Generalize the AArch64 .td nodes for AssertZext and AssertSext. Use
them to match the relevant pextr store instructions.
The test widen_load-2.ll requires a slight change because with the
stores gone, the remaining instructions are scheduled in a different
order.
Add test cases for SSE4 and AVX variants.
Resolves rdar://13414672.
Patch by Adam Nemet <anemet@apple.com>.
llvm-svn: 200957
mode.
Basically the idea is to transform code like this:
%idx = add nsw i32 %a, 1
%sextidx = sext i32 %idx to i64
%gep = gep i8* %myArray, i64 %sextidx
load i8* %gep
Into:
%sexta = sext i32 %a to i64
%idx = add nsw i64 %sexta, 1
%gep = gep i8* %myArray, i64 %idx
load i8* %gep
That way the computation can be folded into the addressing mode.
This transformation is done as part of the addressing mode matcher.
If the matching fails (not profitable, addressing mode not legal, etc.), the
matcher will revert the related promotions.
<rdar://problem/15519855>
llvm-svn: 200947
This solves a problem where a def machine operand has no uses but has
not been marked dead. In this case, the initial RP analysis was being
extra precise and determining from LiveIntervals the the register was
actually dead. This caused us to omit the register from the RP
tracker's block live out. That's all good, but the per-instruction
summary still accounted for it as a valid def. This could cause an
assertion in the tracker later when we underflow pressure.
This is from a bug report on an out-of-tree target. It is not
reproducible on well-behaved targets. I'm just making an obvious fix
without unit test.
llvm-svn: 200941
There was a problem with the old pattern, so we were copying some
larger immediates into registers when we could have been encoding
them in the instruction.
llvm-svn: 200932
In a previous commit (r199818) we added a const_cast to an existing
subtarget info instead of creating a new one so that we could reuse
it when creating the TargetAsmParser for parsing inline assembly.
This cast was necessary because we needed to reuse the existing STI
to avoid generating incorrect code when the inline asm contained
mode-switching directives (e.g. .code 16).
The root cause of the failure was that there was an implicit sharing
of the STI between the parser and the MCCodeEmitter. To fix a
different but related issue, we now explicitly pass the STI to the
MCCodeEmitter (see commits r200345-r200351).
The const_cast is no longer necessary and we can now create a fresh
STI for the inline asm parser to use.
Differential Revision: http://llvm-reviews.chandlerc.com/D2709
llvm-svn: 200929
The most important part of this is probably adding any cost at all for
operations like zext <8 x i8> to <8 x i32>. Before they were being
recorded as extremely costly (24, I believe) which made LLVM fall back
on a 4-wide vectorisation of a loop.
It also rebalances the values for sext, zext and trunc. Lacking any
other sane metric that might work across CPU microarchitectures I went
for instructions. This seems to be in reasonable accord with the rest
of the table (sitofp, ...) though no doubt at least one value is
sub-optimal for some bizarre reason.
Finally, separate AVX and AVX2 values are provided where appropriate.
The CodeGen is quite different in many cases.
rdar://problem/15981990
llvm-svn: 200928
The aim in this patch is to reduce work that VirtRegRewriter needs to do when
telling MachineRegisterInfo which physregs are in use. Up until now
VirtRegRewriter::rewrite has been doing rewriting and populating def info and
then proceeding to set whether a physreg is used based this info for every
physreg that the target provides. This can be expensive when a target has an
unusually high number of supported physregs, and is a noticeable chunk of
compile time for small programs on such targets.
So to reduce compile time, this patch simply adds the use of a SparseSet to the
rewrite function that is used to flag each physreg that is encountered in a
MachineFunction. Afterward, rather than iterating over the set of all physregs
for a given target to set the physregs used in MachineRegisterInfo, the new way
is to iterate over the set of physregs that were actually encountered and set
in the SparseSet. This improves compile time because the existing rewrite
function was iterating over all MachineOperands already, and because the
iterations afterward to setPhysRegUsed is reduced by use of the SparseSet data.
llvm-svn: 200919
I believe VZEXT_MOVL means "zero all vector elements except the first" (and
should have identical input & output types) whereas VZEXT means "zero extend
each element of a vector (discarding higher elements if necessary)".
For example:
(v4i32 (vzext (v16i8 ...)))
should zero extend the low 4 bytes of the incoming vector to 32-bits,
discarding higher bytes.
However, somewhere in the past, these two concepts had become confused, even
leading to a nonsensical VSEXT_MOVL.
This re-merges the nodes where appropriate (all VSEXT_MOVL -> VSEXT, VZEXT_MOVL
-> VZEXT when it's an actual extension).
rdar://problem/15981990
llvm-svn: 200918
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
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
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
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
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
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
This patch adds NaCl target for Mips. It also forbids indexed loads and
stores if the target is NaCl.
Patch by Sasa Stankovic.
Differential Revision: http://llvm-reviews.chandlerc.com/D2690
llvm-svn: 200855
Summary:
The check performed in the comparator is invalid, as some STL
implementations enforce strict weak ordering by calling the comparator with the
same value. This check was also in a wrong place: the assertion would only fire
when -help was used. The new check is performed each time the category is
registered (we are not going to have thousands of them, so it's fine to do it in
O(N^2)).
Reviewers: jordan_rose
Reviewed By: jordan_rose
CC: cfe-commits, alexmc
Differential Revision: http://llvm-reviews.chandlerc.com/D2699
llvm-svn: 200853
In Thumb1 mode, bl instruction might be selected for branches between
basic blocks in the function if the offset is greater than 2KB.
However, this might cause SEGV because the destination symbol
is not marked as thumb function and the execution mode will be reset
to ARM mode.
Since we are sure that these symbols are in the same data fragment, we
can simply resolve these local symbols, and don't emit any relocation
information for this bl instruction.
llvm-svn: 200842
Manman Ren