April 2011 - Wine-Devel - list.winehq.org

Re: wineoss.drv: Add mmdevapi driver.

by Gerald Pfeifer

I believe commit be332326ba8fc3def406c5f29adf04fbe9a83976 Author: Andrew Eikum <aeikum(a)codeweavers.com> Date: Wed Apr 27 09:12:36 2011 -0500 is causing the following build failures on my nightly FreeBSD testers: mmdevdrv.c:463:20: error: 'AFMT_S24_PACKED' undeclared (first use in this function) mmdevdrv.c:476:16: error: 'AFMT_FLOAT' undeclared (first use in this function) mmdevdrv.c:854:24: error: 'AFMT_FLOAT' undeclared (first use in this function) mmdevdrv.c:863:24: error: 'AFMT_S24_PACKED' undeclared (first use in this function) mmdevdrv.c:1084:24: error: 'SNDCTL_DSP_HALT' undeclared (first use in this function) Looks like some autoconfigury is not working or incomplete? Gerald

April 29, 2011

re: msvbvm60: add stub dll

by Dan Kegel

Austin wrote: > Fixes http://bugs.winehq.org/show_bug.cgi?id=19816 That bug is "Add implementation of MSVBVM60". I don't think a stub fixes it. What apps benefit from having a stub? - Dan

April 29, 2011

Re: [PATCH] dinput: Mark internal symbols as hidden

by Marcus Meissner

On Fri, Apr 29, 2011 at 08:12:36PM +0200, André Hentschel wrote: > Am 29.04.2011 16:18, schrieb Marcus Meissner: > > On Fri, Apr 29, 2011 at 07:38:30AM -0600, Vitaliy Margolen wrote: > >> On 04/29/2011 01:31 AM, Marcus Meissner wrote: > > > > will generate without the visibility line a PLT relocation resolving call: > > > > 0000045c <f>: > > 45c: 55 push %ebp > > 45d: 89 e5 mov %esp,%ebp > > 45f: 53 push %ebx > > 460: 83 ec 04 sub $0x4,%esp > > 463: e8 ef ff ff ff call 457 <__i686.get_pc_thunk.bx> > > 468: 81 c3 8c 1b 00 00 add $0x1b8c,%ebx > > 46e: e8 15 ff ff ff call 388 <g(a)plt> > > 473: 83 c4 04 add $0x4,%esp > > 476: 5b pop %ebx > > 477: 5d pop %ebp > > 478: c3 ret > > > > > > but with the visibility hidden just a relative call, no relocations: > > > > 0000042c <f>: > > 42c: 55 push %ebp > > 42d: 89 e5 mov %esp,%ebp > > 42f: 83 ec 08 sub $0x8,%esp > > 432: e8 05 00 00 00 call 43c <g> > > 437: c9 leave > > 438: c3 ret > > > > 0000043c <g>: > > > > Ciao, Marcus > > > > Hi Marcus, > That's a very cool search&replace job, Wine reduces size and get's faster. > The question is how much faster? Hope we'll see some benchmarks from you :) I do not think there will be noticable speed-up. It however has the possibility to make funny Windows code who has troubles with %ebx changes work better. Also the size of the .sos gets smaller a bit (but just in the kilobyte range). Ciao, Marcus

April 29, 2011

Re: [PATCH] dinput: Mark internal symbols as hidden

by André Hentschel

Am 29.04.2011 16:18, schrieb Marcus Meissner: > On Fri, Apr 29, 2011 at 07:38:30AM -0600, Vitaliy Margolen wrote: >> On 04/29/2011 01:31 AM, Marcus Meissner wrote: > > will generate without the visibility line a PLT relocation resolving call: > > 0000045c <f>: > 45c: 55 push %ebp > 45d: 89 e5 mov %esp,%ebp > 45f: 53 push %ebx > 460: 83 ec 04 sub $0x4,%esp > 463: e8 ef ff ff ff call 457 <__i686.get_pc_thunk.bx> > 468: 81 c3 8c 1b 00 00 add $0x1b8c,%ebx > 46e: e8 15 ff ff ff call 388 <g(a)plt> > 473: 83 c4 04 add $0x4,%esp > 476: 5b pop %ebx > 477: 5d pop %ebp > 478: c3 ret > > > but with the visibility hidden just a relative call, no relocations: > > 0000042c <f>: > 42c: 55 push %ebp > 42d: 89 e5 mov %esp,%ebp > 42f: 83 ec 08 sub $0x8,%esp > 432: e8 05 00 00 00 call 43c <g> > 437: c9 leave > 438: c3 ret > > 0000043c <g>: > > Ciao, Marcus > Hi Marcus, That's a very cool search&replace job, Wine reduces size and get's faster. The question is how much faster? Hope we'll see some benchmarks from you :) -- Best Regards, André Hentschel

April 29, 2011

Fix WineHQ bug #19762

by Joerg-Cyril.Hoehle＠t-systems.com

Kai Wasserbäch wrote: >as bug #19762 is open for some time now, fully bisected, the solution applied for >several versions in my Debian packages ([0]) without reports about breakage, I >hearby propose the reversion of commit 67631163. http://bugs.winehq.org/show_bug.cgi?id=19762#c21 That commit is another example of a commit that introduced a regression while fixing another regression in another app. This does not prove that commit wrong, however, the burden is now on you (us) to prove it in order to change git. It's not an impossible task. Even code that seemed to be backed by tests in Wine was changed when the tests were proven wrong/incomplete. For a start, it may be worth testing that other app and see if it still needs that commit in current Wine. Does anybody know the bug number of that other app's original issue? Finally, are there any tests that support either behavior? Regards, Jörg Höhle

