[FEATURE] Support external Metal command queue via qd.init#618
Merged
Conversation
…etal_command_queue=ptr) Allow callers (e.g. Genesis) to share PyTorch MPS's MTLCommandQueue with Quadrants so both frameworks dispatch GPU work on the same Metal queue. This eliminates the need for qd.sync() + torch.mps.synchronize() at every DLPack zero-copy interop point — Metal's sequential command buffer semantics guarantee ordering automatically. Changes: - CompileConfig: add external_metal_command_queue (uint64_t, default 0) - MetalStream: add owns_queue_ flag; create_with_external_queue() factory - MetalDevice: accept optional external queue; create_with_external_queue() - MetalProgramImpl: route config→device creation - export_lang.cpp: expose new field to Python - field.py: gate qd.sync() and torch.mps.synchronize() on shared queue - test_metal_shared_queue.py: 4 tests covering init, zero-copy, and sync gating Co-authored-by: Cursor <cursoragent@cursor.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 922de4dd45
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Comment on lines
+1030
to
+1031
| [external_queue retain]; | ||
| return new MetalStream(device, external_queue, /*owns_queue=*/false); |
There was a problem hiding this comment.
In MetalStream::create_with_external_queue, the code retains external_queue but marks the stream as non-owning, so MetalStream::destroy() never releases that retain. Any workflow that repeatedly calls qd.init(...external_metal_command_queue=...) and qd.reset() will leak one MTLCommandQueue retain per cycle. Either avoid retaining for non-owned queues or pair this retain with a release during destruction.
Useful? React with 👍 / 👎.
New page metal_shared_queue.md with full setup guide: extracting PyTorch MPS MTLCommandQueue, init ordering, sync implications, ownership, and fallback. Cross-referenced from init_options.md and interop.md. Co-authored-by: Cursor <cursoragent@cursor.com>
hughperkins
force-pushed
the
hp/metal-shared-queue
branch
from
729f417 to
da31112
Compare
The external queue is caller-owned (PyTorch keeps it alive). The retain was unbalanced — owns_queue_=false skipped the release in destroy(), leaking one retain per init/reset cycle. Co-authored-by: Cursor <cursoragent@cursor.com>
duburcqa
reviewed
May 4, 2026
| import quadrants as qd | ||
|
|
||
| queue_ptr = get_mps_command_queue() # see below | ||
| qd.init(arch=qd.metal, external_metal_command_queue=queue_ptr) |
Contributor
There was a problem hiding this comment.
Would it be possible to rather add an option 'share_torch_command_queue'?
This is a bit invasive (yet very shallow, in python scope), but then it brings guarantees for to_torch. Otherwise, someone specifying anything by torch mps queue, would have synchronise wrongly skipped.
Collaborator
Author
There was a problem hiding this comment.
I'd rather keep it general on the whole. and have the user pass in the command queue, keeps everything clean, on quadrants side. Open to an extra boolean like something like 'external_metal_command_queue_is_torch_queue=True`, or some such.
Contributor
There was a problem hiding this comment.
'external_metal_command_queue_is_torch_queue=True`, or some such.
yes. I think this is necessary in this case.
duburcqa
reviewed
May 4, 2026
|
|
||
| PyTorch does not expose its MPS command queue through a public Python API. The following helper extracts it at runtime using `ctypes` and the Objective-C runtime, with no build-time PyTorch dependency: | ||
|
|
||
| ```python |
Contributor
There was a problem hiding this comment.
Not sure about the value of this in user-facing doc.
Collaborator
Author
There was a problem hiding this comment.
well, you (Genesis), the user, have to provide this bit of code.
Contributor
There was a problem hiding this comment.
well, you (Genesis), the user, have to provide this bit of code.
What? Wow that's painful. But ok I guess
Contributor
There was a problem hiding this comment.
I guess you are against making this first case in Quadrants? This is acceptable, even though it would be nice to hide this complexity out of Genesis to avoid having to care about the internal.
There could be some generic cross-gpu-backend force_shared_command_queue. Since the same also applies to CUDA and AMDGPU, it just happens that it works by default.
Collaborator
Author
There was a problem hiding this comment.
These are the current parameters:
Thoughts?
Contributor
There was a problem hiding this comment.
I know that these are the current parameters. I just find it a bit disappointing to have to manage command queue sharing for all backends manually on Genesis side. But the current approach does work in practice.
duburcqa
reviewed
May 4, 2026
|
|
||
| ```python | ||
| import torch | ||
| torch.zeros(1, device="mps") # trigger MPS init |
Contributor
There was a problem hiding this comment.
I think instantiating the device is sufficient. As you want.
Contributor
There was a problem hiding this comment.
Could you move device instantiation (ie torch init) in get_mps_command_queue ? I don't think this should be the responsibility of the user.
Collaborator
Author
There was a problem hiding this comment.
get_mps_command_queue() is part of Genesis, not part of torch. (It doesnt itself involve Quadrants).
duburcqa
reviewed
May 4, 2026
| | `f.to_torch(copy=False)` | `qd.sync()` called internally | no sync needed | | ||
| | `f.to_torch(copy=True)` | `qd.sync()` + `torch.mps.synchronize()` | no sync needed | | ||
| | Quadrants kernel after torch write | manual `torch.mps.synchronize()` required | automatic (same queue) | | ||
| | `f.to_numpy()` | `qd.sync()` (always needed for CPU readback) | `qd.sync()` (still needed) | |
Contributor
There was a problem hiding this comment.
Really? I'm quite sure torch is doing this automatically when converting to numpy, and it should work just great when sharing the same command queue.
Collaborator
Author
There was a problem hiding this comment.
Per Opus:
If to_numpy() went through PyTorch (DLPack → MPS tensor → .cpu().numpy()), then PyTorch's own GPU→CPU transfer would synchronize
automatically on the shared queue, and qd.sync() would indeed be redundant.
But the actual Field.to_numpy() code path doesn't go through torch — it uses Quadrants' own tensor_to_ext_arr kernel to copy GPU→CPU, then calls qd.sync() to wait for that
kernel. Since that's a Quadrants-internal command buffer, the CPU still needs to wait for it regardless of queue sharing.
Collaborator
Author
There was a problem hiding this comment.
from Opus:
Here's the step-by-step walkthrough of f.to_numpy() on Metal:
Line 626 — arr = np.zeros(shape=self.shape, dtype=dtype)
• Allocates a numpy array on CPU memory. This is the destination.
Line 629 — tensor_to_ext_arr(self, arr)
• This is a Quadrants kernel. It reads from self (a field backed by a Metal GPU buffer) and writes to arr (a numpy array in CPU memory).
• Under the hood, Quadrants compiles this to a Metal compute shader. The kernel is dispatched as a command buffer on Quadrants' Metal command queue.
• On Apple Silicon, Metal uses unified memory — the GPU and CPU share the same physical RAM. The numpy array's backing memory is accessible to the GPU. So the Metal shader can
write directly into the numpy buffer.
