GPGPU Programming for Games and Science

GPU Computing 153

mov o0 .w, v2 . x

mov o1 . xyz , v1 . xy zx

ret

// Approximately 4 instruction slots used

LISTING 4.9: The output assembly listing for the vertex shader of

Billboards.hlsl.

The input and output signatures are similar to what we have seen before,

except that the HLSL compiler has taken the liberty to optimize the output.

The

POSITION0 and TEXCOORD0 outputs are stored in the same register,

the position in the first three components (

xyz) and the billboard size in the

last component (

w). When the geometry shader consumes the vertex shader

output, only two registers per vertex are fetched rather than three if the

compiler had decided not to