Merge remote-tracking branch 'qatar/master'

* qatar/master:
  proresdsp: port x86 assembly to cpuflags.
  lavr: x86: improve non-SSE4 version of S16_TO_S32_SX macro
  lavfi: better channel layout negotiation
  alac: check for truncated packets
  alac: reverse lpc coeff order, simplify filter
  lavr: add x86-optimized mixing functions
  x86: add support for fmaddps fma4 instruction with abstraction to avx/sse
  tscc2: fix typo in array index
  build: use COMPILE template for HOSTOBJS
  build: do full flag handling for all compiler-type tools
  eval: fix printing of NaN in eval fate test.
  build: Rename aandct component to more descriptive aandcttables
  mpegaudio: bury inline asm under HAVE_INLINE_ASM.
  x86inc: automatically insert vzeroupper for YMM functions.
  rtmp: Check the buffer length of ping packets
  rtmp: Allow having more unknown data at the end of a chunk size packet without failing
  rtmp: Prevent reading outside of an allocate buffer when receiving server bandwidth packets

Conflicts:
	Makefile
	configure
	libavcodec/x86/proresdsp.asm
	libavutil/eval.c
Merged-by: Michael Niedermayer <michaelni@gmx.at>

common.mak

@@ -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))

doc/Makefile

@@ -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)
 

libavcodec/Makefile

@@ -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

libavcodec/alac.c

@@ -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",

libavcodec/tscc2.c

@@ -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;
                 }

libavcodec/x86/dsputil_yasm.asm

@@ -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

libavcodec/x86/fft_mmx.asm

@@ -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
 
@@ -957,9 +954,6 @@ cglobal imdct_half, 3,12,8; FFTContext *s, FFTSample *output, const FFTSample *i
     %1 r0, r1, r6, rtcos, rtsin
 %if ARCH_X86_64 == 0
     add esp, 12
-%endif
-%if mmsize == 32
-    vzeroupper
 %endif
     RET
 %endmacro

libavcodec/x86/mpegaudiodec_mmx.c

@@ -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

libavcodec/x86/proresdsp.asm

@@ -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

libavfilter/avfiltergraph.c

@@ -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;
 
-    chlayout = link->out_channel_layouts->channel_layouts[0];
-
     for (i = 0; i < filter->nb_outputs; i++) {
         AVFilterLink *outlink = filter->outputs[i];
-        int best_idx, best_score = INT_MIN;
+        int best_idx, best_score = INT_MIN, best_count_diff = INT_MAX;
 
         if (outlink->type != AVMEDIA_TYPE_AUDIO ||
             outlink->in_channel_layouts->nb_channel_layouts < 2)
             continue;
 
         for (j = 0; j < outlink->in_channel_layouts->nb_channel_layouts; j++) {
+            uint64_t  in_chlayout = link->out_channel_layouts->channel_layouts[0];
             uint64_t out_chlayout = outlink->in_channel_layouts->channel_layouts[j];
-            int matched_channels  = av_get_channel_layout_nb_channels(chlayout &
-                                                                      out_chlayout);
-            int extra_channels     = av_get_channel_layout_nb_channels(out_chlayout &
-                                                                       (~chlayout));
-            int score = matched_channels - extra_channels;
+            int  in_channels      = av_get_channel_layout_nb_channels(in_chlayout);
+            int out_channels      = av_get_channel_layout_nb_channels(out_chlayout);
+            int count_diff        = out_channels - in_channels;
+            int matched_channels, extra_channels;
+            int score = 0;
+
+            /* channel substitution */
+            for (k = 0; k < FF_ARRAY_ELEMS(ch_subst); k++) {
+                uint64_t cmp0 = ch_subst[k][0];
+                uint64_t cmp1 = ch_subst[k][1];
+                if (( in_chlayout & cmp0) && (!(out_chlayout & cmp0)) &&
+                    (out_chlayout & cmp1) && (!( in_chlayout & cmp1))) {
+                    in_chlayout  &= ~cmp0;
+                    out_chlayout &= ~cmp1;
+                    /* add score for channel match, minus a deduction for
+                       having to do the substitution */
+                    score += 10 * av_get_channel_layout_nb_channels(cmp1) - 2;
+                }
+            }
 
