mesa3d-dev

[Mesa3d-dev] s_texture.c

[Daniel V. <vo...@lo...>]

hi`

I tried running UT with the Mesa software renderer and when I turned on
trilinear filtering it crashed in sample_2d_linear being called with a
NULL pointer for the image from sample_2d_linear_mipmap_linear. The
problem seems to be that UT uses GL_TEXTURE_MAX_LEVEL and tObj->_P was 7
and tObj->_M was 5. From what I can read in mtypes.h shouldn't _M be
used instead of _P in the following code? Changing the code to use _M at
least made it work ;) BTW, why is _M a float and not an integer?


s_texture.c:840 [sample_2d_linear_mipmap_linear]

if (level >= tObj->_P) {
   sample_2d_linear( tObj, tObj->Image[tObj->_P], s, t, rgba);
}
else {
   GLchan t0[4], t1[4];  /* texels */
   const GLfloat f = myFrac(lambda);
   sample_2d_linear( tObj, tObj->Image[level  ], s, t, t0 );
   sample_2d_linear( tObj, tObj->Image[level+1], s, t, t1 );
   rgba[RCOMP] = (GLchan) (GLint) ((1.0F-f) * t0[RCOMP] + f * t1[RCOMP]);
   rgba[GCOMP] = (GLchan) (GLint) ((1.0F-f) * t0[GCOMP] + f * t1[GCOMP]);
   rgba[BCOMP] = (GLchan) (GLint) ((1.0F-f) * t0[BCOMP] + f * t1[BCOMP]);
   rgba[ACOMP] = (GLchan) (GLint) ((1.0F-f) * t0[ACOMP] + f * t1[ACOMP]);
}

 

                            Daniel Vogel, Programmer, Loki Software Inc.

Re: [Mesa3d-dev] s_texture.c

[Brian P. <br...@va...>]

Daniel Vogel wrote:
> 
> hi`
> 
> I tried running UT with the Mesa software renderer and when I turned on
> trilinear filtering it crashed in sample_2d_linear being called with a
> NULL pointer for the image from sample_2d_linear_mipmap_linear. The
> problem seems to be that UT uses GL_TEXTURE_MAX_LEVEL and tObj->_P was 7
> and tObj->_M was 5. From what I can read in mtypes.h shouldn't _M be
> used instead of _P in the following code? Changing the code to use _M at
> least made it work ;) BTW, why is _M a float and not an integer?
> 
> s_texture.c:840 [sample_2d_linear_mipmap_linear]
> 
> if (level >= tObj->_P) {
>    sample_2d_linear( tObj, tObj->Image[tObj->_P], s, t, rgba);
> }
> else {
>    GLchan t0[4], t1[4];  /* texels */
>    const GLfloat f = myFrac(lambda);
>    sample_2d_linear( tObj, tObj->Image[level  ], s, t, t0 );
>    sample_2d_linear( tObj, tObj->Image[level+1], s, t, t1 );
>    rgba[RCOMP] = (GLchan) (GLint) ((1.0F-f) * t0[RCOMP] + f * t1[RCOMP]);
>    rgba[GCOMP] = (GLchan) (GLint) ((1.0F-f) * t0[GCOMP] + f * t1[GCOMP]);
>    rgba[BCOMP] = (GLchan) (GLint) ((1.0F-f) * t0[BCOMP] + f * t1[BCOMP]);
>    rgba[ACOMP] = (GLchan) (GLint) ((1.0F-f) * t0[ACOMP] + f * t1[ACOMP]);
> }


Are you settting GL_TEXTURE_BASE_LEVEL?

I think I see an error in the calculation in _P.  I think I should
add the base level to it.  However, that would increase _P.

I'll have to review the spec and write a test program.  I do think
there is a bug here.  Thanks for flagging it.

-Brian

Re: [Mesa3d-dev] s_texture.c

[Daniel V. <vo...@lo...>]

On Thu, Dec 14, 2000 at 10:16:58AM -0700, Brian Paul wrote:
> 
> Are you settting GL_TEXTURE_BASE_LEVEL?

No, I only set GL_TEXTURE_MAX_LEVEL and cut off the last two mip levels.
It also looks like the filters use the wrong base mipmap. Everything
looks overblurry from an angle and a naive guess would be that it's
taking everything 2 levels too far down.

> I think I see an error in the calculation in _P.  I think I should
> add the base level to it.  However, that would increase _P.

from mtypes.h:

   GLint _P;			/* Highest mipmap level */
   GLfloat _M;			/* = MIN(MaxLevel, P) - BaseLevel */

What do _P and _M actually mean? Is _P the highest mip level based on
the size of the image (e.g. 5 for 32x32) and _M supposed to be the one
after "clipping"?



