This eliminates an N*N*logN algorithm from the loop simplify pass, replacing
it with a much simpler and faster alternative. In a debug build, this reduces
gccas time on eon from 85s to 42s.
llvm-svn: 14851
null, uint 1) to uint)' to a constant integer. We can only do this with
primitive LLVM types, because other types have target-specific sizes.
llvm-svn: 14837
this LLVM function:
int %foo() {
ret int cast (int** getelementptr (int** null, int 1) to int)
}
into:
foo:
mov %EAX, 0
lea %EAX, DWORD PTR [%EAX + 4]
ret
now we compile it into:
foo:
mov %EAX, 4
ret
This sequence is frequently generated by the MSIL front-end, and soon the malloc lowering pass and
Java front-ends as well..
-Chris
llvm-svn: 14834
* Simplify a lot of code because type's cannot be in function symbol tables
* Fix memory leaks in handling of redefined function prototypes
* Don't use SymbolTable directly for stuff that we can go through the Module
for.
* Fix some minor bugs on obscure testcases like:
test/Feature/globalredefinition.ll
* Do not create GlobalVariable objects for forward referenced Functions!
* When forward referencing a function in a constant expression, do not create
a placeholder, add a bunch of references to it, then turn around and
replaceAllUsesOfWith on it with a new global, deleting the placeholder.
Instead, when we find the real definition of the global, just use the
placeholder instead of creating a new object.
This substantially simplifies the asmwriter and should even speed it up on
cases heavy in constantexprs (like C++, Java, MSIL)...
llvm-svn: 14830
replaceAllUsesWith'ing any forward references, just use the forward
reference if it exists.
This introduces GetForwardRefForGlobal, which will eventually completely
replace the horrible DeclareNewGlobalValue function.
llvm-svn: 14828
to substantially simplify the result. In particular, we no longer create
GlobalVariables and then immediately destroy them when they are duplciate
definitions.
The real point of this patch though is that it gets us closer to the
DeclareNewGlobalValue calls...
llvm-svn: 14827
for a basic block, use it when the block is defined instead of deleting it
and creating a new one. Also, only create at most ONE forward reference
for any block, instead of one for each forward reference.
llvm-svn: 14807
forward reference blocks if they have been created (instead of creating a new
block, replaceAllUsesOfWith, then nuking the placeholder). This is not yet
implemented.
llvm-svn: 14791
the funciton symbol table to check for conflicts instead of having to
keep a shadow named LocalSymtab. Totally eliminate LocalSymtab. Verified
that this did not cause a regression on the testcase for PR107.
llvm-svn: 14788
1. Split setValueName into two separate functions, one that is only used
at function scope and doesn't have to deal with duplicates, and one
that can be used either at global or function scope but that does deal
with conflicts. Conflicts were only in there because of the crappy old
CFE and probably should be entirely eliminated.
2. Insert BasicBlock's into the parent functions when they are created
instead of when they are complete. This effects name lookup (for the
better), which will be exploited in the next patch.
llvm-svn: 14787
"load (cast foo)". This allows us to compile C++ code like this:
class Bclass {
public: virtual int operator()() { return 666; }
};
class Dclass: public Bclass {
public: virtual int operator()() { return 667; }
} ;
int main(int argc, char** argv) {
Dclass x;
return x();
}
Into this:
int %main(int %argc, sbyte** %argv) {
entry:
call void %__main( )
ret int 667
}
Instead of this:
int %main(int %argc, sbyte** %argv) {
entry:
%x = alloca "struct.std::bad_typeid" ; <"struct.std::bad_typeid"*> [#uses=3]
call void %__main( )
%tmp.1.i.i = getelementptr "struct.std::bad_typeid"* %x, uint 0, uint 0, uint 0 ; <int (...)***> [#uses=1]
store int (...)** getelementptr ([3 x int (...)*]* %vtable for Bclass, int 0, long 2), int (...)*** %tmp.1.i.i
%tmp.3.i = getelementptr "struct.std::bad_typeid"* %x, int 0, uint 0, uint 0 ; <int (...)***> [#uses=1]
store int (...)** getelementptr ([3 x int (...)*]* %vtable for Dclass, int 0, long 2), int (...)*** %tmp.3.i
%tmp.5 = load int ("struct.std::bad_typeid"*)** cast (int (...)** getelementptr ([3 x int (...)*]* %vtable for Dclass, int 0, long 2) to int
("struct.std::bad_typeid"*)**) ; <int ("struct.std::bad_typeid"*)*> [#uses=1]
%tmp.6 = call int %tmp.5( "struct.std::bad_typeid"* %x ) ; <int> [#uses=1]
ret int %tmp.6
ret int 0
}
In order words, we now resolve the virtual function call.
llvm-svn: 14783
- Remove tabs
- Standardize use of space around ( and ).
