Merge: 7082b6e34a 88f83280dc

Merged riscv_new branch

* moved align helper so it actually gets inlined

+ created branch for keyboard/video/mouse unit unicode support development

+ added tests directory

* Fixed comparison.

* Fixed warnings.

+ add svn:needs-lock to kbd_us.ods, because it's a binary file (and therefore,
unmerge-able)

+ llvm_lifetime_start/llvm_lifetime_end intrinsic declarations

reverted accidental commit to text.inc

* access the Default8087CW global variable directly from within inline asm,
instead of copying it to a local var first, since the concern about PIC code
(this asm code started out as a copy of the i386 rtl code) doesn't exist on
the i8086, since there's no such thing as PIC code on the i8086

* optimize the i8086 backwards move by using movsw instead of movsb in the Move
and MoveData procedures

+ added asm optimized version of fpc_pwidechar_length for i8086

* place the "external" directives in the interface instead of in the
implementation, because we don't support "external" in the implementation
for C varargs functions for LLVM (can't generate a stub for them)
o having "external" in the implementation also causes useless overhead
with code generators that do support this

+ added the initial implementation of my C structure checker code generator
tool, named h2paschk. This is currently a separate tool, that requires a
manually crafted input file with a special syntax (example is included as a
comment in the source code), that contains descriptions of all the records
that need to be checked, but it is planned to add support to h2pas to generate
these files automatically, since this would be very useful.

+ added h2paschk structure checker files for the DirectDraw and DirectInput
header translations, used by PTCPas.

Avoid C 64-bit compiler warnings by using %lu instead of %u for sizeof and offsetof

+ Add handling of simple types and constants
+ Add ReadID helper procedure inside ProcessH2PasChk method

Use %X instead of %x for C hexadecimal constants as Free Pascal hexstr uses uppercase characters A to F for values 10 to 15

* fixed floating point div-by-zero RTE code on Darwin (208 instead of 200,
fixes webtbs/tw32671)

* fixed Darwin/AArch64 flt/int-div-by-zero error codes (fix for 42045, as
noted by AlexeyT)

Allways use local pt variable to ameliorate readability of code

put #ifdef TARGETOS / #endif around the gcclib section, otherwise the various host paths can cause a hard to debug mess during crosscompiling in some cases

Add support for softfloat in RISCV RTL.

* TRttiDataCommon is required outside of the RTTI and DynArray include files as well

* Tenum_typedata needs to include the Common record as well

* disable UCS4String if dynamic arrays are disabled

* tests with Delphi showed that an attribute class must explicitely declare a parameterless constructor if it should be used, cause TCustomAttribute.Create is private
Note: this also means that TCustomAttribute itself can not be used as an attribute
* adjusted existing tests
+ added test

* handle records with managed fields correctly after attribute merge fixes

- Fix overflow checking in software multiplication routines. Overflow checking should also be enabled for the unsigned multiplications.
- Call FPC_Overflow instead of multiple calls to save on size.

* fixed distribution of random(int64) based on patch by Pangea
(mantis 0035878)

* simplified random(int64) code

Try to fix alignment issues in RTTI code for 64-bit targets which require proper alignment

Avoid range check error when compiled -Criot option

Fix bug report 35937 by simple assembly code change and add test

* fix for Mantis 0035965: correct alignment also for the class info used to retrieve the unit name
+ added test

Avoid range check error for functions returning THandle type by using unusedhandle system constant

* m68k has slightly different alignment requirements, so that aligntoptr might result in incorrect addresses for a record's element entries, thus we introduce a separate AlignTypeData function that handles this correctly

* also correctly align the pointer for the record elements inside FPC_COPY

* remove packed from tdynarray, it is not needed and especially 64 bit CPUs which require proper aligment, suffer from it

* call DoUnhandledException instead of RunError(217) when the the psabi eh
unwinder returns an error (this is usually because there is no outer
exception catching frame anymore -> unhandled exception) (based on patch by
Cyrax, mantis 0035751)

