907dc2bc37
Fix PR2355: bug in ChangeCompareStride. When the loop termination compare is the only use of its iv stride, the stride can be eliminated by moving it to another stride. If the scale is negative, swap the predicate instead of using a inverse predicate.
...
llvm-svn: 54415
2008-08-06 18:04:43 +00:00
f5b353c1fd
optimize a common idiom generated by clang for bitfield access, PR2638.
...
llvm-svn: 54408
2008-08-06 07:35:52 +00:00
7bdaecb7f4
Zap sitofp/fptoui pairs. In all cases when the sign difference
...
matters, the result is undefined anyway.
llvm-svn: 54396
2008-08-06 05:13:06 +00:00
bf42893567
Reinstate this optimization, but without the miscompile. Thanks to Bill for
...
tracking down that this was breaking llvm-gcc bootstrap on Linux.
llvm-svn: 54394
2008-08-06 04:54:03 +00:00
1fcc804cfd
Pass the computed iteration count value to RewriteLoopExitValues
...
instead of having it call getIterationCount again.
llvm-svn: 54380
2008-08-05 22:34:21 +00:00
ee12a7aeff
Revert r53282. This was causing a miscompile on Linux. Also, the transformation
...
looks bogus. Please see PR2629 for details on why this is breaking things.
llvm-svn: 54372
2008-08-05 21:23:45 +00:00
3da016d137
Trim #includes.
...
llvm-svn: 54350
2008-08-05 15:32:23 +00:00
c1e48b582d
Fix comment typos.
...
llvm-svn: 54266
2008-08-01 12:23:49 +00:00
fecbc8cff1
Add vector shifts to the IR, patch by Eli Friedman.
...
CodeGen & Clang work coming next.
llvm-svn: 54161
2008-07-29 15:49:41 +00:00
98b5c16e3b
Add -unroll-allow-partial command line option that enabled the loop unroller to
...
partially unroll a loop when fully unrolling would not fit under the threshold.
Patch by Mikael Lepistö.
llvm-svn: 54160
2008-07-29 13:21:23 +00:00
fd3070459b
Restructure ArgumentPromotion a bit. Instead of just having a single boolean
...
that says "unconditional loads from this argument are safe", we now keep track
of the safety per set of indices from which loads happen. This prevents
ArgPromotion from promoting loads that aren't really valid. As an added effect,
this will now disregard the the type of the indices passed to a GEP, so
"load GEP %A, i32 1" and "load GEP %A, i64 1" will result in a single argument,
not two.
This fixes PR2598, for which a testcase has been added as well.
llvm-svn: 54159
2008-07-29 10:00:13 +00:00
813bf7af7f
Don't remove volatile loads. Thanks to Duncan for noticing this one.
...
llvm-svn: 54144
2008-07-28 20:52:42 +00:00
3f3389745d
Add support for eliminating stores that store the same value that was just loaded.
...
This fixes PR2599.
llvm-svn: 54133
2008-07-28 16:14:26 +00:00
2ce6f2ad5e
Rename SDOperand to SDValue.
...
llvm-svn: 54128
2008-07-27 21:46:04 +00:00
5f36a32e7b
Put the LICM of constant GlobalVariables, introduced in r53945, under a
...
command-line option, and disable it by default. It introduced performance
regressions because CodeGen is currently not able to remat such loads.
llvm-svn: 53997
2008-07-24 23:57:25 +00:00
8a8fb908dc
"Allow LICM to sink or lift loads from constant memory. Also add a test
...
case for this.
This allows instructions like loads from global variables declared to
be constant to be moved out of loops."
Patch by Stefanus Du Toit!
llvm-svn: 53945
2008-07-23 05:06:28 +00:00
fa1211f69b
Enable first-class aggregates support.
...
Remove the GetResultInst instruction. It is still accepted in LLVM assembly
and bitcode, where it is now auto-upgraded to ExtractValueInst. Also, remove
support for return instructions with multiple values. These are auto-upgraded
to use InsertValueInst instructions.
The IRBuilder still accepts multiple-value returns, and auto-upgrades them
to InsertValueInst instructions.
llvm-svn: 53941
2008-07-23 00:34:11 +00:00
7ad3cd8c9d
Fix a bug in LSR's dead-PHI cleanup. If a PHI has a def-use chain that
...
leads into a cycle involving a different PHI, LSR got stuck running
around that cycle looking for the original PHI. To avoid this, keep
track of visited PHIs and stop searching if we see one more than once.
This fixes PR2570.
llvm-svn: 53879
2008-07-21 21:45:02 +00:00
2c741145a7
Supress a gcc-4.3 warning.
...
llvm-svn: 53771
2008-07-18 21:06:02 +00:00
04a6e0ba8c
Make PRE actually handle critical edges (by splitting them). Confirmed that bootstrap passes with this change.
...
llvm-svn: 53762
2008-07-18 18:03:38 +00:00
9858691f25
Reapply r53735. My last patch fixed the failures Dan observed.
...
llvm-svn: 53761
2008-07-18 17:49:43 +00:00
1468bec06e
Add some checks that got lost in the shuffle. This fixes 464.h264ref.
...
llvm-svn: 53760
2008-07-18 17:46:41 +00:00
29c3adaae0
Revert r53735. It broke SPEC 464.h264ref.
...
llvm-svn: 53757
2008-07-18 16:44:49 +00:00
fd7102037d
Use MergeBlockIntoPredecessor to simplify some code.
...
llvm-svn: 53735
2008-07-17 20:00:46 +00:00
27405efdc0
Make MergeBlockIntoPredecessor more aggressive when the same successor appears
...
more than once.
llvm-svn: 53731
2008-07-17 19:42:29 +00:00
addbe3eed1
Enable PRE. My last batch of changes fixed the miscompile.
...
llvm-svn: 53730
2008-07-17 19:41:00 +00:00
8b69d77a7a
Make GlobalOpt preserve address spaces when scalar replacing aggregate globals.
...
llvm-svn: 53716
2008-07-17 11:59:53 +00:00
c600c53d1f
Fix PR2553
...
llvm-svn: 53715
2008-07-17 06:07:20 +00:00
97cd0298cc
Inliner tweak. Function calls should cost more than one instruction!
...
llvm-svn: 53712
2008-07-17 01:31:49 +00:00
c062381c7b
Factor MergeBlockIntoPredecessor out into BasicBlockUtils.
...
llvm-svn: 53705
2008-07-17 00:01:40 +00:00
ac31096311
There's no need to iterate block merging and PRE. In fact, iterating the latter
...
could cause problems for memdep when it breaks critical edges.
llvm-svn: 53691
2008-07-16 17:52:31 +00:00
c1d7477ed2
Redo InstCombiner::visitExtractValueInst. Instead of using the (complicate)
...
FindInsertedValue, it now performs a number of simple transformations that
should result in the same effect when applied iteratively.
llvm-svn: 53673
2008-07-16 12:55:45 +00:00
c97094552c
Fix PR2296. Do not transform x86_sse2_storel_dq into a full-width store.
...
llvm-svn: 53666
2008-07-16 07:28:14 +00:00
24768e3dc4
Revert this, as it seems to still be broken.
...
llvm-svn: 53627
2008-07-15 17:59:02 +00:00
9d1f497a28
Enable local PRE by default.
...
llvm-svn: 53616
2008-07-15 16:28:23 +00:00
53d546e40b
Have GVN do a pre-pass over the CFG that folds away unconditional branches where possible. This allows local PRE to be more aggressive.
...
llvm-svn: 53615
2008-07-15 16:28:06 +00:00
c893bf472d
Allow deadargelim to change return types even though now values were dead. This
...
again canonicalizes {i32} into i32 and {} into void.
llvm-svn: 53610
2008-07-15 14:42:31 +00:00
5e8c022e21
Revert r53606. It turns out that explicitely tracking the liveness of the
...
return value as a whole in deadargelim is really not needed now that we simply
rebuild the old return value and actually prevents some canonicalization from
taking place.
This revert stops deadargelim from changing {i32} into i32 for now, but I'll
fix that next.
llvm-svn: 53609
2008-07-15 14:39:36 +00:00
c1da874478
Make deadargelim a bit less smart, so it doesn't choke on nested structs as
...
return values that are still (partially) live. Instead of updating all uses of
a call instruction after removing some elements, it now just rebuilds the
original struct (With undef gaps where the unused values were) and leaves it to
instcombine to clean this up.
The added testcase still fails currently, but this is due to instcombine which
isn't good enough yet. I will fix that part next.
llvm-svn: 53608
2008-07-15 14:03:10 +00:00
04d4c328ac
Don't use isa when we can reuse a previous dyn_cast.
...
llvm-svn: 53607
2008-07-15 13:39:08 +00:00
84194b6768
Make DeadArgElim keep liveness of the return value as a whole in addition to
...
only the liveness of partial return values (for functions returning a struct).
This is more explicit to prevent unwanted changes in the return value.
In particular, deadargelim now canonicalizes a function returning {i32} to
returning i32 and {} to void, if the struct returned is not used in its
entirety, but only the single element is used.
llvm-svn: 53606
2008-07-15 13:36:06 +00:00
79a8eb547c
Let DAE keep a list of live functions, instead of simply marking all arguments
...
and return values live for those functions. This doesn't change anything yet,
but prepares for the coming commits.
llvm-svn: 53601
2008-07-15 09:11:16 +00:00
e9af814669
Split DAE::MarkLive into MarkLive and PropagateLiveness.
...
llvm-svn: 53600
2008-07-15 09:00:17 +00:00
2ce5709e31
Pass around const RetOrArg references instead of copying values. Also, mark
...
RetOrArg::getDescription() as const.
llvm-svn: 53599
2008-07-15 08:56:49 +00:00
f2860b9fb3
Simplify debug code by using RetOrArg::getDescription().
...
llvm-svn: 53598
2008-07-15 08:53:36 +00:00
90d08addb0
Fix indentation (intentionally left out of the previous commit).
...
llvm-svn: 53592
2008-07-15 08:47:32 +00:00
06642d3812
Move the deadargelim code for intrinsically alive functions into its own
...
method, to slightly simplify control flow.
llvm-svn: 53591
2008-07-15 08:45:12 +00:00
162668fa78
Fix uninitialized use of the Changed variable.
...
llvm-svn: 53564
2008-07-14 17:55:01 +00:00
8882b1c41c
Reapply r53540, now with the matching header!
...
llvm-svn: 53557
2008-07-14 17:32:59 +00:00
68b0383057
Revert r53540 - it does not compile.
...
llvm-svn: 53549
2008-07-14 07:59:28 +00:00
2831ad28be
If a function calls setjmp, never inline it into other functions. This is
...
a hack around the fact that we don't represent the CFG correctly for sj/lj.
It fixes PR2486.
llvm-svn: 53540
2008-07-14 00:46:56 +00:00
6f5ea6e49c
simplify some code, shuffle and insertelt always return a vector.
...
llvm-svn: 53538
2008-07-14 00:32:20 +00:00
16395e51f4
Fix PR2506 by being a bit more careful about reverse fact propagation when
...
disproving a condition. This actually compiles the existing testcase
(udiv_select_to_select_shift) to:
define i64 @test(i64 %X, i1 %Cond) {
entry:
%divisor1.t = lshr i64 %X, 3 ; <i64> [#uses=1]
%quotient2 = lshr i64 %X, 3 ; <i64> [#uses=1]
%sum = add i64 %divisor1.t, %quotient2 ; <i64> [#uses=1]
ret i64 %sum
}
instead of:
define i64 @test(i64 %X, i1 %Cond) {
entry:
%quotient1.v = select i1 %Cond, i64 3, i64 4 ; <i64> [#uses=1]
%quotient1 = lshr i64 %X, %quotient1.v ; <i64> [#uses=1]
%quotient2 = lshr i64 %X, 3 ; <i64> [#uses=1]
%sum = add i64 %quotient1, %quotient2 ; <i64> [#uses=1]
ret i64 %sum
}
llvm-svn: 53534
2008-07-14 00:15:52 +00:00
80b03a1b49
Fix mishandling of the infinite loop case when merging two blocks. This
...
fixes PR2540.
llvm-svn: 53533
2008-07-13 22:23:11 +00:00
834ab4ec1b
more refactoring. Use early exits instead of really complex logic.
...
No functionality change.
llvm-svn: 53532
2008-07-13 22:04:41 +00:00
5eed37224a
improve comments.
...
llvm-svn: 53531
2008-07-13 21:55:46 +00:00
9aada1d755
factor another large hunk of code out into its own function.
...
No functionality change.
llvm-svn: 53530
2008-07-13 21:53:26 +00:00
55eaae1e0c
Final bit of simplification for FoldBranchToCommonDest.
...
llvm-svn: 53528
2008-07-13 21:20:19 +00:00
1b317ea48a
simplify logic a bit
...
llvm-svn: 53527
2008-07-13 21:15:11 +00:00
2e25b8f444
Refactor some code out into its own helper function, getting rid of crazy multiline
...
conditionals and commenting the code better.
