76a63ed099
now that all the pieces are in place, teach instcombine's
...
simplifydemandedbits to simplify instructions with *multiple
uses* in contexts where it can get away with it. This allows
it to simplify the code in multi-use-or.ll into a single 'add
double'.
This change is particularly interesting because it will cover
up for some common codegen bugs with large integers created due
to the recent SROA patch. When working on fixing those bugs,
this should be disabled.
llvm-svn: 63481
2009-01-31 08:40:03 +00:00
94cfb281c3
make sure to set Changed=true when instcombine hacks on the code,
...
not doing so prevents it from properly iterating and prevents it
from deleting the entire body of dce-iterate.ll
llvm-svn: 63476
2009-01-31 07:04:22 +00:00
3537a62704
Fixed optimization of combining two shuffles where the first shuffle inputs
...
has a different number of elements than the output.
llvm-svn: 62998
2009-01-26 04:39:00 +00:00
f4395ea97a
testcase for PR3381.
...
Also it was an empty struct, not a void after all.
llvm-svn: 62920
2009-01-24 17:16:04 +00:00
72cd68fe64
Make InstCombineStoreToCast handle aggregates more aggressively,
...
handling the case in Transforms/InstCombine/cast-store-gep.ll, which
is a heavily reduced testcase from Clang on x86-64.
llvm-svn: 62904
2009-01-24 01:00:13 +00:00
1f86498f93
Do not use host floating point types when emitting
...
ASCII IR; loading and storing these can change the
bits of NaNs on some hosts. Remove or add warnings
at a few other places using host floating point;
this is a bad thing to do in general.
llvm-svn: 62712
2009-01-21 20:32:55 +00:00
287b4bc44e
Disable on x86_64 until I figure out what's wrong.
...
llvm-svn: 62660
2009-01-21 02:08:30 +00:00
b5721632ee
Make special cases (0 inf nan) work for frem.
...
Besides APFloat, this involved removing code
from two places that thought they knew the
result of frem(0., x) but were wrong.
llvm-svn: 62645
2009-01-21 00:35:19 +00:00
e75fdb0510
Calls to fmod, it turns out, are constant-folded by
...
invoking the host fmod, not by lowering to frem and
constant-folding that. Fix this so it tests what I
want to test.
llvm-svn: 62622
2009-01-20 21:58:13 +00:00
d067ecd1c7
Move & restructure test per review.
...
llvm-svn: 62538
2009-01-19 22:33:12 +00:00
db2d9613d2
Fix PR3335 by not turning a store to one address space into a store to another.
...
llvm-svn: 62351
2009-01-16 20:12:52 +00:00
beac6f8b0c
Clean up previous cast optimization a bit. Also make zext elimination a bit more aggressive: if it's not necessary to emit an AND (i.e. high bits are already zero), it's profitable to evaluate the operand at a different type.
...
llvm-svn: 62297
2009-01-16 02:11:43 +00:00
60e19a46f2
- Teach CanEvaluateInDifferentType of this xform: sext (zext ty1), ty2 -> zext ty2
...
- Looking at the number of sign bits of the a sext instruction to determine whether new trunc + sext pair should be added when its source is being evaluated in a different type.
llvm-svn: 62263
2009-01-15 17:01:23 +00:00
59af77376c
Make instcombine ensure that all allocas are explicitly aligned at at
...
least their preferred alignment.
llvm-svn: 62176
2009-01-13 20:18:38 +00:00
f50aa6ae5c
Implement rdar://6480391, extending of equality icmp's to avoid a truncation.
...
I noticed this in the code compiled for a routine using std::map, which produced
this code:
%25 = tail call i32 @memcmp(i8* %24, i8* %23, i32 6) nounwind readonly
%.lobit.i = lshr i32 %25, 31 ; <i32> [#uses=1]
%tmp.i = trunc i32 %.lobit.i to i8 ; <i8> [#uses=1]
%toBool = icmp eq i8 %tmp.i, 0 ; <i1> [#uses=1]
br i1 %toBool, label %bb3, label %bb4
which compiled to:
call L_memcmp$stub
shrl $31, %eax
testb %al, %al
jne LBB1_11 ##
with this change, we compile it to:
call L_memcmp$stub
testl %eax, %eax
js LBB1_11
This triggers all the time in common code, with patters like this:
%169 = and i32 %ply, 1 ; <i32> [#uses=1]
%170 = trunc i32 %169 to i8 ; <i8> [#uses=1]
%toBool = icmp ne i8 %170, 0 ; <i1> [#uses=1]
%7 = lshr i32 %6, 24 ; <i32> [#uses=1]
%9 = trunc i32 %7 to i8 ; <i8> [#uses=1]
%10 = icmp ne i8 %9, 0 ; <i1> [#uses=1]
etc
llvm-svn: 61985
2009-01-09 07:47:06 +00:00
fef138b140
Fix part 3/2 of PR3290, making instcombine zap (gep(bitcast)) when possible.
