llvm-project/clang/lib/CodeGen
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
..
ABIInfo.h Merge ABIInfo StructRet/ByVal into Indirect. 2009-02-05 08:00:50 +00:00
CGBuilder.h Disable generation of basic block names in NDEBUG mode. 2008-11-12 00:01:12 +00:00
CGBuiltin.cpp Reapply Daniel's patch to match up with llvm 63765. 2009-02-05 01:50:47 +00:00
CGCXX.cpp Provide a new kind of iterator, the specific_decl_iterator, that 2009-01-09 17:18:27 +00:00
CGCall.cpp Set load/store alignment when doing ABI coercions. 2009-02-07 02:46:03 +00:00
CGCall.h Unbreak CGFunctionInfo::Profile method and reenable caching of ABI 2009-02-05 00:00:23 +00:00
CGDebugInfo.cpp When making dummy file entries, the directory name should also be 2009-02-07 00:40:41 +00:00
CGDebugInfo.h reimplement debug info generation in terms of DebugInfo.h instead of 2008-11-10 06:08:34 +00:00
CGDecl.cpp Follow Eli's advice and store the VLA size with the native size_t type. Fixes PR3491. 2009-02-05 19:43:10 +00:00
CGExpr.cpp Pull CodeGenFunction::GetUndefRValue() out of EmitUnsupportedRValue. 2009-02-05 07:09:07 +00:00
CGExprAgg.cpp Make CodeGen produce an error if we come across a non-constant initializer list that involves the GNU array-range designator extension 2009-01-29 19:42:23 +00:00
CGExprComplex.cpp Introduce a new expression node, ImplicitValueInitExpr, that 2009-01-29 17:44:32 +00:00
CGExprConstant.cpp Fix for PR3447: use padded sizes for computations on struct/union 2009-02-01 08:12:19 +00:00
CGExprScalar.cpp Follow Eli's advice and store the VLA size with the native size_t type. Fixes PR3491. 2009-02-05 19:43:10 +00:00
CGObjC.cpp Use correct signature for calling enumeration mutation function. 2009-02-04 22:00:33 +00:00
CGObjCGNU.cpp Patch fixes messaging for GNU runtime. 2009-02-04 20:31:19 +00:00
CGObjCMac.cpp Fixed an objc2 nonfragile-abi code gen bug. 2009-02-06 23:46:26 +00:00
CGObjCRuntime.h ir-gen for nonfragile ivar bitfield access (objc2 nonfragile abi). 2009-02-03 19:03:09 +00:00
CGStmt.cpp If an input constraint refers to an output constraint, it should have the same constraint info as the output constraint. Fixes PR3417 2009-01-27 20:38:24 +00:00
CGValue.h Remove tabs. 2008-12-16 19:57:09 +00:00
CMakeLists.txt CMake: Builds and installs clang binary and libs (no docs yet). It 2008-10-26 00:56:18 +00:00
CodeGenFunction.cpp Follow Eli's advice and store the VLA size with the native size_t type. Fixes PR3491. 2009-02-05 19:43:10 +00:00
CodeGenFunction.h Pull CodeGenFunction::GetUndefRValue() out of EmitUnsupportedRValue. 2009-02-05 07:09:07 +00:00
CodeGenModule.cpp Use 'compile' instead of 'codegen' when reporting error to user. 2009-02-06 19:18:03 +00:00
CodeGenModule.h More ABI API cleanup. 2009-02-02 22:03:45 +00:00
CodeGenTypes.cpp Thread CGFunctionInfo construction through CodeGenTypes. 2009-02-02 23:23:47 +00:00
CodeGenTypes.h Memoize CGFunctionInfo construction. 2009-02-03 00:07:12 +00:00
Makefile Make a major restructuring of the clang tree: introduce a top-level 2008-03-15 23:59:48 +00:00
ModuleBuilder.cpp Remove ScopedDecl, collapsing all of its functionality into Decl, so 2009-01-20 01:17:11 +00:00
README.txt Mention an optimization opportunity pointed out by Chris. 2008-12-04 09:05:45 +00:00

README.txt

IRgen optimization opportunities.

//===---------------------------------------------------------------------===//

The common pattern of
--
short x; // or char, etc
(x == 10)
--
generates an zext/sext of x which can easily be avoided.

//===---------------------------------------------------------------------===//

Bitfields accesses can be shifted to simplify masking and sign
extension. For example, if the bitfield width is 8 and it is
appropriately aligned then is is a lot shorter to just load the char
directly.

//===---------------------------------------------------------------------===//

It may be worth avoiding creation of alloca's for formal arguments
for the common situation where the argument is never written to or has
its address taken. The idea would be to begin generating code by using
the argument directly and if its address is taken or it is stored to
then generate the alloca and patch up the existing code.

In theory, the same optimization could be a win for block local
variables as long as the declaration dominates all statements in the
block.

//===---------------------------------------------------------------------===//