• tensor_to_ext_arr() returns immediately after submitting the command buffer. It does not wait for the GPU to finish.
Line 630 — quadrants.lang.runtime_ops.sync()
• This calls MetalStream::command_sync(), which does [cmdbuf waitUntilCompleted] for all pending command buffers.
• This is where the CPU blocks until the GPU has actually finished writing to arr.
• Without this sync, return arr on the next line would hand back a numpy array whose contents are still being written to by the GPU — you'd read garbage or partial data.
Why the shared queue doesn't help here: the sync isn't about coordinating between two frameworks (Quadrants vs PyTorch). It's about waiting for the GPU to finish before the CPU
reads the result. Even with a shared queue, the CPU still needs to wait for the Metal command buffer to complete. There's no PyTorch involvement in this path at all — it's
purely Quadrants kernel → CPU readback.
Contrast with to_torch(copy=False) where the shared queue does help: there the consumer is another GPU command buffer (PyTorch's), so Metal's in-queue ordering suffices without
any CPU-side wait.
Contributor
There was a problem hiding this comment.
But the actual Field.to_numpy() code path doesn't go through torch — it uses Quadrants' own tensor_to_ext_arr kernel to copy GPU→CPU
If this is true, it is a mistake that should be addressed.
Collaborator
Author
There was a problem hiding this comment.
why? that would add a torch dependency.
Contributor
There was a problem hiding this comment.
Because it is vastly more efficient. Without that, we cannot migrate Genesis to Quadrants' own 'to_numpy'. it is simply unusable.
Contributor
There was a problem hiding this comment.
I'm fine with keeping 'qd_to_numpy' in Genesis if for you it is not acceptable.
Collaborator
Author
There was a problem hiding this comment.
alright, well, one less thing to think about for me, if we just leave it in genesis for now :)
(otehrwise, we could have the quadrants vesrion switch depending on whether torch is avialble, add warnings if its not, etc, but taht would need a bunch of thought 😅 )
duburcqa
reviewed
May 4, 2026
duburcqa
reviewed
May 4, 2026
Comment on lines
+108
to
+111
| try: | ||
| queue_ptr = get_mps_command_queue() | ||
| except (AssertionError, OSError): | ||
| queue_ptr = 0 # 0 means "create a new queue" (the default) |
Contributor
There was a problem hiding this comment.
I would expect Quadrants to handle this, returning 0 with a warning. But this design is also ok-ish.
duburcqa
reviewed
May 4, 2026
| except RuntimeError as e: | ||
| raise ValueError(f"Zero-copy not available: {e}") from None | ||
| if impl.current_cfg().arch == _ARCH_METAL: | ||
| if impl.current_cfg().arch == _ARCH_METAL and not impl.current_cfg().external_metal_command_queue: |
Contributor
There was a problem hiding this comment.
This is not very efficient. It should probably be cached?
…semantics Co-authored-by: Cursor <cursoragent@cursor.com>
Coverage Report (
|
| File | Coverage | Missing |
|---|---|---|
🟢 python/quadrants/lang/field.py |
100% | |
🔴 tests/python/test_metal_shared_queue.py |
14% | 16-17,19,21,23-24,26-29,31-34,36-39,41-47,49-51,56,58-63,69-70,72,74-77,79-81,87,89,91-94,96-98,104,106,108,110-113,115-119,121-123,129,131,133,135,137-140,142,144-146,148-149,151-152 |
Diff coverage: 16% · Overall: 74% · 101 lines, 85 missing
CI runs non-torch tests first (without PyTorch installed), then installs torch and runs needs_torch tests. Without the marker, these tests ran in Phase 1 and failed on import. Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
hughperkins
force-pushed
the
hp/metal-shared-queue
branch
from
930708b to
3e3aa5d
Compare
Avoid repeated pybind attribute lookups on the hot path by caching the result of the arch + external_metal_command_queue check. The cache is keyed on the runtime object identity so it auto-invalidates on qd.init() / qd.reset(). Co-authored-by: Cursor <cursoragent@cursor.com>
hughperkins
force-pushed
the
hp/metal-shared-queue
branch
from
3e3aa5d to
bc79efc
Compare
Move the hardcoded _frozen_dc_plans.clear() call out of impl.reset() and register it via impl.on_reset() from _func_base.py instead, using the same lazy registration pattern as field.py to avoid circular imports. Co-authored-by: Cursor <cursoragent@cursor.com>
When an external Metal command queue is provided, the interop syncs (qd.sync / torch.mps.synchronize) are now only skipped when the caller explicitly declares the queue belongs to PyTorch MPS by setting external_metal_command_queue_is_torch_queue=True. Defaults to False, preserving sync behaviour for non-PyTorch external queues. Co-authored-by: Cursor <cursoragent@cursor.com>
Coverage Report (
|
| File | Coverage | Missing |
|---|---|---|
🔴 python/quadrants/lang/_func_base.py |
67% | 105,108 |
🔴 python/quadrants/lang/field.py |
69% | 149,152,159,165,168 |
🔴 python/quadrants/lang/impl.py |
60% | 589,592 |
🔴 tests/python/test_metal_shared_queue.py |
15% | 18-19,21,23,25-26,28-31,33-36,38-41,43-49,51-53,58,60-65,71-72,74,76-79,81-83,89,91,93-96,98-100,106,108,110,112-115,117-121,123-125,131,133,135,137,139-142,144,146-148,150-151,153-154 |
Diff coverage: 26% · Overall: 74% · 127 lines, 94 missing
hughperkins
added a commit
to hughperkins/genesis-world
that referenced
this pull request
May 4, 2026
At gs.init() time on Apple Metal, extract PyTorch MPS's MTLCommandQueue and pass it to Quadrants via external_metal_command_queue. This lets both frameworks dispatch GPU work on the same queue, so Metal's sequential command buffer ordering eliminates the need for explicit CPU-side syncs (qd.sync / torch.mps.synchronize) at every interop point. Raises an exception if the queue cannot be extracted, since correct synchronisation on Metal depends on the shared queue. Removes all 18 torch.mps.synchronize() guards from solvers and misc.py. Depends on Genesis-Embodied-AI/quadrants#618. Co-authored-by: Cursor <cursoragent@cursor.com>
hughperkins
added a commit
to hughperkins/genesis-world
that referenced
this pull request
May 4, 2026
At gs.init() time on Apple Metal, extract PyTorch MPS's MTLCommandQueue and pass it to Quadrants via external_metal_command_queue. This lets both frameworks dispatch GPU work on the same queue, so Metal's sequential command buffer ordering eliminates the need for explicit CPU-side syncs (qd.sync / torch.mps.synchronize) at every interop point. Raises an exception if the queue cannot be extracted, since correct synchronisation on Metal depends on the shared queue. Removes all torch.mps.synchronize() guards from solvers and misc.py. Depends on Genesis-Embodied-AI/quadrants#618. Co-authored-by: Cursor <cursoragent@cursor.com>
Move the Metal/PyTorch MPS synchronisation caching logic out of the hot core file field.py into its own module (python/quadrants/lang/ _metal_interop.py). This addresses the feature-factorization CI check which flags self-contained feature code being added to heavily-tracked core files. No behavioural change — field.py, matrix.py, and _ndarray.py now import from _metal_interop instead of defining the helpers inline. Co-authored-by: Cursor <cursoragent@cursor.com>
- Use Callable[[], None] instead of lowercase callable in impl.py (fixes pyright reportGeneralTypeIssues) - Unwrap docstrings and comments to 120-char lines instead of 80 Co-authored-by: Cursor <cursoragent@cursor.com>
Collaborator
Author
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 88475b448f
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Comment on lines
+1246
to
+1248
| MTLDevice_id mtl_device = MTLCreateSystemDefaultDevice(); | ||
| auto *queue = reinterpret_cast<MTLCommandQueue_id>(external_queue_ptr); | ||
| return new MetalDevice(mtl_device, queue); |
There was a problem hiding this comment.
