oota-llvm.git
16 years agosimplify some code by using a helper function. This really really
Chris Lattner [Mon, 16 Jun 2008 06:38:26 +0000 (06:38 +0000)]
simplify some code by using a helper function.  This really really
wants a 'nocapture' predicate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52304 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agomove a bunch of predicates up into their own section
Chris Lattner [Mon, 16 Jun 2008 06:30:22 +0000 (06:30 +0000)]
move a bunch of predicates up into their own section
in this file, no other changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52303 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoOther parts of this code treat noalias arguments as objects for
Chris Lattner [Mon, 16 Jun 2008 06:28:01 +0000 (06:28 +0000)]
Other parts of this code treat noalias arguments as objects for
the purposes of escape analysis.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52302 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoIf we are checking to see if the result of a call aliases a
Chris Lattner [Mon, 16 Jun 2008 06:19:11 +0000 (06:19 +0000)]
If we are checking to see if the result of a call aliases a
pointer derived from a local allocation, if the local allocation
never escapes, the pointers can't alias.  This implements PR2436

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52301 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAssign indices to empty basic blocks. This will be necessary for StrongPHIEliminatio...
Owen Anderson [Mon, 16 Jun 2008 06:18:41 +0000 (06:18 +0000)]
Assign indices to empty basic blocks.  This will be necessary for StrongPHIElimination in the near future.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52300 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRefactor basicaa's main alias function somethin' fierce.
Chris Lattner [Mon, 16 Jun 2008 06:10:11 +0000 (06:10 +0000)]
Refactor basicaa's main alias function somethin' fierce.
This fixes several minor bugs (such as returning noalias
for comparisons between external weak functions an null) but
is mostly a cleanup.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52299 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agothis is unneeded now.
Chris Lattner [Mon, 16 Jun 2008 04:25:39 +0000 (04:25 +0000)]
this is unneeded now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52298 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSwitch from generating the int128 typedefs based on targetdata to generating
Chris Lattner [Mon, 16 Jun 2008 04:25:29 +0000 (04:25 +0000)]
Switch from generating the int128 typedefs based on targetdata to generating
them based on the end-compiler's capabilities.  This fixes PR2453

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52297 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoresolve PR2453 by adding a run line.
Chris Lattner [Mon, 16 Jun 2008 04:22:39 +0000 (04:22 +0000)]
resolve PR2453 by adding a run line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52296 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix the crash on SimplifyLibCalls/2005-05-20-sprintf-crash.ll
Chris Lattner [Mon, 16 Jun 2008 04:10:21 +0000 (04:10 +0000)]
Fix the crash on SimplifyLibCalls/2005-05-20-sprintf-crash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52295 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agofix pr2460
Chris Lattner [Mon, 16 Jun 2008 04:02:40 +0000 (04:02 +0000)]
fix pr2460

