255 Commits

Author SHA1 Message Date
Daniel Dunbar
b98d1f7140 x86_64 ABI: Qualified id types are passed as pointers.
- <rdar://problem/6622451> Bad x86_64 code gen for message call taking one argument.

llvm-svn: 65510
2009-02-26 07:21:35 +00:00
Daniel Dunbar
4208835eec Temporarily disable clearing of insert point (to indicate unreachable
code) when calling noreturn functions; general expression emission
isn't ready to do the right thing in all cases.

llvm-svn: 65473
2009-02-25 20:59:29 +00:00
Daniel Dunbar
1234749853 Add low level support for generating invoke instead of calls.
- No functionality change.

llvm-svn: 65325
2009-02-23 17:26:39 +00:00
Daniel Dunbar
bb525c3c7f x86_64 ABI: Actually, we can always pass things we want to pass in
memory using Indirect; this was a holdover from when CGCall wasn't as
robust.

llvm-svn: 65278
2009-02-22 08:17:51 +00:00
Daniel Dunbar
075d642e24 x86_64 ABI: Make sure to pass vectors that we want to pass in memory
as byval. Otherwise LLVM will have its own opinion about where to put
things.

We now pass all gcc dg.compat tests on x86_64.

llvm-svn: 65266
2009-02-22 07:22:25 +00:00
Daniel Dunbar
c64c481d18 x86_64 ABI: Pass 32-bit vectors as Integer to match gcc. We don't care
about these much but <2 x i16> shows up in the gcc test suite.

llvm-svn: 65264
2009-02-22 04:48:22 +00:00
Daniel Dunbar
f25afad9e5 x86_64 ABI: Classify <1 x i64> as INTEGER (match gcc not llvm-gcc).
Also, make sure to pass <1 x i64> as i64 (not <1 x i64>, which doesn't
quite work yet in the backend).

llvm-svn: 65262
2009-02-22 04:16:10 +00:00
Daniel Dunbar
5006f4a5f0 Take advantage of noreturn attribute to add unreachable instruction &
clear insertion point. The rest of IRgen should theoretically take
advantage of this to avoid emitting dead code. Theory != Practice.

llvm-svn: 65141
2009-02-20 18:54:31 +00:00
Daniel Dunbar
cdbb5e336d Set call attribute for direct calls (i.e. noreturn).
- Remove an unused variant of EmitCallExpr overload.

llvm-svn: 65130
2009-02-20 18:06:48 +00:00
Chris Lattner
90669d0500 switch ObjCMethodDecl's parameter list from being explicitly managed to an ObjCList.
llvm-svn: 65114
2009-02-20 06:23:21 +00:00
Daniel Dunbar
8cdb9dae45 i386 ABI: Offset computation in va_arg was incorrect for sizeof(Ty)>4.
We are down to only failing gcc.dg/compat/vector-[12] (8 tests total).

llvm-svn: 64967
2009-02-18 22:28:45 +00:00
Daniel Dunbar
e60ec0abca x86_64 ABI: Fix thinko in computation of bound for "passed in SSE regs" test.
Two more gcc/x86_64 failures down.

llvm-svn: 64963
2009-02-18 22:19:44 +00:00
Daniel Dunbar
e3bba6e3d0 x86_64 ABI: "is passed in regs" computation for va_arg was broken for
things passed in mixed registers.

This knocks out 8 x86_64 failures.

llvm-svn: 64958
2009-02-18 22:05:01 +00:00
Daniel Dunbar
617e89231d x86_64 ABI: Two bug fixes.
1. Return of _Complex long double used wrong type.

2. va_arg of types passed in two SSE registers didn't account for
extra space in register save area.

