I'm writing some application which displays lines of different thickness using D3D 12. Lines are drawn as quads (two triangles). In order to reduce upload bandwidth to GPU i'm not constructing vertices on CPU. The vertex buffer is empty, instance buffer includes two float2 paramaters (line points). In vertex shader i use SV_VERTEXID to construct vertices from instance buffer. The problem is i want to overlap instance buffer stride by one point. That is, input signature includes two float2 variables instance data and instance buffer stride is only one float2 variable. This works perfectly on WARP devices, but on AMD GPU it does not work.