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52294 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoThe transforms in visitEXTRACT_VECTOR_ELT are
Duncan Sands [Sun, 15 Jun 2008 20:12:31 +0000 (20:12 +0000)]
The transforms in visitEXTRACT_VECTOR_ELT are
not valid if the load is volatile.  Hopefully
all wrong DAG combiner transforms of volatile
loads and stores have now been caught.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52293 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoLegalizeTypes support for INSERT_VECTOR_ELT with
Duncan Sands [Sun, 15 Jun 2008 20:00:14 +0000 (20:00 +0000)]
LegalizeTypes support for INSERT_VECTOR_ELT with
a non-constant index.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52292 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix PR2434. When scanning for exising binary operator to reuse don't
Wojciech Matyjewicz [Sun, 15 Jun 2008 19:07:39 +0000 (19:07 +0000)]
Fix PR2434. When scanning for exising binary operator to reuse don't
take into account the instrucion pointed by InsertPt. Thanks to it,
returning the new value of InsertPt to the InsertBinop() caller can be
avoided. The bug was, actually, in visitAddRecExpr() method which wasn't
correctly handling changes of InsertPt. There shouldn't be any
performance regression, as -gvn pass (run after -indvars) removes any
redundant binops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52291 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd a missing semicolon.
Wojciech Matyjewicz [Sun, 15 Jun 2008 18:02:47 +0000 (18:02 +0000)]
Add a missing semicolon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52290 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd an "exe" suffix only if the output file has no suffix at all.
Argyrios Kyrtzidis [Sun, 15 Jun 2008 15:20:16 +0000 (15:20 +0000)]
Add an "exe" suffix only if the output file has no suffix at all.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52289 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix the sys::Path::getSuffix() implementation.
Argyrios Kyrtzidis [Sun, 15 Jun 2008 15:15:19 +0000 (15:15 +0000)]
Fix the sys::Path::getSuffix() implementation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52288 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMake sure all produced executable files have "exe" suffix on Windows.
Argyrios Kyrtzidis [Sun, 15 Jun 2008 13:48:12 +0000 (13:48 +0000)]
Make sure all produced executable files have "exe" suffix on Windows.
With this more general way, -native and -native-cbe options are handled too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52287 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMake sure that the current executable filename has "exe" suffix on Windows.
Argyrios Kyrtzidis [Sun, 15 Jun 2008 12:07:01 +0000 (12:07 +0000)]
Make sure that the current executable filename has "exe" suffix on Windows.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52286 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAppend "exe" suffix to executable files.
Argyrios Kyrtzidis [Sun, 15 Jun 2008 12:01:16 +0000 (12:01 +0000)]
Append "exe" suffix to executable files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52285 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix the environment block that is passed to the CreateProcess function.
Argyrios Kyrtzidis [Sun, 15 Jun 2008 03:54:39 +0000 (03:54 +0000)]
Fix the environment block that is passed to the CreateProcess function.
This bug made llvm-ld unable to function with "-native" option, since the process that was used to call 'gcc' was crashing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52284 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove a redundant AfterLegalize check. Turn
Duncan Sands [Sat, 14 Jun 2008 17:48:34 +0000 (17:48 +0000)]
Remove a redundant AfterLegalize check.  Turn
on some code when !AfterLegalize - but since
this whole code section is turned off by an
"if (0)" it's not really turning anything on.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52276 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoChange 'while' loop to 'do' loop.
Wojciech Matyjewicz [Sat, 14 Jun 2008 16:48:22 +0000 (16:48 +0000)]
Change 'while' loop to 'do' loop.

