Merging the HAL rewrite branch to main #4571
Merged
Conversation
This file contains 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
benvanik
added
runtime
|
This is still marked as a draft and builds are failing. Do you want me to start reviewing now or wait? |
|
Can wait - leaving as draft until I get the CIs passing. All of these have been reviewed by @ScottTodd as they were going into the branch, so the big thing to look for here is whether this messes you up in any way that the CI's don't check :) |
benvanik
force-pushed
the
staging-hal-rewrite
branch
17 times, most recently
from
4ffb506 to
685c5cd
Compare
Doesn't seem to change anything but there's not really a reason to detach and I'd rather never have to think about this code again.
This would cause all kind of badness when multiple executables shared the same hal.interface bindings but had differing push constants.
The max size on the stack is 3KB, which is fine for something transient like this that is only used for construction of the executor.
They were all implemented using mapping anyway and this way we only have that code once and make the restriction that buffers be mappable part of their usage uniformly.
* Use `IREE_RESTRICT` for older MSVC versions. * Update iree-run-module-vulkan-gui-main.cc * Include <atomic> for std::atomic use in GetTempFile. Tested CMake build of `all` (vmla+vulkan only) on my Windows machine, with samples and Vulkan enabled.
This case is not yet exercised from our code but I'll be fixing it ASAP anyway.
Unfortunately not caught by any of our open source builds.
---
```
typedef size_t iree_host_size_t;
typedef uint64_t iree_device_size_t;
```
`iree_device_size_t` should use `% " PRIu64 "` instead of `%zu`
---
`ppoll` is available since Android API 21 (or 28 for `ppoll64`? seeing conflicting docs)
---
`iree_hal_buffer_calculate_range` outputs into a `iree_device_size_t`, which isn't directly compatible with
```
typedef struct {
uint8_t* data;
iree_host_size_t data_length;
} iree_byte_span_t;
```
Possibly: hard to test without 128 cores. I did try turning the topology mask into a uint8_t on my 64 core machine and it passed --config=ubsan.
The wait functionality isn't done yet.
The classic condvar shutdown notification race.
Fallback to 1 thread for when cpuinfo_get_cores_count()==0 and core count when cpuinfo_get_l2_caches_count()==0. Issue #4654 is tracking making this better.
benvanik
force-pushed
the
staging-hal-rewrite
branch
from
59b7a7c to
6de6fdf
Compare
ScottTodd
approved these changes
Jan 29, 2021
Branches are stable. Ready to merge :D |
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Major branch features:
iree/hal/local/backend replacing host supporting multithreading and pluggable executable loadersiree/modules/vmla/, enabling the ability for host code to use it tooiree/hal/cts/using the public C APIs instead of the internal C++ onesNOTE: this removes the experimental metal backend (to be revived in #4370).
NOTE: this disables hal.executable linking for vmla/dylib pending a fix in #4536 (linking had prior to this been relying on those two backends to not using executable layouts and now they do). Will be fixed up soon.
First major milestone of #4369.
Fixes #4391.
Fixes #4472.