What is the reason behind a triangle's width drawn by GPU occupying 15 pixels instead of 16 pixels?

I want to draw a triangle in a 160*160 color attachment. the coordination is

-1.0,1.0,0.0
-0.9,0.9,0.0
-0.8,1.0,0.0

In my opinion, the width of triangle is 0.2, so it should occupy 16 pixels.
but the result is 15.

I want to know how GPU process this situation? Can i find some algorithm or formula？

The edges of your triangle pass exactly through pixel centers (at half-integer coordinates). Therefore whether those pixels on the edge belong to the triangle or not is not well determined.

One applicable rule is that triangles that share an edge must rasterize the pixels on that edge exactly once, in order to prevent artifacts with blending, among others. The way an implementation would do that is to include the left edges and exclude the right edges (for example) when rasterizing each triangle individually.

Note that this applies only to the tie-breaking when the sample is exactly on the edge. If you raised your triangle just by quarter of a pixel, all samples would be unambiguously inside or outside the triangle, and the apparent width of it would be as you'd expect.

Here's a diagram for clarity, where the triangle is drawn onto a 60x60 framebuffer. The diamonds indicate the location of the samples within each pixel. The white diamonds lay outside the triangle and thus aren't shaded. The black diamonds are determined to be within the triangles and are therefore shaded. The red diamonds happen to be exactly on the right edge, and are considered to be outside of the triangle as explained earlier, so they aren't shaded either. The result is that the triangle is only five pixels in width on the screen rather than six that you'd expect: