RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 19:23:23 +01:00
Code Issues Projects Releases Wiki Activity
64,726 Commits 4 Branches 5 Tags 704 MiB
ea0a5aee7b
Commit Graph

64726 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Owen Anderson
c0664eb99a Lower the unrolling theshold to 150. Empirical tests indicate that this is a sweet spot in the performance per 
code size increase curve.

llvm-svn: 113595
 2010-09-10 17:57:00 +00:00
Dale Johannesen
545bd92baf x86mmx is 64 bits. 
llvm-svn: 113594
 2010-09-10 17:51:47 +00:00
Jim Grosbach
bee860c7c2 add a comment explicitly calling out that allocation orders may include 
reserved regs and that register allocators need to explicitly check for
them.

llvm-svn: 113593
 2010-09-10 17:51:34 +00:00
Bill Wendling
bbdc6c49bf Mark the sse_load_f32 and sse_load_f64 load patterns as having memoperands so 
that the memoperands are properly set after DAG building and general mucking
about.

llvm-svn: 113585
 2010-09-10 10:34:22 +00:00
Bill Wendling
6cc7baa725 Reword since this may not be a bug but intended behavior. 
llvm-svn: 113584
 2010-09-10 10:31:11 +00:00
NAKAMURA Takumi
9c328040a6 Makefile.rules: LOADABLE_MODULE/Win32: "all" components may be linked 
only when ENABLE_SHARED=1.

Loadable module for Win32 requires all symbols resolved for linking.

llvm-svn: 113579
 2010-09-10 06:27:02 +00:00
NAKAMURA Takumi
7857655635 Don't build redundant libLLVMgold.a. 
Building archive would be executed due to definition of BUILD_ARCHIVE,
even if BUILD_ARCHIVE = "0".

llvm-svn: 113578
 2010-09-10 06:26:57 +00:00
Evan Cheng
e2425602c5 Fix test so it passes on non-Darwin hosts. 
llvm-svn: 113577
 2010-09-10 06:20:01 +00:00
Bob Wilson
01b7bf3510 Fix merging base-updates for VLDM/VSTM: Before I switched these instructions 
to use AddrMode4, there was a count of the registers stored in one of the
operands.  I changed that to just count the operands but forgot to adjust for
the size of D registers.  This was noticed by Evan as a performance problem
but it is a potential correctness bug as well, since it is possible that this
could merge a base update with a non-matching immediate.

llvm-svn: 113576
 2010-09-10 05:15:04 +00:00
Evan Cheng
c9cb37516d Teach if-converter to be more careful with predicating instructions that would 
take multiple cycles to decode.
For the current if-converter clients (actually only ARM), the instructions that
are predicated on false are not nops. They would still take machine cycles to
decode. Micro-coded instructions such as LDM / STM can potentially take multiple
cycles to decode. If-converter should take treat them as non-micro-coded
simple instructions.

llvm-svn: 113570
 2010-09-10 01:29:16 +00:00
Eric Christopher
54e138031e Fix build error. 
llvm-svn: 113566
 2010-09-10 00:35:09 +00:00
Eric Christopher
92506682eb Update comments, reorganize some code, rename variables to be 
more clear.  No functional change.

llvm-svn: 113565
 2010-09-10 00:34:35 +00:00
Eric Christopher
06f4af2ef4 64-bit fp loads can come straight out of the constant pool, not as 
bad as I'd thought.

llvm-svn: 113561
 2010-09-09 23:50:00 +00:00
Dan Gohman
a9936d3f97 Discard metadata produced by LLVM 2.7. The value enumeration it used 
is different from what the code now uses in a two ways: NamedMDNodes
were considered Values and included in the numbering, and the
function-local metadata counter wasn't reset between functions.

