Commit Graph

25024 Commits

Author SHA1 Message Date
Owen Anderson fe44aedf8a Add my most recent work.
llvm-svn: 28604
2006-05-31 22:15:45 +00:00
Evan Cheng cf70c7f42d Sign extender
llvm-svn: 28603
2006-05-31 22:05:11 +00:00
Reid Spencer d8df61a4d0 Squelch this warning:
/bzlib.c:126: warning: string length `1056' is greater than the length `509' ISO
C89 compilers are required to support

llvm-svn: 28602
2006-05-31 21:53:42 +00:00
Chris Lattner 9b32cf59e0 Fix utostr once and for all, by making there only be one function named
utostr.  To keep the efficiency in the 32-bit case, make it check to see if
the value is 32-bits and if so switch over to the faster 32-bit case.

llvm-svn: 28601
2006-05-31 21:25:50 +00:00
Chris Lattner 1df0e98ac2 Swap the order of operands created here. For +&|^, the order doesn't matter,
but for sub, it really does!  Fix fixes a miscompilation of fibheap_cut in
llvmgcc4.

llvm-svn: 28600
2006-05-31 21:14:00 +00:00
Owen Anderson dad8c57340 Extract a huge loop into a helper method. Fix a few iterator-invalidation bugs.
llvm-svn: 28599
2006-05-31 20:55:06 +00:00
Andrew Lenharth 440dcd03fa 4 billion names is enough for anyone. And really fix the build on alpha this time
llvm-svn: 28598
2006-05-31 20:40:36 +00:00
Andrew Lenharth 53250e7fcb fix utostr
llvm-svn: 28597
2006-05-31 20:18:56 +00:00
Andrew Lenharth 0d4c10eae6 Fix build breakage on alpha, without causing it on x86. as a bonus, all platforms can invent the same number of unique names now
llvm-svn: 28596
2006-05-31 20:18:28 +00:00
Andrew Lenharth 40d4af63f2 revert for now
llvm-svn: 28595
2006-05-31 19:16:26 +00:00
Evan Cheng 25e44e008d Rename instructions for consistency sake.
llvm-svn: 28594
2006-05-31 19:00:07 +00:00
Andrew Lenharth 26baab1593 make 64-bit safe and fix the build on alpha
llvm-svn: 28593
2006-05-31 18:56:42 +00:00
Evan Cheng 0c0996a97b commuteInstruction() does not always create a new MI!
llvm-svn: 28592
2006-05-31 18:03:39 +00:00
Reid Spencer 5dd0111188 Add a brief description for the tblgen program. More detail is needed but
the current author is not the person to provide it. Now that the file
exists, perhaps others will chime in and embellish.

llvm-svn: 28591
2006-05-31 17:54:39 +00:00
Reid Spencer a9355fa5d8 Update the documentation for llvm2cpp after the -gen-* options were added.
llvm-svn: 28590
2006-05-31 17:32:21 +00:00
Reid Spencer de2daad0ac Major reorganization and extension of the code. The diff on this will be a
mess as functions were moved around into a better ordering. The code was
extended to provide various -gen-* options to better control what the
generated output should be. Currently it is possible to generate entire
modules (three different ways), functions, global variables, and types.

llvm-svn: 28589
2006-05-31 17:31:38 +00:00
Reid Spencer 9fef163d19 Make the getNamedFunction and getNamedGlobal methods be const. They don't
change the module in any way and we should enforce that.

