programs on targets with large register files. The root of the compile time
overhead was in the use of llvm::SmallVector to hold PhysRegEntries, which
resulted in slow-down from calling llvm::SmallVector::assign(N, 0). In contrast
std::vector uses the faster __platform_bzero to zero out primitive buffers when
assign is called, while SmallVector uses an iterator.
The fix for this was simply to replace the SmallVector with a dynamically
allocated buffer and to initialize or reinitialize the buffer based on the
total registers that the target architecture requires. The changes support
cases where a pass manager may be reused for different targets, and note that
the PhysRegEntries is allocated using calloc mainly for good for, and also to
quite tools like Valgrind (see comments for more info on this).
There is an rdar to track the fact that SmallVector doesn't have platform
specific speedup optimizations inside of it for things like this, and I'll
create a bugzilla entry at some point soon as well.
TL;DR: This fix replaces the expensive llvm::SmallVector<unsigned
char>::assign(N, 0) with a call to calloc for N bytes which is much faster
because SmallVector's assign uses iterators.
llvm-svn: 200917
large register files. The omission of Queries.clear() is perfectly safe because
LiveIntervalUnion::Query doesn't contain any data that needs freeing and
because LiveRegMatrix::runOnFunction happens to reset the OwningArrayPtr
holding Queries every time it is run, so there's no need to zero out the
queries either. Not having to do this for very large numbers of physregs
is a noticeable constant cost reduction in compilation of small programs.
llvm-svn: 200913
clang/test/FixIt/fixit-unicode-with-utf8-output.c has begun complained since LLVM r200885.
Although it is changes for StringRef, it brought LLVM_ON_WIN32 to Support/Locale.cpp.
Before r200885, LLVM_ON_WIN32 was undefined in Locale.cpp!
FIXME: We should consider i18n on win32.
llvm-svn: 200909
build but spectacularly changed behavior of the C++98 build. =]
This shows my one problem with not having unittests -- basic API
expectations aren't well exercised by the integration tests because they
*happen* to not come up, even though they might later. I'll probably add
a basic unittest to complement the integration testing later, but
I wanted to revive the bots.
llvm-svn: 200905
The primary motivation for this pass is to separate the call graph
analysis used by the new pass manager's CGSCC pass management from the
existing call graph analysis pass. That analysis pass is (somewhat
unfortunately) over-constrained by the existing CallGraphSCCPassManager
requirements. Those requirements make it *really* hard to cleanly layer
the needed functionality for the new pass manager on top of the existing
analysis.
However, there are also a bunch of things that the pass manager would
specifically benefit from doing differently from the existing call graph
analysis, and this new implementation tries to address several of them:
- Be lazy about scanning function definitions. The existing pass eagerly
scans the entire module to build the initial graph. This new pass is
significantly more lazy, and I plan to push this even further to
maximize locality during CGSCC walks.
- Don't use a single synthetic node to partition functions with an
indirect call from functions whose address is taken. This node creates
a huge choke-point which would preclude good parallelization across
the fanout of the SCC graph when we got to the point of looking at
such changes to LLVM.
- Use a memory dense and lightweight representation of the call graph
rather than value handles and tracking call instructions. This will
require explicit update calls instead of some updates working
transparently, but should end up being significantly more efficient.
The explicit update calls ended up being needed in many cases for the
existing call graph so we don't really lose anything.
- Doesn't explicitly model SCCs and thus doesn't provide an "identity"
for an SCC which is stable across updates. This is essential for the
new pass manager to work correctly.
- Only form the graph necessary for traversing all of the functions in
an SCC friendly order. This is a much simpler graph structure and
should be more memory dense. It does limit the ways in which it is
appropriate to use this analysis. I wish I had a better name than
"call graph". I've commented extensively this aspect.
This is still very much a WIP, in fact it is really just the initial
bits. But it is about the fourth version of the initial bits that I've
implemented with each of the others running into really frustrating
problms. This looks like it will actually work and I'd like to split the
actual complexity across commits for the sake of my reviewers. =] The
rest of the implementation along with lots of wiring will follow
somewhat more rapidly now that there is a good path forward.
Naturally, this doesn't impact any of the existing optimizer. This code
is specific to the new pass manager.
A bunch of thanks are deserved for the various folks that have helped
with the design of this, especially Nick Lewycky who actually sat with
me to go through the fundamentals of the final version here.
llvm-svn: 200903
necessary until we add analyses to the driver, but I have such an
analysis ready and wanted to split this out. This is actually exercised
by the existing tests of the new pass manager as the analysis managers
are cross-checked and validated by the function and module managers.
llvm-svn: 200901
During DAGCombine visitShiftByConstant assumes that certain binary operations
with only constant operands can always be folded successfully. This is no longer
true when the constant is opaque. This commit fixes visitShiftByConstant by not
performing the optimization for opaque constants. Otherwise we would end up in
an infinite DAGCombine loop.
llvm-svn: 200900
In the following code:
struct A { static const int sz; };
template<class T> void f() { T arr[A::sz]; }
the array 'arr' is represented as a variable size array in the template.
