Reimplement heuristic for estimating complete-unroll optimization effects.
Summary:
This patch reimplements heuristic that tries to estimate optimization beneftis
from complete loop unrolling.
In this patch I kept the minimal changes - e.g. I removed code handling
branches and folding compares. That's a promising area, but now there
are too many questions to discuss before we can enable it.
Petar Jovanovic [Tue, 12 May 2015 17:14:05 +0000 (17:14 +0000)]
[Mips] Return false for isFPCloseToIncomingSP()
On Mips, frame pointer points to the same side of the frame as the stack
pointer. This function is used to decide where to put register scavenging
spill slot. So far, it was put on the wrong side of the frame, and thus it
was too far away from $fp when frame was larger than 2^15 bytes.
Tom Stellard [Tue, 12 May 2015 17:13:02 +0000 (17:13 +0000)]
R600/SI: add pass to mark CF live ranges as non-spillable
Spilling can insert instructions almost anywhere, and this can mess
up control flow lowering in a multitude of ways, due to instruction
reordering. Let's sort this out the easy way: never spill registers
involved with control flow, i.e. saved EXEC masks.
Unfortunately, this does not work at all with optimizations disabled,
as the register allocator ignores spill weights. This should be
addressed in a future commit.
The test was reduced from the "stacks" shader of [1]. Some issues
trigger the machine verifier while another one is checked manually.
[1] http://madebyevan.com/webgl-path-tracing/
v2: only insert pass with optimizations enabled, merge test runs.
Sunil Srivastava [Tue, 12 May 2015 16:47:30 +0000 (16:47 +0000)]
Changed renaming of local symbols by inserting a dot vefore the numeric suffix.
One code change and several test changes to match that
details in http://reviews.llvm.org/D9481
Keith Walker [Tue, 12 May 2015 15:25:08 +0000 (15:25 +0000)]
[DWARF] Add CIE header fields address_size and segment_size when generating dwarf-4
The DWARF-4 specification added 2 new fields in the CIE header called
address_size and segment_size.
Create these 2 new fields when generating dwarf-4 CIE entries, print out
the new fields when dumping the CIE and update tests
Tom Stellard [Tue, 12 May 2015 15:00:53 +0000 (15:00 +0000)]
R600/SI: Update tablegen defs to avoid restoring spilled sgprs to m0
We had code to do this in SIRegisterInfo::eliminateFrameIndex(), but
it is easier to just change the definition of SI_SPILL_S32_RESTORE to
only allow numbered sgprs.
Tom Stellard [Tue, 12 May 2015 14:18:14 +0000 (14:18 +0000)]
R600/SI: Remove explicit m0 operand from s_sendmsg
Instead add m0 as an implicit operand. This allows us to avoid using
the M0Reg register class and eliminates a number of unnecessary spills
when using s_sendmsg instructions. This impacts one shader in the
shader-db:
AVX-512, X86: Added lowering for shift operations for SKX.
The other changes in the LowerShift() are not functional,
just to make the code more convenient.
So, the functional changes for SKX only.
John Brawn [Tue, 12 May 2015 13:13:38 +0000 (13:13 +0000)]
[ARM] Use AEABI aligned function variants
AEABI defines aligned variants of memcpy etc. that can be faster than
the default version due to not having to do alignment checks. When
emitting target code for these functions make use of these aligned
variants if possible. Also convert memset to memclr if possible.
Igor Laevsky [Tue, 12 May 2015 13:12:14 +0000 (13:12 +0000)]
Reverse ordering of base and derived pointer during safepoint lowering.
According to the documentation in StackMap section for the safepoint we should have:
"The first Location in each pair describes the base pointer for the object. The second is the derived pointer actually being relocated."
But before this change we emitted them in reverse order - derived pointer first, base pointer second.
Andrea Di Biagio [Tue, 12 May 2015 12:34:22 +0000 (12:34 +0000)]
[X86] Remove useless target specific combine on TRUNCATE dag nodes.
Before revision 171146, function 'PerformTruncateCombine' used to perform
a premature lowering of TRUNCATE dag nodes.
Revision 171146 then moved all the logic implemented by PerformTruncateCombine
to a custom lowering hook. However, that revision forgot to delete
function PerformTruncateCombine from the code.
This patch removes function 'PerformTruncateCombine' since it has no effect
on the SelectionDAG. No functional change intended.
AVX-512: select operation for i1 vectors
like: select i1 %cond, <16 x i1> %a, <16 x i1> %b.
I added pseudo-CMOV patterns to resolve the "select".
