0 Replies Latest reply on Jun 22, 2009 7:23 AM by sgratton

    bug report involving global buffers and domains not starting at 0,0

    sgratton
      uncertainty over the domain for which output occurs

       

      Hi there,

       

      An old issue, first mentioned in:

       

      http://forums.amd.com/devforum/messageview.cfm?catid=328&threadid=98181

       

      still seems to be present unfortunately.  Consider a simple IL program that writes wWinCoord.x to either o0 or g[int(wWinCoord.x)] or both, operating on a 1d domain (CALdomain dom={a,0,w,0}.  If

      A: you a run a program with a output buffer o0 for output, non-zero output appears in the output buffer for elements from a to a+w-1 

      B: you run a program with a global buffer g[] for output, non-zero output appears from a to w-1

      C: you run a program with both o0 and g[] for output, non-zero output appears in both buffers form a to a+w-1.

      The kind of program I have in mind in case C is...

      il_ps_2_0

      dcl_input_position_interp(linear_noperspective) vWinCoord0.xy__

      dcl_output_generic o0

      ftoi r0,vWinCoord0.xy00

      itof r1,r0

      mov g[r0.x],r1.x

      mov o0,r1.x

      ret_dyn

      end

       

      For case A comment out the mov g[r0.x],r1.x  whereas for case B comment out dcl_output_generic o0 and mov o0,r1.x.

      Note the difference in the element for which output ends.  When output buffers are involved, it is as though the last two elements of a CALdomain give you the width and height whereas without output buffers the last two elements give you the bottom right corner. 

      Best,

      Steven.