...
llvm-svn: 61980
2009-01-09 05:44:56 +00:00
f3e696bc5a
ValueTracker can't assume that an alloca with no specified alignment
...
will get its preferred alignment. It has to be careful and cautiously assume
it will just get the ABI alignment. This prevents instcombine from rounding
up the alignment of a load/store without adjusting the alignment of the alloca.
llvm-svn: 61934
2009-01-08 19:28:38 +00:00
4e735eb157
make m_ConstantInt(int64_t) safely match ConstantInt's that are larger than i64.
...
This fixes an instcombine crash on PR3235.
llvm-svn: 61775
2009-01-05 23:45:50 +00:00
0a09d2de13
XFAIL this test. The xform was removed.
...
llvm-svn: 61624
2009-01-04 06:32:28 +00:00
aedb54a947
Add transformation:
...
xor (or (icmp, icmp), true) -> and(icmp, icmp)
This is possible because of De Morgan's law.
llvm-svn: 61537
2009-01-01 01:18:23 +00:00
0f0e63fe73
Make all the vector elements positive in an srem of constant vector.
...
llvm-svn: 61195
2008-12-18 06:31:11 +00:00
582fe6b0ca
Use m_Specific() instead of double matching.
...
llvm-svn: 60341
2008-12-01 08:09:47 +00:00
9e6b243428
simplify these patterns using m_Specific. No need to grep for
...
xor in testcase (or is a substring).
llvm-svn: 60328
2008-12-01 05:16:26 +00:00
9d02a70a7d
Teach inst combine to merge GEPs through PHIs. This is really
...
important because it is sinking the loads using the GEPs, but
not the GEPs themselves. This triggers 647 times on 403.gcc
and makes the .s file much much nicer. For example before:
je LBB1_87 ## bb78
LBB1_62: ## bb77
leal 84(%esi), %eax
LBB1_63: ## bb79
movl (%eax), %eax
...
LBB1_87: ## bb78
movl $0, 4(%esp)
movl %esi, (%esp)
call L_make_decl_rtl$stub
jmp LBB1_62 ## bb77
after:
jne LBB1_63 ## bb79
LBB1_62: ## bb78
movl $0, 4(%esp)
movl %esi, (%esp)
call L_make_decl_rtl$stub
LBB1_63: ## bb79
movl 84(%esi), %eax
The input code was (and the GEPs are merged and
the PHI is now eliminated by instcombine):
br i1 %tmp233, label %bb78, label %bb77
bb77:
%tmp234 = getelementptr %struct.tree_node* %t_addr.3, i32 0, i32 0, i32 22
br label %bb79
bb78:
call void @make_decl_rtl(%struct.tree_node* %t_addr.3, i8* null) nounwind
%tmp235 = getelementptr %struct.tree_node* %t_addr.3, i32 0, i32 0, i32 22
br label %bb79
bb79:
%iftmp.12.0.in = phi %struct.rtx_def** [ %tmp235, %bb78 ], [ %tmp234, %bb77 ]
%iftmp.12.0 = load %struct.rtx_def** %iftmp.12.0.in
llvm-svn: 60322
2008-12-01 02:34:36 +00:00
5b902c5b1e
Implement ((A|B)&1)|(B&-2) -> (A&1) | B transformation. This also takes care of
...
permutations of this pattern.
llvm-svn: 60312
2008-12-01 01:07:11 +00:00
de89bc275c
Add instruction combining for ((A&~B)|(~A&B)) -> A^B and all permutations.
...
llvm-svn: 60291
2008-11-30 13:52:49 +00:00
9eef421e12
Implement (A&((~A)|B)) -> A&B transformation in the instruction combiner. This
...
takes care of all permutations of this pattern.
llvm-svn: 60290
2008-11-30 13:08:13 +00:00
2d2e7861b5
getSExtValue() doesn't work for ConstantInts with bitwidth > 64 bits. Use all
...
APInt calls instead.
This fixes PR3144.
llvm-svn: 60288
2008-11-30 12:38:24 +00:00
361c0e5f9c
Strengthen check for div inst-combining.
