Amd/dev/brcahoon/swdev 563385 7.2 #707
Closed
Conversation
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
…) (#541) This PR introduces a new pass "lower-workdistribute" Fortran array statements are lowered to fir as fir.do_loop unordered. "lower-workdistribute" pass works mainly on identifying "fir.do_loop unordered" that is nested in target{teams{workdistribute{fir.do_loop unordered}}} and lowers it to target{teams{parallel{wsloop{loop_nest}}}}. It hoists all the other ops outside target region. Relaces heap allocation on target with omp.target_allocmem and deallocation with omp.target_freemem from host. Also replaces runtime function "Assign" with omp.target_memcpy from host. This pass implements following rewrites and optimisations: - **FissionWorkdistribute**: finds the parallelizable ops within teams {workdistribute} region and moves them to their own teams{workdistribute} region. - **WorkdistributeRuntimeCallLower**: finds the FortranAAssign calls nested in teams {workdistribute{}} and lowers it to unordered do loop if src is scalar and dest is array. Other runtime calls are not handled currently. - **WorkdistributeDoLower**: finds the fir.do_loop unoredered nested in teams {workdistribute{fir.do_loop unoredered}} and lowers it to teams {parallel { distribute {wsloop {loop_nest}}}}. - **TeamsWorkdistributeToSingle**: hoists all the ops inside teams {workdistribute{}} before teams op. The work in this PR is C-P and updated from @ivanradanov commits from coexecute implementation: [flang_workdistribute_iwomp_2024](https://github.com/ivanradanov/llvm-project/commits/flang_workdistribute_iwomp_2024) Paper related to this work by @ivanradanov ["Automatic Parallelization and OpenMP Offloadingof Fortran Array Notation"](https://www.osti.gov/servlets/purl/[2449728](https://www.osti.gov/servlets/purl/2449728))
will work on script changes for aomp, and npsdb after it lands
- add_subdirectory(utils) in offload CMakeLists.txt - usage of new macro add_openmp_util to install utils into llvm/bin
rocm 7.2 changed pci layout/info
really messes up xnack=1 performance
necessitates forced path to numactl
-nr use numactl ROCR_VISIBLE_DEVICES
-nm use numactl OMPI_COMM_WORLD_LOCAL_RANK
…#167486) Attempt to only define used subregisters when creating IMPLICIT_DEF fix ups for live interval subranges. This avoids the appearance at the MIR level of entire (wide) registers becoming live rather than relying only on transient LiveIntervals dead definitions for unused subregisters. (cherry picked from commit b1c4b55)
Collaborator
Author
|
oops, i stacked this on another PR, will redo it |
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.
CP'ed from amd-mainline version: #638