* Make the function pointer argument explicit for a call nodes
* Eliminate unreachable global values
* Merge call nodes that are identical
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2266 91177308-0d34-0410-b5e6-96231b3b80d8
* Fold call nodes that are indistinguishable for each other. This is a big
win for external functions like sqrt, which would multiply dramatically
before.
* Global nodes with no edges to or from them are now eliminated from the graph.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2257 91177308-0d34-0410-b5e6-96231b3b80d8
* Build new datatypes correctly
* Transform instructions that return null pointers from functions to return
a null index.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2244 91177308-0d34-0410-b5e6-96231b3b80d8
* Add new function printTypeAtLeastOneLevel used to...
* Print the symbol table *WITH SYMBOLIC TYPES*. Now we get:
%tree = type { int, %tree*, %tree* }
in the type definition section of the disassembled output instead of
%tree = type { int, \2*, \2* }
the different for the health benchmark and power are simply amazing.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2240 91177308-0d34-0410-b5e6-96231b3b80d8
* Handle more complex structure so that power works almost
* Fix bug with pooldestroy call where we passed in the size of the object
instead of the pool to destroy. bisort is now pool allocated successfully!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2238 91177308-0d34-0410-b5e6-96231b3b80d8
* Do not print the allocation size for a non array allocation (this used to work,
but was broken).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2235 91177308-0d34-0410-b5e6-96231b3b80d8
* Split the CleanGCC pass into two passes, a global pass and an IP pass.
Before it was just a global pass, but it did illegal things to the
module, which broke other passes that were being scheduled with it by
gccld.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2224 91177308-0d34-0410-b5e6-96231b3b80d8
be 'Argument' instead of FunctionArgument.
Rename some yacc type names to be more concise. Change jump table to use
a vector instead of a list.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2214 91177308-0d34-0410-b5e6-96231b3b80d8
* Delete the DelaySlotInfo objects created by the SchedulingManager class.
These leaked objects were accounting for 3/4 of the memory leaked by the
backend, so this is a relatively major win.
* Reorganize SchedulingManager::getDelaySlotInfoForInstr so that it has
better code locality (making it easier to read).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2197 91177308-0d34-0410-b5e6-96231b3b80d8
* Move InsertPrologEpilogCode class to PrologEpilogCodeInserter.cpp
-> Among other things, this eliminates the need for the minstrVec static
global array.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2195 91177308-0d34-0410-b5e6-96231b3b80d8
new PrologEpilogCodeInserter.cpp file, and include the bodies of the
GetInstructionsFor(Prolog|Epilog) functions from SparcInstrSelection.cpp
into the class. This eliminates the need for a static global vector of
instructions and a class of errors that is really unneccesary.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2194 91177308-0d34-0410-b5e6-96231b3b80d8
the bodies of the GetInstructionsFor(Prolog|Epilog) functions from
SparcInstrSelection.cpp into the class. This eliminates the need for a static
global vector of instructions and a class of errors that is really unneccesary.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@2193 91177308-0d34-0410-b5e6-96231b3b80d8
