In PIC mode we were previously computing global variable addresses (or GOT
entry addresses) by adding the PC, the PC-relative GOT displacement and
the GOT-relative symbol/GOT entry displacement. Because the latter two
displacements are fixed, we ended up performing one more addition than
necessary.
This change causes us to compute addresses using a single PC-relative
displacement, resulting in a shorter code sequence. This reduces code size
by about 4% in a recent build of Chromium for Android.
As a result of this change we no longer need to compute the GOT base address
in the ARM backend, which allows us to remove the Global Base Reg pass and
SDAG lowering for the GOT.
We also now no longer use the GOT when addressing a symbol which is known
to be defined in the same linkage unit. Specifically, the symbol must have
either hidden visibility or a strong definition in the current module in
order to not use the the GOT.
This is a change from the previous behaviour where we would use the GOT to
address externally visible symbols defined in the same module. I think the
only cases where this could matter are cases involving symbol interposition,
but we don't really support that well anyway.
Differential Revision: http://reviews.llvm.org/D13650
llvm-svn: 251322
This patch enables the ptrace syscall interceptors for arm and adds support
for both PTRACE_GETVFPREGS and PTRACE_SETVFPREGS used to get the VFP register
from ARM.
The ptrace tests is also updated with arm and PTRACE_GETVFPREGS tests.
llvm-svn: 251321
This issue is triggered in PGO mode when bootstrapping LLVM. It seems that it is not guaranteed that edge weights are always greater than zero which are read from profile data.
llvm-svn: 251317
The problem was that the @skipIfNoSBHeaders on darwin was trying to use self.lib_dir when it hadn't been set yet.
I looked at the code and places were required to set "self.lib_dir" for no real reason as all places that used it just used the LLDB_LIB_DIR environment variable. So I removed all uses of self.lib_dir and replaced them to use 'os.environ["LLDB_LIB_DIR"]'. Did the same for self.implib_dir.
llvm-svn: 251315
The regex for -isystem matching is broken. -[D,I,Usystem] matches "-D", "-,",
"-I", "-U", "-s" "-y", etc. Besides that, "-isystem /foo" gets interpreted as
"-i" with a non-empty value "system" and thus the next "/foo" argument is not
read. This patch corrects the regex.
This fixes PR13237 <https://llvm.org/bugs/show_bug.cgi?id=13237>.
A patch by Peter Wu!
Differential Revision: http://reviews.llvm.org/D13800
llvm-svn: 251312
Also since we always read in the DWARF data or mmap it, we don't need to make a copy of the strings for the directories and file names, we can just store "cosnt char *" values. Every place that uses the prologues use them temporarily and then throw them away so no one is expecting the directory and filename strings to live longer than the parse functions.
llvm-svn: 251310
Python3 has no analogue to sys.maxint since ints in Python 3 have
arbitrary size. However, the distinction was not actually important
in any of these cases, and in a few cases using maxint was already
a bug to begin with.
llvm-svn: 251306
This avoids the need to query the PC for private resume operations (public resumes have the PC
from the bigger jStopInfo packet) and speeds up the stepping on an android target by about 10%
(it some cases even more).
llvm-svn: 251301
There was a threading issue in the ICF code for COFF. That seems like
a venign bug in the sense that it doesn't produce an incorrect output,
but it oftentimes misses reducible sections. As a result, mergeable
sections could remain in outputs, which makes the output nondeterministic.
Basically the algorithm we are using for ICF is this: We group sections
so that identical sections will eventually be in the same group. Initially,
all sections are in one group. We split the group by relocation targets
until we get a convergence (if relocation targets are in different gruops,
the sections are different). Once a group is split, they will never be
merged.
Each section has a group ID. That variable itself is atomic, so there's
no threading issue at the level that we can use thread sanitizer.
The point is, when we split a group, we re-assign new group IDs to group
of sections. That are multiple separate writes to atomic varaibles.
Thus, splitting a group is not an atomic operation, and there's a small
chance that the other thread observes inconsistent group IDs.
Over-splitting is always "safe", so it will never create incorrect output.
