OpenCL S3D fails on OS X
Closed this issue · 6 comments
Gives the following output:
$ ../bin/Serial/OpenCL/S3D -s 1 -p 0 -d 1
Chose device: name='GeForce GT 330M' index=1 id=0x1022600
Compiling kernels (phase 1)...Error: CL_BUILD_PROGRAM_FAILURE in S3D.cpp line 306
Retsize: 69
Log: kernel referenced an external function exp, that could not be found.
Error: CL_BUILD_PROGRAM_FAILURE in S3D.cpp line 315
Retsize: 69
Log: kernel referenced an external function exp, that could not be found.
Error: CL_BUILD_PROGRAM_FAILURE in S3D.cpp line 317
Retsize: 69
Log: kernel referenced an external function exp, that could not be found.
done.
Generating OpenCL Kernel Objects (phase 1)...Error: CL_INVALID_PROGRAM_EXECUTABLE in S3D.cpp line 337
Error: CL_INVALID_PROGRAM_EXECUTABLE in S3D.cpp line 355
Error: CL_INVALID_PROGRAM_EXECUTABLE in S3D.cpp line 361
done.
Error: CL_INVALID_KERNEL in S3D.cpp line 390
Error: CL_INVALID_KERNEL in S3D.cpp line 392
Error: CL_INVALID_KERNEL in S3D.cpp line 394
Error: CL_INVALID_KERNEL in S3D.cpp line 404
Error: CL_INVALID_KERNEL in S3D.cpp line 406
Error: CL_INVALID_KERNEL in S3D.cpp line 408
Error: CL_INVALID_KERNEL in S3D.cpp line 411
Error: CL_INVALID_KERNEL in S3D.cpp line 411
Error: CL_INVALID_KERNEL in S3D.cpp line 411
Error: CL_INVALID_KERNEL in S3D.cpp line 411
Error: CL_INVALID_KERNEL in S3D.cpp line 411
Error: CL_INVALID_KERNEL in S3D.cpp line 411
Executing kernels (phase 1)...Memory fault
So this complains about source for ratt10.cl and ratxb.cl, but exp() is a built-in OpenCL math function. Am I missing something here? Confirmed exp presence in 1.0 and 1.1 spec.
I think I know what's going on. A missing CPREC around a constant, so it is treating it as double precision and there isn't a double precision exp(). Give me a few minutes and I'll see if I'm right.
Phil
On Mar 6, 2012, at 10:54 , Kyle Spafford wrote:
So this complains about source for ratt10.cl and ratxb.cl, but exp() is a built-in OpenCL math function. Am I missing something here? Confirmed exp presence in 1.0 and 1.1 spec.
Reply to this email directly or view it on GitHub:
https://github.com/spaffy/shoc/issues/6#issuecomment-4346807
Philip C. Roth | +1 865 241-1543 | http://ft.ornl.gov/~rothpc
Hmm I see several missing CPREC wrappers, so I will follow up on those after your testing anyway.
Yes, I think that's what the problem is. Fixing ratt will take a while, the others are already done.
On Mar 6, 2012, at 11:01 , Roth, Philip C. wrote:
I think I know what's going on. A missing CPREC around a constant, so it is treating it as double precision and there isn't a double precision exp(). Give me a few minutes and I'll see if I'm right.
Phil
On Mar 6, 2012, at 10:54 , Kyle Spafford wrote:
So this complains about source for ratt10.cl and ratxb.cl, but exp() is a built-in OpenCL math function. Am I missing something here? Confirmed exp presence in 1.0 and 1.1 spec.
Reply to this email directly or view it on GitHub:
https://github.com/spaffy/shoc/issues/6#issuecomment-4346807Philip C. Roth | +1 865 241-1543 | http://ft.ornl.gov/~rothpc
Philip C. Roth | +1 865 241-1543 | http://ft.ornl.gov/~rothpc
I think all these are now committed.
Yes, this issue is resolved.