From c8a7469c7910f5cfa6ea42cc3a188ed0463cb5de Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Jakub=20Klinkovsk=C3=BD?= <klinkjak@fjfi.cvut.cz>
Date: Tue, 16 Jul 2019 22:40:46 +0200
Subject: [PATCH] Fixed ParallelFor and added tests
This reverts commit cf34dca066ff8368c3d591e7bf74145a4ea025f9.
---
src/TNL/Devices/Cuda.h | 4 +-
src/TNL/Devices/Cuda_impl.h | 4 +-
src/TNL/ParallelFor.h | 12 +-
src/UnitTests/CMakeLists.txt | 10 ++
src/UnitTests/ParallelForTest.cpp | 1 +
src/UnitTests/ParallelForTest.cu | 1 +
src/UnitTests/ParallelForTest.h | 290 ++++++++++++++++++++++++++++++
7 files changed, 312 insertions(+), 10 deletions(-)
create mode 100644 src/UnitTests/ParallelForTest.cpp
create mode 100644 src/UnitTests/ParallelForTest.cu
create mode 100644 src/UnitTests/ParallelForTest.h
diff --git a/src/TNL/Devices/Cuda.h b/src/TNL/Devices/Cuda.h
index 80108bf530..7831014155 100644
--- a/src/TNL/Devices/Cuda.h
+++ b/src/TNL/Devices/Cuda.h
@@ -75,10 +75,10 @@ class Cuda
/****
* This functions helps to count number of CUDA grids depending on the
- * number of the CUDA threads and maximum grid size.
+ * number of the CUDA blocks and maximum grid size.
* It is obsolete and it will be replaced by setupThreads.
*/
- static inline int getNumberOfGrids( const int threads,
+ static inline int getNumberOfGrids( const int blocks,
const int gridSize = getMaxGridSize() );
#ifdef HAVE_CUDA
diff --git a/src/TNL/Devices/Cuda_impl.h b/src/TNL/Devices/Cuda_impl.h
index a40375c39c..234f45b720 100644
--- a/src/TNL/Devices/Cuda_impl.h
+++ b/src/TNL/Devices/Cuda_impl.h
@@ -111,10 +111,10 @@ inline int Cuda::getNumberOfBlocks( const int threads,
return roundUpDivision( threads, blockSize );
}
-inline int Cuda::getNumberOfGrids( const int threads,
+inline int Cuda::getNumberOfGrids( const int blocks,
const int gridSize )
{
- return roundUpDivision( threads, gridSize );
+ return roundUpDivision( blocks, gridSize );
}
#ifdef HAVE_CUDA
diff --git a/src/TNL/ParallelFor.h b/src/TNL/ParallelFor.h
index 950a577bb4..04af274080 100644
--- a/src/TNL/ParallelFor.h
+++ b/src/TNL/ParallelFor.h
@@ -205,7 +205,7 @@ struct ParallelFor< Devices::Cuda, Mode >
dim3 gridSize;
gridSize.x = TNL::min( Devices::Cuda::getMaxGridSize(), Devices::Cuda::getNumberOfBlocks( end - start, blockSize.x ) );
- if( Devices::Cuda::getNumberOfGrids( end - start ) == 1 )
+ if( (std::size_t) blockSize.x * gridSize.x >= (std::size_t) end - start )
ParallelForKernel< false ><<< gridSize, blockSize >>>( start, end, f, args... );
else {
// decrease the grid size and align to the number of multiprocessors
@@ -257,8 +257,8 @@ struct ParallelFor2D< Devices::Cuda, Mode >
gridSize.y = TNL::min( Devices::Cuda::getMaxGridSize(), Devices::Cuda::getNumberOfBlocks( sizeY, blockSize.y ) );
dim3 gridCount;
- gridCount.x = Devices::Cuda::getNumberOfGrids( sizeX );
- gridCount.y = Devices::Cuda::getNumberOfGrids( sizeY );
+ gridCount.x = roundUpDivision( sizeX, blockSize.x * gridSize.x );
+ gridCount.y = roundUpDivision( sizeY, blockSize.y * gridSize.y );
if( gridCount.x == 1 && gridCount.y == 1 )
