Commit Graph

4366 Commits

Author SHA1 Message Date
Peter Collingbourne 422542625b Implement -cl-mad-enable
llvm-svn: 120881
2010-12-04 01:51:33 +00:00
Peter Collingbourne b8d9995c0e Implement -cl-unsafe-math-optimizations
llvm-svn: 120879
2010-12-04 01:51:14 +00:00
Peter Collingbourne 0ba5ac8544 Implement -cl-finite-math-only
llvm-svn: 120878
2010-12-04 01:51:05 +00:00
Bob Wilson 7795599f4b Add support for vmul_p8 Neon intrinsic. Radar 8446141.
llvm-svn: 120812
2010-12-03 17:29:39 +00:00
Bob Wilson 4fa993fc51 Add a separate rightShift flag instead of reusing the existing "poly" variable
to distinguish vsri/vsli.

llvm-svn: 120806
2010-12-03 17:10:22 +00:00
Chris Lattner 5cd8475564 Reflow to a style doug prefers, increasing indentation :-)
llvm-svn: 120746
2010-12-02 22:52:04 +00:00
Chris Lattner c5cc2fb980 fix PR8726 by teaching the aggregate init optimization code to handle
structs with references in them correctly.

llvm-svn: 120722
2010-12-02 18:29:00 +00:00
Fariborz Jahanian 50198098b9 IR Gen. part of API support for __block cxx
objects imported into blocks. //rdar://8594790.
Will have a test case coming (as well as one
sent to llvm test suite).

llvm-svn: 120713
2010-12-02 17:02:11 +00:00
Chris Lattner 27a3631bac Improve codegen for initializer lists to use memset more aggressively
when an initializer is variable (I handled the constant case in a previous
patch).  This has three pieces:

1. Enhance AggValueSlot to have a 'isZeroed' bit to tell CGExprAgg that
   the memory being stored into has previously been memset to zero.
2. Teach CGExprAgg to not emit stores of zero to isZeroed memory.
3. Teach CodeGenFunction::EmitAggExpr to scan initializers to determine
   whether they are profitable to emit a memset + inividual stores vs
   stores for everything.

The heuristic used is that a global has to be more than 16 bytes and
has to be 3/4 zero to be candidate for this xform.  The two testcases
are illustrative of the scenarios this catches.  We now codegen test9 into:

 call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 400, i32 4, i1 false)
 %.array = getelementptr inbounds [100 x i32]* %Arr, i32 0, i32 0
 %tmp = load i32* %X.addr, align 4
 store i32 %tmp, i32* %.array

and test10 into:

  call void @llvm.memset.p0i8.i64(i8* %0, i8 0, i64 392, i32 8, i1 false)
  %tmp = getelementptr inbounds %struct.b* %S, i32 0, i32 0
  %tmp1 = getelementptr inbounds %struct.a* %tmp, i32 0, i32 0
  %tmp2 = load i32* %X.addr, align 4
  store i32 %tmp2, i32* %tmp1, align 4
  %tmp5 = getelementptr inbounds %struct.b* %S, i32 0, i32 3
  %tmp10 = getelementptr inbounds %struct.a* %tmp5, i32 0, i32 4
  %tmp11 = load i32* %X.addr, align 4
  store i32 %tmp11, i32* %tmp10, align 4

Previously we produced 99 stores of zero for test9 and also tons for test10.
This xforms should substantially speed up -O0 builds when it kicks in as well
as reducing code size and optimizer heartburn on insane cases.  This resolves
PR279.

llvm-svn: 120692
2010-12-02 07:07:26 +00:00
Chris Lattner 4fc8fb09d0 attempt to fix a buildbot failure, apparently apache fails to build.
llvm-svn: 120688
2010-12-02 04:27:29 +00:00
Eric Christopher 8a2ee390be Add support for the common and nocommon attributes.
rdar://8560647

llvm-svn: 120650
2010-12-02 02:45:55 +00:00
Chris Lattner e6af88628f Enhance the init generation logic to emit a memset followed by a few stores when
a global is larger than 32 bytes and has fewer than 6 non-zero values in the
initializer.  Previously we'd turn something like this:

char test8(int X) {
  char str[10000] = "abc";

into a 10K global variable which we then memcpy'd from.  Now we generate:

