updates/002_201830nov_phoronix_articles.mdwn

   1 Many thanks to Michael Larabel, who has been writing early articles on
   2 this project even before we had a chance to set up this pre-launch
   3 page. What follows are some of my observations and responses about the
   4 articles.
   5
   6 #### List of Phoronix Articles
   7
   8 For the last few months, Michael has been covering various aspects of
   9 this project. The first article covered a lot of the technical
  10 details, the second article covered an announcement of Kazan, which
  11 implements the Vulkan 3D API, and the most recent article picks up on
  12 [the first project
  13 update](why-make-a-quad-core-64-bit-soc-surely-there-are-enough-already):
  14
  15 - [There's A New Libre GPU Effort Building On RISC-V, Rust, LLVM & Vulkan](https://www.phoronix.com/scan.php?page=news_item&px=Libre-GPU-RISC-V-Vulkan) (posted 28 September 2018)
  16 - [The Kazan Vulkan CPU/Software-Based Implementation Being Rewritten In Rust](https://www.phoronix.com/scan.php?page=news_item&px=Kazan-Vulkan-Rust) (posted 04 October 2018)
  17 - [The EOMA68 Libre Computer Developer Wants To Tackle A Quad-Core RISC-V Libre SoC Design](http://www.phoronix.com/scan.php?page=news_item&px=Quad-Core-Libre-SoC-Proposal) (posted 29 November 2018)
  18
  19 There has been quite a lot going on, including an enormous amount of
  20 planning for nearly six to eight months, so there are quite
  21 a few catch-up updates to write.  It's worthwhile doing one that incorporates
  22 responses to Michael and to some of the people who also kindly asked
  23 questions and made comments on the
  24 [Phoronix Forum](https://www.phoronix.com/forums/node/1064199).
  25
  26 #### Comments and Responses
  27
  28 I have no illusions about the cost of development of this project:
  29 it's going to be somewhere north of USD $6 million, with contingency
  30 of up to USD $10 million.  This is just how it is.  Interestingly,
  31 that means there's provision for both attracting investment and
  32 really, really good talent, and to properly pay for that talent. The
  33 project's origins reflect what can be achieved with the current
  34 resources. I've been kindly sponsored with a ZC706 FPGA board (worth
  35 over USD $2,500), which will allow one major hurdle to be cleared that
  36 will meet the criteria of many investors: making sure the design is
  37 FPGA proven.
  38
  39 Secondly, I note Michael is a bit incredulous of the goal of achieving
  40 mobile-class 3D performance.  It's actually extremely modest: 100
  41 million pixels per second, 30 million triangles per second, and around
  42 5 to 6 GFLOPs.  These statistics were taken from the benchmarks for
  43 Vivante's GC800.  Achieving these kind of numbers is dead easy.
  44 Achieving them within a power envelope of under 2.5 watts? Not so
  45 easy!
  46
  47 To that end, I spent a considerable amount of time speaking to Jeff
  48 Bush, who developed
  49 [Nyuzi](https://www.phoronix.com/scan.php?page=news_item&px=LGPL-GPGPU-NyuziProcessor).
  50 Jeff's work is fascinating and extremely valuable because, despite it
  51 being such low 3D peformance, the technical documentation and academic
  52 analysis of *why* that performance is so low is absolutely, absolutely
  53 critical.  The
  54 [paper](http://www.cs.binghamton.edu/~millerti/nyuziraster.pdf) that
  55 Jeff co-authored makes a comparison of software and hardware
  56 rasterisation, and he actually developed a fixed-function hardware
  57 renderer called [ChiselGPU](https://github.com/jbush001/ChiselGPU) in
  58 order to do the comparisons.
  59
  60 One fascinating insight that came out of Jeff's work was that just getting
  61 data through the L1/L2 cache has a massive impact on power consumption.
  62 A way to deal with that is to increase the number of registers in the
  63 design until such time as the data being processed (for example, a tile,
  64 or an inter-dependent four-wide bank of 4x3 floating-point numbers) can
  65 all fit into the register file, so as not to need to be pushed back down
  66 to the L1 cache and back.  Some GPUs have a "scratch RAM" area to deal
  67 with this.  Staggeringly, even for (or, especially for) a mobile-class
  68 GPU, we had to increase the register file size to a whopping 128 64-bit
  69 entries, that can be broken down into **256** 32-bit single-precision
  70 floating-point entries!  What are we *doing*?!  This is supposed to be
  71 a modest design!
  72
  73 A little digging around the Internet reveals that even mobile-class
  74 GPUs genuinely have this number of registers.  More than that, though,
  75 it turns out that we may have a hidden advantage through implementing
  76 Kazan as a Vulkan Driver.  In [this mailing list
  77 discussion](http://lists.libre-riscv.org/pipermail/libre-riscv-dev/2018-October/000065.html)
  78 which follows on from a proposal on the LLVM mailing list about making
  79 matrices a first-class type, Jacob informed me that the Vulkan API
  80 also passes in large batches of data that contain matrices, and also
  81 arrays of data structures that need to be processed together.  The
  82 problem here is that you want the elements of the arrays (or the
  83 matrix) to be processed as if they were linear, preferably without
  84 having to move them around.  Matrix multiplication, for example,
  85 typically requires the second matrix to be transposed (X swapped with
  86 Y) in order to access the elements in a linear fashion.  What we
  87 decided to do instead was to add [1D/2D/3D data
  88 shaping](http://lists.libre-riscv.org/pipermail/libre-riscv-dev/2018-October/000087.html).
