1. New parameter attribute called 'inreg'. It has meaning "place this
parameter in registers, if possible". This is some generalization of
gcc's regparm(n) attribute. It's currently used only in X86-32 backend.
2. Completely rewritten CC handling/lowering code inside X86 backend.
Merged stdcall + c CCs and fastcall + fast CC.
3. Dropped CSRET CC. We cannot add struct return variant for each
target-specific CC (e.g. stdcall + csretcc and so on).
4. Instead of CSRET CC introduced 'sret' parameter attribute. Setting in
on first attribute has meaning 'This is hidden pointer to structure
return. Handle it gently'.
5. Fixed small bug in llvm-extract + add new feature to
FunctionExtraction pass, which relinks all internal-linkaged callees
from deleted function to external linkage. This will allow further
linking everything together.
NOTEs: 1. Documentation will be updated soon.
2. llvm-upgrade should be improved to translate csret => sret.
Before this, there will be some unexpected test fails.
llvm-svn: 33597
When a value is found to have the same name as another, try harder to
disambiguate when its a type plane collapse issue and when it isn't. We
traverse the type to see if it contains an integer. If it does not then
we issue the error because it can't be resulting from integer type planes
collapsing. Otherwise we just rename it, even if that's a bit of overkill.
llvm-svn: 33594
PR645
PR761
PR1082
PR1122
Completely rewrite llvm-upgrade. This should be its final design. Any future
changes will use this same design. The changes involve the following:
1. Make this work very much like the 1.9 AsmParser
2. Retain old upgrades dating back to release 1.2 time frame.
3. Merge in some of the upgrades between 1.9 and 2.0 (e.g. icmp/fcmp).
4. Attach a Signedness value (Signless, Unsigned, Signed) to every type,
Value, Constant, Instruction, and list of those things in the Parser.
Use these to make signedness decisions for instruction upgrades.
5. Implement unique name upgrade for function values and global values.
6. Identify rename cases that might cause problems and warn about them.
For example: renaming a global variable with external linkage.
7. Generate a 2.0 IR using VMCore. This is necessary for numerous
reasons and has the advantage that it never goes out of date.
8. Use the AsmPrinter to make the output nice.
9. Clean up error and warning messages from 1.9 form.
llvm-svn: 33531
instcombine to clean up the code, which makes more code suitable for SRoA.
This helps C++ code in particular, e.g. speeding up tramp3d by 31%.
llvm-svn: 33235
Increment the counter after the second use, not after the first use. This
fixes PR1113.
Also, rename some classes for simplicity and to more naturally be
reminscient of LLVM 1.9. This in preparation for additional classes that
will provide a scaled down model of the LLVM 1.9 IR.
llvm-svn: 33226
rename Type::getIntegralTypeMask to Type::getIntegerTypeMask.
This makes naming much more consistent. For example, there are now no longer any
instances of IntegerType that are not considered isInteger! :)
llvm-svn: 33225
like a lot, its really only two changes:
1. Move stuff that should be private to .y out of the .h file.
2. Make all semantic values pointers.
This cleans up the silly destroy methods and changes them to delete. It
also moves the TypeInfo and ValueInfo classes into the .y where we can
modify them more readily because they have no clients any more.
This shouldn't result in any functional changes in llvm-upgrade.
llvm-svn: 33211
Implement upgrading of the varargs intrinsics. These must now have i8*
arguments. This patch bitcasts arguments to i8* and fixes the prototypes.
llvm-svn: 33163
Implement the arbitrary bit-width integer feature. The feature allows
integers of any bitwidth (up to 64) to be defined instead of just 1, 8,
16, 32, and 64 bit integers.
This change does several things:
1. Introduces a new Derived Type, IntegerType, to represent the number of
bits in an integer. The Type classes SubclassData field is used to
store the number of bits. This allows 2^23 bits in an integer type.
2. Removes the five integer Type::TypeID values for the 1, 8, 16, 32 and
64-bit integers. These are replaced with just IntegerType which is not
a primitive any more.
