Skip to content

[HIPIFY][merge] merge amd-develop into amd-staging#2321

Merged
ronlieb merged 15 commits intoamd-stagingfrom
upstream_merge_2301262305
Jan 23, 2026
Merged

[HIPIFY][merge] merge amd-develop into amd-staging#2321
ronlieb merged 15 commits intoamd-stagingfrom
upstream_merge_2301262305

Conversation

@ranapratap55
Copy link
Collaborator

No description provided.

emankov and others added 15 commits January 20, 2026 21:08
[Synopsis]
+ Transformation maps reside in Stack and occupy up to `64kB` each of stack memory now
+ That leads to multiple compiler warnings about possible Stack Overflow

[Solution]
+ Move maps' initialization logic into a lambda. This prevents the compiler from creating a massive temporary array on the stack, which was the root cause of the "excessive stack usage" warning.
[HIPIFY][fix] `Stack Overflow` possible Issues - Part 3 - `Runtime`
+ Updated the regenerated `hipify-perl`, and `DNN` `CUDA2HIP` documentation
[HIPIFY][DNN] `cuDNN 9.17.0` is the latest supported version
[Reason]
+ The framework `Caffe2` is deprecated. While some legacy code exists within older versions of `PyTorch`, it is not a supported path for new development. The old `rocm-caffe2` repositories you might find on GitHub (targeting ROCm 1.7 or Ubuntu 16.04) are abandoned and will not function on modern hardware (like Instinct `MI300` or `Ryzen AI`) or modern ROCm versions.
[HIPIFY][CAFFE2] Total removal
[Synopsis]
+ Transformation maps reside in Stack and occupy up to `64kB` each of stack memory now
+ That leads to multiple compiler warnings about possible Stack Overflow

[Solution]
+ Move maps' initialization logic into a lambda. This prevents the compiler from creating a massive temporary array on the stack, which was the root cause of the "excessive stack usage" warning.

[Fix]
+ Removed wrong duplicates of `CU_DEVICE_P2P_ATTRIBUTE_ARRAY_ACCESS_ACCESS_SUPPORTED`, `CU_COREDUMP_LIGHTWEIGHT`, and `CUDA_ERROR_ILLEGAL_STATE`
[HIPIFY][fix] `Stack Overflow` possible Issues - Part 4 - `Driver`
+ Regenerated and updated the corresponding `CUDA2HIP` tables
[HIPIFY][fix] Remove duplicates from maps
[Synopsis]
+ Transformation maps reside in Stack and occupy up to `64kB` each of stack memory now
+ That leads to multiple compiler warnings about possible Stack Overflow

[Solution]
+ Move maps' initialization logic into a lambda. This prevents the compiler from creating a massive temporary array on the stack, which was the root cause of the "excessive stack usage" warning.
[HIPIFY][fix] `Stack Overflow` possible Issues - Part 5 - `DNN`
… - Part 2 - `Data Types`

+ Updated the regenerated `hipify-perl` and `Driver` `CUDA2HIP` docs accordingly
[HIPIFY][CUDA 13.1][Driver] `CUDA 13.1.0` support - Step 1 - `Driver` - Part 2 - `Data Types`
@ranapratap55 ranapratap55 self-assigned this Jan 23, 2026
@ranapratap55 ranapratap55 requested review from a team, emankov and searlmc1 as code owners January 23, 2026 17:52
@emankov emankov requested a review from SyamaAmd January 23, 2026 18:04
Copy link

@ronlieb ronlieb left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

passed all theRock builds

@ronlieb ronlieb merged commit f1328e9 into amd-staging Jan 23, 2026
41 of 43 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants