RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 19:52:54 +01:00
Code Issues Projects Releases Wiki Activity
51,004 Commits 4 Branches 5 Tags 704 MiB
f7a22e19cf
Commit Graph

1105 Commits

Author SHA1 Message Date
Devang Patel
10c075a316 Revert 79977. It causes llvm-gcc bootstrap failures on some platforms. 
llvm-svn: 80073
 2009-08-26 05:01:18 +00:00
Dan Gohman
e298d55930 Special-case static allocas in IndVarSimplify's loop invariant 
sinking code, since they are special. If the loop preheader happens
to be the entry block of a function, don't sink static allocas
out of it. This fixes PR4775.

llvm-svn: 80010
 2009-08-25 17:42:10 +00:00
Dan Gohman
bf08e82d8e Remove obsolete -f flags. 
llvm-svn: 79992
 2009-08-25 15:38:29 +00:00
Devang Patel
7d42bfab6c Update DebugInfo interface to use metadata, instead of special named llvm.dbg.... global variables, to encode debugging information in llvm IR. This is mostly a mechanical change that tests metadata support very well. 
This change speeds up llvm-gcc by more then 6% at "-O0 -g" (measured by compiling InstructionCombining.cpp!)

llvm-svn: 79977
 2009-08-25 05:24:07 +00:00
Dan Gohman
d240c19451 Change getelementptr folding to use APInt instead of uint64_t for 
offset computations. This fixes a truncation bug on targets that
don't have 64-bit pointers.

llvm-svn: 79639
 2009-08-21 16:52:54 +00:00
Dan Gohman
e3245061a9 Add targetdata strings to these tests, since SimplifyLibCalls uses 
TargetData to find the pointer size.

llvm-svn: 79490
 2009-08-19 23:18:49 +00:00
Dan Gohman
cc511acf87 Fix a bug in the over-index constant folding. When over-indexing an 
array member of a struct, it's possible to land in an arbitrary position
inside that struct, such that attempting to find further getelementptr
indices will fail. In such cases, folding cannot be done.

llvm-svn: 79485
 2009-08-19 22:46:59 +00:00
Dan Gohman
bc59c24278 Canonicalize indices in a constantexpr GEP. If Indices exceed the 
static extents of the static array type, it causes GlobalOpt and
other passes to be more conservative. This canonicalization also
allows the constant folder to add "inbounds" to GEPs.

llvm-svn: 79440
 2009-08-19 18:18:36 +00:00
Nick Lewycky
47bc7e0bd0 Fix up PHI nodes correctly in the presence of unreachable BBs, part two. Also 
delete a newed pointer, and improve readability a little bit.

llvm-svn: 79411
 2009-08-19 07:16:57 +00:00
Dan Gohman
807652ac3a Fix SimplifyLibcalls and ValueTracking to check mayBeOverridden 
before performing optimizations based on constant string values.

llvm-svn: 79384
 2009-08-19 00:11:12 +00:00
Dan Gohman
b0cf049a1e Generalize ScalarEvolution to be able to analyze GEPs when 
TargetData is not present. It still uses TargetData when available.
This generalization also fixed some limitations in the TargetData
case; the attached testcase covers this.

llvm-svn: 79344
 2009-08-18 16:46:41 +00:00
Dan Gohman
0b1af29372 Fix a bug that caused globalopt to miscompile tramp3d: don't miss 
unruly indices for arrays that are members of structs.

llvm-svn: 79337
 2009-08-18 14:58:19 +00:00
Nick Lewycky
db9fe40304 Test the pass the test is actually for, instead of one that doesn't exist. 
llvm-svn: 79257
 2009-08-17 17:41:29 +00:00
Nick Lewycky
afaae7957e Don't crash on critical edge. Patch by Andre Tavares. 
llvm-svn: 79252
 2009-08-17 17:00:57 +00:00
Nick Lewycky
3e2d98a039 Add a test that shows that SSI is working correctly. 
llvm-svn: 79230
 2009-08-17 07:32:08 +00:00
Nick Lewycky
e791519ea3 Don't crash trying to promote VLAs. 
llvm-svn: 79226
 2009-08-17 05:37:31 +00:00
Eli Friedman
ca19f19760 Fix for PR3016: detect the tricky case, where there are 
unfoldable references to a PHI node in the block being folded, and disable
the transformation in that case.  The correct transformation of such PHI
nodes depends on whether BB dominates Succ, and dominance is expensive 
to compute here.  (Alternatively, it's possible to check whether any 
uses are live, but that's also essentially a dominance calculation. 
Another alternative is to use reg2mem, but it probably isn't a good idea to
use that in simplifycfg.)