Added tests for KNL and SKX.
Eric Christopher [Tue, 12 May 2015 01:26:05 +0000 (01:26 +0000)]
Migrate existing backends that care about software floating point
to use the information in the module rather than TargetOptions.
We've had and clang has used the use-soft-float attribute for some
time now so have the backends set a subtarget feature based on
a particular function now that subtargets are created based on
functions and function attributes.
For the one middle end soft float check go ahead and create
an overloadable TargetLowering::useSoftFloat function that
just checks the TargetSubtargetInfo in all cases.
Also remove the command line option that hard codes whether or
not soft-float is set by using the attribute for all of the
target specific test cases - for the generic just go ahead and
add the attribute in the one case that showed up.
Sanjoy Das [Mon, 11 May 2015 23:47:30 +0000 (23:47 +0000)]
Refactoring gc_relocate related code in CodeGenPrepare.cpp
Summary:
The original code inserted new instructions by following a
Create->Remove->ReInsert flow. This patch removes the unnecessary
Remove->ReInsert part by setting up the InsertPoint correctly at the
very beginning. This change does not introduce any functionality change.
Ahmed Bougacha [Mon, 11 May 2015 23:09:46 +0000 (23:09 +0000)]
[MemCpyOpt] Look at any dependency -not just source- for memset+memcpy.
This fixes another miscompile introduced by r235232: when there was a
dependency on the memcpy destination other than the memset, we would
ignore it, because we only looked at the source dependency.
It was a mistake to use SrcDepInfo. Instead, just use DepInfo.
David Blaikie [Mon, 11 May 2015 22:20:48 +0000 (22:20 +0000)]
Readdress r236990, use of static members on a non-static variable.
The TargetRegistry is just a namespace-like class, instantiated in one
place to use a range-based for loop. Instead, expose access to the
registry via a range-based 'targets()' function instead. This makes most
uses a bit awkward/more verbose - but eventually we should just add a
range-based find_if function which will streamline these functions. I'm
happy to mkae them a bit awkward in the interim as encouragement to
improve the algorithms in time.
James Y Knight [Mon, 11 May 2015 22:17:13 +0000 (22:17 +0000)]
Fix tablegen's PrintFatalError function to run registered file
cleanups.
Also, change code in tablegen which printed a message and then called
"exit(1)" to use PrintFatalError, instead.
This fixes instances where an empty output file was left behind after
a failed tablegen invocation, which would confuse subsequent ninja
runs into not attempting to rebuild.
Alexey Samsonov [Mon, 11 May 2015 21:20:20 +0000 (21:20 +0000)]
Fix input validation issues in llvm-as/llvm-dis
Summary:
1. llvm-as/llvm-dis tools do not check for input filename length.
2. llvm-dis does not verify the `Streamer` variable against `nullptr` properly, so the `M` variable could be uninitialized (e.g. if the input file does not exist) leading to null dref.
Sanjay Patel [Mon, 11 May 2015 21:07:09 +0000 (21:07 +0000)]
propagate IR-level fast-math-flags to DAG nodes; 2nd try; NFC
This is a less ambitious version of:
http://reviews.llvm.org/rL236546
because that was reverted in:
http://reviews.llvm.org/rL236600
because it caused memory corruption that wasn't related to FMF
but was actually due to making nodes with 2 operands derive from a
plain SDNode rather than a BinarySDNode.
This patch adds the minimum plumbing necessary to use IR-level
fast-math-flags (FMF) in the backend without actually using
them for anything yet. This is a follow-on to:
http://reviews.llvm.org/rL235997
...which split the existing nsw / nuw / exact flags and FMF
into their own struct.
Davide Italiano [Mon, 11 May 2015 21:02:34 +0000 (21:02 +0000)]
[LoopIdiomRecognize] Transform backedge-taken count check into an assertion.
runOnCountable() allowed the caller to call on a loop without a
predictable backedge-taken count. Change the code so that only loops
with computable backdge-count can call this function, in order to catch
abuses.
[lib/Fuzzer] add a trace-based mutatation logic. Same idea as with DFSan-based mutator, but instead of relying on taint tracking, try to find the data directly in the input. More (logic and comments) to go.
