Summary:
If a constraint is something like:
```
$0 = [1,1]
```
it'll now be created as:
```
assert($0 == 1)
```
instead of:
```
assert($0 >= 1 && $0 <= 1)
```
In general, ~3% speedup when solving per query in my machine. Biggest improvement was when verifying sqlite3, total time went down from 3000s to 2200s.
I couldn't create a test for this as there is no way to dump the formula yet. D48221 adds a method to dump the formula but there is no way to do it from the command line.
Also, a test that prints the formula will most likely fail in the future, as different solvers print the formula in different formats.
Reviewers: NoQ, george.karpenkov, ddcc
Reviewed By: george.karpenkov
Subscribers: xazax.hun, szepet, a.sidorin
Differential Revision: https://reviews.llvm.org/D48227
llvm-svn: 335116
Sign-extension of small types (e.g. short) was not handled correctly.
The reason for that was that when we were assigning the a value to the
Scalar object, we would accidentally promote the type to int (even
though the assignment code in AssignTypeToMatch tried to cast the value
to the appropriate type, it would still invoke the "int" version of
operator=). Instead, I use the APInt version of operator=, where the
bitwidth is specified explicitly. Among other things, this allows us to
fold the individual size cases into one.
llvm-svn: 335114
This patch teaches llvm-mca how to identify register writes that implicitly zero
the upper portion of a super-register.
On X86-64, a general purpose register is implemented in hardware as a 64-bit
register. Quoting the Intel 64 Software Developer's Manual: "an update to the
lower 32 bits of a 64 bit integer register is architecturally defined to zero
extend the upper 32 bits". Also, a write to an XMM register performed by an AVX
instruction implicitly zeroes the upper 128 bits of the aliasing YMM register.
This patch adds a new method named clearsSuperRegisters to the MCInstrAnalysis
interface to help identify instructions that implicitly clear the upper portion
of a super-register. The rest of the patch teaches llvm-mca how to use that new
method to obtain the information, and update the register dependencies
accordingly.
I compared the kernels from tests clear-super-register-1.s and
clear-super-register-2.s against the output from perf on btver2. Previously
there was a large discrepancy between the estimated IPC and the measured IPC.
Now the differences are mostly in the noise.
Differential Revision: https://reviews.llvm.org/D48225
llvm-svn: 335113
The SPIR target currently allows for half precision floating point types to be
emitted using the LLVM intrinsic functions which convert half types to floats
and doubles. However, this is illegal in SPIR as the only intrinsic allowed by
SPIR is memcpy, as per section 3 of the SPIR specification. Currently this is
leading to an assert being hit in the Clang CodeGen when attempting to emit a
constant or literal _Float16 type in a comparison operation on a SPIR or SPIR64
target. This assert stems from the CodeGen attempting to emit a constant half
value as an integer because the backend has specified that it is using these
half conversion intrinsics (which represents half as i16). This patch prevents
SPIR targets from using these intrinsics by overloading the responsible target
info method, marks SPIR targets as having a legal half type and provides
additional regression testing for the _Float16 type on SPIR targets.
Patch by: Stephen McGroarty
Differential Revision: https://reviews.llvm.org/D48188
llvm-svn: 335111
This fixes compilation with MinGW after SVN r333798, which added
a few functions within _WIN32 ifdefs, functions returning
std::error_code. Include everything that is needed instead of
hoping that this header being inclued transitively (which it apparently
is in MSVC builds).
llvm-svn: 335107
Summary: Introducing a Prototype object to capture Variables that must be set but keeps degrees of freedom as Invalid. This allows exploring non constraint variables later on.
Reviewers: courbet
Subscribers: tschuett, llvm-commits
Differential Revision: https://reviews.llvm.org/D48316
llvm-svn: 335105
Summary:
The only reason python was used in the Host module was to compute the
python path. I resolve this the same way as D47384 did for clang, by
moving the path computation into the python plugin and modifying
SBHostOS class to call into this module for ePathTypePythonDir.
Reviewers: zturner, jingham, davide
Subscribers: mgorny, lldb-commits
Differential Revision: https://reviews.llvm.org/D48215
llvm-svn: 335104
Summary:
The comment with the OpenCL clause about this clearly
says: "No type shall be qualified by qualifiers for
two or more different address spaces."
This must mean that two or more qualifiers for the
_same_ address space is allowed. However, it is
likely unintended by the programmer, so emit a
warning.
For dependent address space types, reject them like
before since we cannot know what the address space
will be.
Patch by Bevin Hansson (ebevhan).
