0_Introduction/simpleCUDA2GL/README.md

@@ -30,8 +30,7 @@ cudaHostAlloc, cudaGraphicsUnmapResources, cudaMalloc, cudaFree, cudaGraphicsRes
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## References (for more details)
-

Samples/0_Introduction/simpleCUDA2GL/main.cpp

@@ -50,8 +50,8 @@
 #endif
 
 // CUDA includes
-#include <cuda_runtime.h>
 #include <cuda_gl_interop.h>
+#include <cuda_runtime.h>
 
 // CUDA utilities and system includes
 #include <helper_cuda.h>
@@ -124,8 +124,7 @@ StopWatchInterface *timer = NULL;
 GLuint shDraw;
 
 ////////////////////////////////////////////////////////////////////////////////
-extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes,
+extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes, unsigned int *g_odata, int imgw);
-                                   unsigned int *g_odata, int imgw);
 
 // Forward declarations
 void runStdProgram(int argc, char **argv);
@@ -140,8 +139,7 @@ void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource);
 void deletePBO(GLuint *pbo);
 #endif
 
-void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x,
+void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x, unsigned int size_y);
-                      unsigned int size_y);
 void deleteTexture(GLuint *tex);
 
 // rendering callbacks
@@ -155,7 +153,8 @@ void mainMenu(int i);
 ////////////////////////////////////////////////////////////////////////////////
 //! Create PBO
 ////////////////////////////////////////////////////////////////////////////////
-void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource) {
+void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource)
+{
     // set up vertex data parameter
     num_texels    = image_width * image_height;
     num_values    = num_texels * 4;
@@ -171,33 +170,32 @@ void createPBO(GLuint *pbo, struct cudaGraphicsResource **pbo_resource) {
     glBindBuffer(GL_ARRAY_BUFFER, 0);
 
     // register this buffer object with CUDA
-  checkCudaErrors(cudaGraphicsGLRegisterBuffer(pbo_resource, *pbo,
+    checkCudaErrors(cudaGraphicsGLRegisterBuffer(pbo_resource, *pbo, cudaGraphicsMapFlagsNone));
-                                               cudaGraphicsMapFlagsNone));
 
     SDK_CHECK_ERROR_GL();
 }
 
-void deletePBO(GLuint *pbo) {
+void deletePBO(GLuint *pbo)
+{
     glDeleteBuffers(1, pbo);
     SDK_CHECK_ERROR_GL();
     *pbo = 0;
 }
 #endif
 
-const GLenum fbo_targets[] = {
+const GLenum fbo_targets[] = {GL_COLOR_ATTACHMENT0_EXT,
-    GL_COLOR_ATTACHMENT0_EXT, GL_COLOR_ATTACHMENT1_EXT,
+                              GL_COLOR_ATTACHMENT1_EXT,
-    GL_COLOR_ATTACHMENT2_EXT, GL_COLOR_ATTACHMENT3_EXT};
+                              GL_COLOR_ATTACHMENT2_EXT,
+                              GL_COLOR_ATTACHMENT3_EXT};
 
 #ifndef USE_TEXSUBIMAGE2D
-static const char *glsl_drawtex_vertshader_src =
+static const char *glsl_drawtex_vertshader_src = "void main(void)\n"
-    "void main(void)\n"
                                                  "{\n"
                                                  "	gl_Position = gl_Vertex;\n"
                                                  "	gl_TexCoord[0].xy = gl_MultiTexCoord0.xy;\n"
                                                  "}\n";
 
-static const char *glsl_drawtex_fragshader_src =
+static const char *glsl_drawtex_fragshader_src = "#version 130\n"
-    "#version 130\n"
                                                  "uniform usampler2D texImage;\n"
                                                  "void main()\n"
                                                  "{\n"
@@ -227,15 +225,15 @@ static const char *glsl_draw_fragshader_src =
 #endif
 
 // copy image and process using CUDA
-void generateCUDAImage() {
+void generateCUDAImage()
+{
     // run the Cuda kernel
     unsigned int *out_data;
 
 #ifdef USE_TEXSUBIMAGE2D
     checkCudaErrors(cudaGraphicsMapResources(1, &cuda_pbo_dest_resource, 0));
     size_t num_bytes;
-  checkCudaErrors(cudaGraphicsResourceGetMappedPointer(
+    checkCudaErrors(cudaGraphicsResourceGetMappedPointer((void **)&out_data, &num_bytes, cuda_pbo_dest_resource));
-      (void **)&out_data, &num_bytes, cuda_pbo_dest_resource));
 // printf("CUDA mapped pointer of pbo_out: May access %ld bytes, expected %d\n",
 // num_bytes, size_tex_data);
 #else
@@ -258,8 +256,7 @@ void generateCUDAImage() {
     glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, pbo_dest);
 
     glBindTexture(GL_TEXTURE_2D, tex_cudaResult);
-  glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, image_width, image_height, GL_RGBA,
+    glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, image_width, image_height, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
-                  GL_UNSIGNED_BYTE, NULL);
     SDK_CHECK_ERROR_GL();
     glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, 0);
     glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, 0);
@@ -268,21 +265,20 @@ void generateCUDAImage() {
     // map buffer objects to get CUDA device pointers
     cudaArray *texture_ptr;
     checkCudaErrors(cudaGraphicsMapResources(1, &cuda_tex_result_resource, 0));
-  checkCudaErrors(cudaGraphicsSubResourceGetMappedArray(
+    checkCudaErrors(cudaGraphicsSubResourceGetMappedArray(&texture_ptr, cuda_tex_result_resource, 0, 0));
-      &texture_ptr, cuda_tex_result_resource, 0, 0));
 
     int num_texels    = image_width * image_height;
     int num_values    = num_texels * 4;
     int size_tex_data = sizeof(GLubyte) * num_values;
-  checkCudaErrors(cudaMemcpyToArray(texture_ptr, 0, 0, cuda_dest_resource,
+    checkCudaErrors(cudaMemcpyToArray(texture_ptr, 0, 0, cuda_dest_resource, size_tex_data, cudaMemcpyDeviceToDevice));
-                                    size_tex_data, cudaMemcpyDeviceToDevice));
 
     checkCudaErrors(cudaGraphicsUnmapResources(1, &cuda_tex_result_resource, 0));
 #endif
 }
 
