Mesa (gallium-cylindrical-wrap): softpipe: Properly implement cylindrical wrapping.

[Brian Paul]

michal wrote:
> Brian Paul wrote on 2010-02-05 18:56:
>> michal wrote:
>>   
>>> Micha?? Kr??l wrote on 2010-02-05 17:34:
>>>     
>>>> Module: Mesa
>>>> Branch: gallium-cylindrical-wrap
>>>> Commit: f90399107eb9b54ea2c76f973ab487c38e61a5f0
>>>> URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=f90399107eb9b54ea2c76f973ab487c38e61a5f0
>>>>
>>>> Author: Michal Krol <michal at vmware.com>
>>>> Date:   Fri Feb  5 17:12:48 2010 +0100
>>>>
>>>> softpipe: Properly implement cylindrical wrapping.
>>>>
>>>> @@ -801,21 +900,23 @@ line_linear_coeff(const struct setup_context *setup,
>>>>  /**
>>>>   * Compute a0, dadx and dady for a perspective-corrected interpolant,
>>>>   * for a line.
>>>> + * v[0] and v[1] are vmin and vmax, respectively.
>>>>   */
>>>>  static void
>>>>  line_persp_coeff(const struct setup_context *setup,
>>>>                   struct tgsi_interp_coef *coef,
>>>> -                 uint vertSlot, uint i)
>>>> +                 uint i,
>>>> +                 const float v[2])
>>>>  {
>>>>     /* XXX double-check/verify this arithmetic */
>>>> -   const float a0 = setup->vmin[vertSlot][i] * setup->vmin[0][3];
>>>> -   const float a1 = setup->vmax[vertSlot][i] * setup->vmax[0][3];
>>>> +   const float a0 = v[0] * setup->vmin[0][3];
>>>> +   const float a1 = v[1] * setup->vmax[0][3];
>>>>     const float da = a1 - a0;
>>>>     const float dadx = da * setup->emaj.dx * setup->oneoverarea;
>>>>     const float dady = da * setup->emaj.dy * setup->oneoverarea;
>>>>     coef->dadx[i] = dadx;
>>>>     coef->dady[i] = dady;
>>>> -   coef->a0[i] = (setup->vmin[vertSlot][i] -
>>>> +   coef->a0[i] = (v[0] - /* XXX: <-- shouldn't that be a0? */
>>>>   
>>>>       
>>> I have put this XXX comment, because the code in line_persp_coeff() is 
>>> not consistent wih tri_persp_coeff(). Can someone double check this?
>>>     
>> It works.  I added a line drawing mode to cylwrap.c to test.
>>
>>   
> Just noticed there are two XXX comments in the snippet. Are we talking 
> about the new one in the last line of the cited code?
> 
> /* XXX: <-- shouldn't that be a0? */

Yes, it looks like the code is wrong there.  I'll do some additional 
testing.  This is probably wrong in 7.7 too.

-Brian