PlayStation 2 Clipping

Source: playstation2-linux.com
Topic: Playstation 2

Short Desciption:
PlayStation 2 Clipping. Colin Hughes. Sony Computer Entertainment Europe .... lq.xyz vf82+12(vi1). lq.xy vf130+6(vi1). iaddi vi1vi13. div Qvf0wvf11w ...

 

Content Inside:
PlayStation 2 ClippingPage 1makebettergamesPlayStation 2 ClippingColin HughesSony Computer Entertainment EuropePage 2makebettergamesStandard graphics pipeline Vertices passed to T&L unit Transformed to clip space Hardware clips to viewport Programmer has no hassles. Not the case with PS2Page 3makebettergamesWhy no clipping Clipping in hardware is very expensive. At least one FP divider and vector multiplier required Up to 3 for full pipelined hardware relying on symmetry to clip +& together This hardware lies idle for a great deal of the time PS2 designed to maximize use of hardware Have software handle infrequent casesPage 4makebettergamesPS2 HardwarePage 5makebettergamesPS2 Overview MIPs CPU core 128 bit register set multimedia extensions Vector Units SIMD FP processors operate on 4xSP Individual data & code memory Can run independently of core cpu VU0 closely tied to core VU1 closely tied to GSPage 6makebettergamesPS2 Overview Graphics Synthesisor Rasteriser with 4MB embedded memory Optimised for high fill rate Very simple. No processing of vertex data Only 2D scissoring supported This means work for the programmer!!Page 7makebettergamesInternal datapathsEE CoreVU0VU1VIF0VIF1GIFI$D$SPTimersDMAcontro llerRambusinterfaceSIFIPUPage 8makebettergamesVU1VIF1GIFRambusinterfaceF or a simple renderer all weare interested in are thefollowing componentsPage 9makebettergamesPS2 simple graphics pipeline Geometry sent to VU1 via DMA along with all state information. VU1 applies all transform lighting and clipping. Back to old days.. Polys sent directly from VU1 to GS Special internal link from VU1 memory to GIF.Page 10makebettergamesSimple transform Without worrying about clipping transform is vector matrix operation followed by homogeneous divideMul ACCvf31vf1wMadd ACCvf30vf1zMadd ACCvf29vf1yMadd vf2vf28vf1xdiv Qvf0wvf2w(vf0w = 1.0 )Mul vf2vf2QPage 11makebettergamesHow does this compare? 6 instructions 5 multiples and one divide Compare to 4 for Vertex shader no ...