April 29, 2011

Re: [PATCH] dinput: Mark internal symbols as hidden

by Chris Robinson

On Friday, April 29, 2011 7:18:47 AM Marcus Meissner wrote: > If they are not declared hidden, the -fPIC compile generates PLT > relocations for those symbols, even if they are internal to the dll (the > compiler does not know what to export during our build). > > With the hidden attribute these get turned into just relative calls or > simpler relocations. Wouldn't it be better to compile with -fvisibility=hidden, then? Default to hidden, then explicitly mark the functions that can be exported from the .so (if Wine even needs PE exports to have "default" visibilty; I don't think it does). Functions that are *never* called from outside the module may be better marked as "internal": "Internal visibility is like hidden visibility, but with additional processor specific semantics. Unless otherwise specified by the psABI, GCC defines internal visibility to mean that a function is never called from another module. Compare this with hidden functions which, while they cannot be referenced directly by other modules, can be referenced indirectly via function pointers. By indicating that a function cannot be called from outside the module, GCC may for instance omit the load of a PIC register since it is known that the calling function loaded the correct value." http://gcc.gnu.org/onlinedocs/gcc-4.6.0/gcc/Function-Attributes.html#Functi… Attributes FWIW, you can even mark COM functions as "hidden". As noteded above, hidden functions can still be called from the outside (through function pointers, like vtables), they just aren't able to be directly referenced.

April 29, 2011

Re: Fix WineHQ bug #19762

by Henri Verbeet

On 29 April 2011 12:29, Kai Wasserbäch <curan(a)debian.org> wrote: > several versions in my Debian packages ([0]) without reports about breakage, I > hearby propose the reversion of commit 67631163. Wylda tried to reach the I don't think reverting that commit is the right thing to do, it's just going to break something else instead.

April 29, 2011

Re: [PATCH] dinput: Mark internal symbols as hidden

by Marcus Meissner

On Fri, Apr 29, 2011 at 07:38:30AM -0600, Vitaliy Margolen wrote: > On 04/29/2011 01:31 AM, Marcus Meissner wrote: > >--- > > dlls/dinput/device_private.h | 148 +++++++++++++++++++++--------------------- > > 1 files changed, 74 insertions(+), 74 deletions(-) > > What exactly is this supposed to to? This is inside header file, > everything declared in it is internal to dinput. Why do we need some > extra declarations on it? If they are not declared hidden, the -fPIC compile generates PLT relocations for those symbols, even if they are internal to the dll (the compiler does not know what to export during our build). With the hidden attribute these get turned into just relative calls or simpler relocations. On x86 this will also save use of the %ebx register. So e.g. this piece of code: extern int g(void) __attribute__((__visibility__("hidden"))); int f() { return g(); } will generate without the visibility line a PLT relocation resolving call: 0000045c <f>: 45c: 55 push %ebp 45d: 89 e5 mov %esp,%ebp 45f: 53 push %ebx 460: 83 ec 04 sub $0x4,%esp 463: e8 ef ff ff ff call 457 <__i686.get_pc_thunk.bx> 468: 81 c3 8c 1b 00 00 add $0x1b8c,%ebx 46e: e8 15 ff ff ff call 388 <g(a)plt> 473: 83 c4 04 add $0x4,%esp 476: 5b pop %ebx 477: 5d pop %ebp 478: c3 ret but with the visibility hidden just a relative call, no relocations: 0000042c <f>: 42c: 55 push %ebp 42d: 89 e5 mov %esp,%ebp 42f: 83 ec 08 sub $0x8,%esp 432: e8 05 00 00 00 call 43c <g> 437: c9 leave 438: c3 ret 0000043c <g>: Ciao, Marcus

April 29, 2011

Re: [PATCH] dinput: Mark internal symbols as hidden

by Vitaliy Margolen

On 04/29/2011 01:31 AM, Marcus Meissner wrote: > --- > dlls/dinput/device_private.h | 148 +++++++++++++++++++++--------------------- > 1 files changed, 74 insertions(+), 74 deletions(-) What exactly is this supposed to to? This is inside header file, everything declared in it is internal to dinput. Why do we need some extra declarations on it? Vitaliy.

April 29, 2011

Re: [1/5] wined3d: Add missing temporary variable declaration to ARB shader backend.

by Stefan Dösinger

Am 29.04.2011 um 00:39 schrieb Matteo Bruni: > 2011/4/29 Stefan Dösinger <stefandoesinger(a)gmx.at>: >>> + shader_addline(buffer, "TEMP TB;\n"); >> Which shader instruction uses this? Afair vertex shaders should only use TA, >> but there's no proper infrastructure that manages that. >> > > AFAICS, TB could get used in shader_hw_pow(): > > ... > shader_arb_get_src_param(ins, &ins->src[1], 1, src1); > ... > > where the '1' in the third argument means that > shader_arb_get_src_param() is allowed to use TB as temporary storage. > There may be other pieces of code with the same assumption, I didn't > look deeper. Ok

April 29, 2011