1efabaaa2a
Allow readonly functions to unwind exceptions. Teach
...
the optimizers about this. For example, a readonly
function with no uses cannot be removed unless it is
also marked nounwind.
llvm-svn: 71071
2009-05-06 06:49:50 +00:00
1b5055ab7f
Return null instead of false, as appropriate.
...
llvm-svn: 70054
2009-04-25 17:28:45 +00:00
5638e0d642
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
46c97e626f
Allow i16 type indices to gep.
...
llvm-svn: 69946
2009-04-24 02:37:54 +00:00
428d490332
Before trying to introduce/eliminate cast/ext/trunc to make indices type as
...
pointer type, make sure that the pointer size is a valid sequential index type.
llvm-svn: 69574
2009-04-20 06:05:54 +00:00
eb510d6b3d
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
321741af5f
fix rdar://6762290, a crash compiling cxx filt with clang.
...
llvm-svn: 68500
2009-04-07 05:03:34 +00:00
826b6f0f7c
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
c3b2111d97
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
9e94538005
oops, I intended to remove this, not comment it out. Thanks Duncan!
...
llvm-svn: 67657
2009-03-24 23:48:25 +00:00
306813cbbb
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
d9eb41177a
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
1f15ca7c7a
Factorize out a concept - no functionality change.
...
llvm-svn: 67454
2009-03-21 21:27:31 +00:00
0a981d1d36
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
a09e0afe74
Don't load values out of global constants with weak
...
linkage: the value may be replaced with something
different at link time. (Frontends that want to
allow values to be loaded out of weak constants can
give their constants weak_odr linkage).
llvm-svn: 67407
2009-03-20 21:53:29 +00:00
595923ff75
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
e549493a55
Remove a condition which is always true.
...
llvm-svn: 67089
2009-03-17 17:55:15 +00:00
c65830519e
One more place where debug info affects codegen.
...
llvm-svn: 66930
2009-03-13 19:23:20 +00:00
fa54bc2052
Oops...I committed too much.
...
llvm-svn: 66867
2009-03-13 04:39:26 +00:00
b02eadf660
Temporarily XFAIL this test.
...
llvm-svn: 66866
2009-03-13 04:37:11 +00:00
08ccba73a7
Skip interleaved debug info when fast-forwarding through
...
allocations. Apparently the assumption is there is an
instruction (terminator?) following the allocation so I
am allowing the same assumption.
llvm-svn: 66716
2009-03-11 22:19:43 +00:00
703703aacb
Removing a dead debug intrinsic shouldn't trigger
...
another instcombine pass if we weren't going to make
one without debug info.
llvm-svn: 66576
2009-03-10 21:19:49 +00:00
dc35e5b43a
change the MemIntrinsic get/setAlignment method to take an unsigned
...
instead of a Constant*, which is what the clients of it really want.
llvm-svn: 66364
2009-03-08 03:59:00 +00:00
334268a211
Introduce a new MemTransferInst pseudo class, which is a common
...
parent between MemCpyInst and MemMoveInst, simplify some code to
use it.
llvm-svn: 66361
2009-03-08 03:37:16 +00:00
78ab338024
Fix another case where debug info was affecting
...
codegen. I convinced myself it was OK to skip all
pointer bitcasts here too.
llvm-svn: 66122
2009-03-05 02:06:48 +00:00
ad6b47377f
Fix another case where a dbg.declare meant something
...
had 2 uses instead of 1.
llvm-svn: 66112
2009-03-05 00:39:02 +00:00
c8b5a6ef7d
Always skip ptr-to-ptr bitcasts when counting,
...
per Chris' suggestion. Slightly faster.
llvm-svn: 65999
2009-03-04 01:53:05 +00:00
0365d3b8b5
Make my earlier patch to skip debug intrinsics
...
when counting work; it was only off by 1.
llvm-svn: 65993
2009-03-04 01:20:34 +00:00
09c3e8ec00
Instruction counters must skip the bitcasts that
...
feed into llvm.dbg.declare nodes, as well as
the debug directives themselves.
llvm-svn: 65976
2009-03-03 22:36:47 +00:00
77456b7ab4
When removing a store to an alloca that has only one
...
use, check also for the case where it has two uses,
the other being a llvm.dbg.declare. This is needed so
debug info doesn't affect codegen.
llvm-svn: 65970
2009-03-03 21:26:39 +00:00
92b551bc2b
Fix a bunch of Doxygen syntax issues. Escape special characters,
...
and put @file directives on their own comment line.
llvm-svn: 65920
2009-03-03 02:55:14 +00:00
0192552340
Don't count DebugInfo instructions in another limit
...
