so that they can still be freed after the reference has been changed
(e.g. in case of array indexing or record field accesses) (mantis #33628)
git-svn-id: trunk@38814 -
tcpuparamanager, very similar to the existing get_volatile_registers_XXX. The
new methods are called get_saved_registers_XXX, where XXX is the register
type ("int", "address", "fpu" or "mm")
git-svn-id: trunk@38794 -
* disable fix for #33439 during bootstrapping with 3.0.x, as 3.0.x cannot compile the currency division without the fix above
git-svn-id: trunk@38558 -
inline assembly, and fixed check after r35959 (mantis #32318)
o can also subscript parameters passed by value on the stack
o can also subscript local variables, the parameters passed by reference
that are subsequently copied into a local
git-svn-id: trunk@37886 -
* changed most of the variables in the assembler readers used to store constants from aint to tcgint
as aint has only the size of the accumular while some CPUs (AVR) allow larger constants in instructions
+ allow access to absolute symbols with address type in inline assembler
* allow absolute addresses in avr inline assembler
+ tests
git-svn-id: trunk@37411 -
determine whether it's in a register if it's a pure assembler routine
* you can't "index" implicit pointers either using their fields
git-svn-id: trunk@36287 -
* compinnr.inc include file converted to a unit
* inline number field size stored in ppu increased from byte to longint
* inlines in the parse tree (when written with the -vp option) now printed with
their enum name, instead of number
git-svn-id: trunk@36174 -
The linker will always change BL to BLX if necessary, but not vice versa (linker version dependent).
"BLX label" ALWAYS changes the instruction set. It changes a processor in ARM state to Thumb state,
or a processor in Thumb state to ARM state.
git-svn-id: trunk@36086 -
directly in inline assembly: that's only possible if it's a register
parameter where the address of the record was passed (rather than the
record itself), or if a parameter has been explicitly typecasted in
Intel-style assembly using ".size"
git-svn-id: trunk@35959 -
- all reg allocs and PIC labeels before the instruction;
- all reg deallocs and reg syncs after the instruction.
It fixes bug #31135.
git-svn-id: trunk@35545 -
ad hoc set constants containing varying number cdecl-like calling
conventions
o added pocall_sysv_abi_cdecl and pocall_ms_abi_cdecl to cstylearrayofconst
o also allow C-style blocks with mwpascal instead of cdecl (mwpascal = cdecl
with "const" = "constref" for record parameters)
o did not touch cases related to name mangling and import/export names,
because those are a real mess and easily break things left and right :/
git-svn-id: trunk@35479 -
directly in inline assembly: that's only possible if it's a register
parameter where the address of the record was passed (rather than the
record itself)
git-svn-id: trunk@35424 -
o separate information for reading and writing, because e.g. in a
try-block, only the writes to local variables and parameters are
volatile (they have to be committed immediately in case the next
instruction causes an exception)
o for now, only references to absolute memory addresses are marked
as volatile
o the volatily information is (should be) properly maintained throughout
all code generators for all archictures with this patch
o no optimizers or other compiler infrastructure uses the volatility
information yet
o this functionality is not (yet) exposed at the language level, it
is only for internal code generator use right now
git-svn-id: trunk@34996 -
procedure node_change_local_switch(p : tnode;cs : tlocalswitch;enable : boolean);
to add/remove cs local switch from all p tnode tree localsettings.
modify second_prefetch implementation in arm, ppcgen and x86 subdirectories
to use node_change_local_settings to temporarily disabled
checkpointer local switch, as prefetch is allowed even on unaccessible addresses.
(previous code was buggy because I frogot that each node has it own localswitches).