Sanjoy Das [Mon, 11 May 2015 18:49:34 +0000 (18:49 +0000)]
[RewriteStatepointsForGC] Fix a bug on creating gc_relocate for pointer to vector of pointers
Summary:
In RewriteStatepointsForGC pass, we create a gc_relocate intrinsic for
each relocated pointer, and the gc_relocate has the same type with the
pointer. During the creation of gc_relocate intrinsic, llvm requires to
mangle its type. However, llvm does not support mangling of all possible
types. RewriteStatepointsForGC will hit an assertion failure when it
tries to create a gc_relocate for pointer to vector of pointers because
mangling for vector of pointers is not supported.
This patch changes the way RewriteStatepointsForGC pass creates
gc_relocate. For each relocated pointer, we erase the type of pointers
and create an unified gc_relocate of type i8 addrspace(1)*. Then a
bitcast is inserted to convert the gc_relocate to the correct type. In
this way, gc_relocate does not need to deal with different types of
pointers and the unsupported type mangling is no longer a problem. This
change would also ease further merge when LLVM erases types of pointers
and introduces an unified pointer type.
Some minor changes are also introduced to gc_relocate related part in
InstCombineCalls, CodeGenPrepare, and Verifier accordingly.
Summary:
r235215 adds support for f16 to be considered as a load/store type and
promote f16 operations to f32.
This patch has miscellaneous fixes for the X86 backend so all f16
operations are handled:
1. Set loadextaction for f16 vectors to expand.
2. Handle FP_EXTEND in a switch statement when handling v2f32
3. Do not fold (FP_TO_SINT (load f16)) into FP_TO_INT*_IN_MEM or
(store (SINT_TO_FP )) to a FILD.
Aaron Ballman [Mon, 11 May 2015 13:10:17 +0000 (13:10 +0000)]
Replacing a range-based for loop with an old-style for loop. This code was previously causing a warning with MSVC about a compiler-generated local variable because TargetRegistry::begin() and end() are static member functions. NFC.
Adam Nemet [Mon, 11 May 2015 08:30:28 +0000 (08:30 +0000)]
[Docs] Fix scoped noalias example
Summary:
As far as I understand the entire point of this example is to show that
if noalias is not a superset/equal to the alias.scope list on a scope
domain then load could reference locations that the store is not known
to not-alias i.e may alias.
The QPX single-precision load/store intrinsics have implied
truncation/extension from/to the declared value type of <4 x double> to the
memory type of <4 x float>. When we can prove the alignment of the pointer
argument, and thus replace the intrinsic with a regular load or store, we need
to load or store the correct data type (<4 x float>) instead of (<4 x double>).
Tobias Grosser [Sat, 9 May 2015 09:08:56 +0000 (09:08 +0000)]
Add polly support to sort_includes.py
Changes:
- Add "isl/" as a system library prefix. Even though isl is regularly
imported into polly, it is still used like an external library.
- Add "json/" as a system library prefix. Polly uses json-cpp as external
library.
- Distinguish between llvm and subproject libraries. Always sort subprojects
before LLVM. This was already the case with clang, as 'clang' comes before
'llvm', but we also want 'polly' to be sorted before 'llvm'.
The sorting of headers that are not part of Polly or isl remains unchanged.
James Y Knight [Sat, 9 May 2015 03:13:37 +0000 (03:13 +0000)]
Fix MergeConsecutiveStore for non-byte-sized memory accesses.
The bug showed up as a compile-time assertion failure:
Assertion `NumBits >= MIN_INT_BITS && "bitwidth too small"' failed
when building msan tests on x86-64.
Prior to r236850, this bug was masked due to a bogus alignment check,
which also accidentally rejected non-byte-sized accesses. Afterwards,
an invalid ElementSizeBytes == 0 got further into the function, and
triggered the assertion failure.
It would probably be a good idea to allow it to handle merging stores
of unusual widths as well, but for now, to un-break it, I'm just
making the minimal fix.
Pete Cooper [Sat, 9 May 2015 00:51:03 +0000 (00:51 +0000)]
[Fast-ISel] Don't mark the first use of a remat constant as killed.
When emitting something like 'add x, 1000' if we remat the 1000 then we should be able to
mark the vreg containing 1000 as killed. Given that we go bottom up in fast-isel, a later
use of 1000 will be higher up in the BB and won't kill it, or be impacted by the lower kill.
However, rematerialised constant expressions aren't generated bottom up. The local value save area
grows downwards. This means that if you remat 2 constant expressions which both use 1000 then the
first will kill it, then the second, which is *lower* in the BB will read a killed register.
This is the case in the attached test where the 2 GEPs both need to generate 'add x, 6680' for the constant offset.
Note that this commit only makes kill flag generation conservative. There's nothing else obviously wrong with
the local value save area growing downwards, and in fact it needs to for handling arbitrarily complex constant expressions.