create_with_external_queue() still calls MTLCreateSystemDefaultDevice() instead of deriving the MTLDevice from the provided queue, so a non-default external queue can be paired with resources allocated on a different device. In multi-GPU/non-default-queue setups this can lead to invalid Metal command encoding or runtime failures because command buffers and buffers/textures must come from the same device.
Useful? React with 👍 / 👎.
Collaborator
Author
There was a problem hiding this comment.
Per Opus:
My thoughts: Technically correct but practically irrelevant. Apple Silicon Macs have exactly one GPU — there is no multi-GPU scenario on macOS. MTLCreateSystemDefaultDevice()
always returns the same (only) device. PyTorch MPS also uses MTLCreateSystemDefaultDevice(), so the queue will always belong to the same device. That said, it would be
slightly more defensive to extract the device from the queue via [queue device]. Low priority — fine to leave as-is or address as a follow-up.
Comment on lines
+152
to
+154
| with patch("quadrants.lang.runtime_ops.sync") as mock_sync: | ||
| x.to_torch(copy=False) | ||
| mock_sync.assert_not_called() |
There was a problem hiding this comment.
This test patches quadrants.lang.runtime_ops.sync, but x.to_torch(copy=False) goes through _try_zerocopy_torch() and calls impl.get_runtime().sync() when sync is needed. Because of that mismatch, the assertion can pass even if the production code regresses and still synchronizes, so the test gives false confidence about the shared-queue fast path.
Useful? React with 👍 / 👎.
Coverage Report (
|
| File | Coverage | Missing |
|---|---|---|
🔴 python/quadrants/lang/_func_base.py |
67% | 105,108 |
🔴 python/quadrants/lang/_metal_interop.py |
45% | 12,14-15,17,19,22,31,37,40,51,62,64 |
🔴 python/quadrants/lang/_ndarray.py |
67% | 9 |
🔴 python/quadrants/lang/field.py |
67% | 12 |
🔴 python/quadrants/lang/impl.py |
50% | 5,589,592 |
🔴 python/quadrants/lang/matrix.py |
50% | 14 |
🔴 tests/python/test_metal_shared_queue.py |
13% | 18-19,21,23,25-26,28-31,33-36,38-41,43-49,51-53,58,60-64,69,75-77,79,81-84,86-88,94-95,97,99-102,104-106,112,114,116,118-121,123-127,129-131,137,139,141,143,145-148,150,152-154,156-157,159-160,166,168,170-172,174,176-179,181,183,185-190,192-193,195-196 |
Diff coverage: 23% · Overall: 67% · 168 lines, 129 missing
Co-authored-by: Cursor <cursoragent@cursor.com> # Conflicts: # python/quadrants/lang/field.py
Coverage Report (
|
| File | Coverage | Missing |
|---|---|---|
🔴 python/quadrants/lang/_func_base.py |
67% | 105,108 |
🔴 python/quadrants/lang/_metal_interop.py |
45% | 12,14-15,17,19,22,31,37,40,51,62,64 |
🔴 python/quadrants/lang/_ndarray.py |
67% | 9 |
🔴 python/quadrants/lang/field.py |
67% | 12 |
🔴 python/quadrants/lang/impl.py |
50% | 5,589,592 |
🔴 python/quadrants/lang/matrix.py |
50% | 14 |
🔴 tests/python/test_metal_shared_queue.py |
13% | 18-19,21,23,25-26,28-31,33-36,38-41,43-49,51-53,58,60-64,69,75-77,79,81-84,86-88,94-95,97,99-102,104-106,112,114,116,118-121,123-127,129-131,137,139,141,143,145-148,150,152-154,156-157,159-160,166,168,170-172,174,176-179,181,183,185-190,192-193,195-196 |
Diff coverage: 23% · Overall: 74% · 168 lines, 129 missing
Collaborator
Author
|
re: deleted comment: Per Opus: "we don't need it anymore — the comment was a FIXME describing the problem, and the shared queue feature is the fix. The function now has a proper docstring explaining when |
The test was patching quadrants.lang.runtime_ops.sync, but the actual code path (to_torch -> _try_zerocopy_torch) calls impl.get_runtime().sync(). Patch the runtime object directly so the test correctly verifies that sync is skipped when using a shared queue. Co-authored-by: Cursor <cursoragent@cursor.com>
…etal-shared-queue
Coverage Report (
|
| File | Coverage | Missing |
|---|---|---|
🔴 python/quadrants/lang/_func_base.py |
67% | 105,108 |
🔴 python/quadrants/lang/_metal_interop.py |
45% | 12,14-15,17,19,22,31,37,40,51,62,64 |
🔴 python/quadrants/lang/_ndarray.py |
67% | 9 |
🔴 python/quadrants/lang/field.py |
67% | 12 |
🔴 python/quadrants/lang/impl.py |
50% | 5,589,592 |
🔴 python/quadrants/lang/matrix.py |
50% | 14 |
🔴 tests/python/test_metal_shared_queue.py |
13% | 18-19,21,23,25-26,28-31,33-36,38-41,43-49,51-53,58,60-64,69,75-77,79,81-84,86-88,94-95,97,99-102,104-106,112,114,116,118-121,123-127,129-131,137,139,141,143,145-148,150,152,154-158,160-161,163-164,170,172,174-176,178,180-183,185,187,189-194,196-197,199-200 |
Diff coverage: 23% · Overall: 74% · 171 lines, 132 missing
PR change report (
|
| File | LoC | Added | Removed |
|---|---|---|---|
tests/python/test_metal_shared_queue.py |
133 | +133 | |
python/quadrants/lang/_metal_interop.py |
27 | +27 | |
python/quadrants/lang/_func_base.py |
619 | +7 | |
python/quadrants/lang/impl.py |
938 | +6 | -3 |
quadrants/runtime/program_impls/metal/metal_program.cpp |
31 | +6 | -1 |
quadrants/rhi/metal/metal_device.h |
427 | +5 | -2 |
quadrants/python/export_lang.cpp |
1048 | +4 | -1 |
python/quadrants/lang/field.py |
447 | +3 | -6 |
python/quadrants/lang/_ndarray.py |
334 | +3 | -4 |
python/quadrants/lang/matrix.py |
1113 | +2 | -2 |
quadrants/program/compile_config.h |
90 | +2 |
Total: 11 file(s) changed, +198 -19 code lines.
