Commit Graph

26218 Commits

Author SHA1 Message Date
Nick Lewycky
504288e7af It turns out that "align 1" and unaligned are different. Add a bias to the
alignment attribute such that 0 means unaligned.

This will probably require a rebuild of llvm-gcc because of the change to
Attributes.h. If you see many test failures on "make check", please rebuild
your llvm-gcc.

llvm-svn: 61030
2008-12-15 07:29:55 +00:00
Mon P Wang
2f96113348 Added support to LegalizeType for expanding the operands of scalar to vector
and insert vector element.  Modified extract vector element to extend the
result to match the expected promoted type.

llvm-svn: 61029
2008-12-15 06:57:02 +00:00
Chris Lattner
30c1871282 gvn now hoists this load out of the hot non-call path.
llvm-svn: 61028
2008-12-15 06:34:48 +00:00
Chris Lattner
b467a5b4a5 Enable Load PRE. This teaches GVN to push partially redundant loads up the
CFG when there is exactly one predecessor where the load is not available.
This is designed to not increase code size but still eliminate partially
redundant loads.  This fires 1765 times on 403.gcc even though it doesn't
do critical edge splitting yet (the most common reason for it to fail).

llvm-svn: 61027
2008-12-15 05:28:29 +00:00
Chris Lattner
be89ad1615 if we have a phi translation failure of the start block,
return *just* a clobber of the start block, not other 
random stuff as well.

llvm-svn: 61026
2008-12-15 04:58:29 +00:00
Owen Anderson
90af4c9640 Ifdef out some code that I didn't mean to enable by default yet.
llvm-svn: 61024
2008-12-15 03:52:17 +00:00
Chris Lattner
22cfa14eed make GVN try to rename inputs to the resultant replaced values, which
cleans up the generated code a bit.  This should have the added benefit of
not randomly renaming functions/globals like my previous patch did. :)

llvm-svn: 61023
2008-12-15 03:46:38 +00:00
Chris Lattner
c92b131639 Implement initial support for PHI translation in memdep. This means that
memdep keeps track of how PHIs affect the pointer in dep queries, which 
allows it to eliminate the load in cases like rle-phi-translate.ll, which
basically end up being:

BB1:
   X = load P
   br BB3
BB2:
   Y = load Q
   br BB3
BB3:
   R = phi [P] [Q]
   load R

turning "load R" into a phi of X/Y.  In addition to additional exposed
opportunities, this makes memdep safe in many cases that it wasn't before
(which is required for load PRE) and also makes it substantially more 
efficient.  For example, consider:


bb1:  // has many predecessors.
   P = some_operator()
   load P

In this example, previously memdep would scan all the predecessors of BB1
to see if they had something that would mustalias P.  In some cases (e.g.
test/Transforms/GVN/rle-must-alias.ll) it would actually find them and end
up eliminating something.  In many other cases though, it would scan and not
find anything useful.  MemDep now stops at a block if the pointer is defined
in that block and cannot be phi translated to predecessors.  This causes it
to miss the (rare) cases like rle-must-alias.ll, but makes it faster by not
scanning tons of stuff that is unlikely to be useful.  For example, this
speeds up GVN as a whole from 3.928s to 2.448s (60%)!.  IMO, scalar GVN 
should be enhanced to simplify the rle-must-alias pointer base anyway, which
would allow the loads to be eliminated.

In the future, this should be enhanced to phi translate through geps and 
bitcasts as well (as indicated by FIXMEs) making memdep even more powerful.

llvm-svn: 61022
2008-12-15 03:35:32 +00:00
Owen Anderson
c2d2c0bdf3 Add support for slow-path GVN with full phi construction for scalars. This is disabled for now, as it actually pessimizes code in the abscence
of phi translation for load elimination.  This slow down GVN a bit, by about 2% on 403.gcc.

llvm-svn: 61021
2008-12-15 02:03:00 +00:00
Nick Lewycky
120e01b631 Fix whitespace in comment.
Remove TODO; icmp isn't a binary operator, so this function will never deal
with them.

llvm-svn: 61020
2008-12-15 01:35:36 +00:00
Nick Lewycky
8bdae4db80 Introducing nocapture, a parameter attribute for pointers to indicate that the
callee will not introduce any new aliases of that pointer.

