[LLVMdev] [RFC] BlockFrequency is the wrong metric; we need a new one

Duncan P. N. Exon Smith dexonsmith at apple.com
Tue Feb 4 16:23:57 PST 2014


On Feb 3, 2014, at 7:47 AM, Diego Novillo <dnovillo at google.com> wrote:

> Now, when we have sample information, we have raw block counts.
> Wouldn't it make sense to just use them in that case?

Raw block counts will give the same type of information as the current metric of
block frequency (although more accurate for the profiled use case).  If you have
profile data, running BranchProbabilityInfo followed by BlockFrequencyInfo
should recreate block frequencies roughly equivalent to the raw data scaled to a
“standard” entry frequency for each function, modulo adjustments for sanity and
precision.

Where this metric is unhelpful, raw block counts will also be unhelpful.



More information about the llvm-dev mailing list