Add and update samples with CUDA 11.2 support

Samples/simpleAWBarrier/Makefile

@@ -301,11 +301,9 @@ ifeq ($(TARGET_OS),linux)
     GCCVERSION += 00
 # Remove spaces from the version number
     GCCVERSION := $(subst $(space),$(empty),$(GCCVERSION))
-# Crop the version number to 3 decimals.
-    GCCVERSION := $(shell expr `echo $(GCCVERSION)` | cut -b1-3)
 #$(warning $(GCCVERSION))
 
-    IS_MIN_VERSION := $(shell expr `echo $(GCCVERSION)` \>= 500)
+    IS_MIN_VERSION := $(shell expr `echo $(GCCVERSION)` \>= 50000)
 
     ifeq ($(IS_MIN_VERSION), 1)
         $(info >>> GCC Version is greater or equal to  5.0.0 <<<)
@@ -338,7 +336,7 @@ GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
 endif
 endif
 
-ALL_CCFLAGS += --std=c++11
+ALL_CCFLAGS += --std=c++11 --threads 0
 
 ifeq ($(SAMPLE_ENABLED),0)
 EXEC ?= @echo "[@]"

Samples/simpleAWBarrier/README.md

@@ -30,7 +30,7 @@ cudaMalloc, cudaFree, cudaMemcpyAsync
 
 ## Prerequisites
 
-Download and install the [CUDA Toolkit 11.1](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
+Download and install the [CUDA Toolkit 11.2](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
 Make sure the dependencies mentioned in [Dependencies]() section above are installed.
 
 ## Build and Run

Samples/simpleAWBarrier/simpleAWBarrier.cu

@@ -31,7 +31,7 @@
 
 // Includes CUDA
 #include <cuda_runtime.h>
-#include <cuda_awbarrier.h>
+#include <cuda/barrier>
 #include <cooperative_groups.h>
 
 // Utilities and timing functions
@@ -40,12 +40,11 @@
 // CUDA helper functions
 #include <helper_cuda.h>         // helper functions for CUDA error check
 
-namespace nvcuda_namespace = nvcuda::experimental;
 namespace cg = cooperative_groups;
 
 
 #if __CUDA_ARCH__ >= 700
-template <bool writeSquareRoot> __device__ void reduceBlockData(nvcuda_namespace::awbarrier &barrier,
+template <bool writeSquareRoot> __device__ void reduceBlockData(cuda::barrier<cuda::thread_scope_block> &barrier,
                             cg::thread_block_tile<32> &tile32, double &threadSum, double *result)
 {
     extern __shared__ double tmp[];
@@ -60,11 +59,12 @@ template <bool writeSquareRoot> __device__ void reduceBlockData(nvcuda_namespace
         tmp[tile32.meta_group_rank()] = threadSum;
     }
 
-    const auto token = barrier.arrive();
+    auto token = barrier.arrive();
 
-    // The warp which would arrive last at the barrier will 
-    // perform last round of reduction
-    if (tile32.any(token.pending_count() == 1)) {
+    barrier.wait(std::move(token));
+
+    // The warp 0 will perform last round of reduction
+    if (tile32.meta_group_rank() == 0) {
 
         double beta  = tile32.thread_rank() < tile32.meta_group_size() ? tmp[tile32.thread_rank()] : 0.0;
 
@@ -82,22 +82,21 @@ template <bool writeSquareRoot> __device__ void reduceBlockData(nvcuda_namespace
                 *result = beta;
         }
     }
-
-    barrier.wait(token);
 }
 #endif
 
 __global__ void normVecByDotProductAWBarrier(float *vecA, float *vecB, double *partialResults, int size)
 {
 #if __CUDA_ARCH__ >= 700
+#pragma diag_suppress static_var_with_dynamic_init
     cg::thread_block cta = cg::this_thread_block();
     cg::grid_group grid = cg::this_grid();;
     cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
 
-    __shared__ nvcuda_namespace::awbarrier barrier;
+    __shared__ cuda::barrier<cuda::thread_scope_block> barrier;
 
     if (threadIdx.x == 0) {
-        nvcuda_namespace::init(&barrier, blockDim.x);
+        init(&barrier, blockDim.x);
     }
 
     cg::sync(cta);
@@ -253,6 +252,7 @@ int runNormVecByDotProductAWBarrier(int argc, char **argv, int deviceId)
         }
     }
 
+    printf("Result = %s\n", matches == size ? "PASSED" : "FAILED");
     checkCudaErrors(cudaFree(d_vecA));
     checkCudaErrors(cudaFree(d_vecB));
     checkCudaErrors(cudaFree(d_partialResults));

Samples/simpleAWBarrier/simpleAWBarrier_vs2015.sln

@@ -1,20 +0,0 @@
-
-Microsoft Visual Studio Solution File, Format Version 14.00
-# Visual Studio 2015
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "simpleAWBarrier", "simpleAWBarrier_vs2015.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
-EndProject
-Global
-	GlobalSection(SolutionConfigurationPlatforms) = preSolution
-		Debug|x64 = Debug|x64
-		Release|x64 = Release|x64
-	EndGlobalSection
-	GlobalSection(ProjectConfigurationPlatforms) = postSolution
-		{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
-		{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
-		{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
-		{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
-	EndGlobalSection
-	GlobalSection(SolutionProperties) = preSolution
-		HideSolutionNode = FALSE
-	EndGlobalSection
-EndGlobal