The later problem breaks lazy deserialization, so instead of trying
to emulate the old numbering, just drop the old metadata. The only
in-tree use case is debug info with LTO, where the QOI loss is
considered acceptable.

llvm-svn: 113557
 2010-09-09 23:12:39 +00:00
Daniel Dunbar
547563df12 llvm-mc: Don't crash when using -n and we see a directive before the initial section. 
- This is annoying, because we have to scatter this check everywhere that could emit real data, but I see no better solution.

llvm-svn: 113552
 2010-09-09 22:42:59 +00:00
Daniel Dunbar
07c0ca2a51 llvm-mc: Make sure we exit != 0 if any errors are encountered. 
llvm-svn: 113551
 2010-09-09 22:42:56 +00:00
Jakob Stoklund Olesen
79f7387096 XFAIL test under valgrind. It is not really our problem if sh is leaking. 
llvm-svn: 113550
 2010-09-09 22:02:13 +00:00
Eric Christopher
3c86728bd6 SIToFP and FPToSI conversions work only on fp-reg to fp-reg. Move 
some data around and implement a couple of move routines to do this.

llvm-svn: 113546
 2010-09-09 21:44:45 +00:00
Dan Gohman
3c81335a10 MDNodes are not Constants. 
llvm-svn: 113539
 2010-09-09 20:54:24 +00:00
Dan Gohman
2a0bee3d07 Print invalid metadata references as <badref>, for consistency with 
regular value references.

llvm-svn: 113538
 2010-09-09 20:53:58 +00:00
Eric Christopher
d5f271d31e New "move to fp reg" routine. Use it. 
llvm-svn: 113537
 2010-09-09 20:49:25 +00:00
Eric Christopher
7208671884 "Strike that, reverse it." -- Mr. Wonka. 
Truncate when truncating, extend when extending.

llvm-svn: 113536
 2010-09-09 20:36:19 +00:00
Owen Anderson
f7276d42d3 What the loop unroller cares about, rather than just not unrolling loops with calls, is 
not unrolling loops that contain calls that would be better off getting inlined.  This mostly
comes up when an interleaved devirtualization pass has devirtualized a call which the inliner
will inline on a future pass.  Thus, rather than blocking all loops containing calls, add
a metric for "inline candidate calls" and block loops containing those instead.

llvm-svn: 113535
 2010-09-09 20:32:23 +00:00
Eric Christopher
38194e51cc Add FPTrunc, fix some bugs where I forgot to update the value map. 
llvm-svn: 113533
 2010-09-09 20:26:31 +00:00
Owen Anderson
db6a08beef Revert r113439, which relaxed the requirement that loops containing calls cannot be unrolled. After some discussion, 
there seems to be a better way to achieve the same effect.

llvm-svn: 113528
 2010-09-09 20:02:23 +00:00
Owen Anderson
82f5f82ccd r113526 introduced an unintended change to the loop unrolling threshold. Revert it. 
llvm-svn: 113527
 2010-09-09 19:11:57 +00:00
Owen Anderson
46189436ea Fix typo in code to cap the loop code size reduction calculation. 
llvm-svn: 113526
 2010-09-09 19:08:59 +00:00
Owen Anderson
a182e2b20b Use code-size reduction metrics to estimate the amount of savings we'll get when we unroll a loop. 
Next step is to recalculate the threshold values given this new heuristic.

llvm-svn: 113525
 2010-09-09 19:07:31 +00:00
Eric Christopher
2ba28d3889 Basic FP->Int, Int->FP conversions. 
llvm-svn: 113523
 2010-09-09 18:54:59 +00:00
Bruno Cardoso Lopes
49efee5c95 Add one more pattern to fallback movddup 
llvm-svn: 113522
 2010-09-09 18:48:34 +00:00
Dan Gohman
7fbaed00bf Extend the getDependence query with support for PHI translation. 
llvm-svn: 113521
 2010-09-09 18:37:31 +00:00
Dan Gohman
f24287edb1 Add a getPointerOperand() helper function to VAArgInst, for consistency 
with LoadInst and StoreInst.

llvm-svn: 113520
 2010-09-09 18:32:40 +00:00
Evan Cheng
b79a03a3ca For each instruction itinerary class, specify the number of micro-ops each 
instruction in the class would be decoded to. Or zero if the number of
uOPs must be determined dynamically.