3. Adjust the rest of LLVM to account for this change.
Note that while this incremental change lays the foundation for arbitrary
bit-width integers, LLVM has not yet been converted to actually deal with
them in any significant way. Most optimization passes, for example, will
still only deal with the byte-width integer types. Future increments
will rectify this situation.
llvm-svn: 33113
recommended that getBoolValue be replaced with getZExtValue and that
get(bool) be replaced by get(const Type*, uint64_t). This implements
those changes.
llvm-svn: 33110
Take an incremental step towards type plane elimination. This change
separates types from values in the symbol tables by finally making use
of the TypeSymbolTable class. This yields more natural interfaces for
dealing with types and unclutters the SymbolTable class.
llvm-svn: 32956
Solve several related problems by making variable names more unique and
dealing with recursive phi nodes. Unfortunately, this doesn't solve the
main issue reported in the PR, but its a step in that direction.
llvm-svn: 32953
Emit the correct syntax for icmp and fcmp constant expressions. The
predicate goes outside the parentheses not inside. This was a change made
in the assembler without the corresponding change made in llvm-upgrade.
llvm-svn: 32942
1. Completely revise the type system so that types are handled as const
objects and not created multiple times, cloned, or otherwise copied.
This gets around memory issues, saves memory, and also emulates LLVM's
no-two-types-of-the-same-shape-created semantics.
2. Adjust the handling of global names. Basically, we cannot rename them
for a variety of reasons: linking, forward references, etc.
3. Detect global names that have name conflicts as the result of collapsed
type planes or redefinitions that llvm-as no longer accepts. These will
produce warnings on stderr and one of the globals will be renamed.
4. Rename ParserInternals.h as UpgradeInternals.h so it doesn't conflict
in the debugger with ParserInternals.h from lib/AsmParser.
5. Move the guts of the TypeInfo class into the grammar so we aren't
implementing large functions in a header file. This also helps with
debugging a bit.
llvm-svn: 32906
by correctly handling up references and enumerated types. Previously there
was some confusion of these two. Thanks to Chris Lattner for demistifying
llvm up references.
llvm-svn: 32846
Revise the upgrade parser to keep track of types more faithfully and use
this information to resolve name conflicts resulting from collapsed type
planes. The type planes have collapsed because the integer types are now
signless so that uint and int became i32. Where two planes existed for uint
and int, only i32 exists. Any variable names depending on the type planes
to pmake the identifier unique would cause a conflict. This patch resolves
that conflict for many but not all cases.
Situations involving the integer types and pointers to them are handled
by this patch. However, there are corner cases that are not handled
well, such as:
%t1 = type { uint, int }
%t2 = type { int, uint }
void %myfunc(%t1* one, %t2* two) {
%var = load %t1* one
%var = load %t2* two
}
In the scenario above, %t1 and %t2 are really the same type: { i32, i32 }
Consequently attempting to name %var twice will yield a redefinition error
when assembled.
While this patch is sufficien to allow the llvm/test suite to pass, More
work needs to be to complete the handling of these corner cases.
llvm-svn: 32810
Don't attempt to parse both the old and new grammars. It is near impossible
to get it right. Remove support for the new define keyword and don't
attempt to insert parameter attributes because there isn't enough
contextual information for it.
llvm-svn: 32784
Fix several bugs and update for new assembly syntax. Changes made include:
1. Fixing rules for icmp/fcmp instructions to not require a closing paren
at the end. This was a cut-and-paste error from a previous commit.
2. Changing things like Out << " " to Out << ' '
3. Adding the "define" keyword for function definitions
4. Adding support for packed structures
llvm-svn: 32771
This patch removes the SetCC instructions and replaces them with the ICmp
and FCmp instructions. The SetCondInst instruction has been removed and
been replaced with ICmpInst and FCmpInst.
llvm-svn: 32751
div -> fdiv/udiv/sdiv
rem -> frem/urem/srem
except -> unwind
uninitialized -> external
as well as tracking the element type of pointers, packed, and array.
llvm-svn: 32127
adding a temporary wrapper around the ostream to make it friendly to
functions expecting an LLVM stream. This should be fixed in the future.
llvm-svn: 31990
append "exit <retcode>" to the end of the output file. This is used by
the nightly tester to make bugpoint match the output generated by the
RunSafely.sh script so it doesn't find false positives.
llvm-svn: 31960
The long awaited CAST patch. This introduces 12 new instructions into LLVM
to replace the cast instruction. Corresponding changes throughout LLVM are
provided. This passes llvm-test, llvm/test, and SPEC CPUINT2000 with the
exception of 175.vpr which fails only on a slight floating point output
difference.
llvm-svn: 31931
Remove predicate simplifier from default gcc3 pipeline. New design is too
slow to enable by default.