Add a safety measure. It isn't safe to assume in ScalarEvolutionExpander that
all loops are in canonical form (but it should be safe for loops that have
AddRecs).
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52275 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoadd missing atomic intrinsic from gcc
Andrew Lenharth [Sat, 14 Jun 2008 05:48:15 +0000 (05:48 +0000)]
add missing atomic intrinsic from gcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52270 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix a case where tailcallelim wouldn't set the changed bit when it made a change.
Chris Lattner [Sat, 14 Jun 2008 00:49:48 +0000 (00:49 +0000)]
Fix a case where tailcallelim wouldn't set the changed bit when it made a change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52267 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoTeach the spiller to commute instructions in order to fold a reload. This hits 410...
Evan Cheng [Fri, 13 Jun 2008 23:58:02 +0000 (23:58 +0000)]
Teach the spiller to commute instructions in order to fold a reload. This hits 410 times on 444.namd and 122 times on 252.eon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52266 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove unnecessary target lines.
Eli Friedman [Fri, 13 Jun 2008 22:12:16 +0000 (22:12 +0000)]
Remove unnecessary target lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52261 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove unnecessary target lines.
Eli Friedman [Fri, 13 Jun 2008 22:10:32 +0000 (22:10 +0000)]
Remove unnecessary target lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52260 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDon't skip over instructions other than loads that might read memory
Eli Friedman [Fri, 13 Jun 2008 22:02:12 +0000 (22:02 +0000)]
Don't skip over instructions other than loads that might read memory
when trying to sink stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52259 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoProtect ChangeCompareStride from situations in which it is possible
Dan Gohman [Fri, 13 Jun 2008 21:43:41 +0000 (21:43 +0000)]
Protect ChangeCompareStride from situations in which it is possible
for it to generate use-before-def IR, such as in this testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52258 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMake sure SimplifyStoreAtEndOfBlock doesn't mess with loops; the
Eli Friedman [Fri, 13 Jun 2008 21:17:49 +0000 (21:17 +0000)]
Make sure SimplifyStoreAtEndOfBlock doesn't mess with loops; the
structure checks are incorrect if the blocks aren't distinct.
Fixes PR2435.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52257 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDisable some DAG combiner optimizations that may be
Duncan Sands [Fri, 13 Jun 2008 19:07:40 +0000 (19:07 +0000)]
Disable some DAG combiner optimizations that may be
wrong for volatile loads and stores.  In fact this
is almost all of them!  There are three types of
problems: (1) it is wrong to change the width of
a volatile memory access.  These may be used to
do memory mapped i/o, in which case a load can have
an effect even if the result is not used.  Consider
loading an i32 but only using the lower 8 bits.  It
is wrong to change this into a load of an i8, because
you are no longer tickling the other three bytes.  It
is also unwise to make a load/store wider.  For
example, changing an i16 load into an i32 load is
wrong no matter how aligned things are, since the
fact of loading an additional 2 bytes can have
i/o side-effects.  (2) it is wrong to change the
number of volatile load/stores: they may be counted
by the hardware.  (3) it is wrong to change a volatile
load/store that requires one memory access into one
that requires several.  For example on x86-32, you
can store a double in one processor operation, but to
store an i64 requires two (two i32 stores).  In a
multi-threaded program you may want to bitcast an i64
to a double and store as a double because that will
occur atomically, and be indivisible to other threads.
So it would be wrong to convert the store-of-double
into a store of an i64, because this will become two
i32 stores - no longer atomic.  My policy here is
to say that the number of processor operations for
an illegal operation is undefined.  So it is alright
to change a store of an i64 (requires at least two
stores; but could be validly lowered to memcpy for
example) into a store of double (one processor op).
In short, if the new store is legal and has the same
size then I say that the transform is ok.  It would
also be possible to say that transforms are always
ok if before they were illegal, whether after they
are illegal or not, but that's more awkward to do
and I doubt it buys us anything much.
However this exposed an interesting thing - on x86-32
a store of i64 is considered legal!  That is because
operations are marked legal by default, regardless of
whether the type is legal or not.  In some ways this
is clever: before type legalization this means that
operations on illegal types are considered legal;
after type legalization there are no illegal types
so now operations are only legal if they really are.
But I consider this to be too cunning for mere mortals.
Better to do things explicitly by testing AfterLegalize.
So I have changed things so that operations with illegal
types are considered illegal - indeed they can never
map to a machine operation.  However this means that
the DAG combiner is more conservative because before
it was "accidentally" performing transforms where the
type was illegal because the operation was nonetheless
marked legal.  So in a few such places I added a check
on AfterLegalize, which I suppose was actually just
forgotten before.  This causes the DAG combiner to do
slightly more than it used to, which resulted in the X86
backend blowing up because it got a slightly surprising
node it wasn't expecting, so I tweaked it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52254 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix an error in the two-address example.
Dan Gohman [Fri, 13 Jun 2008 17:55:57 +0000 (17:55 +0000)]
Fix an error in the two-address example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52253 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUse recently added getTruncateOrZeroExtend method to make the code shorter.
Wojciech Matyjewicz [Fri, 13 Jun 2008 17:02:03 +0000 (17:02 +0000)]
Use recently added getTruncateOrZeroExtend method to make the code shorter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52251 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoXFAIL some tests that became failing due to the extra error reporting recently. PR...
Matthijs Kooijman [Fri, 13 Jun 2008 16:52:35 +0000 (16:52 +0000)]
XFAIL some tests that became failing due to the extra error reporting recently. PR's are created for these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52250 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoCrash less. The i64 restriction in BinomialCoefficient caused some problems
Nick Lewycky [Fri, 13 Jun 2008 04:38:55 +0000 (04:38 +0000)]
Crash less. The i64 restriction in BinomialCoefficient caused some problems
with code that was expecting different bit widths for different values.

