1
0
mirror of https://github.com/andre-wojtowicz/blas-benchmarks synced 2024-11-24 09:40:27 +01:00

added naive cublas optimize

This commit is contained in:
Andrzej Wójtowicz 2016-05-30 17:49:09 +02:00
parent 095e5278d2
commit 5c64266f39
2 changed files with 54 additions and 2 deletions

View File

@ -3,8 +3,8 @@ runs = 10
f = function() f = function()
{ {
set.seed (1) set.seed (1)
m <- 4000 m <- 3500
n <- 4000 n <- 3500
A <- matrix (runif (m*n),m,n) A <- matrix (runif (m*n),m,n)
# Matrix multiply # Matrix multiply

View File

@ -546,6 +546,58 @@ function cublas_install {
echo "Finished installing cuBLAS" echo "Finished installing cuBLAS"
} }
function cublas_optimize {
if [ $(nvidia-detect | grep -c "No NVIDIA GPU detected.") -eq 1 ]; then
echo "No NVIDIA GPU detected, cuBLAS installation aborted"
return 1
fi
echo "Started optimizing cuBLAS"
dims_arr=(128 256 512 1024 2048 4096)
best_dim_val=-1
best_dim_time=-1
for dim in ${dims_arr[*]}
do
echo -n "$dim - "
echo "
NVBLAS_LOGFILE /dev/null
NVBLAS_CPU_BLAS_LIB ${DIR_MKL}/libRblas.so
NVBLAS_GPU_LIST ALL0
NVBLAS_TILE_DIM $dim
#NVBLAS_AUTOPIN_MEM_ENABLED" > ${DIR_CUBLAS}/nvblas.conf
cu_out=$(cublas_check | tail -2 | head -1)
echo $cu_out
if [ $best_dim_val -eq -1 ]; then
echo "first"
best_dim_val=$dim
best_dim_time=$cu_out
elif [ $(echo "$cu_out < $best_dim_time" | bc) -eq 1 ] ; then
echo "better"
best_dim_time=$cu_out
best_dim_val=$dim
fi
done
echo "best: ${best_dim_val} ${best_dim_time}"
echo "
NVBLAS_LOGFILE /dev/null
NVBLAS_CPU_BLAS_LIB ${DIR_MKL}/libRblas.so
NVBLAS_GPU_LIST ALL0
NVBLAS_TILE_DIM $best_dim_val
#NVBLAS_AUTOPIN_MEM_ENABLED" > ${DIR_CUBLAS}/nvblas.conf
echo "Finished optimizing cuBLAS"
}
# CUDA 7.5 currently supports linux kernel 3.x # CUDA 7.5 currently supports linux kernel 3.x
# #
#function cublas_online_install { #function cublas_online_install {