From 99b90ba46b4b69034fe051d37aaf9a14b302391d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1=C5=A1=20Oberhuber?= <oberhuber.tomas@gmail.com>
Date: Wed, 10 Feb 2021 13:27:22 +0100
Subject: [PATCH] Deleting legacy constants in CSRAdaptiveKernel.
---
.../Algorithms/Segments/CSRAdaptiveKernel.h | 21 -------------------
.../Algorithms/Segments/CSRAdaptiveKernel.hpp | 2 +-
.../details/CSRAdaptiveKernelParameters.h | 2 +-
3 files changed, 2 insertions(+), 23 deletions(-)
diff --git a/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.h b/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.h
index 5ade54d02e..22cf447ecb 100644
--- a/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.h
+++ b/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.h
@@ -67,27 +67,6 @@ struct CSRAdaptiveKernel
static TNL::String getKernelType();
-
- static constexpr Index THREADS_ADAPTIVE = details::CSRAdaptiveKernelParameters< sizeof( Index ) >::CudaBlockSize(); //sizeof(Index) == 8 ? 128 : 256;
-
- // How many shared memory use per block in CSR Adaptive kernel
- static constexpr Index SHARED_PER_BLOCK = details::CSRAdaptiveKernelParameters< sizeof( Index ) >::StreamedSharedMemory(); //20000; //24576; TODO:
-
- // Number of elements in shared memory
- static constexpr Index SHARED = SHARED_PER_BLOCK/sizeof(double);
-
- // Number of warps in block for CSR Adaptive
- static constexpr Index WARPS = THREADS_ADAPTIVE / 32;
-
- // Number of elements in shared memory per one warp
- static constexpr Index SHARED_PER_WARP = SHARED / WARPS;
-
- // Max length of row to process one warp for CSR Light, MultiVector
- static constexpr Index MAX_ELEMENTS_PER_WARP = 384;
-
- // Max length of row to process one warp for CSR Adaptive
- static constexpr Index MAX_ELEMENTS_PER_WARP_ADAPT = details::CSRAdaptiveKernelParameters< sizeof( Index ) >::MaxAdaptiveElementsPerWarp();
-
template< typename Offsets >
void init( const Offsets& offsets );
diff --git a/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.hpp b/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.hpp
index ff2db147be..13c653c6c5 100644
--- a/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.hpp
+++ b/src/TNL/Algorithms/Segments/CSRAdaptiveKernel.hpp
@@ -165,7 +165,7 @@ initValueSize( const Offsets& offsets )
if( type == details::Type::LONG )
{
const Index blocksCount = inBlocks.size();
- const Index warpsPerCudaBlock = THREADS_ADAPTIVE / TNL::Cuda::getWarpSize();
+ const Index warpsPerCudaBlock = details::CSRAdaptiveKernelParameters< sizeof( Index ) >::CudaBlockSize() / TNL::Cuda::getWarpSize();
Index warpsLeft = roundUpDivision( blocksCount, warpsPerCudaBlock ) * warpsPerCudaBlock - blocksCount;
if( warpsLeft == 0 )
warpsLeft = warpsPerCudaBlock;
diff --git a/src/TNL/Algorithms/Segments/details/CSRAdaptiveKernelParameters.h b/src/TNL/Algorithms/Segments/details/CSRAdaptiveKernelParameters.h
index 2546580f8f..56f203a74b 100644
--- a/src/TNL/Algorithms/Segments/details/CSRAdaptiveKernelParameters.h
+++ b/src/TNL/Algorithms/Segments/details/CSRAdaptiveKernelParameters.h
@@ -24,7 +24,7 @@ struct CSRAdaptiveKernelParameters
*
* \return CUDA block size.
*/
- static constexpr int CudaBlockSize() { return 256; }; //sizeof( Value ) == 8 ? 128 : 256; };
+ static constexpr int CudaBlockSize() { return 128; }; //sizeof( Value ) == 8 ? 128 : 256; };
//std::max( ( int ) ( 1024 / sizeof( Value ) ), ( int ) Cuda::getWarpSize() ); };
/**
--
GitLab