Make getTruncateOrZeroExtend a method on ScalarEvolution, and use it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52248 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agofix a minor deviation from the original in my previous commit
Gabor Greif [Thu, 12 Jun 2008 21:51:29 +0000 (21:51 +0000)]
fix a minor deviation from the original in my previous commit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52247 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoop_iterator-ify some loops, low hanging fruit only, there is more
Gabor Greif [Thu, 12 Jun 2008 21:37:33 +0000 (21:37 +0000)]
op_iterator-ify some loops, low hanging fruit only, there is more

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52246 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix some tests.
Evan Cheng [Thu, 12 Jun 2008 21:23:38 +0000 (21:23 +0000)]
Fix some tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52245 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDo not speculatively execute an instruction by hoisting it to its predecessor BB...
Evan Cheng [Thu, 12 Jun 2008 21:15:59 +0000 (21:15 +0000)]
Do not speculatively execute an instruction by hoisting it to its predecessor BB if any of its operands are defined but not used in BB. The transformation will prevent the operand from being sunk into the use block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52244 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRevert 52223.
Evan Cheng [Thu, 12 Jun 2008 20:55:39 +0000 (20:55 +0000)]
Revert 52223.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52243 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSwitch GVN to use ScopedHashTable.
Owen Anderson [Thu, 12 Jun 2008 19:25:32 +0000 (19:25 +0000)]
Switch GVN to use ScopedHashTable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52242 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoImprove description of -emit-llvm.
Dale Johannesen [Thu, 12 Jun 2008 17:04:09 +0000 (17:04 +0000)]
Improve description of -emit-llvm.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52238 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDon't try to compile tests for the ev56 alpha subtarget, which hasn't been
Matthijs Kooijman [Thu, 12 Jun 2008 13:44:26 +0000 (13:44 +0000)]
Don't try to compile tests for the ev56 alpha subtarget, which hasn't been
supported since r33492.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52237 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoPass -silence-passes to bugpoint in testcases, this makes two out of three bugpoint...
Matthijs Kooijman [Thu, 12 Jun 2008 13:12:11 +0000 (13:12 +0000)]
Pass -silence-passes to bugpoint in testcases, this makes two out of three bugpoint testcases work again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52236 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoLet bugpoint display generated messages on stderr only if no interpreter was
Matthijs Kooijman [Thu, 12 Jun 2008 13:09:43 +0000 (13:09 +0000)]
Let bugpoint display generated messages on stderr only if no interpreter was
found, this ensures that messages like "Found gcc" end up on stdout where they
belong.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52235 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd -silence-passes option to bugpoint. This option suppresses output generated
Matthijs Kooijman [Thu, 12 Jun 2008 13:02:26 +0000 (13:02 +0000)]
Add -silence-passes option to bugpoint. This option suppresses output generated
when bugpoint is running passes in a child process.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52234 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix redirection of stderr in sys::Program::ExecuteAndWait. There was logic
Matthijs Kooijman [Thu, 12 Jun 2008 12:53:35 +0000 (12:53 +0000)]
Fix redirection of stderr in sys::Program::ExecuteAndWait. There was logic
error that caused it to redirect stderr to stdout too often.

This fix is applied identically to the win32 code as well, but that is
untested.
--Thi line, and those below, will be ignored--

M    System/Unix/Program.inc
M    System/Win32/Program.inc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52233 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoMake I/O redirection handling in sys::Program a bit more consistent. No
Matthijs Kooijman [Thu, 12 Jun 2008 10:47:18 +0000 (10:47 +0000)]
Make I/O redirection handling in sys::Program a bit more consistent. No
functional changes. Win32 code is untested, but should work fine.