git-svn-id: trunk@34576 -
it was AT_NONE, which is invalid and should never be used
* explicitly pass the correct value for all calls to those methods elsewhere
in the compiler
git-svn-id: trunk@34250 -
symbol is defined
o removed all places where AB_INDIRECT symbols were explicitly generated
o only generate AB_INDIRECT symbols for AT_DATA on systems_indirect_var_imports
o for some symbols an indirect symbol is always required (because they are
dereferenced by code in RTL units) -> use new AT_DATA_FORCEINDIRECT type
git-svn-id: trunk@34165 -
tasmdata.DefineAsmSymbol() and all routines that call it
o will be used to automatically generate AB_INDIRECT sybols when
necessary
git-svn-id: trunk@34164 -
moved the code to create the procedure start/end etc. that was at the
call sites of g_external_wrapper() into the new g_external_wrapper()
-> got rid of the x86-specific code in expunix, and fixed expunix for
llvm (e.g. tests/test/tlib1{a,b})
git-svn-id: trunk@34131 -
* updated TAOptObj.RegUsedAfterInstruction with the arm implementation and removed the arm specific implementation
* RegLoadedWithNewValue and InstructionLoadsFromReg are now a methods of TAoptBase
* moved RegEndOfLife to TAOptObj
* during this refactoring, fixed also TCpuAsmOptimizer.RegLoadedWithNewValue for arm regarding post/preindexed
memory references: those modify the register but do not load it with a new value in the sense of RegLoadedWithNewValue
git-svn-id: trunk@33000 -
"don't free even if not registered"; use for defs that may not be written
to a ppu file, but that must nevertheless survive the compilation of the
current module
* mark all defs created for para locations as "don't free even if not
registered", because we don't discard and recalculate all para locations
after a module has been compiled (since that's not needed)
o solves issues if the paralocations for a routine in the interface of
unit A are calculated while the implementation of unit B gets
compiled, and a new reusable type is allocated at that point which
is not used anywhere else (after r32160)
git-svn-id: trunk@32235 -
for all non-ppc(32/64) Darwin platforms
o pass the macosx-version-min/iphoneos-version-min to clang as an assembler,
so that it properly sets this information starting with Xcode 7 (solves
errors when targeting the iOS simulator, and warnings about object files
being compiled for a different OS X version when targeting (Mac) OS X)
o the old assembler is still selectable via -Aas-darwin (required with
Xcode 3.1.x and older)
o since the first Xcode version that shipped with Clang is Xcode 3.2, which
is available for Mac OS X 10.6, most users should not encounter any issues
with the new default (in fact, it fixes some tests for x86 because Clang
supports some instructions that "as" doesn't). Clang does not support
Stabs however, so -gs does require the use of -Aas-darwin
git-svn-id: trunk@31830 -
GNU as from binutils for Darwin, and Apple's "as" is based on an old
version of GNU as)
o this will not cause any backward compatibility problems, as the
previous identifier for as_darwin was "AS-Darwin" and the compiler
compared the uppercase value of the -A parameter to the identifier,
so it was not explicitly selectable earlier. The new name is "AS",
so it is explicitly selectable via -Aas like on other platforms.
git-svn-id: trunk@31806 -
- Fixed access to symbols with offset.
- Always use register R9 for GOT pointer to prevent bugs when free register limit is reached in a function.
- GOT is not needed for function calls by name.
git-svn-id: trunk@31681 -
that we can use assembler writers with different conventions from the
currently set target_asm (e.g. an x86 assembler writer for inline assembly
in LLVM IR)
git-svn-id: trunk@31628 -
labels of LOC_JUMP in the node's location. This generates some extra jumps
for short circuit boolean and/or-expressions if optimizations are off, but
with optimisations enabled the generated code is the same (except for JVM
because the jump threading optimisation isn't enabled there yet).
git-svn-id: trunk@31431 -
something without a location at all; it's a hint/optional operation,
so if there's nothing to prefetch just don't do anything
(mantis #27811)
git-svn-id: trunk@30502 -
Update ARM internal assembler to support most ARM, Thumb and Thumb-2 instructions.
Changed generation of VFP instructions to use UAL mnemonics.
Added divided and unified assembler syntax support to ARM assembly reader.
git-svn-id: trunk@30181 -
o in particular, add tdef size information to the update_reference*()
methods, and factored out offset adjustments into its own method
o also make sure the passed size to update_reference*() corresponds to the
actual size of the index, as it's no longer guaranteed to be ptruint
since the previous commit
git-svn-id: trunk@29967 -
* apparently, wince (or GNU AS for it) does not support blx imm
* set FPC_IN_SYSTEM directive, so assembler code in divide.inc is compiled right
git-svn-id: trunk@29779 -
Precisize rules for selection of thumb instructions.
Add short-cut notation support for most simple Thumb2 instructions ( add r1,#4 instead of add r1,r1,#4 ).
git-svn-id: branches/laksen/armiw@29343 -
Move ThumbFunc flag from section to symbol.
Make .w forms optional the other way around. If .w is explicitly put on an instruction the assembler should always chose a wide form.
git-svn-id: branches/laksen/armiw@29341 -
Fixed bug in Mov instruction.
Added initial scanning of IT/LastInIT detection for proper instruction selection.