No functionality change.
llvm-svn: 53526
2008-07-13 21:12:01 +00:00
f76aa23b54
Enhance analysis of srem.
...
Remove dead code analyzing urem. 'urem' of power-of-2 is canonicalized to an
'and' instruction.
llvm-svn: 53506
2008-07-12 05:04:38 +00:00
3707f1daba
Use find instead of lower_bound.
...
llvm-svn: 53474
2008-07-11 20:58:19 +00:00
8e462e9a82
Don't call lookupNumber more than we have to.
...
llvm-svn: 53470
2008-07-11 20:05:13 +00:00
45e127ab20
Document 'mask' in this calculation.
...
llvm-svn: 53454
2008-07-11 08:16:26 +00:00
da405e1155
Remove misleading constant from comment.
...
llvm-svn: 53452
2008-07-11 07:36:19 +00:00
f95b64acaa
Add another optimization from PR2330. Also catch some missing cases that are
...
similar.
llvm-svn: 53451
2008-07-11 07:20:53 +00:00
3994bed1a9
a missed optimization that Eli spotted
...
llvm-svn: 53449
2008-07-11 06:40:29 +00:00
13a6911ea2
another bug in the same line.
...
llvm-svn: 53448
2008-07-11 06:38:16 +00:00
de89b507dd
fix a bug spotted by Eli's eagle eyes
...
llvm-svn: 53447
2008-07-11 06:36:01 +00:00
bd25b8507c
simplify and merge a bunch of code. Instead of comparing against
...
the min/max values for an integer type, compare against the min/max
values we can prove contain the input. This might be a tighter bound,
so this is general goodness.
llvm-svn: 53446
2008-07-11 05:40:05 +00:00
38a50c9528
fold away (x <= cst) earlier, allowing us to not have to
...
handle them in some code.
llvm-svn: 53445
2008-07-11 05:08:55 +00:00
6af608b8ce
Fix folding of icmp's of i1 where the comparison is signed. The code
...
was using the algorithm for folding unsigned comparisons which is
completely wrong. This has been broken since the signless types change.
llvm-svn: 53444
2008-07-11 04:20:58 +00:00
4fa8bb3430
Fix a bogus optimization: folding (slt (zext i1 A to i32), 1) -> (slt i1 A, true)
...
This cause a regression in InstCombine/JavaCompare, which was doing the right
thing on accident. To handle the missed case, generalize the comparisons based
on masked bits a little bit to handle comparisons against the max value. For
example, we can now xform (slt i32 (and X, 4), 4) -> (setne i32 (and X, 4), 4)
llvm-svn: 53443
2008-07-11 04:09:09 +00:00
e0f3ab82c4
Restructure dead argument elimination, try #3 :-)
...
Rewrite the DeadArgumentElimination pass, to use a more explicit tracking of
dependencies between return values and/or arguments. Also make the handling of
arguments and return values the same.
The pass now looks properly inside returned structs, but only at the first
level (ie, not inside nested structs).
This version fixed a few more bugs and was cleaned up a bit. It now passes all
of LLVM's testing, and should still pass SPEC2006. There is still a minor bug
with regard to returning nested structs. Since there is currently nothing that
emits such IR, I will fix that in a seperate commit (partly because it requires
a non-trivial fix).
llvm-svn: 53400
2008-07-10 10:24:08 +00:00
6193a564ab
Fix overzealous optimization. Thanks to Duncan Sands for pointing out my error!
...
llvm-svn: 53393
2008-07-10 05:51:40 +00:00
bb89c2a3f6
Simplify, suggested by Chris Lattner.
...
llvm-svn: 53283
2008-07-09 07:35:26 +00:00
f9c27c343a
Fold (a < 8) && (b < 8) into (a|b) < 8 for unsigned less or greater than.
...
llvm-svn: 53282
2008-07-09 07:29:11 +00:00
364661c43e
Fold ((1 << a) & 1) to (a == 0).
...
llvm-svn: 53276
2008-07-09 05:20:13 +00:00
0d3645e673
Reduce x - y to -y when we know the 'x' part will get masked off anyways.
...
llvm-svn: 53271
2008-07-09 04:32:37 +00:00
51cbf928ab
If loop induction variable's start value is less then its exit value then do not split the loop.
...
llvm-svn: 53265
2008-07-09 00:12:01 +00:00
501d78fdc0
Fix PR2496, a really nasty bug which involved sinking volatile loads
...
into phis. This is actually the same bug as PR2262 /
2008-04-29-VolatileLoadDontMerge.ll, but I missed checking the first
predecessor for multiple successors. Testcase here:
InstCombine/2008-07-08-VolatileLoadMerge.ll
llvm-svn: 53240
2008-07-08 17:18:32 +00:00
03001cb820
Fix two serious LSR bugs.
...
1. LSR runOnLoop is always returning false regardless if any transformation is made.
2. AddUsersIfInteresting can create new instructions that are added to DeadInsts. But there is a later early exit which prevents them from being freed.
llvm-svn: 53193
2008-07-07 19:51:32 +00:00
38740a98b2
Make DenseMap's insert return a pair, to more closely resemble std::map.
...
llvm-svn: 53177
2008-07-07 17:46:23 +00:00
9f1a4dc672
Fix missed optimization opportunity when analyzing cast of mul and select.
...
llvm-svn: 53151
2008-07-05 21:19:34 +00:00
3ea90a7d55
Use information already present in the ValueTable to fast-fail when we know there won't be a value number match. This speeds up GVN on a case where there are very few redundancies by ~25%.
...
llvm-svn: 53108
2008-07-03 17:44:33 +00:00
eb611ddeb2
Do not try to update dominator info while manipulating CFG. This code does not handle all cases and keeps invalid dom info around some cases, which misleads other passes down stream.
...
Right now, dom info is recaluclated in the end if the loop is switched.
llvm-svn: 53106
2008-07-03 17:37:52 +00:00
d57cdc3c60
Remove the ability for ADCE to remove unreachable blocks in loop nests, because, as Eli pointed out, SimplifyCFG already does this.
...
llvm-svn: 53104
2008-07-03 17:21:41 +00:00
a96eabaab7
Remove unused function.
...
llvm-svn: 53090
2008-07-03 07:10:03 +00:00
f94b9826b5
Preserve dom info.
...
llvm-svn: 53089
2008-07-03 07:04:22 +00:00
226edd1826
Remove extra FIXME
...
llvm-svn: 53087
2008-07-03 06:50:04 +00:00
c4dcf82a16
Reconstruct dom info, if loop is unswitched.
...
llvm-svn: 53086
2008-07-03 06:48:21 +00:00
e491bb8845
LoopUnswitch does not preserve dominator info in all cases.
...
llvm-svn: 53085
2008-07-03 05:55:03 +00:00
7dcfff392a
Undo previous patch. It is not that simple to fix dom info here.
...
llvm-svn: 53062
2008-07-03 00:08:13 +00:00
5adfcb5783
Preserve dom info while simplifing loop after the unswitch.
...
llvm-svn: 53052
2008-07-02 22:58:54 +00:00
488b89f608
Use df_ext_iterator to capture the reachable set without allocating an extra set.
...
Also, move large sets and vectors out of instance variables and onto the stack,
and give them more reasonable sizes.
llvm-svn: 53044
2008-07-02 18:41:09 +00:00
6acc782dad
Avoid a redundant call.
...
llvm-svn: 53040
2008-07-02 18:15:31 +00:00
323b5755a6
Add support to ADCE for pruning unreachable blocks. This addresses the final
...
part of PR2509.
llvm-svn: 53038
2008-07-02 18:05:19 +00:00
9edcf24da9
Use DenseSet rather than SmallPtrSet for the alive set. Using SmallPtrSet
...
with a huge "size" parameter is actually quite inefficient.
llvm-svn: 53034
2008-07-02 17:32:04 +00:00
b22a640fe4
A better fix for PR2503 that doesn't pessimize GVN in the presence of unreachable blocks.
...
llvm-svn: 53032
2008-07-02 17:20:16 +00:00
ed50fb5b61
reuse vectors.
...
llvm-svn: 53007
2008-07-02 01:44:29 +00:00
57d94d6304
Fix comment.
...
llvm-svn: 53006
2008-07-02 01:31:19 +00:00
e149d4ed4d
Preserve loop data so that it is not fetched everytime it is needed.
...
Keep track of currentLoop.
llvm-svn: 53005
2008-07-02 01:18:13 +00:00
da3db11db3
- Re-apply 52748 and friends with fix. GetConstantStringInfo() returns an empty string for ConstantAggregateZero case which surprises selectiondag.
...
- Correctly handle memcpy from constant string which is zero-initialized.
llvm-svn: 52891
2008-06-30 07:31:25 +00:00
a7c583d584
Revert (52748 and friends):
...
Move GetConstantStringInfo to lib/Analysis. Remove
string output routine from Constant. Update all
callers. Change debug intrinsic api slightly to
accomodate move of routine, these now return values
instead of strings.
This unbreaks llvm-gcc bootstrap.
llvm-svn: 52884
2008-06-29 17:57:03 +00:00
3f1c75c4d8
Remove unused function.
...
llvm-svn: 52749
2008-06-26 01:19:35 +00:00
d0ab9c47e6
Move GetConstantStringInfo to lib/Analysis. Remove
...
string output routine from Constant. Update all
callers. Change debug intrinsic api slightly to
accomodate move of routine, these now return values
instead of strings.
llvm-svn: 52748
2008-06-26 00:31:12 +00:00
88ca48b09d
Restore DeadArgElim back to 52570. It's breaking 447.dealII.
...
llvm-svn: 52736
2008-06-25 18:10:09 +00:00
1b03c2ac98
Pacify gcc-4.3.
...
llvm-svn: 52723
2008-06-25 16:31:18 +00:00
2e2001d8b9
Fix a (false) warning on darwin.
...
llvm-svn: 52705
2008-06-25 08:12:16 +00:00
4e1cf1e7d7
Fix some cosmetics in comments.
...
llvm-svn: 52704
2008-06-25 08:10:21 +00:00
5fd28b54c7
- Use O(1) check of basic block size limit.
...
- Avoid speculatively execute vector ops.
llvm-svn: 52703
2008-06-25 07:50:12 +00:00
c9c81fb0df
Fix PR2488, a case where we deleted stack restores too aggressively.
...
llvm-svn: 52702
2008-06-25 05:59:28 +00:00
04c8bd7e11
Revert 52645, the loop unroller changes. It caused a regression in 252.eon.
...
llvm-svn: 52688
2008-06-24 20:44:42 +00:00
4be44e62b3
Fix a typo in a comment.
...
llvm-svn: 52687
2008-06-24 18:00:21 +00:00
c702e1d32f
Commit the new DeadArgElim pass again, this time with the gcc bootstrap failures fixed.
...
Also add a testcase to reproduce the gcc bootstrap failure in very much reduced form.
llvm-svn: 52677
2008-06-24 16:30:26 +00:00
19a6469e1b
Rename a few variables to be more consistent.
...
llvm-svn: 52672
2008-06-24 09:14:10 +00:00
abd8f41c81
Use use_empty() instead of getNumUses(), avoiding a use list traversal.
...
llvm-svn: 52651
2008-06-23 23:23:49 +00:00
ac563833ae
Fix spelling and grammar in a comment.
...
llvm-svn: 52648
2008-06-23 22:11:52 +00:00
48c5c7e860
Revamp the loop unroller, extending it to correctly update PHI nodes
...
in the presence of out-of-loop users of in-loop values and the trip
count is not a known multiple of the unroll count, and to be a bit
simpler overall. This fixes PR2253.
llvm-svn: 52645
2008-06-23 21:29:41 +00:00
403e567043
Disable PRE. It's breaking bootstrapping.
...
llvm-svn: 52643
2008-06-23 21:22:35 +00:00
54e02194a1
Tighten the conditions under which we do PRE, remove some unneeded code, and correct our preserved analyses list, since we
...
do now change the CFG by splitting critical edges during PRE.
llvm-svn: 52631
2008-06-23 17:49:45 +00:00
4d754bc97b
minor tidying of comments.
...
llvm-svn: 52630
2008-06-23 17:11:23 +00:00
00fdbd01e5
At Chris' suggestion, move the liveness and worklist datastructures into
...
instance variables so they can be allocated just once, and reuse the worklist
as the dead list as well.
llvm-svn: 52618
2008-06-23 06:13:12 +00:00
5ca5e02480
Improve LSR's dead-phi detection to handle use-def cycles
...
with more than two nodes.
llvm-svn: 52617
2008-06-22 20:44:02 +00:00
90071075e2
Use Loop::block_iterator.
...
llvm-svn: 52616
2008-06-22 20:18:58 +00:00
6ff85681e4
Fix PR2369 by making scalarrepl more careful about promoting
...
structures. Its default threshold is to promote things that are
smaller than 128 bytes, which is sane. However, it is not sane
to do this for things that turn into 128 *registers*. Add a cap
on the number of registers introduced, defaulting to 128/4=32.
llvm-svn: 52611
2008-06-22 17:46:21 +00:00
d3449df326
Fix for PR2479: correctly optimize expressions like (a > 13) & (a ==
...