In the unix variant, rename RedirectFD to RedirectIO and let that function
handle empty and null paths instead of doing that in the caller 3 times. This
is the same as win32 already does it.

In the win32 variant, use Path::isEmpty() instead of checking the resulting
c_str() manually. This is the same as unix already does it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52230 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoIf queried as to whether an operation is legal
Duncan Sands [Thu, 12 Jun 2008 09:01:30 +0000 (09:01 +0000)]
If queried as to whether an operation is legal
for a particular MVT, return false if the type
is illegal rather than barfing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52229 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd line continuation character so the avoid dup loop header test actually runs.
Matthijs Kooijman [Thu, 12 Jun 2008 08:49:04 +0000 (08:49 +0000)]
Add line continuation character so the avoid dup loop header test actually runs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52228 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoop_iterator-ify some loops, fix 80col violations
Gabor Greif [Wed, 11 Jun 2008 21:38:51 +0000 (21:38 +0000)]
op_iterator-ify some loops, fix 80col violations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52226 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoProperly lower DYNAMIC_STACKALLOC - bracket all black magic with
Anton Korobeynikov [Wed, 11 Jun 2008 20:16:42 +0000 (20:16 +0000)]
Properly lower DYNAMIC_STACKALLOC - bracket all black magic with
CALLSEQ_BEGIN & CALLSEQ_END.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52225 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFor now, avoid generating FP select instructions in order to speculatively execute...
Evan Cheng [Wed, 11 Jun 2008 19:18:20 +0000 (19:18 +0000)]
For now, avoid generating FP select instructions in order to speculatively execute integer arithmetic instructions. FP selects are more likely to be expensive (even compared to branch on fcmp). This is not a wonderful solution but I rather err on the side of conservative.

This fixes the heapsort performance regressions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52224 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAvoid duplicating loop header which leads to unnatural loops (and just seem like...
Evan Cheng [Wed, 11 Jun 2008 19:07:54 +0000 (19:07 +0000)]
Avoid duplicating loop header which leads to unnatural loops (and just seem like general badness to me, likely to cause code explosion).
Patch by Florian Brandner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52223 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoDon't send checkpoints to stderr for the vmcore.ml test.
Gordon Henriksen [Wed, 11 Jun 2008 14:58:01 +0000 (14:58 +0000)]
Don't send checkpoints to stderr for the vmcore.ml test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52218 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoTeach instruction combining about the extractvalue. It can succesfully fold
Matthijs Kooijman [Wed, 11 Jun 2008 14:05:05 +0000 (14:05 +0000)]
Teach instruction combining about the extractvalue. It can succesfully fold
useless insert-extract chains, similar to how it folds them for vectors.

Add a testcase for this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52217 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoremove bogus comment
Gabor Greif [Wed, 11 Jun 2008 13:53:41 +0000 (13:53 +0000)]
remove bogus comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52216 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoop_iterator-ify dropAllReferences
Gabor Greif [Wed, 11 Jun 2008 11:45:26 +0000 (11:45 +0000)]
op_iterator-ify dropAllReferences

