Chandler Carruth [Sat, 31 Jan 2015 11:17:59 +0000 (11:17 +0000)]
[PM] Switch the TargetMachine interface from accepting a pass manager
base which it adds a single analysis pass to, to instead return the type
erased TargetTransformInfo object constructed for that TargetMachine.
This removes all of the pass variants for TTI. There is now a single TTI
*pass* in the Analysis layer. All of the Analysis <-> Target
communication is through the TTI's type erased interface itself. While
the diff is large here, it is nothing more that code motion to make
types available in a header file for use in a different source file
within each target.
I've tried to keep all the doxygen comments and file boilerplate in line
with this move, but let me know if I missed anything.
With this in place, the next step to making TTI work with the new pass
manager is to introduce a really simple new-style analysis that produces
a TTI object via a callback into this routine on the target machine.
Once we have that, we'll have the building blocks necessary to accept
a function argument as well.
Owen Anderson [Sat, 31 Jan 2015 09:13:36 +0000 (09:13 +0000)]
Replace another std::set in the core of CodeGenRegister, this time with sorted arrays.
The hot path through this region of code does lots of batch inserts into sets. By storing them as sorted arrays, we can defer the sorting to the end of the batch, which is dramatically more efficient. This reduces tblgen runtime by 25% on my worst-case target.
Owen Anderson [Sat, 31 Jan 2015 07:49:41 +0000 (07:49 +0000)]
Change more of the guts of CodeGenRegister's RegUnit tracking to be based on bit vectors.
This is a continuation of my prior work to move some of the inner workings for CodeGenRegister to use bit vectors when computing about register units. This is highly beneficial to TableGen runtime on targets with large, dense register files. This patch represents a ~40% runtime reduction over and above my earlier improvement on a stress test of this case.
llvm-readobj: add a test case for ARM_MOV32(T) base relocation
Add a trivial binary (int main() { return 0; }) built for Windows on ARM to
ensure that we can correctly identify ARM_MOV32(T) base relocations. Addresses
post-commit review comments.
This adds some comments and splits the flag calculation on type boundaries to
make the table more readable. Addresses some post-commit review comments to SVN
r227603. NFC.
Chandler Carruth [Sat, 31 Jan 2015 03:43:40 +0000 (03:43 +0000)]
[PM] Change the core design of the TTI analysis to use a polymorphic
type erased interface and a single analysis pass rather than an
extremely complex analysis group.
The end result is that the TTI analysis can contain a type erased
implementation that supports the polymorphic TTI interface. We can build
one from a target-specific implementation or from a dummy one in the IR.
I've also factored all of the code into "mix-in"-able base classes,
including CRTP base classes to facilitate calling back up to the most
specialized form when delegating horizontally across the surface. These
aren't as clean as I would like and I'm planning to work on cleaning
some of this up, but I wanted to start by putting into the right form.
There are a number of reasons for this change, and this particular
design. The first and foremost reason is that an analysis group is
complete overkill, and the chaining delegation strategy was so opaque,
confusing, and high overhead that TTI was suffering greatly for it.
Several of the TTI functions had failed to be implemented in all places
because of the chaining-based delegation making there be no checking of
this. A few other functions were implemented with incorrect delegation.
The message to me was very clear working on this -- the delegation and
analysis group structure was too confusing to be useful here.
The other reason of course is that this is *much* more natural fit for
the new pass manager. This will lay the ground work for a type-erased
per-function info object that can look up the correct subtarget and even
cache it.
Yet another benefit is that this will significantly simplify the
interaction of the pass managers and the TargetMachine. See the future
work below.
The downside of this change is that it is very, very verbose. I'm going
to work to improve that, but it is somewhat an implementation necessity
in C++ to do type erasure. =/ I discussed this design really extensively
with Eric and Hal prior to going down this path, and afterward showed
them the result. No one was really thrilled with it, but there doesn't
seem to be a substantially better alternative. Using a base class and
virtual method dispatch would make the code much shorter, but as
discussed in the update to the programmer's manual and elsewhere,
a polymorphic interface feels like the more principled approach even if
this is perhaps the least compelling example of it. ;]
Ultimately, there is still a lot more to be done here, but this was the
huge chunk that I couldn't really split things out of because this was
the interface change to TTI. I've tried to minimize all the other parts
of this. The follow up work should include at least:
1) Improving the TargetMachine interface by having it directly return
a TTI object. Because we have a non-pass object with value semantics
and an internal type erasure mechanism, we can narrow the interface
of the TargetMachine to *just* do what we need: build and return
a TTI object that we can then insert into the pass pipeline.