* fixed compilation for __ARM_EABI_UNWINDER__ with -dexcdebug

* InitThread variant without parameters as user threads normally cannot know the max. stack size

* the final build of packages and rtl are compiled with the general tls model on i386 as they could be used in dyn. libraries.
This does not hurt much as the linker optimized the threadvar access if possible

* arm uses the same approach as i386: during the final build of the rtl and the packages, the general dynamic tls model is selected

* fix random() for 32 Bit CPUs

Define FPC_IS_SYSTEM at start of all system.pp source

* remove unused constants

* replaced the arrays in the generic set helpers with arrays of 32 bytes
instead of high(cardinal), so LLVM doesn't get tricked into massively
unrolling and parallelising those loops

TInterfacedObject.Destroy: fix case for correct auto-completion

Add riscv32 mathu inculde file, redirecting to ../riscv64/mathu.inc file (this file seems to have no 64-bit specific code)

* fixed (harmless) range error when expanding ~

+ x86-64: assembler implementation for u128_div_u64_to_u64 for SysV ABI
* reminder => remainder (thanks to Stefan Kanthak for pointing it out)

* some inconsistency regarding FPC_NONE usage fixed

Try to correctly handle m68k exceptions bits inside SetExceptionMask function

Wipe out any pre-existing exception on SetExceptionMask call for m68k cpu

* ensure that the includes for the new mm related files are indeed enabled

* ensure that the new mm related procs are only used for 3.3.1 and newer as 3.2 and older don't support the corresponding SSE types

* disabled mm procs again, they are not usable yet neither supposed to be
used

Pass old value of OPT to FPCMAKEOPT

* first couple of improvements propose by Stefan Kanthak at https://skanthak.homepage.t-online.de/division.html#fpc

Add '.built.on.$(SOURCESUFFIX)' for snapshots generated on another target using singlezipinstall

* Makefile fixed

* fixed generic round(...) for large values based on a comment by Alexander Hofmann on fpc-devel
* extended test

* attempt at crt1 translation to get freebsd 12 running.

atari: added the fpwidestring unit to the RTL build

* use getpagesize function instead of reading external variables to restore
ability to compile against SDKs < OS X 10.9 (mantis 0036653)

* allow FPC 3.2.0 as bootstrap version

Try to fix problems for more recent GDB versions.
Move GDB_NEW_UI macro def. to GDB_VER_GE_712
Add GDB_INTERP_LOOKUP_HAS_UI and GDB_HAS_SAVED_COMMAND_LINE_BUT_NO_SIZE
to same GDB_VER_GE_712.
Use stdio_filopen to create gdb_stdout and gdb_stderr
if GDB_HAS_GET_STDIN macro is set.

* Fix compilation of gdbint unit for go32v2, should resolve bug report 36923.
Replace cint type by longint (as gdbint unit does not use ctypes unit).

* disable some x86_64 specific assembly routines

* disable CPU and Signals units for Windows on AArch64 for now

* use accessor functions to work with the TContext record

* the prefetch bug handling is specific to x86_64

* rename RunErrorCodex64 to a more appropriate RunErrorCodeSEH

* enable is_prefetch only for x86 code

* inside the Win64 system unit convert FPC_USE_WIN64_SEH to a SYSTEM_USE_WIN_SEH

+ implement RTL functionality for SEH for aarch64-win64

+ implement a main_wrapper for the default exception handler for aarch64-win64

* Minor improvement. No need to move the terminating null char, since it is written later anyway. Remove a local var.

* Fixed 0044960. Also copy a trailing implicit #0 of the original string to the new larger string.

* Do not report about unimplemented unicodestring manager when HAS_WIDESTRINGMANAGER is defined (on Windows). In such case unimplemented parts are never executed unless you call them directly.
* Do not assign DefaultGetStandardCodePage when HAS_WIDESTRINGMANAGER is defined (on Windows).