Reviewers: Anastasia
Reviewed By: Anastasia
Subscribers: bader, cfe-commits
Differential Revision: https://reviews.llvm.org/D47630
llvm-svn: 335103
Summary:
Related to https://bugs.llvm.org/show_bug.cgi?id=37793, https://reviews.llvm.org/D46760#1127287
We'd like to do this canonicalization https://rise4fun.com/Alive/Gmc
But it is currently restricted by rL155136 / rL155362, which says:
```
// This is a constant shift of a constant shift. Be careful about hiding
// shl instructions behind bit masks. They are used to represent multiplies
// by a constant, and it is important that simple arithmetic expressions
// are still recognizable by scalar evolution.
//
// The transforms applied to shl are very similar to the transforms applied
// to mul by constant. We can be more aggressive about optimizing right
// shifts.
//
// Combinations of right and left shifts will still be optimized in
// DAGCombine where scalar evolution no longer applies.
```
I think these tests show that for *constants*, SCEV has no issues with that canonicalization.
Reviewers: mkazantsev, spatel, efriedma, sanjoy
Reviewed By: mkazantsev
Subscribers: sanjoy, javed.absar, llvm-commits, stoklund, bixia
Differential Revision: https://reviews.llvm.org/D48229
llvm-svn: 335101
Summary:
First off: i do not have any access to that processor,
so this is purely theoretical, no benchmarks.
I have been looking into b**d**ver2 scheduling profile, and while cross-referencing
the existing b**t**ver2, znver1 profiles, and the reference docs
(`Software Optimization Guide for AMD Family {15,16,17}h Processors`),
i have noticed that only b**t**ver2 scheduling profile specifies these.
Also, there is no mca test coverage.
Reviewers: RKSimon, craig.topper, courbet, GGanesh, andreadb
Reviewed By: GGanesh
Subscribers: gbedwell, vprasad, ddibyend, shivaram, Ashutosh, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D47676
llvm-svn: 335099
Summary:
I ran llvm-exegesis on SKX, SKL, BDW, HSW, SNB.
Atom is from Agner and SLM is a guess.
I've left AMD processors alone.
Reviewers: RKSimon, craig.topper
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D48079
llvm-svn: 335097
The clang test inadvertently turned a floating point value into a double by having the wrong return type on the test function relative to the intrinsic it was testing.
This resulted in an extra fptosi instruction that propagated into this test when I copied the clang output.
llvm-svn: 335094
Summary:
If we get an error building the SelectionDAG for inline assembly we try to continue and still build the DAG.
But if the return type for the inline assembly is a struct we end up crashing because we try to create an UNDEF node with a struct type which isn't valid.
Instead we need to create an UNDEF for each element of the struct and join them with merge_values.
This patch relies on single operand merge_values being handled gracefully by getMergeValues. If the return type is void there will be no VTs returned by ComputeValueVTs and now we just return instead of calling setValue. Hopefully that's ok, I assumed nothing would need to look up the mapped value for void node.
Fixes PR37359
Reviewers: rengolin, rovka, echristo, efriedma, bogner
Reviewed By: efriedma
Subscribers: craig.topper, llvm-commits
Differential Revision: https://reviews.llvm.org/D46560
llvm-svn: 335093
Summary:
After r335018, the static tables are guaranteed sorted by the EVEX opcode to convert. We can use this to do a binary search and remove the need for any secondary data structures.
Right now one table is 736 entries and the other is 482 entries. It might make sense to merge the two tables as a follow up. The effort it takes to select the table is probably similar to the extra binary search step it would require for a larger table.
I haven't done any measurements to see if this has any effect on compile time, but I don't imagine that EVEX->VEX conversion is a place we spend a lot of time.
Reviewers: RKSimon, spatel, chandlerc
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D48312
llvm-svn: 335092
... instead of prepending it at the beginning (the original behavior
since implemented in r122535 2010-12-23). This builds up an
AttributeList in the the order in which the attributes appear in the
source.
The reverse order caused nodes for attributes in the AST (e.g. LoopHint)
to be in the reverse, and therefore printed in the wrong order by
-ast-dump. Some TODO comments mention this. The order was explicitly
reversed for enable_if attribute overload resolution and name mangling,
which is not necessary anymore with this patch.
The change unfortunately has some secondary effects, especially for
diagnostic output. In the simplest cases, the CHECK lines or expected
diagnostic were changed to the the new output. If the kind of
error/warning changed, the attribute's order was changed instead.
It also causes some 'previous occurrence here' hints to be textually
after the main marker. This typically happens when attributes are
merged, but are incompatible. Interchanging the role of the the main
and note SourceLocation will also cause the case where two different
declaration's attributes (in contrast to multiple attributes of the
same declaration) are merged to be reversed. There is no easy fix
because sometimes previous attributes are merged into a new
declaration's attribute list, sometimes new attributes are added to a
previous declaration's attribute list. Since 'previous occurrence here'
pointing to locations after the main marker is not rare, I left the
markers as-is; it is only relevant when the attributes are declared in
the same declaration anyway, which often is on the same line.