Add new testcases for problems encountered in development.
llvm-svn: 31895
miscompilation. This is useful for working around GCC+CBE bugs or for handling
programs that CBE doesn't support (e.g. inline asm) when searching for
optimizer bugs.
llvm-svn: 31588
This patch converts the old SHR instruction into two instructions,
AShr (Arithmetic) and LShr (Logical). The Shr instructions now are not
dependent on the sign of their operands.
llvm-svn: 31542
Turn on -Wunused and -Wno-unused-parameter. Clean up most of the resulting
fall out by removing unused variables. Remaining warnings have to do with
unused functions (I didn't want to delete code without review) and unused
variables in generated code. Maintainers should clean up the remaining
issues when they see them. All changes pass DejaGnu tests and Olden.
llvm-svn: 31380
Make necessary changes to support DIV -> [SUF]Div. This changes llvm to
have three division instructions: signed, unsigned, floating point. The
bytecode and assembler are bacwards compatible, however.
llvm-svn: 31195
initializers as well. This is only a first pass. It can be slow because
it clones the module for each pass. An obvious improvement is not to do that.
llvm-svn: 31182
This patch implements the first increment for the Signless Types feature.
All changes pertain to removing the ConstantSInt and ConstantUInt classes
in favor of just using ConstantInt.
llvm-svn: 31063
DLL* linkages got full (I hope) codegeneration support in C & both x86
assembler backends.
External weak linkage added for future use, we don't provide any
codegeneration, etc. support for it.
llvm-svn: 30374
This pass:
1. Splits TargetMachine into TargetMachine (generic targets, can be implemented
any way, like the CBE) and LLVMTargetMachine (subclass of TM that is used by
things using libcodegen and other support).
2. Instead of having each target fully populate the passmgr for file or JIT
output, move all this to common code, and give targets hooks they can
implement.
3. Commonalize the target population stuff between file emission and JIT
emission.
4. All (native code) codegen stuff now happens in a FunctionPassManager, which
paves the way for "fast -O0" stuff in the CFE later, and now LLC could
lazily stream .bc files from disk to use less memory.
5. There are now many fewer #includes and the targets don't depend on the
scalar xforms or libanalysis anymore (but codegen does).
6. Changing common code generator pass ordering stuff no longer requires
touching all targets.
7. The JIT now has the option of "-fast" codegen or normal optimized codegen,
which is now orthogonal to the fact that JIT'ing is being done.
llvm-svn: 30081
Close out this long standing bug by removing the remaining overloaded
virtual functions in LLVM. The -Woverloaded-virtual option is now turned on.
llvm-svn: 29934
memory pressure. This order spaces out large executables with small ones in
between so that in a -j2 or -j3 build, it only attempts to build only one
large executable at time. If you're doing -j4, you probably have enuogh
memory anyway.
llvm-svn: 29835
Adjust usage of the ExecuteAndWait function to use the last argument which
is the ErrMsg string. This is necessitated because this function no longer
throws exceptions on error.
llvm-svn: 29791
Make sys::Program::ExecuteAndWait not throw exceptions and update any
affected code. It now return -9999 to signal that the program couldn't be
executed. Only one case (in bugpoint) actually examines the result code.
llvm-svn: 29785
one-off (and broken) RunOptimizations function. Also, run some cleanup
passes after the user's loaded passes run. This make sure to clean up
any cruft left around by thos passes.
This patch was inspired by a patch submitted by Bram Adams.
llvm-svn: 29781
Rid the Assembly Parser of exceptions. This is a really gross hack but it
will do until the Assembly Parser is re-written as a recursive descent.
The basic premise is that wherever the old "ThrowException" function was
called (new name: GenerateError) we set a flag (TriggerError). Every
production checks that flag and calls YYERROR if it is set. Additionally,
each call to ThrowException in the grammar is replaced with GEN_ERROR
which calls GenerateError and then YYERROR immediately. This prevents
the remaining production from continuing after an error condition.
llvm-svn: 29763
Shrinkify LLVM's footprint by removing the analyze tool and moving its
functionality into the opt tool. THis eliminates one of the largest tools
from LLVM and doesn't make opt much bigger because it already included
most of the analysis passes. To get the old analyze functionality pass
the -analyze option to opt. Note that the integeration here is dead
simple. The "main" of analyze was just copied to opt and invoked if the
-analyze option was given. There may be opportunities for further
integration such as removing the distinction between transform passes
and analysis passes.
To use the analysis functionality, if you previously did this:
analyze $FNAME -domset -disable-verify
you would now do this:
opt -analyze $FNAME -domset -disable-verify
Pretty simple.
llvm-svn: 29762
