[FFmpeg-devel] [PATCH] nasm support

[Måns Rullgård]

Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:

> Hello,
> this allows compiling FFmpeg asm code with nasm.
> Advantages:
> 1) nasm support OS/2 OMF format
> 2) there's a Cygwin package for nasm
> Tested on 64 bit Linux, compiles and passes tests.
> Obviously it would be a good idea for someone to test it
> on 32 bit...

What was the reason for only supporting yasm in the first place?

> Index: configure
> ===================================================================
> --- configure	(revision 24867)
> +++ configure	(working copy)
> @@ -2347,6 +2347,7 @@
>          strip="lxlite"
>          ln_s="cp -f"
>          add_ldflags -Zomf -Zbin-files -Zargs-wild -Zmap
> +        objformat="obj"
>          SHFLAGS='$(SUBDIR)$(NAME).def -Zdll -Zomf'
>          FFSERVERLDFLAGS=""
>          LIBSUF="_s.a"
> @@ -2569,8 +2570,21 @@
>          elf) enabled debug && append YASMFLAGS "-g dwarf2" ;;
>      esac
>      if ! disabled yasm; then
> -        check_yasm "pabsw xmm0, xmm0" && enable yasm ||
> -            warn "yasm not found, performance will suffer"
> +        check_yasm "pabsw xmm0, xmm0" && enable yasm
> +        if ! enabled yasm ; then
> +            yasmexe="nasm"
> +            if enabled x86_64 ; then
> +                case "$objformat" in
> +                    elf)   objformat=elf64 ;;
> +                    win32) objformat=win64 ;;
> +                esac
> +            fi
> +            YASMFLAGS="-f $objformat -DARCH_$(toupper $subarch)"
> +            enabled     pic           && append YASMFLAGS "-DPIC"
> +            test -n "$extern_prefix"  && append YASMFLAGS "-DPREFIX"
> +            check_yasm "pabsw xmm0, xmm0" && enable yasm ||
> +                warn "yasm/nasm not found, performance will suffer"
> +        fi
>      fi

This diff is not against current svn, and such duplication is not
acceptable.

> @@ -270,6 +270,7 @@
>  %define mova movaps
>  
>  %define Z(x) [r0+mmsize*x]
> +%define Z2(x) [r0+mmsize*x]

Why?

> @@ -388,7 +389,8 @@
>  FFT48_3DN _3dn
>  
>  
> -%define Z(x) [zq + o1q*(x&6)*((x/6)^1) + o3q*(x/6) + mmsize*(x&1)]
> +%define Z(x) [zq + o1q*(x&6) + mmsize*(x&1)]
> +%define Z2(x) [zq + o3q + mmsize*(x&1)]

What is the problem with the existing macro?  The change isn't exactly
obvious to me.

-- 
M?ns Rullg?rd
mans at mansr.com