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52215 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSometimes (rarely) nodes held in LegalizeTypes
Duncan Sands [Wed, 11 Jun 2008 11:42:12 +0000 (11:42 +0000)]
Sometimes (rarely) nodes held in LegalizeTypes
maps can be deleted.  This happens when RAUW
replaces a node N with another equivalent node
E, deleting the first node.  Solve this by
adding (N, E) to ReplacedNodes, which is already
used to remap nodes to replacements.  This means
that deleted nodes are being allowed in maps,
which can be delicate: the memory may be reused
for a new node which might get confused with the
old deleted node pointer hanging around in the
maps, so detect this and flush out maps if it
occurs (ExpungeNode).  The expunging operation
is expensive, however it never occurs during
a llvm-gcc bootstrap or anywhere in the nightly
testsuite.  It occurs three times in "make check":
Alpha/illegal-element-type.ll,
PowerPC/illegal-element-type.ll and
X86/mmx-shift.ll.  If expunging proves to be too
expensive then there are other more complicated
ways of solving the problem.
In the normal case this patch adds the overhead
of a few more map lookups, which is hopefully
negligable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52214 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoBetter test for availability of __gnu_cxx::stdio_filebuf.
Gordon Henriksen [Wed, 11 Jun 2008 10:46:24 +0000 (10:46 +0000)]
Better test for availability of __gnu_cxx::stdio_filebuf.
If this doesn't work, I'll write a configure test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52213 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoClarify a comment.
Matthijs Kooijman [Wed, 11 Jun 2008 09:00:12 +0000 (09:00 +0000)]
Clarify a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52212 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoop_iterator-ify loops
Gabor Greif [Tue, 10 Jun 2008 22:03:26 +0000 (22:03 +0000)]
op_iterator-ify loops

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52191 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSmall doc fix.
Duncan Sands [Tue, 10 Jun 2008 20:54:01 +0000 (20:54 +0000)]
Small doc fix.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52190 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUse %link not %llvmgxx (which includes -c) to do the link.
Dale Johannesen [Tue, 10 Jun 2008 18:01:54 +0000 (18:01 +0000)]
Use %link not %llvmgxx (which includes -c) to do the link.
The test still fails because an expected symbol is not
present, and I don't see why it should be.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52188 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSuppress ObjC FE warnings, which cause the test to fail.
Dale Johannesen [Tue, 10 Jun 2008 18:00:45 +0000 (18:00 +0000)]
Suppress ObjC FE warnings, which cause the test to fail.
Warnings are legitimate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52187 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAdd -w to inhibit gcc warnings, which causes the
Dale Johannesen [Tue, 10 Jun 2008 18:00:09 +0000 (18:00 +0000)]
Add -w to inhibit gcc warnings, which causes the
harness to fail the tests.  The warning all appear
legitimate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52186 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix parameter spelling: sse not sse1
Dale Johannesen [Tue, 10 Jun 2008 17:57:58 +0000 (17:57 +0000)]
Fix parameter spelling:  sse not sse1

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52185 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoIgnore stderr for some more tests that expect warnings there.
Matthijs Kooijman [Tue, 10 Jun 2008 16:13:38 +0000 (16:13 +0000)]
Ignore stderr for some more tests that expect warnings there.

This fixes 2 testcases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52184 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix some more quoting issues in RUN lines, this time regarding unintended
Matthijs Kooijman [Tue, 10 Jun 2008 16:10:32 +0000 (16:10 +0000)]
Fix some more quoting issues in RUN lines, this time regarding unintended
variable expansions involving the $ character.

This fixes 4 tests that were not running properly before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52183 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix some escaping and quoting in RUN lines, mainly involving { and <. In two
Matthijs Kooijman [Tue, 10 Jun 2008 16:04:47 +0000 (16:04 +0000)]
Fix some escaping and quoting in RUN lines, mainly involving { and <. In two
cases quoting of <{ didn't work out, so I changed the grep to check for }>
instead.

This fixes 7 testcases that were not properly running before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52182 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove double pipes in RUN commandlines.
Matthijs Kooijman [Tue, 10 Jun 2008 15:11:36 +0000 (15:11 +0000)]
Remove double pipes in RUN commandlines.

This fixes 5 testcases that were not being run properly before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52180 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove trailing whitespace after line continuations in test cases to them work.
Matthijs Kooijman [Tue, 10 Jun 2008 15:07:07 +0000 (15:07 +0000)]
Remove trailing whitespace after line continuations in test cases to them work.

This fixes two test cases that were not being run properly before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52179 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoLet some more tests ignore expected output on stderr.
Matthijs Kooijman [Tue, 10 Jun 2008 15:04:14 +0000 (15:04 +0000)]
Let some more tests ignore expected output on stderr.
Also, use > %t instead of -o %t for output in one test since that also works
when %t already exists.

