This one folds
mov r1, r2, lsl #2
ldr/ldrb r0, [r0, r1]
into
ldr/ldrb r0, [r0, r2, lsl #2]
There is still some room for improvement, maybe it would be better to do this before
the register allocator runs, as we'll currently waste a register (r1 in the above example)
in many cases. That would also allow to to fold more operations, because currently if r2
gets reused between the mov and ldr we'll not be able to do the optimization.
git-svn-id: trunk@23408 -
m68k/aasmcpu.pas, taicpu.spilling_get_operation_type:
* add all Sxx instructions as "operand_write" instructions
m68k/n68kadd.pas, t68kaddnode.getresflags:
* use the correct operation in case of swapped nodes
m68k/cgcpu.pas, tcg68k.g_flags2reg:
- don't move a 0 to the register, because this will CLR it and thus the flags won't be valid anymore...
- NEG would have been the wrong operation (NOT would have been correct), but it isn't needed anyway...
* simplify the method by handling the address register case only when necessary
git-svn-id: trunk@23383 -
- removed a_jmp_cond, it's not virtual and not applicable to this CPU.
* a_loadfpu_reg_cgpara: use direct register moves for simple destinations.
* g_concatcopy: don't take address of source/destination for small amounts of data if possible, for single 32-bit moves this reduces number of generated instructions from 4 to 2.
* g_intf_wrapper: 'Self' is in R4 (a0), not R2. Fixes test/tinterface1.pp.
* mips/cpupara.pas: for functions with result returned in parameter, pass its address in a0 only if result is a record. ABI does not specify behavior for types except records/unions. At the same time, Pascal code relies on the fact that results like strings/interfaces does not change their locations, i.e. "function foo(<self>): IInterface" can be invoked as "procedure foo(<self>, out obj)". This fixes test/tdel1.pp and some Variant-related tests.
git-svn-id: trunk@23377 -
ARM status: roughly corresponds to i386 one, passes the test suite. Handles libraries, can link static libc code including basic PIC and TLS IE/LE stuff. Completely misses Thumb support. Also does not handle ABI-specific stuff, for this reason internally linked .so cannot be used for linking executables with ld. Little-endian only. Tested only on "versatilepb" QEMU virtual machine.
MIPS status: can link the compiler and at least some dynamic executables including fpmake. Some PIC support is present but almost untested. Specific header flags and sections are also not handled yet. Written to handle both endian, but tested for big-endian only ("malta" QEMU VM), including cross-linking from x86_64.
git-svn-id: trunk@23376 -
* For unresolved weak symbols, provide dynamic binding only if they are referenced via GOT or PLT (ld-compatible behavior).
* Made more TElfExeOutput properties/methods usable by descendant classes.
* x86_64/cpuelf.pas: some refactoring, handle DTPOFF relocations, prohibit TPOFF relocations in shared objects.
* i386/cpuelf.pas: set symref_from_text flags for copy relocations to work correctly.
git-svn-id: trunk@23375 -
use the correct flag for the copy loop: we jump back to the copy code as long as the value is positive aka BPL instead of BMI
This fixes around 30 tests (it fixes a quite bit more, but now some other tests seem to be broken...)
git-svn-id: trunk@23373 -
Only referencing data labels from code should change "GOT is needed" property. Writing data labels or referencing them from data should have no effect on it.
git-svn-id: trunk@23363 -
Note: the error messages for incorrect "misstyled" floating point numbers (e.g. "2e10foo") have changed because of this.
scanner.pas, tscannerfile.readtoken:
instead of tokenizing "2.", "2.e10", "2.e+10" and "2.e-10" as "_REALNUMBER" tokenize them as "_INTCONST _POINT", "_INTCONST _POINT _ID", "_INTCONST _POINT _ID _PLUS _INTCONST" "_INTCONST _POINT _ID _PLUS _INTCONST"; tokenizing of normal floating constants is not changed
pexpr.pas:
factor:
* extract the code for creating a new constant floating point from "factor" into a new function "real_const_node_from_pattern"
+ allow the parsing of postfixoperators for integer constants if a "." is encountered
+ postfixoperators: check for a "misstyled" floating point number if an ordinal const (not an enum and not a boolean) is encountered (the code is already partially prepared for type helper support)
+ Added tests
git-svn-id: trunk@23356 -
pdecvar.pas, read_record_fields:
* if the def of the field is a static array then use the array's element def (the final element def if it is a multi dimensional array) to check for whether this is the current record type
+ added tests
git-svn-id: trunk@23352 -
* adjust "is_generic" so that it will really work for true generics only
+ added an analogous method "is_specialization" for specializations
Both methods are not used yet, but this will change in the future...
