llvm-project

llvm-project copied to clipboard

This moves the condition into the if statement so that short-circuiting can skip it when the ID is negative. Otherwise it can use a negative index for `LocalSLocEntryTable[LastFileIDLookup.ID]`.

pfultz2

[WIP] This adds thorough handling of HIP's `__constant__` hint. On one hand, variables thus annotated are placed in the constant address space - this is handled in the ME. On...

AlexVlx

Allow compilation for Amdgpu-target Gfx1010

ghostplant

Patch cherry-picked into ATD branch

DominikAdamski

[Feature]: Better preprocessor macros to detect RDNA/CDNA family at compile time

1

### Suggestion Description As new instructions/features are added with each new arch, it is useful to know the target architecture at compile time to employ separate code paths. For example:...

benrichard-amd

[Issue]: Atomic optimizer reorder causes memory access fault in Blender

13

### Problem Description Following https://projects.blender.org/blender/blender/issues/112084, I've bisected the `rocm-6.0.x` branch and found that commit 30a3adf50e2d49dfc97c1b614d9b93638eba672d caused any Blender render (using HIP, of course) to crash with message along the lines...

GZGavinZhao

This patch updates the definition of all OpenMP loop operations (`omp.distribute`, `omp.parallel`, `omp.wsloop`, `omp.simdloop`, `omp.taskloop`) to expect a nested `omp.loopnest` and act as wrapper operations instead. Loop information is removed...

skatrak

This patch introduces the following changes: - Complete OpenMP directive sets to include masked and taskloop-related ones. - Refactor clause processing to store related operands in structures. These are combined...

skatrak

This is a simple suggested fix (might not be the correct fix) for a problem where TargetOp's nested in TargetDataOp's may have some dependencies on operations inside of the TargetDataOp...

agozillon

For simple loops, we can now choose to map `do concurrent` to either the host (i.e. `omp parallel do`) or the device (i.e. `omp target teams distribute parallel do`). In...

ergawy