15).
See also PR1800, which is about the signed case.
llvm-svn: 52608
2008-06-21 23:36:13 +00:00
158ff2c4a9
Use Instruction::eraseFromParent().
...
llvm-svn: 52606
2008-06-21 22:08:46 +00:00
8459e0bc59
Fix warning when assertions disabled.
...
llvm-svn: 52590
2008-06-21 19:49:01 +00:00
42bbca11cc
Enable PRE.
...
llvm-svn: 52574
2008-06-21 07:26:53 +00:00
33067210d1
Back out Matthijs' DAE patches. It's miscompiling gcc driver.
...
llvm-svn: 52570
2008-06-21 00:31:44 +00:00
3ada1e118b
Clean up a use of std::distance.
...
llvm-svn: 52544
2008-06-20 17:11:32 +00:00
a5dd67f002
Tidy up some commments and use the getAggregateOperand and
...
getInsertedValueOperand accessors. Thanks Matthijs!
llvm-svn: 52543
2008-06-20 16:41:17 +00:00
b5210efb31
Fix the conditions under which SCCP should examine insertvalue
...
instructions. Thanks to Matthijs Kooijman for pointing this out!
llvm-svn: 52542
2008-06-20 16:39:44 +00:00
c456f9dfc6
80 column and trailing whitespace fixes.
...
llvm-svn: 52539
2008-06-20 15:34:07 +00:00
0c50b953c5
Don't let DeadArgumentElimination attempt to update callers when the return
...
type wasn't changed.
llvm-svn: 52538
2008-06-20 15:25:43 +00:00
9dc59b7666
Don't let DeadArgElimination change the return type ({} into void and {T}
...
into T) when no return values are actually dead.
llvm-svn: 52537
2008-06-20 15:16:45 +00:00
013b6a9a42
Explicitely track if any arguments or return values were removed in
...
DeadArgumentElimination and assert that the function type does not change if
nothing was changed. This should catch subtle changes in function type that are
not intended.
llvm-svn: 52536
2008-06-20 14:28:52 +00:00
e91aed6ce1
Remove debug output.
...
llvm-svn: 52535
2008-06-20 14:03:35 +00:00
8d32dee428
Recommit r52459, rewriting of the dead argument elimination pass.
...
This is a fixed version that no longer uses multimap::equal_range, which
resulted in a pointer invalidation problem.
Also, DAE::InspectedFunctions was not really necessary, so it got removed.
Lastly, this version no longer applies the extra arg hack on functions who did
not have any arguments to start with.
llvm-svn: 52532
2008-06-20 09:36:16 +00:00
78fbcafb53
Really disable PRE.
...
llvm-svn: 52531
2008-06-20 08:59:13 +00:00
f3ecd2d290
Fix PR2471, which is a bug involving an invalid promotion from a conditional load.
...
llvm-svn: 52525
2008-06-20 05:12:56 +00:00
1b3ea963f7
Change around the data structures used to store availability sets, resulting in a GVN+PRE that is faster that GVN alone was before.
...
llvm-svn: 52521
2008-06-20 01:15:47 +00:00
041f9d03ff
Teach SCCP about insertvalue and extractvalue, and about
...
propagating constants across aggregate return values when
insertvalue and extractvalue are used.
llvm-svn: 52520
2008-06-20 01:15:44 +00:00
3b18fd7b02
Teach InlineFunction how to differentiate between multiple-value
...
return statements and aggregate returns so that it handles both
correctly.
llvm-svn: 52519
2008-06-20 01:03:44 +00:00
9598f930f3
Disable PRE for now. It seems to be breaking llvm-gcc bootstrapping.
...
llvm-svn: 52518
2008-06-20 01:01:07 +00:00
e780d66657
Add a hidden -disable-pre flag for testing purposes. This should be removed
...
once benchmarking is completed.
llvm-svn: 52506
2008-06-19 19:57:25 +00:00
fdf9f168b5
PRE requires that critical edges be split.
...
llvm-svn: 52505
2008-06-19 19:54:19 +00:00
cd6fb1d0a8
Remove dead code causing a warning.
...
llvm-svn: 52502
2008-06-19 18:00:44 +00:00
d6530872f3
Use the common API for adding instructions to basic blocks instead of
...
using BasicBlock::getInstList.
llvm-svn: 52500
2008-06-19 17:53:32 +00:00
ff21db851d
Be sure to remove values from the value numbering table after we delete them.
...
This fixes a failure on povray.
llvm-svn: 52499
2008-06-19 17:53:26 +00:00
ed2250990a
Use Instruction::moveBefore instead of manipulating the instruction list
...
directly.
llvm-svn: 52498
2008-06-19 17:47:47 +00:00
9eea470fcf
Avoid using BasicBlock::getInstList directly in a few places.
...
llvm-svn: 52497
2008-06-19 17:37:25 +00:00
45d3701fce
Revert support for insertvalue and extractvalue instructions for the moment.
...
GVN expects that all inputs which to an instruction fall somewhere in the value
hierarchy, which isn't true for these.
llvm-svn: 52496
2008-06-19 17:25:39 +00:00
68f539e807
Delete dead code.
...
llvm-svn: 52494
2008-06-19 17:18:39 +00:00
0c71732497
Use a CallSite to find the nth argument of a call/invoke instruction instead of
...
using getOperand() directly. This makes things work with invoke instructions as
well.
llvm-svn: 52489
2008-06-19 08:53:24 +00:00
3ea800fbad
Add support for extractvalue and insertvalue instructions in GVN.
...
llvm-svn: 52472
2008-06-18 21:59:00 +00:00
6a903bc601
Add local PRE to GVN. This only operates in cases where it would not increase code size, namely when the instantiated expression
...
would only need to be created in one predecessor.
llvm-svn: 52471
2008-06-18 21:41:49 +00:00
78119b4742
Fix the regressions on sext-misc.ll my patch yesterday caused.
...
llvm-svn: 52466
2008-06-18 18:11:55 +00:00
9094cc957e
Revert r52459, which was causing an infinite loop or massive slowdown on MultiSource/Applications/SPASS, and possibly others as well.
...
Please reapply once this is fixed.
llvm-svn: 52465
2008-06-18 17:32:16 +00:00
be928e3b21
Move LSR's private isZero function to a public SCEV member
...
function, and make use of it in several places.
llvm-svn: 52463
2008-06-18 16:23:07 +00:00
964557fdf5
Rewrite the DeadArgumentElimination pass, to use a more explicit tracking of
...
dependencies between return values and/or arguments. Also make the handling of
arguments and return values the same.
The pass now looks properly inside returned structs, but only at the first
level (ie, not inside nested structs).
Also add a testcase for testing various variations of (multiple) dead rerturn
values.
llvm-svn: 52459
2008-06-18 11:12:53 +00:00
fd17357643
Reapply r52397 (make IPConstProp promote returned arguments), but fixed this
...
time. Sorry for the trouble!
This time, also add a testcase, which I should have done in the first place...
llvm-svn: 52455
2008-06-18 08:30:37 +00:00
97034598b1
Reapply r52396, it was unrelated to the breakage (that was caused by r52397, my
...
commit after this).
llvm-svn: 52453
2008-06-18 08:09:27 +00:00
ef36dcd10b
implement some simple bswap optimizations, rdar://5992453
...
llvm-svn: 52442
2008-06-18 04:33:20 +00:00
b5ee8b3e89
make truncate/sext elimination capable of changing phi's. This
...
implements rdar://6013816 and the testcase in Transforms/InstCombine/sext-misc.ll.
llvm-svn: 52440
2008-06-18 04:00:49 +00:00
cd6b697945
Preserve dominance frontier while trivially unswitching loop.
...
llvm-svn: 52438
2008-06-18 02:16:38 +00:00
75f3732b23
We don't want to find dependencies within the same block in this case. It leads to incorrect results because
...
we're detecting something at or after the call we're querying on.
llvm-svn: 52433
2008-06-17 22:27:06 +00:00
aecc3750d1
revert recent patch which is causing widespread breakage.
...
llvm-svn: 52415
2008-06-17 17:06:43 +00:00
4b50fde2c4
Fix typo that changed the logic to something wrong.
...
Spotted by Nick Lewycky.
llvm-svn: 52411
2008-06-17 15:55:30 +00:00
332836d68d
Learn IPConstProp to propagate arguments that are directly returned. Strictly
...
speaking these are not constant values. However, when a function always returns
one of its arguments, then from the point of view of each caller the return
value is constant (or at least a known value) and can be replaced.
llvm-svn: 52397
2008-06-17 12:20:24 +00:00
f03c1ae407
Learn IPConstProp to look at individual return values and propagate them
...
individually.
Also learn IPConstProp how returning first class aggregates work, in addition
to old style multiple return instructions.
Modify the return-constants testscase to confirm this behaviour.
llvm-svn: 52396
2008-06-17 12:02:52 +00:00
ab0dccba6b
Refine the change in r52258 for avoiding use-before-def conditions
...
when changing the stride of a comparison so that it's slightly
more precise, by having it scan the instruction list to determine
if there is a use of the condition after the point where the
condition will be inserted.
llvm-svn: 52371
2008-06-16 22:34:15 +00:00
319e9a4f63
Switch over to SetVector to ensure same order of iterations do not vary across runs.
...
llvm-svn: 52361
2008-06-16 21:08:17 +00:00
a72cdcd1a2
Iterating over SmallPtrSet is not deterministic.
...
llvm-svn: 52339
2008-06-16 18:17:09 +00:00
86cda9e050
Pass around Instruction* instead of Instruction& in FindInsertedValue and friends.
...
llvm-svn: 52318
2008-06-16 13:13:08 +00:00
5cb387735d
80 column fixes.
...
llvm-svn: 52316
2008-06-16 12:57:37 +00:00
e92e18be5a
Move FindScalarValue from InstructionCombining.cpp to ValueTracking.cpp. While
...
I'm at it, rename it to FindInsertedValue.
The only functional change is that newly created instructions are no longer
added to instcombine's worklist, but that is not really necessary anyway (and
I'll commit some improvements next that will completely remove the need).
llvm-svn: 52315
2008-06-16 12:48:21 +00:00
1c9922703f
Fix the crash on SimplifyLibCalls/2005-05-20-sprintf-crash.ll
...
llvm-svn: 52295
2008-06-16 04:10:21 +00:00
a88cd4ea2a
Fix a case where tailcallelim wouldn't set the changed bit when it made a change.
...
llvm-svn: 52267
2008-06-14 00:49:48 +00:00
5de0a77a9b
Don't skip over instructions other than loads that might read memory
...
when trying to sink stores.
llvm-svn: 52259
2008-06-13 22:02:12 +00:00
9ad8c54aab
Protect ChangeCompareStride from situations in which it is possible
...
for it to generate use-before-def IR, such as in this testcase.
llvm-svn: 52258
2008-06-13 21:43:41 +00:00
9833a1b407
Make sure SimplifyStoreAtEndOfBlock doesn't mess with loops; the
...
structure checks are incorrect if the blocks aren't distinct.
Fixes PR2435.
llvm-svn: 52257
2008-06-13 21:17:49 +00:00
25a7f5de92
Use recently added getTruncateOrZeroExtend method to make the code shorter.
...
llvm-svn: 52251
2008-06-13 17:02:03 +00:00
431e9560b7
fix a minor deviation from the original in my previous commit
...
llvm-svn: 52247
2008-06-12 21:51:29 +00:00
f6d8e77027
op_iterator-ify some loops, low hanging fruit only, there is more
...
llvm-svn: 52246
2008-06-12 21:37:33 +00:00
89553cc42e
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.
...
llvm-svn: 52244
2008-06-12 21:15:59 +00:00
70fe16353a
Revert 52223.
...
llvm-svn: 52243
2008-06-12 20:55:39 +00:00
accdca1b03
Switch GVN to use ScopedHashTable.
...
llvm-svn: 52242
2008-06-12 19:25:32 +00:00
0babc61631
op_iterator-ify some loops, fix 80col violations
...
llvm-svn: 52226
2008-06-11 21:38:51 +00:00
933c743042
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.
llvm-svn: 52224
2008-06-11 19:18:20 +00:00
f3c2902ead
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.
llvm-svn: 52223
2008-06-11 19:07:54 +00:00
b2fc72bfbf
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.
llvm-svn: 52217
2008-06-11 14:05:05 +00:00
3453c7bcb5
Clarify a comment.
...
llvm-svn: 52212
2008-06-11 09:00:12 +00:00
945f2f7fed
op_iterator-ify loops
...
llvm-svn: 52191
2008-06-10 22:03:26 +00:00
9c9f531a47
lower calls to abs to inline code, PR2337
...
llvm-svn: 52138
2008-06-09 08:26:51 +00:00
dbd595f22d
Fix PR2411, where ip constant prop would propagate the
...
result of a weak function.
llvm-svn: 52137
2008-06-09 07:58:07 +00:00
11dd424539
Remove comparison methods for MVT. The main cause
...
of apint codegen failure is the DAG combiner doing
the wrong thing because it was comparing MVT's using
< rather than comparing the number of bits. Removing
the < method makes this mistake impossible to commit.
