glTexImage2D vs. gluBuild2DMipmaps

问题

Very basic OpenGL texture creation code:

int width, height;
BYTE * data;
FILE * file;
// open texture data
file = fopen( filename, "rb" );
if ( file == NULL ) return 0;
// allocate buffer
width = 256;
height = 256;
data =(BYTE*) malloc( width * height * 3 );
// read texture data
fread( data, width * height * 3, 1, file );
fclose( file );
glGenTextures( 1, &texture );
glBindTexture( GL_TEXTURE_2D, texture );
//gluBuild2DMipmaps( GL_TEXTURE_2D, 3, width,   height, GL_RGB, GL_UNSIGNED_BYTE, data );
glTexImage2D( GL_TEXTURE_2D,0, GL_RGB, width,   height,0, GL_RGB, GL_UNSIGNED_BYTE, data );
free( data );
return texture;

and rendering:

glEnable( GL_TEXTURE_2D );
glBindTexture( GL_TEXTURE_2D, texture );

glPushMatrix();
glRotatef( theta, 0.0f, 0.0f, 1.0f );
glBegin( GL_QUADS );
glTexCoord2d(0.0,0.0); glVertex2d(-1.0,-1.0);
glTexCoord2d(1.0,0.0); glVertex2d(+1.0,-1.0);
glTexCoord2d(1.0,1.0); glVertex2d(+1.0,+1.0);
glTexCoord2d(0.0,1.0); glVertex2d(-1.0,+1.0);
glEnd();
glPopMatrix();

SwapBuffers( hDC );

Wen using glTexImage2D, nothing draws but with gluBuild2DMipmaps works I saw result with gDebugger the texture created correctly. What is the problem?

回答1:

When you use glTexImage2D (...) in your code you do not build a mipmap complete texture. It only creates storage and supplies data for texture LOD 0. If you are using GL_..._MIPMAP_... as your minification filter you need to have LODs (levels-of-detail) for every quarter-resolution step.

For example, a texture with dimensions 32x32 requires log₂ 32 = 5 extra mipmap LODs, as described below:

LOD 0: 32x32  ( 1 /    1 )  [1024 Texels]
LOD 1: 16x16  ( 1 /    4 )  [ 256 Texels]
LOD 2: 8x8    ( 1 /   16 )  [  64 Texels]
LOD 3: 4x4    ( 1 /   64 )  [  16 Texels]
LOD 4: 2x2    ( 1 /  256 )  [   4 Texels]
LOD 5: 1x1    ( 1 / 1024 )  [   1 Texel ]

`gluBuild2DMipmaps (...)` does a couple of things:

It builds the set of quarter-resolution mipmap LODs (hence the name).
It properly sets the number of LODs in your texture
- Default LOD level range in an OpenGL Texture: 1001 (min = 0, max = 1000)
- After gluBuild2DMipmaps (...): log₂ (max (Res_x, Res_y)) + 1 (min = 0, max = ...).

This produces a mipmap complete texture, that can be used with a mipmap minification filter.

Miscellaneous things to note:

The default minification filter in OpenGL is: GL_NEAREST_MIPMAP_LINEAR, so you need a mipmap complete texture unless you change it to GL_LINEAR or GL_NEAREST.

LOD indices in OpenGL work somewhat counter-intuitively. Lower numbers represent higher resolution images, and this is why using a negative LOD bias is sometimes referred to as "texture sharpening."

The geometric series: 1 + ¹/₄ + ¹/₁₆ + ¹/₆₄ + ... + ¹/_N converges to ⁴/₃; mipmaps require ~33% extra storage.