Commit Graph

1889 Commits

Author SHA1 Message Date
fd6edef87b Teach ScalarEvolution to consider loop preheaders in the search for
an if statement that guards a loop, to allow indvars to avoid smax
operations in more situations.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56232 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-15 22:18:04 +00:00
8dae138d06 Fix WriteAsOperand to not emit a leading space character. Adjust
its callers to emit a space character before calling it when a
space is needed.

This fixes several spurious whitespace issues in
ScalarEvolution's debug dumps. See the test changes for
examples.

This also fixes odd space-after-tab indentation in the output
for switch statements, and changes calls from being printed like
this:
  call void @foo( i32 %x )
to this:
  call void @foo(i32 %x)


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56196 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-14 17:21:12 +00:00
b8ca4ff643 Fix PR2792: treat volatile loads as writing memory somewhere.
Treat stores as reading memory, just to play safe.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56188 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-13 12:45:50 +00:00
2bb4a4d315 Rather than marking all internal globals "Ref"
when a readonly declaration is called, set a
flag.  This is faster and uses less memory.
In theory it is less accurate, because before
only those internal globals that were read
by someone were being marked "Ref", but now
all are.  But in practice, thanks to other
passes, all internal globals of the kind
considered here will be both read and stored
to: those only read will have been turned
into constants, and those only stored to will
have been deleted.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56143 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-12 07:29:58 +00:00
af8bc26449 Fix comment typo.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56116 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-11 19:41:10 +00:00
892b84007c Intrinsics don't touch internal global variables
(unless passed one via a parameter), even if they
are IntrWriteMem.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56115 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-11 19:35:55 +00:00
1abe60b9be Intrinsics don't read these kinds of global
variables.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56105 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-11 15:43:12 +00:00
7ca9d81bce Simplify this some more. No functionality change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56003 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-09 19:56:34 +00:00
24a052124b Optimization suggested by Matthijs Kooijman.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55988 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-09 13:44:24 +00:00
99c1a7c9e1 Correct callgraph construction. It has two problems:
(1) code left over from the days of ConstantPointerRef:
if a use of a function is a GlobalValue then that is
not considered a reason to add an edge from the external
node, even though the use may be as an initializer for
an externally visible global!  There might be some point
to this behaviour when the use is by an alias (though the
code predated aliases by some centuries), but I think
PR2782 is a better way of handling that.  (2) If function
F calls function G, and also G is a parameter to the
call, then an F->G edge is not added to the callgraph.
While this doesn't seem to matter much, adding such an
edge makes the callgraph more regular.
In addition, the new code should be faster as well as
simpler.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55987 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-09 12:40:47 +00:00
28f02128ae Didn't mean to commit this change to how the
callgraph is printed.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55912 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-08 16:04:03 +00:00
d7b9851c4e Reapply 55859. This doesn't change anything as
long as the callgraph is correct.  It checks
for wrong callgraphs more strictly.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55894 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-08 11:05:51 +00:00
fec2c2bf5e When PruneEH turned an invoke into an ordinary
call (thus changing the call site) it didn't
inform the callgraph about this.  But the
call site does matter - as shown by the testcase,
the callgraph become invalid after the inliner
ran (with an edge between two functions simply
missing), resulting in wrong deductions by
GlobalsModRef.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55872 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-06 17:19:29 +00:00
6e6074508c Revert r55859. This is breaking the build in the abscence of its companion commit.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55865 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-05 23:36:01 +00:00
c94fe21a56 Delete the removeCallEdgeTo callgraph method,
because it does not maintain a correct list
of callsites.  I discovered (see following
commit) that the inliner will create a wrong
callgraph if it is fed a callgraph with
correct edges but incorrect callsites.  These
were created by Prune-EH, and while it wasn't
done via removeCallEdgeTo, it could have been
done via removeCallEdgeTo, which is an accident
waiting to happen.  Use removeCallEdgeFor
instead.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55859 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-05 21:43:04 +00:00
f423abc289 Neaten this up a bit. No functionality change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55789 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-04 19:16:20 +00:00
ae73dc1448 Tidy up several unbeseeming casts from pointer to intptr_t.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55779 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-04 17:05:41 +00:00
34f2a0ca17 If a SCC has a node without a function, then the SCC
analysis would bail out without removing function
records for other members of the SCC (which may exist
if those functions read or wrote global variables).
Since these are initialized to "readnone", this
resulted in incorrect alias analysis results.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55714 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-03 19:37:16 +00:00
b070beee77 Fix maxo bado thinko.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55700 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-03 16:10:55 +00:00
d0ac373660 Since onlyReadsMemory returns true if in fact
doesNotAccessMemory, check doesNotAccessMemory
first, since otherwise functions may be
marked readonly rather than readnone.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55697 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-03 15:31:24 +00:00
9a036b945c Cleanup GlobalsModRef a bit. When analysing the
callgraph, when one member of a SCC calls another
then the analysis would drop to mod-ref because
there is (usually) no function info for the callee
yet; fix this.  Teach the analysis about function
attributes, in particular the readonly attribute
(which requires being careful about globals).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55696 91177308-0d34-0410-b5e6-96231b3b80d8
2008-09-03 12:55:42 +00:00
db57ef1cb3 rename destroy -> releaseMemory to properly hook into passmgr.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55508 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-28 22:56:53 +00:00
1e1ef3e2bd Clear the intervals list in "destroy", patch by
Prakash Prabhu!