  89 The elements *stay in-place* in the original registers, however a
  90 "remapper" engine makes them appear, as far as the parallel
  91 (SIMD/Vector) engine is concerned, as if the registers are contiguous.
  92 In the case of transposing, register numbers 0 1 2 3 / 4 5 6 7 / 8 9
  93 10 11 get "remapped" to 0 4 8 / 1 5 9 / 2 6 10 / 3 7 11 without the
  94 need for "MV" instructions.  Thanks to Mitch Alsup, the designer of
  95 the 66000 ISA, we learned that this re-invented wheel has [also been
  96 implemented](https://groups.google.com/d/msg/comp.arch/bGBeaNjAKvc/_vbqyxTUAQAJ)
  97 in production GPUs and vectorisation systems.
  98
  99 The point is, that by picking Vulkan and implementing both the hardware
 100 design and the software at the same time, we are both constrained
 101 *and guided* towards a successful design.  In addition, as we know from
 102 previous successful (truly) open projects, the very fact you are
 103 at liberty to talk about what you're doing (as compared to a secretive
 104 proprietary company) means that people with specialist expertise are
 105 more than happy to come forward and comment, and help guide you away
 106 from areas that have caught out billion-dollar companies in the past.
 107
 108 The project thus becomes a *synthesis of the expertise and efforts
 109 of much more than just the people who are implementing it*.
 110
 111 Just having the opportunity to do that is extremely humbling.  Mitch
 112 Alsup, the designer of the famous Motorola 68000 series of processors,
 113 is giving us some feedback and input! Like... wow!  For example, [he
 114 made an extremely valuable
 115 recommendation](https://libre-riscv.org/3d_gpu/microarchitecture/) on
 116 how to save on register file space, only needing 1R1W (1 x read-port,
 117 1 x write-port) SRAM, by stretching out the pipeline phases to load
 118 operands sequentially rather than in parallel.  I cannot express how
 119 grateful I am for his input, and for all the other people who have
 120 helped.
 121
 122 So, given this community, I believe we're in good hands.  Ultimately,
 123 what is being presented here is more of an opportunity for anyone who
 124 has wanted something like this to succeed (or even exist), empowering
 125 them to go from "I wish someone would do this" to "I can help make it
 126 happen."  This is one of the reasons why, if I am honest, I get
 127 slightly aggravated by people who write, "oh this project could not
 128 possibly succeed" or "this person could not possibly achieve this
 129 goal," as such comments entirely miss the point.
 130
 131 As this update is quite long, I'll answer more on the [Phoronix
 132 comments](https://www.phoronix.com/forums/forum/phoronix/latest-phoronix-articles/1064199-the-eoma68-libre-computer-developer-wants-to-tackle-a-quad-core-risc-v-libre-soc-design),
 133 however, I particularly wanted to address one comment, here:
 134 *"This religious war target seems to be way off based on skill set."*.
 135 As you can see above, I believe that one aspect of that comment has
 136 been addressed, above: as it's a libre project, unlike a proprietary
 137 company we're at liberty to get out on the Internet and ask peoples'
 138 advice before even committing to a design.
 139
 140 The second aspect is the very silly "religious war" implication.  It's
 141 absolutely nothing of the kind.  What many people do not know about me
 142 is that I very, very deliberately pick projects that nobody else is doing.
 143 I pick projects that make an ethical difference.  That, if successful,
 144 many peoples' lives would be much better, less painful.  It has absolutely
 145 sod-all to do with "religious frothing fervour" (foam, foam).
 146
 147 Many companies choose to make ethical compromises in order to make a profit.
 148 People are finally beginning to wake up to the consequences of this kind
 149 of concentration of financial and informational power.  In India,
 150 people have been *murdered* based on Whatsapp viral hear-say.  In the USA,
 151 democratic elections have been interfered with (e.g., Cambridge Analytica).
 152 I could very easily go to any of these massively-unethical corporations,
 153 and make an absolute fortune in the process of empowering them to do a
 154 hundred more Cambridge Analyticas.  *I choose not to do so.*
 155
 156 There are plenty of companies that make decisions without
 157 a moral compass, because, financially, it is easier to do so. And, more
 158 poignantly, it is legally permissible and *actively encouraged* by
 159 legal frameworks, tax incentives, and Government-sanction monopolies
 160 known by the name "patents."
 161
 162 What I am doing here is to demonstrate that none of that is necessary.
 163 That it is possible to design - and get funding for - a desirable
 164 product that happens also to be ethical.  This is why the goal is
 165 as it is: a mobile-class processor, because that's the kind of product
 166 that could sell in large volumes at around the USD $4 mark.
 167 You won't see any corporation taking on such a goal, as they're required
 168 to prioritise profits over ethics.  So, it's down to you,
 169 if you want this project to succeed, to help make it happen.