...
llvm-svn: 60276
2008-11-30 04:33:53 +00:00
70635adea3
Instcombine was illegally transforming -X/C into X/-C when either X or C
...
overflowed on negation. This commit checks to make sure that neithe C nor X
overflows. This requires that the RHS of X (a subtract instruction) be a
constant integer.
llvm-svn: 60275
2008-11-30 03:42:12 +00:00
4ab50b93c8
Chris prefers icmp/select over udiv!
...
llvm-svn: 60187
2008-11-27 22:41:10 +00:00
69941fd0a0
Add a couple of missed optimizations on integer vectors. Multiply and divide
...
by 1, as well as multiply by -1.
llvm-svn: 60182
2008-11-27 20:21:08 +00:00
07d726ec4d
Optimize (x/y)*y into x-(x%y) in general. Div and rem are about the same, and
...
a subtract is cheaper than a multiply. This generalizes an existing transform.
llvm-svn: 59800
2008-11-21 07:33:58 +00:00
c3f3b059d0
Handle the case where there is no "not". It is possible it got
...
folded into the select.
llvm-svn: 59389
2008-11-16 04:25:26 +00:00
645f60141b
make this actually test what it is trying to.
...
llvm-svn: 59386
2008-11-16 04:21:51 +00:00
3f547be28f
If the LHS of the FCMP is coming from a UIToFP instruction, then we don't want
...
to generate signed ICMP instructions to replace the FCMP. This would violate
the following:
define i1 @test1(i32 %val) {
%1 = uitofp i32 %val to double
%2 = fcmp ole double %1, 0.000000e+00
ret i1 %2
}
would be transformed into:
define i1 @test1(i32 %val) {
%1 = icmp slt i33 %val, 1
ret i1 %1
}
which is obviously wrong. This patch modifes InstCombiner::FoldFCmp_IntToFP_Cst
to handle when the LHS comes from UIToFP.
llvm-svn: 58929
2008-11-09 04:26:50 +00:00
8d8acf327b
Fix demanded bits analysis with srem by negative number. Based on a patch
...
by Richard Osborne.
llvm-svn: 58555
2008-11-02 02:41:50 +00:00
83eea0b17f
Fix this recently moved code to use the correct type. CI is now a
...
ConstantInt, and SI is the original cast instruction. This fixes
PR2996.
llvm-svn: 58549
2008-11-02 00:17:33 +00:00
13cbcf1c18
Canonicalize sext(i1) to i1?-1:0, and update various instcombine
...
optimizations accordingly.
llvm-svn: 58457
2008-10-30 20:40:10 +00:00
2c34c130bf
(A & sext(C)) | (B & ~sext(C) -> C ? A : B
...
llvm-svn: 58351
2008-10-28 22:38:57 +00:00
730d5dade6
Don't try to create a mask when we don't need one. Fixes a crash.
...
llvm-svn: 58075
2008-10-24 06:14:27 +00:00
bc0278400c
Teach instcombine's visitLoad to scan back several instructions
...
to find opportunities for store-to-load forwarding or load CSE,
in the same way that visitStore scans back to do DSE. Also, define
a new helper function for testing whether the addresses of two
memory accesses are known to have the same value, and use it in
both visitStore and visitLoad.
These two changes allow instcombine to eliminate loads in code
produced by front-ends that frequently emit obviously redundant
addressing for memory references.
llvm-svn: 57608
2008-10-15 23:19:35 +00:00
d885f6e139
Combine (fcmp cc0 x, y) | (fcmp cc1 x, y) into a single fcmp when possible.
...
llvm-svn: 57515
2008-10-14 18:44:08 +00:00
ce70752b11
- Somehow I forgot about one / une.
...
- Renumber fcmp predicates to match their icmp counterparts.
- Try swapping operands to expose more optimization opportunities.
llvm-svn: 57513
2008-10-14 18:13:38 +00:00
67786cce66
Optimize anding of two fcmp into a single fcmp if the operands are the same. e.g. uno && ueq -> ueq
...
ord && olt -> olt
ord && ueq -> oeq
llvm-svn: 57507
2008-10-14 17:15:11 +00:00
da435910e8
Fix PR2697 by rewriting the '(X / pos) op neg' logic. This also changes
...
a couple other cases for clarity, but shouldn't affect correctness.