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55458 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-28 03:33:03 +00:00
944fac71e0 Switch the asmprinter (.ll) and all the stuff it requires over to
use raw_ostream instead of std::ostream.  Among other goodness,
this speeds up llvm-dis of kc++ with a release build from 0.85s
to 0.49s (88% faster).

Other interesting changes:
 1) This makes Value::print be non-virtual.
 2) AP[S]Int and ConstantRange can no longer print to ostream directly, 
    use raw_ostream instead.
 3) This fixes a bug in raw_os_ostream where it didn't flush itself 
    when destroyed.
 4) This adds a new SDNode::print method, instead of only allowing "dump".


A lot of APIs have both std::ostream and raw_ostream versions, it would
be useful to go through and systematically anihilate the std::ostream 
versions.

This passes dejagnu, but there may be minor fallout, plz let me know if
so and I'll fix it.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55263 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-23 22:23:09 +00:00
d611cf6dcd Temporarily reverting r55137. This was causing the bootstrap to go into an
infinite loop.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55149 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-21 22:40:10 +00:00
dce51c372a Fix ComputeMaskedBits to handle phis correctly. We need to take the
minimum of the known zeros.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@55137 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-21 20:45:12 +00:00
2c69184fd6 Don't use the result of WriteAsOperand or WriteTypeSymbolic.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54977 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-19 04:42:37 +00:00
3688f268cb Remove GCSE, ValueNumbering, and LoadValueNumbering. These have been deprecated for almost a year; it's finally time for them to go away.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54822 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-15 21:31:02 +00:00
a60832b018 Fix a bogus srem rule - a negative value srem'd by a power-of-2
can have a non-negative result; for example, -16%16 is 0. Also,
clarify the related comments. This fixes PR2670.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54767 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-13 23:12:35 +00:00
81b06be055 Teach constant folding that an inttoptr of a
ptrtoint can be turned into a bitcast if the
integer is at least as wide as a pointer.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54752 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-13 20:20:35 +00:00
3837218ea0 Extend ScalarEvolution's executesAtLeastOnce logic to be able to
continue past the first conditional branch when looking for a
relevant test. This helps it avoid using MAX expressions in
loop trip counts in more cases.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54697 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-12 20:17:31 +00:00
afcde473c5 "This patch adds a virtual call to AbstractLatticeFunction to derive a
type lattice value for an Argument*, giving clients the opportunity to
use something other than Top for it if they choose to."

Patch by John McCall!



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54589 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-09 17:23:35 +00:00
d9cc749318 Canonicalize nested AddRecs in by nesting them in order of loop depth.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54545 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-08 18:33:12 +00:00
0dabb7e8f3 Don't call getAnalysisUsage unless -debug-pass is enabled. This speeds
up the passmgr by avoiding useless work.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54528 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-08 15:14:09 +00:00
b42a626122 PR2621: Improvements to the SCEV AddRec binomial expansion. This
version uses a new algorithm for evaluating the binomial coefficients 
which is significantly more efficient for AddRecs of more than 2 terms 
(see the comments in the code for details on how the algorithm works).  
It also fixes some bugs: it removes the arbitrary length restriction for 
AddRecs, it fixes the silent generation of incorrect code for AddRecs 
which require a wide calculation width, and it fixes an issue where we 
were incorrectly truncating the iteration count too far when evaluating 
an AddRec expression narrower than the induction variable.

