@@ -11,7 +11,7 @@ ifndef V

 Q      = @

 ECHO   = printf "$(1)\t%s\n" $(2)

 BRIEF  = CC CXX AS YASM AR LD HOSTCC STRIP CP

-SILENT = DEPCC YASMDEP RM RANLIB

+SILENT = DEPCC DEPAS DEPHOSTCC YASMDEP RM RANLIB

 MSG    = $@

 M      = @$(call ECHO,$(TAG),$@);

 $(foreach VAR,$(BRIEF), \

@@ -26,15 +26,16 @@ ALLFFLIBS = avcodec avdevice avfilter avformat avresample avutil postproc swscal

 IFLAGS     := -I. -I$(SRC_PATH)/

 CPPFLAGS   := $(IFLAGS) $(CPPFLAGS)

 CFLAGS     += $(ECFLAGS)

-CCFLAGS     = $(CFLAGS)

+CCFLAGS     = $(CPPFLAGS) $(CFLAGS)

+ASFLAGS    := $(CPPFLAGS) $(ASFLAGS)

 CXXFLAGS   := $(CFLAGS) $(CXXFLAGS)

 YASMFLAGS  += $(IFLAGS) -I$(SRC_PATH)/libavutil/x86/ -Pconfig.asm

-HOSTCFLAGS += $(IFLAGS)

+HOSTCCFLAGS = $(IFLAGS) $(HOSTCFLAGS)

 LDFLAGS    := $(ALLFFLIBS:%=-Llib%) $(LDFLAGS)

 define COMPILE

-       $($(1)DEP)

-       $($(1)) $(CPPFLAGS) $($(1)FLAGS) $($(1)_DEPFLAGS) -c $($(1)_O) $<

+       $(call $(1)DEP,$(1))

+       $($(1)) $($(1)FLAGS) $($(1)_DEPFLAGS) -c $($(1)_O) $<

 endef

 COMPILE_C = $(call COMPILE,CC)

@@ -101,7 +102,7 @@ checkheaders: $(filter-out $(SKIPHEADERS:.h=.ho),$(ALLHEADERS:.h=.ho))

 alltools: $(TOOLS)

 $(HOSTOBJS): %.o: %.c

-	$(HOSTCC) $(HOSTCFLAGS) -c -o $@ $<

+	$(call COMPILE,HOSTCC)

 $(HOSTPROGS): %$(HOSTEXESUF): %.o

 	$(HOSTCC) $(HOSTLDFLAGS) -o $@ $< $(HOSTLIBS)

@@ -117,4 +118,4 @@ CLEANSUFFIXES     = *.d *.o *~ *.ho *.map *.ver *.gcno *.gcda

 DISTCLEANSUFFIXES = *.pc

 LIBSUFFIXES       = *.a *.lib *.so *.so.* *.dylib *.dll *.def *.dll.a

--include $(wildcard $(OBJS:.o=.d) $(TESTOBJS:.o=.d))

+-include $(wildcard $(OBJS:.o=.d) $(HOSTOBJS:.o=.d) $(TESTOBJS:.o=.d))

@@ -265,6 +265,7 @@ Optimization options (experts only):

   --disable-sse            disable SSE optimizations

   --disable-ssse3          disable SSSE3 optimizations

   --disable-avx            disable AVX optimizations

+  --disable-fma4           disable FMA4 optimizations

   --disable-armv5te        disable armv5te optimizations

   --disable-armv6          disable armv6 optimizations

   --disable-armv6t2        disable armv6t2 optimizations

@@ -1173,6 +1174,7 @@ ARCH_EXT_LIST='

     armv6t2

     armvfp

     avx

+    fma4

     mmi

     mmx

     mmx2

@@ -1336,7 +1338,7 @@ HAVE_LIST="

 # options emitted with CONFIG_ prefix but not available on command line

 CONFIG_EXTRA="

-    aandct

+    aandcttables

     avutil

     golomb

     gplv3

@@ -1450,6 +1452,7 @@ mmx2_deps="mmx"

 sse_deps="mmx"

 ssse3_deps="sse"

 avx_deps="ssse3"

+fma4_deps="avx"

 aligned_stack_if_any="ppc x86"

 fast_64bit_if_any="alpha ia64 mips64 parisc64 ppc64 sparc64 x86_64"

@@ -1477,7 +1480,7 @@ ac3_fixed_encoder_select="mdct ac3dsp"

 alac_encoder_select="lpc"

 amrnb_decoder_select="lsp"

 amrwb_decoder_select="lsp"

-amv_encoder_select="aandct"

+amv_encoder_select="aandcttables"

 atrac1_decoder_select="mdct sinewin"

 atrac3_decoder_select="mdct"

 binkaudio_dct_decoder_select="mdct rdft dct sinewin"

@@ -1487,13 +1490,13 @@ cook_decoder_select="mdct sinewin"

 cscd_decoder_suggest="zlib"

 dca_decoder_select="mdct"

 dirac_decoder_select="dwt golomb"

-dnxhd_encoder_select="aandct"

+dnxhd_encoder_select="aandcttables"

 dxa_decoder_select="zlib"

 eac3_decoder_select="ac3_decoder"

 eac3_encoder_select="mdct ac3dsp"

-eamad_decoder_select="aandct"

-eatgq_decoder_select="aandct"

-eatqi_decoder_select="aandct"

+eamad_decoder_select="aandcttables"

+eatgq_decoder_select="aandcttables"

