From a7b9a7901d99708da7ada09482c71aa456df24f0 Mon Sep 17 00:00:00 2001 From: Luke Kenneth Casson Leighton Date: Sat, 24 Aug 2019 01:09:42 +0100 Subject: [PATCH] add open 3d alliance slides --- .../open_3d_alliance_2019aug26.tex | 224 ++++++++++++++++++ 1 file changed, 224 insertions(+) create mode 100644 simple_v_extension/open_3d_alliance_2019aug26.tex diff --git a/simple_v_extension/open_3d_alliance_2019aug26.tex b/simple_v_extension/open_3d_alliance_2019aug26.tex new file mode 100644 index 000000000..f3c042baa --- /dev/null +++ b/simple_v_extension/open_3d_alliance_2019aug26.tex @@ -0,0 +1,224 @@ +\documentclass[slidestop]{beamer} +\usepackage{beamerthemesplit} +\usepackage{graphics} +\usepackage{pstricks} + +\title{Open 3D Alliance RISC-V} +\author{Luke Kenneth Casson Leighton} + + +\begin{document} + +\frame{ + \begin{center} + \huge{Open 3D Alliance: RISC-V}\\ + \vspace{32pt} + \Large{An open invitation to collaborate on 3D Graphics}\\ + \Large{Hardware and Software}\\ + \Large{for mobile, embedded, and innovative purposes}\\ + \vspace{24pt} + \Large{With thanks to Pixilica, GoWin, and Western Digital}\\ + \vspace{16pt} + \large{\today} + \end{center} +} + + +\frame{\frametitle{Why collaborate?} + + \begin{itemize} + \item 3D is hard. It's also not the same as HPC\vspace{15pt} + \item NVIDIA, AMD, Imagination - cannot meet "unusual" needs\vspace{15pt} + \item Working together on flexible standards, everyone wins\vspace{15pt} + \item Without collaboration: 10-20 man-years development\vspace{10pt} + \item With collaboration: cross-verification (avoids mistakes) + \end{itemize} +} + + +\frame{\frametitle{What is the goal?} + + \begin{itemize} + \item You get to decide! No, really!\vspace{12pt} + \item Outlined here: some ideas and cost/time-saving approaches\vspace{12pt} + \item Two new platforms: 3D "Embedded", 3D "UNIX"\vspace{12pt} + \item Flexible optional extensions (Transcendentals, Vectors,\\ + Texturisation, Pixel/Z-Buffers - all optional)\vspace{12pt} + \item Good software support absolutely essential\\ + (basically, that means Vulkan)\vspace{15pt} + \end{itemize} +} + +\frame{\frametitle{Libre RISC-V Team} + + \begin{itemize} + \item Small team, sponsored by Purism and the NLNet Foundation\vspace{8pt} + \item Therefore, focus is on efficiency: leap-frogging ahead\\ + without requiring huge resources.\vspace{8pt} + \item OpenGL API? Gallium3D / Vulkan is better\vspace{8pt} + \item Gallium3D turns out to be a single-threaded interpreter\\ + (Vulkan is compiled, and can be parallelised)\vspace{8pt} + \item Independent teams have provided OpenGL to Vulkan adaptors\vspace{8pt} + \item Same approach on hardware: seek highest bang-per-buck\\ + Save design time, save implementation time\vspace{8pt} + \end{itemize} +} + +\frame{\frametitle{What (optional) things are needed?} + + \begin{itemize} + \item Vectorisation. (SIMD? RVV? Other?)\vspace{12pt} + \item Transcendentals (SIN, COS, EXP, LOG)\vspace{12pt} + \item Texture opcodes, Pixel/Z-Buffers\vspace{12pt} + \item Pixel conversion (YUV/RGB etc.)\vspace{12pt} + \item Optional accuracy (embedded space needs less accuracy)\vspace{12pt} + \item Options give implementors flexibility. No imposition:\\ + imposition risks fragmentation (however, collaboration does\\ + need some hard easily-logically-justifiable rules) + \end{itemize} +} + +\frame{\frametitle{What is essential (not really optional)} + + \begin{itemize} + \item The software, basically. Anything other than Vulkan\\ + is a 10+ man-year effort\vspace{12pt} + \item Respecting that standards are hard to get right\\ + (and that consequences of mistakes are severe:\\ + no opportunity for corrections after a freeze)\vspace{12pt} + \item Respecting that, for collaboration and interoperability,\\ + some things go into a standard that you might not "need"\vspace{12pt} + \item Mutually respectful open and fully transparent collaboration.\\ + No NDAs, no "closed forums". We need the help of experts\\ + (such as Mitch Alsup) in this highly technical specialist area. + \end{itemize} +} + +\frame{\frametitle{What has our team done already?} + + \begin{itemize} + \item Decided to go the "Hybrid" Route (Separate GPUs requires a\\ + full-blown RPC/IPC mechanism to transfer all 3D API calls\\ + to and from userspace memory to GPU memory... and back). + \item Developed Simple-V (a "Parallelising API)\\ + (Simple-V is very hard to describe, because it is unique:\\ + there is no common Computer Science terminology) + \item Started on Kazan (a Vulkan SPIR-V to LLVM compiler) + \item Started work on a highly flexible IEEE754 FPU + \item Started work on a "Precise" CDC 6600 style OoO Engine,\\ + with help from Mitch Alsup, the designer of the M68000 + \item Variable-issue, predicated SIMD backend, Vector front-end\\ + "precise" exceptions, branch shadowing, much more + \end{itemize} +} + +\frame{\frametitle{Why Simple-V? Why not RVV?