* reduced the stack margin for stack checking to 64 bytes on the Z80

* Windows: Do not use CommandLineToArgvW for parsing command line parameters, because it treats \" as an escape sequence for the double quote.
- Implemented parsing of the command line which is backward compatible.
- Do not use GetCommandLineA.

Allow system unit to be compiled without RTTI feature

Use 0 instead of 0xffffffff to disable page width

- removed old float-to-string and string-to-float conversion code

+ add ability to generate a flat binary file instead of a tape file

* simplify code: simply raising an exception is enough, it's not necessary to call ShowException and Terminate

* fix compilation for non 64 Bit targets

Also use union with double field to ensure correct alignment for float64 type from genmath.inc source

Put floating point type as first variant of floatXX definitions, to ensure correct alignment

Fix conversion between TFPURoundMode and bits of get_fsr according to SPARC-V8 specifications

New: rtl/inc/genmathh.inc source file, to provide a unique position
for the interface definition of float64 and float32 records,
available to all systems if FPUNONE is not defined.

rtl/inc/genmath.inc: Remove float64/float32, as they are now in rtl/inc/genmathh.inc
rtl/inc/softfpu.pp: Use provided definitions of float32 and float64
if available (by checking presence of FPC_SYSTEM_HAS_floatXX macro)

rtl/inc/systemh.inc: include genmathh.inc if FPUNONE is not defined.
rtl/java/jsystemh_types.inc: Likewise.

+ readd Randomize with parameter

* import library function with a name that does not conflict with the functions exported in the interface

* Added a simple implementation of random. It is TP/Delphi compatible. Enabled it for 8/16 bit CPUs by default.

* Improved generic implementations of fpc_mul_dword, fpc_mul_word, fpc_mul_byte to use less iterations and registers.

* Optimized Handle_I8086_Error by using smaller int types.

FExpand: always uppercase drive letter also in BasePath

* Fix bug 0037601: add some attribte definitions

* Add StoredAttribute class, bug ID 0037602

* Change some methods from using var to using out (bugMantis 0037376)

* Xtensa: disable simple random again, committed by accident

* fix for broken wince compilation after 0046863

* restore ability to cycle with 3.0.4 (at least until 3.2.2 is released) after 0046890

* make use of AllocMem in fpc_dynarray_setlength

Fix compilation for watcom RTL

* inlinable frame handling dummies for avr
* inline frame handler if possible

Fix NDS RTL compilation for default armv4 CPU, which does not have CLZ instruction

+ introduce TExitCode

* make framecount and framebufsize dependent on the pointer size

* replace some longint by ObjpasInt

* enable the compiler to optimize away parts of System.PushExceptObject, if System.get_frame is not implemented, i.e. returning nil

Use // comments as curly brackets not recognized as comment start inside aarch64 assembly blocks

* adjust softfpu unit so that it can be really used as a standalone unit instead of mainly as part of the System unit

* the define to check for inclusion in the System unit is FPC_IS_SYSTEM, *not* FPC_IN_SYSTEM

Fix compilation failure of i386-netbsd target after last commit

+ initial implementation of a FullName field in file records to overcome length limitions of the name field

fix casting typo in 0047263

Fix compilation of RTL when USE_FILEREC_FULLNAME is not defined

Put GetFullName functions inside $ifdef FPC_WIDESTRING_EQUAL_UNICODESTRING

* protect GetFullName with FPC_HAS_FEATURE_UNICODESTRINGS instead of _WIDESTRINGS

* make sure things still build with 3.0 after 0047353

Adapt to change of chr function in commit 47353, to change chr() a real intrinsics

* Fix bug ID 0038008: allow UTF8 to unicode conversion to react on/ignore invalid input

Add missing CPULLVM versions for fpc_abs_real and fpc_sqr_real in x86_64/math.inc

Fix LLVM compilation by casting untyped real constants to double type, as LLVM does not support extended type constant apparently