-            if (score > best_score) {
+            /* no penalty for LFE channel mismatch */
+            if ( (in_chlayout & AV_CH_LOW_FREQUENCY) &&
+                (out_chlayout & AV_CH_LOW_FREQUENCY))
+                score += 10;
+            in_chlayout  &= ~AV_CH_LOW_FREQUENCY;
+            out_chlayout &= ~AV_CH_LOW_FREQUENCY;
+
+            matched_channels = av_get_channel_layout_nb_channels(in_chlayout &
+                                                                 out_chlayout);
+            extra_channels   = av_get_channel_layout_nb_channels(out_chlayout &
+                                                                 (~in_chlayout));
+            score += 10 * matched_channels - 5 * extra_channels;
+
+            if (score > best_score ||
+                (count_diff < best_count_diff && score == best_score)) {
                 best_score = score;
                 best_idx   = j;
+                best_count_diff = count_diff;
             }
         }
         FFSWAP(uint64_t, outlink->in_channel_layouts->channel_layouts[0],

libavformat/rtmpproto.c

@@ -515,6 +515,12 @@ static int gen_pong(URLContext *s, RTMPContext *rt, RTMPPacket *ppkt)
     uint8_t *p;
     int ret;
 
+    if (ppkt->data_size < 6) {
+        av_log(s, AV_LOG_ERROR, "Too short ping packet (%d)\n",
+               ppkt->data_size);
+        return AVERROR_INVALIDDATA;
+    }
+
     if ((ret = ff_rtmp_packet_create(&pkt, RTMP_NETWORK_CHANNEL, RTMP_PT_PING,
                                      ppkt->timestamp + 1, 6)) < 0)
         return ret;
@@ -885,9 +891,9 @@ static int handle_chunk_size(URLContext *s, RTMPPacket *pkt)
     RTMPContext *rt = s->priv_data;
     int ret;
 
-    if (pkt->data_size != 4) {
+    if (pkt->data_size < 4) {
         av_log(s, AV_LOG_ERROR,
-               "Chunk size change packet is not 4 bytes long (%d)\n",
+               "Too short chunk size change packet (%d)\n",
                pkt->data_size);
         return AVERROR_INVALIDDATA;
     }
@@ -913,6 +919,12 @@ static int handle_ping(URLContext *s, RTMPPacket *pkt)
     RTMPContext *rt = s->priv_data;
     int t, ret;
 
+    if (pkt->data_size < 2) {
+        av_log(s, AV_LOG_ERROR, "Too short ping packet (%d)\n",
+               pkt->data_size);
+        return AVERROR_INVALIDDATA;
+    }
+
     t = AV_RB16(pkt->data);
     if (t == 6) {
         if ((ret = gen_pong(s, rt, pkt)) < 0)
@@ -950,6 +962,13 @@ static int handle_server_bw(URLContext *s, RTMPPacket *pkt)
 {
     RTMPContext *rt = s->priv_data;
 
+    if (pkt->data_size < 4) {
+        av_log(s, AV_LOG_ERROR,
+               "Too short server bandwidth report packet (%d)\n",
+               pkt->data_size);
+        return AVERROR_INVALIDDATA;
+    }
+
     rt->server_bw = AV_RB32(pkt->data);
     if (rt->server_bw <= 0) {
         av_log(s, AV_LOG_ERROR, "Incorrect server bandwidth %d\n",

libavresample/utils.c

@@ -246,9 +246,10 @@ static int handle_buffered_output(AVAudioResampleContext *avr,
     return 0;
 }
 