Disabled "wide" format flag detection again for now.
git-svn-id: branches/laksen/armiw@29338 -
Fixed a bunch of instruction encodings by comparing bulks of handwritten tests to binutils assembled versions.
Fixed emission of regsets of S and D registers above 15.
Fixed assembler reader for RRX shiftmode.
There can be a size postfix after a condition code in UAL assembler syntax. This has been added to the assembler reader.
git-svn-id: branches/laksen/armiw@29277 -
Switched codegeneration of VFPv2 and VFPv3 to use UAL mnemonics and syntax.
Updated VFP code in RTL to use UAL syntax too.
Added preliminary ELF support for ARM.
Added support for linking of WinCE COFF files. Should work for with a standard ARMv4-I target.
git-svn-id: branches/laksen/armiw@29247 -
Add some VFP registers.
Rebuilt tables.
Added a lot of VFPv3 and Advanced SIMD(not supported yet) oppostfixes.
Implemented code in aasmcpu to generate binary code from the instructions. Only ARM32 supported so far.
git-svn-id: branches/laksen/armiw@29246 -
A more radical approach is to remove them altogether. Tested with i386-win32 (the oldest peephole optimizer), mips-linux (the newest one) and arm-linux (the most complex one) targets. The fallout was limited to two minor issues fixed in r28629 and r28708, respectively.
git-svn-id: trunk@28711 -
r28315 introduced an arm optimization which requires
tst rX, #imm
to work. This is not available on 16bit thumb, I've disabled that
optimization on thumb for now.
git-svn-id: trunk@28360 -
StrLdr2StrMov now uses GetNextInstructionUsingRef to find an instruction
which uses the same Reference. In one of our internal testcases it
speeded up a function by 15% as fpc generated a lot of spilling.
git-svn-id: trunk@28344 -
It's the counterpart to GetNextInstructionUsingReg and finds the next
instruction to use the same reference. By default it stops searching
when hitting a store instructions to avoid aliasing issues.
git-svn-id: trunk@28343 -
This is a very common construct in normal code and also heavily used in
softfpu code.
The ARM-cg will now just test for the MSB of reghi to be set, instead of
a full comparison against constant 0.
git-svn-id: trunk@28315 -
There was an interference between the load scheduler and then
Str/LdrAdd/Sub2Str/Ldr peephole optimizer.
ldrb r0, [r2]
ldrb r1, [r2, #1]
orr r3, r0, r1
add r2, r2, #2
got changed pre-regalloc to:
ldrb r1, [r2, #1]
ldrb r0, [r2]
orr r3, r0, r1
add r2, r2, #2
and the peephole optimizer collapsed the add into the second ldrd:
ldrb r1, [r2, #1]
ldrb r0, [r2], #2
orr r3, r0, r1
Then the post-peephole optimizer changed that into:
ldrb r0, [r2], #2
ldrb r1, [r2, #1]
orr r3, r0, r1
so r1 got loaded from a modified base-register.
This patch prevents the scheduler from moving an ldr-instruction if it
uses Pre/Post-indexing and the instruction before it uses the
base-register.
git-svn-id: trunk@28284 -
to free up space for more ait_* types in taitype (can't have
more than 32 because they have to fit in a small set)
o factored out writing of floating point numbers as an array of
byte in the external assemblers
git-svn-id: branches/hlcgllvm@28105 -
Background: these operations are defined as flipping or clearing the upper bit of number, respectively, and never result in precision loss or raise floating-point exceptions.
git-svn-id: trunk@27411 -
o made all (non-abstract) tdef and tsym constructors virtual
o added c*def/c*sym classref types for every (non-abstract) t*def/t*sym
class
o added cpusym unit for every architecture that derives a tcpu*def/tcpu*sym
class from the base classes, and initialises the c*def/c*sym classes with
them. This is done so that the llvm target will be able to derive from
the tcpu*def/sym classes without umpteen ifdefs, and it also means that
the WPO can devirtualise everything because the c* variables are only
initialised with one class type
o replaced all t*def/t*sym constructor calls with c*def/c*sym constructor
calls
git-svn-id: trunk@27361 -
on the stack, this can be enabled by -OoFORCENOSTACKFRAME. This reduces the required
entry/exit code and makes an extra register available to the compiler. However, since this is
based on an estimation of the required stack size, it might have two drawbacks:
either the stack frame is estimated to big, the program requires a bigger stack than needed
or it is estimated too small, then the compiler throws an internalerror during compilation. These
issues can be overcome as soon as the compiler supports recompiling subroutines if needed.
git-svn-id: trunk@27239 -