Also, remove some incorrect code from CanPropagatePredecessorsForPHIs 
which is made unnecessary with this patch: it didn't consider the case 
where a PHI node in BB has multiple uses.

llvm-svn: 79174
 2009-08-16 04:23:49 +00:00
Nick Lewycky
de61ef6c5e SSI construction should just go ahead and ignore instructions in unreachable 
blocks.

llvm-svn: 79132
 2009-08-15 20:12:18 +00:00
Mon P Wang
d56e4482fc When InstCombine simplifies a load -> extract element to gep -> load, place 
the new load by the old load instead of by the extract element because
a store could have occurred between the load and extract element.

llvm-svn: 78891
 2009-08-13 05:12:13 +00:00
Dan Gohman
9f80d2be6b Make LLVM Assembly dramatically easier to read by aligning the comments, 
using formatted_raw_ostream's PadToColumn.

Before:

bb1:            ; preds = %bb
  %2 = sext i32 %i.01 to i64            ; <i64> [#uses=1]
  %3 = getelementptr double* %p, i64 %2         ; <double*> [#uses=1]
  %4 = load double* %3, align 8         ; <double> [#uses=1]
  %5 = fmul double %4, 1.100000e+00             ; <double> [#uses=1]
  %6 = sext i32 %i.01 to i64            ; <i64> [#uses=1]
  %7 = getelementptr double* %p, i64 %6         ; <double*> [#uses=1]

After:

bb1:                                        ; preds = %bb
  %2 = sext i32 %i.01 to i64                ; <i64> [#uses=1]
  %3 = getelementptr double* %p, i64 %2     ; <double*> [#uses=1]
  %4 = load double* %3, align 8             ; <double> [#uses=1]
  %5 = fmul double %4, 1.100000e+00         ; <double> [#uses=1]
  %6 = sext i32 %i.01 to i64                ; <i64> [#uses=1]
  %7 = getelementptr double* %p, i64 %6     ; <double*> [#uses=1]

Several tests required whitespace adjustments.

llvm-svn: 78816
 2009-08-12 17:23:50 +00:00
Dan Gohman
00ee3a9a1a Transform -X/C to X/-C, implementing a README.txt entry. 
llvm-svn: 78812
 2009-08-12 16:37:02 +00:00
Dan Gohman
d5b6e35080 Optimize (x/C)*C to x if the division is exact. 
llvm-svn: 78811
 2009-08-12 16:33:09 +00:00
Dan Gohman
8d69df5773 Optimize exact sdiv by a constant power of 2 to ashr. 
llvm-svn: 78714
 2009-08-11 20:47:47 +00:00
Dan Gohman
08e747855c Don't assume that external global variables are aligned at their preferred 
alignment. Only the minimum alignment guaranteed by the ABI may be assumed.

llvm-svn: 78668
 2009-08-11 15:50:03 +00:00
Dan Gohman
aba682a290 Add -disable-output. Thanks Bill! 
llvm-svn: 78009
 2009-08-03 22:24:22 +00:00
Dan Gohman
39f93f6443 Add a new Constant::getIntegerValue helper function, and convert a 
few places in InstCombine to use it, to fix problems handling pointer
types. This fixes the recent llvm-gcc bootstrap error.

llvm-svn: 78005
 2009-08-03 22:07:33 +00:00
Eli Friedman
7bb0485237 PR4662: Fix a crash introduced by the recent LLVMContext changes. 
llvm-svn: 77716
 2009-07-31 19:36:47 +00:00
Daniel Dunbar
89cb72a6bc Fix PR4645 which was fallout from the fix for PR4641. 
- Call RAUW to delete all instructions (this is a patch from Nick Lewycky).

llvm-svn: 77512
 2009-07-29 22:00:43 +00:00
Nick Lewycky
1961298b63 Just discard the output, no need to turn it back into text. 
llvm-svn: 77439
 2009-07-29 06:14:52 +00:00
Chris Lattner
e5f1099d05 don't dump .bc file to stdout, and simplify this to a trivial testcase. 
llvm-svn: 77436
 2009-07-29 05:32:07 +00:00
Nick Lewycky
e0524c1795 Bulk erasing instructions without RAUWing them is unsafe. Instead, break them 
into a new BB that has no predecessors.

llvm-svn: 77433
 2009-07-29 05:17:50 +00:00
Dan Gohman
0d0dd7b732 Teach instcombine to respect and preserve inbounds. Add inbounds 
to a few tests where it is required for the expected transformation.

