Ros-dev

ros-dev@reactos.org

1 participants
5793 discussions

[ros-diffs] [jimtabor] 26877: - Fixed arithmetic semantics.
by jimtabor 24 May '07

24 May '07

Hi, Here is a test program. Thanks, James #include <stdio.h> #include <stdbool.h> #include <math.h> #include <float.h> //#define APIENTRY __stdcall #define APIENTRY //#define FASTCALL __fastcall #define FASTCALL #ifdef __GNUC__ #define FLOAT_TO_INT(in,out) \ __asm__ __volatile__ ("fistpl %0" : "=m" (out) : "t" (in) : "st"); #else #define FLOAT_TO_INT(in,out) \ __asm fld in \ __asm fistp out #endif /* the following deal with IEEE single-precision numbers */ #define EXCESS 126L #define SIGNBIT 0x80000000L #define SIGN(fp) ((fp) & SIGNBIT) #define EXP(fp) (((fp) >> 23L) & 0xFF) #define MANT(fp) ((fp) & 0x7FFFFFL) #define PACK(s,e,m) ((s) | ((e) << 23L) | (m)) typedef union // float_long { float f; unsigned long l; char c[4]; } gxf_long; #ifdef _X86_ typedef long FLOATL; #else /* !_X86_ */ typedef float FLOATL; #endif typedef struct _FLOATOBJ { unsigned long ul1; unsigned long ul2; } FLOATOBJ, *PFLOATOBJ; typedef struct _EFLOAT_S { long lMant; long lExp; } EFLOAT_S; //// //// void FASTCALL EF_Negate(EFLOAT_S * efp) { efp->lMant = -efp->lMant; } long FASTCALL EFtoF( EFLOAT_S * efp) { long Mant, Exp, Sign = 0; if (!efp->lMant) return 0; printf("EFtoF %x : %x\n", efp->lMant, efp->lExp); Mant = efp->lMant; Exp = efp->lExp; Sign = SIGN(efp->lMant); //// M$ storage emulation if( Sign ) Mant = -Mant; Mant = ((Mant & 0x3fffffff)>>7); Exp += (EXCESS-1); //// Mant = MANT(Mant); return PACK(Sign, Exp, Mant); } void FASTCALL FtoEF( EFLOAT_S * efp, FLOATL f) { long Mant, Exp, Sign = 0; gxf_long worker; printf("FtoEF\n"); #ifdef _X86_ worker.l = f; // It's a float stored in a long. #else worker.f = f; #endif Exp = EXP(worker.l); Mant = MANT(worker.l); if (SIGN(worker.l)) Sign = -1; //// M$ storage emulation Mant = ((Mant << 7) | 0x40000000); Mant ^= Sign; // invert/mask it Mant -= Sign; // sub it Exp -= (EXCESS-1); //// efp->lMant = Mant; efp->lExp = Exp; } void APIENTRY FLOATOBJ_Neg ( PFLOATOBJ pf ) { EFLOAT_S * efp = (EFLOAT_S *)pf; EF_Negate(efp); } long APIENTRY FLOATOBJ_GetFloat ( PFLOATOBJ pf ) { EFLOAT_S * efp = (EFLOAT_S *)pf; return EFtoF(efp); } void APIENTRY FLOATOBJ_SetFloat( PFLOATOBJ pf, FLOATL f ) { EFLOAT_S * efp = (EFLOAT_S *)pf; FtoEF( efp, f); } long APIENTRY FLOATOBJ_GetLong( PFLOATOBJ pf ) { EFLOAT_S * efp = (EFLOAT_S *)pf; gxf_long f; long l; f.l = EFtoF( efp ); FLOAT_TO_INT(f.f, l); // Let FPP handle it the fasty haxy way. return l; } void APIENTRY FLOATOBJ_SetLong( PFLOATOBJ pf, long l ) { EFLOAT_S * efp = (EFLOAT_S *)pf; gxf_long f; f.f = (float) l; // Convert it now. #ifdef _X86_ FtoEF( efp, f.l ); #else FtoEF( efp, f.f ); #endif } long FT_FixedFromFloat(float f) { short value = f; unsigned short fract = (f - value) * 0xFFFF; return (long)((long)value << 16 | (unsigned long)fract); } double F2F(PFLOATOBJ f) { printf("F2F --> %x : %x\n", f->ul1, f->ul2); if(SIGN(f->ul1)) // negate mant return (double) -(-f->ul1 * pow(2,(double)f->ul2-32)); else return (double) f->ul1 * pow(2,(double)f->ul2-32); } void Test_It(float fx) { FLOATOBJ pf; gxf_long f; f.f = fx; printf("Start 1 %f\n", f.f); printf(" 2 %x\n", f.l); #ifdef _X86_ FLOATOBJ_SetFloat( &pf, f.l ); #else FLOATOBJ_SetFloat( &pf, f.f ); #endif // FLOATOBJ_Neg ( &pf ); f.l = FLOATOBJ_GetFloat ( &pf ); printf("Fin 1 %f\n",f.f); printf("Fin 2 %x\n",f.l); printf("F2F --> %f\n\n",F2F( &pf )); } // // gcc -Os fpms.c -o f -lm // or // gcc -Os -D_X86_ fpms.c -o f -lm // main() { gxf_long f; f.f = 1.0; Test_It(f.f); f.f = -1.0; Test_It(f.f); f.f = 2.0; Test_It(f.f); f.f = -2.0; Test_It(f.f); f.f = 1.5; Test_It(f.f); f.f = -1.5; Test_It(f.f); f.f = 1235.0; Test_It(f.f); f.f = -1235.0; Test_It(f.f); f.f = 1235678.9; // Max it Test_It(f.f); f.f = -1235678.9; Test_It(f.f); }

1 0

Suggestion for making things a bit smaller
by carlo.bramix 23 May '07

23 May '07

When building the release versions of executables, DLLs, etc, I would suggest to add the "-s" switch to the linker flags. It will do stripping automatically and release-mode objects will be a bit smaller. Alternatively, we could "strip -s file.exe" but just adding the flag would be much simpler. Sincerely, Carlo Bramini. ------------------------------------------------------ Passa a Infostrada. ADSL e Telefono senza limiti e senza canone Telecom http://click.libero.it/infostrada

2 1

Re: [ros-dev] QEMU support broken!!!
by carlo.bramix 23 May '07

23 May '07

>From livecd-26861-rel it runs again under QEMU. Last tested image is livecd-26871-rel. Great job! Sincerely, Carlo Bramini. ---------- Initial Header ----------- >From : ros-dev-bounces(a)reactos.org To : "ros-dev" ros-dev(a)reactos.org Cc : Date : Fri, 11 May 2007 17:21:55 +0200 Subject : Re: [ros-dev] QEMU support broken!!! > I just wanted to tell you that I downloaded and tested livecd-26700-rel.7z and it doesn't work yet. > > Sincerely, > > Carlo Bramini > > ---------- Initial Header ----------- > > >From : ros-dev-bounces(a)reactos.org > To : "ReactOS Development List" ros-dev(a)reactos.org > Cc : > Date : Fri, 11 May 2007 07:16:05 +0000 > Subject : Re: [ros-dev] QEMU support broken!!! > > > I also have a problem with booting LiveCD. > > Bugcheck in cmcontrl.c, line 151. > > > > _______________________________________________ > > Ros-dev mailing list > > Ros-dev(a)reactos.org > > http://www.reactos.org/mailman/listinfo/ros-dev > > > > > ------------------------------------------------------ > Passa a Infostrada. ADSL e Telefono senza limiti e senza canone Telecom > http://click.libero.it/infostrada > > > > _______________________________________________ > Ros-dev mailing list > Ros-dev(a)reactos.org > http://www.reactos.org/mailman/listinfo/ros-dev > ------------------------------------------------------ Passa a Infostrada. ADSL e Telefono senza limiti e senza canone Telecom http://click.libero.it/infostrada