llvm-svn: 28588
2006-05-31 16:40:28 +00:00
Vladimir Prus 735e3f76a8 Clarify type naming.
llvm-svn: 28587
2006-05-31 16:03:20 +00:00
Vladimir Prus c1eae29b91 Improve InstVisitor docs.
llvm-svn: 28586
2006-05-31 15:30:18 +00:00
Evan Cheng 9d91caa053 Eliminate a memory leak.
llvm-svn: 28585
2006-05-31 07:13:03 +00:00
Evan Cheng 64d2846017 visitVBinOp: Can't fold divide by zero!
llvm-svn: 28584
2006-05-31 06:08:35 +00:00
Reid Spencer 03977ec9b1 Don't generate module definitions when the -fragment option is given.
llvm-svn: 28583
2006-05-31 04:43:19 +00:00
Evan Cheng 8abf45e22d Select vector_shuffle v1, undef <2, 3, ?, ?> to MOVHLPS.
llvm-svn: 28582
2006-05-31 00:51:37 +00:00
Evan Cheng 550cb663e8 Remove dead code.
llvm-svn: 28581
2006-05-31 00:50:42 +00:00
Evan Cheng 6821bbb1ea Update vector extract test cases.
llvm-svn: 28580
2006-05-31 00:48:09 +00:00
Evan Cheng ddced95d8f A new entry
llvm-svn: 28579
2006-05-30 23:56:31 +00:00
Evan Cheng 57399704b3 MAXP{D|S} and MINP{D|S} are commutable.
llvm-svn: 28578
2006-05-30 23:47:30 +00:00
Evan Cheng c0f90bef47 Commute shufps / shufpd.
llvm-svn: 28577
2006-05-30 23:34:30 +00:00
Reid Spencer b5ddb66ce0 1. No need to thwart this test with an environment variable. Turning it off
is the default and handled by the makefile system and runtest
2. Redirect stderr of llvm-as and llvm2cpp so that warning messages about
   instrinsics don't cause Tcl to report the run as failed.

llvm-svn: 28576
2006-05-30 23:07:17 +00:00
Evan Cheng f21045a5cd Somehow I lost a condition when I was shuffling some code around. Anyway,
only transform a shufps to pshufd when the first two operands are the same.

llvm-svn: 28575
2006-05-30 22:13:36 +00:00
Evan Cheng c8c172eaae Fix a build breaker.
llvm-svn: 28574
2006-05-30 21:45:53 +00:00
Evan Cheng a4fc5b8699 Oops. PSHUFD is only available with SSE2.
llvm-svn: 28573
2006-05-30 21:30:59 +00:00
Chris Lattner a5d4587296 Add a note
llvm-svn: 28572
2006-05-30 21:29:15 +00:00
Chris Lattner b9342afa56 Always reserve space for 8 spilled GPRs. GCC apparently assumes that this
space will be available, even if the callee isn't varargs.

llvm-svn: 28571
2006-05-30 21:21:04 +00:00
Reid Spencer c3b9370754 llvm2cpp is ready to be compiled so add it to the makefile, but make it
optional.

llvm-svn: 28570
2006-05-30 21:20:55 +00:00
Reid Spencer ad7da68343 Add the new command line arguments.
llvm-svn: 28569
2006-05-30 21:19:29 +00:00
Reid Spencer 30945ca559 Fix more bugs. This version now passes all of the Feature test except for
a floating point conversion problem with NAN in intrinsics.ll. llvm2cpp
makes an attempt to provide the correct conversion, based on the
implementation in the CBackend writer, but it doesn't get NAN correct.

llvm-svn: 28568
2006-05-30 21:18:23 +00:00
Evan Cheng baace007a3 Added a test case for parameter passing of vector values.
llvm-svn: 28567
2006-05-30 20:37:00 +00:00
Evan Cheng 1ec0e4233e Fix the test failure on non-Darwin targets.
llvm-svn: 28566
2006-05-30 20:35:46 +00:00
Evan Cheng 66f849bd7b Allow shufps x, x, mask to be converted to pshufd x, mask to save a move.
llvm-svn: 28565
2006-05-30 20:26:50 +00:00
Evan Cheng b33e54ead7 Remove bogus comment.
llvm-svn: 28564
2006-05-30 20:24:48 +00:00
Reid Spencer f9aa688f5b Add llvm2cpp program.
llvm-svn: 28563
2006-05-30 19:56:31 +00:00
Reid Spencer 82ebabafde Provide a simpler interface for getting a ConstantArray from a character
string. Instead of specifying the length, just specify whether the user
wants a terminating null or not. The default is "true" to retain the same
behavior as previously provided by this function.

llvm-svn: 28562
2006-05-30 18:15:07 +00:00
Evan Cheng d12c97d23a Make sure the register pressure reduction schedulers work for non-uniform
latency targets, e.g. PPC32.

llvm-svn: 28561
2006-05-30 18:05:39 +00:00
Evan Cheng 61e9f0d680 When a priority_queue is empty, the behavior of top() operator is
non-deterministic. Returns NULL when it's empty!