This fixes 6 testcases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52178 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove some debug output from tblgen. This fixes a test.
Matthijs Kooijman [Tue, 10 Jun 2008 14:53:49 +0000 (14:53 +0000)]
Remove some debug output from tblgen. This fixes a test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52177 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix some llvm-gcc warnings in testcases, mostly by adding includes or adding
Matthijs Kooijman [Tue, 10 Jun 2008 14:37:44 +0000 (14:37 +0000)]
Fix some llvm-gcc warnings in testcases, mostly by adding includes or adding
declarations. These are the fixes that I was pretty confident about, there are
still a lot of other llvm-gcc warnings of which I'm not sure if they can be
safely ignored or fixed, without breaking the test case.

This fixes 11 testcases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52176 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFor all RUN lines starting with "not", redirect stderr to /dev/null so tests
Matthijs Kooijman [Tue, 10 Jun 2008 12:57:32 +0000 (12:57 +0000)]
For all RUN lines starting with "not", redirect stderr to /dev/null so tests
don't fail when (expected) error output is produced. This fixes 17 tests.

While I was there, I also made all RUN lines of the form "not llvm-as..." a bit
more consistent, they now all redirect stderr and stdout to /dev/null and use
input redirect to read their input.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52174 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSuppress the (stderr) output of -aa-eval, this fixes 5 tests.
Matthijs Kooijman [Tue, 10 Jun 2008 12:39:15 +0000 (12:39 +0000)]
Suppress the (stderr) output of -aa-eval, this fixes 5 tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52173 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoChange llvm.exp so it no longer ignores some errors when executing dejagnu
Matthijs Kooijman [Tue, 10 Jun 2008 12:28:43 +0000 (12:28 +0000)]
Change llvm.exp so it no longer ignores some errors when executing dejagnu
tests. This breaks 80 tests in the tree.

The interesting part here is that this no longer ignores syntax errors
in RUN command lines. Some tests have not been working all the time because of
this.

The tricky part is that it now also views any stderr output as an error. This
can be suppressed in tcl 8.5, but let's not add this dependency. Instead, all
testcases should be changed to redirect stderr if they expect stderr output.
This holds in particular for lines like:
  ; RUN: not llvm-as < %s
where an error is expected (but I think I can solve this by modifying the not
script). Also, compilations resulting in warnings will now also fail (so
the warnings should be fixed, disabled or redirected...).

I'll continue with fixing the testcases that are broken now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52172 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoadjust calls to ConstantFP::get to new API
Gabor Greif [Tue, 10 Jun 2008 01:52:17 +0000 (01:52 +0000)]
adjust calls to ConstantFP::get to new API

