Commits · faf75c8c9a11e36c44b8bc16022b00609d2aca5b · Roger Ferrer / llvm-epi-0.8

May 11, 2009
- Convert a subtract into a negate and an add when it helps x86 · faf75c8c
  Dan Gohman authored May 11, 2009
```
address folding.

llvm-svn: 71446
```
  faf75c8c
May 09, 2009

Rename PaddedSize to AllocSize, in the hope that this · af9eaa83

Duncan Sands authored May 09, 2009

will make it more obvious what it represents, and stop
it being confused with the StoreSize.

llvm-svn: 71349

af9eaa83

May 08, 2009

Factor out cycle-finder code and make it generic. · 65a58168
Anton Korobeynikov authored May 08, 2009
```
llvm-svn: 71241
```
65a58168
Fix PR4152: asm constraint validation happens before dag combine, so we · f1d9b914
Chris Lattner authored May 08, 2009
```
need to work a bit to combine things like (x+c1+c2) into x+c3.

llvm-svn: 71232
```
f1d9b914

Optimize code placement in loop to eliminate unconditional branches or move... · 2fa28110

Evan Cheng authored May 08, 2009

Optimize code placement in loop to eliminate unconditional branches or move unconditional branch to the outside of the loop. e.g.

///       A:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       <fallthrough to B>                                                                                                                                                 
///                                                                                                                                                                          
///       B:  --> loop header                                                                                                                                                
///       ...                                                                                                                                                                
///       jcc <cond> C, [exit]                                                                                                                                               
///                                                                                                                                                                          
///       C:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       jmp B                                                                                                                                                              
///                                                                                                                                                                          
/// ==>                                                                                                                                                                      
///                                                                                                                                                                          
///       A:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       jmp B                                                                                                                                                              
///                                                                                                                                                                          
///       C:  --> new loop header                                                                                                                                            
///       ...                                                                                                                                                                
///       <fallthough to B>                                                                                                                                                  
///                                                                                                                                                                          
///       B:                                                                                                                                                                 
///       ...                                                                                                                                                                
///       jcc <cond> C, [exit] 

llvm-svn: 71209

2fa28110

May 06, 2009

Use X86AddrNumOperands instead of magic constant one · 72b6582c

Dale Johannesen authored May 06, 2009

more place.  This fixes a bunch of x86-64 JIT regressions.
(Introduced when the value of the magic constant changed
in 68645.  At the time apparently nobody noticed; failures
were hidden in 70343-70439 by an unrelated bug, so showed
up again as "new" failures in 70440.)

llvm-svn: 71106

72b6582c

May 05, 2009

Add basic support for code generation of · be9fa506
Chris Lattner authored May 05, 2009
```
addrspace(257) -> FS relative on x86.  Patch by Zoltan Varga!

llvm-svn: 70992
```
be9fa506

Revert part of 70929 that has to do with determining whether a SIB byte is... · a35aed56

Evan Cheng authored May 05, 2009

Revert part of 70929 that has to do with determining whether a SIB byte is needed. It causes a lot of x86_64 JIT failures.

llvm-svn: 70986

a35aed56

- Avoid the longer SIB encoding on x86_64 when it's not needed. · c298ccb9

Evan Cheng authored May 04, 2009

- Synchronize instruction length computation code in X86InstrInfo with code in X86CodeEmitter.cpp
Patch by Zoltan Varga.