Revert wrong change, llvm supports extended constants

Use pascal code for LLVM fpc_sqr_real, because fmultp instruction is not accepted by llvm

* patch by Zoran Vučenović: fixes TDoubleRec.SetFrac, resolves 0038202
+ test

- remove could which prevented that rte is triggered even if we are inside
an exception block but *without* using sysutils. The remove code
was once added to have primitive exception handling even if
no sysutils is used. But if this is desired, an appropriate handler
to ErrorProc should be assigned. Resolves 0038201

* fpc_ln_real raises an exception for invalid operands if they exception is not masked

+ RTMSupport function
+ if available use RTM to support InterlockedCompareExchange128 on i386

* i386: disable InterlockedCompareExchange128 implementation if PIC is enabled

* darwin/AArch64: always clear fpu exception flags regardless of the
triggered signal, as the kernel currently triggers SIGILL rather
than SIGFPE for FPU exceptions

* fixed round(currency) so it takes the rounding mode into account on
platforms where currency is "type int64"
* inline trunc(currency), trunc(comp) and round(comp) on platforms where
currency and comp are "type int64"

* while not official supported, bootstrapping with 3.0.4 fixed

* fpc_AnsiStr_Concat: if one of the strings is empty and CP_NONE is passed, the code page
of the other string is taken, this is also like fpc_AnsiStr_Concat_multi works if length(sarr)=2

* handle rawbytestrings in Win32Ansi2UnicodeMove properly, resolves 0038299

Fix spelling error of revision 48023

Introduce USE_REAL_INSTRUCTIONS to handle expressions not accepted by old assemblers

* do not execute xend twice which was introduced in 0048430

* fixed the parse-collations script/bat file

* fixed the file name in the cldrparser help screen

* updated readme with URLs of the Unicode 9.0 and CLDR 30 data files (to the best of my knowledge)

+ added tool for parsing GraphemeBreakProperty.txt and converting it to code

+ added unit GraphemeBreakProperty, containing the grapheme break property character data from Unicode 9.0

+ added modified LGPL license header to the new graphemebreakproperty.pp unit

+ added and implemented a TUnicodeStringExtendedGraphemeClustersEnumerator class
in the graphemebreakproperty unit - an enumerator that splits a UTF-16 string
into extended grapheme clusters (i.e. user-perceived characters), conforming
to the Unicode 9.0 algorithm:
https://www.unicode.org/reports/tr29/tr29-29.html

+ added an eawparser tool - tool for parsing the EastAsianWidth.txt unicode
property table and converting it to code

+ added unit EastAsianWidth, which contains the EastAsianWidth unicode data property

Fixed free_fun function signature in zlib

fix! 2d00cab5

+ RiscV32: setjmp/longjmp implementation, based on the RiscV64 one

* compilation on Windows fixed

* RiscV32: fpc_longjmp needs nostackframe directive
* RiscV: unified procedure directives of fpc_*jmp

* overflow checked mul helpers do not need to check all bits either

* instead of using discardresult, wrap the uniquestring functions in procedures,
that are declared as inline

* JVM compilation fixed

* support building with FPC 3.2.2

Add missing go32v2 zips and corresponding sources

Update EMX packages

Update OS/2 packages

* rework the basic event on Windows so that peventstate directly represents the Event handle

Replace p2fp by p2fm as short version for pas2fpm utils package and add some missing -3.3.1 to long source zip names

Remove double source entry for utils lexyacc package

Use utils also for short version for OS/2 and emx

Fix short name for utils-fpcres for OS/2 and EMX

* thumb2: Make fillchar smaller, faster and more reliable. Some gas versions remove the "mov r0,r0" making the calculated branch break