+eatqi_decoder_select="aandcttables"

 exr_decoder_select="zlib"

 ffv1_decoder_select="golomb"

 flac_decoder_select="golomb"

@@ -1505,9 +1508,9 @@ flashsv2_decoder_select="zlib"

 flv_decoder_select="h263_decoder"

 flv_encoder_select="h263_encoder"

 fraps_decoder_select="huffman"

-h261_encoder_select="aandct"

+h261_encoder_select="aandcttables"

 h263_decoder_select="h263_parser"

-h263_encoder_select="aandct"

+h263_encoder_select="aandcttables"

 h263_vaapi_hwaccel_select="vaapi h263_decoder"

 h263i_decoder_select="h263_decoder"

 h263p_encoder_select="h263_encoder"

@@ -1523,9 +1526,9 @@ iac_decoder_select="fft mdct sinewin"

 imc_decoder_select="fft mdct sinewin"

 jpegls_decoder_select="golomb"

 jpegls_encoder_select="golomb"

-ljpeg_encoder_select="aandct"

+ljpeg_encoder_select="aandcttables"

 loco_decoder_select="golomb"

-mjpeg_encoder_select="aandct"

+mjpeg_encoder_select="aandcttables"

 mlp_decoder_select="mlp_parser"

 mp1_decoder_select="mpegaudiodsp"

 mp1float_decoder_select="mpegaudiodsp"

@@ -1544,13 +1547,13 @@ mpeg_xvmc_decoder_deps="X11_extensions_XvMClib_h"

 mpeg_xvmc_decoder_select="mpegvideo_decoder"

 mpeg1_vdpau_decoder_select="vdpau mpeg1video_decoder"

 mpeg1_vdpau_hwaccel_select="vdpau mpeg1video_decoder"

-mpeg1video_encoder_select="aandct"

+mpeg1video_encoder_select="aandcttables"

 mpeg2_crystalhd_decoder_select="crystalhd"

 mpeg2_dxva2_hwaccel_deps="dxva2api_h"

 mpeg2_dxva2_hwaccel_select="dxva2 mpeg2video_decoder"

 mpeg2_vdpau_hwaccel_select="vdpau mpeg2video_decoder"

 mpeg2_vaapi_hwaccel_select="vaapi mpeg2video_decoder"

-mpeg2video_encoder_select="aandct"

+mpeg2video_encoder_select="aandcttables"

 mpeg4_crystalhd_decoder_select="crystalhd"

 mpeg4_decoder_select="h263_decoder mpeg4video_parser"

 mpeg4_encoder_select="h263_encoder"

@@ -1580,11 +1583,11 @@ rv40_decoder_select="golomb h264chroma h264pred h264qpel"

 shorten_decoder_select="golomb"

 sipr_decoder_select="lsp"

 snow_decoder_select="dwt"

-snow_encoder_select="aandct dwt"

+snow_encoder_select="aandcttables dwt"

 sonic_decoder_select="golomb"

 sonic_encoder_select="golomb"

 sonic_ls_encoder_select="golomb"

-svq1_encoder_select="aandct"

+svq1_encoder_select="aandcttables"

 svq3_decoder_select="golomb h264chroma h264dsp h264pred h264qpel"

 svq3_decoder_suggest="zlib"

 theora_decoder_select="vp3_decoder"

@@ -1965,6 +1968,8 @@ ldflags_filter=echo

 AS_O='-o $@'

 CC_O='-o $@'

 CXX_O='-o $@'

+LD_O='-o $@'

+HOSTCC_O='-o $@'

 host_cflags='-D_ISOC99_SOURCE -D_XOPEN_SOURCE=600 -O3 -g'

 host_libs='-lm'

@@ -1975,8 +1980,8 @@ target_path='$(CURDIR)'

 # since the object filename is not given with the -MM flag, the compiler

 # is only able to print the basename, and we must add the path ourselves

-DEPEND_CMD='$(DEPCC) $(DEPFLAGS) $< | sed -e "/^\#.*/d" -e "s,^[[:space:]]*$(*F)\\.o,$(@D)/$(*F).o," > $(@:.o=.d)'

-DEPFLAGS='$(CPPFLAGS) $(CFLAGS) -MM'

+DEPCMD='$(DEP$(1)) $(DEP$(1)FLAGS) $($(1)DEP_FLAGS) $< | sed -e "/^\#.*/d" -e "s,^[[:space:]]*$(*F)\\.o,$(@D)/$(*F).o," > $(@:.o=.d)'

+DEPFLAGS='-MM'

 # find source path

 if test -f configure; then

@@ -2319,120 +2324,150 @@ tms470_flags(){

     done

 }

-if   $cc -v 2>&1 | grep -q '^gcc.*LLVM'; then

-    cc_type=llvm_gcc

-    gcc_extra_ver=$(expr "$($cc --version | head -n1)" : '.*\((.*)\)')

-    cc_ident="llvm-gcc $($cc -dumpversion) $gcc_extra_ver"

-    CC_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

-    AS_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

-    cflags_speed='-O3'

-    cflags_size='-Os'

-elif $cc -v 2>&1 | grep -qi ^gcc; then

-    cc_type=gcc

-    gcc_version=$($cc --version | head -n1)

-    gcc_basever=$($cc -dumpversion)

-    gcc_pkg_ver=$(expr "$gcc_version" : '[^ ]* \(([^)]*)\)')

