1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
Commit Graph

955 Commits

Author SHA1 Message Date
Dale Johannesen
626b0a32f7 Fix PR 4086, a bug in FP IV elimination.
llvm-svn: 70247
2009-04-27 21:03:15 +00:00
Dan Gohman
ff30ebd710 Teach getZeroExtendExpr and getSignExtendExpr to use trip-count
information to simplify [sz]ext({a,+,b}) to {zext(a),+,[zs]ext(b)},
as appropriate.

These functions and the trip count code each call into the other, so
this requires careful handling to avoid infinite recursion. During
the initial trip count computation, conservative SCEVs are used,
which are subsequently discarded once the trip count is actually
known.

Among other benefits, this change lets LSR automatically eliminate
some unnecessary zext-inreg and sext-inreg operation where the
operand is an induction variable.

llvm-svn: 70241
2009-04-27 20:16:15 +00:00
Dan Gohman
820b45049b Handle ands with ~0 correctly too. This fixes PR4052.
llvm-svn: 70176
2009-04-27 01:41:10 +00:00
Dan Gohman
a7fae1f865 Add several more icmp simplifications. Transform signed comparisons
into unsigned ones when the operands are known to have the same
sign bit value.

llvm-svn: 70053
2009-04-25 17:12:48 +00:00
Dan Gohman
9eb5ba6eb7 Handle ands with 0 and shifts by 0 correctly. These aren't
common, but indvars shouldn't crash on them. This fixes PR4054.

llvm-svn: 70051
2009-04-25 17:05:40 +00:00
Dan Gohman
ea9a6d22d3 Fix an error in this test.
llvm-svn: 69893
2009-04-23 15:22:28 +00:00
Dan Gohman
c0f47d6ec1 Change SCEVExpander's expandCodeFor to provide more flexibility
with the persistent insertion point, and change IndVars to make
use of it. This fixes a bug where IndVars was holding on to a
stale insertion point and forcing the SCEVExpander to continue to
use it.

This fixes PR4038.

llvm-svn: 69892
2009-04-23 15:16:49 +00:00
Owen Anderson
caa90b2561 Testcase for PR2639.
llvm-svn: 69867
2009-04-23 04:30:52 +00:00
Owen Anderson
bf7354995a Testcase for PR2537.
llvm-svn: 69866
2009-04-23 04:26:42 +00:00
Owen Anderson
f04f0e15c7 Fix typo.
llvm-svn: 69865
2009-04-23 04:24:19 +00:00
Owen Anderson
a1a09bc01f Testcase for PR3085.
llvm-svn: 69863
2009-04-23 04:21:14 +00:00
Owen Anderson
d4b3279a3f Add testcase from PR3086.
llvm-svn: 69862
2009-04-23 04:14:03 +00:00
Evan Cheng
bdfff0ba69 Make sure both operands have binary instructions have the same type.
llvm-svn: 69844
2009-04-22 23:39:28 +00:00
Evan Cheng
2af546d5fa Avoid deferencing use_begin() if value does not have a use.
llvm-svn: 69836
2009-04-22 22:45:37 +00:00
Dan Gohman
0ab6ecf6a1 SCEVExpander's InsertCastOfTo knows how to move existing cast
instructions in order to avoid inserting new ones. However, if
the cast instruction is the SCEVExpander's InsertPt, this
causes subsequently emitted instructions to be inserted near
the cast, and not at the location of the original insert point.
Fix this by adjusting the insert point in such cases.
This fixes PR4009.

llvm-svn: 69808
2009-04-22 16:11:16 +00:00
Chris Lattner
95aad4d625 fix a crash on a pointless but valid zero-length memset, rdar://6808691
llvm-svn: 69680
2009-04-21 16:52:12 +00:00
Dale Johannesen
040d118b17 Another testcase for IV shortening.
llvm-svn: 69247
2009-04-16 00:45:21 +00:00
Dale Johannesen
427e9aade9 Enhance induction variable code to remove the
sext around sext(shorter IV + constant), using a
longer IV instead, when it can figure out the
add can't overflow.  This comes up a lot in
subscripting; mainly affects 64 bit.

llvm-svn: 69123
2009-04-15 01:10:12 +00:00
Devang Patel
7323064183 While inlining, clone llvm.dbg.func.start intrinsic and adjust
llvm.dbg.region.end instrinsic. This nested llvm.dbg.func.start/llvm.dbg.region.end pair now enables DW_TAG_inlined_subroutine support in code generator.

