Re: [FFmpeg-devel] [PATCH v3 2/5] build: detect Metal.framework and build .metal files

[Aman Karmani <ffmpeg@tmm1.net>]

On Fri, Dec 17, 2021 at 12:54 AM Martin Storsjö <martin@martin.st> wrote:

> On Thu, 16 Dec 2021, Aman Karmani wrote:
>
> > From: Aman Karmani <aman@tmm1.net>
> >
> > Signed-off-by: Aman Karmani <aman@tmm1.net>
> > ---
> > .gitignore         | 3 +++
> > configure          | 8 +++++++-
> > ffbuild/common.mak | 9 +++++++++
> > 3 files changed, 19 insertions(+), 1 deletion(-)
> >
> > diff --git a/.gitignore b/.gitignore
> > index 9ed24b542e..1a5bb29ad5 100644
> > --- a/.gitignore
> > +++ b/.gitignore
> > @@ -19,6 +19,9 @@
> > *.swp
> > *.ver
> > *.version
> > +*.metal.air
> > +*.metallib
> > +*.metallib.c
> > *.ptx
> > *.ptx.c
> > *.ptx.gz
> > diff --git a/configure b/configure
> > index 5fffcb8afe..ab00b2d7cb 100755
> > --- a/configure
> > +++ b/configure
> > @@ -309,6 +309,7 @@ External library support:
> >                            if openssl, gnutls or libtls is not used [no]
> >   --enable-mediacodec      enable Android MediaCodec support [no]
> >   --enable-mediafoundation enable encoding via MediaFoundation [auto]
> > +  --disable-metal          disable Apple Metal framework [autodetect]
> >   --enable-libmysofa       enable libmysofa, needed for sofalizer filter
> [no]
> >   --enable-openal          enable OpenAL 1.1 capture support [no]
> >   --enable-opencl          enable OpenCL processing [no]
> > @@ -382,6 +383,7 @@ Toolchain options:
> >   --dep-cc=DEPCC           use dependency generator DEPCC [$cc_default]
> >   --nvcc=NVCC              use Nvidia CUDA compiler NVCC or clang
> [$nvcc_default]
> >   --ld=LD                  use linker LD [$ld_default]
> > +  --metalcc=METALCC        use metal compiler METALCC [$metalcc_default]
> >   --pkg-config=PKGCONFIG   use pkg-config tool PKGCONFIG
> [$pkg_config_default]
> >   --pkg-config-flags=FLAGS pass additional flags to pkgconf []
> >   --ranlib=RANLIB          use ranlib RANLIB [$ranlib_default]
> > @@ -2564,6 +2566,7 @@ CMDLINE_SET="
> >     ln_s
> >     logfile
> >     malloc_prefix
> > +    metalcc
> >     nm
> >     optflags
> >     nvcc
> > @@ -3835,6 +3838,7 @@ host_cc_default="gcc"
> > doxygen_default="doxygen"
> > install="install"
> > ln_s_default="ln -s -f"
> > +metalcc_default="xcrun metal"
> > nm_default="nm -g"
> > pkg_config_default=pkg-config
> > ranlib_default="ranlib"
> > @@ -4435,7 +4439,7 @@ if enabled cuda_nvcc; then
> > fi
> >
> > set_default arch cc cxx doxygen pkg_config ranlib strip sysinclude \
> > -    target_exec x86asmexe
> > +    target_exec x86asmexe metalcc
> > enabled cross_compile || host_cc_default=$cc
> > set_default host_cc
> >
> > @@ -6326,6 +6330,7 @@ check_apple_framework CoreFoundation
> > check_apple_framework CoreMedia
> > check_apple_framework CoreVideo
> > check_apple_framework CoreAudio
> > +check_apple_framework Metal
> >
> > enabled avfoundation && {
> >     disable coregraphics applicationservices
> > @@ -7620,6 +7625,7 @@ ARFLAGS=$arflags
> > AR_O=$ar_o
> > AR_CMD=$ar
> > NM_CMD=$nm
> > +METALCC=$metalcc
> > RANLIB=$ranlib
> > STRIP=$strip
> > STRIPTYPE=$striptype
> > diff --git a/ffbuild/common.mak b/ffbuild/common.mak
> > index 0eb831d434..05440911f4 100644
> > --- a/ffbuild/common.mak
> > +++ b/ffbuild/common.mak
> > @@ -112,6 +112,15 @@ COMPILE_LASX = $(call COMPILE,CC,LASXFLAGS)
> > $(BIN2CEXE): ffbuild/bin2c_host.o
> >       $(HOSTLD) $(HOSTLDFLAGS) $(HOSTLD_O) $^ $(HOSTEXTRALIBS)
> >
> > +%.metal.air: %.metal
> > +     $(METALCC) $(patsubst $(SRC_PATH)/%,$(SRC_LINK)/%,$<) -o $@
> > +
> > +%.metallib: %.metal.air
> > +     $(METALCC)lib --split-module-without-linking $(patsubst
> $(SRC_PATH)/%,$(SRC_LINK)/%,$<) -o $@
>
> Hmm, so does this try to run "xcrun metallib" instead of "xcrun metal"? I
> think that can be kinda brittle, e.g. if someone wants to configure a
> custom build env, where METALCC expands to e.g.
> "my-wrapped-metal-compiler.sh".


> I guess it feels a bit boring to need to define two separate variables,
> but if it really is two separate tools, then I think that'd be the best
> for clarity.
>

Good catch, I forgot about this little hack.

I agree and will split the variables.


>
> // Martin
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".