配置OpenGL纹理过滤参数后，着色器访问纹理时如何选择过滤模式？

Great question! Let's break this down clearly, regardless of whether you're sampling textures in a vertex or fragment shader.

Core Decision Logic #

The key takeaway is: OpenGL selects the filter mode based on the relative size of a texel (texture pixel) versus the screen area (or shader sample area) it maps to—not which shader you're sampling from.

Here's the simple rule:

• When the screen area covered by a single texture coordinate sample is smaller than one texel (you’re magnifying the texture, e.g., stretching a 100x100 texture over a 200x200 quad), OpenGL uses your GL_TEXTURE_MAG_FILTER setting.
• When that area is larger than one texel (you’re minifying the texture, e.g., shrinking a 100x100 texture onto a 50x50 quad), OpenGL uses your GL_TEXTURE_MIN_FILTER setting.

How OpenGL Calculates This (Across Both Shaders) #

This all relies on texture coordinate differentials—dFdx and dFdy—which measure how much texture coordinates change between adjacent screen samples. OpenGL uses these values to calculate the "footprint" of the current sample in texture space.

• Fragment Shader Sampling: Hardware directly tracks texture coordinate changes between neighboring fragments, so the differential calculation is precise. The footprint matches the actual size of the fragment on screen.
• Vertex Shader Sampling: Even if you sample the texture in the vertex shader (before rasterization), OpenGL approximates the necessary differentials using the vertex’s screen position changes. It estimates how the texture coordinate would spread across the primitive being rendered, letting it still determine if magnification or minification applies to that vertex’s sample.

Quick Extra Notes #

• GL_TEXTURE_MIN_FILTER often supports mipmapping (e.g., GL_LINEAR_MIPMAP_LINEAR), since minification benefits from lower-resolution mip levels. GL_TEXTURE_MAG_FILTER never uses mipmaps—there’s no need to downsample when you’re already enlarging the texture.
• If you force a specific mip level (using textureLod() instead of texture()), the filter mode choice still follows the magnification/minification rule—you just override which mip level gets used.

内容的提问来源于stack exchange，提问作者user3752629