llvm-svn: 70929

c298ccb9

May 04, 2009
- X86FastISel doesn't support the -tailcallopt ABI. · bb525f7e
  Dan Gohman authored May 04, 2009
```
llvm-svn: 70902
```
  bb525f7e
May 01, 2009
- Fix compilation for some targets other than x86. · 31af6179
  Argyrios Kyrtzidis authored Apr 30, 2009
```
llvm-svn: 70522
```
  31af6179
- Make DebugLoc independent of DwarfWriter. · a5037484
  Argyrios Kyrtzidis authored Apr 30, 2009
```
-Replace DebugLocTuple's Source ID with CompileUnit's GlobalVariable*
-Remove DwarfWriter::getOrCreateSourceID
-Make necessary changes for the above (fix callsites, etc.)

llvm-svn: 70520
```
  a5037484
Apr 30, 2009
- Set mayLoad on MOVZX32_NOREXrm8 too. · db3a57ec
  Dan Gohman authored Apr 30, 2009
```
llvm-svn: 70466
```
  db3a57ec
- Mark MOV8mr_NOREX and MOV8rm_NOREX as mayStore / mayLoad respectively. · 99578674
  Evan Cheng authored Apr 30, 2009
```
llvm-svn: 70461
```
  99578674
- Instead of passing in an unsigned value for the optimization level, use an enum, · 026e5d76
  Bill Wendling authored Apr 29, 2009
```
which better identifies what the optimization is doing. And is more flexible for
future uses.

llvm-svn: 70440
```
  026e5d76
- Fix infinite recursion in the C++ code which handles movddup by making it unnecessary. · 7e6e3527
  Nate Begeman authored Apr 29, 2009
```
llvm-svn: 70425
```
  7e6e3527
Apr 29, 2009

Implement review feedback for vector shuffle work. · 5f829d89
Nate Begeman authored Apr 29, 2009
```
llvm-svn: 70372
```
5f829d89

Second attempt: · 084669a1

Bill Wendling authored Apr 29, 2009

Massive check in. This changes the "-fast" flag to "-O#" in llc. If you want to
use the old behavior, the flag is -O0. This change allows for finer-grained
control over which optimizations are run at different -O levels.

Most of this work was pretty mechanical. The majority of the fixes came from
verifying that a "fast" variable wasn't used anymore. The JIT still uses a
"Fast" flag. I'll change the JIT with a follow-up patch.

llvm-svn: 70343

084669a1

Apr 28, 2009

Properly print 'P' modifier on inline asm memory operands. · dac88bae
Anton Korobeynikov authored Apr 28, 2009
```
This should fix PR3379 and PR4064.
Patch inspired by Edwin Török!

llvm-svn: 70328
```
dac88bae
r70270 isn't ready yet. Back this out. Sorry for the noise. · 56f2987a
Bill Wendling authored Apr 28, 2009
```
llvm-svn: 70275
```
56f2987a

Massive check in. This changes the "-fast" flag to "-O#" in llc. If you want to · d0ae1594

Bill Wendling authored Apr 28, 2009

use the old behavior, the flag is -O0. This change allows for finer-grained
control over which optimizations are run at different -O levels.

Most of this work was pretty mechanical. The majority of the fixes came from
verifying that a "fast" variable wasn't used anymore. The JIT still uses a
"Fast" flag. I'm not 100% sure if it's necessary to change it there...

llvm-svn: 70270

d0ae1594

Apr 27, 2009

2nd attempt, fixing SSE4.1 issues and implementing feedback from duncan. · 8d6d4b92

Nate Begeman authored Apr 27, 2009

PR2957

ISD::VECTOR_SHUFFLE now stores an array of integers representing the shuffle
mask internal to the node, rather than taking a BUILD_VECTOR of ConstantSDNodes
as the shuffle mask.  A value of -1 represents UNDEF.

In addition to eliminating the creation of illegal BUILD_VECTORS just to 
represent shuffle masks, we are better about canonicalizing the shuffle mask,
resulting in substantially better code for some classes of shuffles.

llvm-svn: 70225

8d6d4b92

Rename GR8_ABCD to GR8_ABCD_L and create GR8_ABCD_H, and use these · 29869721

Dan Gohman authored Apr 27, 2009

to precisely describe the h-register subreg register classes.
Thanks to Jakob Stoklund Olesen for spotting this and for the
initial patch!