Instead, add helper methods for comparing bits and use
them.
llvm-svn: 52098
2008-06-08 20:54:56 +00:00
b4866ef30c
Limit the icmp+phi merging optimization to the cases where it is profitable:
...
don't make i1 phis when it won't be possible to eliminate them.
llvm-svn: 52097
2008-06-08 20:52:11 +00:00
89200c9177
Speculatively execute a block when the the block is the then part of a triangle shape and it contains a single, side effect free, cheap instruction. The branch is eliminated by adding a select instruction. i.e.
...
Turn
BB:
%t1 = icmp
br i1 %t1, label %BB1, label %BB2
BB1:
%t3 = add %t2, c
br label BB2
BB2:
=>
BB:
%t1 = icmp
%t4 = add %t2, c
%t3 = select i1 %t1, %t2, %t3
llvm-svn: 52073
2008-06-07 08:52:29 +00:00
8549e4ca07
LoopSimplify preserves AA.
...
llvm-svn: 52053
2008-06-06 17:50:58 +00:00
13237ac3b9
Wrap MVT::ValueType in a struct to get type safety
...
and better control the abstraction. Rename the type
to MVT. To update out-of-tree patches, the main
thing to do is to rename MVT::ValueType to MVT, and
rewrite expressions like MVT::getSizeInBits(VT) in
the form VT.getSizeInBits(). Use VT.getSimpleVT()
to extract a MVT::SimpleValueType for use in switch
statements (you will get an assert failure if VT is
an extended value type - these shouldn't exist after
type legalization).
This results in a small speedup of codegen and no
new testsuite failures (x86-64 linux).
llvm-svn: 52044
2008-06-06 12:08:01 +00:00
1152ca9101
As Chris suggested, handle the situation if ShAmt larger than BitWidth,
...
otherwise, opt might crash.
llvm-svn: 52041
2008-06-06 08:32:05 +00:00
fbe1dc240c
If BitWidth equals to ShtAmt, the RHSKnownZero[BitWidth-ShiftAmt-1] will
...
crash the opt. Just fix this.
Test case in llvm/test/Transforms/InstCombine/2008-06-05-ashr-crash.ll
llvm-svn: 52003
2008-06-05 14:23:44 +00:00
812989b147
Learn ScalarReplAggregrates how stores and loads of first class aggregrates
...
work and how to replace them into individual values. Also, when trying to
replace an aggregrate that is used by load or store with a single (large)
integer, don't crash (but don't replace the aggregrate either).
Also adds a testcase for both structs and arrays.
llvm-svn: 51997
2008-06-05 12:51:53 +00:00
e0c5adc158
Let StructRetPromotion check if all if its users are really calls or invokesn,
...
not other instructions. This fixes a crash with the added testcase.
llvm-svn: 51992
2008-06-05 08:57:20 +00:00
463f86639d
Let StructRetPromotion check if it's users are really calling it and not
...
passing its pointer. Fixes test with added testcase.
llvm-svn: 51991
2008-06-05 08:48:32 +00:00
230d6fbfeb
Use use_iterator::getOperandNo instead of CallSite::hasArgument to check if a
...
function is passed as an argument instead of called. Also do this check a bit
earlier.
llvm-svn: 51990
2008-06-05 08:34:25 +00:00
5afc2740b7
Update comments and documentation to reflect that GCSE and ValueNumbering are
...
deprecated by the GVN and GVNPRE passes.
llvm-svn: 51983
2008-06-05 07:55:49 +00:00
61c7f2a633
Remove unneeded #include.
...
llvm-svn: 51955
2008-06-04 18:28:10 +00:00
2353f35989
Replace two manual loops with calls to CallSite::hasArguments (no functional changes).
...
llvm-svn: 51947
2008-06-04 16:57:50 +00:00
fc3c489b52
Change packed struct layout so that field sizes
...
are the same as in unpacked structs, only field
positions differ. This only matters for structs
containing x86 long double or an apint; it may
cause backwards compatibility problems if someone
has bitcode containing a packed struct with a
field of one of those types.
The issue is that only 10 bytes are needed to
hold an x86 long double: the store size is 10
bytes, but the ABI size is 12 or 16 bytes (linux/
darwin) which comes from rounding the store size
up by the alignment. Because it seemed silly not
to pack an x86 long double into 10 bytes in a
packed struct, this is what was done. I now
think this was a mistake. Reserving the ABI size
for an x86 long double field even in a packed
struct makes things more uniform: the ABI size is
now always used when reserving space for a type.
This means that developers are less likely to
make mistakes. It also makes life easier for the
CBE which otherwise could not represent all LLVM
packed structs (PR2402).
Front-end people might need to adjust the way
they create LLVM structs - see following change
to llvm-gcc.
llvm-svn: 51928
2008-06-04 08:21:45 +00:00
2df82e7cec
LoopIndexSplit can sometimes result in cases where a block in its own domfrontier.
...
Don't crash when we encounter one of these.
llvm-svn: 51915
2008-06-03 18:29:48 +00:00
2ad7e7341c
Fix whitespace in whitespace-significant pseudocode in a comment.
...
llvm-svn: 51890
2008-06-03 00:57:21 +00:00
7314d0ee3c
Update dom tree. Fix PR 2372.
...
llvm-svn: 51887
2008-06-02 22:52:56 +00:00
a12a6de683
move CannotBeNegativeZero to ValueTracking. Simplify some signbit comparisons.
...
llvm-svn: 51864
2008-06-02 01:29:46 +00:00
965c769b3c
move ComputeMaskedBits, MaskedValueIsZero, and ComputeNumSignBits
...
out of instcombine into a new file in libanalysis. This also teaches
ComputeNumSignBits about the number of sign bits in a constantint.
llvm-svn: 51863
2008-06-02 01:18:21 +00:00
38099c1b6e
Fix two issues that Eli Friedman pointed out, where would misoptimized code like:
...
char a[200];
init(a, a+200);
OR
int a[200];
char* b = (char*)a;
char* c = (char*)a;
foo(b, c);
llvm-svn: 51850
2008-06-01 22:26:26 +00:00
d071a8708e
Don't remove the memcpy when call slot substitution fails.
...
llvm-svn: 51848
2008-06-01 21:52:16 +00:00
0397cd2ec4
When simplifying a call to a bitcast function, tighten up
...
the conditions for performing the transform when only the
function declaration is available: no longer allow turning
i32 into i64 for example. Only allow changing between
pointer types, and between pointer types and integers of
the same size. For return values ptr -> intptr was already
allowed; I added ptr -> ptr and intptr -> ptr while there.
As shown by a recent objc testcase, changing the way
parameters/return values are passed can be fatal when calling
code written in assembler that directly manipulates call
arguments and return values unless the transform has no
impact on the way they are passed at the codegen level.
While it is possible to imagine an ABI that treats integers
of pointer size differently to pointers, I don't think LLVM
supports any so the transform should now be safe while still
being useful.
llvm-svn: 51834
2008-06-01 07:38:42 +00:00
035fe6f716
Peer through sext/zext when looking for not(cmp).
...
llvm-svn: 51819
2008-05-31 19:01:33 +00:00
26b8cd84b3
Add more i1 optimizations. add, sub, mul, s/udiv on i1 are now simplified away.
...
llvm-svn: 51817
2008-05-31 17:59:52 +00:00
df9242a833
Adding i1 is always Xor.
...
llvm-svn: 51816
2008-05-31 17:10:28 +00:00
5df4326d78
rewrite operand loops to use iterators
...
llvm-svn: 51789
2008-05-30 21:24:22 +00:00
1f59d9937f
Since LCSSA switched over to DenseMap, we have to be more careful to avoid iterator invalidation. Fixes PR2385.
...
llvm-svn: 51777
2008-05-30 17:31:01 +00:00
57da7d2308
Use eraseFromParent() instead of doing that manually in two places.
...
llvm-svn: 51770
2008-05-30 12:35:46 +00:00
86ff8536f9
const-ify getOpcode.
...
llvm-svn: 51698
2008-05-29 19:53:46 +00:00
9e064a2180
Add a newline at the end of this file.
...
llvm-svn: 51680
2008-05-29 14:38:23 +00:00
7686b555e2
Replace the old ADCE implementation with a new one that more simply solves
...
the one case that ADCE catches that normal DCE doesn't: non-induction variable
loop computations.
This implementation handles this problem without using postdominators.
llvm-svn: 51668
2008-05-29 08:45:13 +00:00
f4aece5976
Remove debugging code.
...
llvm-svn: 51666
2008-05-29 08:15:48 +00:00
3a9fba5a72
convert more operand loops to iterator formulation
...
llvm-svn: 51663
2008-05-29 01:59:18 +00:00
ecdefb5df7
Implement PR2370: memmove(x,x,size) -> noop.
...
llvm-svn: 51636
2008-05-28 05:30:41 +00:00
698348dfac
Fix some constructs that gcc-4.4 warns about.
...
llvm-svn: 51591
2008-05-27 11:50:51 +00:00
3ebe82b57a
InequalityGraph::node() can create new nodes, invalidating iterators across
...
the set of nodes. Fix makeEqual to handle this by creating the new node first
then iterating across them second.
llvm-svn: 51573
2008-05-27 00:59:05 +00:00
6be65d2a84
Grammaro.
...
llvm-svn: 51572
2008-05-26 22:49:36 +00:00
dd7daee850
Factor code to copy global value attributes like
...
the section or the visibility from one global
value to another: copyAttributesFrom. This is
particularly useful for duplicating functions:
previously this was done by explicitly copying
each attribute in turn at each place where a
new function was created out of an old one, with
the result that obscure attributes were regularly
forgotten (like the collector or the section).
Hopefully now everything is uniform and nothing
is forgotten.
llvm-svn: 51567
2008-05-26 19:58:59 +00:00
d3f21d165f
Use a DenseMap instead of an std::map, speeding up the testcase in PR2368 by about a third.
...
llvm-svn: 51565
2008-05-26 10:07:43 +00:00
f6ccd2580c
"ret (constexpr)" can't be folded into a Constant. Add a method to
...
Analysis/ConstantFolding to fold ConstantExpr's, then make instcombine use it
to try to use targetdata to fold constant expressions on void instructions.
Also extend the icmp(inttoptr, inttoptr) folding to handle the case where
int size != ptr size.
llvm-svn: 51559
2008-05-25 20:56:15 +00:00
87a099a057
Fix a serious brain-o. Obviously no-one reviewed my patch :(
...
This fixes PR2359
llvm-svn: 51536
2008-05-24 04:06:28 +00:00
5c207c83c6
Fix PR2358 by resolving calls with undef arguments to overdefined.
...
llvm-svn: 51535
2008-05-24 03:59:33 +00:00
02912418f1
Remove x86.sse2.loadh.pd and x86.sse2.loadl.pd. These will be lowered into load and shuffle instructions.
...
llvm-svn: 51521
2008-05-24 00:07:06 +00:00
f96e1371e8
Tidy up BasicBlock::getFirstNonPHI, and change a bunch of places to
...
use it instead of duplicating its functionality.
llvm-svn: 51499
2008-05-23 21:05:58 +00:00
f52b23c0eb
Replace some weird usage of UserOp1 introduced in r49492 by a plain if.
...
llvm-svn: 51482
2008-05-23 16:17:48 +00:00
aef2b8198b
Restucture a part of the SimplifyCFG pass and include a testcase.
...
The SimplifyCFG pass looks at basic blocks that contain only phi nodes,
followed by an unconditional branch. In a lot of cases, such a block (BB) can
be merged into their successor (Succ).
This merging is performed by TryToSimplifyUncondBranchFromEmptyBlock. It does
this by taking all phi nodes in the succesor block Succ and expanding them to
include the predecessors of BB. Furthermore, any phi nodes in BB are moved to
Succ and expanded to include the predecessors of Succ as well.
Before attempting this merge, CanPropagatePredecessorsForPHIs checks to see if
all phi nodes can be properly merged. All functional changes are made to
this function, only comments were updated in
TryToSimplifyUncondBranchFromEmptyBlock.
In the original code, CanPropagatePredecessorsForPHIs looks quite convoluted
and more like stack of checks added to handle different kinds of situations
than a comprehensive check. In particular the first check in the function did
some value checking for the case that BB and Succ have a common predecessor,
while the last check in the function simply rejected all cases where BB and
Succ have a common predecessor. The first check was still useful in the case
that BB did not contain any phi nodes at all, though, so it was not completely
useless.
Now, CanPropagatePredecessorsForPHIs is restructured to to look a lot more
similar to the code that actually performs the merge. Both functions now look
at the same phi nodes in about the same order. Any conflicts (phi nodes with
different values for the same source) that could arise from merging or moving
phi nodes are detected. If no conflicts are found, the merge can happen.
Apart from only restructuring the checks, two main changes in functionality
happened.
Firstly, the old code rejected blocks with common predecessors in most cases.
