oota-llvm.git
15 years agoAdd instruction patterns and encodings for the x86 bt instructions.
Dan Gohman [Tue, 23 Dec 2008 22:45:23 +0000 (22:45 +0000)]
Add instruction patterns and encodings for the x86 bt instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61400 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUpdate test
Anton Korobeynikov [Tue, 23 Dec 2008 22:26:37 +0000 (22:26 +0000)]
Update test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61399 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRestore debug printing
Anton Korobeynikov [Tue, 23 Dec 2008 22:26:18 +0000 (22:26 +0000)]
Restore debug printing

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61398 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoSometimes APInt syntax is really ugly... :(
Anton Korobeynikov [Tue, 23 Dec 2008 22:26:01 +0000 (22:26 +0000)]
Sometimes APInt syntax is really ugly... :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61397 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoIndent stuff properly
Anton Korobeynikov [Tue, 23 Dec 2008 22:25:45 +0000 (22:25 +0000)]
Indent stuff properly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61396 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoInitial checkin of APInt'ififcation of switch lowering
Anton Korobeynikov [Tue, 23 Dec 2008 22:25:27 +0000 (22:25 +0000)]
Initial checkin of APInt'ififcation of switch lowering

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61395 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoSilence unused variable warnings.
Devang Patel [Tue, 23 Dec 2008 21:56:28 +0000 (21:56 +0000)]
Silence unused variable warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61392 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix typo.
Devang Patel [Tue, 23 Dec 2008 21:55:38 +0000 (21:55 +0000)]
Fix typo.
Silence unused variable warning.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61391 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoSilience unused warnings.
Devang Patel [Tue, 23 Dec 2008 21:55:04 +0000 (21:55 +0000)]
Silience unused warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61390 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoClean up the atomic opcodes in SelectionDAG.
Dan Gohman [Tue, 23 Dec 2008 21:37:04 +0000 (21:37 +0000)]
Clean up the atomic opcodes in SelectionDAG.

This removes all the _8, _16, _32, and _64 opcodes and replaces each
group with an unsuffixed opcode. The MemoryVT field of the AtomicSDNode
is now used to carry the size information. In tablegen, the size-specific
opcodes are replaced by size-independent opcodes that utilize the
ability to compose them with predicates.

This shrinks the per-opcode tables and makes the code that handles
atomics much more concise.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61389 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoadd some notes for simplifylibcalls optimizations
Chris Lattner [Tue, 23 Dec 2008 20:52:52 +0000 (20:52 +0000)]
add some notes for simplifylibcalls optimizations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61385 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoTestcase to show we can tie together integers and pointers of
Chris Lattner [Tue, 23 Dec 2008 18:52:26 +0000 (18:52 +0000)]
Testcase to show we can tie together integers and pointers of
the same size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61380 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoTweak --version to include the date and time.
Steve Naroff [Tue, 23 Dec 2008 18:41:47 +0000 (18:41 +0000)]
Tweak --version to include the date and time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61378 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRename BuildSchedUnits to BuildSchedGraph, and refactor the
Dan Gohman [Tue, 23 Dec 2008 18:36:58 +0000 (18:36 +0000)]
Rename BuildSchedUnits to BuildSchedGraph, and refactor the
code in ScheduleDAGSDNodes' BuildSchedGraph into separate functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61376 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoComment MO_FPImmediate and doxygenate surrounding comments.
Dan Gohman [Tue, 23 Dec 2008 18:20:16 +0000 (18:20 +0000)]
Comment MO_FPImmediate and doxygenate surrounding comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61374 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse isTerminator() instead of isBranch()||isReturn() in
Dan Gohman [Tue, 23 Dec 2008 17:28:50 +0000 (17:28 +0000)]
Use isTerminator() instead of isBranch()||isReturn() in
several places. isTerminator() returns true for a superset
of cases, and includes things like FP_REG_KILL, which are
nither return or branch but aren't safe to move/remat/etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61373 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAvoid an unnecessary call to allnodes_size(), which is linear.
Dan Gohman [Tue, 23 Dec 2008 17:24:50 +0000 (17:24 +0000)]
Avoid an unnecessary call to allnodes_size(), which is linear.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61372 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMinor code simplifications.
Dan Gohman [Tue, 23 Dec 2008 17:22:32 +0000 (17:22 +0000)]
Minor code simplifications.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61371 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agorevert r61368.
Zhongxing Xu [Tue, 23 Dec 2008 05:43:56 +0000 (05:43 +0000)]
revert r61368.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61369 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRemove dead code.
Zhongxing Xu [Tue, 23 Dec 2008 05:30:44 +0000 (05:30 +0000)]
Remove dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61368 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdded shuffle and splat test cases for r61365.
Mon P Wang [Tue, 23 Dec 2008 04:05:08 +0000 (04:05 +0000)]
Added shuffle and splat test cases for r61365.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61366 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFixed code generation for v8i16 and v16i8 splats on X86.
Mon P Wang [Tue, 23 Dec 2008 04:03:27 +0000 (04:03 +0000)]
Fixed code generation for v8i16 and v16i8 splats on X86.
Fixed lowering of v8i16 shuffles for v8i16 when we fall back to extract/insert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61365 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix the time regression I introduced in 464.h264ref with
Dale Johannesen [Tue, 23 Dec 2008 02:12:52 +0000 (02:12 +0000)]
Fix the time regression I introduced in 464.h264ref with
my last patch to this file.