llvm-svn: 69118
2009-04-15 00:17:06 +00:00
Evan Cheng
dba98a0669 Optimize conditional branch on i1 phis with non-constant inputs.
This turns:

eq:
        %3 = icmp eq i32 %1, %2
        br label %join

ne:
        %4 = icmp ne i32 %1, %2
        br label %join

join:
        %5 = phi i1 [%3, %eq], [%4, %ne]
        br i1 %5, label %yes, label %no

=>

eq:
        %3 = icmp eq i32 %1, %2
        br i1 %3, label %yes, label %no

ne:
        %4 = icmp ne i32 %1, %2
        br i1 %4, label %yes, label %no

llvm-svn: 69102
2009-04-14 23:40:03 +00:00
Chris Lattner
c1bfdc9bb2 Add a new "available_externally" linkage type. This is intended
to support C99 inline, GNU extern inline, etc.  Related bugzilla's
include PR3517, PR3100, & PR2933.  Nothing uses this yet, but it
appears to work.

llvm-svn: 68940
2009-04-13 05:44:34 +00:00
Chris Lattner
f03202e76d add some optimizations for strncpy/strncat and factor some
code.  Patch by Benjamin Kramer!

llvm-svn: 68885
2009-04-12 05:06:39 +00:00
Chris Lattner
7d75f78b92 Instcombine should not promote whole computation trees to "strange"
integer types, unless they are already strange.  This prevents it from
turning the code produced by SROA into crazy libcalls and stuff that 
the code generator can't handle.  In the attached example, the result
was an i96 multiply that caused the x86 backend to assert.

Note that if TargetData had an idea of what the legal types are for
a target that this could be used to stop instcombine from introducing
i64 muls, as Scott wanted.

llvm-svn: 68598
2009-04-08 05:41:03 +00:00
Chris Lattner
2f520929d4 fix rdar://6762290, a crash compiling cxx filt with clang.
llvm-svn: 68500
2009-04-07 05:03:34 +00:00
Ed Schouten
ff25f858fd Let the strcat optimizer return the pointer to the start of the buffer,
instead of the place where it started to perform the string copy.

- PR3661
- Patch by Benjamin Kramer!

llvm-svn: 68443
2009-04-06 13:06:48 +00:00
Owen Anderson
851ce6d1d5 Reapply r68211, with the miscompilations it caused fixed.
llvm-svn: 68262
2009-04-01 23:53:49 +00:00
Dan Gohman
a134448980 Revert r68172. It caused regressions in
Applications/Burg/burg
  Applications/ClamAV/clamscan
and many other tests.

llvm-svn: 68211
2009-04-01 16:37:47 +00:00
Owen Anderson
d7c837bb4b Enhance GVN to propagate simple conditionals. This fixes PR3921.
llvm-svn: 68172
2009-04-01 01:20:45 +00:00
Evan Cheng
c419350132 Throttle back "fold select into operand" transformation. InstCombine should not generate selects of two constants unless they are selects of 0 and 1.
e.g.
define i32 @t1(i32 %c, i32 %x) nounwind {
       %t1 = icmp eq i32 %c, 0
       %t2 = lshr i32 %x, 18
       %t3 = select i1 %t1, i32 %t2, i32 %x
       ret i32 %t3
}

was turned into

define i32 @t2(i32 %c, i32 %x) nounwind {
       %t1 = icmp eq i32 %c, 0
       %t2 = select i1 %t1, i32 18, i32 0
       %t3 = lshr i32 %x, %t2
       ret i32 %t3
}

For most targets, that means materializing two constants and then a select. e.g. On x86-64

movl    %esi, %eax
shrl    $18, %eax
testl   %edi, %edi
cmovne  %esi, %eax
ret

=>

xorl    %eax, %eax
testl   %edi, %edi
movl    $18, %ecx
cmovne  %eax, %ecx
movl    %esi, %eax
shrl    %cl, %eax
ret

Also, the optimizer and codegen can reason about shl / and / add, etc. by a constant. This optimization will hinder optimizations using ComputeMaskedBits.

llvm-svn: 68142
2009-03-31 20:42:45 +00:00
Devang Patel
ec65625744 Loop Index Split can eliminate a loop if it can determin if loop body is executed only once. There was a bug in determining IV based value of the iteration for which the loop body is executed. Fix it.
llvm-svn: 68071
2009-03-30 22:24:10 +00:00
Devang Patel
8c31ea5290 Before deleting a basic block, give other loop passes a chance cleanup analysis values, related to the instructions in the basic block.
llvm-svn: 67719
2009-03-25 23:57:48 +00:00
Chris Lattner
c055403764 Fix PR3874 by restoring a condition I removed, but making it more
precise than it used to be.