Down to 18 failures on gcc/compat/x86_64. Combined 32/64 results are:
--
		=== gcc Summary ===

# of expected passes		1292
# of unexpected failures	34
# of unsupported tests		2
--

llvm-svn: 64880
2009-02-18 03:44:19 +00:00
Daniel Dunbar
0ee13255ee x86_64 ABI: Fix assert on return of _Complex long double.
llvm-svn: 64756
2009-02-17 07:55:55 +00:00
Daniel Dunbar
7f0e2368bb x86_64 ABI: Implement classification for bit-fields.
llvm-svn: 64727
2009-02-17 02:45:44 +00:00
Daniel Dunbar
f2c6198eef x86_64 ABI: Handle va_arg arguments with alignment > 8.
llvm-svn: 64701
2009-02-16 23:38:56 +00:00
Daniel Dunbar
3d88672f64 x86_64 ABI: Need to use canonical types when comparing against
ASTContext types.

llvm-svn: 64533
2009-02-14 02:45:45 +00:00
Daniel Dunbar
019ef0bbfe x86_64 ABI: Pass simple types directly when possible. This is
important for both keeping the generated LLVM simple and for ensuring
that integer types are passed/promoted correctly.

llvm-svn: 64529
2009-02-14 02:09:24 +00:00
Daniel Dunbar
abe6ef932f x86_64 ABI: Support va_arg passed in mixed registers.
- Now at 1274 passes on gcc compat suite vs 1262.

llvm-svn: 64469
2009-02-13 17:46:31 +00:00
Daniel Dunbar
753cc07d13 x86_64: Initial varargs support.
- Doesn't yet handle case where values are passed in mixed (general
   purpose & floating point) registers; otherwise largely
   functional. Code still needs some cleaning.

Fixes:
MultiSource/Applications/lua/lua
MultiSource/Applications/siod/siod
MultiSource/Applications/sqlite3/sqlite3
SingleSource/Regression/C/PR640
SingleSource/UnitTests/2003-07-09-SignedArgs
SingleSource/UnitTests/2007-03-02-VaCopy

gcc compat test suite results (Darwin x86-32 & -64):
--
# of expected passes		1262
# of unexpected failures	56
# of unresolved testcases	34
# of unsupported tests		2

Compare to: http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20090209/012050.html

llvm-svn: 64370
2009-02-12 09:04:14 +00:00
Daniel Dunbar
2d0746fb97 Pull CodeGenFunction::EmitVAArg into target specific ABIInfo classes.
llvm-svn: 64235
2009-02-10 20:44:09 +00:00
Daniel Dunbar
9403cd6d85 Tweak x86-64 ABI to allow reuse for vararg handling.
llvm-svn: 64221
2009-02-10 17:06:09 +00:00
Daniel Dunbar
9bfb4de38b ABI: Correctly handle load/store of values which have a different LLVM
memory representation (e.g., bool).
 - This upgrades (downgrades) MultiSource/Applications/ClamAV/clamscan
   to a miscompile and fixes
   SingleSource/UnitTests/2003-05-31-CastToBool.

llvm-svn: 64194
2009-02-10 01:51:39 +00:00
Daniel Dunbar
d5f1f55e28 Make sure to initialize local variables, even if they were ignored by
ABI.

llvm-svn: 64187
2009-02-10 00:06:49 +00:00
Daniel Dunbar
ee9e4c274b Set load/store alignment when doing ABI coercions.
- Currently, this is producing poor code, but we prefer correctness
   to performance for now. Eventually we should be able to generally
   avoid having to set the alignment when we control the alignment of
   the alloca.

 - This knocks out 33/1000 failures on my single argument ABI tests,
   down to 22/1000 and 18 of these appear to be gcc bugs. Woot.

llvm-svn: 64001
2009-02-07 02:46:03 +00:00
Daniel Dunbar
5d3dbd64e1 Implement Direct ABIInfo semantics.
- No intended functionality change, this is essentially enabling
   direct passing of complex and aggregate values, which no ABI is
   using.

llvm-svn: 63863
2009-02-05 11:13:54 +00:00
Daniel Dunbar
cea3af4e54 Simplify test for whether we need an alloca to hold an indirect return
value.
 - No functionality change.

llvm-svn: 63859
2009-02-05 09:24:53 +00:00
Daniel Dunbar
747865af0c Implement ABI Indirect sematics for arguments.
- No intended functionality change, all current ABI implementations
   were only using indirect for complex/aggregate types, which were
   being passed indirectly with the Direct ABIInfo kind.

llvm-svn: 63858
2009-02-05 09:16:39 +00:00
Daniel Dunbar
b8b1c679c4 Merge ABIInfo StructRet/ByVal into Indirect.
- No (intended) functionality change, the semantic changes are to come.