The issue there was that all uses of an IV inside a loop
are actually references to Base[IV*2], and there was one
use outside that was the same but LSR didn't see the base
or the scaling because it didn't recurse into uses outside
the loop; thus, it used base+IV*scale mode inside the loop
instead of pulling base out of the loop.  This was extra bad
because register pressure later forced both base and IV into
memory.  Doing that recursion, at least enough
to figure out addressing modes, is a good idea in general;
the change in AddUsersIfInteresting does this.  However,
there were side effects....

It is also possible for recursing outside the loop to
introduce another IV where there was only 1 before (if
the refs inside are not scaled and the ref outside is).
I don't think this is a common case, but it's in the testsuite.
It is right to be very aggressive about getting rid of
such introduced IVs (CheckForIVReuse and the handling of
nonzero RewriteFactor in StrengthReduceStridedIVUsers).
In the testcase in question the new IV produced this way
has both a nonconstant stride and a nonzero base, neither
of which was handled before.  And when inserting
new code that feeds into a PHI, it's right to put such
code at the original location rather than in the PHI's
immediate predecessor(s) when the original location is outside
the loop (a case that couldn't happen before)
(RewriteInstructionToUseNewBase); better to avoid making
multiple copies of it in this case.

Also, the mechanism for keeping SCEV's corresponding to GEP's
no longer works, as the GEP might change after its SCEV
is remembered, invalidating the SCEV, and we might get a bad
SCEV value when looking up the GEP again for a later loop.
This also couldn't happen before, as we weren't recursing
into GEP's outside the loop.