The new code performs some extra checks so common predecessors can be handled
in a lot of cases. Wherever common predecessors still pose problems, the
blocks are left untouched.
Secondly, the old code rejected the merge when values (phi nodes) from BB were
used in any other place than Succ. However, it does not seem that there is any
situation that would require this check. Even more, this can be proven.
Consider that BB is a block containing of a single phi node "%a" and a branch
to Succ. Now, since the definition of %a will dominate all of its uses, BB
will dominate all blocks that use %a. Furthermore, since the branch from BB to
Succ is unconditional, Succ will also dominate all uses of %a.
Now, assume that one predecessor of Succ is not dominated by BB (and thus not
dominated by Succ). Since at least one use of %a (but in reality all of them)
is reachable from Succ, you could end up at a use of %a without passing
through it's definition in BB (by coming from X through Succ). This is a
contradiction, meaning that our original assumption is wrong. Thus, all
predecessors of Succ must also be dominated by BB (and thus also by Succ).
This means that moving the phi node %a from BB to Succ does not pose any
problems when the two blocks are merged, and any use checks are not needed.
llvm-svn: 51478
2008-05-23 09:09:41 +00:00
f399bbf980
Indent fix.
...
llvm-svn: 51477
2008-05-23 07:57:02 +00:00
3bf5512d87
Constant integer vectors may also be negated.
...
llvm-svn: 51476
2008-05-23 04:54:45 +00:00
8f3127c5b5
Typo.
...
llvm-svn: 51475
2008-05-23 04:39:38 +00:00
4f3d878507
Revert X + X --> X * 2 optz'n which pessimizes heavily on x86.
...
llvm-svn: 51474
2008-05-23 04:34:58 +00:00
452fb32927
Implement X + X for vectors.
...
llvm-svn: 51472
2008-05-23 04:14:51 +00:00
2ec9a01173
Fix a recently added optimization to not crash on vectors.
...
llvm-svn: 51471
2008-05-23 03:26:47 +00:00
6d5f120c5c
Generalize the new code in instcombine's ComputeNumSignBits for handling
...
and/or to handle more cases (such as this add-sitofp.ll testcase), and
port it to selectiondag's ComputeNumSignBits.
llvm-svn: 51469
2008-05-23 02:28:01 +00:00
53b2698531
Use isSingleValueType instead of isFirstClassType to
...
exclude struct and array types.
llvm-svn: 51467
2008-05-23 01:52:21 +00:00
fecb88249f
Allow for switch with no cases. Was causing fault
...
in gcc.dg/pr27531-1.c.
llvm-svn: 51464
2008-05-23 01:01:31 +00:00
30ab45d01e
Use isSingleValueType instead of isFirstClassType to
...
exclude struct and array types.
llvm-svn: 51459
2008-05-23 00:17:26 +00:00
7a0566b9cd
Use isSingleValueType instead of isFirstClassType to
...
exclude struct and array types.
llvm-svn: 51456
2008-05-23 00:12:03 +00:00
c5ec1e19eb
rewrite the validity checking for memory promotion to be simpler,
...
more aggressive, and more correct. Verify that we only attempt to
promote loads and stores.
llvm-svn: 51406
2008-05-22 03:22:42 +00:00
f12c08dcd8
Use 'continue' to reduce nesting in this loop. No functionality change.
...
llvm-svn: 51399
2008-05-22 00:53:38 +00:00
e62632e0bb
When LSR is replacing an instruction, call
...
ScalarEvolution::deleteValueFromRecords on it before doing the
replaceAllUsesWith, because ScalarEvolution looks at the instruction's
users to find SCEV references to the instruction's SCEV object in its
internal maps.
Move all of LSR's loop-related state clearing after processing the loop
and before cleaning up dead PHI nodes. This eliminates all of LSR's SCEV
references just before the calls to ScalarEvolution::deleteValueFromRecords
so that when ScalarEvolution drops its own SCEV references, the reference
counts will reach zero and the SCEVs will be deleted immediately.
These changes fix some compiler aborts involving ScalarEvolution holding
onto and reusing SCEV objects for instructions that have been deleted.
No regression test unfortunately; because the symptoms were due to
dangling pointers, reduced testcases ended up being fairly arbitrary.
llvm-svn: 51359
2008-05-21 00:54:12 +00:00
81ab753b14
Port SelectionDAG's ComputeNumSignBits-using code to instcombine,
...
now that instcombine also has ComputeNumSignBits.
llvm-svn: 51350
2008-05-20 21:01:12 +00:00
5148a4ba66
Fix typo.
...
llvm-svn: 51303
2008-05-20 07:26:45 +00:00
7ac943fffd
Teach instcombine 4 new xforms:
...
(add (sext x), cst) --> (sext (add x, cst'))
(add (sext x), (sext y)) --> (sext (add int x, y))
(add double (sitofp x), fpcst) --> (sitofp (add int x, intcst))
(add double (sitofp x), (sitofp y)) --> (sitofp (add int x, y))
This generally reduces conversions. For example MiBench/telecomm-gsm
gets these simplifications:
HACK2: %tmp67.i142.i.i = sext i16 %tmp6.i141.i.i to i32 ; <i32> [#uses=1]
%tmp23.i139.i.i = sext i16 %tmp2.i138.i.i to i32 ; <i32> [#uses=1]
%tmp8.i143.i.i = add i32 %tmp67.i142.i.i, %tmp23.i139.i.i ; <i32> [#uses=3]
HACK2: %tmp67.i121.i.i = sext i16 %tmp6.i120.i.i to i32 ; <i32> [#uses=1]
%tmp23.i118.i.i = sext i16 %tmp2.i117.i.i to i32 ; <i32> [#uses=1]
%tmp8.i122.i.i = add i32 %tmp67.i121.i.i, %tmp23.i118.i.i ; <i32> [#uses=3]
HACK2: %tmp67.i.i190.i = sext i16 %tmp6.i.i189.i to i32 ; <i32> [#uses=1]
%tmp23.i.i187.i = sext i16 %tmp2.i.i186.i to i32 ; <i32> [#uses=1]
%tmp8.i.i191.i = add i32 %tmp67.i.i190.i, %tmp23.i.i187.i ; <i32> [#uses=3]
HACK2: %tmp67.i173.i.i.i = sext i16 %tmp6.i172.i.i.i to i32 ; <i32> [#uses=1]
%tmp23.i170.i.i.i = sext i16 %tmp2.i169.i.i.i to i32 ; <i32> [#uses=1]
%tmp8.i174.i.i.i = add i32 %tmp67.i173.i.i.i, %tmp23.i170.i.i.i ; <i32> [#uses=3]
HACK2: %tmp67.i152.i.i.i = sext i16 %tmp6.i151.i.i.i to i32 ; <i32> [#uses=1]
%tmp23.i149.i.i.i = sext i16 %tmp2.i148.i.i.i to i32 ; <i32> [#uses=1]
%tmp8.i153.i.i.i = add i32 %tmp67.i152.i.i.i, %tmp23.i149.i.i.i ; <i32> [#uses=3]
HACK2: %tmp67.i.i.i.i = sext i16 %tmp6.i.i.i.i to i32 ; <i32> [#uses=1]
%tmp23.i.i5.i.i = sext i16 %tmp2.i.i.i.i to i32 ; <i32> [#uses=1]
%tmp8.i.i7.i.i = add i32 %tmp67.i.i.i.i, %tmp23.i.i5.i.i ; <i32> [#uses=3]
This also fixes a bug in ComputeNumSignBits handling select and
makes it more aggressive with and/or.
llvm-svn: 51302
2008-05-20 05:46:13 +00:00
9c27f96d04
fix two issues Neil noticed, thanks!
...
llvm-svn: 51296
2008-05-20 03:50:52 +00:00
e5572706e8
Refine the fix in r51169 to only apply when the operand val being
...
replaced is a PHI. This prevents it from inserting uses before defs
in the case that it isn't a PHI and it depends on other instructions
later in the block. This fixes the 447.dealII regression on x86-64.
llvm-svn: 51292
2008-05-20 03:01:48 +00:00
d717761a2b
Make AssociativeOpt static.
...
llvm-svn: 51290
2008-05-20 01:14:05 +00:00
ee7bf41c06
Do not erase induction variable increment if it is used outside the loop.
...
llvm-svn: 51280
2008-05-19 22:23:55 +00:00
123438cc05
Add a ComputeNumSignBits function for use by instcombine, based on the
...
code in SelectionDAG.
llvm-svn: 51279
2008-05-19 22:14:15 +00:00
b42712288e
switch to Type::getFPMantissaWidth instead of reinventing it.
...
llvm-svn: 51275
2008-05-19 21:17:23 +00:00
ba9acbe6dc
minor cleanups, teach instcombine that sitofp/uitofp cannot
...
produce a negative zero.
llvm-svn: 51272
2008-05-19 20:27:56 +00:00
e35fe0f1c6
convert fptosi(sitofp x) -> x if the fp value has enough bits in its mantissa
...
to accurately represent the integer. This triggers 9 times in 471.omnetpp,
though 8 of those seem to be inlined from the same place.
llvm-svn: 51271
2008-05-19 20:25:04 +00:00
5920a78034
Fold FP comparisons where one operand is converted from an integer
...
type and the other operand is a constant into integer comparisons.
This happens surprisingly frequently (e.g. 10 times in 471.omnetpp),
which are things like this:
%tmp8283 = sitofp i32 %tmp82 to double
%tmp1013 = fcmp ult double %tmp8283, 0.0
Clearly comparing tmp82 against i32 0 is cheaper here.
this also triggers 8 times in gobmk, including this one:
%tmp375376 = sitofp i32 %tmp375 to double
%tmp377 = fcmp ogt double %tmp375376, 8.150000e+01
which is comparing an integer against 81.5 :).
llvm-svn: 51268
2008-05-19 20:18:56 +00:00
6e70830af9
remove debug output
...
llvm-svn: 51264
2008-05-19 20:03:53 +00:00
fc365b60dc
be more aggressive about transforming add -> or when the operands have no
...
intersecting bits. This triggers all over the place, for example in lencode,
with adds of stuff like:
%tmp580 = mul i32 %tmp579, 2
%tmp582 = and i32 %b8, 1
and
%tmp28 = shl i32 %abs.i, 1
%sign.0 = select i1 %tmp23, i32 1, i32 0
and
%tmp344 = shl i32 %tmp343, 2
%tmp346 = and i32 %tmp96, 3
etc.
llvm-svn: 51263
2008-05-19 20:01:56 +00:00
eec7a3c071
Fix PR2341 - when the length is 4 use an i32 not
...
an i16! Cleaned up trailing whitespace while there.
llvm-svn: 51240
2008-05-19 09:27:24 +00:00
65720c968c
Teach GVN to not assert on vector comparisons
...
llvm-svn: 51230
2008-05-18 19:49:05 +00:00
4b2a724fb8
Fix PR2339
...
llvm-svn: 51226
2008-05-18 04:11:26 +00:00
79376f4e02
Move isTrueWhenEqual to ICmpInst.
...
llvm-svn: 51215
2008-05-17 07:33:39 +00:00
5610dabac9
Less conservative verison of previous patch,
...
suggested by Duncan.
llvm-svn: 51211
2008-05-16 23:18:52 +00:00
e7f5bc2c3b
Weak functions not declared non-throwing might be
...
replaced at linktime with a body that throws, even
if the body in this file does not. Make PruneEH
be more conservative in this case.
g++.dg/eh/weak1.C
llvm-svn: 51207
2008-05-16 21:31:48 +00:00
e1f6e4b21d
API change for {BinaryOperator|CmpInst|CastInst}::create*() --> Create. Legacy interfaces will be in place for some time. (Merge from use-diet branch.)
...
llvm-svn: 51200
2008-05-16 19:29:10 +00:00
67933e6692
Bill pointed out that system headers should be
...
included after local headers.
llvm-svn: 51187
2008-05-16 09:30:00 +00:00
173a53f87c
Do not dup malloc, vector instructions, etc. Throttle the default theshold way down.
...
llvm-svn: 51183
2008-05-16 07:55:50 +00:00
c7d6eceb69
Remove ADCE's ability to delete loops. This ability is now implemented in a
...
safer manner by loop deletion.
llvm-svn: 51182
2008-05-16 04:34:51 +00:00
ad5f211b48
Clean ups for loop deletion based on Chris' feedback.
...
Also, use SCEV to determine the trip count of the loop, which is more powerful
and accurate that Loop::getTripCount.
llvm-svn: 51179
2008-05-16 04:32:45 +00:00
5c953b7d27
implement PR2328.
...
llvm-svn: 51176
2008-05-16 02:59:42 +00:00
0a0fa7cf78
Fix a bug in LoopStrengthReduce that caused it to emit IR with
...
use-before-def. The problem comes up in code with multiple PHIs where
one PHI is being rewritten in terms of the other, but the other needs
to be casted first. LLVM rules requre the cast instruction to be
inserted after any PHI instructions, but when instructions were
inserted to replace the second PHI value with a function of the first,
they were ended up going before the cast instruction. Avoid this
problem by remembering the location of the cast instruction, when one
is needed, and inserting the expansion of the new value after it.