 // display image to the screen as textured quad
-void displayImage(GLuint texture) {
+void displayImage(GLuint texture)
+{
     glBindTexture(GL_TEXTURE_2D, texture);
     glEnable(GL_TEXTURE_2D);
     glDisable(GL_DEPTH_TEST);
@@ -332,7 +328,8 @@ void displayImage(GLuint texture) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Display callback
 ////////////////////////////////////////////////////////////////////////////////
-void display() {
+void display()
+{
     sdkStartTimer(&timer);
 
     if (enable_cuda) {
@@ -358,9 +355,7 @@ void display() {
             sprintf(currentOutputPPM, "kilt.ppm");
             g_CheckRender->savePPM(currentOutputPPM, true, NULL);
 
-      if (!g_CheckRender->PPMvsPPM(currentOutputPPM,
+            if (!g_CheckRender->PPMvsPPM(currentOutputPPM, sdkFindFilePath(ref_file, pArgv[0]), MAX_EPSILON, 0.30f)) {
-                                   sdkFindFilePath(ref_file, pArgv[0]),
-                                   MAX_EPSILON, 0.30f)) {
                 g_TotalErrors++;
             }
 
@@ -374,8 +369,7 @@ void display() {
     if (++fpsCount == fpsLimit) {
         char  cTitle[256];
         float fps = 1000.0f / sdkGetAverageTimerValue(&timer);
-    sprintf(cTitle, "CUDA GL Post Processing (%d x %d): %.1f fps", window_width,
+        sprintf(cTitle, "CUDA GL Post Processing (%d x %d): %.1f fps", window_width, window_height, fps);
-            window_height, fps);
         glutSetWindowTitle(cTitle);
         // printf("%s\n", cTitle);
         fpsCount = 0;
@@ -384,7 +378,8 @@ void display() {
     }
 }
 
-void timerEvent(int value) {
+void timerEvent(int value)
+{
     glutPostRedisplay();
     glutTimerFunc(REFRESH_DELAY, timerEvent, 0);
 }
@@ -392,7 +387,8 @@ void timerEvent(int value) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Keyboard events handler
 ////////////////////////////////////////////////////////////////////////////////
-void keyboard(unsigned char key, int /*x*/, int /*y*/) {
+void keyboard(unsigned char key, int /*x*/, int /*y*/)
+{
     switch (key) {
     case (27):
         Cleanup(EXIT_SUCCESS);
@@ -404,7 +400,8 @@ void keyboard(unsigned char key, int /*x*/, int /*y*/) {
 
         if (enable_cuda) {
             glClearColorIuiEXT(128, 128, 128, 255);
-      } else {
+        }
+        else {
             glClearColor(0.5, 0.5, 0.5, 1.0);
         }
 
@@ -413,7 +410,8 @@ void keyboard(unsigned char key, int /*x*/, int /*y*/) {
     }
 }
 
-void reshape(int w, int h) {
+void reshape(int w, int h)
+{
     window_width  = w;
     window_height = h;
 }
@@ -423,8 +421,8 @@ void mainMenu(int i) { keyboard((unsigned char)i, 0, 0); }
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x,
+void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x, unsigned int size_y)
-                      unsigned int size_y) {
+{
     // create a texture
     glGenTextures(1, tex_cudaResult);
     glBindTexture(GL_TEXTURE_2D, *tex_cudaResult);
@@ -436,24 +434,22 @@ void createTextureDst(GLuint *tex_cudaResult, unsigned int size_x,
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
 
 #ifdef USE_TEXSUBIMAGE2D
-  glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, size_x, size_y, 0, GL_RGBA,
+    glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, size_x, size_y, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
-               GL_UNSIGNED_BYTE, NULL);
     SDK_CHECK_ERROR_GL();
 #else
-  glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8UI_EXT, size_x, size_y, 0,
+    glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8UI_EXT, size_x, size_y, 0, GL_RGBA_INTEGER_EXT, GL_UNSIGNED_BYTE, NULL);
-               GL_RGBA_INTEGER_EXT, GL_UNSIGNED_BYTE, NULL);
     SDK_CHECK_ERROR_GL();
     // register this texture with CUDA
     checkCudaErrors(cudaGraphicsGLRegisterImage(
-      &cuda_tex_result_resource, *tex_cudaResult, GL_TEXTURE_2D,
+        &cuda_tex_result_resource, *tex_cudaResult, GL_TEXTURE_2D, cudaGraphicsMapFlagsWriteDiscard));
-      cudaGraphicsMapFlagsWriteDiscard));
 #endif
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void deleteTexture(GLuint *tex) {
+void deleteTexture(GLuint *tex)
+{
     glDeleteTextures(1, tex);
     SDK_CHECK_ERROR_GL();
 
@@ -463,7 +459,8 @@ void deleteTexture(GLuint *tex) {
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
 #if defined(__linux__)
     char *Xstatus = getenv("DISPLAY");
     if (Xstatus == NULL) {
@@ -487,8 +484,7 @@ int main(int argc, char **argv) {
     if (checkCmdLineFlag(argc, (const char **)argv, "device")) {
         printf("[%s]\n", argv[0]);
         printf("   Does not explicitly support -device=n\n");
-    printf(
+        printf("   This sample requires OpenGL.  Only -file=<reference> are "
-        "   This sample requires OpenGL.  Only -file=<reference> are "
                "supported\n");
         printf("exiting...\n");
         exit(EXIT_WAIVED);
@@ -497,7 +493,8 @@ int main(int argc, char **argv) {
     if (ref_file) {
         printf("(Test with OpenGL verification)\n");
         runStdProgram(argc, argv);
-  } else {
+    }
+    else {
         printf("(Interactive OpenGL Demo)\n");
         runStdProgram(argc, argv);
     }
@@ -508,7 +505,8 @@ int main(int argc, char **argv) {
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void FreeResource() {
+void FreeResource()
+{
     sdkDeleteTimer(&timer);
 