I owe some testcases for this, want to get it in for nightly runs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61362 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoOne more permutation of subtracting off a base value.
Dale Johannesen [Tue, 23 Dec 2008 01:59:54 +0000 (01:59 +0000)]
One more permutation of subtracting off a base value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61361 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDon't forget to remove phi nodes from the value numbering table after we collapse...
Owen Anderson [Tue, 23 Dec 2008 00:49:51 +0000 (00:49 +0000)]
Don't forget to remove phi nodes from the value numbering table after we collapse them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61358 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMake the fuse-failed debug output human-readable.
Dan Gohman [Tue, 23 Dec 2008 00:19:20 +0000 (00:19 +0000)]
Make the fuse-failed debug output human-readable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61356 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoComment clean-ups. No functionality change.
Bill Wendling [Mon, 22 Dec 2008 22:32:22 +0000 (22:32 +0000)]
Comment clean-ups. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61354 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoCheck that the instruction isn't in the value numbering scope.
Bill Wendling [Mon, 22 Dec 2008 22:28:56 +0000 (22:28 +0000)]
Check that the instruction isn't in the value numbering scope.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61353 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoSimplification: Negate the operator== method instead of implementing a full operator...
Bill Wendling [Mon, 22 Dec 2008 22:16:31 +0000 (22:16 +0000)]
Simplification: Negate the operator== method instead of implementing a full operator!= method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61352 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd verification that deleted instruction isn't hiding in the PHI map.
Bill Wendling [Mon, 22 Dec 2008 22:14:07 +0000 (22:14 +0000)]
Add verification that deleted instruction isn't hiding in the PHI map.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61350 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoVerify removed in a few more places.
Bill Wendling [Mon, 22 Dec 2008 21:57:30 +0000 (21:57 +0000)]
Verify removed in a few more places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61349 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd verification functions to GVN which check to see that an instruction was
Bill Wendling [Mon, 22 Dec 2008 21:36:08 +0000 (21:36 +0000)]
Add verification functions to GVN which check to see that an instruction was
truely deleted. These will be expanded with further checks of all of the data
structures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61347 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRefactor a bunch of code out of AsmPrinter::EmitGlobalConstant into separate
Dan Gohman [Mon, 22 Dec 2008 21:14:27 +0000 (21:14 +0000)]
Refactor a bunch of code out of AsmPrinter::EmitGlobalConstant into separate
functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61345 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoOptimize setDepthDirty and setHeightDirty a little, as they showed
Dan Gohman [Mon, 22 Dec 2008 21:11:33 +0000 (21:11 +0000)]
Optimize setDepthDirty and setHeightDirty a little, as they showed
up on a profile.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61344 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd an assertion to the ScheduleDAGInstrs class to catch SUnits
Dan Gohman [Mon, 22 Dec 2008 21:08:08 +0000 (21:08 +0000)]
Add an assertion to the ScheduleDAGInstrs class to catch SUnits
reallocations. We don't do cloning on MachineInstr schedule DAGs,
but this is a worthwhile sanity check regardless.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61343 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd an accesor for the isNormalMemory field in the SDep class.
Dan Gohman [Mon, 22 Dec 2008 21:06:56 +0000 (21:06 +0000)]
Add an accesor for the isNormalMemory field in the SDep class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61342 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd an assertion to catch SUnits reallocations. And add a doxygen
Dan Gohman [Mon, 22 Dec 2008 21:06:20 +0000 (21:06 +0000)]
Add an assertion to catch SUnits reallocations. And add a doxygen
comment for the ScheduleDAGSDNodes class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61341 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoClarify a comment.
Dan Gohman [Mon, 22 Dec 2008 19:44:39 +0000 (19:44 +0000)]
Clarify a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61338 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse ignore & grep instead of XFAIL.
Mikhail Glushenkov [Sun, 21 Dec 2008 07:47:49 +0000 (07:47 +0000)]
Use ignore & grep instead of XFAIL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61307 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoTurn strcmp into memcmp, such as strcmp(P, "x") --> memcmp(P, "x", 2).
Nick Lewycky [Sun, 21 Dec 2008 00:19:21 +0000 (00:19 +0000)]
Turn strcmp into memcmp, such as strcmp(P, "x") --> memcmp(P, "x", 2).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61297 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix fast-isel to not emit invalid assembly when presented with a
Dan Gohman [Sat, 20 Dec 2008 17:19:40 +0000 (17:19 +0000)]
Fix fast-isel to not emit invalid assembly when presented with a
constant shift count that doesn't fit in the shift instruction's
immediate field. This fixes PR3242.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61281 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoReword the comment for ConstantInt's getLimitedValue.
Dan Gohman [Sat, 20 Dec 2008 17:06:39 +0000 (17:06 +0000)]
Reword the comment for ConstantInt's getLimitedValue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61280 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRemove redundant test for vector-nature. Scan the vector first to see whether
Nick Lewycky [Sat, 20 Dec 2008 16:48:00 +0000 (16:48 +0000)]
Remove redundant test for vector-nature. Scan the vector first to see whether
our optz'n will apply to it, then build the replacement vector only if needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61279 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoLiveInterval::removeKills and isKill don't need 'this' and
Dan Gohman [Sat, 20 Dec 2008 16:44:40 +0000 (16:44 +0000)]
LiveInterval::removeKills and isKill don't need 'this' and
can be static member functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61278 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse SmallVector's pop_back_val.
Dan Gohman [Sat, 20 Dec 2008 16:42:33 +0000 (16:42 +0000)]
Use SmallVector's pop_back_val.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61277 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse the correct Preds and Succs lists in setHeightDirty()
Dan Gohman [Sat, 20 Dec 2008 16:34:57 +0000 (16:34 +0000)]
Use the correct Preds and Succs lists in setHeightDirty()
and setDepthDirty(), respectively. This fixes PR3241.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61276 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMore precise XFAIL.
Bill Wendling [Fri, 19 Dec 2008 22:28:23 +0000 (22:28 +0000)]
More precise XFAIL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61265 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUn-XFAIL this test because it's passing and John doesn't seem interested in un-XFAILi...
Bill Wendling [Fri, 19 Dec 2008 22:25:01 +0000 (22:25 +0000)]
Un-XFAIL this test because it's passing and John doesn't seem interested in un-XFAILing it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61264 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse ~0u instead of -1u as the special value, to hopefully avoid
Dan Gohman [Fri, 19 Dec 2008 22:23:43 +0000 (22:23 +0000)]
Use ~0u instead of -1u as the special value, to hopefully avoid
warnings on compilers that warn about such things.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61263 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix PR3149. If an early clobber def is a physical register and it is tied to an input...
Evan Cheng [Fri, 19 Dec 2008 20:58:01 +0000 (20:58 +0000)]
Fix PR3149. If an early clobber def is a physical register and it is tied to an input operand, it effectively extends the live range of the physical register. Currently we do not have a good way to represent this.