                            Daniel Vogel, Programmer, Loki Software Inc.

Re: [Mesa3d-dev] s_texture.c

[Brian P. <br...@va...>]

Daniel Vogel wrote:
> 
> On Thu, Dec 14, 2000 at 10:16:58AM -0700, Brian Paul wrote:
> >
> > Are you settting GL_TEXTURE_BASE_LEVEL?
> 
> No, I only set GL_TEXTURE_MAX_LEVEL and cut off the last two mip levels.
> It also looks like the filters use the wrong base mipmap. Everything
> looks overblurry from an angle and a naive guess would be that it's
> taking everything 2 levels too far down.
> 
> > I think I see an error in the calculation in _P.  I think I should
> > add the base level to it.  However, that would increase _P.
> 
> from mtypes.h:
> 
>    GLint _P;                    /* Highest mipmap level */
>    GLfloat _M;                  /* = MIN(MaxLevel, P) - BaseLevel */
> 
> What do _P and _M actually mean? Is _P the highest mip level based on
> the size of the image (e.g. 5 for 32x32) and _M supposed to be the one
> after "clipping"?

I should rename _P to _q (which is what the 1.2 spec uses now).  q is the
highest-numbered mipmap level that can actually be used, after taking in
account the image sizes and the value of GL_TEXTURE_MAX_LEVEL.

_M is the max usable lambda value.  It should be = _MaxLevel - BaseLevel.
If you're looking at equation 3.21 in the 1.2 spec, then _M = q - b.  I
should probably rename that variable too (_MaxLambda).

The incoming lambda value is added to BaseLevel to select a mipmap level.

The idea is that lambda=0 will correspond to the BaseLevel texture
and lambda = _M will correspond to the GL_MAX_LEVEL texture.

I'm not sure why you're seeing blurry images.  I haven't tried running
UT with 3.5 yet.

-Brian

Re: [Mesa3d-dev] s_texture.c

[Daniel V. <vo...@lo...>]

On Thu, Dec 14, 2000 at 12:49:45PM -0700, Brian Paul wrote:
> 
> I should rename _P to _q (which is what the 1.2 spec uses now).  q is the
> highest-numbered mipmap level that can actually be used, after taking in
> account the image sizes and the value of GL_TEXTURE_MAX_LEVEL.
> 
> _M is the max usable lambda value.  It should be = _MaxLevel - BaseLevel.
> If you're looking at equation 3.21 in the 1.2 spec, then _M = q - b.  I
> should probably rename that variable too (_MaxLambda).
> 
> The incoming lambda value is added to BaseLevel to select a mipmap level.
> 
> The idea is that lambda=0 will correspond to the BaseLevel texture
> and lambda = _M will correspond to the GL_MAX_LEVEL texture.

Thanks.

> I'm not sure why you're seeing blurry images.  I haven't tried running
> UT with 3.5 yet.

Most noticeable with the floor textures in DM-Stalwart.



                            Daniel Vogel, Programmer, Loki Software Inc.

Re: [Mesa3d-dev] s_texture.c

[Brian P. <br...@va...>]

Daniel Vogel wrote:
> 
> On Thu, Dec 14, 2000 at 12:49:45PM -0700, Brian Paul wrote:
> >
> > I should rename _P to _q (which is what the 1.2 spec uses now).  q is the
> > highest-numbered mipmap level that can actually be used, after taking in
> > account the image sizes and the value of GL_TEXTURE_MAX_LEVEL.
> >
> > _M is the max usable lambda value.  It should be = _MaxLevel - BaseLevel.
> > If you're looking at equation 3.21 in the 1.2 spec, then _M = q - b.  I
> > should probably rename that variable too (_MaxLambda).
> >
> > The incoming lambda value is added to BaseLevel to select a mipmap level.
> >
> > The idea is that lambda=0 will correspond to the BaseLevel texture
> > and lambda = _M will correspond to the GL_MAX_LEVEL texture.
> 
> Thanks.
> 
> > I'm not sure why you're seeing blurry images.  I haven't tried running
> > UT with 3.5 yet.
> 
> Most noticeable with the floor textures in DM-Stalwart.


I just checked in some changes.  Want to give it a try again?

-Brian

Re: [Mesa3d-dev] s_texture.c

[Daniel V. <vo...@lo...>]

On Thu, Dec 14, 2000 at 01:29:32PM -0700, Brian Paul wrote:
> 
> I just checked in some changes.  Want to give it a try again?

Fixes the segfault but is still blurry.



                            Daniel Vogel, Programmer, Loki Software Inc.