If 'A::sz' gets value below in the translation unit, the array in
instantiation can turn into constant size array.
This change fixes PR18633.
Differential Revision: http://llvm-reviews.chandlerc.com/D2688
llvm-svn: 200899
225 is the default value of inline-threshold. This change will make sure
we have the same inlining behavior as prior to r200886.
As Chandler points out, even though we don't have code in our testing
suite that uses cold attribute, there are larger applications that do
use cold attribute.
r200886 + this commit intend to keep the same behavior as prior to r200886.
We can later on tune the inlinecold-threshold.
The main purpose of r200886 is to help performance of instrumentation based
PGO before we actually hook up inliner with analysis passes such as BPI and BFI.
For instrumentation based PGO, we try to increase inlining of hot functions and
reduce inlining of cold functions by setting inlinecold-threshold.
Another option suggested by Chandler is to use a boolean flag that controls
if we should use OptSizeThreshold for cold functions. The default value
of the boolean flag should not change the current behavior. But it gives us
less freedom in controlling inlining of cold functions.
llvm-svn: 200898
using-declaration, and they declare the same function (either because
the using-declaration is in the same namespace as the declaration it
imports, or because they're both extern "C"), they do not conflict.
llvm-svn: 200897
This is a nop. doesSectionRequireSymbols is only used from
isSymbolLinkerVisible. isSymbolLinkerVisible only use from ELF was in
if (!Asm.isSymbolLinkerVisible(Symbol) && !Symbol.isUndefined())
return false;
if (Symbol.isTemporary())
return false;
If the symbol is a temporary this code returns false and it is irrelevant if
we take the first if or not. If the symbol is not a temporary,
Asm.isSymbolLinkerVisible returns true without ever calling
doesSectionRequireSymbols.
This was an horrible leftover from when support for ELF was first added.
llvm-svn: 200894
Ideally only those transform passes that run at -O0 remain enabled,
in reality we get as close as we reasonably can.
Passes are responsible for disabling themselves, it's not the job of
the pass manager to do it for them.
llvm-svn: 200892
Added command line option inlinecold-threshold to set threshold for inlining
functions with cold attribute. Listen to the cold attribute when it would
decrease the inline threshold.
llvm-svn: 200886
This option will:
- load the given pch file
- verify it is not out of date by stat'ing dependencies, and
- return 0 on success and non-zero on error
llvm-svn: 200884
find a register.
The idea is to choose a color for the variable that cannot be allocated and
recolor its interferences around. Unlike the current register allocation scheme,
it is allowed to change the color of an already assigned (but maybe not
splittable or spillable) live interval while propagating this change to its
neighbors.
In other word, there are two things that may help finding an available color:
- Already assigned variables (RS_Done) can be recolored to different color.
- The recoloring allows to catch solutions that needs to touch more that just
the neighbors of the current allocated variable.
E.g.,
vA can use {R1, R2 }
vB can use { R2, R3}
vC can use {R1 }
Where vA, vB, and vC cannot be split anymore (they are reloads for instance) and
they all interfere.
vA is assigned R1
vB is assigned R2
vC tries to evict vA but vA is already done.
=> Regular register allocation heuristic fails.
Last chance recoloring kicks in:
vC does as if vA was evicted => vC uses R1.
vC is marked as fixed.
vA needs to find a color.
None are available.
vA cannot evict vC: vC is a fixed virtual register now.
vA does as if vB was evicted => vA uses R2.
vB needs to find a color.
R3 is available.
Recoloring => vC = R1, vA = R2, vB = R3.
<rdar://problem/15947839>
llvm-svn: 200883
We now properly detect when a result object has an immediate output stream and don't echo the results a second time.
<rdar://problem/15954906>
llvm-svn: 200882
I think this was just over-eagerness on my part. The analysis results
need to often be non-const because they need to (in some cases at least)
be updated by the transformation pass in order to remain correct. It
also makes lazy analyses (a common case) needlessly annoying to write in
order to make their entire state mutable.
llvm-svn: 200881
Also emit the "Executing commands" message so it properly only comes out when desired and so it comes out in the right place.
<rdar://problem/15992208>
llvm-svn: 200875
We collect a maximal function count among all functions in the pgo data file.
For functions that are hot, we set its InlineHint attribute. For functions that
are cold, we set its Cold attribute.
We currently treat functions with >= 30% of the maximal function count as hot
and functions with <= 1% of the maximal function count are treated as cold.
These two numbers are from preliminary tuning on SPEC.
This commit should not affect non-PGO builds and should boost performance on
instrumentation based PGO.
llvm-svn: 200874
Function references always use $1? like function pointers and never $E?
like var decl references. Static methods are mangled like function
pointers.
llvm-svn: 200869
Because in C++, "anonymous" doesn't mean "nameless" for records. In
other words, RecordDecl::isAnonymousStructOrUnion only returns true if
the record lacks a name *and* is not used as the type in an object's
declaration.
llvm-svn: 200868
Puyan Lotfi