Difference between revisions of "NV2A/Vertex Shader"

Revision as of 21:29, 21 April 2017

The Xbox implements the 2 GL extensions NV_vertex_program and NV_vertex_program1_1 (with some modifications). This article will mainly focus on actual encoding on hardware as the behaviour is mostly outlined in those GL extensions already.

Operating modes

• Fixed / Programmable
• Writeable / Read-Only constants
• Low-constants only / all constants
• Vertex processing / State program

Registers

Input registers

There are 16 input registers v[0] to v[15].

They normally map to the vertex attributes. However, in the case of vertex state programs, v[0] is fed from LAUNCH_DATA (PGRAPH Methods 0x1E80, 0x1E84, 0x1E88, 0x1E8C for XYZW respectively) instead.

Output registers

output registers: o[0] to o[?] (initialized to XYZ=0x00000000 W=0x3F800000) The alias names from the GL extension match.

Address register

A0.x exists as documented in the GL extension.

Temporary registers

There are 12 temporary registers: R0 to R11 (initialized to XYZW=0x00000000), as documented in the GL extension. Additionally, o[POS] is mirrored as R12 and can be used as source operand; so effectively you have 13 temporaries

Constant space

There are 192 constant registers in two seperate blocks with 96 constants each. They can be accessed through the PGRAPH RDI: select=0x17. Each constant slot is 4x DWORD, ordered as WZYX. Alternatively they can be uploaded through PGRAPH method [FIXME], with 4x DWORD, ordered XYZW.

In nvidia vertex programs only 96 constants are normally accessible. Microsoft exposed the 96 additional constant registers in D3D shaders through c[-96] to c[-1]. This documentation uses the GL terminology instead and expose the new registers as c[96] to c[191]. This means c[0] to c[191] valid.

Instructions

In total, there are 136 instruction slots.

Each slot consists of 16 bytes, we consider those as 4 seperate little-endian DWORDS describing the operation. Word 0 is inused.