This fixes a bug that surfaced in 255.vortex on x86-64 when
instcombine was removed from the middle of the loop optimization
passes.
llvm-svn: 51169
2008-05-15 23:26:57 +00:00
61724355af
Remove useless check.
...
Patch by Matthijs Kooijman.
llvm-svn: 51154
2008-05-15 18:04:29 +00:00
783cb2d76d
Use of UINT_MAX requires climits, at least when
...
compiling with gcc 4.3.
llvm-svn: 51145
2008-05-15 11:22:50 +00:00
697e94cc22
Fix a bunch of 80col violations that arose from the Create API change. Tweak makefile targets to find these better.
...
llvm-svn: 51143
2008-05-15 10:04:30 +00:00
3716952f10
Situations can arise when you have a function called that returns a 'void', but
...
is bitcast to return a floating point value. The result of the instruction may
not be used by the program afterwards, and LLVM will happily remove all
instructions except the call. But, on some platforms, if a value is returned as
a floating point, it may need to be removed from the stack (like x87). Thus, we
can't get rid of the bitcast even if there isn't a use of the value.
llvm-svn: 51134
2008-05-14 22:45:20 +00:00
e15051d64b
rename SimplifyCFG.cpp -> SimplifyCFGPass.cpp
...
llvm-svn: 51130
2008-05-14 20:38:44 +00:00
f2763e233e
Simplify internalize pass. Add test case.
...
Patch by Matthijs Kooijman!
llvm-svn: 51114
2008-05-14 20:01:01 +00:00
3dc2d92ebd
Split the loop unroll mechanism logic out into a utility function.
...
Patch by Matthijs Kooijman!
llvm-svn: 51083
2008-05-14 00:24:14 +00:00
17816b321f
Fix Analysis/BasicAA/pure-const-dce.ll. This turned out to be a correctness
...
bug as well as a missed optimization. We weren't properly checking for local
dependencies before moving on to non-local ones when doing non-local read-only
call CSE.
llvm-svn: 51082
2008-05-13 23:18:30 +00:00
e695ab227c
Fix for PR 2323, infinite loop in tail dup.
...
llvm-svn: 51063
2008-05-13 20:06:43 +00:00
8c2391d00d
Make the non-local CSE safety checks slightly more thorough.
...
llvm-svn: 51035
2008-05-13 13:41:23 +00:00
69057b80c7
Add support for non-local CSE of read-only calls.
...
llvm-svn: 51024
2008-05-13 08:17:22 +00:00
0479aa5c0b
Change class' public PassInfo variables to by initialized with the
...
address of the PassInfo directly instead of calling getPassInfo.
This eliminates a bunch of dynamic initializations of static data.
Also, fold RegisterPassBase into PassInfo, make a bunch of its
data members const, and rearrange some code to initialize data
members in constructors instead of using setter member functions.
llvm-svn: 51022
2008-05-13 02:05:11 +00:00
53c5c62d6d
80 col / tabs fixes
...
llvm-svn: 51021
2008-05-13 01:48:26 +00:00
d78c400b5b
Clean up the use of static and anonymous namespaces. This turned up
...
several things that were neither in an anonymous namespace nor static
but not intended to be global.
llvm-svn: 51017
2008-05-13 00:00:25 +00:00
f792860255
Go back to passing the analyses around as parameters.
...
llvm-svn: 50995
2008-05-12 20:15:55 +00:00
4afb1c864a
Move the various analyses used by GVN into static variables so we don't have to keep passing them around or refetching them.
...
llvm-svn: 50963
2008-05-12 08:15:27 +00:00
47fed61526
Fix various DOUTs to not call the extremely expensive Value::getName()
...
method. DOUT statements are disabled when assertions are off, but the
side effects of getName() are still evaluated. Just call getNameSTart,
which is close enough and doesn't cause heap traffic.
llvm-svn: 50958
2008-05-11 01:55:59 +00:00
82146fa267
Simplify code by using SwitchInst::findCaseValue instead of reimplementing it.
...
llvm-svn: 50957
2008-05-10 23:56:54 +00:00
a4ee1f516f
don't sink invokes, even if they are readonly. This fixes a
...
crash on kimwitu++.
llvm-svn: 50901
2008-05-09 15:07:33 +00:00
437435dcbc
Fix a type and formatting.
...
llvm-svn: 50900
2008-05-09 12:20:10 +00:00
aaba10e843
Implement PR2298. This transforms:
...
~x < ~y --> y < x
-x == -y --> x == y
llvm-svn: 50882
2008-05-09 05:19:28 +00:00
e7f0afe168
restore doxygen comment.
...
llvm-svn: 50881
2008-05-09 04:43:13 +00:00
829046b0b4
Improve pass documentation and comments.
...
Patch by Matthijs Kooijman!
llvm-svn: 50861
2008-05-08 17:46:35 +00:00
49a594e6ab
More than just loads can read from memory: readonly calls like strlen
...
also need to be checked for memory modifying instructions before we
can sink them. THis fixes the second half of PR2297.
llvm-svn: 50860
2008-05-08 17:37:37 +00:00
4fa09669d8
Make instcombine's DSE respect loads as well as stores. It is not safe to
...
delete the first store in:
store x -> p
load p
store y -> p
This is for PR2297.
llvm-svn: 50859
2008-05-08 17:20:30 +00:00
4758caa926
Check linkage.
...
llvm-svn: 50851
2008-05-08 15:08:39 +00:00
fc2edad4ae
Turn StripPointerCast() into a method
...
llvm-svn: 50836
2008-05-07 22:54:15 +00:00
5a3eecdfd8
Fix a bug in the ComputeMaskedBits logic for multiply.
...
llvm-svn: 50793
2008-05-07 00:35:55 +00:00
82c02b28f3
Make StripPointerCast a common function (should we mak it method of Value instead?)
...
llvm-svn: 50775
2008-05-06 22:52:30 +00:00
0e1ab4a9be
We need to update PHIs containing the exiting block, not the exit block. We really should come up with better names for these.
...
llvm-svn: 50770
2008-05-06 20:55:16 +00:00
7ffc3c9a95
Fix typo.
...
llvm-svn: 50713
2008-05-06 05:40:11 +00:00
de68fabb35
fix typo Duncan noticed
...
llvm-svn: 50699
2008-05-06 02:31:18 +00:00
6a2da37c0e
Make several variable declarations static.
...
llvm-svn: 50696
2008-05-06 01:53:16 +00:00
a8b7e78f54
Remove uses of llvm/System/IncludeFile.h that are no longer needed.
...
llvm-svn: 50695
2008-05-06 01:32:53 +00:00
cf0e3acf16
Correct the value of LowBits in srem and urem handling in
...
ComputeMaskedBits.
llvm-svn: 50692
2008-05-06 00:51:48 +00:00
4ead264c08
Fix: Some classes were derived from a class in an anonymous namespace, but they
...
themselves weren't in the anonymous namespace.
llvm-svn: 50673
2008-05-05 21:37:59 +00:00
8ed8e3d0e6
Fix a crash when threading a block that includes a MRV call result.
...
DemoteRegToStack doesn't work with MRVs yet, because it relies on the
ability to load/store things.
This fixes PR2285.
llvm-svn: 50667
2008-05-05 20:21:22 +00:00
2d7a4d70c3
processStore may delete the instruction, avoid
...
using dyn_cast<> on already freed memory.
llvm-svn: 50618
2008-05-04 08:51:25 +00:00
fa0e3c4a92
Handle multiple return values.
...
llvm-svn: 50604
2008-05-03 01:12:15 +00:00
a1ec89fbf1
Do not sink getresult.
...
llvm-svn: 50600
2008-05-03 00:36:30 +00:00
1962c2be6a
Fix a mistake in the computation of leading zeros for udiv.
...
llvm-svn: 50591
2008-05-02 21:30:02 +00:00
5f0563ceb6
strength reduce exp2 into ldexp, rdar://5852514
...
llvm-svn: 50586
2008-05-02 18:43:35 +00:00
a700b2bd0f
add a FIXME so we remember to eventually remove this code.
...
llvm-svn: 50582
2008-05-02 17:18:31 +00:00
86ceb0db9c
Porting r50563 from Tak to mainline.
...
llvm-svn: 50564
2008-05-02 00:43:20 +00:00
78ffe6e939
Don't try to create PHIs of struct types. Fallout
...
from x86-64 calling convention work.
llvm-svn: 50545
2008-05-01 22:27:44 +00:00
4be6ae4e6c
Fix an overaggressive SimplifyDemandedBits optimization on urem. This
...
fixes the 254.gap regression on x86 and the 403.gcc regression on x86-64.
llvm-svn: 50537
2008-05-01 19:13:24 +00:00
bb41aab426
1) add '-debug' output
...
2) Return NULL instead of false in several places for tidiness.
3) fix a bug optimizing sprintf(p, "%c", x)
llvm-svn: 50521
2008-05-01 06:39:12 +00:00
b9b5d6ddaa
Delete the IPO simplify-libcalls and completely reimplement it as
...
a FunctionPass. This makes it simpler, fixes dozens of bugs, adds
a couple of minor features, and shrinks is considerably: from
2214 to 1437 lines.
llvm-svn: 50520
2008-05-01 06:25:24 +00:00
0ced13ccd9
This condition got inverted accidentally.
...
llvm-svn: 50473
2008-04-30 07:16:33 +00:00
2dc4426675
move lowering of llvm.memset -> store from simplify libcalls
...
to instcombine.
llvm-svn: 50472
2008-04-30 06:39:11 +00:00
438e35c4d1
use string length computation to generalize several xforms.
...
llvm-svn: 50464
2008-04-30 03:07:53 +00:00
ad5367f8ed
Revert r50441. The original code was correct. Add some more comments so that I don't make the same mistake in the future.
...
llvm-svn: 50446
2008-04-29 21:51:00 +00:00
ff7d7b18e5
Fix a bug in memcpyopt where the memcpy-memcpy transform was never being applied because
...
we were checking for it in the wrong order. This caused a miscompilation because the
return slot optimization assumes that the call it is dealing with is NOT a memcpy.
llvm-svn: 50444
2008-04-29 21:26:06 +00:00
f07de734cf
We should be returning true here since we've changed the function.
...
llvm-svn: 50442
2008-04-29 21:02:46 +00:00
e674600266
A lot of cleanups and documentation improvements, as well as a few corner case fixes. Most
...
of this was suggested by Chris.
llvm-svn: 50441
2008-04-29 20:59:33 +00:00
2306a1e098
Rename DeadLoopElimination to LoopDeletion, part 2.
...
llvm-svn: 50437
2008-04-29 20:06:54 +00:00
e9f05bd1f0
Rename DeadLoopElimination to LoopDeletion, part one.
...
llvm-svn: 50436
2008-04-29 19:58:07 +00:00
d9e3b5c5bd
don't eliminate load from volatile value on paths where the load is dead.
...
This fixes the second half of PR2262
llvm-svn: 50430
2008-04-29 17:28:22 +00:00
9233c124c9
fix a subtle volatile handling bug.
...
llvm-svn: 50428
2008-04-29 17:13:43 +00:00
92f4702254
Implement more aggressive support for analyzing string length. This
...
generalizes the previous code to handle the case when the string is not
an immediate to the strlen call (for example, crazy stuff like
strlen(c ? "foo" : "bart"+1) -> 3). This implements
gcc.c-torture/execute/builtins/strlen-2.c. I will generalize other
cases in simplifylibcalls to use the same routine later.
llvm-svn: 50408
2008-04-29 06:56:02 +00:00
304ef22f6e
Clarify what we mean by a dead loop.
...
llvm-svn: 50406
2008-04-29 06:34:55 +00:00
e331a65c79
don't delete the last store to an alloca if the store is volatile.
...
llvm-svn: 50390
2008-04-29 04:58:38 +00:00
586216e5bb
Add some more comments.
...
llvm-svn: 50384
2008-04-29 00:45:15 +00:00
41377175d3
Remove debugging code.
...
llvm-svn: 50383
2008-04-29 00:39:24 +00:00
94ad702412
Add dead loop elimination, which removes dead loops for which we can compute
...
the trip count.
llvm-svn: 50382
2008-04-29 00:38:34 +00:00
8cb19d967f
Fix DSE to not eliminate volatile loads with no uses.
...
llvm-svn: 50370
2008-04-28 19:51:27 +00:00
72ec3f4562
Teach InstCombine's ComputeMaskedBits what SelectionDAG's
...
ComputeMaskedBits knows about cttz, ctlz, and ctpop. Teach
SelectionDAG's ComputeMaskedBits what InstCombine's knows
about SRem. And teach them both some things about high bits
in Mul, UDiv, URem, and Sub. This allows instcombine and
dagcombine to eliminate sign-extension operations in
several new cases.
llvm-svn: 50358
2008-04-28 17:02:21 +00:00
8be72700b8
Fix PR2256, yet another miscompilation in simplifycfg of i
...
multiple return values.