172     %ECX<def> = MOV32rr %reg1039<kill>
180     INLINEASM <es:subl $5,$1
        sbbl $3,$0>, 10, %EAX<def>, 14, %ECX<earlyclobber,def>, 9, %EAX<kill>,
36, <fi#0>, 1, %reg0, 0, 9, %ECX<kill>, 36, <fi#1>, 1, %reg0, 0
188     %EAX<def> = MOV32rr %EAX<kill>
196     %ECX<def> = MOV32rr %ECX<kill>
204     %ECX<def> = MOV32rr %ECX<kill>
212     %EAX<def> = MOV32rr %EAX<kill>
220     %EAX<def> = MOV32rr %EAX
228     %reg1039<def> = MOV32rr %ECX<kill>

The early clobber operand ties ECX input to the ECX def.

The live interval of ECX is represented as this:
%reg20,inf = [46,47:1)[174,230:0)  0@174-(230) 1@46-(47)

The right way to represent this is something like
%reg20,inf = [46,47:2)[174,182:1)[181:230:0)  0@174-(182) 1@181-230 @2@46-(47)

Of course that won't work since that means overlapping live ranges defined by two val#.

The workaround for now is to add a bit to val# which says the val# is redefined by a early clobber def somewhere. This prevents the move at 228 from being optimized away by SimpleRegisterCoalescing::AdjustCopiesBackFrom.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61259 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoThe fields for the stoppoint debug intrinsic have not changed, so update the
John Criswell [Fri, 19 Dec 2008 19:56:36 +0000 (19:56 +0000)]
The fields for the stoppoint debug intrinsic have not changed, so update the
version number assertions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61257 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoThis test works again for Darwin because a patch was reverted.
Bill Wendling [Fri, 19 Dec 2008 19:08:13 +0000 (19:08 +0000)]
This test works again for Darwin because a patch was reverted.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61254 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd dyn_cast_or_null bindings for some additional classes missed in r61252.
Gordon Henriksen [Fri, 19 Dec 2008 18:51:17 +0000 (18:51 +0000)]
Add dyn_cast_or_null bindings for some additional classes missed in r61252.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61253 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoC bindings for dyn_cast_or_null.
Gordon Henriksen [Fri, 19 Dec 2008 18:39:45 +0000 (18:39 +0000)]
C bindings for dyn_cast_or_null.
This operation can be used to build dyn_cast, isa, and cast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61252 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdd support for writing LLVM IR to a specified BitstreamWriter.
Chris Lattner [Fri, 19 Dec 2008 18:37:59 +0000 (18:37 +0000)]
Add support for writing LLVM IR to a specified BitstreamWriter.
Patch by Lukasz Janyst!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61251 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMove the patterns which have i8 immediates before the patterns
Dan Gohman [Fri, 19 Dec 2008 18:25:21 +0000 (18:25 +0000)]
Move the patterns which have i8 immediates before the patterns
that have i32 immediates so that they get selected first. This
currently only matters in the JIT, as assemblers will
automatically use the smallest encoding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61250 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUse dyn_cast intead of isa + cast in the generated DAGISel code. This
Dan Gohman [Fri, 19 Dec 2008 18:13:39 +0000 (18:13 +0000)]
Use dyn_cast intead of isa + cast in the generated DAGISel code. This
reduces the amount of code slightly when assertions are enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61249 91177308-0d34-0410-b5e6-96231b3b80d8

