The fundamental problem is that SROA didn't allow for overly wide loads
where the bits past the end of the alloca were masked away and the load
was sufficiently aligned to ensure there is no risk of page fault, or
other trapping behavior. With such widened loads, SROA would delete the
load entirely rather than clamping it to the size of the alloca in order
to allow mem2reg to fire. This was exposed by a test case that neatly
arranged for GVN to run first, widening certain loads, followed by an
inline step, and then SROA which miscompiles the code. However, I see no
reason why this hasn't been plaguing us in other contexts. It seems
deeply broken.
Diagnosing all of the above took all of 10 minutes of debugging. The
really annoying aspect is that fixing this completely breaks the pass.
;] There was an implicit reliance on the fact that no loads or stores
extended past the alloca once we decided to rewrite them in the final
stage of SROA. This was used to encode information about whether the
loads and stores had been split across multiple partitions of the
original alloca. That required threading explicit tracking of whether
a *use* of a partition is split across multiple partitions.
Once that was done, another problem arose: we allowed splitting of
integer loads and stores iff they were loads and stores to the entire
alloca. This is a really arbitrary limitation, and splitting at least
some integer loads and stores is crucial to maximize promotion
opportunities. My first attempt was to start removing the restriction
entirely, but currently that does Very Bad Things by causing *many*
common alloca patterns to be fully decomposed into i8 operations and
lots of or-ing together to produce larger integers on demand. The code
bloat is terrifying. That is still the right end-goal, but substantial
work must be done to either merge partitions or ensure that small i8
values are eagerly merged in some other pass. Sadly, figuring all this
out took essentially all the time and effort here.
So the end result is that we allow splitting only when the load or store
at least covers the alloca. That ensures widened loads and stores don't
hurt SROA, and that we don't rampantly decompose operations more than we
have previously.
All of this was already fairly well tested, and so I've just updated the
tests to cover the wide load behavior. I can add a test that crafts the
pass ordering magic which caused the original PR, but that seems really
brittle and to provide little benefit. The fundamental problem is that
widened loads should Just Work.
llvm-svn: 177055
isa and a cast inside the assert. The efficiency concern isn't really
important here. The code should likely be cleaned up a bit more,
especially getting a message into the assert.
Please review Rafael.
llvm-svn: 177053
template instantiation will still consider them to be definitions
if we instantiate the containing class before we get around
to parsing the friend.
This seems like a legitimate use of "late template parsed" to me,
but I'd appreciate it if someone responsible for the MS feature
would look over this.
This file already appears to access AST nodes directly, which
is arguably not kosher in the parser, but the performance of this
path matters enough that perpetuating the sin is justifiable.
Probably we ought to reconsider this policy for very simple
manipulations like this.
The reason this entire thing is necessary is that
function template instantiation plays some very gross games
in order to not associate an instantiated function template
with the class it came from unless it's a definition, and
the reason *that's* necessary is that the AST currently
cannot represent the instantiation history of individual
function template declarations, but instead tracks it in
common for the entire function template. That probably
prevents us from correctly reporting ill-formed calls to
ambiguously instantiated friend function templates.
rdar://12350696
llvm-svn: 177003
constructs default arguments. It can now take default arguments from
cl::opt'ions. Add a new -default-gcov-version=... option, and actually test it!
Sink the reverse-order of the version into GCOVProfiling, hiding it from our
users.
llvm-svn: 177002
Before this patch we would compute the linkage lazily and cache it. When the
AST was modified in ways that could change the value, we would invalidate the
cache.
That was fairly brittle, since any code could ask for the a linkage before
the correct value was available.
We should change the API to one where the linkage is computed explicitly and
trying to get it when it is not available asserts.
This patch is a first step in that direction. We still compute the linkage
lazily, but instead of invalidating a cache, we assert that the AST
modifications didn't change the result.
llvm-svn: 176999
emitProfileNotes(), similar to emitProfileArcs(). Also update its comment.
Also add a comment on Version[4] (there will be another comment in clang later),
and compress lines that exceeded 80 columns.
llvm-svn: 176994
Evgeniy Stepanov