Coverage Report (
|
| File | Coverage | Missing |
|---|---|---|
🔴 python/quadrants/lang/_func_base.py |
67% | 105,108 |
🔴 python/quadrants/lang/_metal_interop.py |
45% | 12,14-15,17,19,22,31,37,40,51,62,64 |
🔴 python/quadrants/lang/_ndarray.py |
67% | 9 |
🔴 python/quadrants/lang/field.py |
67% | 12 |
🔴 python/quadrants/lang/impl.py |
50% | 5,589,592 |
🔴 python/quadrants/lang/matrix.py |
50% | 14 |
🔴 tests/python/test_metal_shared_queue.py |
13% | 18-19,21,23,25-26,28-31,33-36,38-41,43-49,51-53,58,60-64,69,75-77,79,81-84,86-88,94-95,97,99-102,104-106,112,114,116,118-121,123-127,129-131,137,139,141,143,145-148,150,152,154-158,160-161,163-164,170,172,174-176,178,180-183,185,187,189-194,196-197,199-200 |
Diff coverage: 23% · Overall: 74% · 171 lines, 132 missing
Expose a public helper that extracts PyTorch MPS's MTLCommandQueue* via ctypes/ObjC runtime. Downstream projects (e.g. Genesis) can call qd.interop.get_mps_command_queue() instead of inlining the extraction logic, keeping a single source of truth co-located with the feature. - New package: python/quadrants/interop/ - Tests updated to use the new public API Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…I/gstaichi into hp/metal-shared-queue
Replace inline ctypes extraction examples with the new public API. The function is now provided by quadrants.interop so users no longer need to copy-paste low-level ctypes code. Co-authored-by: Cursor <cursoragent@cursor.com>
The function now calls torch.zeros(1, device='mps') internally to ensure PyTorch MPS is initialised before extracting the queue pointer. Callers no longer need to handle init ordering themselves. Co-authored-by: Cursor <cursoragent@cursor.com>
PR change report (
|
| File | LoC | Added | Removed |
|---|---|---|---|
tests/python/test_metal_shared_queue.py |
117 | +117 | |
python/quadrants/interop/_torch_mps.py |
42 | +42 | |
python/quadrants/lang/_metal_interop.py |
27 | +27 | |
python/quadrants/lang/_func_base.py |
619 | +7 | |
python/quadrants/lang/impl.py |
938 | +6 | -3 |
quadrants/runtime/program_impls/metal/metal_program.cpp |
31 | +6 | -1 |
quadrants/rhi/metal/metal_device.h |
427 | +5 | -2 |
quadrants/python/export_lang.cpp |
1048 | +4 | -1 |
python/quadrants/lang/field.py |
447 | +3 | -6 |
python/quadrants/lang/_ndarray.py |
334 | +3 | -4 |
python/quadrants/lang/matrix.py |
1113 | +2 | -2 |
python/quadrants/interop/__init__.py |
2 | +2 | |
quadrants/program/compile_config.h |
90 | +2 | |
python/quadrants/__init__.py |
63 | +1 |
Total: 14 file(s) changed, +227 -19 code lines.
Coverage Report (
|
| File | Coverage | Missing |
|---|---|---|
🟢 python/quadrants/interop/__init__.py |
100% | |
🔴 python/quadrants/interop/_torch_mps.py |
12% | 28-29,31-34,36-37,39-40,43,45-47,49-52,54-57,59-62,64-70,72-75 |
🟢 python/quadrants/lang/_func_base.py |
100% | |
🟢 python/quadrants/lang/_metal_interop.py |
91% | 62,64 |
🟢 python/quadrants/lang/_ndarray.py |
100% | |
🟢 python/quadrants/lang/field.py |
100% | |
🟢 python/quadrants/lang/impl.py |
100% | |
🟢 python/quadrants/lang/matrix.py |
100% | |
🔴 tests/python/test_adstack.py |
0% | 200-201,203-204,207-211,213 |
🔴 tests/python/test_metal_shared_queue.py |
17% | 15-18,23,25-29,34,40-42,44,46-49,51-53,59-60,62,64-67,69-71,77,79,81,83-86,88-92,94-96,102,104,106,108,110-113,115,117,119-123,125-126,128-129,135,137-141,147,149,151-153,155,157-160,162,164,166-171,173-174,176-177 |
Diff coverage: 32% · Overall: 68% · 209 lines, 143 missing
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Contributor
|
ok to merge |
PR change report (
|
| File | LoC | Added | Removed |
|---|---|---|---|
tests/python/test_metal_shared_queue.py |
117 | +117 | |
python/quadrants/interop/_torch_mps.py |
42 | +42 | |
python/quadrants/lang/_metal_interop.py |
27 | +27 | |
python/quadrants/lang/_func_base.py |
619 | +7 | |
python/quadrants/lang/impl.py |
938 | +6 | -3 |
quadrants/runtime/program_impls/metal/metal_program.cpp |
31 | +6 | -1 |
quadrants/rhi/metal/metal_device.h |
427 | +5 | -2 |
quadrants/python/export_lang.cpp |
1048 | +4 | -1 |
python/quadrants/lang/field.py |
447 | +3 | -6 |
python/quadrants/lang/_ndarray.py |
334 | +3 | -4 |
python/quadrants/lang/matrix.py |
1113 | +2 | -2 |
python/quadrants/interop/__init__.py |
2 | +2 | |
quadrants/program/compile_config.h |
90 | +2 | |
tests/python/test_api.py |
457 | +2 | |
python/quadrants/__init__.py |
63 | +1 |
Total: 15 file(s) changed, +229 -19 code lines.
hughperkins
added a commit
that referenced
this pull request
May 9, 2026
* black auto-reformats in `subgroup.py` and `test_simt.py` (line-length=120 per `.pre-commit-config.yaml`). * clang-format auto-reformats in `codegen_amdgpu.cpp` and `spirv_codegen.cpp`. * Drop unused `from quadrants.lang.simt import subgroup` from `_algorithms.py` (left over after the switch to `subgroup_inclusive_add_warp_i32`); ruff re-sorts the remaining import block. * Extend the file-level pyright comment in `subgroup.py` from `reportInvalidTypeForm=false` to also disable `reportOperatorIssue` so that `p & shuffle_xor(...)` / `p | shuffle_xor(...)` in the new voting ops don't trip pyright on `Expr` operator overloads — same false-positive class the existing suppression already covers. Pre-commit (black, clang-format, ruff, pylint, trailing-whitespace, end-of-file) clean. Pyright is down to 6 pre-existing errors in files this branch does not touch (`_tensor_wrapper.py`, `_func_base.py`, `_metal_interop.py`, all from PR #618 / streams work) — net 0 new errors attributable to this branch.