15 years ago- CodeGenPrepare does not split loop back edges but it only knows about back edges...
Evan Cheng [Fri, 19 Dec 2008 18:03:11 +0000 (18:03 +0000)]
- CodeGenPrepare does not split loop back edges but it only knows about back edges of single block loops. It now does a DFS walk to find loop back edges.
- Use SplitBlockPredecessors to factor out common predecessors of the critical edge destination. This is disabled for now due to some regressions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61248 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix some release-assert warnings
Chris Lattner [Fri, 19 Dec 2008 17:03:38 +0000 (17:03 +0000)]
Fix some release-assert warnings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61244 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix bug 3202.
Rafael Espindola [Fri, 19 Dec 2008 10:55:56 +0000 (10:55 +0000)]
Fix bug 3202.
The EH_frame and .eh symbols are now private, except for darwin9 and earlier.
The patch also fixes the definition of PrivateGlobalPrefix on pcc linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61242 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoUpdate the .cvs files for nocapture.
Nick Lewycky [Fri, 19 Dec 2008 09:41:54 +0000 (09:41 +0000)]
Update the .cvs files for nocapture.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61241 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoCommit missed files from nocapture change.
Nick Lewycky [Fri, 19 Dec 2008 09:38:31 +0000 (09:38 +0000)]
Commit missed files from nocapture change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61240 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoResubmit support for the 'nocapture' attribute.
Nick Lewycky [Fri, 19 Dec 2008 06:39:12 +0000 (06:39 +0000)]
Resubmit support for the 'nocapture' attribute.

The problematic part of this patch is that we were out of attribute bits,
requiring some fancy bit hacking to make it fit (by shrinking alignment)
without breaking existing users or the file format.

This change will require users to rebuild llvm-gcc to match llvm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61239 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoPerform this loop only when the -debug flag is specified.
Bill Wendling [Fri, 19 Dec 2008 02:09:57 +0000 (02:09 +0000)]
Perform this loop only when the -debug flag is specified.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61238 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoInitialize the ImplicitDefed member, to avoid getting stale
Dan Gohman [Fri, 19 Dec 2008 00:46:20 +0000 (00:46 +0000)]
Initialize the ImplicitDefed member, to avoid getting stale
data from a previous block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61237 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix RegScavenger::forward() to work on basic blocks containing exactly
Dan Gohman [Fri, 19 Dec 2008 00:45:13 +0000 (00:45 +0000)]
Fix RegScavenger::forward() to work on basic blocks containing exactly
one instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61236 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDelete the RegScavenging constructor that takes a MachineBasicBlock
Dan Gohman [Fri, 19 Dec 2008 00:34:32 +0000 (00:34 +0000)]
Delete the RegScavenging constructor that takes a MachineBasicBlock
argument. Nothing was using it, and it set the MBB member without
calling enterBasicBlock, which was problematic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61234 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix test to account for generating some vector code for mul v2i64 instead
Mon P Wang [Thu, 18 Dec 2008 23:42:37 +0000 (23:42 +0000)]
Fix test to account for generating some vector code for mul v2i64 instead
of incorrectly generating pmuldq