1 0

Re: [ros-dev] revision 26772, Mon May 14 03:18:42 2007 UTC
by jimtabor 23 May '07

23 May '07

Hi! > It should probably be > + Exp -= (1 - EXCESS); > in EFtoF() and > + Exp += (1 - EXCESS); > in FtoEF(). The other way round. > > No, this would break compatibility. The way to verify this. Write a driver, install it and send a packet to it, so to activate the test. Have the driver dump to WinDbg. > Suggestion to make it a little clearer: > #define BIAS 127 > Exp -+= (BIAS - 2); I know about the IEEE bias. I'm just using my old 68k code that I'm comfortable with. I'll change this later with the Dc handle stuff I'm still testing. Thanks, James

3 2

Re: [ros-dev] [ekohl] 26843: Update WIDL to rev 20070519.
by Aleksey Bragin 20 May '07

20 May '07

Hi, 2nd stage shows "Enable UsermodePnP failed", which doesn't appear with older WIDL. WBR, Aleksey Bragin. On May 20, 2007, at 3:02 AM, ekohl(a)svn.reactos.org wrote: > Author: ekohl > Date: Sun May 20 03:02:35 2007 > New Revision: 26843 > > URL: http://svn.reactos.org/svn/reactos?rev=26843&view=rev > Log: > Update WIDL to rev 20070519. >

1 0

Win32/Gdi32 Rewrite
by jimtabor 18 May '07

18 May '07

Hi all, On or about September 2006 I was asked by Aleksey (PC) and Alex if I could head up the win32 and gdi32 part of the rewrite. The discussion had taken place on irc. It was important that the rewrite would have one developer working alone (preferred method) or be the leading developer working with a small group of developers (no more than three). The rewrite was needed so that this section of ReactOS would be compatible with the new kernel and the very thing we are trying to do. Be compatible. This included interfaces, structures and the data that was contained in those structures. Just not only that, also dropping in the place of drivers and dlls, the real thing and still be compatible. Yes! Running with ReactOS. I do not recall any requirement for me to write a technical manual or any kind of web wiki. Even a How-To-Black-Box an OS interrupt calling hooky thingy with kmtest driver callback winpooch as an application interface. ReactOS developers should have a healthy tool box. This point on, the discussion is over. I have a job to do. I would like help and not debate everything I do. Some of you, need to, really think about, the future of this project. How can I help with this and how can I help make this work for compatibility with the very thing we are working toward. If you are thinking about something new and different other than something innovative, compatible and helpful? Maybe this project is not for you. Aros, FreeMint, Linux, Syllable, Visopsys, X, etc, these projects might be what you are looking for. If it breaks compatibility with interfaces and data stored in structures? I will reject it and post an email why it was rejected. Wine is not enough, okay? Thanks, James