Differential Revision: https://reviews.llvm.org/D48100
llvm-svn: 335084
This patch introduces two helpers to make it easier to ignore debug
intrinsics:
- Instruction::getNextNonDebugInstruction()
This is just like Instruction::getNextNode(), except that it skips debug
info.
- skipDebugInfo(BasicBlock::iterator)
A free function which advances a BasicBlock iterator past any debug
info. This is a no-op when the iterator already points to a non-debug
instruction.
Part of: llvm.org/PR37728
Related to: https://reviews.llvm.org/D47874
Differential Revision: https://reviews.llvm.org/D48305
llvm-svn: 335083
Diagnostics for narrowing conversions in initializer lists are currently
DefaultIgnored in Microsoft mode. But MSVC 2015 did add warnings about
narrowing conversions (C2397), so clang-cl can remove its special case code if
MSCompatibilityVersion is new enough.
(In MSVC, C2397 is just a warning and in clang it's default-mapped to an error,
but it can be remapped, and disabled with -Wno-c++11-narrowing, so that should
be fine.)
Fixes PR37314.
https://reviews.llvm.org/D48296
llvm-svn: 335082
The recommit ensures that the tests that failed on bots don't trigger the warning.
Xcode 10 removes support for libstdc++, but the users just get a confusing
include not file warning when including an STL header (when building for iOS6
which uses libstdc++ by default for example).
This patch adds a new warning that lets the user know that the libstdc++ include
path was not found to ensure that the user is more aware of why the error occurs.
rdar://40830462
Differential Revision: https://reviews.llvm.org/D48297
llvm-svn: 335081
passes to the recursive search function so we only recursively
search the kext bundle directory, instead of its parent directory.
<rdar://problem/41227170>
Differential Revision: https://reviews.llvm.org/D48302
llvm-svn: 335079
Summary:
OnExit ensures we call `ResetDeclMap` before this method ends. However,
we also have a few manual calls to ResetDeclMap in there that are actually unnecessary
because of this (calling the method multiple times has no effect). This patch also moves
the class out of the method that we can reuse it for the upcoming method that handles
parsing for completion.
Subscribers: lldb-commits
Differential Revision: https://reviews.llvm.org/D48337
llvm-svn: 335078
This patch covers up a fairly fundemental issue around remat and register allocation which shows up with psuedo instructions with more vreg uses than there are physical registers. This patch essentially just disables remat for STATEPOINTs which are the only case we've seen so far, but long term we need a better fix.
For STATEPOINTs specifically, this is a strict improvement. It unblocks progress towards enabling a currently off-by-default mode which integrates deopt bundle operand lowering with register allocator spilling so that we end up with smaller stack sizes and more optimally placed spills. Assming no other issues turn up during my next round of integration testing - which based on experience so far, is admittedly unlikely - we might finally be able to enable something I've been working towards in small bits and pieces for years now. :)
For psuedo ops in general, there are a couple of ideas for a "proper fix" discussed on the bug, but I'm far enough outside my knowledge area to not be able to see any of them through to a successful conclusion. If anyone wants to help out here, please do.
Differential Revision: https://reviews.llvm.org/D41098
llvm-svn: 335077
insertOutlinerPrologue was not used by any target, and prologue-esque code was
beginning to appear in insertOutlinerEpilogue. Refactor that into one function,
buildOutlinedFrame.
This just removes insertOutlinerPrologue and renames insertOutlinerEpilogue.
llvm-svn: 335076
Summary:
This fixes liveness tracking information after `drop` instruction
insertion in ExplicitLocals pass.
When a drop instruction is inserted to drop a dead register operand, the
original operand should be marked not dead anymore because it is now
used by the new drop instruction. And the operand to the new drop
instruction should be marked killed instead. This bug caused some
programs to fail when `llc` is run with `-verify-machineinstrs` option.
Reviewers: dschuff
Subscribers: sbc100, jgravelle-google, sunfish, llvm-commits
Differential Revision: https://reviews.llvm.org/D48253
llvm-svn: 335074
We only need to use 512 bit vectors all the way through v8i64 reductions since those max instructions are new to avx512f and only available in 512 bits until SKX.
For v16i32 and floating point we have legacy 128/256 bit instructions we can use.
I've tried to use other intrinsics to reduce the verbosity of the code and avoid having to mention all the shuffles. I've also removed all the -1 shuffle indices so the output sequence is fully specified and not left to backend optimization.
Differential Revision: https://reviews.llvm.org/D47401
llvm-svn: 335070
Clement Courbet