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61228 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDidn't mean to commit this.
Bill Wendling [Thu, 18 Dec 2008 22:19:50 +0000 (22:19 +0000)]
Didn't mean to commit this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61222 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoTeach LowerSubregs to preserve kill/dead information when lowering
Dan Gohman [Thu, 18 Dec 2008 22:14:08 +0000 (22:14 +0000)]
Teach LowerSubregs to preserve kill/dead information when lowering
subreg instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61220 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRe-XFAIL this test until debug stuff settles down.
Bill Wendling [Thu, 18 Dec 2008 22:13:31 +0000 (22:13 +0000)]
Re-XFAIL this test until debug stuff settles down.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61219 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMake LowerSubregs' debug output for EXTRACT_SUBREG consistent with
Dan Gohman [Thu, 18 Dec 2008 22:11:34 +0000 (22:11 +0000)]
Make LowerSubregs' debug output for EXTRACT_SUBREG consistent with
that of INSERT_SUBREG and SUBREG_TO_REG.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61218 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix a copy+pasto in an assertion message.
Dan Gohman [Thu, 18 Dec 2008 22:07:25 +0000 (22:07 +0000)]
Fix a copy+pasto in an assertion message.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61217 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix indentation level.
Dan Gohman [Thu, 18 Dec 2008 22:06:01 +0000 (22:06 +0000)]
Fix indentation level.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61216 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoWhen emitting instructions that define EFLAGS and the EFLAGS value isn't
Dan Gohman [Thu, 18 Dec 2008 22:03:42 +0000 (22:03 +0000)]
When emitting instructions that define EFLAGS and the EFLAGS value isn't
used, mark the defs as dead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61215 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoWhen setting up the frame pointer, add it as a live-in register to all
Dan Gohman [Thu, 18 Dec 2008 22:01:52 +0000 (22:01 +0000)]
When setting up the frame pointer, add it as a live-in register to all
non-entry blocks, so that it doesn't appear use-before-def anywhere.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61214 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoPrint subreg information in MachineInstr::dump.
Dan Gohman [Thu, 18 Dec 2008 21:51:27 +0000 (21:51 +0000)]
Print subreg information in MachineInstr::dump.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61213 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFixed x86 code generation of multiple for v2i64. It was incorrect for SSE4.1.
Mon P Wang [Thu, 18 Dec 2008 21:42:19 +0000 (21:42 +0000)]
Fixed x86 code generation of multiple for v2i64.  It was incorrect for SSE4.1.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61211 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdded some basic test cases for r61209
Mon P Wang [Thu, 18 Dec 2008 20:05:58 +0000 (20:05 +0000)]
Added some basic test cases for r61209

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61210 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoAdded support for vector widening.
Mon P Wang [Thu, 18 Dec 2008 20:03:17 +0000 (20:03 +0000)]
Added support for vector widening.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61209 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRemove dead comments.
Evan Cheng [Thu, 18 Dec 2008 09:01:18 +0000 (09:01 +0000)]
Remove dead comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61201 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoOops! Left out a line.
Nick Lewycky [Thu, 18 Dec 2008 06:42:28 +0000 (06:42 +0000)]
Oops! Left out a line.
Simplifying the sdiv might allow further simplifications for our users.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61196 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMake all the vector elements positive in an srem of constant vector.
Nick Lewycky [Thu, 18 Dec 2008 06:31:11 +0000 (06:31 +0000)]
Make all the vector elements positive in an srem of constant vector.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61195 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix PR2929 by making bugpoint/code extract propagate the nothrow
Chris Lattner [Thu, 18 Dec 2008 05:52:56 +0000 (05:52 +0000)]
Fix PR2929 by making bugpoint/code extract propagate the nothrow
bit from the original function to the cloned one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61194 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix typo in error message.
Mikhail Glushenkov [Thu, 18 Dec 2008 04:06:58 +0000 (04:06 +0000)]
Fix typo in error message.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61191 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRecommit this grammar fix that was backed out along with nocapture.
Nick Lewycky [Thu, 18 Dec 2008 02:15:05 +0000 (02:15 +0000)]
Recommit this grammar fix that was backed out along with nocapture.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61189 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoGive MachineLICM a name, for -time-passes etc.
Dan Gohman [Thu, 18 Dec 2008 01:37:56 +0000 (01:37 +0000)]
Give MachineLICM a name, for -time-passes etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61184 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMove post-RA scheduling before branch folding for now, because branch
Dan Gohman [Thu, 18 Dec 2008 01:36:42 +0000 (01:36 +0000)]
Move post-RA scheduling before branch folding for now, because branch
folding's tail merging doesn't currently preserve liveness information
which post-RA scheduling requires.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61183 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRe-apply r61158 in a form that no longer breaks tests.
Owen Anderson [Thu, 18 Dec 2008 01:27:19 +0000 (01:27 +0000)]
Re-apply r61158 in a form that no longer breaks tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61182 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoRevert previous patch, appears to break bootstrap.
Dale Johannesen [Thu, 18 Dec 2008 01:23:41 +0000 (01:23 +0000)]
Revert previous patch, appears to break bootstrap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61181 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoMark the x86 fp stack registers as "reserved". This tells LiveVariables
Dan Gohman [Thu, 18 Dec 2008 01:05:09 +0000 (01:05 +0000)]
Mark the x86 fp stack registers as "reserved". This tells LiveVariables
and the RegisterScavenger not to expect traditional liveness
techniques are applicable to these registers, since we don't fully
modify the effects of push and pop after stackification.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61179 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoFix the time regression I introduced in 464.h264ref with
Dale Johannesen [Thu, 18 Dec 2008 00:57:22 +0000 (00:57 +0000)]
Fix the time regression I introduced in 464.h264ref with
my last patch to this file.