 // unregister this buffer object with CUDA
@@ -530,18 +528,18 @@ void FreeResource() {
     printf("simpleCUDA2GL Exiting...\n");
 }
 
-void Cleanup(int iExitCode) {
+void Cleanup(int iExitCode)
+{
     FreeResource();
-  printf("PPM Images are %s\n",
+    printf("PPM Images are %s\n", (iExitCode == EXIT_SUCCESS) ? "Matching" : "Not Matching");
-         (iExitCode == EXIT_SUCCESS) ? "Matching" : "Not Matching");
     exit(iExitCode);
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-GLuint compileGLSLprogram(const char *vertex_shader_src,
+GLuint compileGLSLprogram(const char *vertex_shader_src, const char *fragment_shader_src)
-                          const char *fragment_shader_src) {
+{
     GLuint v, f, p = 0;
 
     p = glCreateProgram();
@@ -563,7 +561,8 @@ GLuint compileGLSLprogram(const char *vertex_shader_src,
             // #endif
             glDeleteShader(v);
             return 0;
-    } else {
+        }
+        else {
             glAttachShader(p, v);
         }
     }
@@ -585,7 +584,8 @@ GLuint compileGLSLprogram(const char *vertex_shader_src,
             // #endif
             glDeleteShader(f);
             return 0;
-    } else {
+        }
+        else {
             glAttachShader(p, f);
         }
     }
@@ -611,7 +611,8 @@ GLuint compileGLSLprogram(const char *vertex_shader_src,
 //! Allocate the "render target" of CUDA
 ////////////////////////////////////////////////////////////////////////////////
 #ifndef USE_TEXSUBIMAGE2D
-void initCUDABuffers() {
+void initCUDABuffers()
+{
     // set up vertex data parameter
     num_texels    = image_width * image_height;
     num_values    = num_texels * 4;
@@ -625,7 +626,8 @@ void initCUDABuffers() {
 ////////////////////////////////////////////////////////////////////////////////
 //!
 ////////////////////////////////////////////////////////////////////////////////
-void initGLBuffers() {
+void initGLBuffers()
+{
 // create pbo
 #ifdef USE_TEXSUBIMAGE2D
     createPBO(&pbo_dest, &cuda_pbo_dest_resource);
@@ -636,8 +638,7 @@ void initGLBuffers() {
     shDraw = compileGLSLprogram(NULL, glsl_draw_fragshader_src);
 
 #ifndef USE_TEXSUBIMAGE2D
-  shDrawTex = compileGLSLprogram(glsl_drawtex_vertshader_src,
+    shDrawTex = compileGLSLprogram(glsl_drawtex_vertshader_src, glsl_drawtex_fragshader_src);
-                                 glsl_drawtex_fragshader_src);
 #endif
     SDK_CHECK_ERROR_GL();
 }
@@ -645,7 +646,8 @@ void initGLBuffers() {
 ////////////////////////////////////////////////////////////////////////////////
 //! Run standard demo loop with or without GL verification
 ////////////////////////////////////////////////////////////////////////////////
-void runStdProgram(int argc, char **argv) {
+void runStdProgram(int argc, char **argv)
+{
     // First initialize OpenGL context, so we can properly set the GL for CUDA.
     // This is necessary in order to achieve optimal performance with OpenGL/CUDA
     // interop.
@@ -683,8 +685,7 @@ void runStdProgram(int argc, char **argv) {
         g_CheckRender->EnableQAReadback(true);
     }
 
-  printf(
+    printf("\n"
-      "\n"
            "\tControls\n"
            "\t(right click mouse button for Menu)\n"
            "\t[esc] - Quit\n\n");
@@ -699,7 +700,8 @@ void runStdProgram(int argc, char **argv) {
 ////////////////////////////////////////////////////////////////////////////////
 //! Initialize GL
 ////////////////////////////////////////////////////////////////////////////////
-bool initGL(int *argc, char **argv) {
+bool initGL(int *argc, char **argv)
+{
     // Create GL context
     glutInit(argc, argv);
     glutInitDisplayMode(GLUT_RGBA | GLUT_ALPHA | GLUT_DOUBLE | GLUT_DEPTH);
@@ -707,8 +709,8 @@ bool initGL(int *argc, char **argv) {
     iGLUTWindowHandle = glutCreateWindow("CUDA OpenGL post-processing");
 
     // initialize necessary OpenGL extensions
-  if (!isGLVersionSupported(2, 0) ||
+    if (!isGLVersionSupported(2, 0)
-      !areGLExtensionsSupported("GL_ARB_pixel_buffer_object "
+        || !areGLExtensionsSupported("GL_ARB_pixel_buffer_object "
                                      "GL_EXT_framebuffer_object")) {
         printf("ERROR: Support for necessary OpenGL extensions missing.");
         fflush(stderr);
@@ -729,8 +731,7 @@ bool initGL(int *argc, char **argv) {
     // projection
     glMatrixMode(GL_PROJECTION);
     glLoadIdentity();
-  gluPerspective(60.0, (GLfloat)window_width / (GLfloat)window_height, 0.1f,
+    gluPerspective(60.0, (GLfloat)window_width / (GLfloat)window_height, 0.1f, 10.0f);
-                 10.0f);
 
     glPolygonMode(GL_FRONT_AND_BACK, GL_FILL);
 

Samples/0_Introduction/simpleCUDA2GL/simpleCUDA2GL.cu

@@ -35,14 +35,16 @@ __device__ float clamp(float x, float a, float b) { return max(a, min(b, x)); }
 __device__ int clamp(int x, int a, int b) { return max(a, min(b, x)); }
 
 // convert floating point rgb color to 8-bit integer
-__device__ int rgbToInt(float r, float g, float b) {
+__device__ int rgbToInt(float r, float g, float b)
+{
     r = clamp(r, 0.0f, 255.0f);
     g = clamp(g, 0.0f, 255.0f);
     b = clamp(b, 0.0f, 255.0f);
     return (int(b) << 16) | (int(g) << 8) | int(r);
 }
 
