Suppose an inline instance has hot total sample count but 0 entry count, and
it is an indirect call target. If the indirect call has no other call target
and inline instance associated with it and it is promoted, currently the
conditional branch generated by indirect call promotion will have invalid
branch profile which is !{!"branch_weights", i32 0, i32 0} -- because the
entry count of the promoted target is 0 and the total entry count of all
targets is also 0. This caused a SEGV in Control Height Reduction and may
cause problem in other passes.
Function entry count of an inline instance is computed by a heuristic --
using either the sample of the starting line or starting inner inline
instance. The patch changes the heuristic a little bit so that when total
sample count is larger than 0, the computed entry count will be at least 1.
Then the new branch profile will be !{!"branch_weights", i32 1, i32 0}.
Differential Revision: https://reviews.llvm.org/D72790
As it's causing some bot failures (and per request from kbarton).
This reverts commit r358543/ab70da07286e618016e78247e4a24fcb84077fda.
llvm-svn: 358546
The patch saves a function offset table which maps function name index to the
offset of its function profile to the start of the binary profile. By using
the function offset table, for those function profiles which will not be used
when compiling a module, the profile reader does't have to read them. For
profile size around 10~20M, it saves ~10% compile time.
Differential Revision: https://reviews.llvm.org/D51863
llvm-svn: 342283
The patch tries to make sample profile loader independent of profile format
change. It moves compact format related code into FunctionSamples and
SampleProfileReader classes, and sample profile loader only has to interact
with those two classes and will be unaware of profile format changes.
The cleanup also contain some fixes to further remove the difference between
compactbinary format and binary format. After the cleanup using different
formats originated from the same profile will generate the same binaries,
which we verified by compiling two large server benchmarks w/wo thinlto.
Differential Revision: https://reviews.llvm.org/D51643
llvm-svn: 341591