The issue there was that all uses of an IV inside a loop
are actually references to Base[IV*2], and there was one
use outside that was the same but LSR didn't see the base
or the scaling because it didn't recurse into uses outside
the loop; thus, it used base+IV*scale mode inside the loop
instead of pulling base out of the loop.  This was extra bad
because register pressure later forced both base and IV into
memory.  Doing that recursion, at least enough
to figure out addressing modes, is a good idea in general;
the change in AddUsersIfInteresting does this.  However,
there were side effects....

It is also possible for recursing outside the loop to
introduce another IV where there was only 1 before (if
the refs inside are not scaled and the ref outside is).
I don't think this is a common case, but it's in the testsuite.
It is right to be very aggressive about getting rid of
such introduced IVs (CheckForIVReuse and the handling of
nonzero RewriteFactor in StrengthReduceStridedIVUsers).
In the testcase in question the new IV produced this way
has both a nonconstant stride and a nonzero base, neither
of which was handled before.  (This patch does not handle
all the cases where this can happen.)  And when inserting
new code that feeds into a PHI, it's right to put such
code at the original location rather than in the PHI's
immediate predecessor(s) when the original location is outside
the loop (a case that couldn't happen before)
(RewriteInstructionToUseNewBase); better to avoid making
multiple copies of it in this case.

Everything above is exercised in
CodeGen/X86/lsr-negative-stride.ll (and ifcvt4 in ARM which is
the same IR).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61178 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoreapply this hunk from Bill's reversion in r61169, it is conservative
Chris Lattner [Thu, 18 Dec 2008 00:51:32 +0000 (00:51 +0000)]
reapply this hunk from Bill's reversion in r61169, it is conservative
and safe and orthogonal from turning off load pre.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61177 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoXFAIL on Linux.
Bill Wendling [Thu, 18 Dec 2008 00:35:21 +0000 (00:35 +0000)]
XFAIL on Linux.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61176 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agomake instnamer name unnamed blocks as well as instructions and args.
Chris Lattner [Thu, 18 Dec 2008 00:33:11 +0000 (00:33 +0000)]
make instnamer name unnamed blocks as well as instructions and args.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61175 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoDo not XFAIL.
Bill Wendling [Thu, 18 Dec 2008 00:27:15 +0000 (00:27 +0000)]
Do not XFAIL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61174 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoTemporarily revert r61027. It was causing a bootstrap failure in "release" mode
Bill Wendling [Wed, 17 Dec 2008 23:31:20 +0000 (23:31 +0000)]
Temporarily revert r61027. It was causing a bootstrap failure in "release" mode
with everyone's favorite error messages:

Comparing stages 2 and 3
warning: ./cc1-checksum.o differs
warning: ./cc1plus-checksum.o differs
Bootstrap comparison failure!
./c-decl.o differs
./cp/decl.o differs
./df-core.o differs
./gcc.o differs
./i386.o differs
./stor-layout.o differs
./tree-pretty-print.o differs
./tree.o differs
make[2]: *** [compare] Error 1
make[1]: *** [stage3-bubble] Error 2

See PR3227.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61169 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoXFAIL for now.
Devang Patel [Wed, 17 Dec 2008 22:54:54 +0000 (22:54 +0000)]
XFAIL for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61167 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoXfail these tests for now.
Devang Patel [Wed, 17 Dec 2008 22:53:09 +0000 (22:53 +0000)]
Xfail these tests for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61166 91177308-0d34-0410-b5e6-96231b3b80d8

15 years agoToday the front-ends (llvm-gcc and clang) generate multiple llvm.dbg.compile_units...
Devang Patel [Wed, 17 Dec 2008 22:39:29 +0000 (22:39 +0000)]
Today the front-ends (llvm-gcc and clang) generate multiple llvm.dbg.compile_units to identify source file for various debug entities. Each llvm.dbg.compile_unit matches one file on the disk. However, the backend only supports one DW_TAG_compile_unit per .o file. The backend selects first compile_unit from the vector to construct DW_TAG_compile_unit entry, which is not correct in all cases.

First step to resolve this is, record file name and directory directly in debug info for various debug entities.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@61164 91177308-0d34-0410-b5e6-96231b3b80d8