duburcqa
pushed a commit
to hughperkins/genesis-world
that referenced
this pull request
May 12, 2026
At gs.init() time on Apple Metal, extract PyTorch MPS's MTLCommandQueue and pass it to Quadrants via external_metal_command_queue. This lets both frameworks dispatch GPU work on the same queue, so Metal's sequential command buffer ordering eliminates the need for explicit CPU-side syncs (qd.sync / torch.mps.synchronize) at every interop point. Raises an exception if the queue cannot be extracted, since correct synchronisation on Metal depends on the shared queue. Removes all torch.mps.synchronize() guards from solvers and misc.py. Depends on Genesis-Embodied-AI/quadrants#618. Co-authored-by: Cursor <cursoragent@cursor.com>
npoulad1
added a commit
to ROCm/quadrants
that referenced
this pull request
Jun 8, 2026
* [Misc] Warn user to disable caching when print_ir/QD_DUMP_IR enabled (Genesis-Embodied-AI#425) Co-authored-by: v01dxyz <v01dxyz@v01d.xyz> * [Build] Pin torch version to CUDA 12.8 for CUDA tests (Genesis-Embodied-AI#428) * [Misc] Fixing up taichi-dev urls (Genesis-Embodied-AI#429) * [Perf] Rename cuda_graph to gpu_graph across the codebase (Genesis-Embodied-AI#430) * Misc: fix typo integeral -> integral (Genesis-Embodied-AI#434) Co-authored-by: v01dxyz <v01dxyz@v01d.xyz> * [Perf] CUDA graph 4: call from multiple locations (Genesis-Embodied-AI#420) * [Bug] Fix fastcache not restoring graph_do_while_arg (Genesis-Embodied-AI#435) * [Perf] Cache last-call result in perf_dispatch for single-compatible case (Genesis-Embodied-AI#438) * Fix gpu_graph fallback on old Nvidia GPU. (Genesis-Embodied-AI#443) * Fix shared memory offset not reset between CUDA kernels. (Genesis-Embodied-AI#442) * [Misc] Allow disabling GPU graph via QD_GPU_GRAPH=0 env var (Genesis-Embodied-AI#439) * [Misc] Add named top-level loops (Genesis-Embodied-AI#440) * [Misc] Rename gpu_graph to graph (Genesis-Embodied-AI#446) * [Misc] Add cross-platform shuffle (Genesis-Embodied-AI#447) * [Bug] Fix graph_do_while on Windows: search for cudadevrt.lib (Genesis-Embodied-AI#456) * [Bug] Also search default CUDA toolkit install location on Windows (Genesis-Embodied-AI#461) * [SPIRV] Feature Parity Atomics & Shared Array (Genesis-Embodied-AI#432) * [Misc] Change clang format to 120 characters (Genesis-Embodied-AI#463) * [Misc] CUDA graph 5 Add fatbin (Genesis-Embodied-AI#464) * [Bug] Reuse VkInstance across init/reset cycles (Genesis-Embodied-AI#465) * [Perf] Tiles 1: _load, _store, _eye_ (Genesis-Embodied-AI#466) * [Misc] Remove dead InternalFuncStmt type_check override (Genesis-Embodied-AI#471) * [Perf] Tiles 2: add cholesky and ger (Genesis-Embodied-AI#472) * [Perf] Tiles 2b: add triangular solve (Genesis-Embodied-AI#474) * [Misc] Refactor: use _get_col/_set_col in tiles load/store/init (Genesis-Embodied-AI#475) * [Build] Fix flaky test_clock_accuracy (Genesis-Embodied-AI#436) * Fix AARCH64 emitting invalid asm in CUDA kernels. (Genesis-Embodied-AI#473) Co-authored-by: Hugh Perkins <hughperkins@gmail.com> * [AMDGPU] Enable HIP memory pool and surface pool-exhaustion errors. (Genesis-Embodied-AI#485) * [AMDGPU] Scope hsaco tmp dir per-user to avoid collisions. (Genesis-Embodied-AI#484) * [Perf] Tiles 3: Add slice syntax, qd.outer() and initial doc (Genesis-Embodied-AI#477) * [AMDGPU] Fix gradient computation. (Genesis-Embodied-AI#486) * Enable all backends that are supported in unit tests. (Genesis-Embodied-AI#488) * Fix SPIRV ID overflow for large kernels due to autodiff. (Genesis-Embodied-AI#489) * [Misc] Fix purity checker to allow accessing constants from quadrants modules (Genesis-Embodied-AI#487) * [Misc] Increase tolerance for clock monotonic test (Genesis-Embodied-AI#492) * [CI] Serialize api doc workflow (Genesis-Embodied-AI#494) * [CI] Increase tolerance for clock test (Genesis-Embodied-AI#506) * [CI] Increase clock test tolerance to 20% (Genesis-Embodied-AI#509) * [Perf] Add tensor_type parametrization to tile16 tests (Genesis-Embodied-AI#504) * [Perf] Tiles 4b: Migrate tiles16 tests to enable fastcache (Genesis-Embodied-AI#505) * [Perf] Tiles 4c: add Tiles16x16 proxy (Genesis-Embodied-AI#507) * [Perf] Tiles 4d: Consolidate slice error tests using parametrize (Genesis-Embodied-AI#508) * [Perf] Tiles 4: add SharedArray slice support (Genesis-Embodied-AI#482) * [Perf] Tiles 5: add Cholesky benchmark demo (Genesis-Embodied-AI#483) * [Doc] Add user guide page for subgroup shuffle (Genesis-Embodied-AI#512) * [Perf] Implement cross-platform shuffle_down (Genesis-Embodied-AI#510) * [Perf] Add portable subgroup reduce_add and reduce_all_add (Genesis-Embodied-AI#511) * [Perf] Add first warmup config to perf dispatch (Genesis-Embodied-AI#422) * [AutoDiff] Autodiff 1: Add baseline adstack regression test for unary_collections (Genesis-Embodied-AI#500) * [AutoDiff] Autodiff 2: Implement derivative for tan (Genesis-Embodied-AI#501) * [AutoDiff] Autodiff 3: Recompute tanh/exp on the operand in the reverse pass (Genesis-Embodied-AI#502) * [AutoDiff] Autodiff 4: Mark rsqrt as non-linear for adstack promotion (Genesis-Embodied-AI#503) * [AutoDiff] Autodiff 5: Fix adjoint-alloca placement for GlobalLoads outside the current range-for (Genesis-Embodied-AI#496) * [AutoDiff] Autodiff 6: Adstack regression tests (Genesis-Embodied-AI#491) * [AutoDiff] Autodiff 7: Fix header size in AdStackAllocaStmt to match u64 runtime layout (Genesis-Embodied-AI#534) * [AutoDiff] Autodiff 8: Surface LLVM adstack push/pop overflow as a Python exception (Genesis-Embodied-AI#535) * [AutoDiff] Autodiff 9: Guard against LLVM worker-thread stack overflow from large per-task adstack budget (Genesis-Embodied-AI#495) * [AutoDiff] Autodiff 10: Implement adstack for SPIR-V (Genesis-Embodied-AI#490) * [AutoDiff] Autodiff 11: Latent adstack-adjacent fixes (AMDGPU hipFree, flush() keeps ctx_buffers_, always-preallocate) (Genesis-Embodied-AI#536) * [Doc] Add AGENTS.md with