-int avresample_convert(AVAudioResampleContext *avr, void **output,
-                       int out_plane_size, int out_samples, void **input,
-                       int in_plane_size, int in_samples)
+int attribute_align_arg avresample_convert(AVAudioResampleContext *avr,
+                                           void **output, int out_plane_size,
+                                           int out_samples, void **input,
+                                           int in_plane_size, int in_samples)
 {
     AudioData input_buffer;
     AudioData output_buffer;

libavresample/x86/audio_convert.asm

@@ -145,12 +145,7 @@ cglobal conv_s32_to_flt, 3,3,3, dst, src, len
     mova  [dstq+lenq+mmsize], m2
     add     lenq, mmsize*2
     jl .loop
-%if mmsize == 32
-    vzeroupper
-    RET
-%else
     REP_RET
-%endif
 %endmacro
 
 INIT_XMM sse2
@@ -218,12 +213,7 @@ cglobal conv_flt_to_s32, 3,3,5, dst, src, len
     mova  [dstq+lenq+3*mmsize], m3
     add     lenq, mmsize*4
     jl .loop
-%if mmsize == 32
-    vzeroupper
-    RET
-%else
     REP_RET
-%endif
 %endmacro
 
 INIT_XMM sse2

libavresample/x86/audio_mix.asm

@@ -51,12 +51,7 @@ cglobal mix_2_to_1_fltp_flt, 3,4,6, src, matrix, len, src1
     add        srcq, mmsize*2
     sub        lend, mmsize*2/4
     jg .loop
-%if mmsize == 32
-    vzeroupper
-    RET
-%else
     REP_RET
-%endif
 %endmacro
 
 INIT_XMM sse
@@ -175,12 +170,7 @@ cglobal mix_1_to_2_fltp_flt, 3,5,4, src0, matrix0, len, src1, matrix1
     add       src0q, mmsize
     sub        lend, mmsize/4
     jg .loop
-%if mmsize == 32
-    vzeroupper
-    RET
-%else
     REP_RET
-%endif
 %endmacro
 