-__global__ void cudaProcess(unsigned int *g_odata, int imgw) {
+__global__ void cudaProcess(unsigned int *g_odata, int imgw)
+{
     extern __shared__ uchar4 sdata[];
 
     int tx = threadIdx.x;
@@ -56,7 +58,7 @@ __global__ void cudaProcess(unsigned int *g_odata, int imgw) {
     g_odata[y * imgw + x] = rgbToInt(c4.z, c4.y, c4.x);
 }
 
-extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes,
+extern "C" void launch_cudaProcess(dim3 grid, dim3 block, int sbytes, unsigned int *g_odata, int imgw)
-                                   unsigned int *g_odata, int imgw) {
+{
     cudaProcess<<<grid, block, sbytes>>>(g_odata, imgw);
 }

Samples/0_Introduction/simpleCallback/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleCallback/README.md

@@ -27,6 +27,6 @@ cudaHostAlloc, cudaStreamDestroy, cudaFree, cudaSetDevice, cudaGetDeviceCount, c
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/simpleCallback/multithreading.cpp

@@ -29,18 +29,21 @@
 
 #if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
 // Create thread
-CUTThread cutStartThread(CUT_THREADROUTINE func, void *data) {
+CUTThread cutStartThread(CUT_THREADROUTINE func, void *data)
+{
     return CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)func, data, 0, NULL);
 }
 
 // Wait for thread to finish
-void cutEndThread(CUTThread thread) {
+void cutEndThread(CUTThread thread)
+{
     WaitForSingleObject(thread, INFINITE);
     CloseHandle(thread);
 }
 
 // Wait for multiple threads
-void cutWaitForThreads(const CUTThread *threads, int num) {
+void cutWaitForThreads(const CUTThread *threads, int num)
+{
     WaitForMultipleObjects(num, threads, true, INFINITE);
 
     for (int i = 0; i < num; i++) {
@@ -49,7 +52,8 @@ void cutWaitForThreads(const CUTThread *threads, int num) {
 }
 
 // Create barrier.
-CUTBarrier cutCreateBarrier(int releaseCount) {
+CUTBarrier cutCreateBarrier(int releaseCount)
+{
     CUTBarrier barrier;
 
     InitializeCriticalSection(&barrier.criticalSection);
@@ -61,7 +65,8 @@ CUTBarrier cutCreateBarrier(int releaseCount) {
 }
 
 // Increment barrier. (execution continues)
-void cutIncrementBarrier(CUTBarrier *barrier) {
+void cutIncrementBarrier(CUTBarrier *barrier)
+{
     int myBarrierCount;
     EnterCriticalSection(&barrier->criticalSection);
     myBarrierCount = ++barrier->count;
@@ -73,16 +78,15 @@ void cutIncrementBarrier(CUTBarrier *barrier) {
 }
 
 // Wait for barrier release.
-void cutWaitForBarrier(CUTBarrier *barrier) {
+void cutWaitForBarrier(CUTBarrier *barrier) { WaitForSingleObject(barrier->barrierEvent, INFINITE); }
-  WaitForSingleObject(barrier->barrierEvent, INFINITE);
-}
 
 // Destroy barrier
 void cutDestroyBarrier(CUTBarrier *barrier) {}
 
 #else
 // Create thread
-CUTThread cutStartThread(CUT_THREADROUTINE func, void *data) {
+CUTThread cutStartThread(CUT_THREADROUTINE func, void *data)
+{
     pthread_t thread;
     pthread_create(&thread, NULL, func, data);
     return thread;
@@ -92,14 +96,16 @@ CUTThread cutStartThread(CUT_THREADROUTINE func, void *data) {
 void cutEndThread(CUTThread thread) { pthread_join(thread, NULL); }
 
 // Wait for multiple threads
-void cutWaitForThreads(const CUTThread *threads, int num) {
+void cutWaitForThreads(const CUTThread *threads, int num)
+{
     for (int i = 0; i < num; i++) {
         cutEndThread(threads[i]);
     }
 }
 
 // Create barrier.
-CUTBarrier cutCreateBarrier(int releaseCount) {
+CUTBarrier cutCreateBarrier(int releaseCount)
+{
     CUTBarrier barrier;
 
     barrier.count        = 0;
@@ -112,7 +118,8 @@ CUTBarrier cutCreateBarrier(int releaseCount) {
 }
 
 // Increment barrier. (execution continues)
-void cutIncrementBarrier(CUTBarrier *barrier) {
+void cutIncrementBarrier(CUTBarrier *barrier)
+{
     int myBarrierCount;
     pthread_mutex_lock(&barrier->mutex);
     myBarrierCount = ++barrier->count;
@@ -124,7 +131,8 @@ void cutIncrementBarrier(CUTBarrier *barrier) {
 }
 
 // Wait for barrier release.
-void cutWaitForBarrier(CUTBarrier *barrier) {
+void cutWaitForBarrier(CUTBarrier *barrier)
+{
     pthread_mutex_lock(&barrier->mutex);
 
     while (barrier->count < barrier->releaseCount) {
@@ -135,7 +143,8 @@ void cutWaitForBarrier(CUTBarrier *barrier) {
 }
 
 // Destroy barrier
-void cutDestroyBarrier(CUTBarrier *barrier) {
+void cutDestroyBarrier(CUTBarrier *barrier)
+{
     pthread_mutex_destroy(&barrier->mutex);
     pthread_cond_destroy(&barrier->conditionVariable);
 }

Samples/0_Introduction/simpleCallback/multithreading.h

@@ -37,7 +37,8 @@
 typedef HANDLE CUTThread;
 typedef unsigned(WINAPI *CUT_THREADROUTINE)(void *);
 
-struct CUTBarrier {
+struct CUTBarrier
+{
     CRITICAL_SECTION criticalSection;
     HANDLE           barrierEvent;
     int              releaseCount;
@@ -57,7 +58,8 @@ typedef void *(*CUT_THREADROUTINE)(void *);
 #define CUT_THREADPROC void *
 #define CUT_THREADEND  return 0
 
-struct CUTBarrier {
+struct CUTBarrier
+{
     pthread_mutex_t mutex;
     pthread_cond_t  conditionVariable;
     int             releaseCount;
@@ -67,7 +69,8 @@ struct CUTBarrier {
 #endif
 
 #ifdef __cplusplus
-extern "C" {
+extern "C"
+{
 #endif
 
     // Create thread.