llvm-svn: 28560
2006-05-30 18:04:34 +00:00
Rafael Espindola 5bc60da112 Expand ret into "CopyToReg;BRIND"
llvm-svn: 28559
2006-05-30 17:33:19 +00:00
Chris Lattner 11c25cfa13 Enable -fno-use-cxa-atexit on darwin/ppc also.
llvm-svn: 28558
2006-05-30 16:38:06 +00:00
Reid Spencer b0cffafb59 Undo a patch that breaks llvm-as because the warning message is written to
stdout when the output of llvm-as is also written to stdout. We'll have to
fix tcl some other way.

llvm-svn: 28557
2006-05-30 16:34:59 +00:00
Reid Spencer b5c64fc3b6 Write the WARNING message to cout instead of cerr. Writing to cerr causes
Tcl to claim that the program had an error and thus produces errors in
the dejagnu testing when its really just a warning.

llvm-svn: 28556
2006-05-30 16:05:59 +00:00
Vladimir Prus 413441b9c9 Make doc comment visible in doxygen output. Clarify Type construction.
llvm-svn: 28555
2006-05-30 15:49:30 +00:00
Reid Spencer 37cf1765ed Fix many small bugs in llvm2cpp. This patch gets llvm2cpp working with
everything except PHI nodes and one odd recurse/opaque type situation.
PHI nodes forward reference INSTRUCTIONS (aaaaaaaahhhhhhh!) :)

llvm-svn: 28554
2006-05-30 10:21:41 +00:00
Reid Spencer 84504f215b Properly document the second form of ConstArray::get()
llvm-svn: 28553
2006-05-30 08:26:13 +00:00
Reid Spencer 6f61453ff8 Adjust the interface to ConstantArray::get. The previous
implementation always added a null byte to the end of the string. It turns
out that this is not always wanted. By adding a length parameter we preserve
this behavior when length==0 (default value) but also allow other lengths
(not null terminated) to be created.

llvm-svn: 28552
2006-05-30 08:23:18 +00:00
Evan Cheng 02420144ab Add a note about integer multiplication by constants.
llvm-svn: 28551
2006-05-30 07:37:37 +00:00
Evan Cheng 734e1e241b A addressing mode folding enhancement:
Fold c2 in (x << c1) | c2 where (c2 < c1)
e.g.
int test(int x) {
  return (x << 3) + 7;
}

This can be codegen'd as:
leal 7(,%eax,8), %eax

llvm-svn: 28550
2006-05-30 06:59:36 +00:00
Evan Cheng 905e88caaa Add a lea instruction selection test case.
llvm-svn: 28549
2006-05-30 06:53:55 +00:00
Evan Cheng 749138582e Some new entries about truncate / anyext
llvm-svn: 28548
2006-05-30 06:23:50 +00:00
Reid Spencer c267698f23 First complete version of llvm2cpp that doesn't crash on any of the Feature
tests. The output in a few cases still doesn't compile, however.

llvm-svn: 28547
2006-05-30 03:43:49 +00:00
Andrew Lenharth c3cff44880 move calltarget to dsa
llvm-svn: 28546
2006-05-29 23:39:48 +00:00
Andrew Lenharth e465381401 Since there was interest on the mailing list, this is a utility pass that
uses DSA to make find targets of calls.  It provides a very convinient
interface to DSA results to do things with indirect calls, such as
write a devirtualizer (which I have and may commit one of these days).

llvm-svn: 28545
2006-05-29 22:58:38 +00:00
Chris Lattner d65928994a Silly cleanup
llvm-svn: 28544
2006-05-29 18:52:52 +00:00
Chris Lattner dc0b3afcdd silly cleanup
llvm-svn: 28543
2006-05-29 18:52:05 +00:00
Reid Spencer 131ca2869c Fix a bug with diffing the wrong files. Make output more readable.
llvm-svn: 28542
2006-05-29 18:09:38 +00:00
Reid Spencer c194e7e843 Next batch of implementation:
1. Get rid of old AsmWriter cruft that's not needed.
2. Implement several instructions. Enough to get by globalvars.ll and
   alignment.ll in the Feature test suite.
3. Handle constants properly (don't repeat definitions).
4. Make the output compatible with llvm-dis for diff purposes.