The attributes had all bits allocated already, so I decided to collapse
alignment. Alignment was previously stored as a 16-bit integer from bits 16 to
32 of the attribute, but it was required to be a power of 2. Now it's stored in
log2 encoded form in five bits from 16 to 21. That gives us 11 more bits of
space.

You may have already noticed that you only need four bits to encode a 16-bit
power of two, so why five bits? Because the AsmParser accepted 32-bit
alignments, even though we couldn't store them (they were silently discarded).
Now we can store them in memory, but not in the bitcode.

The bitcode format was already storing these as 64-bit VBR integers. So, the
bitcode format stays the same, keeping the alignment values stored as 16 bit
raw values. There's some hideous code in the reader and writer that deals with
this, waiting to be ripped out the moment we run out of bits again and have to
replace the parameter attributes table encoding.

llvm-svn: 61019
2008-12-15 01:34:58 +00:00
Chris Lattner
10a0fb1e83 silence warning when asserts disabled.
llvm-svn: 61014
2008-12-14 21:38:24 +00:00
Chris Lattner
05dda70cd4 silence warning when asserts disabled.
llvm-svn: 61013
2008-12-14 21:37:33 +00:00
Chris Lattner
9458712db4 eliminate warning when asserts disabled.
llvm-svn: 61012
2008-12-14 21:36:23 +00:00
Owen Anderson
47efff5b14 Generalize GVN's phi construciton routine to work for things other than loads.
llvm-svn: 61009
2008-12-14 19:10:35 +00:00
Duncan Sands
ef671b5627 Reapply r60997, this time without forgetting that
target constants are allowed to have an illegal
type.

llvm-svn: 61006
2008-12-14 09:43:15 +00:00
Bill Wendling
380fbdc9f8 Temporarily revert r60997. It was causing this failure:
Running /Users/void/llvm/llvm.src/test/CodeGen/Generic/dg.exp ...
FAIL: /Users/void/llvm/llvm.src/test/CodeGen/Generic/asm-large-immediate.ll
Failed with exit(1) at line 1
while running:  llvm-as < /Users/void/llvm/llvm.src/test/CodeGen/Generic/asm-large-immediate.ll |  llc | /usr/bin/grep 68719476738
Assertion failed: ((TypesNeedLegalizing || getTypeAction(VT) == Legal) && "Illegal type introduced after type legalization?"), function HandleOp, file /Users/void/llvm/llvm.src/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp, line 493.
0   llc               0x0085392e char const* std::find<char const*, char>(char const*, char const*, char const&) + 98
1   llc               0x00853e63 llvm::sys::PrintStackTraceOnErrorSignal() + 593
2   libSystem.B.dylib 0x96cac09b _sigtramp + 43
3   libSystem.B.dylib 0xffffffff _sigtramp + 1765097359
4   libSystem.B.dylib 0x96d24ec2 raise + 26
5   libSystem.B.dylib 0x96d3447f abort + 73
6   libSystem.B.dylib 0x96d26063 __assert_rtn + 101
7   llc               0x004f9018 llvm::cast_retty<llvm::SubprogramDesc, llvm::DebugInfoDesc*>::ret_type llvm::cast<llvm::Sub
...

llvm-svn: 61001
2008-12-13 23:53:00 +00:00
Duncan Sands
7cddec2a2f LegalizeDAG is not supposed to introduce illegal
types into the DAG if they were not already there.
Check this with an assertion.

llvm-svn: 60997
2008-12-13 22:33:38 +00:00
Chris Lattner
0be74c4208 These messages should always be emitted when NDEBUG is unset, not when
NDEBUG is unset and -debug is passed.