Samples/0_Introduction/simpleCallback/simpleCallback.cu

@@ -43,8 +43,8 @@
 #include <stdio.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 #include "multithreading.h"
 
@@ -53,10 +53,10 @@ const int N_elements_per_workload = 100000;
 
 CUTBarrier thread_barrier;
 
-void CUDART_CB myStreamCallback(cudaStream_t event, cudaError_t status,
+void CUDART_CB myStreamCallback(cudaStream_t event, cudaError_t status, void *data);
-                                void *data);
 
-struct heterogeneous_workload {
+struct heterogeneous_workload
+{
     int id;
     int cudaDeviceID;
 
@@ -67,13 +67,16 @@ struct heterogeneous_workload {
     bool success;
 };
 
-__global__ void incKernel(int *data, int N) {
+__global__ void incKernel(int *data, int N)
+{
     int i = blockIdx.x * blockDim.x + threadIdx.x;
 
-  if (i < N) data[i]++;
+    if (i < N)
+        data[i]++;
 }
 
-CUT_THREADPROC launch(void *void_arg) {
+CUT_THREADPROC launch(void *void_arg)
+{
     heterogeneous_workload *workload = (heterogeneous_workload *)void_arg;
 
     // Select GPU for this CPU thread
@@ -81,11 +84,8 @@ CUT_THREADPROC launch(void *void_arg) {
 
     // Allocate Resources
     checkCudaErrors(cudaStreamCreate(&workload->stream));
-  checkCudaErrors(
+    checkCudaErrors(cudaMalloc(&workload->d_data, N_elements_per_workload * sizeof(int)));
-      cudaMalloc(&workload->d_data, N_elements_per_workload * sizeof(int)));
+    checkCudaErrors(cudaHostAlloc(&workload->h_data, N_elements_per_workload * sizeof(int), cudaHostAllocPortable));
-  checkCudaErrors(cudaHostAlloc(&workload->h_data,
-                                N_elements_per_workload * sizeof(int),
-                                cudaHostAllocPortable));
 
     // CPU thread generates data
     for (int i = 0; i < N_elements_per_workload; ++i) {
@@ -97,25 +97,28 @@ CUT_THREADPROC launch(void *void_arg) {
     dim3 block(512);
     dim3 grid((N_elements_per_workload + block.x - 1) / block.x);
 
-  checkCudaErrors(cudaMemcpyAsync(workload->d_data, workload->h_data,
+    checkCudaErrors(cudaMemcpyAsync(workload->d_data,
+                                    workload->h_data,
                                     N_elements_per_workload * sizeof(int),
-                                  cudaMemcpyHostToDevice, workload->stream));
+                                    cudaMemcpyHostToDevice,
-  incKernel<<<grid, block, 0, workload->stream>>>(workload->d_data,
+                                    workload->stream));
-                                                  N_elements_per_workload);
+    incKernel<<<grid, block, 0, workload->stream>>>(workload->d_data, N_elements_per_workload);
-  checkCudaErrors(cudaMemcpyAsync(workload->h_data, workload->d_data,
+    checkCudaErrors(cudaMemcpyAsync(workload->h_data,
+                                    workload->d_data,
                                     N_elements_per_workload * sizeof(int),
-                                  cudaMemcpyDeviceToHost, workload->stream));
+                                    cudaMemcpyDeviceToHost,
+                                    workload->stream));
 
     // New in CUDA 5.0: Add a CPU callback which is called once all currently
     // pending operations in the CUDA stream have finished
-  checkCudaErrors(
+    checkCudaErrors(cudaStreamAddCallback(workload->stream, myStreamCallback, workload, 0));
-      cudaStreamAddCallback(workload->stream, myStreamCallback, workload, 0));
 
     CUT_THREADEND;
     // CPU thread end of life, GPU continues to process data...
 }
 
-CUT_THREADPROC postprocess(void *void_arg) {
+CUT_THREADPROC postprocess(void *void_arg)
+{
     heterogeneous_workload *workload = (heterogeneous_workload *)void_arg;
     // ... GPU is done with processing, continue on new CPU thread...
 
@@ -140,8 +143,8 @@ CUT_THREADPROC postprocess(void *void_arg) {
     CUT_THREADEND;
 }
 
-void CUDART_CB myStreamCallback(cudaStream_t stream, cudaError_t status,
+void CUDART_CB myStreamCallback(cudaStream_t stream, cudaError_t status, void *data)
-                                void *data) {
+{
     // Check status of GPU after stream operations are done
     checkCudaErrors(status);
 
@@ -149,7 +152,8 @@ void CUDART_CB myStreamCallback(cudaStream_t stream, cudaError_t status,
     cutStartThread(postprocess, data);
 }
 
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     int N_gpus, max_gpus = 0;
     int gpuInfo[32]; // assume a maximum of 32 GPUs in a system configuration
 
@@ -168,10 +172,8 @@ int main(int argc, char **argv) {
         cudaSetDevice(devid);
         cudaGetDeviceProperties(&deviceProp, devid);
         SMversion = deviceProp.major << 4 + deviceProp.minor;
-    printf("GPU[%d] %s supports SM %d.%d", devid, deviceProp.name,
+        printf("GPU[%d] %s supports SM %d.%d", devid, deviceProp.name, deviceProp.major, deviceProp.minor);
-           deviceProp.major, deviceProp.minor);
+        printf(", %s GPU Callback Functions\n", (SMversion >= 0x11) ? "capable" : "NOT capable");
-    printf(", %s GPU Callback Functions\n",
-           (SMversion >= 0x11) ? "capable" : "NOT capable");
 
         if (SMversion >= 0x11) {
             gpuInfo[max_gpus++] = devid;
@@ -181,8 +183,7 @@ int main(int argc, char **argv) {
     printf("%d GPUs available to run Callback Functions\n", max_gpus);
 
     heterogeneous_workload *workloads;
-  workloads = (heterogeneous_workload *)malloc(N_workloads *
+    workloads = (heterogeneous_workload *)malloc(N_workloads * sizeof(heterogeneous_workload));
-                                               sizeof(heterogeneous_workload));
     ;
     thread_barrier = cutCreateBarrier(N_workloads);
 