  %str = alloca [10000 x i8], align 16
  %tmp = getelementptr inbounds [10000 x i8]* %str, i64 0, i64 0
  call void @llvm.memset.p0i8.i64(i8* %tmp, i8 0, i64 10000, i32 16, i1 false)
  store i8 97, i8* %tmp, align 16
  %0 = getelementptr [10000 x i8]* %str, i64 0, i64 1
  store i8 98, i8* %0, align 1
  %1 = getelementptr [10000 x i8]* %str, i64 0, i64 2
  store i8 99, i8* %1, align 2

Which is much smaller in space and also likely faster.

This is part of PR279

llvm-svn: 120645
2010-12-02 01:58:41 +00:00
John McCall b7bd14fa08 Simplify the ASTs by consolidating ObjCImplicitGetterSetterExpr and ObjCPropertyRefExpr
into the latter.

llvm-svn: 120643
2010-12-02 01:19:52 +00:00
John McCall f3735e01ce Restore the lvalue-to-rvalue conversion patch with a minimal fix.
llvm-svn: 120555
2010-12-01 04:43:34 +00:00
Chris Lattner b85025fedd add some infrastructure that will let us codegen
http://llvm.org/bugs/show_bug.cgi?id=279#c3 better.

No functionality change yet.

llvm-svn: 120530
2010-12-01 02:05:19 +00:00
Chris Lattner 46a7ad7605 get some i32/i8/i1 constants from Builder, which is much less painful than
using ConstantInt::get.

llvm-svn: 120527
2010-12-01 01:47:15 +00:00
John McCall 5a39bd2443 A CGRecordLayout object persists. Since its contained types may
refer to opaque types, they must be held via PATypeHolders.  I'm
not sure why this hasn't blown up before.

llvm-svn: 120491
2010-11-30 23:21:46 +00:00
John McCall bcd38217f3 CGRecordLayoutBuilder does not need to be exported from this module.
llvm-svn: 120489
2010-11-30 23:17:27 +00:00
Fariborz Jahanian 3a4ea9a76c Declaring local static in global block
literal declaration caused crash in CodeGen.
This patch fixes it. pr8707

llvm-svn: 120486
2010-11-30 23:07:14 +00:00
John McCall 4f22c9de48 L-value to r-value conversion is not ready for prime-time.
llvm-svn: 120433
2010-11-30 17:58:55 +00:00
John McCall 824fd1dcff Introduce an r-value to l-value cast kind. I'm not promising anything
about the reliability of this yet.