llvm-svn: 67662
2009-03-25 00:28:58 +00:00
Chris Lattner
aabd3eeeff canonicalize inttoptr and ptrtoint instructions which cast pointers
to/from integer types that are not intptr_t to convert to intptr_t
then do an integer conversion to the dest type.  This exposes the
cast to the optimizer.

llvm-svn: 67638
2009-03-24 18:35:40 +00:00
Chris Lattner
51a4134e1c two changes:
1. Make instcombine always canonicalize trunc x to i1 into an icmp(x&1).  This 
   exposes the AND to other instcombine xforms and is more of what the code
   generator expects.
2. Rewrite the remaining trunc pattern match to use 'match', which 
   simplifies it a lot.
   

llvm-svn: 67635
2009-03-24 18:15:30 +00:00
Chris Lattner
623662e8e1 Fix instcombine to not introduce undefined shifts when merging two
shifts together.  This fixes PR3851.

llvm-svn: 67411
2009-03-20 22:41:15 +00:00
Chris Lattner
6dce8d4135 aha, DAE does have to think about PHI nodes. Many thanks to "Dr Evil" (aka Duncan)
for pointing this out :)

llvm-svn: 67212
2009-03-18 16:48:45 +00:00
Chris Lattner
0542f9f1ba Fix PR3826 - InstComb assert with vector shift, by not calling ComputeNumSignBits on a vector.
llvm-svn: 67211
2009-03-18 16:32:19 +00:00
Zhou Sheng
90fc23d03d Fix a bug.
If I->use_empty(), this method should return false.

llvm-svn: 67180
2009-03-18 07:56:13 +00:00
Chris Lattner
7bef74e92f Fix PR3807 by inserting 'insertelement' instructions in the normal dest of
an invoke instead of after the invoke (in its block), which is invalid.

llvm-svn: 67139
2009-03-18 00:31:45 +00:00
Chris Lattner
120540fec6 remove a test that depends on -debug.
llvm-svn: 66937
2009-03-13 20:31:48 +00:00
John Criswell
d1fd7b1f72 Do not attempt to do parial redundancy elimination on void values.
Also fixed a punctuation error in the header comment.
This fixes PR3775.

llvm-svn: 66542
2009-03-10 15:04:53 +00:00
Evan Cheng
c147115cdf If a function is marked alwaysinline, it must be inlined (possibly for correctness). Do so even if the callee has dynamic alloca and the caller doesn't.
llvm-svn: 66539
2009-03-10 07:57:50 +00:00
Devang Patel
54b17d998a Ignore debug info while evaluating function.
llvm-svn: 66490
2009-03-09 23:04:12 +00:00
Chris Lattner
f05ebf0849 teach SROA to handle promoting vector allocas with a memset into them into
a vector type instead of into an integer type.

llvm-svn: 66368
2009-03-08 04:17:04 +00:00
Chris Lattner
54d2292fe5 Enhance SROA to "promote to scalar" allocas which are
memcpy/memmove'd into or out of.  This fixes a serious
perf issue that Nate ran into.

llvm-svn: 66366
2009-03-08 04:04:21 +00:00
Chris Lattner
fc43115972 fix a serious pessimization that Tron on IRC pointed out where we would
"boolify" pointers, generating really awful code because getting the pointer
value requires a load itself.  Before:

_foo:
	movb	$1, _X.b
	ret
_get:
	xorl	%ecx, %ecx
	movb	_X.b, %al
	testb	%al, %al
	movl	$_Y, %eax
	cmove	%ecx, %eax
	ret

With the xform disabled:

_foo:
	movl	$_Y, _X
	ret
_get:
	movl	_X, %eax
	ret

llvm-svn: 66351
2009-03-07 23:32:02 +00:00
Duncan Sands
26567dea82 While thinking about the one-definition-rule and trying
to find a tiny mouse hole to squeeze through, it struck
me that globals without a name can be considered internal
since they can't be referenced from outside the current
module.  This patch makes GlobalOpt give them internal
linkage.  Also done for aliases even though they always
have names, since in my opinion anonymous aliases should
be allowed for consistency with global variables and
functions.  So if that happens one day, this code is ready!

llvm-svn: 66267
2009-03-06 10:21:56 +00:00
Devang Patel
12e9aa7629 While converting an aggregate to scalare, ignore and remove aggregate's debug info.
llvm-svn: 66262
2009-03-06 07:03:54 +00:00
Devang Patel
3ae9d5d375 While hoisting instruction to speculatively execute simple bb, ignore dbg intrinsics.
llvm-svn: 66255
2009-03-06 06:00:17 +00:00
Devang Patel
9abfbc2e14 Skip DbgInfoIntrinsic.
llvm-svn: 66244
2009-03-06 02:59:27 +00:00