Samples/0_Introduction/simpleCooperativeGroups/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleCooperativeGroups/README.md

@@ -27,6 +27,6 @@ cudaDeviceSynchronize, cudaGetErrorString
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/simpleCooperativeGroups/simpleCooperativeGroups.cu

@@ -38,8 +38,8 @@
  *
  */
 
-#include <stdio.h>
 #include <cooperative_groups.h>
+#include <stdio.h>
 
 using namespace cooperative_groups;
 
@@ -49,7 +49,8 @@ using namespace cooperative_groups;
  * calculates the sum of val across the group g. The workspace array, x,
  * must be large enough to contain g.size() integers.
  */
-__device__ int sumReduction(thread_group g, int *x, int val) {
+__device__ int sumReduction(thread_group g, int *x, int val)
+{
     // rank of this thread in the group
     int lane = g.thread_rank();
 
@@ -85,7 +86,8 @@ __device__ int sumReduction(thread_group g, int *x, int val) {
  *
  * Creates cooperative groups and performs reductions
  */
-__global__ void cgkernel() {
+__global__ void cgkernel()
+{
     // threadBlockGroup includes all threads in the block
     thread_block threadBlockGroup     = this_thread_block();
     int          threadBlockGroupSize = threadBlockGroup.size();
@@ -107,24 +109,22 @@ __global__ void cgkernel() {
 
     // master thread in group prints out result
     if (threadBlockGroup.thread_rank() == 0) {
-    printf(
+        printf(" Sum of all ranks 0..%d in threadBlockGroup is %d (expected %d)\n\n",
-        " Sum of all ranks 0..%d in threadBlockGroup is %d (expected %d)\n\n",
+               (int)threadBlockGroup.size() - 1,
-        (int)threadBlockGroup.size() - 1, output, expectedOutput);
+               output,
+               expectedOutput);
 
-    printf(" Now creating %d groups, each of size 16 threads:\n\n",
+        printf(" Now creating %d groups, each of size 16 threads:\n\n", (int)threadBlockGroup.size() / 16);
-           (int)threadBlockGroup.size() / 16);
     }
 
     threadBlockGroup.sync();
 
     // each tiledPartition16 group includes 16 threads
-  thread_block_tile<16> tiledPartition16 =
+    thread_block_tile<16> tiledPartition16 = tiled_partition<16>(threadBlockGroup);
-      tiled_partition<16>(threadBlockGroup);
 
     // This offset allows each group to have its own unique area in the workspace
     // array
-  int workspaceOffset =
+    int workspaceOffset = threadBlockGroup.thread_rank() - tiledPartition16.thread_rank();
-      threadBlockGroup.thread_rank() - tiledPartition16.thread_rank();
 
     // input to reduction, for each thread, is its' rank in the group
     input = tiledPartition16.thread_rank();
@@ -138,10 +138,10 @@ __global__ void cgkernel() {
 
     // each master thread prints out result
     if (tiledPartition16.thread_rank() == 0)
-    printf(
+        printf("   Sum of all ranks 0..15 in this tiledPartition16 group is %d "
-        "   Sum of all ranks 0..15 in this tiledPartition16 group is %d "
                "(expected %d)\n",
-        output, expectedOutput);
+               output,
+               expectedOutput);
 
     return;
 }
@@ -149,7 +149,8 @@ __global__ void cgkernel() {
 /**
  * Host main routine
  */
-int main() {
+int main()
+{
     // Error code to check return values for CUDA calls
     cudaError_t err;
 
@@ -166,8 +167,7 @@ int main() {
     err = cudaDeviceSynchronize();
 
     if (err != cudaSuccess) {
-    fprintf(stderr, "Failed to launch kernel (error code %s)!\n",
+        fprintf(stderr, "Failed to launch kernel (error code %s)!\n", cudaGetErrorString(err));
-            cudaGetErrorString(err));
         exit(EXIT_FAILURE);
     }
 

Samples/0_Introduction/simpleCubemapTexture/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries

Samples/0_Introduction/simpleCubemapTexture/README.md

@@ -27,6 +27,6 @@ cudaMemcpy, cudaCreateChannelDesc, cudaFreeArray, cudaFree, cudaPitchedPtr, cuda
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/simpleCubemapTexture/simpleCubemapTexture.cu

@@ -36,17 +36,17 @@
  */
 
 // includes, system
-#include <stdlib.h>
-#include <stdio.h>
-#include <string.h>
 #include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
 
 // includes CUDA
 #include <cuda_runtime.h>
 
 // helper functions and utilities to work with CUDA
-#include <helper_functions.h>
 #include <helper_cuda.h>
+#include <helper_functions.h>
 
 static const char *sSDKname = "simpleCubemapTexture";
 
@@ -56,8 +56,8 @@ static const char *sSDKname = "simpleCubemapTexture";
 //! Transform a cubemap face of a linear buffe using cubemap texture lookups
 //! @param g_odata  output data in global memory
 ////////////////////////////////////////////////////////////////////////////////
-__global__ void transformKernel(float *g_odata, int width,
+__global__ void transformKernel(float *g_odata, int width, cudaTextureObject_t tex)
-                                cudaTextureObject_t tex) {
+{
     // calculate this thread's data point
     unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
     unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
@@ -110,15 +110,15 @@ __global__ void transformKernel(float *g_odata, int width,
         }
 
         // read from texture, do expected transformation and write to global memory
-    g_odata[face * width * width + y * width + x] =
+        g_odata[face * width * width + y * width + x] = -texCubemap<float>(tex, cx, cy, cz);
-        -texCubemap<float>(tex, cx, cy, cz);
     }
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Program main
 ////////////////////////////////////////////////////////////////////////////////
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     // use command-line specified CUDA device, otherwise use device with highest
     // Gflops/s
     int devID = findCudaDevice(argc, (const char **)argv);
@@ -129,13 +129,11 @@ int main(int argc, char **argv) {
     cudaDeviceProp deviceProps;
 
     checkCudaErrors(cudaGetDeviceProperties(&deviceProps, devID));
-  printf("CUDA device [%s] has %d Multi-Processors ", deviceProps.name,
+    printf("CUDA device [%s] has %d Multi-Processors ", deviceProps.name, deviceProps.multiProcessorCount);
-         deviceProps.multiProcessorCount);
     printf("SM %d.%d\n", deviceProps.major, deviceProps.minor);
 
     if (deviceProps.major < 2) {
-    printf(
+        printf("%s requires SM 2.0 or higher for support of Texture Arrays.  Test "
-        "%s requires SM 2.0 or higher for support of Texture Arrays.  Test "
                "will exit... \n",
                sSDKname);
 
