re: netbsd-7 on intel GPU (improvements and regressions)

[matthew green <mrg%eterna.com.au@localhost>]

> At last I found why it happens on i386 and not amd64; the attached
> patch makes it work on i386 for me.
> On amd64 version, an assembly version of the functions in
> dist/src/mapi/entry.c is used only when GLX_USE_TLS is defined.
> On i386, there are assembly versions for both GLX_USE_TLS and !GLX_USE_TLS
> cases. I guess the assembly version is wrong, but I've not looked closely.
> The attached patch makes i386 use the C version as amd64 does, and makes it
> work.
> 
[ ... ]

> --- dist/src/mapi/entry.c	5 Jan 2015 20:59:09 -0000	1.1.1.1.2.2
> +++ dist/src/mapi/entry.c	16 Jan 2015 12:06:55 -0000
> @@ -39,7 +39,7 @@
>  #define ENTRY_CURRENT_TABLE_GET U_STRINGIFY(u_current_get_table_internal)
>  #endif
>  
> -#if defined(USE_X86_ASM) && defined(__GNUC__)
> +#if defined(USE_X86_ASM) && defined(__GNUC__) && defined(GLX_USE_TLS)
>  #   ifdef GLX_USE_TLS
>  #      include "entry_x86_tls.h"
>  #   else                 

thanks for figuring this out.  i've commited a change to disable
the 32 bit x86 asm code in entry.c until some x86 guru can look at
it and figure why it (and the 64 bit code that i already have
tested as non-working) is broken.

with that done, how about ... anyone have any good ideas why eg:

   % env LIBGL_ALWAYS_SOFTWARE=1 glxinfo

always crashes when exiting?  it seems to be inside a dtor, but i
didn't get anything useful out of gdb.


.mrg.