 INIT_XMM sse
@@ -236,3 +226,296 @@ MIX_1_TO_2_S16P_FLT
 INIT_XMM avx
 MIX_1_TO_2_S16P_FLT
 %endif
+
+;-----------------------------------------------------------------------------
+; void ff_mix_3_8_to_1_2_fltp/s16p_flt(float/int16_t **src, float **matrix,
+;                                      int len, int out_ch, int in_ch);
+;-----------------------------------------------------------------------------
+
+%macro MIX_3_8_TO_1_2_FLT 3 ; %1 = in channels, %2 = out channels, %3 = s16p or fltp
+; define some names to make the code clearer
+%assign  in_channels %1
+%assign out_channels %2
+%assign stereo out_channels - 1
+%ifidn %3, s16p
+    %assign is_s16 1
+%else
+    %assign is_s16 0
+%endif
+
+; determine how many matrix elements must go on the stack vs. mmregs
+%assign matrix_elements in_channels * out_channels
+%if is_s16
+    %if stereo
+        %assign needed_mmregs 7
+    %else
+        %assign needed_mmregs 5
+    %endif
+%else
+    %if stereo
+        %assign needed_mmregs 4
+    %else
+        %assign needed_mmregs 3
+    %endif
+%endif
+%assign matrix_elements_mm num_mmregs - needed_mmregs
+%if matrix_elements < matrix_elements_mm
+    %assign matrix_elements_mm matrix_elements
+%endif
+%if matrix_elements_mm < matrix_elements
+    %assign matrix_elements_stack matrix_elements - matrix_elements_mm
+%else
+    %assign matrix_elements_stack 0
+%endif
+
+cglobal mix_%1_to_%2_%3_flt, 3,in_channels+2,needed_mmregs+matrix_elements_mm, src0, src1, len, src2, src3, src4, src5, src6, src7
+
+; get aligned stack space if needed
+%if matrix_elements_stack > 0
+    %if mmsize == 32
+    %assign bkpreg %1 + 1
+    %define bkpq r %+ bkpreg %+ q
+    mov           bkpq, rsp
+    and           rsp, ~(mmsize-1)
+    sub           rsp, matrix_elements_stack * mmsize
+    %else
+    %assign pad matrix_elements_stack * mmsize + (mmsize - gprsize) - (stack_offset & (mmsize - gprsize))
+    SUB           rsp, pad
+    %endif
+%endif
+
+; load matrix pointers
+%define matrix0q r1q
+%define matrix1q r3q
+%if stereo
+    mov      matrix1q, [matrix0q+gprsize]
+%endif
+    mov      matrix0q, [matrix0q]
+
+; define matrix coeff names
+%assign %%i 0
+%assign %%j needed_mmregs
+%rep in_channels
+    %if %%i >= matrix_elements_mm
+        CAT_XDEFINE mx_stack_0_, %%i, 1
+        CAT_XDEFINE mx_0_, %%i, [rsp+(%%i-matrix_elements_mm)*mmsize]
+    %else
+        CAT_XDEFINE mx_stack_0_, %%i, 0
+        CAT_XDEFINE mx_0_, %%i, m %+ %%j
+        %assign %%j %%j+1
+    %endif
+    %assign %%i %%i+1
+%endrep
+%if stereo
+%assign %%i 0
+%rep in_channels
+    %if in_channels + %%i >= matrix_elements_mm
+        CAT_XDEFINE mx_stack_1_, %%i, 1
+        CAT_XDEFINE mx_1_, %%i, [rsp+(in_channels+%%i-matrix_elements_mm)*mmsize]
+    %else
+        CAT_XDEFINE mx_stack_1_, %%i, 0
+        CAT_XDEFINE mx_1_, %%i, m %+ %%j
+        %assign %%j %%j+1
+    %endif
+    %assign %%i %%i+1
+%endrep
+%endif
+
+; load/splat matrix coeffs
+%assign %%i 0
+%rep in_channels
+    %if mx_stack_0_ %+ %%i
+        VBROADCASTSS m0, [matrix0q+4*%%i]
+        mova  mx_0_ %+ %%i, m0
+    %else
+        VBROADCASTSS mx_0_ %+ %%i, [matrix0q+4*%%i]
+    %endif
+    %if stereo
+    %if mx_stack_1_ %+ %%i
+        VBROADCASTSS m0, [matrix1q+4*%%i]
+        mova  mx_1_ %+ %%i, m0
+    %else
+        VBROADCASTSS mx_1_ %+ %%i, [matrix1q+4*%%i]
+    %endif
+    %endif
+    %assign %%i %%i+1
+%endrep
+
+; load channel pointers to registers as offsets from the first channel pointer
+%if ARCH_X86_64
+    movsxd       lenq, r2d
+%endif
+    shl          lenq, 2-is_s16
+%assign %%i 1
+%rep (in_channels - 1)
+    %if ARCH_X86_32 && in_channels >= 7 && %%i >= 5
+    mov         src5q, [src0q+%%i*gprsize]
+    add         src5q, lenq
+    mov         src %+ %%i %+ m, src5q
