26.06.2015 Views

Advanced_GPU_talk

Advanced_GPU_talk

Advanced_GPU_talk

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Overlap Example<br />

cudaStream_t streams[8];<br />

for (int i = 0; i < 8; i++) cudaStreamCreate(&stream[i]);<br />

!<br />

cudaMallocHost(host1); cudaMalloc(device1);<br />

!<br />

cudaMemcpyAsync(host1,device1,D2H,stream[0]);<br />

cudaMemcpyAsync(device2,host2,H2D,stream[1]);<br />

for (int i = 2; i < 8; i++) {<br />

H2D<br />

Copy<br />

D2H<br />

Copy<br />

kernel_do_sg(…);<br />

engine<br />

engine<br />

}<br />

kernel<br />

SM1<br />

SM2<br />

SM3<br />

kernel<br />

kernel<br />

Compute<br />

SM4<br />

kernel<br />

kernel<br />

engine<br />

SM5<br />

kernel<br />

SM6

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!