Bill, please pull this into Tak.
llvm-svn: 50332
2008-04-28 00:19:07 +00:00
2237973438
Implement a signficant optimization for inline asm:
...
When choosing between constraints with multiple options,
like "ir", test to see if we can use the 'i' constraint and
go with that if possible. This produces more optimal ASM in
all cases (sparing a register and an instruction to load it),
and fixes inline asm like this:
void test () {
asm volatile (" %c0 %1 " : : "imr" (42), "imr"(14));
}
Previously we would dump "42" into a memory location (which
is ok for the 'm' constraint) which would cause a problem
because the 'c' modifier is not valid on memory operands.
Isn't it great how inline asm turns 'missed optimization'
into 'compile failed'??
Incidentally, this was the todo in
PowerPC/2007-04-24-InlineAsm-I-Modifier.ll
Please do NOT pull this into Tak.
llvm-svn: 50315
2008-04-27 00:37:18 +00:00
4793515a9c
Move a bunch of inline asm code out of line.
...
llvm-svn: 50313
2008-04-27 00:09:47 +00:00
67ca6f6347
When SRoA'ing a global variable, make sure the new globals get the
...
appropriate alignment. This fixes a miscompilation of 252.eon on
x86-64 (rdar://5891920).
Bill, please pull this into Tak.
llvm-svn: 50308
2008-04-26 07:40:11 +00:00
0d1d3df564
change comments per review
...
llvm-svn: 50300
2008-04-25 21:16:07 +00:00
ca95a5f49f
Remove the code from CodeGenPrepare that moved getresult instructions
...
to the block that defines their operands. This doesn't work in the
case that the operand is an invoke, because invoke is a terminator
and must be the last instruction in a block.
Replace it with support in SelectionDAGISel for copying struct values
into sequences of virtual registers.
llvm-svn: 50279
2008-04-25 18:27:55 +00:00
ca270ad96f
Feedback from chris
...
llvm-svn: 50271
2008-04-25 17:45:52 +00:00
4d43d3c72c
Remove 'unwinds to' support from mainline. This patch undoes r47802 r47989
...
r48047 r48084 r48085 r48086 r48088 r48096 r48099 r48109 and r48123.
llvm-svn: 50265
2008-04-25 16:53:59 +00:00
6fed3b2038
Teach the PruningFunctionCloner how to look through loads with
...
ConstantExpression GEPs pointing into constant globals.
llvm-svn: 50256
2008-04-25 06:37:06 +00:00
f7de528463
Don't infininitely thread branches when a threaded edge
...
goes back to the block, e.g.:
Threading edge through bool from 'bb37.us.thread3829' to 'bb37.us' with cost: 1, across block:
bb37.us: ; preds = %bb37.us.thread3829, %bb37.us, %bb33
%D1361.1.us = phi i32 [ %tmp36, %bb33 ], [ %D1361.1.us, %bb37.us ], [ 0, %bb37.us.thread3829 ] ; <i32> [#uses=2]
%tmp39.us = icmp eq i32 %D1361.1.us, 0 ; <i1> [#uses=1]
br i1 %tmp39.us, label %bb37.us, label %bb42.us
llvm-svn: 50251
2008-04-25 04:12:29 +00:00
608eeef5ce
Adjust inline cost computation to be less aggressive.
...
llvm-svn: 50222
2008-04-24 18:42:47 +00:00
97951ac580
code restructuring, not functionality change.
...
llvm-svn: 50203
2008-04-24 00:21:50 +00:00
12f1e007f7
Don't replace multiple result of calls with undef,
...
sccp tracks getresult values, not call values in this
case.
llvm-svn: 50202
2008-04-24 00:19:54 +00:00
769203cb03
code cleanup, no functionality change.
...
llvm-svn: 50201
2008-04-24 00:16:28 +00:00
86bbf338e5
Split some code out of the main SimplifyCFG loop into its own function.
...
Fix said code to handle merging return instructions together correctly
when handling multiple return values.
llvm-svn: 50199
2008-04-24 00:01:19 +00:00
8f83081fea
Check type instead of no. of operands.
...
llvm-svn: 50179
2008-04-23 20:18:29 +00:00
f6e15a4774
Rewrite previous patch to suit Chris's preference.
...
llvm-svn: 50174
2008-04-23 18:34:37 +00:00
a82d691caa
simplify code for propagation of constant arguments into
...
callees.
llvm-svn: 50142
2008-04-23 06:16:27 +00:00
5f1802cfdf
Fix a number of bugs in ipconstantprop, simplify the code, fit in 80 cols,
...
fix read after free bug (PR2238).
llvm-svn: 50141
2008-04-23 05:59:23 +00:00
5a58a4dc6d
Rewrite multiple return value handling in SCCP. Before, the -sccp pass
...
would turn every getresult instruction into undef. This helps with
rdar://5778210
llvm-svn: 50140
2008-04-23 05:38:20 +00:00
493527d8c9
Do not change the type of a ByVal argument to a
...
type of a different size.
llvm-svn: 50121
2008-04-23 01:03:05 +00:00
1c89ca7295
Don't do: "(X & 4) >> 1 == 2 --> (X & 4) == 4" if there are more than one uses of the shift result.
...
llvm-svn: 50118
2008-04-23 00:38:06 +00:00
37e9c187b0
Start doing the significantly useful part of jump threading: handle cases
...
where a comparison has a phi input and that phi is a constant. For example,
stuff like:
Threading edge through bool from 'bb2149' to 'bb2231' with cost: 1, across block:
bb2237: ; preds = %bb2231, %bb2149
%tmp2328.rle = phi i32 [ %tmp2232, %bb2231 ], [ %tmp2232439, %bb2149 ] ; <i32> [#uses=2]
%done.0 = phi i32 [ %done.2, %bb2231 ], [ 0, %bb2149 ] ; <i32> [#uses=1]
%tmp2239 = icmp eq i32 %done.0, 0 ; <i1> [#uses=1]
br i1 %tmp2239, label %bb2231, label %bb2327
or
bb38.i298: ; preds = %bb33.i295, %bb1693
%tmp39.i296.rle = phi %struct.ibox* [ null, %bb1693 ], [ %tmp39.i296.rle1109, %bb33.i295 ] ; <%struct.ibox*> [#uses=2]
%minspan.1.i291.reg2mem.1 = phi i32 [ 32000, %bb1693 ], [ %minspan.0.i288, %bb33.i295 ] ; <i32> [#uses=1]
%tmp40.i297 = icmp eq %struct.ibox* %tmp39.i296.rle, null ; <i1> [#uses=1]
br i1 %tmp40.i297, label %implfeeds.exit311, label %bb43.i301
This triggers thousands of times in spec.
llvm-svn: 50110
2008-04-22 21:40:39 +00:00
d5425e8f8d
Dig through multiple levels of AND to thread jumps if needed.
...
llvm-svn: 50106
2008-04-22 20:46:09 +00:00
3df4c15dc7
Teach jump threading to thread through blocks like:
...
br (and X, phi(Y, Z, false)), label L1, label L2
This triggers once on 252.eon and 6 times on 176.gcc. Blocks
in question often look like this:
bb262: ; preds = %bb261, %bb248
%iftmp.251.0 = phi i1 [ true, %bb261 ], [ false, %bb248 ] ; <i1> [#uses=4]
%tmp270 = icmp eq %struct.rtx_def* %tmp.0.i, null ; <i1> [#uses=1]
%bothcond = or i1 %iftmp.251.0, %tmp270 ; <i1> [#uses=1]
br i1 %bothcond, label %bb288, label %bb273
In this case, it is clear that it doesn't matter if tmp.0.i is null when coming from bb261. When coming from bb248, it is all that matters.
Another random example:
check_asm_operands.exit: ; preds = %check_asm_operands.exit.thr_comm, %bb30.i, %bb12.i, %bb6.i413
%tmp.0.i420 = phi i1 [ true, %bb6.i413 ], [ true, %bb12.i ], [ true, %bb30.i ], [ false, %check_asm_operands.exit.thr_comm ; <i1> [#uses=1]
call void @llvm.stackrestore( i8* %savedstack ) nounwind
%tmp4389 = icmp eq i32 %added_sets_1.0, 0 ; <i1> [#uses=1]
%tmp4394 = icmp eq i32 %added_sets_2.0, 0 ; <i1> [#uses=1]
%bothcond80 = and i1 %tmp4389, %tmp4394 ; <i1> [#uses=1]
%bothcond81 = and i1 %bothcond80, %tmp.0.i420 ; <i1> [#uses=1]
br i1 %bothcond81, label %bb4398, label %bb4397
Here is the case from 252.eon:
bb290.i.i: ; preds = %bb23.i57.i.i, %bb8.i39.i.i, %bb100.i.i, %bb100.i.i, %bb85.i.i110
%myEOF.1.i.i = phi i1 [ true, %bb100.i.i ], [ true, %bb100.i.i ], [ true, %bb85.i.i110 ], [ true, %bb8.i39.i.i ], [ false, %bb23.i57.i.i ] ; <i1> [#uses=2]
%i.4.i.i = phi i32 [ %i.1.i.i, %bb85.i.i110 ], [ %i.0.i.i, %bb100.i.i ], [ %i.0.i.i, %bb100.i.i ], [ %i.3.i.i, %bb8.i39.i.i ], [ %i.3.i.i, %bb23.i57.i.i ] ; <i32> [#uses=3]
%tmp292.i.i = load i8* %tmp16.i.i100, align 1 ; <i8> [#uses=1]
%tmp293.not.i.i = icmp ne i8 %tmp292.i.i, 0 ; <i1> [#uses=1]
%bothcond.i.i = and i1 %tmp293.not.i.i, %myEOF.1.i.i ; <i1> [#uses=1]
br i1 %bothcond.i.i, label %bb202.i.i, label %bb301.i.i
Factoring out 3 common predecessors.
On the path from any blocks other than bb23.i57.i.i, the load and compare
are dead.
llvm-svn: 50096
2008-04-22 07:05:46 +00:00
e369c35a84
refactor some code, no functionality change.
...
llvm-svn: 50094
2008-04-22 06:36:15 +00:00
8fb13cbe4e
remove dead code.
...
llvm-svn: 50080
2008-04-22 03:21:48 +00:00
c3a439351c
optimize "p != gep p, ..." better. This allows us to compile
...
getelementptr-seteq.ll into:
define i1 @test(i64 %X, %S* %P) {
%C = icmp eq i64 %X, -1 ; <i1> [#uses=1]
ret i1 %C
}
instead of:
define i1 @test(i64 %X, %S* %P) {
%A.idx.mask = and i64 %X, 4611686018427387903 ; <i64> [#uses=1]
%C = icmp eq i64 %A.idx.mask, 4611686018427387903 ; <i1> [#uses=1]
ret i1 %C
}
And fixes the second half of PR2235. This speeds up the insertion sort
case by 45%, from 1.12s to 0.77s. In practice, this will significantly
speed up for loops structured like:
for (double *P = Base + N; P != Base; --P)
...
Which happens frequently for C++ iterators.
llvm-svn: 50079
2008-04-22 02:53:33 +00:00
bab7bec9c8
fix grammar-o, thanks to Duncan for noticing.
...
llvm-svn: 50047
2008-04-21 18:25:01 +00:00
a5b96ecef9
Remove unneeded #include's.
...
llvm-svn: 50035
2008-04-21 07:47:38 +00:00
6a7355caa2
Refactor memcpyopt based on Chris' suggestions. Consolidate several functions
...
and simplify code that was fallout from the separation of memcpyopt and gvn.
llvm-svn: 50034
2008-04-21 07:45:10 +00:00
ad0d42ba15
don't assume that the argument passed to fprintf("%s" is a string. This
...
fixes a crash in opt on 433.milc.
llvm-svn: 50023
2008-04-21 03:18:33 +00:00
f6236cc2e9
Use the new SplitBlockPredecessors to implement a todo.
...
llvm-svn: 50022
2008-04-21 02:57:57 +00:00
a5b11705b6
Move SplitBlockPredecessors out of loopsimplify into BasicBlockUtils.h
...
as a global helper function. At the same type, switch it from taking
a vector of predecessors to an arbitrary sequential input. This allows
us to switch LoopSimplify to use a SmallVector for various temporary
vectors that it passed into SplitBlockPredecessors.
llvm-svn: 50020
2008-04-21 01:28:02 +00:00
d418b06abf
Move domtree/frontier updating earlier, allowing us to use it to update phi
...
nodes, removing a hack.
llvm-svn: 50019
2008-04-21 01:05:08 +00:00
96e9e22269
Factor dominator tree and frontier updating into SplitBlockPredecessors
...
instead of doing it after every call.
llvm-svn: 50018
2008-04-21 00:54:38 +00:00
559c867ece
fit some more code in 80 cols.
...
llvm-svn: 50016
2008-04-21 00:25:49 +00:00
aca912d793
simplify code, fit in 80 cols.