Patch by Eli Friedman!
llvm-svn: 57387
2008-10-11 22:55:00 +00:00
917a6c1343
rewrite bswap matching to be more general, allowing arbitrary
...
shifting and masking inside a bswap expr. This allows it to handle
the cases from PR2842, which involve the intermediate 'or'
expressions being shifted, not just the input value.
llvm-svn: 57095
2008-10-05 02:13:19 +00:00
e8ced3ec19
Fix misoptimization of: xor i1 (icmp eq (X, C1), icmp s[lg]t (X, C2))
...
llvm-svn: 56834
2008-09-30 06:08:34 +00:00
dafa9c6e85
Improve instcombine's handling of integer min and max in two ways:
...
- Recognize expressions like "x > -1 ? x : 0" as min/max and turn them
into expressions like "x < 0 ? 0 : x", which is easily recognizable
as a min/max operation.
- Refrain from folding expression like "y/2 < 1" to "y < 2" when the
comparison is being used as part of a min or max idiom, like
"y/2 < 1 ? 1 : y/2". In that case, the division has another use, so
folding doesn't eliminate it, and obfuscates the min/max, making it
harder to recognize as a min/max operation.
These benefit ScalarEvolution, CodeGen, and anything else that wants to
recognize integer min and max.
llvm-svn: 56246
2008-09-16 18:46:06 +00:00
eff71f2953
On 64-bit targets, change 32-bit getelementptr indices to be 64-bit
...
getelementptr indices, inserting an explicit cast if necessary.
This helps expose the sign-extension operation to other optimizations.
llvm-svn: 56133
2008-09-11 23:06:38 +00:00
7d01c0654c
Fix a vectorshuffle instcombine bug introduced by r55995.
...
Patch by Nicolas Capens!
llvm-svn: 56129
2008-09-11 22:47:57 +00:00
c1ae01688f
Fix an icmp+sdiv optimization to check for and handle an overflow
...
condition. This fixes PR2740.
llvm-svn: 56076
2008-09-10 23:30:57 +00:00
86fb5b48de
Make SimplifyDemandedVectorElts simplify vectors with multiple
...
users, and teach it about shufflevector instructions.
Also, fix a subtle bug in SimplifyDemandedVectorElts'
insertelement code.
This is a patch that was originally written by Eli Friedman,
with some fixes and cleanup by me.
llvm-svn: 55995
2008-09-09 18:11:14 +00:00
f023db6444
Don't crash when trying to constant fold a vector with some elements that can't
...
be folded. Instead, fail to fold the entire vector.
We could also return a vector with some elements folded and some not. If anyone
thinks that's a better approach, please speak up!
llvm-svn: 55689
2008-09-03 05:54:33 +00:00
99f4558117
Revert r54876 r54877 r54906 and r54907. Evan found that these caused a 20%
...
slowdown in bzip2.
llvm-svn: 55113
2008-08-21 05:56:10 +00:00
53b44029d6
Consider the case where xor by -1 and xor by 128 have been combined already to
...
produce an xor by 127.
llvm-svn: 54906
2008-08-17 19:58:24 +00:00
18f50b2637
Xor'ing both sides of icmp by sign-bit is equivalent to swapping signedness of
...
the predicate.
Also, make this optz'n apply in more cases where it's safe to do so.
llvm-svn: 54876
2008-08-17 07:34:14 +00:00
2a6adfa4f0
Remove GCSE and LoadVN from the testsuite.
...
llvm-svn: 54832
2008-08-16 00:00:54 +00:00
6134fbccef
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.
llvm-svn: 54767
2008-08-13 23:12:35 +00:00
2aa0ff27aa
Implement support for simplifying vector comparisons by 0.0 and 1.0 like we
...
do for scalars. Patch contributed by Nicolas Capens
This also generalizes the previous xforms to work on long double, now that
isExactlyValue works for long double.
llvm-svn: 54653
2008-08-11 22:06:05 +00:00
ac22cfcae9
Fix a shufflevector instcombine that was emitting invalid masks indices
...
when it meant to be emitting undef indices.
llvm-svn: 54417
2008-08-06 18:17:32 +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
0e966d3e2c
Just grep for through the LL code instead of the ASM code
...
llvm-svn: 54389
2008-08-06 00:10:32 +00:00
bc6786e7ee
Add default architecture.
...
llvm-svn: 54384
2008-08-05 23:36:00 +00:00
3dfa168d22
Testcase for PR2629.
...
llvm-svn: 54377
2008-08-05 22:23:59 +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
c600c53d1f
Fix PR2553
...
llvm-svn: 53715
2008-07-17 06:07:20 +00:00
f22f34f0cc
Add a few cases to instcombine's extractvalue testcase.
...
llvm-svn: 53675
2008-07-16 12:57:25 +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
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
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
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
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
1be09d9e21
make this condition more precise.