1 0

revision 26772, Mon May 14 03:18:42 2007 UTC
by Timo Kreuzer 16 May '07

16 May '07

It should probably be + Exp -= (1 - EXCESS); in EFtoF() and + Exp += (1 - EXCESS); in FtoEF(). The other way round. Suggestion to make it a little clearer: #define BIAS 127 Exp -+= (BIAS - 2); jimtabor wrote: > --- trunk/reactos/dll/win32/gdi32/objects/painting.c 2006/12/28 22:36:53 25235 > +++ trunk/reactos/dll/win32/gdi32/objects/painting.c 2007/05/14 03:18:42 26772 > @@ -1,15 +1,66 @@ > [...] > +/* the following deal with IEEE single-precision numbers */ > +#define EXCESS 126L > +#define SIGNBIT 0x80000000L > +#define SIGN(fp) ((fp) & SIGNBIT) > +#define EXP(fp) (((fp) >> 23L) & 0xFF) > +#define MANT(fp) ((fp) & 0x7FFFFFL) > +#define PACK(s,e,m) ((s) | ((e) << 23L) | (m)) > [...] > +LONG > +FASTCALL > +EFtoF( EFLOAT_S * efp) > +{ > + long Mant, Exp, Sign = 0; > + > + if (!efp->lMant) return 0; > + > + Mant = efp->lMant; > + Exp = efp->lExp; > + Sign = SIGN(Mant); > + > +//// M$ storage emulation > + Mant = ((Mant & 0x3fffffff) >> 7); > + Exp += (1 - EXCESS); > +//// > + Mant = MANT(Mant); > + return PACK(Sign, Exp, Mant); > +} > + > +VOID > +FASTCALL > +FtoEF( EFLOAT_S * efp, FLOATL f) > +{ > + long Mant, Exp; > + gxf_long worker; > + > +#ifdef _X86_ > + worker.l = f; // It's a float stored in a long. > +#else > + worker.f = f; > +#endif > + > + Exp = EXP(worker.l); > + Mant = MANT(worker.l); > + > +//// M$ storage emulation > + Mant = ((Mant << 7) | 0x40000000 | SIGN(worker.l)); > + Exp -= (1 - EXCESS); > +//// > + efp->lMant = Mant; > + efp->lExp = Exp; > +} >

1 0

Re: [ros-dev] Bug into log10 function.
by carlo.bramix 16 May '07

16 May '07

Unfortunately, I just tested it into the latest working SVN build under QEMU (livecd-26665-rel) and this bug it is still into the CRT library. I also read the source code with SVN-WEB and it's still the same of 0.3.1. BTW, from livecd-26771-rel, a bad thing happens with QEMU: immediately after starting the boot, a blu screen appear with errors and everything is dead. Sincerely, Carlo Bramini ---------- This bug have already been fixed by me. log time ago. some time after 0.3.1 was relase or under 0.3.1 release process. plese use svn source code, and try make own build of reactos as well. the trunk is changes so fast, the code in 0.3.1 is already outdated in many areas. Citerar "carlo\\.bramix" <carlo.bramix at libero.it>: > While I was doing the final tests with ReactOS Calc before releasing the new > version, I discovered that it didn't work because a bug into log10 function. > This happens because the two operands are swapped into: > > \ReactOS-0.3.1\lib\crt\math\i386\log10_asm.s > > I attached the patch. > > Sincerely, > > Carlo Bramini. > > > > > ------------------------------------------------------ > Leggi GRATIS le tue mail con il telefonino i-mode di Wind > http://i-mode.wind.it/ > > ------------------------------------------------------ Leggi GRATIS le tue mail con il telefonino i-mode di Wind http://i-mode.wind.it/

1 0

Re: [ros-dev] Differences in Windows and Linux builds
by hto＠mail.cnt.ru 13 May '07

13 May '07

>> - At the boot screen, the version compiled under Windows shows >> "Service Pack 1" correctly, but the version compiled under Linux >> shows "Service Pac 1" > > I see this "Service Pac 1" in ReactOS compiled under Windows with GCC 4.1.2 (from RosBE). In ExpInitializeExecutive(): ... ANSI_STRING CsdString; ... RtlInitAnsiString(&CsdString, MsgEntry->Text); CsdString.Length -= sizeof(UNICODE_NULL); ... Why Unicode?

2 1

reactos.c compilation error
by Mikhail Y. Zvyozdochkin 13 May '07

13 May '07

Hi All Latest revisions of ReactOS produces compilation error at my system. Build Environment: RosBE 0.3.5b2 File: boot\freeldr\freeldr\reactos\reactos.c Error message (after 'make clean' and 'make'): cc1.exe: warnings being treated as errors boot\freeldr\freeldr\reactos\reactos.c: In function 'LoadAndBootReactOS': boot\freeldr\freeldr\reactos\reactos.c:524: warning: dereferencing type-punned pointer will break strict-aliasing rules mingw32-make: *** [obj-i386\boot\freeldr\freeldr\reactos\reactos.o] Error 1 WBR, DarkHobbit

1 0

Jump to page:

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Ros-dev