I suspect that the nondeterminism stems from that point. However, I
cannot prove or fix that at this moment, so I'm going to avoid using
threads here.
llvm-svn: 251300
Instead of XFAIL-ing the tests with the wrong usage of the "interrupt"
attribute, we should check that we emit the correct error messages to
the user.
llvm-svn: 251295
Even though we may not know the value of the shifter operand, it's possible we know the shifter operand is non-zero. This can allow us to infer more known bits - for example:
%1 = load %p !range {1, 5}
%2 = shl %q, %1
We don't know %1, but we do know that it is nonzero so %2[0] is known zero, and importantly %2 is known non-zero.
Calling isKnownNonZero is nontrivially expensive so use an Optional to run it lazily and cache its result.
llvm-svn: 251294
Adds option -c <x,y,z> to the 'language renderscript kernel breakpoint set' command.
Breaks only on the invocation of the kernel with specified coordinate.
Implemented by adding a callback to the kernel breakpoint which checks the coordinates of every invocation.
llvm-svn: 251293
Summary:
Replace (const SCEVAddRecExpr *) with cast<SCEVAddRecExpr>.
Rename SCEVApplyRewriter to SCEVLoopAddRecRewriter (which is a more
appropriate name) since the description is "takes a scalar evolution
expression and applies the Map (Loop -> SCEV) to all AddRecExprs."
Subscribers: llvm-commits, sanjoy
Differential Revision: http://reviews.llvm.org/D14065
llvm-svn: 251292
Vectorization of memory instruction (Load/Store) is possible when the pointer is coming from GEP. The GEP analysis allows to estimate the profit.
In some cases we have a "bitcast" between GEP and memory instruction.
I added code that skips the "bitcast".
http://reviews.llvm.org/D13886
llvm-svn: 251291
Summary:
This patch adds support for using the "interrupt" attribute on Mips
for interrupt handling functions. At this time only mips32r2+ with the
o32 ABI with the static relocation model is supported. Unsupported
configurations will be rejected
Patch by Simon Dardis (+ clang-format & some trivial changes to follow the
LLVM coding standards by me).
Reviewers: mpf, dsanders
Subscribers: dsanders, vkalintiris, llvm-commits
Differential Revision: http://reviews.llvm.org/D10768
llvm-svn: 251286
Summary:
Add a SCEVRewriteVisitor class which contains the common
visiting patterns used when rewriting SCEVs.
SCEVParameterRewriter and SCEVApplyRewriter now inherit
from SCEVRewriteVisitor (and are therefore much simpler).
Reviewers: anemet, mzolotukhin, sanjoy
Subscribers: rengolin, llvm-commits, sanjoy
Differential Revision: http://reviews.llvm.org/D13242
llvm-svn: 251283
Summary:
InstCombine tries to transform GEP(PHI(GEP1, GEP2, ..)) into GEP(GEP(PHI(...))
when possible. However, this may leave the old PHI node around. Even if we
do end up folding the GEPs, having an extra PHI node might not be beneficial.
This change makes the transformation more conservative. We now only do this if
the PHI has only one use, and can therefore be removed after the transformation.
Reviewers: jmolloy, majnemer
Subscribers: mcrosier, mssimpso, llvm-commits
Differential Revision: http://reviews.llvm.org/D13887
llvm-svn: 251281
Summary:
Per discussions on the mailing list, I have implemented a decorator which annotates individual
test methods with categories. I have used this framework to replace the '-a' and '+a'
command-line switches (now '-G pyapi' and '--skip-category pyapi') and the @python_api_test
decorator (now @add_test_categories('pyapi')). The test suite now gives an error message
suggesting the new options if the user specifies the deprecated +/-a switches. If the general
direction is good, I will follow this up with other switches.
Reviewers: tberghammer, tfiala, granata.enrico, zturner
Subscribers: lldb-commits
Differential Revision: http://reviews.llvm.org/D14020
llvm-svn: 251277
the module pointer type passed in by the user.
The previous ownership scheme, where the user pointer was always moved into a
std::shared_ptr, breaks if the user passes in a raw pointer.
Discovered while working on the Orc C API, which should be landing shortly.
I expect to include a test-case with that.
llvm-svn: 251273
GNU as and Darwin give the various binary operators different
precedence. LLVM's MC supported the Darwin semantics but not the GNU
semantics.
This fixes PR25311.
llvm-svn: 251271
Peter Collingbourne