ParallelFor2DKernel< false, false ><<< gridSize, blockSize >>>
@@ -342,9 +342,9 @@ struct ParallelFor3D< Devices::Cuda, Mode >
gridSize.z = TNL::min( Devices::Cuda::getMaxGridSize(), Devices::Cuda::getNumberOfBlocks( sizeZ, blockSize.z ) );
dim3 gridCount;
- gridCount.x = Devices::Cuda::getNumberOfGrids( sizeX );
- gridCount.y = Devices::Cuda::getNumberOfGrids( sizeY );
- gridCount.z = Devices::Cuda::getNumberOfGrids( sizeZ );
+ gridCount.x = roundUpDivision( sizeX, blockSize.x * gridSize.x );
+ gridCount.y = roundUpDivision( sizeY, blockSize.y * gridSize.y );
+ gridCount.z = roundUpDivision( sizeZ, blockSize.z * gridSize.z );
if( gridCount.x == 1 && gridCount.y == 1 && gridCount.z == 1 )
ParallelFor3DKernel< false, false, false ><<< gridSize, blockSize >>>
diff --git a/src/UnitTests/CMakeLists.txt b/src/UnitTests/CMakeLists.txt
index 564a7d1347..f7d3b68b03 100644
--- a/src/UnitTests/CMakeLists.txt
+++ b/src/UnitTests/CMakeLists.txt
@@ -35,6 +35,15 @@ ADD_EXECUTABLE( ObjectTest ObjectTest.cpp )
TARGET_COMPILE_OPTIONS( ObjectTest PRIVATE ${CXX_TESTS_FLAGS} )
TARGET_LINK_LIBRARIES( ObjectTest ${GTEST_BOTH_LIBRARIES} )
+if( BUILD_CUDA )
+ CUDA_ADD_EXECUTABLE( ParallelForTest ParallelForTest.cu OPTIONS ${CXX_TESTS_FLAGS} )
+ TARGET_LINK_LIBRARIES( ParallelForTest ${GTEST_BOTH_LIBRARIES} )
+else()
+ ADD_EXECUTABLE( ParallelForTest ParallelForTest.cpp )
+ TARGET_COMPILE_OPTIONS( ParallelForTest PRIVATE ${CXX_TESTS_FLAGS} )
+ TARGET_LINK_LIBRARIES( ParallelForTest ${GTEST_BOTH_LIBRARIES} )
+endif()
+
ADD_EXECUTABLE( SaveAndLoadMeshfunctionTest SaveAndLoadMeshfunctionTest.cpp )
TARGET_COMPILE_OPTIONS( SaveAndLoadMeshfunctionTest PRIVATE ${CXX_TESTS_FLAGS} )
TARGET_LINK_LIBRARIES( SaveAndLoadMeshfunctionTest ${GTEST_BOTH_LIBRARIES} )
@@ -46,4 +55,5 @@ endif()
ADD_TEST( FileTest ${EXECUTABLE_OUTPUT_PATH}/FileTest${CMAKE_EXECUTABLE_SUFFIX} )
ADD_TEST( StringTest ${EXECUTABLE_OUTPUT_PATH}/StringTest${CMAKE_EXECUTABLE_SUFFIX} )
ADD_TEST( ObjectTest ${EXECUTABLE_OUTPUT_PATH}/ObjectTest${CMAKE_EXECUTABLE_SUFFIX} )
+ADD_TEST( ParallelForTest ${EXECUTABLE_OUTPUT_PATH}/ParallelForTest${CMAKE_EXECUTABLE_SUFFIX} )
ADD_TEST( SaveAndLoadMeshfunctionTest ${EXECUTABLE_OUTPUT_PATH}/SaveAndLoadMeshfunctionTest${CMAKE_EXECUTABLE_SUFFIX} )
diff --git a/src/UnitTests/ParallelForTest.cpp b/src/UnitTests/ParallelForTest.cpp
new file mode 100644
index 0000000000..ed0e90c0b1
--- /dev/null
+++ b/src/UnitTests/ParallelForTest.cpp
@@ -0,0 +1 @@
+#include "ParallelForTest.h"
diff --git a/src/UnitTests/ParallelForTest.cu b/src/UnitTests/ParallelForTest.cu
new file mode 100644
index 0000000000..ed0e90c0b1
--- /dev/null
+++ b/src/UnitTests/ParallelForTest.cu
@@ -0,0 +1 @@
+#include "ParallelForTest.h"
diff --git a/src/UnitTests/ParallelForTest.h b/src/UnitTests/ParallelForTest.h
new file mode 100644
index 0000000000..95455286e7
--- /dev/null
+++ b/src/UnitTests/ParallelForTest.h
@@ -0,0 +1,290 @@
+/***************************************************************************
+ ParallelForTest.h - description
+ -------------------
+ begin : Jul 16, 2019
+ copyright : (C) 2019 by Tomas Oberhuber et al.