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52165 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoConvert several tests to use temporary files instead of redundantly
Dan Gohman [Tue, 10 Jun 2008 00:36:41 +0000 (00:36 +0000)]
Convert several tests to use temporary files instead of redundantly
executing the test commands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52163 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix two more not-grep tests that were missing llvm-dis.
Dan Gohman [Mon, 9 Jun 2008 22:36:45 +0000 (22:36 +0000)]
Fix two more not-grep tests that were missing llvm-dis.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52159 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoTeach isGAPlusOffset to respect a GlobalAddressSDNode's offset
Dan Gohman [Mon, 9 Jun 2008 22:05:52 +0000 (22:05 +0000)]
Teach isGAPlusOffset to respect a GlobalAddressSDNode's offset
value, which is something that apparently isn't used much.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52158 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRe-apply 52002, allowing the verifier to accept non-MRV struct return
Dan Gohman [Mon, 9 Jun 2008 21:26:13 +0000 (21:26 +0000)]
Re-apply 52002, allowing the verifier to accept non-MRV struct return
types on functions, with adjustments so that it accepts both
new-style aggregate returns and old-style MRV returns, including those
with only a single member.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52157 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoCodeGen support for aggregate-value function arguments.
Dan Gohman [Mon, 9 Jun 2008 21:19:23 +0000 (21:19 +0000)]
CodeGen support for aggregate-value function arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52156 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUpdate the build instructions.
Duncan Sands [Mon, 9 Jun 2008 20:24:05 +0000 (20:24 +0000)]
Update the build instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52153 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoVarious tweaks related to apint codegen. No functionality
Duncan Sands [Mon, 9 Jun 2008 15:48:25 +0000 (15:48 +0000)]
Various tweaks related to apint codegen.  No functionality
change for non-funky-sized integers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52151 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoHandle empty aggregate values.
Dan Gohman [Mon, 9 Jun 2008 15:21:47 +0000 (15:21 +0000)]
Handle empty aggregate values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52150 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAsmParser support for immediate constant aggregate values.
Dan Gohman [Mon, 9 Jun 2008 14:45:02 +0000 (14:45 +0000)]
AsmParser support for immediate constant aggregate values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52149 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoCPPBackend support for extractvalue and insertvalue.
Dan Gohman [Mon, 9 Jun 2008 14:12:10 +0000 (14:12 +0000)]
CPPBackend support for extractvalue and insertvalue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52147 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoAbort on an unrecognized opcode.
Dan Gohman [Mon, 9 Jun 2008 14:09:13 +0000 (14:09 +0000)]
Abort on an unrecognized opcode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52146 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoUpdate the CPP backend for the ConstantFP::get API change.
Dan Gohman [Mon, 9 Jun 2008 14:08:11 +0000 (14:08 +0000)]
Update the CPP backend for the ConstantFP::get API change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52144 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoRemove some DAG combiner assumptions about sizes
Duncan Sands [Mon, 9 Jun 2008 11:32:28 +0000 (11:32 +0000)]
Remove some DAG combiner assumptions about sizes
of integer types.  Fix the isMask APInt method to
actually work (hopefully) rather than crashing
because it adds apints of different bitwidths.
It looks like isShiftedMask is also broken, but
I'm leaving that one to the APInt people (it is
not used anywhere).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52142 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoTest that prune-eh doesn't make deductions based
Duncan Sands [Mon, 9 Jun 2008 11:28:41 +0000 (11:28 +0000)]
Test that prune-eh doesn't make deductions based
on bodies of functions with weak linkage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52141 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoadd support for PIC on linux x86-64
Rafael Espindola [Mon, 9 Jun 2008 09:52:31 +0000 (09:52 +0000)]
add support for PIC on linux x86-64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52139 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agolower calls to abs to inline code, PR2337
Chris Lattner [Mon, 9 Jun 2008 08:26:51 +0000 (08:26 +0000)]
lower calls to abs to inline code, PR2337

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52138 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoFix PR2411, where ip constant prop would propagate the
Chris Lattner [Mon, 9 Jun 2008 07:58:07 +0000 (07:58 +0000)]
Fix PR2411, where ip constant prop would propagate the
result of a weak function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52137 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agoSmall formatting changes.
Bill Wendling [Mon, 9 Jun 2008 07:51:45 +0000 (07:51 +0000)]
Small formatting changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52136 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agouse 'continue' to make the function linker simpler. When linking a strong
Chris Lattner [Mon, 9 Jun 2008 07:47:34 +0000 (07:47 +0000)]
use 'continue' to make the function linker simpler.  When linking a strong
function into a weak function, zap the weak function body so that the
strong one overrides it.  This fixes PR2410

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52135 91177308-0d34-0410-b5e6-96231b3b80d8

16 years agominor changes to short circuit the 'no linkage' case earlier for
Chris Lattner [Mon, 9 Jun 2008 07:36:11 +0000 (07:36 +0000)]
minor changes to short circuit the 'no linkage' case earlier for
function bodies.  We now don't try to unify types or handling type
mismatches if when linking an internal foo to an external foo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52134 91177308-0d34-0410-b5e6-96231b3b80d8