* thumb2: Optimize fillchar a bit more with a wider inner loop chunk size

+ Aarch64: use LSE if available for atomic intrinsics

+ lazarus project

* Darwin/BSD: fixed StackBottom calculation

rtl: make STACK_MARGIN a variable instead of a const. this allows adjusting margin size on system unit init, based on the actual stack length of the executable. useful for small systems running with little stack. allow overwriting SysBackTraceStr with a platform-specific implementation

atari: implemented stacktop for working stacktraces, implemented stacksize dependent stackmargin, and a system specific backtracestr. however, offs display still doesn't work properly, due to a linker issue (reported)

* insert explicit type cast for the count parameter of move/fillchar wrapping C functions,
avoids a warning in combination with -O3 and negative count values

* fix bootstrapping on aarch64-linux with 3.2.x

+ added Get/SetTextAutoFlush feature

* LLVM does not have an intrinsic for int/frac, so don't replace frac with
a call to runerror(207) there

* fpcr and fpsr are 64 bit on aarch64

* RiscV: unified cpu initialization and FPU exception handling, resolves 0038893

* Windows: Prevent removal of the TLS directory by the linker when RTL is compiled with -O4 or when inlining is enabled.

* as 3.2.0 is now the oldest compiler we support we can now use {$push} {$pop} to reset the $minenumsize in the RTTI header

Exchange REQUIREDVERSION and REQUIREDVERSION2 values to get 3.2.2 version in reuired release version message

* win64: factored out cpu specific code

* BitSizeOf Documentation sub (issue 0039110)

+ CPU.BMI1Support and CPU.BMI2Support

+ AVX512*Support functions
* call cpuid with eax=7 only if it is supported by the CPU

* x86-64: removed some fwait instructions from fpc_frac_real, they shouldn't be necessary

* x86-64: fpc_frac_real further improved: memory loads of extendeds are expensive

- x86_64: some more fwaits removed

* Gitlab client + example

* Some additional filters

* only .pp is needed

* extended routines int/frac/trunc shouldn't mess with precision or exception flags, should fix gitlab 0039275

* moved fwait

* moved fwait

* Add error check

* replace MT random generator by Xoshiro128**, resolves gitlab 0038237

+ CPU units: check for SHA support

* fixed padding in the filerec to match textrec even on platforms where
sizeof(pointer)<>sizeof(codepointer), like in the medium and compact i8086
memory models

* x86-64: do not jump in inc/declocked to return but return directly

* when compiling with the main branch compiler, p2align with 3 parameters can be used now

* the libc unit might be deprecated, but it's no reason for wrong declarations, in this case values that should be declared as octal but were declared as decimal with the same literal text

Fix compilation of nds and wii OS rtl

* fixed the ptccrt key codes for the ctrl+numpad keys

* Add missing definitions

* Add crypto.h functions

+ initial files for intrinsics support on AArch64

FreeBSD CSU fixes

* Assembler blocks in non-pure assembler functions must always declare all
changed registers
* argc is read as *(long*) in C -> changed plongint to pptrint
* fixed ident section

Also add short format version of build unit BUnit.pp to ignore list

* updated ptcpas and ptcgraph to the latest 0.99.16 release candidate

* ptcpas update: support the F13..F35 function keys on platforms that recognize them

* ptcpas keyboard handling improvements for macOS and X11

* TObject.InitInstance got very big so inlining has no advantage anymore, resolves gitlab 0039494

* patch by Rika: second part of gitlab 0039496, resolves gitlab 0030496
+ extended test

* libffi: fix abi declarations for x86_64

+ LZCNTSupport
+ test extended

* out and var are reserved words

* Prepend T when writing pointer types and PrependTypes is set

+ x86: ADXSupport
+ x86: RDSEEDSupport

* Remove debugging statements

* Add aliases

* Add options to request

* fixes for fpc_Val_SInt_ShortStr by Bart B
+ new test for fpc_Val_SInt_ShortStr
+ benchmark

* reordered condition so they can be evaluated quicker

Java: adapt FPC_ANSISTR_UNIQUE signature

Conform to change made in 800bb3adc23

* updated version by Bart B

* optimze case
* few formatting changes

* fpc_Val_SInt_ShortStr: bug fixes and improvements by Bart B

* patch by Bart B to fix the value of Code for unsigned types if the input is negative, resolves gitlab 0039523