llvm-svn: 28541
2006-05-29 18:08:06 +00:00
Reid Spencer 4967fa722d Fix file header comment.
llvm-svn: 28540
2006-05-29 18:06:28 +00:00
Reid Spencer fe361813aa Remove temporary testing rules.
llvm-svn: 28539
2006-05-29 18:05:59 +00:00
Vladimir Prus 5150b763ce Reset DEBUG_SYMBOL_TABLE back to 0.
llvm-svn: 28538
2006-05-29 12:54:52 +00:00
Vladimir Prus f10bbd052d Fix compile error when DEBUG_SYMBOL_TABLE is defined.
llvm-svn: 28537
2006-05-29 12:45:15 +00:00
Reid Spencer 84f1a3e639 Fix two bugs in the CppWriter.cpp:
1. Return the module from the MakeModule function so it can be verified.
2. Make sure types get generated with their names

llvm-svn: 28536
2006-05-29 02:58:15 +00:00
Reid Spencer 7af13d73f7 Fix the file's comment block.
llvm-svn: 28535
2006-05-29 02:35:29 +00:00
Reid Spencer edcf47ea45 Fix grammar in a comment.
llvm-svn: 28534
2006-05-29 02:34:34 +00:00
Reid Spencer 5531640f4b Replace an old C-style cast with a C++ cast (squelch warning)
llvm-svn: 28533
2006-05-29 02:32:43 +00:00
Reid Spencer d078a1fd4c Correct some grammar and describe current reality.
llvm-svn: 28532
2006-05-29 02:31:47 +00:00
Owen Anderson a03aa6f37f Require both tests to pass.
llvm-svn: 28531
2006-05-29 01:28:16 +00:00
Owen Anderson c01e7bde40 Update the testcase to check the full extent of LCSSA.
llvm-svn: 28530
2006-05-29 01:07:04 +00:00
Owen Anderson 8a8f278f15 Add Use replacement. Assuming there is nothing horribly wrong with this, LCSSA
is now theoretically feature-complete.  It has not, however, been thoroughly
test, and is still considered experimental.

llvm-svn: 28529
2006-05-29 01:00:00 +00:00
Reid Spencer 58d5dce11d Initial Commit of llvm2cpp
This is a safekeeping commit. The program is not finished. It currently
handles modules, types, global variables and function declarations. Blocks
and instructions remain to be done.

llvm-svn: 28528
2006-05-29 00:57:22 +00:00
Owen Anderson 152d063ccb Major think-o. Iterate over all live out-of-loop values, and perform the
other calculations on each individually, rather than trying to delay it and do
them all at the end.

llvm-svn: 28527
2006-05-28 19:33:28 +00:00
Reid Spencer 497120ef63 Fix a problem where dejagnu won't accept the value of global tcl variable
"libdir" for some reason. Changing to llvmlibsdir instead fixes it.

llvm-svn: 28526
2006-05-28 07:22:42 +00:00
Reid Spencer 633301acea Provide an infrastructure for testing the llvm2cpp program (yet to be
committed). This infrastructure is only activated when RUNLLVM2CPP=1 is
specified on the make command line. Currently it is only supported in the
Feature test suite.

llvm-svn: 28525
2006-05-28 04:21:40 +00:00
Owen Anderson 1310e42803 Make LCSSA insert proper Phi nodes throughout the rest of the CFG by computing
the iterated Dominance Frontier of the loop-closure Phi's.  This is the
second phase of the LCSSA pass.  The third phase (coming soon) will be to
update all uses of loop variables to use the loop-closure Phi's instead.

llvm-svn: 28524
2006-05-27 18:47:11 +00:00
Chris Lattner 67c424e010 Fix some regression from the inliner patch I committed last night. This fixes
ldecod, lencod, and SPASS.

llvm-svn: 28523
2006-05-27 17:28:13 +00:00
Chris Lattner 6b852931a0 Fix pastos in comments
llvm-svn: 28522
2006-05-27 06:57:55 +00:00
Chris Lattner be853d77e9 Switch the inliner over to using CloneAndPruneFunctionInto. This effectively
makes it so that it constant folds instructions on the fly.  This is good
for several reasons:

0. Many instructions are constant foldable after inlining, particularly if
   inlining a call with constant arguments.
1. Without this, the inliner has to allocate memory for all of the instructions
   that can be constant folded, then a subsequent pass has to delete them.  This
   gets the job done without this extra work.
2. This makes the inliner *pass* a bit more aggressive: in particular, it
   partially solves a phase order issue where the inliner would inline lots
   of code that folds away to nothing, but think that the resultant function
   is big because of this code that will be gone.  Now the code never exists.

