文件列表:

RBased_DWT.cu (50997, 2016-07-16)
RBased_DWT_common.h (4475, 2016-07-16)
main_example.cu (3010, 2016-07-16)

The code for this project has been developed for research pourposes. Square images of size i\*2 x i\*2 with i > *** have been employed for all experiments. This implementation may not properly work for other image sizes. The "main_example.cu" first applies the Forward DWT and then the Reverse DWT over a set of randomly generated images. It can be compiled, executed and profiled using the nvcc compiler as follows: $nvcc -c -O3 -arch=sm_35 -use_fast_math -Xptxas -v -DDWT53_or_DWT97=1 -DNEXPERIMENTS=1 -DEXPERIMENT_INI=1024 -DNELEMENTS_THREAD_Y=20 -DSHUFFLE=1 -DLEVELS=5 *.cu $nvcc *.o -o test $nvprof --print-gpu-trace ./test Several precompilation parameters are defined in "RBased_DWT_common.h". Most important are: -DDWT53_or_DWT97: =1 the DWT 5/3 is computed, =0 the DWT 9/7 -DNEXPERIMENTS: Number of randomly generated images computed. -DEXPERIMENT_INI: Size in both dimensions of the randomly generated input image employed. -DNELEMENTS_THREAD_Y : Length in samples of the Data block computed by each warp. Width is fixed to *** samples. -DSHUFFLE: If =1 the shuffle instructions are employed in the computation. If =0, an intermediate buffer in shared memory is used instead (for pre-Kepler architectures). -DLEVELS : Number of DWT decomposition levels applied. Depending on the CUDA architecture, the code can be compiled as follows: Fermi 2.0: $nvcc -c -O3 -arch=sm_20 -use_fast_math -Xptxas -v -DDWT53_or_DWT97=1 -DNEXPERIMNETS=1 -DEXPERIMENT_INI=1024 -DNELEMENTS_THREAD_Y=20 -DSHUFFLE=0 -DLEVELS=5 *.cu Kepler 3.5: $nvcc -c -O3 -arch=sm_35 -use_fast_math -Xptxas -v -DDWT53_or_DWT97=1 -DNEXPERIMNETS=1 -DEXPERIMENT_INI=1024 -DNELEMENTS_THREAD_Y=20 -DSHUFFLE=1 -DLEVELS=5 *.cu Maxwell 5.0: $nvcc -c -O3 -arch=sm_50 -use_fast_math -Xptxas -v -DDWT53_or_DWT97=1 -DNEXPERIMNETS=1 -DEXPERIMENT_INI=1024 -DNELEMENTS_THREAD_Y=20 -DSHUFFLE=1 -DLEVELS=5 *.cu

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：