git-svn-id: trunk@23348 -
pgenutil.pas, generate_specialization:
in case of "assigned(parsedtype)" an ">" could immediately follow (only one type parameter), so don't necessarily expect a type identifier
git-svn-id: trunk@23347 -
pgenutil.pas:
generate_specialization:
+ instead of giving an internal error if "symname" is empty and "tt" is "nil" we now do an error recovery by parsing the specialization parameters and returning an errordef (this happens if the "generic" type before the "<" is not found)
* handle "<>" specially by giving an approbiate error message (both when doing a recovery/parsing a generic and during normal specialization)
parse_generic_parameters:
* set the "block_type" to "bt_type" to be on the safe side
* don't continue with inspecting the def (especially hard typecasting) if the found def is not an "objectdef"
Added tests.
git-svn-id: trunk@23344 -
* If possible, use reference base instead of index, this yields shorter instructions.
* Added comment about offset limits for rip-relative addressing.
- Removed code related to taking threadvar address on win32, it is incorrect because Windows TLS is not directly accessible via segment registers (fs:0x2c points to array of pointers to TLS storages of each module, so at least double indirection is needed).
git-svn-id: trunk@23342 -
that the type of the parameters can be determined automatically
o added compilerproc declarations for all helpers called in the compiler
via their assembler name, so we can look up the corresponding procdef
git-svn-id: trunk@23325 -
compiles again there (it's possible to take the address of arbitrary
threadvars on the jvm platform, but not of arbitrary other variables)
- disabled inclusion of generic thread manager and threadvar support
on the jvm target if the threading feature is enabled
git-svn-id: trunk@23322 -
* Netware target may be broken by this change; at least it deserves a good cleanup because it contains a lot of code trying to solve similar task: to write exesections without alignment and deal with objsections that are nevertheless being aligned by TExeOutput.
git-svn-id: trunk@23291 -
* TElfSymtab.WriteSymbol: write type of external symbols as STT_NOTYPE only in object files; external symbols in exe/so files must have type of their resolving symbol.
* TElfObjData.writeReloc_internal: RELOC_RELATIVE needs adjustment by size presumably on x86 only; ARM data relocations don't need it, other CPUs aren't checked yet.
* Use VER_xx and VERSYM_xx constants instead of literal values.
git-svn-id: trunk@23290 -
* Cleanup of Android linker.
* Use only INSERT command in the linker script to inject custom .fpc sections. Whole linker script duplication is nor needed anymore.
* Use BFD linker, since GOLD linker does not support INSERT command. Other incompatibilities may also exist.
* Enable tf_smartlink_sections for i386-android.
* i386-android is broken after this commit. Will be fixed later.
git-svn-id: branches/targetandroid@23284 -
1) tguidconstnode.pass_1 (location is LOC_CREFERENCE)
2) tonnode.pass_1 (generic secondpass references VMT of exception class)
3) Conversion from unsigned to floating-point involves 'magic constant' as GOT-relative data object.
4) Case jumptable address is loaded relative to GOT.