...
llvm-svn: 50015
2008-04-21 00:23:14 +00:00
38806c3e9c
fit in 80 cols
...
llvm-svn: 50014
2008-04-21 00:19:16 +00:00
ff1c6e388c
finish the first cut of a jump threading pass implementation.
...
llvm-svn: 50006
2008-04-20 22:39:42 +00:00
567166c0a8
replace a slow and verbose version of Instruction::isUsedOutsideOfBlock with
...
a call to Instruction::isUsedOutsideOfBlock.
llvm-svn: 50005
2008-04-20 22:18:22 +00:00
9c1f1a82bf
we can only thread blocks when there is a pred we can determine the succ of.
...
llvm-svn: 50003
2008-04-20 21:18:09 +00:00
2115722ffa
improve comments, infrastructure, and add some validity checks for threading.
...
Add a cost function.
llvm-svn: 50002
2008-04-20 21:13:06 +00:00
b3b6007c8b
Add a new Jump Threading pass, which will handle cases
...
such as those in PR2235. Right now the pass is not very
effective. :)
llvm-svn: 50000
2008-04-20 20:35:01 +00:00
ab20784740
g++-4.3 build-fix: CHAR_BIT requires <climits>.
...
llvm-svn: 49989
2008-04-20 08:33:11 +00:00
3b18762f40
Switch to using Simplified ConstantFP::get API.
...
llvm-svn: 49977
2008-04-20 00:41:09 +00:00
eb6bb803a7
Allow argpromote to promote struct arguments with a specified number
...
of elements. Patch by Matthijs Kooijman!
llvm-svn: 49962
2008-04-19 19:50:01 +00:00
f9ae76d89c
Make GVN able to remove unnecessary calls to read-only functions again.
...
llvm-svn: 49842
2008-04-17 05:36:50 +00:00
376acf4aaa
Remove unused variable
...
llvm-svn: 49838
2008-04-17 01:30:44 +00:00
f66cb3696a
Workaround for PR2207, in which pred_iterator assert gets triggered due to a
...
wee problem in Xcode 2.[45]/gcc 4.0.1.
llvm-svn: 49831
2008-04-16 23:46:39 +00:00
c6a47e8a79
VisualStudio project files updated. #include <algorithm> added to make VisualStudio happy. Also had to undefine setjmp because of #include <csetjmp> turning setjmp into _setjmp in VisualStudio.
...
llvm-svn: 49743
2008-04-15 21:27:11 +00:00
4fff979a43
Remove unnecessary <sstream> includes.
...
llvm-svn: 49681
2008-04-14 20:40:47 +00:00
e36714c0b4
Minor whitespace and comment cleanups.
...
llvm-svn: 49671
2008-04-14 18:26:16 +00:00
7629b71dd4
Revert r49614. As Dan pointed out, some of these aren't correct.
...
llvm-svn: 49657
2008-04-14 17:38:21 +00:00
1f6fbc4bc3
Replace calls of the form V1->setName(V2->getName()) with V1->takeName(V2),
...
which is significantly more efficient.
llvm-svn: 49614
2008-04-13 19:15:17 +00:00
1e73f29a7f
Fix PR2213 by simultaneously making GVN more aggressive with the return values
...
of calls and less aggressive with non-readnone calls.
llvm-svn: 49516
2008-04-11 05:11:49 +00:00
99b7b3f03b
Teach InstCombine's ComputeMaskedBits to handle pointer expressions
...
in addition to integer expressions. Rewrite GetOrEnforceKnownAlignment
as a ComputeMaskedBits problem, moving all of its special alignment
knowledge to ComputeMaskedBits as low-zero-bits knowledge.
Also, teach ComputeMaskedBits a few basic things about Mul and PHI
instructions.
This improves ComputeMaskedBits-based simplifications in a few cases,
but more noticeably it significantly improves instcombine's alignment
detection for loads, stores, and memory intrinsics.
llvm-svn: 49492
2008-04-10 18:43:06 +00:00
a29d2536aa
Disable an xform we've had for a long time, pow(x,0.5) -> sqrt.
...
This is not safe for all inputs.
llvm-svn: 49458
2008-04-10 02:07:51 +00:00
802134fc02
Generalize getUnaryFloatFunction to handle any FP unary function, automatically
...
figuring out the suffix to use. implement pow(2,x) -> exp2(x).
llvm-svn: 49437
2008-04-09 17:48:11 +00:00
cca74e5ab9
use the new ConstantFP::get method to make this work with
...
long double and simplify the code.
llvm-svn: 49435
2008-04-09 17:17:35 +00:00
a7dfbc0366
Be conservative if getresult operand is neither call nor invoke.
...
llvm-svn: 49430
2008-04-09 15:58:24 +00:00
ef9a6fd5c2
Factor a bunch of functionality related to memcpy and memset transforms out of
...
GVN and into its own pass.
llvm-svn: 49419
2008-04-09 08:23:16 +00:00
8ee792d1b6
Remove accidentally duplicated code.
...
llvm-svn: 49418
2008-04-09 07:55:01 +00:00
b859fb49ed
many cleanups to the pow optimizer. Allow it to handle powf,
...
add support for pow(x, 2.0) -> x*x.
llvm-svn: 49411
2008-04-09 00:07:45 +00:00
8cd2a3ae2a
Fix insert point handling for multiple return values.
...
llvm-svn: 49367
2008-04-08 02:24:08 +00:00
ed92b41a39
Add operator= implementations to SparseBitVector, allowing it to be used in GVN. This results
...
in both time and memory savings for GVN. For example, one testcase went from 10.5s to 6s with
this patch.
llvm-svn: 49345
2008-04-07 17:38:23 +00:00
813384951e
Use Intrinsic::getDeclaration in more places.
...
llvm-svn: 49338
2008-04-07 13:45:04 +00:00
1416ebf1fe
The "stacksave is not nounwind problem" no longer
...
needs to be fixed here - a previous commit made sure
that intrinsics always get the right attributes.
So remove no-longer needed code, and while there use
Intrinsic::getDeclaration rather than getOrInsertFunction.
llvm-svn: 49337
2008-04-07 13:43:58 +00:00
fbc6adcc59
Use Intrinsic::getDeclaration to get hold of
...
intrinsics. Fix up the argument type (should
be i8*, was an array*).
llvm-svn: 49336
2008-04-07 13:41:19 +00:00
0c1e634cbb
Make GVN more memory efficient, particularly on code that contains a large number of
...
allocations, which GVN can't optimize anyways.
llvm-svn: 49329
2008-04-07 09:59:07 +00:00
87e484f08b
Mark calls to llvm.stacksave, llvm.stackrestore as
...
nounwind. When such calls are inlined into something
else that is invoked, they were getting changed to invokes,
which is badness.
llvm-svn: 49299
2008-04-07 00:08:48 +00:00
a39cfc5c5b
silence a warning when assertions are disabled.
...
llvm-svn: 49283
2008-04-06 21:44:08 +00:00
e9ecc68d8f
API changes for class Use size reduction, wave 1.
...
Specifically, introduction of XXX::Create methods
for Users that have a potentially variable number of
Uses.
llvm-svn: 49277
2008-04-06 20:25:17 +00:00
586740f401
Iterators folloring a SmallVector erased element are invalidated so
...
don't access cached iterators from after the erased element.
Re-apply 49056 with SmallVector support.
llvm-svn: 49106
2008-04-02 18:24:46 +00:00
ac38d444e2
1. Drop default inline threshold back down to 200.
...
2. Do not use # of basic blocks as part of the cost computation since it doesn't really figure into function size.
3. More aggressively inline function with vector code.
llvm-svn: 49061
2008-04-01 23:59:29 +00:00
052838c55d
Reverting 49056 due to the build being broken.
...
llvm-svn: 49060
2008-04-01 23:41:44 +00:00
7f7edc3824
Iterators folloring a SmallVector erased element are invalidated so
...
don't access cached iterators from after the erased element.
llvm-svn: 49056
2008-04-01 22:14:23 +00:00
5e4e051c2a
Revert 49006 for the moment.
...
llvm-svn: 49046
2008-04-01 20:00:57 +00:00
7d02cf3c9c
Emit exception handling info for functions which are
...
not marked nounwind, or for all functions when -enable-eh
is set, provided the target supports Dwarf EH.
llvm-gcc generates nounwind in the right places; other FEs
will need to do so also. Given such a FE, -enable-eh should
no longer be needed.
llvm-svn: 49006
2008-03-31 23:40:23 +00:00
f2b0b0eb17
Don't eliminate bitcast instructions that change the type of a pointer
...
llvm-svn: 48971
2008-03-31 00:22:16 +00:00
0f760dfe09
Fix "Control reaches the end of non-void function" warnings,
...
patch by David Chisnall.
llvm-svn: 48963
2008-03-30 18:22:13 +00:00
4311ad2dae
change iterator invalidation avoidance to just move the iterator backward
...
when something changes, instead of moving forward. This allows us to
simplify memset lowering, inserting the memset at the end of the range of
stuff we're touching instead of at the start.
This, in turn, allows us to make use of the addressing instructions already
used in the function instead of inserting our own. For example, we now
codegen:
%tmp41 = getelementptr [8 x i8]* %ref_idx, i32 0, i32 0 ; <i8*> [#uses=2]
call void @llvm.memset.i64( i8* %tmp41, i8 -1, i64 8, i32 1 )
instead of:
%tmp20 = getelementptr [8 x i8]* %ref_idx, i32 0, i32 7 ; <i8*> [#uses=1]
%ptroffset = getelementptr i8* %tmp20, i64 -7 ; <i8*> [#uses=1]
call void @llvm.memset.i64( i8* %ptroffset, i8 -1, i64 8, i32 1 )
llvm-svn: 48940
2008-03-29 05:15:47 +00:00
ac95515741
make the common case of a single store (which clearly shouldn't be turned
...
into a memset!) faster by avoiding an allocation of an std::list node.
llvm-svn: 48939
2008-03-29 04:52:12 +00:00
d528b21a65
give form-memset a significantly more sane heuristic, enable it by default.
...
llvm-svn: 48937
2008-03-29 04:36:18 +00:00
d62964a7d8
make memset inference significantly more powerful: it can now handle
...
memsets that initialize "structs of arrays" and other store sequences
that are not sequential. This is still only enabled if you pass
-form-memset-from-stores. The flag is not heavily tested and I haven't
analyzed the perf regressions when -form-memset-from-stores is passed
either, but this causes no make check regressions.
llvm-svn: 48909
2008-03-28 06:45:13 +00:00
eb1e3fcbe0
PHI->removeIncomingValue may remove PHInode.
...
Increment iterator in advance.
llvm-svn: 48890
2008-03-27 17:32:46 +00:00
2b72c05992
Handle a special case xor undef, undef -> 0. Technically this should be transformed to undef. But this is such a common idiom (misuse) we are going to handle it.
...
llvm-svn: 48791
2008-03-25 20:07:13 +00:00
a38f58aa5c
Add incoming value from header only if phi node has any use inside the loop.
...
llvm-svn: 48738
2008-03-24 20:16:14 +00:00
3471ae8c5d
Increasing the inline limit from (overly conservative) 200 to 300. Given each BB costs 20 and each instruction costs 5, 200 means a 4 BB function + 24 instructions (actually less because caller's size also contributes to it).
...
Furthermore, double the limit when more than 10% of the callee instructions are vector instructions. Multimedia kernels tend to love inlining.
llvm-svn: 48725
2008-03-24 06:37:48 +00:00
21a8e3d260
Temporarily disabling memset forming optimization. Add an option.
...
llvm-svn: 48720
2008-03-24 05:28:38 +00:00
c3cf9f872a
Transform (zext (or (icmp), (icmp))) to (or (zext (cimp), (zext icmp))) if at least one of the (zext icmp) can be transformed to eliminate an icmp.
...
llvm-svn: 48715
2008-03-24 00:21:34 +00:00
d38b3fb127
Preserve calling convention during function cloning
...
llvm-svn: 48708
2008-03-23 16:03:00 +00:00
53ccb62712
implement an initial hack at a straight-line store -> memset optimization.
...
This fires dozens of times across spec and multisource, but I don't know
if it actually speeds stuff up. Hopefully the testers will show something
nice :)
llvm-svn: 48680
2008-03-22 05:37:16 +00:00
168be766a8
implement the logic for memset insertion and store deletion.
...
llvm-svn: 48679
2008-03-22 04:13:49 +00:00
f5d41c67af
This is a partially implemented and currently disabled start of a store
...
merging optimization. Nothing to see here, hopefully more later :)
llvm-svn: 48670
2008-03-22 00:31:52 +00:00
9988569af8
Don't include <map> in Pass.h, which doesn't need it. This requires
...
adding <map> to many files that actually do need it.
llvm-svn: 48667
2008-03-21 23:51:57 +00:00
804209d17c
the size of a smallvector shouldn't be part of the interface to these methods.
...
llvm-svn: 48662
2008-03-21 22:01:16 +00:00
beb216da0a
make gvn marginally faster by reallocating the lastSeenLoad map for
...
each basic block.
llvm-svn: 48660
2008-03-21 21:33:23 +00:00
Evan Cheng