This should avoid documentation related warnings and make the code and documentation consistent
Details
Details
Diff Detail
Diff Detail
Event Timeline
Comment Actions
We do not want to talk about "fixed point" (as the alternative to "floating point") iterations but "fixpoint" iterations (in the sense of https://xavierleroy.org/courses/DSSS-2017/compiler.Fixpoint.html).
That said, the current Attributor docs are wrong once:
llvm/include/llvm/Transforms/IPO/Attributor.h 2007: /// Maximum number of fixedpoint iterations.
and the code uses the wrong wording a few times:
llvm/lib/Transforms/IPO/Attributor.cpp 1331: unsigned MaxFixedPointIterations; 1333: MaxFixedPointIterations = MaxFixpointIterations.getValue(); 1335: MaxFixedPointIterations = SetFixpointIterations; 1415: } while (!Worklist.empty() && (IterationCounter++ < MaxFixedPointIterations || 1418: if (IterationCounter > MaxFixedPointIterations && !Worklist.empty()) { 1421: << ore::NV("Iterations", MaxFixedPointIterations) 1425: emitRemark<OptimizationRemarkMissed>(F, "FixedPoint", Remark); 1464: IterationCounter != MaxFixedPointIterations) { 1466: << IterationCounter << "/" << MaxFixedPointIterations
But this patch changes it in the wrong way.
Comment Actions
@jdoerfert Thank you, but I do not have commit permissions. Could you please commit this?
Comment Actions
I need a name and email to attribute the patch to. Arc is not great doing this automagically.
This comment was removed by gAlfonso-bit.
This comment was removed by gAlfonso-bit.
This comment was removed by gAlfonso-bit.