(lest they affect codegen).
llvm-svn: 65915
2009-03-03 01:43:03 +00:00
e1bb2f86f9
When sinking an insn in InstCombine bring its debug
...
info with it.
Don't count debug info insns against the scan maximum
in FindAvailableLoadedValue (lest they affect codegen).
llvm-svn: 65910
2009-03-03 01:09:07 +00:00
5795a6091d
Fix PR3694: add an instcombine micro-optimization that helps
...
clean up when using variable length arrays in llvm-gcc.
llvm-svn: 65832
2009-03-02 09:18:21 +00:00
34709f84d8
Silence compiler warning about use of uninitialized variables (in reality these
...
are always set by reference on the path that uses them.) No functional change.
llvm-svn: 65621
2009-02-27 06:37:39 +00:00
af618171f4
Fix PR3667
...
llvm-svn: 65464
2009-02-25 18:20:01 +00:00
4f356bb9b0
Fix a ValueTracking rule: RHS means operand 1, not 0. Add a simple
...
ashr instcombine to help expose this code. And apply the fix to
SelectionDAG's copy of this code too.
llvm-svn: 65364
2009-02-24 02:00:40 +00:00
3a86bcf134
Should reset DBI_Prev if DBI_Next == 0.
...
llvm-svn: 65314
2009-02-23 10:14:11 +00:00
d5420f0957
fix some typos that Duncan noticed
...
llvm-svn: 65306
2009-02-23 05:56:17 +00:00
648c5e9c99
Revert the part of 64623 that attempted to align the source in a
...
memcpy to match the alignment of the destination. It isn't necessary
for making loads and stores handled like the SSE loadu/storeu
intrinsics, and it was causing a performance regression in
MultiSource/Applications/JM/lencod.
The problem appears to have been a memcpy that copies from some
highly aligned array into an alloca; the alloca was then being
assigned a large alignment, which required codegen to perform
dynamic stack-pointer re-alignment, which forced the enclosing
function to have a frame pointer, which led to increased spilling.
llvm-svn: 65289
2009-02-22 18:06:32 +00:00
d44e80d7fc
Don't sign extend the char when expanding char -> int during
...
load(bitcast(char[4] to i32*)) evaluation.
llvm-svn: 65246
2009-02-21 20:50:42 +00:00
bef6b2098e
rename a function to indicate that it checks for profitability as well
...
as legality. Make load sinking and gep sinking more careful: we only
do it when it won't pessimize loads from the stack. This has the added
benefit of not producing code that is unanalyzable to SROA.
llvm-svn: 65209
2009-02-21 00:46:50 +00:00
24f31a0e59
commit a tweaked version of Daniel's patch for PR3599. We now
...
eliminate all the extensions and all but the one required truncate
from the testcase, but the or/and/shift stuff still isn't zapped.
llvm-svn: 64809
2009-02-17 20:47:23 +00:00
f68d29edd5
Fix EnforceKnownAlignment so that it doesn't ever reduce the alignment
...
of an alloca or global variable.
llvm-svn: 64693
2009-02-16 23:02:21 +00:00
9cdfd44521
Change these tests to use regular loads instead of llvm.x86.sse2.loadu.dq.
...
Enhance instcombine to use the preferred field of
GetOrEnforceKnownAlignment in more cases, so that regular IR operations are
optimized in the same way that the intrinsics currently are.
llvm-svn: 64623
2009-02-16 00:44:23 +00:00
318aea93bf
the two non-mask arguments to a shufflevector must be the same width, but they do not have to be the same
...
width as the result value.
llvm-svn: 64335
2009-02-11 22:36:25 +00:00
21eb52a74f
Instrcombine should not change load(cast p) to cast(load p) if the cast
...
changes the address space of the pointer.
llvm-svn: 64035
2009-02-07 22:19:29 +00:00
8542caa3f7
APInt'fy SimplifyDemandedVectorElts so it can analyze vectors with more than 64 elements.
...
llvm-svn: 63631
2009-02-03 10:05:09 +00:00
1aafe4cece
reduce indentation, (~XorCST->getValue()).isSignBit() -> isMaxSignedValue()
...
llvm-svn: 63500
2009-02-02 07:15:30 +00:00
f23908151a
Reinstate this optimization to fold icmp of xor when possible. Don't try to
...
turn icmp eq a+x, b+x into icmp eq a, b if a+x or b+x has other uses. This
may have been increasing register pressure leading to the bzip2 slowdown.
llvm-svn: 63487
2009-01-31 21:30:05 +00:00
Duncan Sands