However, it would be nice if there was a solution which would let us generate more accurate kill flags, or just kill flags completely.
Author: dblaikie
Date: Fri May 8 17:47:50 2015
New Revision: 236912
URL: http://llvm.org/viewvc/llvm-project?rev=236912&view=rev
Log:
[opaque pointer type] Cleanup a few references to pointee types using nearby non-pointee types of the same value
& cleanup a convoluted return expression while I'm here
ScheduleDAGInstrs: In functions with tail calls PseudoSourceValues are not non-aliasing distinct objects
The code that builds the dependence graph assumes that two PseudoSourceValues
don't alias. In a tail calling function two FixedStackObjects might refer to the
same location. Worse 'immutable' fixed stack objects like function arguments are
not immutable and will be clobbered.
Change this so that a load from a FixedStackObject is not invariant in a tail
calling function and don't return a PseudoSourceValue for an instruction in tail
calling functions when building the dependence graph so that we handle function
arguments conservatively.
[lib/Fuzzer] use -fsanitize-coverage=trace-cmp when building LLVM with LLVM_USE_SANITIZE_COVERAGE; in lib/Fuzzer try to reload the corpus to pick up new units from other processes
This new class in a global context contain arch-specific knowledge in order
to provide LLVM libraries, tools and projects with the ability to understand
the architectures. For now, only FPU, ARCH and ARCH extensions on ARM are
supported.
Current behaviour it to parse from free-text to enum values and back, so that
all users can share the same parser and codes. This simplifies a lot both the
ASM/Obj streamers in the back-end (where this came from), and the front-end
parsers for command line arguments (where this is going to be used next).
The previous implementation, using .def/.h includes is deprecated due to its
inflexibility to be built without the backend support and for being too
cumbersome. As more architectures join this scheme, and as more features of
such architectures are added (such as hardware features, type sizes, etc) into
a full blown TargetDescription class, having a set of classes is the most
sane implementation.
The ultimate goal of this refactor both LLVM's and Clang's target description
classes into one unique interface, so that we can de-duplicate and standardise
the descriptions, as well as make it available for other front-ends, tools,
etc.
The FPU parsing for command line options in Clang has been converted to use
this new library and a number of aliases were added for compatibility:
* A bogus neon-vfpv3 alias (neon defaults to vfp3)
* armv5/v6
* {fp4/fp5}-{sp/dp}-d16
Next steps:
* Port Clang's ARCH/EXT parsing to use this library.
* Create a TableGen back-end to generate this information.
* Run this TableGen process regardless of which back-ends are built.
* Expose more information and rename it to TargetDescription.
* Continue re-factoring Clang to use as much of it as possible.
Pete Cooper [Fri, 8 May 2015 20:46:54 +0000 (20:46 +0000)]
[Fast-ISel] Clear kill flags on registers replaced by updateValueMap.
When selecting an extract instruction, we don't actually generate code but instead work out which register we are reading, and rewrite uses of the extract def to the source register. This is done via updateValueMap,.
However, its possible that the source register we are rewriting *to* to also have uses. If those uses are after a kill of the value we are rewriting *from* then we have uses after a kill and the verifier fails.
This code checks for the case where the to register is also used, and if so it clears all kill on the from register. This is conservative, but better that always clearing kills on the from register.
Sanjoy Das [Fri, 8 May 2015 18:58:55 +0000 (18:58 +0000)]
[BasicAA] Fix zext & sext handling
Summary:
There are several unhandled edge cases in BasicAA's GetLinearExpression
method. This changes fixes outstanding issues, including zext / sext of
a constant with the sign bit set, and the refusal to decompose zexts or
sexts of wrapping arithmetic.
However, the copy here should only have the kill flag on the 32-bit path, not the 64-bit one.
Otherwise, we are killing the source of the truncate which could be used later in the program.
This extension offers the backend the opportunity to insert (somewhat) arbitrary code to manage the transition from GC-aware code to code that is not GC-aware and back.
In order to support the injection of transition code, this extension wraps the STATEPOINT ISD node generated by the usual lowering lowering with two additional nodes: GC_TRANSITION_START and GC_TRANSITION_END. The transition arguments that were passed passed to the intrinsic (if any) are lowered and provided as operands to these nodes and may be used by the backend during code generation.
Eventually, the lowering of the GC_TRANSITION_{START,END} nodes should be informed by the GC strategy in use for the function containing the intrinsic call; for now, these nodes are instead replaced with no-ops.
projects / llvm / log