...
llvm-svn: 53442
2008-07-11 03:54:57 +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
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
7212e2014c
Fix a broken test. Neither load is eliminable without changing the CFG.
...
llvm-svn: 53273
2008-07-09 05:01:02 +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
855d2c38ec
new testcase for PR2496
...
llvm-svn: 53239
2008-07-08 17:18:05 +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
c9c81fb0df
Fix PR2488, a case where we deleted stack restores too aggressively.
...
llvm-svn: 52702
2008-06-25 05:59:28 +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
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
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
ac5bc8a3dd
Make testcase check for extractvalue instead of extractelement.
...
llvm-svn: 52317
2008-06-16 13:03:44 +00:00
2c580f0323
Remove unnecessary target lines.
...
llvm-svn: 52261
2008-06-13 22:12:16 +00:00
795749845d
Remove unnecessary target lines.
...
llvm-svn: 52260
2008-06-13 22:10:32 +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
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
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
6c0890a169
Ignore stderr for some more tests that expect warnings there.
...
This fixes 2 testcases.
llvm-svn: 52184
2008-06-10 16:13:38 +00:00
632a55e2cc
Fix two more not-grep tests that were missing llvm-dis.
...
llvm-svn: 52159
2008-06-09 22:36:45 +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
c775e462a8
Add a test case for opt -instcombine bug fix in revision 52003.
...
llvm-svn: 52004
2008-06-05 14:25:11 +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
ecdefb5df7
Implement PR2370: memmove(x,x,size) -> noop.
...
llvm-svn: 51636
2008-05-28 05:30:41 +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
3bf5512d87
Constant integer vectors may also be negated.
...
llvm-svn: 51476
2008-05-23 04:54:45 +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
d01c562e48
Eliminate questionable syntax for stdin redirection. This probably also speeds things up a bit.
...
llvm-svn: 51357
2008-05-20 22:07:21 +00:00
0843435b36
Oops, commit the version of this test that actually works.
...
llvm-svn: 51351
2008-05-20 21:19:36 +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
1e427c3264
sabre brings to my attention that the 'tr' suffix is also obsolete
...
llvm-svn: 51349
2008-05-20 21:00:03 +00:00
f45ff35bfe
Rename the last test with .llx extension to .ll, resolve duplicate test by renaming to isnan2. Now that no test has llx ending there is no need to search for them from dg.exp too.
...
llvm-svn: 51328
2008-05-20 19:52:04 +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
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
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
4b2a724fb8
Fix PR2339
...
llvm-svn: 51226
2008-05-18 04:11:26 +00:00
14b3604dcf
remove empty file?
...
llvm-svn: 51225
2008-05-18 04:10:18 +00:00
eb185ca5e9
Revert constant-folding change that will miscompile in some cases.
...
llvm-svn: 51223
2008-05-17 19:00:05 +00:00
1ba90bb69b
Constant fold inttoptr and ptrtoint.
...
llvm-svn: 51216
2008-05-17 09:03:26 +00:00
5c953b7d27
implement PR2328.
...
llvm-svn: 51176
2008-05-16 02:59:42 +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
8111b67ca8
Testcase for PR2303.
...
llvm-svn: 50951
2008-05-10 16:43:10 +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
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
5a3eecdfd8
Fix a bug in the ComputeMaskedBits logic for multiply.
...
llvm-svn: 50793
2008-05-07 00:35:55 +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
1962c2be6a
Fix a mistake in the computation of leading zeros for udiv.
...
llvm-svn: 50591
2008-05-02 21:30:02 +00:00
2cdcf2bd5f
Update old-style syntax in some "not grep" tests.
...
llvm-svn: 50560
2008-05-01 23:50:07 +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
adf28cb71c
move some tests from libcall optimizer suite.
...
llvm-svn: 50516
2008-05-01 06:13:48 +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
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
53bcf3609a
make this test reduced and *valid*
...
llvm-svn: 50429
2008-04-29 17:25:32 +00:00
9233c124c9
fix a subtle volatile handling bug.
...
llvm-svn: 50428
2008-04-29 17:13:43 +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
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
2161d6c075
distill down the essense of this test.
...
llvm-svn: 50125
2008-04-23 03:03:42 +00:00
c4d3c1cbe0
new test
...
llvm-svn: 50123
2008-04-23 01:22:22 +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
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
b839c05a05
rename *.llx -> *.ll, last batch.
...
llvm-svn: 49971
2008-04-19 22:32:52 +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
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
8bf97c2324
Byebye llvm-upgrade!
...
llvm-svn: 48762
2008-03-25 04:26:08 +00:00
Chris Lattner