[AMDGPU][InstCombine] Fold ballot intrinsic based on llvm.assume hints#160670

Open

TejaX-Alaghari wants to merge 16 commits intollvm:mainfrom

TejaX-Alaghari:DA_Assume

Commits on Nov 13, 2025

[InstCombine] Optimize AMDGPU ballot + assume uniformity patterns
TejaX-Alaghari
committed
[InstCombine] Add constant folding for AMDGPU ballot intrinsics
TejaX-Alaghari
committed
[InstCombine] Implement generic assume-based uniformity optimization
TejaX-Alaghari
committed
[InstCombine] Add focused assume-based optimizations
TejaX-Alaghari
committed
Address @ssahasra's review feedback
TejaX-Alaghari
committed
Address feedback on the location of the opt
TejaX-Alaghari
committed
Refactored the ballot optimization condition into propagateEquality method
TejaX-Alaghari
committed
Implement reviewer's suggestions to -
TejaX-Alaghari
committed
Moved the assume based ballot folding logic to AMDGPUInstCombineIntrinsic.cpp
TejaX-Alaghari
committed
Simplified the logic to check for CompareValue and return nullptr when Ballot arg is modified
TejaX-Alaghari
committed
Skip the optimization when ballot size < wave size
TejaX-Alaghari
committed
Only proceed with the optiization if ballot arg is an instruction
TejaX-Alaghari
committed
Add additional check to demonstrate E2E impact of this optimization
TejaX-Alaghari
committed
Address feedback:1. Add the condition to make sure ballot width matches wave sixe
TejaX-Alaghari
committed
Removed unnecessary temp varaible for holding wave front size
TejaX-Alaghari
committed
Skip only constant args
TejaX-Alaghari
committed