overhead: Merge 3 parallel vectors into 1, change regsUsed hash_set to be a
bitvector. Sped up LLC a little less than 10% in a debug build!
llvm-svn: 4261
interface with code that uses symbols in the ansi-c protected namespace. In
most cases this comes from system header files, such as stdio.h. In particular,
without this change, a reference to the __iob symbol is mangled into ll_iob,
which is not resolved by libc.
llvm-svn: 4165
(1) Padding bytes between structure fields (for alignment) were never
being emitted into the constant pool so the layout did not match!
(2) In printing constants, structures containing structures or arrays
were never handled.
(3) Support new model for external/uninitialized/initialized globals.
Uninitialized globals are no longer emitted since they are external.
Initialized globals may go either in .bss or in .data.
llvm-svn: 4134
(2) Use intelligent multiply selection code for array allocas.
(3) Don't use cache padding for alloca'd stack slots!
(4) Bug fix in handling call arguments: was not copying sixth FP arg
to int reg. when calling a function with no prototype.
llvm-svn: 4130
instructions so that (a) constant folding is done automatically before
code generation, and (b) selection does not have to deal with them.
Also, check for ConstantPointerRefs in additional to GlobalValues
when creating a GEP to load a global address.
llvm-svn: 4126
Don't allow direct FP-to-uint conversion (must be eliminated by preselection).
Address arithmetic for arrays is now entirely 64-bit so no sign-ext needed.
llvm-svn: 3961
Original logs for PreSelection.cpp:
revision 1.2
date: 2002/09/17 23:50:32; author: lattner; state: Exp; lines: +1 -3
Don't put default parameter values into .cpp files, it breaks 3.x compilers
revision 1.1
date: 2002/09/16 15:31:13; author: vadve; state: Exp;
New preselection pass that specializes LLVM code for a target machine,
while remaining in legal portable LLVM form and preserving type
information and type safety.
llvm-svn: 3838
TargetMachine.cpp, since it is entirely machine-independent.
Also, add options to disable optional back-end passes (preselection and
instr. scheduling).
llvm-svn: 3740
-- Add support for ConstantExpr constants (only cast and add operators so far)
-- Avoid generating label Bbss.bss, which sometimes came out twice.
llvm-svn: 3578
SetOperandsForMemInstr significantly). Load and Store
no longer have any indices. Eliminate spurious sign-extension
on a cast to float/double.
llvm-svn: 3498
- Eliminate Sparc*Order classes, moving contents to Sparc*Class classes
- get register name methods now return const char * const pointers instead
of std::string's.
- Added file header comments to SparcRegClassInfo.(cpp|h)
- Moved BIG constant arrays out of SparcRegClassInfo.h, into
SparcRegInfo.cpp. This should allow a LOT of std::string constructors to
not have to be called.
llvm-svn: 3295
and generate actual machine instruction sequences directly.
Also a couple of bug fixes in code for putting constants into registers:
-- Do *not* sign-extend unsigned constant that is shorter than int reg size
-- Fix handling of address constant (a GlobalValue) vs. constant that
must be loaded.
llvm-svn: 2856
-- FP argument to a function with no prototype going on stack
was not being copied to the stack in colorCallArgs().
-- Put caller-saving code *before* argument copying code so that
we don't trash a register before saving it!
-- Two other minor fixes.
llvm-svn: 2855
basic block.
Mark setCCInstr used as dest. of conditional-move as both a def and a use.
BA instruction no longer has the unused CC argument.
llvm-svn: 2836
handle conditional move instructions:
-- cpMem<->Reg functions now support CC registers (int and FP) correctly.
Also, cpMem<->Reg functions now return a vector of machine instructions.
-- Scratch registers must be explicitly provided to cpMem<->Reg when
needed, since CC regs need one to be copied to/from memory.
-- CC regs are saved to a scratch register instead of stack.
-- All regs used by a instruction are now recorded in MachineInstr::regsUsed,
since regs used to save values *across* an instruction are not obvious
either from the operands or from the LiveVar sets.
-- An (explicit or implicit) operand may now be both a def and a use.
This is needed for conditional move operations.
So an operand may need spill code both before and after the instruction.
Other changes:
-- Added several get{Class,Type} functions.
-- Added unified-to-local register number conversion.
-- class MachineCodeForBasicBlock is now an annotation on BasicBlock.
-- Suggest/Color methods may modify the MachineInstr (and always did),
so don't make that argument const!
-- Caller-saving code doesn't need its special purpose code for
handling CC registers since cpMem<->Reg handle those correctly now.
llvm-svn: 2834
Chris Lattner