llvm-project/clang/lib/CodeGen
Chris Lattner 63d06ab65a teach codegen to handle noop casts as lvalues.
llvm-svn: 67164
2009-03-18 04:02:57 +00:00
..
ABIInfo.h Pull CodeGenFunction::EmitVAArg into target specific ABIInfo classes. 2009-02-10 21:44:36 +00:00
CGBlocks.cpp Do up codegen for function static data and externs in functions in block 2009-03-13 23:34:28 +00:00
CGBlocks.h Codegen support for copy helpers for block literals. 2009-03-07 02:35:30 +00:00
CGBuilder.h Disable generation of basic block names in NDEBUG mode. 2008-11-12 00:01:12 +00:00
CGBuiltin.cpp Support IRgen of sqrt -> llvm.sqrt, pow -> llvm.pow. 2009-02-16 22:43:43 +00:00
CGCXX.cpp Pull COdeGenFunction::CreateStaticBlockVarDecl (just for creating the 2009-02-25 19:24:29 +00:00
CGCall.cpp x86_32 ABI: Don't try and expand structures with bitfields. 2009-03-11 22:05:26 +00:00
CGCall.h Unbreak CGFunctionInfo::Profile method and reenable caching of ABI 2009-02-05 00:00:23 +00:00
CGDebugInfo.cpp Fix struct field's debug info. 2009-03-16 23:47:53 +00:00
CGDebugInfo.h Add support to emit debug info for objective-c interfaces. 2009-02-26 21:10:26 +00:00
CGDecl.cpp Mangle param names with .arg. only. 2009-03-07 18:09:52 +00:00
CGExpr.cpp teach codegen to handle noop casts as lvalues. 2009-03-18 04:02:57 +00:00
CGExprAgg.cpp brain thinking memcpy, fingers thinking memset :) 2009-02-28 18:31:01 +00:00
CGExprComplex.cpp Minor cleanup for choose expressions: add a helper that returns the 2009-03-04 05:52:32 +00:00
CGExprConstant.cpp Return 0 if the ConstExprEmitter can't handle an expression. 2009-03-03 16:43:34 +00:00
CGExprScalar.cpp Almost complete implementation of rvalue references. One bug, and a few unclear areas. Maybe Doug can shed some light on some of the fixmes. 2009-03-16 23:22:08 +00:00
CGObjC.cpp ir-gen support for class getter/setter call 2009-03-10 18:03:11 +00:00
CGObjCGNU.cpp Obscure code gen bug related to sending 2009-02-28 20:07:56 +00:00
CGObjCMac.cpp Remove a FIXME. 2009-03-15 06:05:20 +00:00
CGObjCRuntime.h Obscure code gen bug related to sending 2009-02-28 20:07:56 +00:00
CGStmt.cpp introduce a new -fheinous-gnu-extensions flag that enables really 2009-03-13 17:38:01 +00:00
CGValue.h Handle case of none gc'able objects regardless of their 2009-02-21 00:30:43 +00:00
CMakeLists.txt Fix cmake builds. 2009-02-13 15:42:50 +00:00
CodeGenFunction.cpp Initialize the cleanup.dst variable if necessary. Fixes PR3789. 2009-03-17 05:53:35 +00:00
CodeGenFunction.h Do up codegen for function static data and externs in functions in block 2009-03-13 23:34:28 +00:00
CodeGenModule.cpp Adjust for linkage name change. 2009-03-11 20:15:27 +00:00
CodeGenModule.h Backout r66408, we don't want handling of globals to rely on the 2009-03-09 23:53:08 +00:00
CodeGenTypes.cpp Almost complete implementation of rvalue references. One bug, and a few unclear areas. Maybe Doug can shed some light on some of the fixmes. 2009-03-16 23:22:08 +00:00
CodeGenTypes.h Initial implementation of CodeGen for incomplete function types; fixes 2009-03-05 03:16:41 +00:00
Makefile Build system changes to use TableGen to generate the various 2009-03-16 23:06:59 +00:00
Mangle.cpp Almost complete implementation of rvalue references. One bug, and a few unclear areas. Maybe Doug can shed some light on some of the fixmes. 2009-03-16 23:22:08 +00:00
Mangle.h Address Chris's comments regarding C++ name mangling. 2009-02-18 23:53:56 +00:00
ModuleBuilder.cpp Remove some redundant Decl -> Decl castings. 2009-02-17 20:23:54 +00:00
README.txt Add note on theoretical IRgen improvement. 2009-03-15 06:39:56 +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.

NOTE: The main case we care about this for is for -O0 -g compile time
performance, and in that scenario we will need to emit the alloca
anyway currently to emit proper debug info. So this is blocked by
being able to emit debug information which refers to an LLVM
temporary, not an alloca.

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

We should try and avoid generating basic blocks which only contain
jumps. At -O0, this penalizes us all the way from IRgen (malloc &
instruction overhead), all the way down through code generation and
assembly time.

On 176.gcc:expr.ll, it looks like over 12% of basic blocks are just
direct branches!

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

There are some more places where we could avoid generating unreachable code. For
example:
  void f0(int a) { abort(); if (a) printf("hi"); }
still generates a call to printf. This doesn't occur much in real
code, but would still be nice to clean up.

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

Deferred generation of statics incurs some additional
overhead. Currently it is even possible to construct test cases with
O(N^2) behavior! For at least simple cases where we can tell a global
is used, it is probably not worth deferring it. This doesn't solve the
O(N^2) cases, ,though...

PR3810

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