* patch by Bart B to fix fpc_val_smallint_shortstr similiar to gitlab 0039406, resolves gitlab 0039528

o patch by Bart B:
* fixes fpc_Val_UInt_Shortstr, part of gitlab 0015633
* fpc_val_qword_shortstr uses the ValueArray constant

* unified ValueArray

* patch by Bart B: fixes fpc_val_word_shortstr() and refactors fpc_val_longword_shortstr() so that it uses the same algorithm and naming conventions as the other unsigned val-helpers.

* all val helpers now use the ValValueArray

* some fixes for 8 Bit CPUs

Fix compilation of jvm-java and jvm-android rtl

* fpc_AnsiStr_To_ShortStr: set length before moving, should result in better memory access pattern

+ routines to test for AVX512VNNI and AVX512BITALG support

* avoid warning/error on uninitialized variable

* more false positives of unintialized varibles worked around

* rtl/atari: get rid of the old prt0.as startup code, which is not in use any more. this way, we do not require GAS any more for a build

Added TranslatePlaceholderCP before Wide2AnsiMoveProc call.
In the fpc_setstring_ansistr_pwidechar procedure, before calling Wide2AnsiMoveProc, the translation cp = 0 to the DefaultSystemCodePage has been added.
Note: In all other places such translation is present.

Added TranslatePlaceholderCP in fpc_Read_Text_AnsiStr procedure.
This is a small optimization.

atari: implement ARGV protocol for runtime startup

rtl/atari: check for application/acc at startup

* rtl/atari: use the haltproc declared in the startup code instead of calling pterm in the system unit directly.

* rtl/atari: fix haltproc external name. apparently the atari target uses a different Cprefix in the compiler, so copy-pasting this stuff from the Amiga port don't work 1:1

Set float type information in system interface part for mips64 and mips64el CPUs

rtl/atari: implement environment functions also in dos unit

* rtl/atari: add the standalone softfpu units to the Atari build. this fixes the Atari native compiler build after recent changes

atari/sysos: use the symbolic names for error numbers

rtl/atari: fix readln from console

+ skeletons for mips64el specific files

+ store also ecx and edx after cpuid call with eax=7
+ test VNNI and BIGALG instructions

* m68k: implement basic support for interlocked functions using a TAS based global spinlock on CPUs which don't have the CAS instruction

+ skeleton

* setting of MIPS64(EL) in system.inc fixed

* m68k: some CPUs don't have TAS either (for example some CF)

rtl/atari: move common code of getenv

rtl/atari: try to convert argv[0] to absolute pathname

atari: support ARGV protocol in ExecuteProcess

* fix declocked(longint) for aarch64 without LSE, resolves gitlab 0039569

* m68k: fix an instruction which doesn't exist on ColdFire

- removed superfluous comment

Add missing rtl/mips64 files and fix some rtl/mips64el files

* handle frac(+/-Inf or Nan) correctly in the software implementation, resolves gitlab 0039584

* improved fpc_frac_real

* Extra options for more flexibility

* revert a650d346f1646c1b71380430a96c440e222a087d for now as it apparently does
not work with all language encoding

* nds: the embedded stub DOS unit was just a copy of the NDS one with minor fixes, so lets make the NDS port use the current cleaned one as well

+ AVX512VBMISupport and AVX512VBMI2Support

* i386: make use of *fence instructions

* adapted to 64 bit

Make pas2jni use cthreads on Unices

+ introduced PTCKEY_LESS (the 102th key on international keyboards, <> when
used with US keyboard layout)

* use v*csr instructions if they are available

* initialized some fields and data to -1 to make the results more similar to the slave chm (merged_chapter/example1/slave1) example.

Now only
- the checksum at $4
- the number of topics (different issue)
- the offset of framename in #strings differs.