This will be used to determine the cost-effectiveness of predicating a
micro-coded instruction.

llvm-svn: 113513
 2010-09-09 18:18:55 +00:00
Dan Gohman
e1fb5b5738 Fix this comment. 
llvm-svn: 113511
 2010-09-09 18:04:22 +00:00
Benjamin Kramer
7ca791ef4f MCELF: Write relocation fragments in the right endian. 
- This code is gross, but does the job for now.

llvm-svn: 113509
 2010-09-09 18:01:29 +00:00
Roman Divacky
536c4ab8bd Make ELF OS ABI dependent on the OS from target triple. 
llvm-svn: 113508
 2010-09-09 17:57:50 +00:00
Daniel Dunbar
59aca55047 build: Allow subdir Makefiles to provide an alternate location for the SRCDIR 
Makefile, which can be used to allow building out of tree sources.

llvm-svn: 113503
 2010-09-09 17:38:11 +00:00
Dale Johannesen
de53df20d6 Move remaining MMX instructions from SSE to MMX. 
llvm-svn: 113501
 2010-09-09 17:13:07 +00:00
Owen Anderson
9a194c80c4 Refactor code-size reduction estimation methods out of InlineCostAnalyzer and into CodeMetrics. They 
don't use any InlineCostAnalyzer state, and are useful for other clients who don't necessarily want to use
all of InlineCostAnalyzer's logic, some of which is fairly inlining-specific.

No intended functionality change.

llvm-svn: 113499
 2010-09-09 16:56:42 +00:00
Daniel Dunbar
8c7fe0e9d0 MC: Give a (lame) hard error if a .org directive would create an unreasonably 
large object file (> 1GB).

llvm-svn: 113494
 2010-09-09 16:23:33 +00:00
Daniel Dunbar
90172ab0b5 tests: XFAIL a handful of tests on the vg_leak builder, so we can get back to 
green.

llvm-svn: 113491
 2010-09-09 15:50:19 +00:00
Daniel Dunbar
082043aa74 projects: Don't try to build libcxx if one happens to have it checked out into 
llvm/projects.

llvm-svn: 113490
 2010-09-09 15:49:32 +00:00
Benjamin Kramer
8849104392 Add an elf-dumper utility. 
- Output format and some of the code stolen from macho-dump.
- Somewhat incomplete and probably buggy.
- Comes with a very basic test.

llvm-svn: 113488
 2010-09-09 15:00:41 +00:00
NAKAMURA Takumi
aa9b257b29 lib/System/Host.cpp: 7bit-ize. Eliminate "TM" and "(R)" in comments. 
llvm-svn: 113486
 2010-09-09 13:30:48 +00:00
Duncan Sands
4a66e8e24d Get rid of the last use of -m64 in FrontendC. This solution 
of checking for either 4 or 8 is not very satisfactory, but
it would catch the original problem (an alignment of 1).

llvm-svn: 113485
 2010-09-09 12:57:29 +00:00
Duncan Sands
3030644248 Another test that uses -m64. Here too it looks like it can be 
removed.  Not that the XTARGET wasn't doing anything since it
does nothing without an accompanying XFAIL.

llvm-svn: 113484
 2010-09-09 12:48:04 +00:00
Duncan Sands
e89d619d23 On i386, llvm-gcc cannot be assumed to support -m64. Since these 
tests pass here (i686-linux and x86-64-linux) without -m64, simply
remove the -m64.

llvm-svn: 113483
 2010-09-09 12:43:44 +00:00
Duncan Sands
5e4622cc87 This script is executable. 
llvm-svn: 113479
 2010-09-09 09:37:54 +00:00
Kalle Raiskila
0f53cff753 Silence compiler warning. 
llvm-svn: 113478
 2010-09-09 07:30:15 +00:00
Bob Wilson
524123343c Fix NEON VLD pseudo instruction itineraries that were incorrectly copied from 
the VST pseudos.  The VLD/VST scheduling still needs work (see pr6722), but
at least we shouldn't confuse the loads with the stores.

llvm-svn: 113473
 2010-09-09 05:40:26 +00:00