(1) and (2) also provide more accurate setting of pi_needs_got for all targets.
git-svn-id: trunk@23188 -
symconst.pas:
+ extend "tdefoptions" by a "df_genconstraint" which will be used to mark dummy defs that should mainly satisfy the compiler's type checking without to much changes
symsym.pas:
+ add a class "tgenericconstraintdata" which will hold information about the constraints associated with a specific generic type parameter (designed for future extensions)
+ extend "ttypesym" by a reference to a "tgenericconstraintdata" which is written to the ppu only if needed
symtype.pas:
+ add a pointer to "tderef" as this is needed for the ppu reading/writing code for the "tgenericconstraintdata"
pdecl.pas, types_dec:
+ call "parse_generic_parameters" so that constraints are allowed
pgenutil.pas:
+ extend "generate_specialization" by a "parsedpos" to give in the file position of the first parsed parameter (needed for correct error locations when checking the constraints)
+ add an overloaded call of "generate_specialization" to differentiate between the use cases "first parameter parsed by generate_specialization" and "first parameter parsed by other code"; this also allows to write the "fillchar" for the "parampos" parameter only once ( => otherwise a warning is triggered => error in compilation)
+ extend the "parse_generic_specialization_types" by a "poslist" parameter which will contain the positions of all parsed type parameters (can only be used in the case that all parameters are parsed)
* move the original code of "parse_generic_specialization_types" to a new procedure "parse_generic_specialization_types_internal" which take an additional "parsedpos" parameter which will be added to the "poslist" before all others; "parse_generic_specialization_types" calls this new procedure with a dummy argument (which won't be used)
+ extend "parse_generic_parameters" by the parsing of generic constraints which constructs correct defs for the parameters and fills in the new "tgenericconstraintdata" object for each parameter (note: the "constructor" constraint is only parsed for Delphi compatibility and basically means the same as a "class" constraint... (it's a relict of Delphi.NET))
* adjust "insert_generic_parameter_types" as specializations and generics can no longer be differed by whether the type parameters are of type "undefineddef"
pdecsub.pas, parse_proc_head, consume_generic_interface:
* adjust call to "generate_specialization"
+ add a new function "check_generic_constraints" which is used from within "generate_specialization" to ensure that the given specialization parameters are compatible with the constraints of the generic parameters
ptype.pas:
* single_type: adjust call to "generate_specialization"
* read_named_type, expr_type: adjust call to "generate_specialization"
+ write_persistent_type_info: don't write typeinfo for constraints
pexpr.pas, sub_expr:
* adjust call to "generate_specialization"
* adjusted ppudump, because of added "tdefoptions.df_genconstraint" value
+ added tests for generic constraints
* modified test for class helper inside a generic which extends a class type parameter
+ added test for record helper inside a generic which extends a record type parameter
git-svn-id: trunk@23158 -
defcmp.pas, compare_defs_ext:
use "equal_defs" instead of a reference comparison to check whether the type defs are equal
pgenutil.pas, insert_generic_parameter_types:
* add to each (specialized) generic parameter the "sp_generic_para" flag so that it will be added to the "genericparas" list upon ppu loading
* add the generic parameter to the genericparas list as well, so that the comparison in "compare_defs_ext" is triggered at all... (up to now the list was always empty -.-)
ppu.pas:
* increase PPU version
git-svn-id: trunk@23149 -
* Added symstate_undefweak for undefined symbols having only weak references to them, this state is removed when a normal reference to symbol is seen.
* Use a separate boolean property to determine if ExeSymbol is referenced.
* Use a separate property TExeOutput.AllowUndefinedSymbols to suppress error messages when linking dynamic shared objects.
* Don't issue InternalError when an undefined symbol is encountered during unused section removal.
git-svn-id: trunk@23127 -
* Dropped TObjInput.newObjData method. Instead, TObjData instances are created directly in ReadObjData, this gives possibility to create a different type of TObjData based on file header.
- Dropped TObjInput destructor, it does nothing except calling inherited.
git-svn-id: trunk@23118 -
* Require symbol table presence only in dynamic objects; relocatable objects may not contain one (example is crtn.o).
* Attempt to write relocations from dynreloclist regardless of relative_reloc_count, the latter is typically 0 in executables.
git-svn-id: trunk@23115 -
* Methods ReportNonDSOReloc and ReportRelocOverflow moved to TElfExeOutput, they are applicable to all target-specific descendants.
git-svn-id: trunk@23094 -
* Moved target-dependent constants into ElfTarget records, eliminates a lot of $ifdef's in ogelf.pas.
