I just re-ordered the members of clScanlineContour (may not be aligned now, I haven't been able to test)
and added Contour.mError += to lots of lines in MakeContoursFromScanlineGroups
Now when I build trace.cl (either in my app or in KA2) it never finishes (And I can't close KA2 as it's waiting for the build to finish)
any idea why? Just a driver/compiler bug?
Using latest SDK (2.8.1), catalyst control center 13.6 and the 13.6 BETA2 drivers on a ATI Radeon HD 5670
Old clScanlineContour struct;
typedef struct clScanlineContour
{
cl_float4 mColour;
cl_int4 mVertexCount;
cl_int2 mVertexes[clScanlineContour_MAXVERTS]; // x,y
} clScanlineContour;
I thought I boiled this down to a simple change, but it seems instead it just takes a LONG time to compile (on a dual hyperthreaded quad xeon).... maybe close this thread if I don't have any more constructive replies
I will try to reproduce it and let you know.
I'm pretty sure it's just my ridiculous sized struct making the compiler do a lot of work with registers... but I'm curious to see if you find anything more interesting than that
The kernel is taking 15-30 seconds to compile for me. Forwarding it to compiler team.
We were able to reproduce this issue with the latest driver and latest CodeXL. We will keep you updated on the resolution.