+ email : tomas.oberhuber@fjfi.cvut.cz
+ ***************************************************************************/
+
+/* See Copyright Notice in tnl/Copyright */
+
+#include <TNL/Devices/Host.h>
+#include <TNL/Devices/Cuda.h>
+#include <TNL/Containers/Array.h>
+#include <TNL/ParallelFor.h>
+
+#ifdef HAVE_GTEST
+#include <gtest/gtest.h>
+#endif
+
+using namespace TNL;
+
+#ifdef HAVE_GTEST
+TEST( ParallelForTest, 1D_host )
+{
+ using Array = Containers::Array< int, Devices::Host >;
+ Array a;
+ for (int size = 100; size <= 100000000; size *= 100)
+ {
+ Array expected;
+ expected.setSize( size );
+ for (int i = 0; i < size; i++)
+ expected[ i ] = i;
+
+ a.setSize( size );
+ a.setValue( 0 );
+ auto view = a.getView();
+ auto kernel = [=] (int i) mutable
+ {
+ view[i] = i;
+ };
+ ParallelFor< Devices::Host >::exec( 0, size, kernel );
+
+ if( a != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( a[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+ }
+}
+
+TEST( ParallelForTest, 2D_host )
+{
+ using Array = Containers::Array< int, Devices::Host >;
+ Array a;
+ for (int size = 100; size <= 100000000; size *= 100)
+ {
+ Array expected;
+ expected.setSize( size );
+ for (int i = 0; i < size; i++)
+ expected[ i ] = i;
+
+ a.setSize( size );
+ a.setValue( 0 );
+ auto view = a.getView();
+ auto kernel1 = [=] (int i, int j) mutable
+ {
+ view[i] = i;
+ };
+ ParallelFor2D< Devices::Host >::exec( 0, 0, size, 1, kernel1 );
+
+ if( a != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( a[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+
+ a.setValue( 0 );
+ auto kernel2 = [=] (int i, int j) mutable
+ {
+ view[j] = j;
+ };
+ ParallelFor2D< Devices::Host >::exec( 0, 0, 1, size, kernel2 );
+
+ if( a != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( a[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+ }
+}
+
+TEST( ParallelForTest, 3D_host )
+{
+ using Array = Containers::Array< int, Devices::Host >;
+ Array a;
+ for (int size = 100; size <= 100000000; size *= 100)
+ {
+ Array expected;
+ expected.setSize( size );
+ for (int i = 0; i < size; i++)
+ expected[ i ] = i;
+
+ a.setSize( size );
+ a.setValue( 0 );
+ auto view = a.getView();
+ auto kernel1 = [=] (int i, int j, int k) mutable
+ {
+ view[i] = i;
+ };
+ ParallelFor3D< Devices::Host >::exec( 0, 0, 0, size, 1, 1, kernel1 );
+
+ if( a != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( a[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+
+ a.setValue( 0 );
+ auto kernel2 = [=] (int i, int j, int k) mutable
+ {
+ view[j] = j;
+ };
+ ParallelFor3D< Devices::Host >::exec( 0, 0, 0, 1, size, 1, kernel2 );
+
+ if( a != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( a[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+
+ a.setValue( 0 );
+ auto kernel3 = [=] (int i, int j, int k) mutable
+ {
+ view[k] = k;
+ };
+ ParallelFor3D< Devices::Host >::exec( 0, 0, 0, 1, 1, size, kernel3 );
+
+ if( a != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( a[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+ }
+}
+
+#ifdef HAVE_CUDA
+// nvcc does not allow __cuda_callable__ lambdas inside private regions
+void test_1D_cuda()
+{
+ using Array = Containers::Array< int, Devices::Cuda >;
+ using ArrayHost = Containers::Array< int, Devices::Host >;
