Sanjay Patel [Tue, 12 Jan 2016 16:50:17 +0000 (16:50 +0000)]
Add/edit tests to include instruction-level FMF on calls
Prepatory patch before changing LibCallSimplifier to use the FMF.
Also, tighten the CHECK lines and give the tests more meaningful names.
Similar changes to:
http://reviews.llvm.org/rL257414
Keno Fischer [Tue, 12 Jan 2016 13:38:15 +0000 (13:38 +0000)]
[ARM] Fix several state persistence bugs
Summary:
This fixes three bugs, in all of which state is not or incorrecly reset between
objects (i.e. when reusing the same pass manager to create multiple object
files):
1) AttributeSection needs to be reset to nullptr, because otherwise the backend
will try to emit into the old object file's attribute section causing a
segmentation fault.
2) MappingSymbolCounter needs to be reset, otherwise the second object file
will start where the first one left off.
3) The MCStreamer base class resets the Streamer's e_flags settings. Since
EF_ARM_EABI_VER5 is set on streamer creation, we need to set it again
after the MCStreamer was rest.
Also rename Reset (uppser case) to EHReset to avoid confusion with
reset (lower case).
Robert Lougher [Tue, 12 Jan 2016 11:48:25 +0000 (11:48 +0000)]
The isel pattern that selects the memory-register form of VCVTPH2PS
(64 to 128-bit) matches against the pattern fragment 'vzmovl_v2i64'
(a zero-extended 64-bit load).
However, a change in r248784 teaches the instruction combiner that only
the lower 64 bits of the input to a 128-bit vcvtph2ps are used. This means
the instruction combiner will ordinarily optimize away the upper 64-bit
insertelement instruction in the zero-extension and so we no longer select
the memory-register form. To fix this a new pattern has been added.
Axel Naumann [Tue, 12 Jan 2016 07:44:58 +0000 (07:44 +0000)]
Explicitly enable OBJECT library "target".
With this, one can build a lib from the objects of other libs:
set(SOURCES
$<TARGET_OBJECTS:obj.clingInterpreter>
$<TARGET_OBJECTS:obj.clingMetaProcessor>
$<TARGET_OBJECTS:obj.clingUtils>
)
Lang Hames [Tue, 12 Jan 2016 07:26:28 +0000 (07:26 +0000)]
[Orc] Comment out debugging output in OrcRemoteTargetClient::reserveMem to see
whether this affects the GCC 4.7 ICE on
http://lab.llvm.org:8011/builders/clang-x86_64-ubuntu-gdb-75 .
Lang Hames [Tue, 12 Jan 2016 06:48:52 +0000 (06:48 +0000)]
[Orc] Add overloads of RPC::handle and RPC::expect that take member functions as
handlers.
It is expected that RPC handlers will usually be member functions. Accepting them
directly in handle and expect allows for the remove of a lot of lambdas an
explicit error variables.
This patch also uses this new feature to substantially tidy up the
OrcRemoteTargetServer class.
Dan Gohman [Tue, 12 Jan 2016 02:58:12 +0000 (02:58 +0000)]
[WebAssembly] Define a custom segment type for function definitions.
Since function definitions are not loaded into the address space, PT_LOAD is
inappropriate. PT_WEBASSEMBLY_FUNCTIONS is used to identify where the function
definitions are so that they can be processed at program startup time.
Justin Bogner [Tue, 12 Jan 2016 00:55:26 +0000 (00:55 +0000)]
LoopUnroll: Clean up the maze of initialization for unroll parameters. NFC
The layering of where the various loop unroll parameters are
initialized and overridden here was very confusing, making it pretty
difficult to tell just how the various sources interacted. Instead, we
put all of the initialization logic together in a single function so
that it's obvious what overrides what.
[libFuzzer] extend the weak memcmp/strcmp/strncmp interceptors to receive the result of the computations. With that, don't do any mutations if memcmp/etc returned 0
Function::copyAttributesFrom will copy the personality function, prefix
data and prolog data from the source function to the new function, and
is invoked when the IRMover copies the function prototype. This puts a
reference to a constant in the source module on a function in the dest
module, which causes an error when deleting the source module after
importing, since the personality function in the source module still has
uses (this would presumably also be an issue for the prologue and prefix
data). Remove the copies added to the dest copy when creating the new
prototype, as they are mapped properly when/if we link the function body.
Currently WebAssembly has two kinds of relocations; data addresses and
function addresses. This adds ELF relocations for them, as well as an
MC symbol kind to indicate which type of relocation is needed.
Sanjay Patel [Mon, 11 Jan 2016 23:31:48 +0000 (23:31 +0000)]
[LibCallSimplifier] use instruction-level fast-math-flags to transform log calls
Also, add tests to verify that we're checking 'fast' on both calls of each transform pair,
tighten the CHECK lines, and give the tests more meaningful names.
This is a continuation of:
http://reviews.llvm.org/rL255555
http://reviews.llvm.org/rL256871
http://reviews.llvm.org/rL256964
http://reviews.llvm.org/rL257400
http://reviews.llvm.org/rL257404
Sanjay Patel [Mon, 11 Jan 2016 22:34:19 +0000 (22:34 +0000)]
[LibCallSimplifier] use instruction-level fast-math-flags to transform sqrt calls
This is a continuation of adding FMF to call instructions:
http://reviews.llvm.org/rL255555
The intent of the patch is to preserve the current behavior of the transform except
that we use the sqrt instruction's 'fast' attribute as a trigger rather than the
function-level attribute.
But this raises a bug noted by the new FIXME comment.