There are still a few related issues I know of: I think there's 
still an issue with the SCEVExpander expansion of AddRec in terms of
the width of the induction variable used.  The hack to avoid generating 
too-wide integers shouldn't be necessary; instead, the callers should be 
considering the cost of the expansion before expanding it (in addition 
to not expanding too-wide integers, we might not want to expand 
expressions that are really expensive, especially when optimizing for 
size; calculating an length-17 32-bit AddRec currently generates about 250 
instructions of straight-line code on X86).  Also, for long 32-bit 
AddRecs on X86, CodeGen really sucks at scheduling the code.  I'm planning on 
filing follow-up PRs for these issues.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54332 91177308-0d34-0410-b5e6-96231b3b80d8
2008-08-04 23:49:06 +00:00
1fbffe0cef Another SCEV issue from PR2607; essentially the same issue, but this
time applying to the implicit comparison in smin expressions. The 
correct way to transform an inequality into the opposite 
inequality, either signed or unsigned, is with a not expression.

I looked through the SCEV code, and I don't think there are any more 
occurrences of this issue.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54194 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-30 04:36:32 +00:00
068acc3caa Fix for PR2607: SCEV miscomputing the loop count for loops with an
SGT exit condition.  Essentially, the correct way to flip an inequality 
in 2's complement is the not operator, not the negation operator.  
That said, the difference only affects cases involving INT_MIN.

Also, enhance the pre-test search logic to be a bit smarter about 
inequalities flipped with a not operator, so it can eliminate the smax 
from the iteration count for simple loops.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54184 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-30 00:04:08 +00:00
d4310a5d41 Fix a subtle bug when removing instructions from memdep. In very specific
circumstances we could end up remapping a dependee to the same instruction 
that we're trying to remove.  Handle this properly by just falling back to
a conservative solution.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54132 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-28 16:00:58 +00:00
b5557abcf1 Fix minor issues with VICmp/VFCmp constant expressions
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54030 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-25 17:35:37 +00:00
9e13cbc171 Revert r53812 -- premature. LegalizeTypes isn't actually on yet!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53816 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-21 04:03:00 +00:00
2ceb40f3da Switch on the use of arbitrary precision integers in scalar evolution. This will
bail after 256-bits to avoid producing code that the backends can't handle.
Previously, we capped it at 64-bits, preferring to miscompile in those cases.

This change also reverts much of r52248 because the invariants the code was
expecting are now being met.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53812 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-21 02:51:31 +00:00
8b82c49084 This header isn't necessary now.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53811 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-20 17:35:32 +00:00
de0f2382e4 Fix PR2088. Use modulo linear equation solver to compute loop iteration
count.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53810 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-20 15:55:14 +00:00
dddc827125 Don't use ++idx_begin when I actually mean idx_begin + 1, especially since we
also use *idx_begin in the same expression, giving unpredictable results.

This fixes this bug: http://lists.cs.uiuc.edu/pipermail/llvmdev/2008-July/015877.html



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53670 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-16 10:47:35 +00:00
8ae38e1516 Correct this inversion!
I swear that didn't show up in svn diff...


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53587 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-15 03:47:44 +00:00
86dae65baa Fix up comments.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53586 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-15 03:40:27 +00:00
59cff12f88 Stop creating extraneous smax/umax in SCEV. This removes a regression where we
started complicating many loops ('for' loops, in fact).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53508 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-12 07:41:32 +00:00
e6acf36dca Fix spelling of "hierarchy" in comments.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53489 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-11 22:51:32 +00:00
c418bf3dd5 Use find instead of lower_bound.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53474 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-11 20:58:19 +00:00
b0b046848f Remove getValueRange from SCEV. It wasn't doing anything there anyways, and a
more complete version is now available from the LoopVR pass.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@53269 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-09 03:21:51 +00:00