+ Added TElfTarget.loadsection hook and pass unknown sections to it.
git-svn-id: trunk@23090 -
- removed redundant assigned() checks before .Free
- removed DataPos_ExeSection method which only calls inherited
- removed all references to RelocSections variable, it's safe because TExeOutput doesn't handle it since r21971.
git-svn-id: trunk@23085 -
* Write relocation format dependent dynamic tags based on actual sh_type of dynamic relocation sections, instead of global relocs_use_addend flag.
* Don't write DT_REL[A]COUNT tag if .rel[a].dyn section is not present.
git-svn-id: trunk@23083 -
* Changed signature of GOTRelocPass1 to pass relocation index instead of relocation itself, because sometimes (e.g. when doing TLS relaxation) we need to process several subsequent relocations at once.
git-svn-id: trunk@23077 -
property is private, the "used" tracking of the property itself will
indicate whether the symbol is actually used (mantis #22155)
git-svn-id: trunk@23070 -
* Made name of dynamic interpreter variable, it is now set correctly for each target.
- Removed setting header flags for ARM. This functionality belongs to CPU-specific unit anyway, and probably must be implemented using object attributes in modern ABIs.
git-svn-id: trunk@23032 -
* For Win32 and Win64 targets made sec_rodata read-only, too (for these targets it doesn't matter if read-only section contains relocations).
git-svn-id: trunk@23025 -
a win64 compiler support the x87 types extended and comp/currency handled by the FPU, this
requires to use a i386 compiler as starting compiler and cross compilation because the starting
compiler must support the 80 bit extended type
git-svn-id: trunk@23023 -
* ELF linker: don't make unresolved weak symbols dynamic if linking statically (-Xt in command line).
* Now internal linker produces working executable for tw14265 also on i386-linux when compiling manually (still unable to locate required libraries when run from test suite).
git-svn-id: trunk@22987 -
JVM target, since Java bytecode itself has no support for properties and
hence exposing properties to external Java code can only be done through
getters/setters. Use the new parameters to do so:
-CTautogetterprefix=XXX
-CTautosetterprefix=YYY
The getter/setter will get the same visibility as the property. If a
getter/setter with the same naming convention was already specified for a
property and this getter/setter is declared in the same class as the
property, then the visibility of this existing getter/setter is
modified and no new routine is generated.
Newly generated getters/setters are virtual methods, because that is
the only way in Java bytecode to allow redefining these getters/setters
in child classes. However, that also means that using these switches can
change the behvaviour of code, since normally the used property definition
is only determined by the declared type of its associated class instance,
and not by the actual instance type. The compiler will therefore warn when
such an automatically generated getter/setter is overridden by another
automatically generated getter/setter in a child class.
git-svn-id: trunk@22959 -
methods of tpropertysym (refactoring)
* turned code to copy the contents of a property into another one
into a method of tpropertysym (refactoring)
git-svn-id: trunk@22955 -
(older) version of the parameter is still supported by current linker
versions, but the older version isn't supported by older linker
versions
git-svn-id: trunk@22954 -
* TElfObjInput.LoadSymbols: ignore STT_SECTION symbols for sections that were ignored in LoadSections().
* TElfObjInput: changed symtabndx (index of .symtab/.dynsym section) into field and its type to unsigned.
git-svn-id: trunk@22951 -
+ a_op64_reg_reg: add support for "NEG" and "NOT" of 64-bit values
+ a_op64_const_reg: make sure that we know whether a NEG or NOT with a constant is performed
git-svn-id: trunk@22930 -
* since m68k is 1) a big endian system and 2) a system without a fixed stack (thus we use a push-equivalent) we need to swap the order of the hi and lo longwords of a 64-bit value when passing it to a function
git-svn-id: trunk@22928 -
* in case of copying from the parameter location to the local location we need to use the alignment size for the source as byte/word values are passed as LongInts (this is how the ABI is specified)
git-svn-id: trunk@22924 -
- Removed separate ELF section names for PIC. The only difference was .data named .data.rel; however .rel suffix has nothing to do with PIC. It only signifies that the section has relocations so such sections can be grouped together in output file and reduce number of pages for dynamic linker to visit while resolving the relocations at load time. At the same time, no existing link scripts distinguish between .rel and any other suffix (except .rel.ro, but it's a different story), meaning that long section names will break .rel grouping.
