github/ROCm
1
1
Fork 0
mirror of https://github.com/ROCm/ROCm.git synced 2026-01-10 07:08:08 -05:00
Code Issues Packages Projects Releases Wiki Activity

Update RELEASE.md (#263)

Browse Source 
remove HIP 6.3.1 content that is not in the release
This commit is contained in:
randyh62
2024-12-19 11:34:11 -08:00
committed by GitHub
parent 9a602592a0
commit 5f86dba37c
1 changed files with 0 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
RELEASE.md
View File

@@ -432,25 +432,11 @@ See the full [AMD SMI changelog](https://github.com/ROCm/amdsmi/blob/6.3.x/CHANG
#### Added
* New environment variables:
* `DEBUG_CLR_SYSMEM_POOL`: Used to control system pool implementation in runtime commands handling. By default, it is disabled and set to `0`.
* `DEBUG_CLR_MAX_BATCH_SIZE`: Used to limit the software batch size. By default, it is set to `1000`.
* Tracking of HSA handlers:
* Added an atomic counter to track the outstanding HSA handlers.
* Used the environment variable `DEBUG_HIP_BLOCK_SYNC` to wait on CPU for the callbacks if the number exceeds the defined value.
* Codes to capture AQL packets for HIP graph memory copy node between host and device. HIP enqueues AQL packets during graph launch.
* An activeQueues set that tracks only the queues that have a command submitted to them, which allows fast iteration in ``waitActiveStreams``.
#### Optimized
* Multi-threaded dispatches for performance improvement.
* Command submissions and processing between CPU and GPU by introducing a way to limit the software batch size.
* HSA callback performance. The HIP runtime creates and submits commands in the queue and interacts with HSA through a callback function. HIP waits for the CPU status from HSA to optimize handling of events, profiling, commands, and HSA signals for higher performance.
#### Resolved issues
* A deadlock in a specific customer application by preventing hipLaunchKernel latency degradation with number of idle streams.
* Segmentation fault caused by a race condition in multi-threaded producer/consumer scenario with ``hipMallocFromPoolAsync``.
### **HIPIFY** (18.0.0)