-    gcc_ext_ver=$(expr "$gcc_version" : ".*$gcc_pkg_ver $gcc_basever \\(.*\\)")

-    cc_ident=$(cleanws "gcc $gcc_basever $gcc_pkg_ver $gcc_ext_ver")

-    if ! $cc -dumpversion | grep -q '^2\.'; then

-        CC_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

-        AS_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

+probe_cc(){

+    pfx=$1

+    _cc=$2

+

+    unset _type _ident _cc_o _flags _cflags _ldflags _depflags _DEPCMD _DEPFLAGS

+    _flags_filter=echo

+

+    if $_cc -v 2>&1 | grep -q '^gcc.*LLVM'; then

+        _type=llvm_gcc

+        gcc_extra_ver=$(expr "$($_cc --version | head -n1)" : '.*\((.*)\)')

+        _ident="llvm-gcc $($_cc -dumpversion) $gcc_extra_ver"

+        _depflags='-MMD -MF $(@:.o=.d) -MT $@'

+        _cflags_speed='-O3'

+        _cflags_size='-Os'

+    elif $_cc -v 2>&1 | grep -qi ^gcc; then

+        _type=gcc

+        gcc_version=$($_cc --version | head -n1)

+        gcc_basever=$($_cc -dumpversion)

+        gcc_pkg_ver=$(expr "$gcc_version" : '[^ ]* \(([^)]*)\)')

+        gcc_ext_ver=$(expr "$gcc_version" : ".*$gcc_pkg_ver $gcc_basever \\(.*\\)")

+        _ident=$(cleanws "gcc $gcc_basever $gcc_pkg_ver $gcc_ext_ver")

+        if ! $_cc -dumpversion | grep -q '^2\.'; then

+            _depflags='-MMD -MF $(@:.o=.d) -MT $@'

+        fi

+        _cflags_speed='-O3'

+        _cflags_size='-Os'

+    elif $_cc --version 2>/dev/null | grep -q Intel; then

+        _type=icc

+        _ident=$($_cc --version | head -n1)

+        _depflags='-MMD'

+        _cflags_speed='-O3'

+        _cflags_size='-Os'

+        _cflags_noopt='-O1'

+    elif $_cc -v 2>&1 | grep -q xlc; then

+        _type=xlc

+        _ident=$($_cc -qversion 2>/dev/null | head -n1)

+        _cflags_speed='-O5'

+        _cflags_size='-O5 -qcompact'

+    elif $_cc -V 2>/dev/null | grep -q Compaq; then

+        _type=ccc

+        _ident=$($_cc -V | head -n1 | cut -d' ' -f1-3)

+        _DEPFLAGS='-M'

+        debuglevel=3

+        _ldflags='-Wl,-z,now' # calls to libots crash without this

+        _cflags_speed='-fast'

+        _cflags_size='-O1'

+    elif $_cc --vsn 2>/dev/null | grep -q "ARM C/C++ Compiler"; then

+        test -d "$sysroot" || die "No valid sysroot specified."

+        _type=armcc

+        _ident=$($_cc --vsn | head -n1)

+        armcc_conf="$PWD/armcc.conf"

+        $_cc --arm_linux_configure                 \

+             --arm_linux_config_file="$armcc_conf" \

+             --configure_sysroot="$sysroot"        \

+             --configure_cpp_headers="$sysinclude" >>$logfile 2>&1 ||

+             die "Error creating armcc configuration file."

+        $_cc --vsn | grep -q RVCT && armcc_opt=rvct || armcc_opt=armcc

+        _flags="--arm_linux_config_file=$armcc_conf --translate_gcc"

+        as_default="${cross_prefix}gcc"

+        _depflags='-MMD'

+        _cflags_speed='-O3'

+        _cflags_size='-Os'

+    elif $_cc -version 2>/dev/null | grep -q TMS470; then

+        _type=tms470

+        _ident=$($_cc -version | head -n1 | tr -s ' ')

+        _flags='--gcc --abi=eabi -me'

+        _cflags='-D__gnuc_va_list=va_list -D__USER_LABEL_PREFIX__='

+        _cc_o='-fe=$@'

+        as_default="${cross_prefix}gcc"

+        ld_default="${cross_prefix}gcc"

+        _depflags='-ppa -ppd=$(@:.o=.d)'

+        _cflags_speed='-O3 -mf=5'

+        _cflags_size='-O3 -mf=2'

+        _flags_filter=tms470_flags

+    elif $_cc -v 2>&1 | grep -q clang; then

+        _type=clang

+        _ident=$($_cc --version | head -n1)

+        _depflags='-MMD'

+        _cflags_speed='-O3'

+        _cflags_size='-Os'

+    elif $_cc -V 2>&1 | grep -q Sun; then

+        _type=suncc

+        _ident=$($_cc -V 2>&1 | head -n1 | cut -d' ' -f 2-)

+        _DEPCMD='$(DEP$(1)) $(DEP$(1)FLAGS) $($(1)DEP_FLAGS) $< | sed -e "1s,^.*: ,$@: ," -e "\$$!s,\$$, \\\," -e "1!s,^.*: , ," > $(@:.o=.d)'

+        _DEPFLAGS='-xM1'

+        _ldflags='-std=c99'

+        _cflags_speed='-O5'

+        _cflags_size='-O5 -xspace'

+        _flags_filter=suncc_flags

+    elif $_cc -v 2>&1 | grep -q 'PathScale\|Path64'; then

+        _type=pathscale

+        _ident=$($_cc -v 2>&1 | head -n1 | tr -d :)

