Apologies in advance if this has been reported before:
The OpenCL HD Graphics 4000 compiler crashes when I attempt to build the following minimal kernel
__kernel void pix(__write_only image2d_t im, const ulong data)
{
int2 coord = { 1, 2 };
if (data)
write_imagef(im, coord, (float4)(0.0f,1.0f,0.0f,1.0f));
else
write_imagef(im, coord, (float4)(0.0f,0.0f,0.0f,1.0f));
}
The set-up is as follows:
Windows 8 (64 bit), Intel graphics driver version 15.28.12.2932(9.17.10.2932), Intel SDK for OpenCLApplications 2013 Beta.
The compiler crash occurs in both my application and when I try to compile using the offline compiler. The kernel builds and executes successfully for the CPU. Is this a known issue?
链接已复制
Rami,
Thanks for replying. The compiler actually crashes during the compilation step, so it is not possible to make the clGetProgramBuildInfo API call. When I try to build the kernel using the offline kernel builder it compiles correctly when I select Intel CPU as the target, but the compiler crashes when I select Intel HD Graphics as the target. When the compiler crashes I get the following information from Windows:
Problem Event Name: APPCRASH
Application Name: ioc64.exe
Application Version: 3.0.0.43087
Application Timestamp: 5018e60a
Fault Module Name: igdfcl64.dll
Fault Module Version: 8.1.0.2932
Fault Module Timestamp: 50c924e2
Exception Code: c0000005
Exception Offset: 000000000083a94a
OS Version: 6.2.9200.2.0.0.768.101
Locale ID: 2057
If I comment out the the "else" then the compilation succeeds (but obviously the logic of the kernel changes).
Would it be possible for you to cut and paste the kernel I've given in the opening post and try to compile in the offline Kernel Builder, to see if you can reproduce the crash?
Thanks,
Ian
I tried your kernel and indeed the compiler crashes during clBuildProgram() call, so querrying the build log is not possible. I will try a later driver to see if the issue is fixed. I'll keep you posted.
Thanks,
Raghu
I have seen crashes in the compiler too and while I don't have the issue right now it would be good if the compiler was able to catch the exception itself and avoid crashing so we can try to get some error somehow.
Thanks.
Laurent.