llvm-svn: 60986
2008-12-13 18:37:58 +00:00
Bill Wendling
34182ae3ae Temporarily revert r60973. It's inexplicably causing a failure when self-hosting LLVM:
llvm[2]: Linking Release executable opt (without symbols)
...
Undefined symbols:
  "llvm::APFloat::IEEEsingle", referenced from:
      __ZN4llvm7APFloat10IEEEsingleE$non_lazy_ptr in libLLVMCore.a(Constants.o)
      __ZN4llvm7APFloat10IEEEsingleE$non_lazy_ptr in libLLVMCore.a(AsmWriter.o)
      __ZN4llvm7APFloat10IEEEsingleE$non_lazy_ptr in libLLVMCore.a(ConstantFold.o)
  "llvm::APFloat::IEEEdouble", referenced from:
      __ZN4llvm7APFloat10IEEEdoubleE$non_lazy_ptr in libLLVMCore.a(Constants.o)
      __ZN4llvm7APFloat10IEEEdoubleE$non_lazy_ptr in libLLVMCore.a(AsmWriter.o)
      __ZN4llvm7APFloat10IEEEdoubleE$non_lazy_ptr in libLLVMCore.a(ConstantFold.o)
ld: symbol(s) not found

This is in release mode. To replicate, compile llvm and llvm-gcc in optimized
mode. Then build llvm, in optimized mode, with the newly created compiler.

llvm-svn: 60977
2008-12-13 09:28:44 +00:00
Torok Edwin
8898288749 Fix getFieldAs() to use the parameter instead of 6.
Add missing DIType constructor, needed by DIVariable::getType().

llvm-svn: 60976
2008-12-13 08:25:29 +00:00
Mon P Wang
2880dc8e8c Remove assertion to allow promotion of a truncating store operand
llvm-svn: 60975
2008-12-13 08:16:43 +00:00
Mon P Wang
da91e0e191 Added basic support for expanding VSETCC
llvm-svn: 60974
2008-12-13 08:15:14 +00:00
Chris Lattner
8753175cd6 make RLE preserve the name of the load that it replaces. This is just
a pretification of the IR.

llvm-svn: 60973
2008-12-13 07:22:47 +00:00
Duncan Sands
1faa6258eb On big-endian machines it is wrong to do a full
width register load followed by a truncating
store for the copy, since the load will not place
the value in the lower bits.  Probably partial
loads/stores can never happen here, but fix it
anyway.

llvm-svn: 60972
2008-12-13 07:18:38 +00:00
Misha Brukman
5e6eec9337 Fix spelling.
llvm-svn: 60971
2008-12-13 05:21:37 +00:00
Devang Patel
5b7938b1cc Do not print empty DW_AT_comp_dir.
llvm-svn: 60965
2008-12-12 21:57:54 +00:00
Duncan Sands
ddce2cb415 When expanding unaligned loads and stores do not make
use of illegal integer types: instead, use a stack slot
and copying via integer registers.  The existing code
is still used if the bitconvert is to a legal integer
type.

This fires on the PPC testcases 2007-09-08-unaligned.ll
and vec_misaligned.ll.  It looks like equivalent code
is generated with these changes, just permuted, but
it's hard to tell.

With these changes, nothing in LegalizeDAG produces
illegal integer types anymore.  This is a prerequisite
for removing the LegalizeDAG type legalization code.

While there I noticed that the existing code doesn't
handle trunc store of f64 to f32: it turns this into
an i64 store, which represents a 4 byte stack smash.
I added a FIXME about this.  Hopefully someone more
motivated than I am will take care of it.

llvm-svn: 60964
2008-12-12 21:47:02 +00:00
Bill Wendling
13e4a3d0b0 - Use patterns instead of creating completely new instruction matching patterns,
which are identical to the original patterns.

- Change the multiply with overflow so that we distinguish between signed and
  unsigned multiplication. Currently, unsigned multiplication with overflow
  isn't working!

llvm-svn: 60963
2008-12-12 21:15:41 +00:00
Evan Cheng
56d9fc70bd Fix add/sub expansion: don't create ADD / SUB with two results (seems like everyone is doing this these days :-). Patch by Daniel M Gessel!
llvm-svn: 60958
2008-12-12 18:49:09 +00:00
Nick Lewycky
51228d6707 Revert my re-instated reverted commit, fixes the bootstrap build on x86-64 linux.
llvm-svn: 60951
2008-12-12 17:09:07 +00:00
Duncan Sands
06ecf57a87 When using a 4 byte jump table on a 64 bit machine,
do an extending load of the 4 bytes rather than a
potentially illegal (type) i32 load followed by a
sign extend.