While support for .data.rel can be added similar to existing rodata handling (separate sec_data and sec_data_norel sections), this doesn't seem worth the trouble.
git-svn-id: trunk@22915 -
* Simplified .shstrtab handling between passes by using oso_debug_copy attribute on it.
- Don't recreate segment mapping: only debug sections are removed between passes, and they are never part of any segment.
git-svn-id: trunk@22914 -
+ add routine "call_rtl_moddiv_reg_reg" which handles the calling of "fpc_div_longint", "fpc_div_dword", "fpc_mod_longint" and "fpc_mod_dword"
* emit_mod_reg_reg & emit_div_reg_reg: use the new method instead of doing the call oneself
=> "Str(SomeInt, SomeStr)" and "Writeln(SomeInt)" now works
git-svn-id: trunk@22893 -
+ add methods "call_rtl_mul_const_reg" and "call_rtl_mul_reg_reg" which can call the RTL helpers "fpc_mul_longint" and "fpc_mul_longword" (based on AVR code)
* use the new call methods for the RTL to correctly pass the parameters (on the stack, not in registers...)
git-svn-id: trunk@22892 -
* don't do a sign_extend, but use the correct move size to copy the const; this fixes the setting of the line ending style inside of "Assign"
=> output of strings does now work correctly!
git-svn-id: trunk@22890 -
* TExeOutput.WriteExeSectionContent: To write proper gap between exe sections, don't align writer position. The desired position is present in exesection.DataPos, so pad right up to it.
git-svn-id: trunk@22881 -
Up until now we only checked the next instruction, with the new load
scheduler this is insufficient as shift-instructions and next usage
might farther apart.
The new version uses GetNextInstructionUsingReg, this also comes with a
price as we very carefully have to check if one of the used registers is
changed and that the usage of RRX will not break when we fold and flags
get changed in between.
git-svn-id: trunk@22876 -
Therefore, added dedicated routines align_aword and align_qword that handle target platform size and 64-bit unsigned, respectively.
+ Also added TExeOutput.FixedSectionAlign boolean that, when set to False, ignores SectionDataAlign and SectionMemAlign and aligns every exe section to its own SecAlign value. This kind of alignment is used on ELF targets.
git-svn-id: trunk@22858 -
+ Read .gnu.version sections and ignore local symbols and symbols with non-current version.
+ Check that external symbols are actually present in dynamic objects.
git-svn-id: trunk@22832 -
TRegNameTable is defined in compiler/rgbase.pas and is an array of
strings, limited to the maximum length of the used register names.
r22792 added a long register name but did not scale the string-size
enough, resulting in the compiler built breaking for arm.
git-svn-id: trunk@22817 -
hlcg, and use compproc name)
* fixed unaryminus overflow check for targets that use the generic code
for both 32 and 64 bit integers even if their "native" integer size
is 32 bit (by not checking against low(aint), but opsize.low)
git-svn-id: trunk@22807 -
Add simple Mul+Sub/Mul+Add into MLS/MLA optimizations
Fix some other small issues in the optimizer
Implement Interlocked* functions with proper use of LDREX/STREX
git-svn-id: branches/laksen/arm-embedded@22801 -
So we need to do this manually by changing (in that case) from ".balign" to ".balignw" and passing the opcode of the "NOP" instruction as a argument. The Coldfire manual suggests here to use the "TPF" instruction, but somehow QEMU does not interpret that correctly.