In order to do this transform:
sqrt((x * x) * y) ---> fabs(x) * sqrt(y)
...we need all of the sqrt, the first fmul, and the second fmul to be 'fast'.
If any of those ops is strict, we should bail out.
Dan Gohman [Mon, 11 Jan 2016 22:05:44 +0000 (22:05 +0000)]
[WebAssembly] Reorganize address offset folding.
Always expect tglobaladdr and texternalsym to be wrapped in
WebAssemblywrapper nodes. Also, split out a regPlusGA from regPlusImm so
that it can special-case global addresses, as they can be folded in more
cases.
Unfortunately this doesn't enable any new optimizations yet due to
SelectionDAG limitations. I'll be submitting changes to the SelectionDAG
infrastructure, along with tests, in a separate patch.
Teresa Johnson [Mon, 11 Jan 2016 21:37:41 +0000 (21:37 +0000)]
Split resolveCycles(bool AllowTemps) into two interfaces and document
Address review feedback from r255909.
Move body of resolveCycles(bool AllowTemps) to
resolveRecursivelyImpl(bool AllowTemps). Revert resolveCycles back
to asserting on temps, and add new resolveNonTemporaries interface
to invoke the new implementation with AllowTemps=true. Document
the differences between these interfaces, specifically the effect
on RAUW support and uniquing. Call appropriate interface from
ValueMapper.
Reid Kleckner [Mon, 11 Jan 2016 20:35:45 +0000 (20:35 +0000)]
Use ::GetVersionEx directly rather than the Win8.1 SDK helpers
This removes ifdefs and fixes the build for users of the Win8.0 SDK,
which I happen to be. Upgrading is not hard, but executing the same code
everywhere seems better.
Dimitry Andric [Mon, 11 Jan 2016 20:12:53 +0000 (20:12 +0000)]
Ensure -mcpu=xscale works for arm targets, after rL252903 and rL252904
After these revisions, for arm targets, the -mcpu=xscale option caused
an error: "the clang compiler does not support '-mcpu=xscale'". Adding
"v5e" as a SUB_ARCH in ARMTargetParser.def helps.
Submitted by: Andrew Turner
Differential Revision: http://reviews.llvm.org/D16043
[sanitizer] [msan] Fix origin store of array types
This patch fixes the memory sanitizer origin store instrumentation for
array types. This can be triggered by cases where frontend lowers
function return to array type instead of aggregation.
int foo (int p)
{
mypair z = my_make_pair(p, 0);
return z.y + z.x;
}
--
It will be lowered with target set to aarch64-linux and -O0 to:
--
[...]
define i32 @_Z3fooi(i32 %p) #0 {
[...]
%call = call [2 x i64] @_Z12my_make_pairxi(i64 %conv, i32 0)
%1 = bitcast %struct.mypair* %z to [2 x i64]*
store [2 x i64] %call, [2 x i64]* %1, align 8
[...]
--
The origin store will emit a 'icmp' to test each store value again the
TLS origin array. However since 'icmp' does not support ArrayType the
memory instrumentation phase will bail out with an error.
This patch change it by using the same strategy used for struct type on
array.
It fixes the 'test/msan/insertvalue_origin.cc' for aarch64 (the -O0 case).
David Blaikie [Mon, 11 Jan 2016 19:26:01 +0000 (19:26 +0000)]
Fix some GCC 4.7 issues with the new Orc remote JIT support
I'm still seeing GCC ICE locally, but figured I'd throw this at the wall
& see if it sticks for the bots at least. Will continue investigating
the ICE in any case.
Lang Hames [Mon, 11 Jan 2016 17:09:58 +0000 (17:09 +0000)]
XFAIL the remote small code model tests on x86. Small code model is not properly
supported, and only worked previously because we weren't really running them
out-of-process.
Matt Arsenault [Mon, 11 Jan 2016 17:02:00 +0000 (17:02 +0000)]
AMDGPU: Pattern match ffbh pattern to instruction.
The hardware instruction's output on 0 is -1 rather than 32.
Eliminate a test and select to -1. This removes an extra instruction
from the compatability function with HSAIL's firstbit instruction.
Lang Hames [Mon, 11 Jan 2016 16:35:55 +0000 (16:35 +0000)]
[LLI] Replace the LLI remote-JIT support with the new ORC remote-JIT components.
The new ORC remote-JITing support provides a superset of the old code's
functionality, so we can replace the old stuff. As a bonus, a couple of
previously XFAILed tests have started passing.
Alexey Bataev [Mon, 11 Jan 2016 11:52:29 +0000 (11:52 +0000)]
[X86] Reduce complexity of the LEA optimization pass, by Andrey Turetsky.
In the OptimizeLEA pass keep instructions' positions in the basic block saved and use them for calculation of the distance between two instructions instead of std::distance. This reduces complexity of the pass from O(n^3) to O(n^2) and thus the compile time.
Differential Revision: http://reviews.llvm.org/D15692
Craig Topper [Mon, 11 Jan 2016 05:13:41 +0000 (05:13 +0000)]
[TableGen] Allow asm writer to use up to 3 OpInfo tables instead of 2. This allows x86 to use 56 total bits made up of a 32-bit, 16-bit, and 8-bit table. Previously we were using 64 total bits.
This saves 14K from the x86 table size. And saves space on other targets as well.
Craig Topper [Mon, 11 Jan 2016 05:13:38 +0000 (05:13 +0000)]
[TableGen] Remove unnecessary 0 terminator from an array that only existed to prevent ending an array with a comma. But that's perfectly legal and not something we need to prevent. NFC