@@ -157,8 +155,7 @@ int main(int argc, char **argv) {
 
     for (unsigned int layer = 0; layer < num_layers; layer++) {
         for (int i = 0; i < (int)(cubemap_size); i++) {
-      h_data_ref[layer * cubemap_size + i] =
+            h_data_ref[layer * cubemap_size + i] = -h_data[layer * cubemap_size + i] + layer;
-          -h_data[layer * cubemap_size + i] + layer;
         }
     }
 
@@ -167,19 +164,16 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaMalloc((void **)&d_data, size));
 
     // allocate array and copy image data
-  cudaChannelFormatDesc channelDesc =
+    cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
-      cudaCreateChannelDesc(32, 0, 0, 0, cudaChannelFormatKindFloat);
     cudaArray            *cu_3darray;
     //    checkCudaErrors(cudaMalloc3DArray( &cu_3darray, &channelDesc,
     //    make_cudaExtent(width, height, num_layers), cudaArrayLayered ));
-  checkCudaErrors(cudaMalloc3DArray(&cu_3darray, &channelDesc,
+    checkCudaErrors(
-                                    make_cudaExtent(width, width, num_faces),
+        cudaMalloc3DArray(&cu_3darray, &channelDesc, make_cudaExtent(width, width, num_faces), cudaArrayCubemap));
-                                    cudaArrayCubemap));
     cudaMemcpy3DParms myparms = {0};
     myparms.srcPos            = make_cudaPos(0, 0, 0);
     myparms.dstPos            = make_cudaPos(0, 0, 0);
-  myparms.srcPtr =
+    myparms.srcPtr            = make_cudaPitchedPtr(h_data, width * sizeof(float), width, width);
-      make_cudaPitchedPtr(h_data, width * sizeof(float), width, width);
     myparms.dstArray          = cu_3darray;
     myparms.extent            = make_cudaExtent(width, width, num_faces);
     myparms.kind              = cudaMemcpyHostToDevice;
@@ -207,10 +201,12 @@ int main(int argc, char **argv) {
     dim3 dimBlock(8, 8, 1);
     dim3 dimGrid(width / dimBlock.x, width / dimBlock.y, 1);
 
-  printf(
+    printf("Covering Cubemap data array of %d~3 x %d: Grid size is %d x %d, each "
-      "Covering Cubemap data array of %d~3 x %d: Grid size is %d x %d, each "
            "block has 8 x 8 threads\n",
-      width, num_layers, dimGrid.x, dimGrid.y);
+           width,
+           num_layers,
+           dimGrid.x,
+           dimGrid.y);
 
     transformKernel<<<dimGrid, dimBlock>>>(d_data, width,
                                            tex); // warmup (for better timing)
@@ -233,8 +229,7 @@ int main(int argc, char **argv) {
     checkCudaErrors(cudaDeviceSynchronize());
     sdkStopTimer(&timer);
     printf("Processing time: %.3f msec\n", sdkGetTimerValue(&timer));
-  printf("%.2f Mtexlookups/sec\n",
+    printf("%.2f Mtexlookups/sec\n", (cubemap_size / (sdkGetTimerValue(&timer) / 1000.0f) / 1e6));
-         (cubemap_size / (sdkGetTimerValue(&timer) / 1000.0f) / 1e6));
     sdkDeleteTimer(&timer);
 
     // allocate mem for the result on host side
@@ -245,14 +240,13 @@ int main(int argc, char **argv) {
     // write regression file if necessary
     if (checkCmdLineFlag(argc, (const char **)argv, "regression")) {
         // write file for regression test
-    sdkWriteFile<float>("./data/regression.dat", h_odata, width * width, 0.0f,
+        sdkWriteFile<float>("./data/regression.dat", h_odata, width * width, 0.0f, false);
-                        false);
+    }
-  } else {
+    else {
         printf("Comparing kernel output to expected data\n");
 
 #define MIN_EPSILON_ERROR 5e-3f
-    bResult =
+        bResult = compareData(h_odata, h_data_ref, cubemap_size, MIN_EPSILON_ERROR, 0.0f);
-        compareData(h_odata, h_data_ref, cubemap_size, MIN_EPSILON_ERROR, 0.0f);
     }
 
     // cleanup memory

Samples/0_Introduction/simpleDrvRuntime/CMakeLists.txt

@@ -10,8 +10,10 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 set(CMAKE_CUDA_ARCHITECTURES 50 52 60 61 70 72 75 80 86 87 89 90 100 101 120)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
-if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+if(ENABLE_CUDA_DEBUG)
-    # set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")  # enable cuda-gdb (expensive)
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G")        # enable cuda-gdb (may significantly affect performance on some targets)
+else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
 endif()
 
 # Include directories and libraries
@@ -40,6 +42,12 @@ target_link_libraries(simpleDrvRuntime PUBLIC
 set(CUDA_FATBIN_FILE "${CMAKE_CURRENT_BINARY_DIR}/vectorAdd_kernel64.fatbin")
 set(CUDA_KERNEL_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/vectorAdd_kernel.cu")
 
+# Construct GENCODE_FLAGS explicitly from CUDA architectures
+set(GENCODE_FLAGS "")
+foreach(arch ${CMAKE_CUDA_ARCHITECTURES})
+    list(APPEND GENCODE_FLAGS "-gencode=arch=compute_${arch},code=sm_${arch}")
+endforeach()
+
 add_custom_command(
     OUTPUT ${CUDA_FATBIN_FILE}
     COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets  ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}