+ Array a;
+ for (int size = 100; size <= 100000000; size *= 100)
+ {
+ ArrayHost expected;
+ expected.setSize( size );
+ for (int i = 0; i < size; i++)
+ expected[ i ] = i;
+
+ a.setSize( size );
+ a.setValue( 0 );
+ auto view = a.getView();
+ auto kernel = [=] __cuda_callable__ (int i) mutable
+ {
+ view[i] = i;
+ };
+ ParallelFor< Devices::Cuda >::exec( 0, size, kernel );
+
+ ArrayHost ah;
+ ah = a;
+ if( ah != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( ah[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+ }
+}
+
+TEST( ParallelForTest, 1D_cuda )
+{
+ test_1D_cuda();
+}
+
+// nvcc does not allow __cuda_callable__ lambdas inside private regions
+void test_2D_cuda()
+{
+ using Array = Containers::Array< int, Devices::Cuda >;
+ using ArrayHost = Containers::Array< int, Devices::Host >;
+ Array a;
+ for (int size = 100; size <= 100000000; size *= 100)
+ {
+ ArrayHost expected;
+ expected.setSize( size );
+ for (int i = 0; i < size; i++)
+ expected[ i ] = i;
+
+ a.setSize( size );
+ a.setValue( 0 );
+ auto view = a.getView();
+ auto kernel1 = [=] __cuda_callable__ (int i, int j) mutable
+ {
+ view[i] = i;
+ };
+ ParallelFor2D< Devices::Cuda >::exec( 0, 0, size, 1, kernel1 );
+
+ ArrayHost ah;
+ ah = a;
+ if( ah != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( ah[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+
+ a.setValue( 0 );
+ auto kernel2 = [=] __cuda_callable__ (int i, int j) mutable
+ {
+ view[j] = j;
+ };
+ ParallelFor2D< Devices::Cuda >::exec( 0, 0, 1, size, kernel2 );
+
+ ah = a;
+ if( ah != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( ah[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+ }
+}
+
+TEST( ParallelForTest, 2D_cuda )
+{
+ test_2D_cuda();
+}
+
+// nvcc does not allow __cuda_callable__ lambdas inside private regions
+void test_3D_cuda()
+{
+ using Array = Containers::Array< int, Devices::Cuda >;
+ using ArrayHost = Containers::Array< int, Devices::Host >;
+ Array a;
+ for (int size = 100; size <= 100000000; size *= 100)
+ {
+ ArrayHost expected;
+ expected.setSize( size );
+ for (int i = 0; i < size; i++)
+ expected[ i ] = i;
+
+ a.setSize( size );
+ a.setValue( 0 );
+ auto view = a.getView();
+ auto kernel1 = [=] __cuda_callable__ (int i, int j, int k) mutable
+ {
+ view[i] = i;
+ };
+ ParallelFor3D< Devices::Cuda >::exec( 0, 0, 0, size, 1, 1, kernel1 );
+
+ ArrayHost ah;
+ ah = a;
+ if( ah != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( ah[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+
+ a.setValue( 0 );
+ auto kernel2 = [=] __cuda_callable__ (int i, int j, int k) mutable
+ {
+ view[j] = j;
+ };
+ ParallelFor3D< Devices::Cuda >::exec( 0, 0, 0, 1, size, 1, kernel2 );
+
+ ah = a;
+ if( ah != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( ah[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+
+ a.setValue( 0 );
+ auto kernel3 = [=] __cuda_callable__ (int i, int j, int k) mutable
+ {
+ view[k] = k;
+ };
+ ParallelFor3D< Devices::Cuda >::exec( 0, 0, 0, 1, 1, size, kernel3 );
+
+ ah = a;
+ if( ah != expected ) {
+ for (int i = 0; i < size; i++)
+ ASSERT_EQ( ah[i], i ) << "First index at which the result is wrong is i = " << i;
+ }
+ }
+}
+
+TEST( ParallelForTest, 3D_cuda )
+{
+ test_3D_cuda();
+}
+#endif
+#endif
+
+#include "main.h"
--
GitLab