instructions for AI agents (Genesis-Embodied-AI#541) * [Bug] Abort kernel execution on assertion failure instead of segfaulting (Genesis-Embodied-AI#419) * [Type] ndarray typing 1: Add eval_str=True to inspect.signature() calls (Genesis-Embodied-AI#411) * [CI] Suppress reportPrivateImportUsage in torch-using files (Genesis-Embodied-AI#552) * [Misc] QD_DUMP_IR dumps to files with the task_id added to the filename (Genesis-Embodied-AI#441) * [Type] ndarray typing 2: Fix NDArray single-arg subscript crash (Genesis-Embodied-AI#412) * [Test] Flush xdist channel before worker exit so test failure reports are visible (Genesis-Embodied-AI#555) * [CI] Reduce test retries on CI from 3 to 1. (Genesis-Embodied-AI#554) * [AutoDiff] Autodiff 12: Heap-backed adstack on LLVM backends (CPU/CUDA/AMDGPU) (Genesis-Embodied-AI#537) * [AutoDiff] Autodiff 13: Heap-backed adstack on SPIR-V backends (Metal, Vulkan) (Genesis-Embodied-AI#493) * [AutoDiff] Autodiff 14: Resolve bounded-inner-loop adstacks without default_ad_stack_size fallback (Genesis-Embodied-AI#539) * [SPIRV] Vulkan SPIR-V correctness: atomic-view aliasing, PSB stride, narrow storage caps, u1 cast, per-init layer recheck (Genesis-Embodied-AI#513) * [Build] Autodiff 15: Replace 2022 MoltenVK pin with LunarG Vulkan SDK fetch and sanitise MoltenVK cap advertisement (Genesis-Embodied-AI#551) * [Test] Suppress stock pytest-timeout to avoid conflict with pytest_hardtle (Genesis-Embodied-AI#557) * [Vulkan] Use SDK validation layer for debugPrintf instead of apt package (Genesis-Embodied-AI#562) * [Test] Fix flaky perf_dispatch tests by increasing work amounts (Genesis-Embodied-AI#559) * [Test] Add --maxfail CLI option to run_tests.py (default 20) (Genesis-Embodied-AI#558) * [CI] Vulkan debug printf fix to address flaky tests (Genesis-Embodied-AI#563) * [Docs] Add a new page to help for first time contributors (Genesis-Embodied-AI#426) Authored-by: v01dxyz <v01dxyz@v01d.xyz> * [AutoDiff] Autodiff 16: Resolve reverse-mode adstack depths per-launch via runtime-evaluated SizeExpr (Genesis-Embodied-AI#543) * Fix: raise error if device memory allocation fails (Genesis-Embodied-AI#451) (Genesis-Embodied-AI#453) Co-authored-by: v01dxyz <v01dxyz@v01d.xyz> Co-authored-by: Hugh Perkins <hughperkins@gmail.com> * [CI] Add CI job to check line wrapping of comments and docs (Genesis-Embodied-AI#564) * [Misc] Add coverage report to PRs, including kernels (Genesis-Embodied-AI#470) * [CI] CI wrap check feeds only diffs to agent (Genesis-Embodied-AI#567) * Skip 'flaky' test on MacOS CI. (Genesis-Embodied-AI#573) * [Test] Fix missing `import sys` in test_fail_device_memory_allocation (Genesis-Embodied-AI#574) * [CI] Fix Vulkan debugPrintf flake with session-scoped warmup (Genesis-Embodied-AI#571) * [AutoDiff] determine_ad_stack_size: replace whole-CFG Bellman-Ford with SCC + DAG DP (Genesis-Embodied-AI#575) * [Test] Fix macOS OOM skip reason to describe actual root cause (Genesis-Embodied-AI#576) * [Lang] whole_kernel_cse: 2.5x compile time speedup on large kernels (Genesis-Embodied-AI#577) * [CI] Add CI check for unnecessarily deleted comments (Genesis-Embodied-AI#570) * [CI] Migrate coverage report to github Check page (Genesis-Embodied-AI#566) * [Lang] Skip IR verifier between passes unless debug=true (Genesis-Embodied-AI#579) * [Lang] Inline AdStack ops on release LLVM codegen: dramatically reduces compile time for adstack-enabled reverse-mode kernels (Genesis-Embodied-AI#584) * [CUDA] Honor offline_cache=False end-to-end so QD_OFFLINE_CACHE=0 actually gives a cold compile (Genesis-Embodied-AI#580) * [Type] Tensor 24 (Genesis-Embodied-AI#561) Co-authored-by: hugh <hugh@slurm-login-0.slurm-login.tenant-slurm.svc.cluster.local> * [Lang] auto_diff host-walk reductions: dramatically faster front-end compile time on adstack-enabled reverse-mode kernels (Genesis-Embodied-AI#587) * [AutoDiff] Speed up reverse-mode kernel launches on GPU backends (Genesis-Embodied-AI#578) * [Vulkan] Move adstack-sizer scratch out of Function-scope memory to fix SPIR-V pipeline build failures (Genesis-Embodied-AI#588) * [AutoDiff] Improve diagnosis of unsupported reverse-mode AD patterns (Genesis-Embodied-AI#590) * [Bug] Fix: promote Ndarray to AnyArray in build_Name for flattened struct fields (Genesis-Embodied-AI#592) * [SPIR-V] Shrink reverse-grad kernel MSL by ~50% (Genesis-Embodied-AI#591) * [CI] Add CI check that PR changes have test coverage (Genesis-Embodied-AI#596) * [Perf] Enable zero-copy in to_torch() and to_numpy() (Genesis-Embodied-AI#450) * Add BufferView: safe sub-range ndarray access for kernels (Genesis-Embodied-AI#585) Co-authored-by: alanray-tech <alanray-tech@users.noreply.github.com> Co-authored-by: Hugh Perkins <hughperkins@gmail.com> * [Doc] Add user-facing fastcache documentation (Genesis-Embodied-AI#597) Co-authored-by: hugh <hugh@slurm-login-0.slurm-login.tenant-slurm.svc.cluster.local> * [Misc] Upgrade to enable v1 dlpack so to_numpy(copy=False) writable (Genesis-Embodied-AI#598) Co-authored-by: root <root@rtx-209-201.slurm-compute.tenant-slurm.svc.cluster.local> * [AutoDiff] Cut reverse-mode adstack memory usage 10x on all backends (Genesis-Embodied-AI#599) * [Misc] Add CI check for feature file factorization (Genesis-Embodied-AI#606) * [Perf] Skip _recursive_set_args for all-Field frozen dataclass structs (Genesis-Embodied-AI#607) Co-authored-by: Cursor <cursoragent@cursor.com> * [AutoDiff] SNode-arm bound-expr capture rejects fold-attack gate indices (Genesis-Embodied-AI#610) * [Misc] Suppress field fastcache warning for qd.Tensor (Genesis-Embodied-AI#615) Co-authored-by: Cursor <cursoragent@cursor.com> * [AutoDiff] Adstack heap: clip reducer count by per-task loop trip count (compile-time and SizeExpr-evaluated) (Genesis-Embodied-AI#611) * [Misc] Forward copy= through qd.Tensor, add copy=None option (Genesis-Embodied-AI#616) Co-authored-by: Cursor <cursoragent@cursor.com> * [Doc] Update README (Genesis-Embodied-AI#617) Co-authored-by: Cursor <cursoragent@cursor.com> * [CI] Fix coverage report showing def lines as uncovered (Genesis-Embodied-AI#623) Co-authored-by: Cursor <cursoragent@cursor.com> * [Perf] Generic launcher: persistent context, JIT-pointer reuse, Metal compute encoder, LLVM-GPU async memory ops (Part 1/2) (Genesis-Embodied-AI#619) * [CI] Encode Python-first testing policy in coverage-check prompt (Genesis-Embodied-AI#622) Co-authored-by: Cursor <cursoragent@cursor.com> * [CI] Add PR Line change report (Genesis-Embodied-AI#624) Co-authored-by: Cursor <cursoragent@cursor.com> * [CI] Disable quadrants pytest plugin during quadrants internal coverage runs (Genesis-Embodied-AI#629) Co-authored-by: Cursor <cursoragent@cursor.com> * [AutoDiff] Adstack load+store eliminations: EliminateRecomputableAdStackPushes pass + leaf extensions (Genesis-Embodied-AI#621) * [CI] Simplify coverage PR comment to a single linked line (Genesis-Embodied-AI#630) * [CUDA] Add AGX Thor, SM_110 (Genesis-Embodied-AI#631) Co-authored-by: Johnny Nunez and Hugh Perkins * [CI] Lines changed report: collapse PR comment to a single linked totals line (Genesis-Embodied-AI#632) * [FEATURE] Support external Metal command queue via qd.init (Genesis-Embodied-AI#618) Co-authored-by: Cursor <cursoragent@cursor.com> * [Perf] Cache adstack-sizer metadata per task across SPIR-V + LLVM-GPU; per-snode / DeviceAllocation invalidation (Part 2/2) (Genesis-Embodied-AI#620) * [AutoDiff] Disable EliminateRecomputableAdStackPushes pending mutated-SNode chain-leaf fix (Genesis-Embodied-AI#633) * [AutoDiff] Adstack chain-clone safety: mutated-SNode leaf reject + load_top consumer-aware guard (Genesis-Embodied-AI#634) * [Docs] Add user-guide page for qd.simt.block.* primitives (Genesis-Embodied-AI#638) * [Docs] Expand qd.simt.subgroup user-guide page to cover every op (Genesis-Embodied-AI#639) * [Perf] Streams 1-4 (Genesis-Embodied-AI#410) * [Docs] Add user-guide page for matrix decompositions and solvers (Genesis-Embodied-AI#643) * [Bug] Revert "[Perf] Streams 1-4 (Genesis-Embodied-AI#410)" (Genesis-Embodied-AI#650) * [Docs] Add user-guide page for atomics and bit operations (Genesis-Embodied-AI#640) * [Docs] Add user-guide page for qd.simt.grid.* primitives (Genesis-Embodied-AI#641) * [AutoDiff] Adstack max-reducer: parallel multi-axis MaxOverRange dispatch (Genesis-Embodied-AI#635) * [AMDGPU] Fix amdgpu parallel rand init (Genesis-Embodied-AI#658) * [Perf] Adstack: skip max-reducer recognizer on CPU + lift host-eval cap (Genesis-Embodied-AI#655) * [Perf] Re-land Streams 1-4 with bug fixes (Genesis-Embodied-AI#653) * [AMDGPU] Apply device_memory_GB=0.3 cap to AMDGPU tests (Genesis-Embodied-AI#659) * [Perf] Per-launch host sync: drop wait_idle on SPIR-V, pin stream and drop stream_synchronize on CUDA/AMDGPU (Genesis-Embodied-AI#654) * [AMDGPU] Unload hipModule_t in JITModuleAMDGPU destructor (Genesis-Embodied-AI#660) * [AMDGPU] Trim default mempool on qd.reset() (Genesis-Embodied-AI#669) * [AMDGPU] Hoist rand-state buffer to process lifetime (Genesis-Embodied-AI#668) * [Streams] Use events for streams serialization on AMDGPU and CUDA (Genesis-Embodied-AI#667) * [Perf] Adstack max-reducer: launch cache + zero-copy result map; content-stable registry_id (Genesis-Embodied-AI#671) * [SPIR-V] dispatch_max_reducers: register each task with the real kernel name (Genesis-Embodied-AI#675) * [AutoDiff] Debug-mode field/grad/dual: dtype, layout, and access-time invariants (Genesis-Embodied-AI#677) * [Docs] Add user-guide page for qd.algorithms.* device-wide algorithms (Genesis-Embodied-AI#642) Co-authored-by: alanray-tech <alan.ray@genesis-ai.company> * [Docs] Doc for existing atomics: switch support table to per-backend columns (Genesis-Embodied-AI#657) Co-authored-by: alanray-tech <alan.ray@genesis-ai.company> * [GPU] Cross gpu atomics (Genesis-Embodied-AI#666) Co-authored-by: alanray-tech <alan.ray@genesis-ai.company> * [GPU] Make block operations portable cross-gpu (Genesis-Embodied-AI#664) * [Perf] CPU LLVM adstack-cache: skip per-launch bump-writes + ndarray_shapes capture on forward-only handles (Genesis-Embodied-AI#685) * [GPU] Cross-GPU for grid ops (Genesis-Embodied-AI#670) * [Math] Make bitop operations portable cross-gpu (Genesis-Embodied-AI#662) * [AMDGPU] Always use wave64, on both RDNA and CDNA (Genesis-Embodied-AI#687) * [AMDGPU] Use syncscope("agent") for atomix xor to avoid CAS livelock (Genesis-Embodied-AI#672) * [GPU] New bit ops for QIPC (Genesis-Embodied-AI#679) * [GPU] Subgroup ops cross-gpu (Genesis-Embodied-AI#665) * [Graph] Rename CUDA Graph to Graph in docs (Genesis-Embodied-AI#691) * [SPIR-V] Fix FIFO-queue ordering when sharing command queue. (Genesis-Embodied-AI#694) * [Atomics] New QIPC ops for atomics (Genesis-Embodied-AI#690) * Pass dataclass sub-structs into qd.func (Genesis-Embodied-AI#698) * [AMDGPU] HIP graph runtime support for @qd.kernel(graph=True) (Genesis-Embodied-AI#692) * [CI] Add per-file timing report to Mac Metal test job (Genesis-Embodied-AI#695) Co-authored-by: Cursor <cursoragent@cursor.com> * [CI] Enable kernel disk cache during tests (Genesis-Embodied-AI#696) * [Math] New QIPC ops for single-threaded linalg (Genesis-Embodied-AI#683) * [BREAKING][GPU] New QIPC ops for subgroups (Genesis-Embodied-AI#676) * [GPU] New QIPC ops for block (Genesis-Embodied-AI#684) * [GPU] New device-level ops for QIPC (Genesis-Embodied-AI#693) * [algorithms] PrefixSumExecutor: drop unused GRID_SZ local (Genesis-Embodied-AI#701) * [block] sync(): fix unsupported-arch error message (Genesis-Embodied-AI#700) * [volatile_load] add qd.volatile_load primitive (closes Genesis-Embodied-AI#648) (Genesis-Embodied-AI#702) * [AutoDiff] Reject recycled identity_key in AdStackCache::register_adstack_sizing_info (Genesis-Embodied-AI#708) * [Vulkan] Declare GroupNonUniform SPIR-V caps and enable shaderSubgroupExtendedTypes (Genesis-Embodied-AI#707) * Fix duplicate HIP graph driver-function declarations after v1.0.0 merge The amd-integration fork had cherry-picked the HIP graph driver functions (graph_create / graph_destroy / graph_add_kernel_node / graph_instantiate / graph_exec_destroy / graph_launch), and upstream v1.0.0 added the same set. The per-file 3-way merge appended both copies into amdgpu_driver_functions.inc.h, producing redeclaration errors that broke the AMDGPU RHI/runtime compile. Drop the upstream duplicate block; the signatures are identical to the fork's existing declarations. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix AMDGPU launcher coherence and num_instructions visibility after v1.0.0 merge - kernel_launcher.cpp: the 3-way merge spliced upstream v1.0.0's launch_llvm_kernel rewrite (ephemeral arg/context buffers, explicit-stream path, AmdgpuDefaultStream PinGuard) onto the AMD fork's kernarg-by-value + persistent-scratch design, leaving references to undefined `ephemeral_context_ptr`. Restore the fork's coherent launch_llvm_kernel verbatim; it calls the (already merged) enhanced launch_offloaded_tasks, which keeps the max-reducer dispatch and stream-parallel groups adapted onto the AMD launch path. - llvm_context.h: both the fork and upstream added `num_instructions`; the merge kept upstream's private placement, but the AMDGPU codegen force-inline heuristic calls it statically from outside the class. Move it back to the public section. Co-authored-by: Cursor <cursoragent@cursor.com> * Restore async result D2H and hoist kernarg vectors in AMDGPU launcher The v1.0.0 merge resolution regressed two amd-integration baseline optimizations in launch_llvm_kernel / launch_offloaded_tasks: - The per-launch result-buffer copy was a blocking memcpy_device_to_host, forcing a host stall on every value-returning launch and serializing the GPU pipeline. Restore the async D2H (the caller synchronizes lazily when it needs the value); external-array transfers still stream_synchronize once before reading back. - launch_task constructed the kernarg std::vectors from initializer lists ({kernarg_payload} / {kernarg_size}) on every dispatch (heap alloc + free per launch). Hoist arg_ptrs/arg_sizes out of the per-task launch and reuse. Co-authored-by: Cursor <cursoragent@cursor.com> * amdgpu: default to LDS permlane64 emulation; drop host-x86 barrier asm on retarget Two AMDGPU JIT-compile crashes surfaced after the v1.0.0 merge pulled in the QIPC subgroup ops (Genesis-Embodied-AI#676), which made the rigid constraint solver's wave-cooperative reductions route through `amdgpu_cross_half_shuffle_i32`. Both manifested as a SIGSEGV inside `llvm::SIInstrInfo::getInstSizeInBytes` during `JITSessionAMDGPU::compile_module_to_hsaco` (i.e. at first kernel launch), and reproduce on gfx942 / MI300X. Baseline 0.4.6 never emitted these constructs, which is why it was unaffected. 1. Native `llvm.amdgcn.permlane64` lowering crashes the bundled LLVM 22.1.0 AMDGPU backend. Default `amdgpu_permlane64` to the existing LDS-roundtrip software emulation on every target (it produces identical results). Add `QD_AMDGPU_USE_NATIVE_PERMLANE64=1` to opt back into the native instruction once the backend bug is fixed; the old `QD_AMDGPU_FORCE_PERMLANE64_FALLBACK` is now the default and still honored. This is the actual crash fix. 2. The runtime module is compiled by the host x86_64 clang and only retargeted to amdgcn here, so `amdgpu_cross_half_shuffle_i32`'s `__asm__ volatile("" : "+v"(byte))` optimization barrier carries x86 flag clobbers (`~{dirflag},~{fpsr},~{flags}`) that are meaningless on AMDGPU. The IR verifies but the empty-body INLINEASM is invalid on the amdgcn target. Neutralize empty-body barrier asm during retarget (forward the tied value, then erase) so no stale host asm reaches codegen. On the wave64 targets we ship `ds_bpermute` already addresses the full wave, so the hint is a no-op. Co-authored-by: Cursor <cursoragent@cursor.com> * style: apply clang-format (v19.1.7) to AMDGPU fn_attrs and launcher sources CI pre-commit's clang-format hook reformatted these files (long declarations/lambda signatures collapsed onto single lines per the repo's clang-format config). Apply the same formatting so the hook passes. No functional changes. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(amdgpu): use CreateNeg for branchless i32 sgn instead of CreateSub(0, input) clang-tidy (modernize-use-nullptr, -warnings-as-errors) flagged `builder->CreateSub(0, input)` in the i32 sgn path: the literal `0` binds to the `llvm::Value*` LHS parameter as a null pointer, not an integer zero. Replace with `builder->CreateNeg(input)`, which emits `0 - input` with a proper zero constant -- identical intended semantics, and clang-tidy clean. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Robert Dazi <14996868+v01dXYZ@users.noreply.github.com> Co-authored-by: v01dxyz <v01dxyz@v01d.xyz> Co-authored-by: Hugh Perkins <hughperkins@gmail.com> Co-authored-by: Alexis DUBURCQ <alexis.duburcq@gmail.com> Co-authored-by: hugh <hugh@slurm-login-0.slurm-login.tenant-slurm.svc.cluster.local> Co-authored-by: alanray-tech <alan.ray@genesis-ai.company> Co-authored-by: alanray-tech <alanray-tech@users.noreply.github.com> Co-authored-by: root <root@rtx-209-201.slurm-compute.tenant-slurm.svc.cluster.local> Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Johnny <johnnynuca14@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Allow callers (e.g. Genesis) to share PyTorch MPS's MTLCommandQueue with Quadrants so both frameworks dispatch GPU work on the same Metal queue. This eliminates the need for qd.sync() + torch.mps.synchronize() at every DLPack zero-copy interop point — Metal's sequential command buffer semantics guarantee ordering automatically.
Changes:
Issue: #
Brief Summary
copilot:summary
Walkthrough
copilot:walkthrough