2) Make the TTI object be fully specialized for a particular function.
This will include splitting off a minimal form of it which is
sufficient for the inliner and the old pass manager.
3) Add a new pass manager analysis which produces TTI objects from the
target machine for each function. This may actually be done as part
of #2 in order to use the new analysis to implement #2.
4) Work on narrowing the API between TTI and the targets so that it is
easier to understand and less verbose to type erase.
5) Work on narrowing the API between TTI and its clients so that it is
easier to understand and less verbose to forward.
6) Try to improve the CRTP-based delegation. I feel like this code is
just a bit messy and exacerbating the complexity of implementing
the TTI in each target.
Many thanks to Eric and Hal for their help here. I ended up blocked on
this somewhat more abruptly than I expected, and so I appreciate getting
it sorted out very quickly.
Now that -mstack-probe-size is piped through to the backend via the function
attribute as on Windows x86, honour the value to permit handling of non-default
values for stack probes. This is needed /Gs with the clang-cl driver or
-mstack-probe-size with the clang driver when targeting Windows on ARM.
Kevin Enderby [Sat, 31 Jan 2015 00:37:11 +0000 (00:37 +0000)]
Add the -section option to llvm-objdump used with -macho that takes the argument
segname,sectname to specify a Mach-O section to print. The printing is based on
the section type or section attributes.
The printing of the module initialization and termination section types is printed
with this change. Printing of other section types will be added next.
David Blaikie [Fri, 30 Jan 2015 23:52:19 +0000 (23:52 +0000)]
Add PPC test for r227481, but XFAIL because this is actually more work than it appeared to be.
Same sort of bug as on ARM where the cmp+branch are lowered to br_cc
(choosing the branch's debugloc for the br_cc's debugloc) then expanded
out to a cmp and a br, but both using the debug loc of the br_cc, thus
losing fidelity.
Philip Reames [Fri, 30 Jan 2015 23:18:42 +0000 (23:18 +0000)]
Fix statepoint verifier tests to actually test verifier.
Patch by: Igor Laevsky
"Statepoint verifier tests were using wrong names for the statepoint and gc.relocate intrinsics. This change renames them to use correct names and fixes all uncovered issues."
David Blaikie [Fri, 30 Jan 2015 23:04:39 +0000 (23:04 +0000)]
Add ARM test for r227489, but XFAIL because this is actually more work than it appeared to be.
Also revert r227489 since it didn't actually fix the thing I thought I
was fixing (since the test case was targeting the wrong architecture
initially). The change might be correct & demonstrated by other test
cases, but it's not a priority for me to find those test cases right
now.
Reid Kleckner [Fri, 30 Jan 2015 21:03:31 +0000 (21:03 +0000)]
Win64: Put a REX_W prefix on all TAILJMP* instructions
MSDN's x64 software conventions page says that this is one of the fixed
list of legal epilogues:
https://msdn.microsoft.com/en-us/library/tawsa7cb.aspx
Presumably this is how the unwinder distinguishes epilogue jumps from
in-function control flow.
Also normalize the way we place "## TAILCALL" comments on such jumps.
Chad Rosier [Fri, 30 Jan 2015 19:55:40 +0000 (19:55 +0000)]
[AArch64] Make AArch64A57FPLoadBalancing output stable.
Add tie breaker to colorChainSet() sort so that processing order doesn't
depend on std::set order, which depends on pointer order, which is
unstable from run to run.
No test case as this is nearly impossible to reproduce.
Phabricator Review: http://reviews.llvm.org/D7265
Patch by Geoff Berry <gberry@codeaurora.org>!
Adrian Prantl [Fri, 30 Jan 2015 19:37:48 +0000 (19:37 +0000)]
Inliner: Use replaceDbgDeclareForAlloca() instead of splicing the
instruction and generalize it to optionally dereference the variable.
Follow-up to r227544.
If the original FPU specification involved a restricted VFP unit (d16), ensure
that we reset the functionality when we encounter a new FPU type. In
particular, if the user specified vfpv3-d16, but switched to a VFPv3 (which has
32 double precision registers), we would fail to reset the D16 feature, and
treat it as being equivalent to vfpv3-d16.
Zachary Turner [Fri, 30 Jan 2015 18:07:45 +0000 (18:07 +0000)]
Move DebugInfo to DebugInfo/DWARF.