Samples/0_Introduction/simpleDrvRuntime/README.md

@@ -30,6 +30,6 @@ cudaStreamCreateWithFlags, cudaFree, cudaMallocHost, cudaFreeHost, cudaStreamSyn
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 
 ## References (for more details)

Samples/0_Introduction/simpleDrvRuntime/simpleDrvRuntime.cpp

@@ -33,12 +33,12 @@
  */
 
 // Includes
+#include <cstring>
 #include <cuda.h>
 #include <cuda_runtime.h>
+#include <iostream>
 #include <stdio.h>
 #include <string.h>
-#include <cstring>
-#include <iostream>
 
 // includes, project
 #include <helper_cuda.h>
@@ -66,11 +66,10 @@ int CleanupNoFailure(CUcontext &cuContext);
 void RandomInit(float *, int);
 bool findModulePath(const char *, string &, char **, ostringstream &);
 
-static void check(CUresult result, char const *const func,
+static void check(CUresult result, char const *const func, const char *const file, int const line)
-                  const char *const file, int const line) {
+{
     if (result) {
-    fprintf(stderr, "CUDA error at %s:%d code=%d \"%s\" \n", file, line,
+        fprintf(stderr, "CUDA error at %s:%d code=%d \"%s\" \n", file, line, static_cast<unsigned int>(result), func);
-            static_cast<unsigned int>(result), func);
         exit(EXIT_FAILURE);
     }
 }
@@ -78,7 +77,8 @@ static void check(CUresult result, char const *const func,
 #define checkCudaDrvErrors(val) check((val), #val, __FILE__, __LINE__)
 
 // Host code
-int main(int argc, char **argv) {
+int main(int argc, char **argv)
+{
     printf("simpleDrvRuntime..\n");
     int        N = 50000, devID = 0;
     size_t     size = N * sizeof(float);
@@ -100,7 +100,8 @@ int main(int argc, char **argv) {
 
     if (!findModulePath(FATBIN_FILE, module_path, argv, fatbin)) {
         exit(EXIT_FAILURE);
-  } else {
+    }
+    else {
         printf("> initCUDA loading module: <%s>\n", module_path.c_str());
     }
 
@@ -113,8 +114,7 @@ int main(int argc, char **argv) {
     checkCudaDrvErrors(cuModuleLoadData(&cuModule, fatbin.str().c_str()));
 
     // Get function handle from module
-  checkCudaDrvErrors(
+    checkCudaDrvErrors(cuModuleGetFunction(&vecAdd_kernel, cuModule, "VecAdd_kernel"));
-      cuModuleGetFunction(&vecAdd_kernel, cuModule, "VecAdd_kernel"));
 
     // Allocate input vectors h_A and h_B in host memory
     checkCudaErrors(cudaMallocHost(&h_A, size));
@@ -133,10 +133,8 @@ int main(int argc, char **argv) {
     cudaStream_t stream;
     checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
     // Copy vectors from host memory to device memory
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(d_A, h_A, size, cudaMemcpyHostToDevice, stream));
-      cudaMemcpyAsync(d_A, h_A, size, cudaMemcpyHostToDevice, stream));
+    checkCudaErrors(cudaMemcpyAsync(d_B, h_B, size, cudaMemcpyHostToDevice, stream));
-  checkCudaErrors(
-      cudaMemcpyAsync(d_B, h_B, size, cudaMemcpyHostToDevice, stream));
 
     int threadsPerBlock = 256;
     int blocksPerGrid   = (N + threadsPerBlock - 1) / threadsPerBlock;
@@ -144,14 +142,12 @@ int main(int argc, char **argv) {
     void *args[] = {&d_A, &d_B, &d_C, &N};
 
     // Launch the CUDA kernel
-  checkCudaDrvErrors(cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1,
+    checkCudaDrvErrors(
-                                    threadsPerBlock, 1, 1, 0, stream, args,
+        cuLaunchKernel(vecAdd_kernel, blocksPerGrid, 1, 1, threadsPerBlock, 1, 1, 0, stream, args, NULL));
-                                    NULL));
 
     // Copy result from device memory to host memory
     // h_C contains the result in host memory
-  checkCudaErrors(
+    checkCudaErrors(cudaMemcpyAsync(h_C, d_C, size, cudaMemcpyDeviceToHost, stream));
-      cudaMemcpyAsync(h_C, d_C, size, cudaMemcpyDeviceToHost, stream));
     checkCudaErrors(cudaStreamSynchronize(stream));
     // Verify result
     int i;
@@ -171,7 +167,8 @@ int main(int argc, char **argv) {
     exit((i == N) ? EXIT_SUCCESS : EXIT_FAILURE);
 }
 
-int CleanupNoFailure(CUcontext &cuContext) {
+int CleanupNoFailure(CUcontext &cuContext)
+{
     // Free device memory
     checkCudaErrors(cudaFree(d_A));
     checkCudaErrors(cudaFree(d_B));
@@ -195,19 +192,21 @@ int CleanupNoFailure(CUcontext &cuContext) {
     return EXIT_SUCCESS;
 }
 // Allocates an array with random float entries.
-void RandomInit(float *data, int n) {
+void RandomInit(float *data, int n)
+{
     for (int i = 0; i < n; ++i) {
         data[i] = rand() / (float)RAND_MAX;
     }
 }
 
-bool inline findModulePath(const char *module_file, string &module_path,
+bool inline findModulePath(const char *module_file, string &module_path, char **argv, ostringstream &ostrm)
-                           char **argv, ostringstream &ostrm) {
+{
     char *actual_path = sdkFindFilePath(module_file, argv[0]);
 
     if (actual_path) {
         module_path = actual_path;
-  } else {
+    }
+    else {
         printf("> findModulePath file not found: <%s> \n", module_file);
         return false;
     }
@@ -215,7 +214,8 @@ bool inline findModulePath(const char *module_file, string &module_path,
     if (module_path.empty()) {
         printf("> findModulePath could not find file: <%s> \n", module_file);
         return false;
-  } else {
+    }
+    else {
         printf("> findModulePath found file at <%s>\n", module_path.c_str());
         if (module_path.rfind("fatbin") != string::npos) {
             ifstream fileIn(module_path.c_str(), ios::binary);