llvm-svn: 77290
 2009-07-28 01:40:03 +00:00
Chris Lattner
0426853d67 merge vector-casts-0.ll into vector-casts.ll 
llvm-svn: 76864
 2009-07-23 05:33:39 +00:00
Chris Lattner
c687344f0c Make some existing optimizations that would only trigger on scalars 
also apply to vectors.  This allows us to compile this:

#include <emmintrin.h>
__m128i a(__m128 a, __m128 b) { return a==a & b==b; }
__m128i b(__m128 a, __m128 b) { return a!=a | b!=b; }

to:

_a:
	cmpordps	%xmm1, %xmm0
	ret
_b:
	cmpunordps	%xmm1, %xmm0
	ret

with clang instead of to a ton of horrible code.

llvm-svn: 76863
 2009-07-23 05:32:17 +00:00
Chris Lattner
f4474da353 convert a test to filecheck format. This fixes an endemic problem 
with negative tests: this test wasn't checking what it thought it was
because it was grepping .bc, not .ll.

llvm-svn: 76861
 2009-07-23 05:27:48 +00:00
Chris Lattner
7061a4300d rename test 
llvm-svn: 76860
 2009-07-23 05:25:12 +00:00
Dan Gohman
f2c6e6a1bd Add a testcase for PR2831. 
llvm-svn: 76527
 2009-07-21 01:02:18 +00:00
Dan Gohman
74a435e9f1 The upper argument of ConstantRange is exclusive, not inclusive. 
llvm-svn: 76492
 2009-07-20 22:34:18 +00:00
Dan Gohman
00b05492f1 Revert the addition of hasNoPointerOverflow to GEPOperator. 
Getelementptrs that are defined to wrap are virtually useless to
optimization, and getelementptrs that are undefined on any kind
of overflow are too restrictive -- it's difficult to ensure that
all intermediate addresses are within bounds. I'm going to take
a different approach.

Remove a few optimizations that depended on this flag.

llvm-svn: 76437
 2009-07-20 17:43:30 +00:00
Eli Friedman
e507c1afaa Canonicalize bitcasts between types like <1 x i64> and i64 to 
insertelement/extractelement.

I'm not entirely sure this is precisely what we want to do: should we 
prefer bitcast(insertelement) or insertelement(bitcast)?  Similarly. should we 
prefer extractelement(bitcast) or bitcast(extractelement)?

llvm-svn: 76345
 2009-07-18 23:06:53 +00:00
Eli Friedman
debc43cb11 Back out 76300; apparently the preference is to canonicalize the other 
way (bitcast -> insert/extractelement).

llvm-svn: 76325
 2009-07-18 19:04:16 +00:00
Eli Friedman
65a5fe312a Add combine: X sdiv (1 << Y) -> X udiv (1 << Y) when X doesn't have the 
sign bit set.

llvm-svn: 76304
 2009-07-18 09:53:21 +00:00
Eli Friedman
f1878fcda1 Canonicalize insert/extractelement from single-element vectors into 
bitcasts.

It would also be possible to canonicalize the other way; does anyone 
have a preference?

llvm-svn: 76300
 2009-07-18 09:07:47 +00:00
Eli Friedman
7b1597133d Fix simplifylibcalls memset recognition to work on 64-bit platforms 
where int is 32 bits.

llvm-svn: 76293
 2009-07-18 08:34:51 +00:00
Dan Gohman
50e65d8c93 Fill in some holes in ScalarEvolution's loop iteration condition 
analysis. This allows indvars to emit a simpler loop trip count
expression.

llvm-svn: 76085
 2009-07-16 17:34:36 +00:00
Eli Friedman
6aa39dcd93 Switch invars away from using isTrapping when it really shouldn't be 
using it.

llvm-svn: 75852
 2009-07-15 22:48:29 +00:00
Eli Friedman
048d13f9bb Don't restrict the set of instructions where we try to constant-fold the 
operands; it's possible to end up with a constant-foldable operand to 
most instructions, even those which can't trap.

llvm-svn: 75845
 2009-07-15 22:13:34 +00:00
Dan Gohman
5329511fae Fix the expansion of umax and smax in the case where one or more of 
the operands have pointer type, so that the resulting type matches
the original SCEV type, and so that unnecessary ptrtoints are
avoided in common cases.

llvm-svn: 75680
 2009-07-14 20:57:04 +00:00
Dan Gohman
9525cf679f Add a testcase for a bug fixed by r75634. 
llvm-svn: 75644
 2009-07-14 18:15:00 +00:00
Dale Johannesen
35fc3243a8 Revert 75571; I'm convinced this isn't the right thing to do. 
llvm-svn: 75642
 2009-07-14 17:48:25 +00:00