This is the first part of a 2-step process.  The second part will be smart
enough to see when this implicit constant folding propagates a constant into
a branch or switch instruction, making CFG edges dead.

This implements Transforms/Inline/inline_constprop.ll

llvm-svn: 28521
2006-05-27 01:28:04 +00:00
Chris Lattner 64d8692dee Ignore generated files
llvm-svn: 28520
2006-05-27 01:23:30 +00:00
Chris Lattner 3df13f4f22 Implement a new method, CloneAndPruneFunctionInto, as documented.
llvm-svn: 28519
2006-05-27 01:22:24 +00:00
Chris Lattner 53e97ebd2c Implement a new method: CloneAndPruneFunctionInto, as documented.
llvm-svn: 28518
2006-05-27 01:21:50 +00:00
Chris Lattner bc3c879fcf Refactor some code to expose an interface to constant fold and instruction given it's opcode, typeand operands.
llvm-svn: 28517
2006-05-27 01:18:04 +00:00
Chris Lattner 00bb5f8248 Add an interface to constant fold and instruction given it's opcode, type
and operands.

llvm-svn: 28516
2006-05-27 01:17:40 +00:00
Chris Lattner 1a2a101359 New testcase: check that the inliner constant folds instructions on the
fly if it can.

llvm-svn: 28515
2006-05-27 01:16:22 +00:00
Chris Lattner 8f872d2091 Fix a nasty dag combiner bug that caused nondeterminstic crashes (MY FAVORITE!):
SimplifySelectOps would eliminate a Select, delete it, then return true.

The clients would see that it did something and return null.

The top level would see a null return, and decide that nothing happened,
proceeding to process the node in other ways: boom.

The fix is simple: clients of SimplifySelectOps should return the select
node itself.

In order to catch really obnoxious boogs like this in the future, add an
assert that nodes are not deleted.  We do this by checking for a sentry node
type that the SDNode dtor sets when a node is destroyed.

llvm-svn: 28514
2006-05-27 00:43:02 +00:00
Chris Lattner 608290c299 Add a new sentry node type, allowing assertions to catch trivial
use-after-deleted errors.

llvm-svn: 28513
2006-05-27 00:40:15 +00:00
Owen Anderson b4e16996f1 A few small clean-ups, and the addition of an LCSSA statistic.
llvm-svn: 28512
2006-05-27 00:31:37 +00:00
Evan Cheng 21dee4e0b2 Make CALL node consistent with RET node. Signness of value has type MVT::i32
instead of MVT::i1. Either is fine except MVT::i32 is probably a legal type
for most (if not all) platforms while MVT::i1 is not.

llvm-svn: 28511
2006-05-26 23:13:20 +00:00
Evan Cheng a3add0fea8 Change RET node to include signness information of the return values. i.e.
RET chain, value1, sign1, value2, sign2, ...

llvm-svn: 28510
2006-05-26 23:10:12 +00:00
Evan Cheng a2e9953c54 Change RET node to include signness information of the return values. e.g.
RET chain, value1, sign1, value2, sign2

llvm-svn: 28509
2006-05-26 23:09:09 +00:00
Owen Anderson f980a7478f Trivial testcase that LCSSA can already handle.
llvm-svn: 28508
2006-05-26 21:59:20 +00:00
Owen Anderson 6e047ab8fc Fix a copy-and-paste-o that would break some compilers.
llvm-svn: 28507
2006-05-26 21:19:17 +00:00
Owen Anderson f3dd3e2bfd Clean up and refactor LCSSA a bunch. It should also run faster now, though
there's still a lot of work to be done on it.

llvm-svn: 28506
2006-05-26 21:11:53 +00:00
Evan Cheng b92f418408 Vector argument must be passed in memory location aligned on 16-byte boundary.
llvm-svn: 28505
2006-05-26 20:37:47 +00:00