NL.net proposal

Project name

Port of AMDVLK/RADV 3D Driver to the Libre RISC-V SoC

Website / wiki

https://libre-riscv.org/nlnet_2019_amdvlk_port

Please be short and to the point in your answers; focus primarily on the what and how, not so much on the why. Add longer descriptions as attachments (see below). If English isn't your first language, don't worry - our reviewers don't care about spelling errors, only about great ideas. We apologise for the inconvenience of having to submit in English. On the up side, you can be as technical as you need to be (but you don't have to). Do stay concrete. Use plain text in your reply only, if you need any HTML to make your point please include this as attachment.

Abstract: Can you explain the whole project and its expected outcome(s).

The Libre RISCV SoC is being developed to provide a privacy-respecting modern processor, developed transparently and as libre to the bedrock as possible. As a hybrid processor, it is intended to be both a CPU and a GPU. GPUs are normally proprietary (and thus are perfect candidate attack vectors), as is the 3D driver software.

In January 2019, AMD released a fully-functioning libre-licensed Vulkan Driver for their Radeon GPUs. A close examination of its source code shows that it would be relatively straightforward to replace the libraries that generate Radeon GPU assembly code with ones that generate assembly for the Libre RISC-V SoC, instead.

In addition, further investigation shows that RADV, the libre-licensed MESA 3D Driver, also supports SPIR-V (by way of conversion to MESA NIR), and, likewise, may be a good candidate for replacing Radeon with Libre RISC-V assembly.

Thus we intend to do exactly that: leverage the excellent work already done to create a libre-licensed commercial-grade Vulkan 3D driver that takes full advantage of the parallelism and Vectorisation in the hybrid Libre RISC-V SoC.

Have you been involved with projects or organisations relevant to this project before? And if so, can you tell us a bit about your contributions?

Luke Leighton is an ethical technology specialist who has a consistent 24-year track record of developing code in a real-time transparent (fully libre) fashion, and in managing Software Libre teams. He is the lead developer on the Libre RISC-V SoC.

Jacob Lifshay is a software libre 3D expert who developed a Vulkan 3D software render engine under the GSoc2017 Programme. He also developed his own libre-licensed 32-bit RISC-V processor, and has written an optimising javascript compiler. Jacob is a valuable member of the team and is working on Kazan (https://salsa.debian.org/Kazan-team/kazan)

Requested Amount

EUR 50,000.

Explain what the requested budget will be used for?

After a thorough and comprehensive evaluation to see which will be the best to choose (RADV or AMDVLK), we are aiming for a multi-stage process, starting with the basics:

  • The first stage is to remove AMD's "PAL" Library in AMDVLK, or the AMDGPU engine used in RADV, and replace it with a straightforward upstream port of the current LLVM JIT compiler, alongside a "support" library that will call OpenCL / OpenGL functions directly on the main processor. This "effectively" turns the engine into a peer of google swiftshader (a "Software 3D Renderer") which will allow us to carry out rapid testing on stable x86 systems before moving on to the next stage.
  • The second stage is to confirm that the standard RISC-V LLVM JIT (which was recently upstreamed as of LLVM 9.0.0) is properly functional under an emulator or other RV64GC system.
  • The third phase will be to begin the iterative process, tying in closely with the work on Kazan, to experiment in both a software simulator as well as in FPGAs, with the addition of both Vectorisation as well as custom opcodes that will significantly improve performance as well as meet commercially-acceptable power-performance demands.

At the point where commercial power-performance requirements are met we may officially declare the project a "success".

Does the project have other funding sources, both past and present?

The overall project has sponsorship from Purism as well as a prior grant from NLNet. However that is for specifically covering the development of the RTL (the hardware source code), and for a "parallel" 3D Vulkan Driver effort, Kazan.

Kazan is to be written in Rust and its inclusion in the initial 2018 proposal was before AMDVLK was released (Jan 2019). AMDVLK is written in c++, is a significant way along (further than Kazan), however there is a different design focus in each that makes choosing one over the other not only difficult but potentially a costly mistake.

Compare your own project with existing or historical efforts.

Nyuzi is a Software-based 3D Engine that has an LLVM port. The problem is that it has deliberately been designed to be a software-only Vector Processor. As such, with no custom accelerated opcodes dedicated to 3D, its power-performance metric is a whopping 25% that of commercially-acceptable 3D GPUs. It also has no actual 3D Vulkan Driver: the developers focussed only on the "core algorithms" as part of an (extremely useful) academic exercise, only.

Google's swiftshader is a software-based 3D Driver/Engine that is compatible with at least one version of Vulkan. On the face of it, this would be a perfect match for the Libre RISC-V SoC due to it being a hybrid CPU / GPU. The problem is that swiftshader was designed - from the ground up - never to have Vectorisation or any form of accelerated hardware beyond SIMD (NEON, Altivec, SSE/AVX). As Nyuzi clearly shows, this approach is known to give a massive 400% power penalty. Not only that, but our additions would not be welcome due to the primary focus of swiftshader being on non-hardware-accelerated, non-custom processors.

RADV is the free software competitor to AMDVLK. It takes a different route: converting SPIR-V to NIR (New Internal Representation) which will need close evaluation to ensure that it's directly suited to Vector Processing. Like AMDVLK, it does not directly support RISC-V: it was purely intended to support Radeon GPUs.

Our initial proposal - Kazan - is much more interesting to discern and compare against. Kazan is being specifically designed so that the SPIR-V compiler is capable of fully supporting "full-function vectorisation". LLVM IR does NOT normally support this (which is why SPIR-V was created by the Khronos Group in the first place). However, AMDVLK, which is a hard fork of LLVM, has had its LLVM-IR specifically modified to support both full-function vectorisation, predication, and in addition, texturisation, such that this information may be "carried" down to the Radeon assembly level, through the PAL library.

Standard LLVM does not support this full-function vectorisation capability: it is typically left up to any given assembly-level converter (such as the RISC-V Vector Engine) to "opportunistically" turn non-vectorised programs into vectorised ones, whereas AMDVLK explicitly carries this very same information.

Kazan on the other hand intends to perform explicit Vectorisation code-transformations in a different location: inside the SPIR-V compiler itself. This key radical technical difference is why we seek to explore the alternative approach taken by AMD side-by-side with that of Kazan, because it is just not possible to predict in advance which would be "better".

What are significant technical challenges you expect to solve during the project, if any?

This is compiler technology, which is traditionally viewed as particularly challenging. We are slightly fortunate in that much of the pieces of the puzzle already exist: AMDVLK, RADV, the upstreamed acceptance of RISC-V LLVM 9.0.0 being the key ones.

Whilst we know technically what they did and why they did it, the key challenge will be to unravel what exact changes AMD made which caused them to have to "fork" LLVM several years back, to keep track of their efforts to introduce "mainline" LLVM patches on an ongoing piecemeal basis, and at the same time add our own assembler back-end into the same fast-moving target.

Whereas with RADV it is upstreamed in MESA, and has much wider community support, it will need very careful detailed evaluation to ensure that it meets the needs of the Libre RISC-V Vector Engine.

Describe the ecosystem of the project, and how you will engage with relevant actors and promote the outcomes?

As mentioned in the 2018 submission, the Libre RISC-V SoC has a full set of resources for Libre Project Management and development: mailing list, bugtracker, git repository and wiki - all listed here: https://libre-riscv.org/

In addition, we have a Crowdsupply page https://www.crowdsupply.com/libre-risc-v/m-class which provides a public gateway, and heise.de, reddit, phoronix, slashdot and other locations have all picked up the story. The list is updated and maintained here: https://libre-riscv.org/3d_gpu/

Extra info to be submitted