llvm-svn: 120422
2010-11-30 10:13:06 +00:00
Douglas Gregor 2f53a0bbf2 Harden CGDebugInfo::CreateType() a bit for cases where we couldn't generate type info for its component types
llvm-svn: 120401
2010-11-30 06:38:09 +00:00
Michael J. Spencer 8aaf49959c Merge System into Support.
llvm-svn: 120297
2010-11-29 18:12:39 +00:00
Anders Carlsson acf877be12 Don't store the maximum alignment, we can trivially compute it.
llvm-svn: 120268
2010-11-28 23:06:23 +00:00
Anders Carlsson a459adb2ff More work on laying out virtual bases.
llvm-svn: 120257
2010-11-28 19:18:44 +00:00
Anders Carlsson 3378d870d2 BuildVirtualCall doesn't need to take a reference to a pointer.
llvm-svn: 120252
2010-11-28 17:53:32 +00:00
Anders Carlsson e8ba473ed2 Make the destructor be the 'vtable anchor' of the CGCXXABI class.
llvm-svn: 120251
2010-11-28 17:50:09 +00:00
Anders Carlsson fdee32f8ed Move CGCXXABI functions to CGCXXABI.cpp
llvm-svn: 120250
2010-11-28 17:49:03 +00:00
Anders Carlsson 81f6f3626e Add a CGCXXABI.cpp file.
llvm-svn: 120249
2010-11-28 17:46:52 +00:00
Benjamin Kramer 90b5b68535 Hide a bunch of symbols.
llvm-svn: 120153
2010-11-25 18:29:30 +00:00
Anders Carlsson 1f95ee3016 Begin work on actually laying out virtual bases.
llvm-svn: 120140
2010-11-25 01:59:35 +00:00
Anders Carlsson 7f95cd1817 Rename RecordLayout::getPrimaryBaseWasVirtual to isPrimaryBaseVirtual.
llvm-svn: 120133
2010-11-24 23:12:57 +00:00
Anders Carlsson cadbc463e2 Move code off the primary base info iterator. No functionality change.
llvm-svn: 120132
2010-11-24 23:04:03 +00:00
Anders Carlsson 4131f00a1f Add CXXRecordDecl::getIndirectPrimaryBases.
llvm-svn: 120129
2010-11-24 22:50:27 +00:00
Anders Carlsson 72b1968d92 Simplify code.
llvm-svn: 120118
2010-11-24 21:53:50 +00:00
Anders Carlsson 0c75b49dc1 Fix a (probably very old) regression where we weren't using the typedef name for anonymous tag types.
llvm-svn: 120113
2010-11-24 21:05:09 +00:00
Anders Carlsson 0e4dfebe63 Revert r120110. We don't want to call ConvertType from within getCGRecordLayout.
llvm-svn: 120112
2010-11-24 20:22:04 +00:00
Anders Carlsson 449bbc5bde Use ConvertType instead of ConvertTagDeclType, since ConvertType will assign a name to the type.
llvm-svn: 120110
2010-11-24 20:05:21 +00:00
Anders Carlsson 6277615cbb Simplify code.
llvm-svn: 120109
2010-11-24 19:57:04 +00:00
Anders Carlsson c5bcaa373b Fix build.
llvm-svn: 120108
2010-11-24 19:52:29 +00:00
Anders Carlsson 63e0f2050f Make CodeGenTypes::getCGRecordLayout compute the layout lazily if it doesn't exist. This matches ASTContext::getASTRecordLayout and is less confusing.
llvm-svn: 120107
2010-11-24 19:51:04 +00:00
Anders Carlsson 36e2fa8209 CGRecordLayout types are always struct types.
llvm-svn: 120106
2010-11-24 19:37:16 +00:00
John McCall 086a464e24 Switch a lot of call-sites over to using the new value-kind calculations.
llvm-svn: 120084
2010-11-24 05:12:34 +00:00
Anders Carlsson 849ea413c6 Refactor the null-initialization for record types and make it handle bases that aren't i8 arrays.
llvm-svn: 119980
2010-11-22 18:42:14 +00:00
Anders Carlsson 39a6b22023 Remove FIXME; we don't ever want to lay out empty bases.
llvm-svn: 119957
2010-11-22 00:03:08 +00:00
Anders Carlsson a7dd96ce77 Rename BaseLLVMType to NonVirtualBaseLLVMType.
llvm-svn: 119956
2010-11-21 23:59:45 +00:00
Anders Carlsson e64fbe2a6c Add getCGRecordLayout helper function. No functionality change.
llvm-svn: 119955
2010-11-21 23:56:06 +00:00
Francois Pichet df946c3702 Fix warning: enumeration value 'IndirectField' not handled in switch.
llvm-svn: 119924
2010-11-21 06:49:41 +00:00
Fariborz Jahanian 44a41d1c25 Minor cleanup. No change otherwise.
llvm-svn: 119814
2010-11-19 18:17:09 +00:00
Fariborz Jahanian 148d113e55 Fix a bug where write-barriers for assignment through reference
types was not being generated for objc pointers.
// rdar://8681766.

llvm-svn: 119751
2010-11-18 22:39:16 +00:00