} + + \begin{itemize} + \item RVV is designed exclusively for supercomputing\vspace{8pt} + \item Like SIMD, RVV uses dedicated opcodes\\ + (google "SIMD considered harmful")\vspace{8pt} + \item 98\% of FP opcodes are duplicated in RVV. Large portion\\ + of BitManip opcodes duplicated in predicate Masks\vspace{8pt} + \item OP32 space is extremely precious: 48 and 64 bit opcode space\\ + comes with an inherent I-Cache power consumption penalty\vspace{8pt} + \item Simple-V "prefixes" scalar opcodes (all of them)\\ + No need for any new "vector" opcodes (at all).\\ + Can therefore use the RVV major opcode for 3D\vspace{8pt} + \end{itemize} +} + +\frame{\frametitle{Simple-V "Prefixing"} + + \begin{itemize} + \item SV "Prefix" does exactly that: takes RVC and OP32 opcodes\\ + and "prefixes" them with predication and a "vector" tag\vspace{8pt} + \item Three prefix types: SV P32 (prefixed RVC), P48 and P64\vspace{8pt} + \item Prefixed RVC takes 3 "Custom" OP32 opcodes.\\ + P48 takes standard OP32 scalar opcodes and "prefixes" them\\ + P64 adds additional vector context on top of P48\\ + \vspace{8pt} + \item "Prefixing" is a bit like SIMD. Vectors may be specified\\ + of length 2 to 4, elements may be "packed" into registers,\\ + opcode element widths over-ridden.\vspace{8pt} + \item Convenient, but not very space-efficient (and VBLOCK is)\vspace{8pt} + \end{itemize} +} + +\frame{\frametitle{VBLOCK Format} + + \begin{itemize} + \item Again: hard to describe. It is a bit like VLIW (only not really)\\ + A "block" of instructions is "prefixed" with register "tags"\\ + which give extra context to scalar instructions within the block + \item Sub-blocks include: Vector Length, Swizzling, Vector/Width\\ + overrides, and predication. All this is added to scalar opcodes!\\ + \textbf{There are NO vector opcodes} (and no need for any) + \item In the "context", it goes like this: "if a register is used\\ + by a scalar opcode, and the register is listed in the "context",\\ + SV mode is "activated" + \item "Activation" results in a hardware-level "for-loop" issuing\\ + \textbf{multiple} contiguous scalar operations (instead of just one). + \item Implementors are free to implement the "loop" in any fashion\\ + they see fit. SIMD, Multi-issue, single-execution: anything. + \end{itemize} +} + +\frame{\frametitle{Other Standard Proposals} + + \begin{itemize} + \item Ztrans and Ztrig* - Transcendentals and Trigonometrics\\ + (optional so that Embedded implementors have some leeway) + \item ISAMUX / ISANS - stops arguments over OP32 space\\ + (also allows clean "paging" of new opcodes into e.g. RVC) + \item MV.SWIZZLE and MV.X - RV does not have a MV opcode. + \item Zfacc - dynamic FP accuracy. Needed for "fast" Vulkan native\\ + and to switch between fast 3D accuracy and IEEE754 modes. + \item These - and more - need your input! 3D is hard! + \item The key strategic premise: these are required as \textbf{public}\\ + standards, because the \textbf{software} is to be public. + \item This is \textbf{not} understood by the RISC-V Foundation.\\ + ("custom" status not appropriate for high-profile mass-volume\\ + end-user APIs such as Vulkan). + \end{itemize} +} + + +\frame{\frametitle{Summary} + + \begin{itemize} + \item 3D is hard (and pure Vectorisation gets you 25\% of \\ + commercially-acceptable performance). + \item Layered optional extensions are going to be key to\\ + acceptance by a wide variety of 3D Alliance Members. + \item With a custom specialised SPIR-V (Vulkan) Compiler\\ + being an absolutely critical strategic requirement,\\ + RVV and its associated compiler (still not developed)\\ + is, surprisingly, of marginal value. + \item Question everything! Your input, and a willingness to\\ + take active responsibility for tasks that your Company\\ + is critically dependent on, are extremely important. + \item Public and transparent Collaboration is key. There is simply\\ + too much to do. + \end{itemize} +} + + +\frame{ + \begin{center} + {\Huge \vspace{20pt} + The end\vspace{20pt}\\ + Thank you\vspace{20pt}\\ + } + \end{center} + + \begin{itemize} + \item http://lists.libre-riscv.org/pipermail/libre-riscv-dev/ + \item http://libre-riscv.org/simple\_v\_extension/abridged\_spec/ + \item https://libre-riscv.org/ztrans\_proposal/ + \item https://libre-riscv.org/simple\_v\_extension/specification/mv.x/ + \end{itemize} +} + + +\end{document} -- 2.30.2