aggas.pas, tgnuassembler.writetree:
* keep track of the last "tai" and pass that to "doalign"
* doalign: do the special handling mentioned above for m68k
git-svn-id: trunk@22799 -
for "CMP" it is important to note that the first operand (which can be basically a register, a constant or a reference) is substracted from the second operand (which needs to be a data register) and not the other way round
git-svn-id: trunk@22798 -
fixes a couple of arm-embedded stuff,
adds some controllers, start of fpv4_s16 support, for a complete list of
changes see below:
------------------------------------------------------------------------
r22787 | laksen | 2012-10-20 22:00:36 +0200 (Sa, 20 Okt 2012) | 1 line
Properly do NR_DEFAULTFLAGS detection/allocation/deallocation
------------------------------------------------------------------------
r22782 | laksen | 2012-10-20 07:44:55 +0200 (Sa, 20 Okt 2012) | 1 line
Fixed flags detections code for wide->short optimization code for Thumb-2
------------------------------------------------------------------------
r22778 | laksen | 2012-10-19 20:23:14 +0200 (Fr, 19 Okt 2012) | 1 line
Added coprocessor registers, and support for 6 operands(MCR/MRC instructions, etc)
------------------------------------------------------------------------
r22647 | laksen | 2012-10-14 21:28:08 +0200 (So, 14 Okt 2012) | 1 line
Added register specifications to lpc1768.pp. From Joan Duran
------------------------------------------------------------------------
r22646 | laksen | 2012-10-14 21:10:20 +0200 (So, 14 Okt 2012) | 4 lines
Fixed some minor formating issues
Implemented a small heap mananger
Implemented console IO
Changed default LineEnding to CrLf(to ease console IO parsing)
------------------------------------------------------------------------
r22599 | laksen | 2012-10-09 08:58:58 +0200 (Di, 09 Okt 2012) | 1 line
Added all STM32F1 configurations
------------------------------------------------------------------------
r22597 | laksen | 2012-10-08 22:10:45 +0200 (Mo, 08 Okt 2012) | 1 line
Added initial support for the Cortex-M4F FPv4_S16 FPU
------------------------------------------------------------------------
r22596 | laksen | 2012-10-08 22:04:14 +0200 (Mo, 08 Okt 2012) | 1 line
Added FPv4_d16 FPU instructions, and a few extra registers
------------------------------------------------------------------------
r22592 | laksen | 2012-10-08 16:07:40 +0200 (Mo, 08 Okt 2012) | 2 lines
Added support for IT block merging
Added a peephole pattern check for UXTB->UXTH chains
------------------------------------------------------------------------
r22590 | laksen | 2012-10-08 14:30:00 +0200 (Mo, 08 Okt 2012) | 3 lines
Add CBNZ/CBZ instructions
Create preliminary Thumb-2 PeepHoleOptPass2 code, hacked together from the ARM mode code
Added a number of simple size optimizations for common Thumb-2 instructions
------------------------------------------------------------------------
r22582 | laksen | 2012-10-08 06:49:39 +0200 (Mo, 08 Okt 2012) | 3 lines
Fix optimizations of Thumb-2 code
Fix problem with loading of condition operand for IT instructions
Properly split IT blocks when register allocator tries to spill inside a block.
------------------------------------------------------------------------
r22581 | laksen | 2012-10-08 05:15:40 +0200 (Mo, 08 Okt 2012) | 4 lines
Fixed assembler calling command line for cpus>ARMv5TE. EDSP instructions will generate errors while assembling, due to RTL assembler routines
Updated boot code for all Cortex-M3 controllers, and sc32442b to use weak linking for exception tables.
Cortex-M3 devices now also share initialization routine to simplify maintenance
STM32F10x classes now have specific units which fit the interrupt source names and counts
------------------------------------------------------------------------
r22580 | laksen | 2012-10-08 05:10:44 +0200 (Mo, 08 Okt 2012) | 2 lines
Added support for .section, .set, .weak, and .thumb_set directive for GAS assembler reader
IFDEF'ed JVM specific assembler directives, to prevent ait_* set to exceed 32 elements
------------------------------------------------------------------------
r22579 | laksen | 2012-10-08 02:10:52 +0200 (Mo, 08 Okt 2012) | 3 lines
Remove all traces of the interrupt vector table generation mechanism
Clean up cpuinfo tables
Fixed ARMv7M bug(BLX <label> doesn't exist on that version)
git-svn-id: trunk@22792 -
