Databases
StackOverflow2013
Usersdbo
USjxj
 

Note that there are some explanatory texts on larger screens.

plurals
  1. USjxj
    primarykey
    data
    text
    plurals
    1. POwhat's the correct and most efficient way to use mapped(zero-copy) memory mechanism in Nvidia OpenCL environment?
      singulars
      1. PTQuestion
    2. POhow to profile sequential launched multiple OpenCL kernels by one clFinish?
      singulars
      1. PTQuestion
    1. POWhy dead code in OpenCL kernel influence result in Nvidia GTX550ti?
      singulars
      1. PTQuestion
    2. POhow to profile sequential launched multiple OpenCL kernels by one clFinish?
      singulars
      1. PTQuestion
    3. PO
      singulars
      1. PTAnswer
    1. VO
      singulars
      1. POWhy dead code in OpenCL kernel influence result in Nvidia GTX550ti?
      1. USjxj
      1. VTFavorite
    2. VO
      singulars
      1. POhow to profile sequential launched multiple OpenCL kernels by one clFinish?
      1. USjxj
      1. VTFavorite
    3. VO
      singulars
      1. POHow to make multiple properties valid to one file by set_source_files_properties in cmake?
      1. USjxj
      1. VTFavorite
    1. COIn CUDA_C_Programming_Guide 3.2.4.3: "Such a block has therefore in general two addresses: ...". But soon after that: "The only exception is for pointers ... Unified Virtual Address Space". In 3.2.7. Unified Virtual Address Space, the 64bits and compute capability 2.0 can be met easily now. By using device side pointer, there is no way for us to know if the real implementation is the "good" or "bad" as you said. In Nvida's profiling tool, there is only kernel execution period shown, not any other transfer period can be seen because we don't use writing buffer explicitly.
      singulars
      1. PO
      1. USjxj
    2. COHi, thanks for your replying. But I don't think OpenCL2.0 works for my original purpose. Neither CL_USE_HOST_PTR nor CL_MEM_ALLOC_HOST_PTR in OpenCL1.2. I actually inspected Nvidia's SDK examples and best practice guide, then I realize that all OpenCL tricks plays before kernel launch, so this doesn't help overlapping transfer and computation (CUDA's device side pointer of host memory does, because transfer begins after kernel launch on kernel's demand)
      singulars
      1. PO
      1. USjxj
    3. COIt is a good manner for overlapping transfer and computation. Explicit copy operation issued from host side is avoided, which can only copy data from host to device global memory. By device side pointer, data can be transfered between host and shared memory directly. It allows device schedule computation and data transfer as its need, which implies data transfer may be hidden. Traditional way is multi streams (CUDA) and multi command queues(OpenCL). The traditional way needs explicit scheduling in host side, which makes the overall code a little bit ugly/hairy.
      singulars
      1. POIs there device side pointer of host memory for kernel use in OpenCL (like CUDA)?
      1. USjxj
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload