A new kind of git history

Discussions about rebase vs. merge are familiar territory for anybody with an interest in version control in general and git in particular. I want to finally give a more permanent home to an idea that I have expressed in the past and that I've occasionally seen others hint at in those discussions as well.

There are multiple camps in these discussions that have slightly different ideas about how and for what purposes git should be used.

The first major axis of disagreement is whether history needs to be git bisect-able. Outside of my own little hobby projects, I've always worked on projects for which bisectability was important. This has generally been because their scope was such that CI simply had no chance to cover all uses of the software. Bug reports that can be traced to regressions from weeks or even months ago are not frequent per se, but they have always been frequent enough to matter. git bisect is an essential tool for finding those regression points when they happen. Not all projects are like that, but for projects which are, the notion of an "atomic" change to the project's main development branch (or branches) is important.

The second major axis of disagreement is whether the development history of those "atomic" changes is important enough to preserve. The original git development workflow does not consider this to be important: developers send around and review multiple iterations of a change, but only the final version of the change goes into the permanent record of the git repository. I tend to agree with that view. I have very occasionally found it useful to go back and read through the comments on a pull request that lead to a change months ago (or the email thread in projects that use an email workflow), but I have never found it useful to look at older versions of a change.

Some people seem to really care about this kind of history, though. They're the people who argue for a merge-based workflow for pull requests on GitHub (but against force-pushes to the same) and who have built hacks for bisectability and readability of history like --first-parent. I'm calling that a hack because it does not compose well. It works for projects whose atomic change history is essentially linear, but it breaks down once the history becomes more complex. What if the project occasionally has a genuine merge? Now you'd want to apply --first-parent for most merge commits but not all. Things get messy.

One final observation. Even "my" camp, which generally prefers to discard development history leading up to the atomic change in a main development branch, does want to preserve a kind of history that is currently not captured by git's graph. git revert inserts the hash of the commit that was reverted into the commit message. Similarly, git cherry-pick optionally inserts the hash of the commit that was cherry-picked into the commit message.

In other words, there is a kind of history for whose preservation at least in some cases there seems to be a broad consensus. This kind of history is distinct from the history that is captured by commit parent links. Looked at in this light, the idea is almost obvious: make this history an explicit part of git commit metadata.

The gist of it would be this. Every commit has a (often empty) list of historical commit references explaining the origins of the diff that is implicitly represented by the commit; let's call them diff-parents. The diff-parents are an ordered list of references to commits, each of them with a "reverted" bit that can optionally be set.

The history of a revert can be encoded by making the reverted commit a diff-parent with the "reverted" bit set. The history of a cherry-pick can be encoded similarly, with the "reverted" bit clear. When we perform a simple rebase, each new commit has an obvious diff-parent. When commits are squashed during a rebase, the sequence of squashed commits becomes the list of diff-parents of the newly formed commit. GitHub users who like to preserve all development history can use the "squash" option when landing pull requests and have the history be preserved via the list of diff-parents. git commit --amend can similarly record the original commit as diff-parent.

This is an idea and not a fully fleshed-out plan. There are obviously a whole bunch of tricky questions to answer. For example: How does this all fit into git's admittedly often byzantine CLI? Can merge commits be diff-parents, and how would that work? Can we visualize the difference between a commit and its diff-parents? (Hint: Here's an idea)

Diff-parents are a source of potential information leaks. This is not a problem specific to the idea of diff-parents; it is a general problem with the idea of preserving all history. Imagine some developer accidentally commits some credentials in their local clone of a repository and then uses git commit --amend to remove them again. Whoops, the commit that contains the credentials is still referenced as a diff-parent. Will it (and therefore the credentials) be published to the world for all to see when the developers pushes their branch to GitHub? This needs to be taken seriously.

So there are a whole bunch of issues that would have to be addressed for this idea to work well. I believe those issues to be quite surmountable in principle, but given the state of git development (where GitHub, which to many is almost synonymous with git, doesn't even seem to be able to understand how git was originally meant to be used) I am not particularly optimistic. Still, I think it's a good idea, and I'd love to see it or something like it in git.

Building a HIP environment from scratch

HIP is a C++-based, single-source programming language for writing GPU code. "Single-source" means that a single source file can contain both the "host code" which runs on the CPU and the "device code" which runs on the GPU. In a sense, HIP is "CUDA for AMD", except that HIP can actually target both AMD and Nvidia GPUs.

If you merely want to use HIP, your best bet is to look at the documentation and download pre-built packages. (By the way, the documentation calls itself "ROCm" because that's what AMD calls its overall compute platform. It includes HIP, OpenCL, and more.)

I like to dig deep, though, so I decided I want to build at least the user space parts myself to the point where I can build a simple HelloWorld using a Clang from upstream LLVM. It's all open-source, after all!

It's a bit tricky, though, in part because of the kind of bootstrapping problems you usually get when building toolchains: Running the compiler requires runtime libraries, at least by default, but building the runtime libraries requires a compiler. Luckily, it's not quite that difficult, though, because compiling the host libraries doesn't require a HIP-enabled compiler - any C++ compiler will do. And while the device libraries do require a HIP- (and OpenCL-)enabled compiler, it is possible to build code in a "freestanding" environment where runtime libraries aren't available.