- Consolidate the ConstantPlaceHolder class
- Rename two methods to be more meaningful (ParseType, ParseTypes)
- Correct indentation of blocks
- Add documentation
- Convert input dependent asserts to error(...) so it throws instead.
Provide placeholder implementations of read_float and read_double that
still read in platform-specific endianess. When I figure out how to do
this without knowing the endianess of the platform, it will get implemented
correctly.
llvm-svn: 14765
- get rid of PARSE_ERROR macro
- add error(std::string) function
- use error(std::string) for all errors
- make input dependent asserts call error(std::string) instead
- ensure asserts are only for logic bugs, not input discrepancies.
llvm-svn: 14729
global type plane starts with a length and the TypeTyID value to identify
the type plane has having type definitions. This doesn't happen in 1.3
because the types are read from a known position in the file. However, the
TypeTyID must be read in (and ignored) if its a 1.2 bytecode file.
llvm-svn: 14728
The large diff is because of indentation of a whole region
* Fix querying predecessor blocks in SelectPHINodes(), thanks to Brian (v8)
* Add support for external functions malloc() and free()
* Fix some code indentation
Remember, kids: It's not plagiarism if you "creatively borrow" from your
sources. It's called "research"!
llvm-svn: 14723
the top and bottom halves backwards...how embarrassing.
Support 'cast long to long' and other similar no-op casts to long.
Support 'ret long'.
llvm-svn: 14683
Regression.Assembler.2002-01-24-BadSymbolTableAssert
Regression.Assembler.2002-01-24-ValueRefineAbsType
Found through the nightly tester :)
llvm-svn: 14671
is HOPELESSLY broken. The problem is that the embedded getNode call can
change the offset of the node handle in unpredictable ways.
As it turns out, all of the clients of this method really want to set
both the node and the offset, thus it is more efficient (and less buggy)
to just do both of them in one method call. This fixes some obscure bugs
handling non-forwarded node handles.
llvm-svn: 14660
* Make visitSetCondInst() share condition-generating code with EmitComparison()
* There are 13 FPRs for function-passing arguments, not 8
* Do not rely on registers being sequential, use an array lookup
* In unimplemented switch cases, send an error and abort instead of silent
fall-through
* Add doInitialization() for adding function prototypes for external math fns
* Minor changes: fix indentation, spacing, code clarity
llvm-svn: 14653
Don't touch GEPs for which DecomposeArrayRef is not going to do anything
special (e.g., < 2 indices, or 2 indices and the last one is a constant.)
llvm-svn: 14647
* If SetCondInst is folded into BranchInst (and it is the only user), do not
emit code for SetCondInst
* Fix assembly opcodes in comments in visitSetCondInst()
* Fix codegen of conditional branches
llvm-svn: 14643
basic block clear()'s all of the operands lists, including phis. This
caused removePredecessor to get confused later. Because of this, we just
nuke (without prejudice) PHI nodes in unreachable blocks.
llvm-svn: 14635
too soon so the function data was not getting dumped (it was generated
after the call handleFinish). Also cleaned up the output format for
proper indentation.
llvm-svn: 14627
- Remove Tabs
- Add inserters needed since Value doesn't implement them any more
- Move some functions here to avoid cyclic header file dependencies.
llvm-svn: 14618
done by doubling up the data structures so that Type based equivalents are
used. A consequence of this is overloading of function members that take a
Type* instead of a Value*. Various other cleanups related to Type != Value
(bug 122) were also implemented.
llvm-svn: 14613
Separate Types from Values because Type no longer inherits from Value. The
changes for this are too numerous to list. In essence, any data structure
that contained a Value was doubled so that Types could be contained
similarly. New members include Types, TypeMap, CompactionTypes, and
CompactionTypeMap. Functions taking a Value* were overloaded with a variant
that takes a Type* that acts on the new data structures.
llvm-svn: 14608
- Don't use PARSE_ERROR macro unnecessarily (for simple strings)
- Add comments before each function
- Convert for bug 122 (Type != Value)
- Handle new value range on Type::PrimitiveTypeId enumeration by augmenting
the reader with a new read_typeid method and sanitizeTypeId method.
- Remove BytecodeHandler's default method implementations to header file.
llvm-svn: 14604