+        _depflags='-MMD -MF $(@:.o=.d) -MT $@'

+        _cflags_speed='-O2'

+        _cflags_size='-Os'

+        _flags_filter='filter_out -Wdisabled-optimization'

+    elif $_cc -v 2>&1 | grep -q Open64; then

+        _type=open64

+        _ident=$($_cc -v 2>&1 | head -n1 | tr -d :)

+        _depflags='-MMD -MF $(@:.o=.d) -MT $@'

+        _cflags_speed='-O2'

+        _cflags_size='-Os'

+        _flags_filter='filter_out -Wdisabled-optimization|-Wtype-limits|-fno-signed-zeros'

+    elif $_cc -V 2>&1 | grep -q Portland; then

+        _type=pgi

+        _ident="PGI $($_cc -V 2>&1 | awk '/^pgcc/ { print $2; exit }')"

+        opt_common='-alias=ansi -Mlre -Mpre'

+        _cflags_speed="-O3 -Mautoinline -Munroll=c:4 $opt_common"

+        _cflags_size="-O2 -Munroll=c:1 $opt_common"

+        _cflags_noopt="-O1"

+        _flags_filter=pgi_flags

     fi

-    cflags_speed='-O3'

-    cflags_size='-Os'

-elif $cc --version 2>/dev/null | grep -q Intel; then

-    cc_type=icc

-    cc_ident=$($cc --version | head -n1)

-    CC_DEPFLAGS='-MMD'

-    AS_DEPFLAGS='-MMD'

-    cflags_speed='-O3'

-    cflags_size='-Os'

-    cflags_noopt='-O1'

-elif $cc -v 2>&1 | grep -q xlc; then

-    cc_type=xlc

-    cc_ident=$($cc -qversion 2>/dev/null | head -n1)

-    cflags_speed='-O5'

-    cflags_size='-O5 -qcompact'

-elif $cc -V 2>/dev/null | grep -q Compaq; then

-    cc_type=ccc

-    cc_ident=$($cc -V | head -n1 | cut -d' ' -f1-3)

-    DEPFLAGS='$(CPPFLAGS) $(CFLAGS) -M'

-    debuglevel=3

-    add_ldflags -Wl,-z,now # calls to libots crash without this

-    cflags_speed='-fast'

-    cflags_size='-O1'

-elif $cc --vsn 2>/dev/null | grep -q "ARM C/C++ Compiler"; then

-    test -d "$sysroot" || die "No valid sysroot specified."

-    cc_type=armcc

-    cc_ident=$($cc --vsn | head -n1)

-    armcc_conf="$PWD/armcc.conf"

-    $cc --arm_linux_configure                 \

-        --arm_linux_config_file="$armcc_conf" \

-        --configure_sysroot="$sysroot"        \

-        --configure_cpp_headers="$sysinclude" >>$logfile 2>&1 ||

-        die "Error creating armcc configuration file."

-    $cc --vsn | grep -q RVCT && armcc_opt=rvct || armcc_opt=armcc

-    cc="$cc --arm_linux_config_file=$armcc_conf --translate_gcc"

-    as_default="${cross_prefix}gcc"

-    CC_DEPFLAGS='-MMD'

-    AS_DEPFLAGS='-MMD'

-    cflags_speed='-O3'

-    cflags_size='-Os'

-    asflags_filter="filter_out -W${armcc_opt}*"

-elif $cc -version 2>/dev/null | grep -q TMS470; then

-    cc_type=tms470

-    cc_ident=$($cc -version | head -n1 | tr -s ' ')

-    cc="$cc --gcc --abi=eabi -me"

-    CC_O='-fe=$@'

-    as_default="${cross_prefix}gcc"

-    ld_default="${cross_prefix}gcc"

-    add_cflags -D__gnuc_va_list=va_list -D__USER_LABEL_PREFIX__=

-    CC_DEPFLAGS='-ppa -ppd=$(@:.o=.d)'

-    AS_DEPFLAGS='-MMD'

-    cflags_speed='-O3 -mf=5'

-    cflags_size='-O3 -mf=2'

-    cflags_filter=tms470_flags

-elif $cc -v 2>&1 | grep -q clang; then

-    cc_type=clang

-    cc_ident=$($cc --version | head -n1)

-    CC_DEPFLAGS='-MMD'

-    AS_DEPFLAGS='-MMD'

-    cflags_speed='-O3'

-    cflags_size='-Os'

-elif $cc -V 2>&1 | grep -q Sun; then

-    cc_type=suncc

-    cc_ident=$($cc -V 2>&1 | head -n1 | cut -d' ' -f 2-)

-    DEPEND_CMD='$(DEPCC) $(DEPFLAGS) $< | sed -e "1s,^.*: ,$@: ," -e "\$$!s,\$$, \\\," -e "1!s,^.*: , ," > $(@:.o=.d)'

-    DEPFLAGS='$(CPPFLAGS) $(CFLAGS) -xM1'

-    add_ldflags -xc99

-    cflags_speed='-O5'

-    cflags_size='-O5 -xspace'

-    cflags_filter=suncc_flags

-elif $cc -v 2>&1 | grep -q 'PathScale\|Path64'; then

-    cc_type=pathscale

-    cc_ident=$($cc -v 2>&1 | head -n1 | tr -d :)

-    CC_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

-    AS_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

-    cflags_speed='-O2'