llvm-svn: 63850
2009-02-05 08:00:50 +00:00
Daniel Dunbar
c79407fc40 Pull CodeGenFunction::GetUndefRValue() out of EmitUnsupportedRValue.
llvm-svn: 63845
2009-02-05 07:09:07 +00:00
Daniel Dunbar
70245be397 x86-32: Use Ignore to avoid passing empty structs (instead of Expand).
llvm-svn: 63813
2009-02-05 01:50:07 +00:00
Daniel Dunbar
0103574d55 Honor ByVal alignment. Patch by Nate Begeman!
llvm-svn: 63811
2009-02-05 01:31:19 +00:00
Daniel Dunbar
fff09f335d Unbreak CGFunctionInfo::Profile method and reenable caching of ABI
information.

llvm-svn: 63799
2009-02-05 00:00:23 +00:00
Daniel Dunbar
56e7552c73 Add ABIArgInfo::dump()
llvm-svn: 63794
2009-02-04 23:24:38 +00:00
Daniel Dunbar
fb5fdf1f14 Temporarily disable caching of ABI results; this is going horribly
wrong in some cases.

llvm-svn: 63780
2009-02-04 21:36:22 +00:00
Daniel Dunbar
a45bdbbb6a Add asserts that the function signature matches the other arguments provide
to CGCall functions.

llvm-svn: 63775
2009-02-04 21:17:21 +00:00
Daniel Dunbar
6e3b7df125 Handle demotion of coerced arguments (as in void a(x) short x; { ... }).
llvm-svn: 63726
2009-02-04 07:22:24 +00:00
Daniel Dunbar
a33461150d Use ConvertTypeForMem when creating alloca for scalar argument.
llvm-svn: 63681
2009-02-03 23:04:57 +00:00
Daniel Dunbar
fc7c76159c x86_64 ABI: Initial implementation of ABI compliant parameter passing.
- Now only 27/500 failures on ABITest single argument tests; from
   350/500. :)
 - As with return types, a large percentage of these are likely to be
   gcc bugs, not yet reviewed.

Also, fix bug in handling of Ignore ABI type in argument lists.

llvm-svn: 63654
2009-02-03 20:00:13 +00:00
Daniel Dunbar
2f219b0770 ABI handling: Implement coercion for argument types (in addition to
return types).

llvm-svn: 63645
2009-02-03 19:12:28 +00:00
Daniel Dunbar
32931eb21d Change ABIInfo to compute information for a full signature at a time
(the main point of this restructing).

llvm-svn: 63619
2009-02-03 06:51:18 +00:00
Daniel Dunbar
0136282a9c Remove ABIArgInfo::Default kind, ABI is now responsible for specifying
acceptable kind with more precise semantics.

llvm-svn: 63617
2009-02-03 06:30:17 +00:00
Daniel Dunbar
67dace890f Add ABIArgInfo::Direct kind, which passes arguments using whatever the
native IRgen type is. This is like Default, but without any extra
semantics (like automatic tweaking of structures or void).

llvm-svn: 63615
2009-02-03 06:17:37 +00:00
Daniel Dunbar
5a0acdc982 Add two FIXMEs.
llvm-svn: 63613
2009-02-03 06:02:10 +00:00
Daniel Dunbar
b52d077d8b Always use CGFunctionInfo to access ABI information.
llvm-svn: 63612
2009-02-03 05:59:18 +00:00
Daniel Dunbar
313321ea23 Move ABIArgInfo into CGFunctionInfo, computed on creation.
- Still have to convert some consumers over.

llvm-svn: 63610
2009-02-03 05:31:23 +00:00
Daniel Dunbar
6d6b0d309a Move ABIInfo/ABIArgInfo classes into ABIInfo.h
llvm-svn: 63586
2009-02-03 01:05:53 +00:00
Daniel Dunbar
e0be82956b Memoize CGFunctionInfo construction.
llvm-svn: 63576
2009-02-03 00:07:12 +00:00
Daniel Dunbar
3668cb2d3c Change CGFunctionInfo args iterator to not include the return type.
llvm-svn: 63571
2009-02-02 23:43:58 +00:00