In preparation for adding PDB support to LLVM, this moves the
DWARF parsing code to its own subdirectory under DebugInfo, and
renames LLVMDebugInfo to LLVMDebugInfoDWARF.
This is purely a mechanical / build system change.
Differential Revision: http://reviews.llvm.org/D7269
Reviewed by: Eric Christopher
The FPU directive permits the user to switch the target FPU, enabling
instructions that would be otherwise unavailable. However, when configuring the
new subtarget features, we would not enable the implied functions for newer
FPUs. This would result in invalid rejection of valid input. Ensure that we
inherit the implied FPU functionality when enabling newer versions of the FPU.
Fortunately, these are mostly hierarchical, unlike the CPUs.
Toma Tabacu [Fri, 30 Jan 2015 11:18:50 +0000 (11:18 +0000)]
[mips] Manually replace JAL pseudo-instructions with their JALR equivalent, instead of using InstAlias.
Summary:
This is needed by the .cprestore assembler directive.
This directive needs to be able to insert an LW instruction after every JALR replacement of a JAL pseudo-instruction
(and never after a JALR which has NOT been a result of a pseudo-instruction replacement).
The problem with using InstAlias for these is that after it replaces the pseudo-instruction, we can't find out if the resulting JALR instruction
was generated by an InstAlias or not, so we don't know whether or not to insert our LW instruction.
By replacing it manually, we know when the pseudo-instruction replacement happens and we can insert the LW instruction correctly.
Owen Anderson [Fri, 30 Jan 2015 09:05:49 +0000 (09:05 +0000)]
Change a very hot piece of code in TableGen's register unit computations to use bit vectors rather than arrays.
For target descriptions with very large and very dense register files, TableGen
can take an extremely long time to run. This change makes a dent in that (~15%
in my measurements) by accelerating the single hottest operation with better data
structures.
I believe there's still a lot of room to make this even faster with more global
changes that require replacing some of the existing datastructures in this area
with bit vectors, but that's a more involved change and I wanted to get this
simpler improvement in first.
Hao Liu [Fri, 30 Jan 2015 05:02:21 +0000 (05:02 +0000)]
[LoopVectorize] Induction variables: support arbitrary constant step.
Previously, only -1 and +1 step values are supported for induction variables. This patch extends LV to support
arbitrary constant steps.
Initial patch by Alexey Volkov. Some bug fixes are added in the following version.
Differential Revision: http://reviews.llvm.org/D6051 and http://reviews.llvm.org/D7193
Hao Liu [Fri, 30 Jan 2015 02:13:53 +0000 (02:13 +0000)]
[AArch64]Fix PR21675, a bug about lowering llvm.ctpop.i32. We should noot use "DAG.getUNDEF(MVT::v8i8)" to get all zero vector.
Patch by Wei-cheng Wang.
Akira Hatanaka [Fri, 30 Jan 2015 01:16:24 +0000 (01:16 +0000)]
[LTO] Scan all per-function subtargets when collecting runtime library names.
accumulateAndSortLibcalls in LTOCodeGenerator.cpp collects names of runtime
library functions which are used to identify user-defined functions that should
be protected. Previously, this function would only scan the TargetLowering
object belonging to the "main" subtarget for the library function names. This
commit changes it to scan all per-function subtargets.
Reid Kleckner [Thu, 29 Jan 2015 23:58:04 +0000 (23:58 +0000)]
x86: Fix large model calls to __chkstk for dynamic allocas
In the large code model, we now put __chkstk in %r11 before calling it.
Refactor the code so that we only do this once. Simplify things by using
__chkstk_ms instead of __chkstk on cygming. We already use that symbol
in the prolog emission, and it simplifies our logic.
Eric Christopher [Thu, 29 Jan 2015 23:27:36 +0000 (23:27 +0000)]
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl
calls that don't take a Function argument from Mips. Notable
exceptions: the AsmPrinter and MipsTargetObjectFile. The
latter needs to be fixed, and the former will be fixed when the
general AsmPrinter changes happen.
Chandler Carruth [Thu, 29 Jan 2015 23:26:37 +0000 (23:26 +0000)]
[LPM] Remove a PPC64 hack to try to work around a bad interaction
between the linker's TLS optimizations and Clang's TLS code generation.
For now, Clang has been changed to disable linker TLS optimizations
until it (and LLVM more generally) are emitting TLS code sequences
compatible with the old bugs found in the linkers. That's a better fix
to handle bootstrapping on that platform.