-    cflags_size='-Os'

-    cflags_filter='filter_out -Wdisabled-optimization'

-elif $cc -v 2>&1 | grep -q Open64; then

-    cc_type=open64

-    cc_ident=$($cc -v 2>&1 | head -n1 | tr -d :)

-    CC_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

-    AS_DEPFLAGS='-MMD -MF $(@:.o=.d) -MT $@'

-    cflags_speed='-O2'

-    cflags_size='-Os'

-    cflags_filter='filter_out -Wdisabled-optimization|-Wtype-limits|-fno-signed-zeros'

-elif $cc -V 2>&1 | grep -q Portland; then

-    cc_type=pgi

-    cc_ident="PGI $($cc -V 2>&1 | awk '/^pgcc/ { print $2; exit }')"

-    opt_common='-alias=ansi -Mlre -Mpre'

-    cflags_speed="-O3 -Mautoinline -Munroll=c:4 $opt_common"

-    cflags_size="-O2 -Munroll=c:1 $opt_common"

-    cflags_noopt="-O1"

-    cflags_filter=pgi_flags

-fi

+

+    eval ${pfx}_type=\$_type

+    eval ${pfx}_ident=\$_ident

+}

+

+set_ccvars(){

+    eval ${1}_O=\${_cc_o-\${${1}_O}}

+

+    if [ -n "$_depflags" ]; then

+        eval ${1}_DEPFLAGS=\$_depflags

+    else

+        eval ${1}DEP=\${_DEPCMD:-\$DEPCMD}

+        eval ${1}DEP_FLAGS=\${_DEPFLAGS:-\$DEPFLAGS}

+        eval DEP${1}FLAGS=\$_flags

+    fi

+}

+

+probe_cc cc "$cc"

+cflags_filter=$_flags_filter

+cflags_speed=$_cflags_speed

+cflags_size=$_cflags_size

+cflags_noopt=$_cflags_noopt

+add_cflags $_flags $_cflags

+cc_ldflags=$_ldflags

+set_ccvars CC

+

+probe_cc hostcc "$host_cc"

+host_cflags_filter=$_flags_filter

+host_ldflags_filter=$_flags_filter

+add_host_cflags  $_flags $_cflags

+add_host_ldflags $_flags $_ldflags

+set_ccvars HOSTCC

 test -n "$cc_type" && enable $cc_type ||

     warn "Unknown C compiler $cc, unable to select optimal CFLAGS"

@@ -2442,9 +2477,23 @@ test -n "$cc_type" && enable $cc_type ||

 : ${ld_default:=$cc}

 set_default ar as dep_cc ld

-test -n "$CC_DEPFLAGS" || CCDEP=$DEPEND_CMD

-test -n "$CXX_DEPFLAGS" || CXXDEP=$DEPEND_CMD

-test -n "$AS_DEPFLAGS" || ASDEP=$DEPEND_CMD

+probe_cc as "$as"

+asflags_filter=$_flags_filter

+add_asflags $_flags $_cflags

+set_ccvars AS

+

+probe_cc ld "$ld"

+ldflags_filter=$_flags_filter

+add_ldflags $_flags $_ldflags

+test "$cc_type" != "$ld_type" && add_ldflags $cc_ldflags

+LD_O=${_cc_o-$LD_O}

+

+if [ -z "$CC_DEPFLAGS" ] && [ "$dep_cc" != "$cc" ]; then

+    probe_cc depcc "$dep_cc"

+    CCDEP=${_DEPCMD:-$DEPCMD}

+    CCDEP_FLAGS=${_DEPFLAGS:=$DEPFLAGS}

+    DEPCCFLAGS=$_flags

+fi

 add_cflags $extra_cflags

 add_cxxflags $extra_cxxflags

@@ -3140,6 +3189,7 @@ EOF

         check_yasm "pextrd [eax], xmm0, 1" && enable yasm ||

             die "yasm not found, use --disable-yasm for a crippled build"

         check_yasm "vextractf128 xmm0, ymm0, 0" || disable avx

+        check_yasm "vfmaddps ymm0, ymm1, ymm2, ymm3" || disable fma4

     fi

     case "$cpu" in

@@ -3673,6 +3723,7 @@ if enabled x86; then

     echo "SSE enabled               ${sse-no}"

     echo "SSSE3 enabled             ${ssse3-no}"

     echo "AVX enabled               ${avx-no}"

+    echo "FMA4 enabled              ${fma4-no}"

     echo "CMOV enabled              ${cmov-no}"

     echo "CMOV is fast              ${fast_cmov-no}"

     echo "EBX available             ${ebx_available-no}"

@@ -3814,6 +3865,9 @@ CXX=$cxx

 AS=$as

 LD=$ld

 DEPCC=$dep_cc

+DEPCCFLAGS=$DEPCCFLAGS \$(CPPFLAGS)

+DEPAS=$as

+DEPASFLAGS=$DEPASFLAGS \$(CPPFLAGS)