Fix compilation by removing unicodenumtable unit

* CGI demo for css minimizer/classname extractor

* check more cpuid flags in the cpu units

* Changed -1 to $FFFFFFFF on DWord writes to suppress range check errors

* Add examples for syshelpers from Zeljko Avramovic. Fix issue gitlab 0039541

* Libnettle & dynamically loadable version of libgmp

* Add libnettle into build

* Add license message

* Fix library name

+ some more cpu specfic files for mips64(el)

+ enable the 64-bit SAR instruction on the WebAssembly target

* cleanup

Fix compilation ofrtl/nds/classes.pp unit

* align memory blocks in cmem at least to stack alignment boundaries, resolves gitlab 0039730

Remove accidently added binaries in commit 0e142c00

* patch by Rika: SwapEndian([u]int16) has unnecessary “and $FFFF”, resolves gitlab 0039814

Revert " * patch by Rika: SwapEndian([u]int16) has unnecessary “and $FFFF”, resolves gitlab 0039814"

This reverts commit f92102a5f22ae7b978ecf91d59781e16f0b82163.

The and $ffff is needed to avoid range check warnings when the function is inlined with a constant parameter

* x86-64: math helpers do not need a stackframe

* proper naming

FIX: DARWIN: aarch64: tstatfs.fstypename length error (gitlab 0039873)

Add DISABLE_SYSTEMINLINE check to be able to avoid setting SYSTEMINLINE macro

+ add platform independent helper routines for the atomic intrinsics; these only require the implementation of the size specific fpc_atomic_cmp_xhg_* helper, but if only the CPU size specific helper is implemented the others will at least be safe for multi threading inside the same process

* export DynArrayAssign, resolves gitlab 0039897

Rewrite fpc_shortstr_concat_multi for AVR to not use a temporary shortstring to reduce RAM usage.

Make Utf8CodepointLen adhere to the Unicode standard somewhat more and know all of the Unicode 15.0 combining characters.

powerpc64le: fix reading FPU control word after gitlab 000758aa1143

Also, missing rest of the commit message for that change:

The default implementation of SysResetFPU now sets the FPU control word that
was last set from any thread (Delphi and FPC i386/x86-64-compatible). No longer
call SysInitFPU when starting a new thread, only SysResetFPU. Combined, this
means that new FPC-started threads will always start out with that control
word on all architectures.

Added generic fpc_cpuinit implementation that is used unless
FPC_SYSTEM_HAS_FPC_CPUINIT has been defined.

Ensure that the softfloat exception mask is initially set to the same
value as the hardware fpu exception mask on architectures that use both.

SafeLoadLibrary now saves/restores the FPU control word on all platforms.

Removed outdated fpc_mtfsb0_corrected define check in powerpc64.inc

Resolves gitlab 0038230

Thumb(2): fix compilation after FPU control word changes

Resolves gitlab 0039966

Change wrong "$macros off" into "$macro off"

Darwin/AArch64: detect when SIGILL indicates an FPU exception

Parse the ESR (ESR_ELx, Exception Syndrome Register (ELx)), return run error
as in float_raise

Darwin/AArch64: future-proof version check

* Add gstreamer library import unit (gst)

* Remove wrong unit

Introduce CTYPES_INLINE, enabled by default but which can be disabled using -dDISABLE_INLINE for ctypes unit

add loongarch64 rtl support

Fix potential overflow when incrementing 8 bit high value of res.

add ppsuf so that we can compiler loongarch fpc now

+ CMOVSupport function

* new cpu feature tests

Fix use of unaligned for big endian code

* first (modified, added a comment) patch by Rika to improve generic Index*, part of gitlab 0039996

* Add stub for IsManagedType

* Add IsConstValue

Add PIC code for solaris/i386/start.inc

- remove apparently accidently added files

* Free cached queries when committing

* Remove debug statement

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar

* Char -> AnsiChar