llvm-svn: 60945
2008-12-12 08:13:38 +00:00
Duncan Sands
9f8a7550b6 Don't make use of an illegal type (i64) when
lowering f64 function arguments.

llvm-svn: 60944
2008-12-12 08:05:40 +00:00
Mon P Wang
53d0c96c6f Added support for SELECT v8i8 v4i16 for X86 (MMX)
Added support for TRUNC v8i16 to v8i8 for X86 (MMX)

llvm-svn: 60916
2008-12-12 01:25:51 +00:00
Bill Wendling
5d026e47c1 Redo the arithmetic with overflow architecture. I was changing the semantics of
ISD::ADD to emit an implicit EFLAGS. This was horribly broken. Instead, replace
the intrinsic with an ISD::SADDO node. Then custom lower that into an
X86ISD::ADD node with a associated SETCC that checks the correct condition code
(overflow or carry). Then that gets lowered into the correct X86::ADDOvf
instruction.

Similar for SUB and MUL instructions.

llvm-svn: 60915
2008-12-12 00:56:36 +00:00
Evan Cheng
dfa19a4009 Fix a 80 col. violation.
llvm-svn: 60901
2008-12-11 22:02:02 +00:00
Nick Lewycky
312d95be37 Sneaky, sneaky: move the -1 to the outside of the SMax. Reinstate the
optimization of SGE/SLE with unit stride, now that it works properly.

llvm-svn: 60881
2008-12-11 17:40:14 +00:00
Torok Edwin
9d454874f3 fix grammar, thanks Duncan!
llvm-svn: 60875
2008-12-11 11:44:49 +00:00
Torok Edwin
34056e3cc9 introduce BasicBlock::getUniquePredecessor()
llvm-svn: 60872
2008-12-11 10:36:07 +00:00
Mon P Wang
f578029326 Avoid generating a convert_rndsat node when the src and dest type are the same.
llvm-svn: 60869
2008-12-11 03:30:13 +00:00
Bill Wendling
060f17c854 Clarify FIXME.
llvm-svn: 60867
2008-12-11 01:26:44 +00:00
Mon P Wang
80cfaeecfe Whitespace clean up (tabs with spaces)
llvm-svn: 60866
2008-12-11 00:44:22 +00:00
Mon P Wang
4448877ed7 Make fix for r60829 less conservative to allow the proper optimization for
vec_extract-sse4.ll.

llvm-svn: 60865
2008-12-11 00:26:16 +00:00
Bill Wendling
02555039a0 Add a newline after this debug output.
llvm-svn: 60861
2008-12-10 23:24:43 +00:00
Bill Wendling
292263313b If ADD, SUB, or MUL have an overflow bit that's used, don't do transformation on
them. The DAG combiner expects that nodes that are transformed have one value
result.

llvm-svn: 60857
2008-12-10 22:36:00 +00:00
Evan Cheng
fc73640f83 Preliminary ARM debug support based on patch by Mikael of FlexyCore.
llvm-svn: 60851
2008-12-10 21:54:21 +00:00
Evan Cheng
487c9ff802 Some code clean up.
llvm-svn: 60850
2008-12-10 21:49:05 +00:00
Bill Wendling
417d88be16 Only perform SETO/SETC to JO/JC conversion if extractvalue is coming from an arithmetic with overflow instruction.
llvm-svn: 60844
2008-12-10 19:44:24 +00:00
Duncan Sands
81499a8e1c For amusement, implement SADDO, SSUBO, UADDO, USUBO
for promoted integer types, eg: i16 on ppc-32, or
i24 on any platform.  Complete support for arbitrary
precision integers would require handling expanded
integer types, eg: i128, but I couldn't be bothered.

llvm-svn: 60834
2008-12-10 12:30:42 +00:00
Duncan Sands
ecb1273c5b Don't dereference the end() iterator. This was
causing a bunch of failures when running
"make ENABLE_EXPENSIVE_CHECKS=1 check".

llvm-svn: 60832
2008-12-10 09:38:36 +00:00