 YASM=$yasmexe

 YASMDEP=$yasmexe

 AR=$ar

@@ -3825,9 +3879,10 @@ CPPFLAGS=$CPPFLAGS

 CFLAGS=$CFLAGS

 CXXFLAGS=$CXXFLAGS

 ASFLAGS=$ASFLAGS

-AS_O=$CC_O

+AS_O=$AS_O

 CC_O=$CC_O

 CXX_O=$CXX_O

+LD_O=$LD_O

 LDFLAGS=$LDFLAGS

 FFSERVERLDFLAGS=$FFSERVERLDFLAGS

 SHFLAGS=$SHFLAGS

@@ -3842,10 +3897,11 @@ SLIBPREF=$SLIBPREF

 SLIBSUF=$SLIBSUF

 EXESUF=$EXESUF

 EXTRA_VERSION=$extra_version

-DEPFLAGS=$DEPFLAGS

 CCDEP=$CCDEP

 CXXDEP=$CXXDEP

+CCDEP_FLAGS=$CCDEP_FLAGS

 ASDEP=$ASDEP

+ASDEP_FLAGS=$ASDEP_FLAGS

 CC_DEPFLAGS=$CC_DEPFLAGS

 AS_DEPFLAGS=$AS_DEPFLAGS

 HOSTCC=$host_cc

@@ -3853,6 +3909,12 @@ HOSTCFLAGS=$host_cflags

 HOSTEXESUF=$HOSTEXESUF

 HOSTLDFLAGS=$host_ldflags

 HOSTLIBS=$host_libs

+DEPHOSTCC=$host_cc

+DEPHOSTCCFLAGS=$DEPHOSTCCFLAGS \$(HOSTCCFLAGS)

+HOSTCCDEP=$HOSTCCDEP

+HOSTCCDEP_FLAGS=$HOSTCCDEP_FLAGS

+HOSTCC_DEPFLAGS=$HOSTCC_DEPFLAGS

+HOSTCC_O=$HOSTCC_O

 TARGET_EXEC=$target_exec

 TARGET_PATH=$target_path

 SDL_LIBS=$sdl_libs

@@ -28,8 +28,6 @@ doc/%.txt: doc/%.texi

 	$(Q)$(TEXIDEP)

 	$(M)makeinfo --force --no-headers -o $@ $< 2>/dev/null

-doc/print_options.o: libavformat/options_table.h libavcodec/options_table.h

-

 GENTEXI  = format codec

 GENTEXI := $(GENTEXI:%=doc/avoptions_%.texi)

@@ -32,7 +32,7 @@ OBJS = allcodecs.o                                                      \

        utils.o                                                          \

 # parts needed for many different codecs

-OBJS-$(CONFIG_AANDCT)                  += aandcttab.o

+OBJS-$(CONFIG_AANDCTTABLES)            += aandcttab.o

 OBJS-$(CONFIG_AC3DSP)                  += ac3dsp.o

 OBJS-$(CONFIG_CRYSTALHD)               += crystalhd.o

 OBJS-$(CONFIG_ENCODERS)                += faandct.o jfdctfst.o jfdctint.o

@@ -200,6 +200,7 @@ static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,

                            int lpc_order, int lpc_quant)

 {

     int i;

+    int32_t *pred = buffer_out;

     /* first sample always copies */

     *buffer_out = *error_buffer;

@@ -223,37 +224,35 @@ static void lpc_prediction(int32_t *error_buffer, int32_t *buffer_out,

     }

     /* read warm-up samples */

-    for (i = 0; i < lpc_order; i++) {

-        buffer_out[i + 1] = sign_extend(buffer_out[i] + error_buffer[i + 1],

-                                        bps);

-    }

+    for (i = 1; i <= lpc_order; i++)

+        buffer_out[i] = sign_extend(buffer_out[i - 1] + error_buffer[i], bps);

     /* NOTE: 4 and 8 are very common cases that could be optimized. */

-    for (i = lpc_order; i < nb_samples - 1; i++) {

+    for (; i < nb_samples; i++) {

         int j;

         int val = 0;

-        int error_val = error_buffer[i + 1];

+        int error_val = error_buffer[i];

         int error_sign;

-        int d = buffer_out[i - lpc_order];

+        int d = *pred++;

         /* LPC prediction */

         for (j = 0; j < lpc_order; j++)

-            val += (buffer_out[i - j] - d) * lpc_coefs[j];

+            val += (pred[j] - d) * lpc_coefs[j];

         val = (val + (1 << (lpc_quant - 1))) >> lpc_quant;

         val += d + error_val;

-        buffer_out[i + 1] = sign_extend(val, bps);

+        buffer_out[i] = sign_extend(val, bps);

         /* adapt LPC coefficients */

         error_sign = sign_only(error_val);

         if (error_sign) {

-            for (j = lpc_order - 1; j >= 0 && error_val * error_sign > 0; j--) {

+            for (j = 0; j < lpc_order && error_val * error_sign > 0; j++) {

                 int sign;

-                val  = d - buffer_out[i - j];

+                val  = d - pred[j];

                 sign = sign_only(val) * error_sign;

                 lpc_coefs[j] -= sign;

                 val *= sign;

-                error_val -= (val >> lpc_quant) * (lpc_order - j);

+                error_val -= (val >> lpc_quant) * (j + 1);

             }

         }

     }

@@ -356,7 +355,7 @@ static int decode_element(AVCodecContext *avctx, void *data, int ch_index,

             lpc_order[ch]         = get_bits(&alac->gb, 5);

             /* read the predictor table */

-            for (i = 0; i < lpc_order[ch]; i++)

+            for (i = lpc_order[ch] - 1; i >= 0; i--)

                 lpc_coefs[ch][i] = get_sbits(&alac->gb, 16);

         }

@@ -477,16 +476,19 @@ static int alac_decode_frame(AVCodecContext *avctx, void *data,