Also, make getStoreRegOpcode and getLoadRegOpcode aware of the
needs of h registers.

llvm-svn: 70211

29869721

Rename GR8_, GR16_, GR32_, and GR64_ to GR8_ABCD, GR16_ABCD, · ec542ca6
Dan Gohman authored Apr 27, 2009
```
GR32_ABCD, and GR64_ABCD, respectively, to help describe them.

llvm-svn: 70210
```
ec542ca6
Break up long multi-mnemonic strings into separate lines for readability. · ba99bddf
Dan Gohman authored Apr 27, 2009
```
llvm-svn: 70209
```
ba99bddf
Revised 68749 to allow matching of load/stores for address spaces < 256. · e15bf109
Mon P Wang authored Apr 27, 2009
```
llvm-svn: 70197
```
e15bf109

Apr 25, 2009
- add support for detecting process features on win64, patch by · 3ad60b18
  Chris Lattner authored Apr 25, 2009
```
Nicolas Capens!

llvm-svn: 70057
```
  3ad60b18
Apr 24, 2009

Fix PR 4004 by including the call to __tls_get_addr in X86tlsaddr. This is not · c1396a23
Rafael Espindola authored Apr 24, 2009
```
very elegant, but neither is the tls specification :-(

llvm-svn: 69968
```
c1396a23
Revert 69952. Causes testsuite failures on linux x86-64. · b93db668
Rafael Espindola authored Apr 24, 2009
```
llvm-svn: 69967
```
b93db668

PR2957 · bb881d66

Nate Begeman authored Apr 24, 2009

ISD::VECTOR_SHUFFLE now stores an array of integers representing the shuffle
mask internal to the node, rather than taking a BUILD_VECTOR of ConstantSDNodes
as the shuffle mask. A value of -1 represents UNDEF.

In addition to eliminating the creation of illegal BUILD_VECTORS just to
represent shuffle masks, we are better about canonicalizing the shuffle mask,
resulting in substantially better code for some classes of shuffles.

A clean up of x86 shuffle code, and some canonicalizing in DAGCombiner is next.

llvm-svn: 69952

bb881d66

Apr 23, 2009
- Add support for printing MO_ExternalSymbol operands in · 14efb90f
  Dan Gohman authored Apr 23, 2009
```
memory operand tuples. This doesn't ever come up in normal
code however.

llvm-svn: 69848
```
  14efb90f
Apr 21, 2009
- Get rid of what looks like a copy-and-pasted typo. · 7ce5cc6b
  Duncan Sands authored Apr 21, 2009
```
Spotted by gcc-4.5.

llvm-svn: 69673
```
  7ce5cc6b
- TLS_addr64 and TLS_addr32 define RDI and EAX. They don't use them. · 47ed1f52
  Rafael Espindola authored Apr 21, 2009
```
This fixes PR4002.

llvm-svn: 69672
```
  47ed1f52
- Make X86's copyRegToReg able to handle copies to and from subclasses. · 1addf647
  Dan Gohman authored Apr 20, 2009
```
This makes the extra copyRegToReg calls in ScheduleDAGSDNodesEmit.cpp
unnecessary. Derived from a patch by Jakob Stoklund Olesen.

llvm-svn: 69635
```
  1addf647
Apr 20, 2009

Move duplicated AddLiveIn function from X86 and ARM backends to be a method · f8b85477

Bob Wilson authored Apr 20, 2009

in the MachineFunction class, renaming it to addLiveIn for consistency with
the same method in MachineBasicBlock.  Thanks for Anton for suggesting this.

llvm-svn: 69615

f8b85477

Apr 18, 2009
- Fixed a few 64 bit cases in X86InstrInfo::commuteInstruction · 6c8bcf9d
  Mon P Wang authored Apr 18, 2009
```
llvm-svn: 69417
```
  6c8bcf9d
- Recommit r69335 and r69336. These were not causing problems. · 06684350
  Bill Wendling authored Apr 17, 2009
```
llvm-svn: 69394
```
  06684350
Apr 17, 2009
- For general dynamic TLS access we must use · 355fe12c
  Rafael Espindola authored Apr 17, 2009
```
leaq	foo@TLSGD(%rip), %rdi

as part of the instruction sequence. Using a register other than %rdi and then
copying it to %rdi is not valid.

llvm-svn: 69350
```
  355fe12c
- Revert r69335 and r69336. They were causing build failures. · 30527b11
  Bill Wendling authored Apr 17, 2009
```
llvm-svn: 69347
```
  30527b11
- MOV8rr_NOREX is a "Move" instruction. This doesn't currently · 09dbb0b5
  Dan Gohman authored Apr 17, 2009
```
matter, because this instruction isn't generated until after
things that care.

llvm-svn: 69336
```
  09dbb0b5