What follows is pretty much just a list of steps with running commentary on what the individual pieces do, since I didn't find an equivalent recipe in the official documentation. Of course, by the time you read this, it may well be outdated. Good luck!

Components need to be installed, but installing into some arbitrary prefix inside your $HOME works just fine. Let's call it $HOME/prefix. All packages use CMake and can be built using invocations along the lines of:

cmake -S . -B build -GNinja -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=$HOME/prefix -DCMAKE_PREFIX_PATH=$HOME/prefix
ninja -C build install

In some cases, additional variables need to be set.

Step 1: clang and lld

We're going to need a compiler and linker, so let's get llvm/llvm-project and build it with Clang and LLD enabled: -DLLVM_ENABLE_PROJECTS='clang;lld' -DLLVM_TARGETS_TO_BUILD='X86;AMDGPU'

Building LLVM is an art of its own which is luckily reasonably well documented, so I'm going to leave it at that.

Step 2: Those pesky cmake files

Build and install ROCm/rocm-cmake to avoid cryptic error messages down the road when building other components that use those CMake files without documenting the dependency clearly. Not rocket science, but man am I glad for GitHub's search function.

Step 3: libhsa-runtime64.so

This is the lowest level user space host-side library in the ROCm stack. Its services, as far as I understand them, include setting up device queues and loading "code objects" (device ELF files). All communication with the kernel driver goes through here.

Notably though, this library does not know how to dispatch a kernel! In the ROCm world, the so-called Architected Queueing Language is used for that. An AQL queue is setup with the help of the kernel driver (and that does go through libhsa-runtime64.so), and then a small ring buffer and a "door bell" associated with the queue are mapped into the application's virtual memory space. When the application wants to dispatch a kernel, it (or rather, a higher-level library like libamdhip64.so that it links against) writes an AQL packet into the ring buffer and "rings the door bell", which basically just means writing a new ring buffer head pointer to the door bell's address. The door bell virtual memory page is mapped to the device, so ringing the door bell causes a PCIe transaction (for us peasants; MI300A has slightly different details under the hood) which wakes up the GPU.

Anyway, libhsa-runtime64.so comes in two parts for what I am being told are largely historical reasons:

• ROCm/ROCT-Thunk-Interface
• ROCm/ROCR-Runtime; this one has one of those bootstrap issues and needs a -DIMAGE_SUPPORT=OFF

The former is statically linked into the latter...

Step 4: It which must not be named

For Reasons(tm), there is a fork of LLVM in the ROCm ecosystem, ROCm/llvm-project. Using upstream LLVM for the compiler seems to be fine and is what I as a compiler developer obviously want to do. However, this fork has an amd directory with a bunch of pieces that we'll need. I believe there is a desire to upstream them, but also an unfortunate hesitation from the LLVM community to accept something so AMD-specific.

In any case, the required components can each be built individually against the upstream LLVM from step 1:

• hipcc; this is a frontend for Clang which is supposed to be user-friendly, but at the cost of adding an abstraction layer. I want to look at the details under the hood, so I don't want to and don't have to use it; but some of the later components want it
• device-libs; as the name says, these are libraries of device code. I'm actually not quite sure what the intended abstraction boundary is between this one and the HIP libraries from the next step. I think these ones are meant to be tied more closely to the compiler so that other libraries, like the HIP library below, don't have to use __builtin_amdgcn_* directly? Anyway, just keep on building...
• comgr; the "code object manager". Provides a stable interface to LLVM, Clang, and LLD services, up to (as far as I understand it) invoking Clang to compile kernels at runtime. But it seems to have no direct connection to the code-related services in libhsa-runtime64.so.
  

That last one is annoying. It needs a -DBUILD_TESTING=OFF

Worse, it has a fairly large interface with the C++ code of LLVM, which is famously not stable. In fact, at least during my little adventure, comgr wouldn't build as-is against the LLVM (and Clang and LLD) build that I got from step 1. I had to hack out a little bit of code in its symbolizer. I'm sure it's fine.

Step 5: libamdhip64.so

Finally, here comes the library that implements the host-side HIP API. It also provides a bunch of HIP-specific device-side functionality, mostly by leaning on the device-libs from the previous step.

It lives in ROCm/clr, which stands for either Compute Language Runtimes or Common Language Runtime. Who knows. Either one works for me. It's obviously for compute, and it's common because it also contains OpenCL support.

You also need ROCm/HIP at this point. I'm not quite sure why stuff is split up into so many repositories. Maybe ROCm/HIP is also used when targeting Nvidia GPUs with HIP, but ROCm/CLR isn't? Not a great justification in my opinion, but at least this is documented in the README.

CLR also needs a bunch of additional CMake options: -DCLR_BUILD_HIP=ON -DHIP_COMMON_DIR=${checkout of ROCm/HIP} -DHIPCC_BIN_DIR=$HOME/prefix/bin

Step 6: Compiling with Clang

We can now build simple HIP programs with our own Clang against our own HIP and ROCm libraries:

clang -x hip --offload-arch=gfx1100 --rocm-path=$HOME/prefix -rpath $HOME/prefix/lib -lstdc++ HelloWorld.cpp
LD_LIBRARY_PATH=$HOME/prefix/lib ./a.out

Neat, huh?