     ALACContext *alac = avctx->priv_data;

     enum RawDataBlockType element;

     int channels;

-    int ch, ret;

+    int ch, ret, got_end;

     init_get_bits(&alac->gb, avpkt->data, avpkt->size * 8);

+    got_end = 0;

     alac->nb_samples = 0;

     ch = 0;

-    while (get_bits_left(&alac->gb)) {

+    while (get_bits_left(&alac->gb) >= 3) {

         element = get_bits(&alac->gb, 3);

-        if (element == TYPE_END)

+        if (element == TYPE_END) {

+            got_end = 1;

             break;

+        }

         if (element > TYPE_CPE && element != TYPE_LFE) {

             av_log(avctx, AV_LOG_ERROR, "syntax element unsupported: %d", element);

             return AVERROR_PATCHWELCOME;

@@ -501,11 +503,15 @@ static int alac_decode_frame(AVCodecContext *avctx, void *data,

         ret = decode_element(avctx, data,

                              alac_channel_layout_offsets[alac->channels - 1][ch],

                              channels);

-        if (ret < 0)

+        if (ret < 0 && get_bits_left(&alac->gb))

             return ret;

         ch += channels;

     }

+    if (!got_end) {

+        av_log(avctx, AV_LOG_ERROR, "no end tag found. incomplete packet.\n");

+        return AVERROR_INVALIDDATA;

+    }

     if (avpkt->size * 8 - get_bits_count(&alac->gb) > 8) {

         av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n",

@@ -298,8 +298,8 @@ static int tscc2_decode_frame(AVCodecContext *avctx, void *data,

         if (!size) {

             int skip_row = 1, j, off = i * c->mb_width;

             for (j = 0; j < c->mb_width; j++) {

-                if (c->slice_quants[off + i] == 1 ||

-                    c->slice_quants[off + i] == 2) {

+                if (c->slice_quants[off + j] == 1 ||

+                    c->slice_quants[off + j] == 2) {

                     skip_row = 0;

                     break;

                 }

@@ -1158,12 +1158,7 @@ ALIGN 16

     add     src1q, 2*mmsize

     sub     lenq,  2*mmsize

     jge     .loop

-%if mmsize == 32

-    vzeroupper

-    RET

-%else

     REP_RET

-%endif

 %endmacro

 INIT_XMM sse

@@ -1193,12 +1188,7 @@ ALIGN 16

     sub     lenq,   2*mmsize

     jge     .loop

-%if mmsize == 32

-    vzeroupper

-    RET

-%else

     REP_RET

-%endif

 %endmacro

 INIT_XMM sse

@@ -1243,10 +1233,6 @@ cglobal butterflies_float_interleave, 4,4,3, dst, src0, src1, len

 %endif

     add       lenq, mmsize

     jl .loop

-%if mmsize == 32

-    vzeroupper

-    RET

-%endif

 .end:

     REP_RET

 %endmacro

@@ -750,9 +750,6 @@ section .text

 ; The others pass args in registers and don't spill anything.

 cglobal fft_dispatch%2, 2,5,8, z, nbits

     FFT_DISPATCH fullsuffix, nbits

-%if mmsize == 32

-    vzeroupper

-%endif

     RET

 %endmacro ; DECL_FFT

@@ -958,9 +955,6 @@ cglobal imdct_half, 3,12,8; FFTContext *s, FFTSample *output, const FFTSample *i

 %if ARCH_X86_64 == 0

     add esp, 12

 %endif

-%if mmsize == 32

-    vzeroupper

-%endif

     RET

 %endmacro

@@ -36,6 +36,8 @@ void ff_four_imdct36_float_avx(float *out, float *buf, float *in, float *win,

 DECLARE_ALIGNED(16, static float, mdct_win_sse)[2][4][4*40];

+#if HAVE_INLINE_ASM

+

 #define MACS(rt, ra, rb) rt+=(ra)*(rb)

 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)

@@ -178,6 +180,7 @@ static void apply_window_mp3(float *in, float *win, int *unused, float *out,

     *out = sum;

 }

+#endif /* HAVE_INLINE_ASM */

 #define DECL_IMDCT_BLOCKS(CPU1, CPU2)                                       \

 static void imdct36_blocks_ ## CPU1(float *out, float *buf, float *in,      \

@@ -241,9 +244,11 @@ void ff_mpadsp_init_mmx(MPADSPContext *s)

         }

     }

+#if HAVE_INLINE_ASM

     if (mm_flags & AV_CPU_FLAG_SSE2) {

         s->apply_window_float = apply_window_mp3;

     }

+#endif /* HAVE_INLINE_ASM */

 #if HAVE_YASM

     if (0) {

 #if HAVE_AVX

@@ -83,8 +83,7 @@ section .text align=16

 ; %1 = row or col (for rounding variable)

 ; %2 = number of bits to shift at the end

-; %3 = optimization

-%macro IDCT_1D 3

+%macro IDCT_1D 2

     ; a0 = (W4 * row[0]) + (1 << (15 - 1));

     ; a1 = a0;

     ; a2 = a0;

@@ -235,8 +234,8 @@ section .text align=16

 ; void prores_idct_put_10_<opt>(uint8_t *pixels, int stride,

 ;                               DCTELEM *block, const int16_t *qmat);