Samples/simpleAWBarrier/simpleAWBarrier_vs2015.vcxproj

@@ -1,107 +0,0 @@
-<?xml version="1.0" encoding="utf-8"?>
-<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
-  <PropertyGroup>
-    <CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
-  </PropertyGroup>
-  <ItemGroup Label="ProjectConfigurations">
-    <ProjectConfiguration Include="Debug|x64">
-      <Configuration>Debug</Configuration>
-      <Platform>x64</Platform>
-    </ProjectConfiguration>
-    <ProjectConfiguration Include="Release|x64">
-      <Configuration>Release</Configuration>
-      <Platform>x64</Platform>
-    </ProjectConfiguration>
-  </ItemGroup>
-  <PropertyGroup Label="Globals">
-    <ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
-    <RootNamespace>simpleAWBarrier_vs2015</RootNamespace>
-    <ProjectName>simpleAWBarrier</ProjectName>
-    <CudaToolkitCustomDir />
-  </PropertyGroup>
-  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
-  <PropertyGroup>
-    <ConfigurationType>Application</ConfigurationType>
-    <CharacterSet>MultiByte</CharacterSet>
-    <PlatformToolset>v140</PlatformToolset>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)'=='Debug'">
-    <UseDebugLibraries>true</UseDebugLibraries>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)'=='Release'">
-    <WholeProgramOptimization>true</WholeProgramOptimization>
-  </PropertyGroup>
-  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
-  <ImportGroup Label="ExtensionSettings">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.1.props" />
-  </ImportGroup>
-  <ImportGroup Label="PropertySheets">
-    <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
-  </ImportGroup>
-  <PropertyGroup Label="UserMacros" />
-  <PropertyGroup>
-    <IntDir>$(Platform)/$(Configuration)/</IntDir>
-    <IncludePath>$(IncludePath)</IncludePath>
-    <CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
-    <CodeAnalysisRules />
-    <CodeAnalysisRuleAssemblies />
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Platform)'=='x64'">
-    <OutDir>../../bin/win64/$(Configuration)/</OutDir>
-  </PropertyGroup>
-  <ItemDefinitionGroup>
-    <ClCompile>
-      <WarningLevel>Level3</WarningLevel>
-      <PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
-      <AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../Common;</AdditionalIncludeDirectories>
-    </ClCompile>
-    <Link>
-      <SubSystem>Console</SubSystem>
-      <AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
-      <AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
-      <OutputFile>$(OutDir)/simpleAWBarrier.exe</OutputFile>
-    </Link>
-    <CudaCompile>
-      <CodeGeneration>compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;</CodeGeneration>
-      <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
-      <Include>./;../../Common</Include>
-      <Defines>WIN32</Defines>
-    </CudaCompile>
-  </ItemDefinitionGroup>
-  <ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
-    <ClCompile>
-      <Optimization>Disabled</Optimization>
-      <RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
-    </ClCompile>
-    <Link>
-      <GenerateDebugInformation>true</GenerateDebugInformation>
-      <LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
-    </Link>
-    <CudaCompile>
-      <Runtime>MTd</Runtime>
-      <TargetMachinePlatform>64</TargetMachinePlatform>
-    </CudaCompile>
-  </ItemDefinitionGroup>
-  <ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
-    <ClCompile>
-      <Optimization>MaxSpeed</Optimization>
-      <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
-    </ClCompile>
-    <Link>
-      <GenerateDebugInformation>false</GenerateDebugInformation>
-      <LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
-    </Link>
-    <CudaCompile>
-      <Runtime>MT</Runtime>
-      <TargetMachinePlatform>64</TargetMachinePlatform>
-    </CudaCompile>
-  </ItemDefinitionGroup>
-  <ItemGroup>
-    <CudaCompile Include="simpleAWBarrier.cu" />
-
-  </ItemGroup>
-  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
-  <ImportGroup Label="ExtensionTargets">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.1.targets" />
-  </ImportGroup>
-</Project>

Samples/simpleAWBarrier/simpleAWBarrier_vs2017.vcxproj

@@ -38,7 +38,7 @@
   </PropertyGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
   <ImportGroup Label="ExtensionSettings">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.1.props" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
   </ImportGroup>
   <ImportGroup Label="PropertySheets">
     <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -71,6 +71,7 @@
       <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
       <Include>./;../../Common</Include>
       <Defines>WIN32</Defines>
+      <AdditionalOptions>--threads 0</AdditionalOptions>
     </CudaCompile>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
@@ -107,6 +108,6 @@
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.1.targets" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
   </ImportGroup>
 </Project>

Samples/simpleAWBarrier/simpleAWBarrier_vs2019.vcxproj

@@ -34,7 +34,7 @@
   </PropertyGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
   <ImportGroup Label="ExtensionSettings">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.1.props" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.2.props" />
   </ImportGroup>
   <ImportGroup Label="PropertySheets">
     <Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
@@ -67,6 +67,7 @@
       <AdditionalOptions>-Xcompiler "/wd 4819" %(AdditionalOptions)</AdditionalOptions>
       <Include>./;../../Common</Include>
       <Defines>WIN32</Defines>
+      <AdditionalOptions>--threads 0</AdditionalOptions>
     </CudaCompile>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
@@ -103,6 +104,6 @@
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets">
-    <Import Project="$(CUDAPropsPath)\CUDA 11.1.targets" />
+    <Import Project="$(CUDAPropsPath)\CUDA 11.2.targets" />
   </ImportGroup>
 </Project>