mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-25 12:12:47 +01:00
cd4b68bee9
priority function. Instead, just iterate over the AllNodes list, which is already in topological order. This eliminates a fair amount of bookkeeping, and speeds up the isel phase by about 15% on many testcases. The impact on most targets is that AddToISelQueue calls can be simply removed. In the x86 target, there are two additional notable changes. The rule-bending AND+SHIFT optimization in MatchAddress that creates new pre-isel nodes during isel is now a little more verbose, but more robust. Instead of either creating an invalid DAG or creating an invalid topological sort, as it has historically done, it can now just insert the new nodes into the node list at a position where they will be consistent with the topological ordering. Also, the address-matching code has logic that checked to see if a node was "already selected". However, when a node is selected, it has all its uses taken away via ReplaceAllUsesWith or equivalent, so it won't recieve any further visits from MatchAddress. This code is now removed. llvm-svn: 58748 |
||
|---|---|---|
| .. | ||
| CMakeLists.txt | ||
| IA64.h | ||
| IA64.td | ||
| IA64AsmPrinter.cpp | ||
| IA64Bundling.cpp | ||
| IA64InstrBuilder.h | ||
| IA64InstrFormats.td | ||
| IA64InstrInfo.cpp | ||
| IA64InstrInfo.h | ||
| IA64InstrInfo.td | ||
| IA64ISelDAGToDAG.cpp | ||
| IA64ISelLowering.cpp | ||
| IA64ISelLowering.h | ||
| IA64MachineFunctionInfo.h | ||
| IA64RegisterInfo.cpp | ||
| IA64RegisterInfo.h | ||
| IA64RegisterInfo.td | ||
| IA64TargetAsmInfo.cpp | ||
| IA64TargetAsmInfo.h | ||
| IA64TargetMachine.cpp | ||
| IA64TargetMachine.h | ||
| Makefile | ||
| README | ||
TODO:
- Un-bitrot ISel
- Hook up If-Conversion a la ARM target
- Hook up all branch analysis functions
- Instruction scheduling
- Bundling
- Dynamic Optimization
- Testing and bugfixing
- stop passing FP args in both FP *and* integer regs when not required
- allocate low (nonstacked) registers more aggressively
- clean up and thoroughly test the isel patterns.
- fix stacked register allocation order: (for readability) we don't want
the out? registers being the first ones used
- fix up floating point
(nb http://gcc.gnu.org/wiki?pagename=ia64%20floating%20point )
- bundling!
(we will avoid the mess that is:
http://gcc.gnu.org/ml/gcc/2003-12/msg00832.html )
- instruction scheduling (hmmmm! ;)
- counted loop support
- make integer + FP mul/div more clever (we have fixed pseudocode atm)
- track and use comparison complements
INFO:
- we are strictly LP64 here, no support for ILP32 on HP-UX. Linux users
don't need to worry about this.
- i have instruction scheduling/bundling pseudocode, that really works
(has been tested, albeit at the perl-script level).
so, before you go write your own, send me an email!
KNOWN DEFECTS AT THE CURRENT TIME:
- C++ vtables contain naked function pointers, not function descriptors,
which is bad. see http://llvm.cs.uiuc.edu/bugs/show_bug.cgi?id=406
- varargs are broken
- alloca doesn't work (indeed, stack frame layout is bogus)
- no support for big-endian environments
- (not really the backend, but...) the CFE has some issues on IA64.
these will probably be fixed soon.
ACKNOWLEDGEMENTS:
- Chris Lattner (x100)
- Other LLVM developers ("hey, that looks familiar")
CONTACT:
- You can email me at duraid@octopus.com.au. If you find a small bug,
just email me. If you find a big bug, please file a bug report
in bugzilla! http://llvm.cs.uiuc.edu is your one stop shop for all
things LLVM.