-%macro idct_put_fn 2

-cglobal prores_idct_put_10_%1, 4, 4, %2

+%macro idct_put_fn 1

+cglobal prores_idct_put_10, 4, 4, %1

     movsxd      r1,  r1d

     pxor        m15, m15           ; zero

@@ -252,7 +251,7 @@ cglobal prores_idct_put_10_%1, 4, 4, %2

     pmullw      m13,[r3+64]

     pmullw      m12,[r3+96]

-    IDCT_1D     row, 15,  %1

+    IDCT_1D     row, 15

     ; transpose for second part of IDCT

     TRANSPOSE8x8W 8, 0, 1, 2, 4, 11, 9, 10, 3

@@ -267,7 +266,7 @@ cglobal prores_idct_put_10_%1, 4, 4, %2

     ; for (i = 0; i < 8; i++)

     ;     idctSparseColAdd(dest + i, line_size, block + i);

-    IDCT_1D     col, 18,  %1

+    IDCT_1D     col, 18

     ; clip/store

     mova        m3, [pw_4]

@@ -302,13 +301,27 @@ cglobal prores_idct_put_10_%1, 4, 4, %2

     RET

 %endmacro

-INIT_XMM

-idct_put_fn sse2, 16

-INIT_XMM

-idct_put_fn sse4, 16

+%macro SIGNEXTEND 2-3 ; dstlow, dsthigh, tmp

+%if cpuflag(sse4)

+    movhlps     %2,  %1

+    pmovsxwd    %1,  %1

+    pmovsxwd    %2,  %2

+%else ; sse2

+    pxor        %3,  %3

+    pcmpgtw     %3,  %1

+    mova        %2,  %1

+    punpcklwd   %1,  %3

+    punpckhwd   %2,  %3

+%endif

+%endmacro

+

+INIT_XMM sse2

+idct_put_fn 16

+INIT_XMM sse4

+idct_put_fn 16

 %if HAVE_AVX

-INIT_AVX

-idct_put_fn avx,  16

+INIT_XMM avx

+idct_put_fn 16

 %endif

 %endif

@@ -578,11 +578,44 @@ static void swap_samplerates(AVFilterGraph *graph)

         swap_samplerates_on_filter(graph->filters[i]);

 }

+#define CH_CENTER_PAIR (AV_CH_FRONT_LEFT_OF_CENTER | AV_CH_FRONT_RIGHT_OF_CENTER)

+#define CH_FRONT_PAIR  (AV_CH_FRONT_LEFT           | AV_CH_FRONT_RIGHT)

+#define CH_STEREO_PAIR (AV_CH_STEREO_LEFT          | AV_CH_STEREO_RIGHT)

+#define CH_WIDE_PAIR   (AV_CH_WIDE_LEFT            | AV_CH_WIDE_RIGHT)

+#define CH_SIDE_PAIR   (AV_CH_SIDE_LEFT            | AV_CH_SIDE_RIGHT)

+#define CH_DIRECT_PAIR (AV_CH_SURROUND_DIRECT_LEFT | AV_CH_SURROUND_DIRECT_RIGHT)

+#define CH_BACK_PAIR   (AV_CH_BACK_LEFT            | AV_CH_BACK_RIGHT)

+

+/* allowable substitutions for channel pairs when comparing layouts,

+ * ordered by priority for both values */

+static const uint64_t ch_subst[][2] = {

+    { CH_FRONT_PAIR,      CH_CENTER_PAIR     },

+    { CH_FRONT_PAIR,      CH_WIDE_PAIR       },

+    { CH_FRONT_PAIR,      AV_CH_FRONT_CENTER },

+    { CH_CENTER_PAIR,     CH_FRONT_PAIR      },

+    { CH_CENTER_PAIR,     CH_WIDE_PAIR       },

+    { CH_CENTER_PAIR,     AV_CH_FRONT_CENTER },

+    { CH_WIDE_PAIR,       CH_FRONT_PAIR      },

+    { CH_WIDE_PAIR,       CH_CENTER_PAIR     },

+    { CH_WIDE_PAIR,       AV_CH_FRONT_CENTER },

+    { AV_CH_FRONT_CENTER, CH_FRONT_PAIR      },

+    { AV_CH_FRONT_CENTER, CH_CENTER_PAIR     },

+    { AV_CH_FRONT_CENTER, CH_WIDE_PAIR       },

+    { CH_SIDE_PAIR,       CH_DIRECT_PAIR     },

+    { CH_SIDE_PAIR,       CH_BACK_PAIR       },

+    { CH_SIDE_PAIR,       AV_CH_BACK_CENTER  },

+    { CH_BACK_PAIR,       CH_DIRECT_PAIR     },

+    { CH_BACK_PAIR,       CH_SIDE_PAIR       },

+    { CH_BACK_PAIR,       AV_CH_BACK_CENTER  },

+    { AV_CH_BACK_CENTER,  CH_BACK_PAIR       },

+    { AV_CH_BACK_CENTER,  CH_DIRECT_PAIR     },

+    { AV_CH_BACK_CENTER,  CH_SIDE_PAIR       },

+};

+

 static void swap_channel_layouts_on_filter(AVFilterContext *filter)

 {

     AVFilterLink *link = NULL;

-    uint64_t chlayout;

-    int i, j;

+    int i, j, k;

     for (i = 0; i < filter->nb_inputs; i++) {

         link = filter->inputs[i];

@@ -594,27 +627,55 @@ static void swap_channel_layouts_on_filter(AVFilterContext *filter)

     if (i == filter->nb_inputs)

         return;