+    %else
+    mov         src %+ %%i %+ q, [src0q+%%i*gprsize]
+    add         src %+ %%i %+ q, lenq
+    %endif
+    %assign %%i %%i+1
+%endrep
+    mov         src0q, [src0q]
+    add         src0q, lenq
+    neg          lenq
+.loop
+; for x86-32 with 7-8 channels we do not have enough gp registers for all src
+; pointers, so we have to load some of them from the stack each time
+%define copy_src_from_stack ARCH_X86_32 && in_channels >= 7 && %%i >= 5
+%if is_s16
+    ; mix with s16p input
+    mova           m0, [src0q+lenq]
+    S16_TO_S32_SX   0, 1
+    cvtdq2ps       m0, m0
+    cvtdq2ps       m1, m1
+    %if stereo
+    mulps          m2, m0, mx_1_0
+    mulps          m3, m1, mx_1_0
+    %endif
+    mulps          m0, m0, mx_0_0
+    mulps          m1, m1, mx_0_0
+%assign %%i 1
+%rep (in_channels - 1)
+    %if copy_src_from_stack
+        %define src_ptr src5q
+    %else
+        %define src_ptr src %+ %%i %+ q
+    %endif
+    %if stereo
+    %if copy_src_from_stack
+    mov       src_ptr, src %+ %%i %+ m
+    %endif
+    mova           m4, [src_ptr+lenq]
+    S16_TO_S32_SX   4, 5
+    cvtdq2ps       m4, m4
+    cvtdq2ps       m5, m5
+    fmaddps        m2, m4, mx_1_ %+ %%i, m2, m6
+    fmaddps        m3, m5, mx_1_ %+ %%i, m3, m6
+    fmaddps        m0, m4, mx_0_ %+ %%i, m0, m4
+    fmaddps        m1, m5, mx_0_ %+ %%i, m1, m5
+    %else
+    %if copy_src_from_stack
+    mov       src_ptr, src %+ %%i %+ m
+    %endif
+    mova           m2, [src_ptr+lenq]
+    S16_TO_S32_SX   2, 3
+    cvtdq2ps       m2, m2
+    cvtdq2ps       m3, m3
+    fmaddps        m0, m2, mx_0_ %+ %%i, m0, m4
+    fmaddps        m1, m3, mx_0_ %+ %%i, m1, m4
+    %endif
+    %assign %%i %%i+1
+%endrep
+    %if stereo
+    cvtps2dq       m2, m2
+    cvtps2dq       m3, m3
+    packssdw       m2, m3
+    mova [src1q+lenq], m2
+    %endif
+    cvtps2dq       m0, m0
+    cvtps2dq       m1, m1
+    packssdw       m0, m1
+    mova [src0q+lenq], m0
+%else
+    ; mix with fltp input
+    %if stereo || mx_stack_0_0
+    mova           m0, [src0q+lenq]
+    %endif
+    %if stereo
+    mulps          m1, m0, mx_1_0
+    %endif
+    %if stereo || mx_stack_0_0
+    mulps          m0, m0, mx_0_0
+    %else
+    mulps          m0, [src0q+lenq], mx_0_0
+    %endif
+%assign %%i 1
+%rep (in_channels - 1)
+    %if copy_src_from_stack
+        %define src_ptr src5q
+        mov   src_ptr, src %+ %%i %+ m
+    %else
+        %define src_ptr src %+ %%i %+ q
+    %endif
+    ; avoid extra load for mono if matrix is in a mm register
+    %if stereo || mx_stack_0_ %+ %%i
+    mova           m2, [src_ptr+lenq]
+    %endif
+    %if stereo
+    fmaddps        m1, m2, mx_1_ %+ %%i, m1, m3
+    %endif
+    %if stereo || mx_stack_0_ %+ %%i
+    fmaddps        m0, m2, mx_0_ %+ %%i, m0, m2
+    %else
+    fmaddps        m0, mx_0_ %+ %%i, [src_ptr+lenq], m0, m1
+    %endif
+    %assign %%i %%i+1
+%endrep
+    mova [src0q+lenq], m0
+    %if stereo
+    mova [src1q+lenq], m1
+    %endif
+%endif
+
+    add          lenq, mmsize
+    jl .loop
+; restore stack pointer
+%if matrix_elements_stack > 0
+    %if mmsize == 32
+    mov           rsp, bkpq
+    %else
+    ADD           rsp, pad
+    %endif
+%endif
+; zero ymm high halves
+%if mmsize == 32
+    vzeroupper
+%endif
+    RET
+%endmacro
+
+%macro MIX_3_8_TO_1_2_FLT_FUNCS 0
+%assign %%i 3
+%rep 6
+    INIT_XMM sse
+    MIX_3_8_TO_1_2_FLT %%i, 1, fltp
+    MIX_3_8_TO_1_2_FLT %%i, 2, fltp
+    INIT_XMM sse2
+    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
+    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
+    INIT_XMM sse4
+    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
+    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
+    ; do not use ymm AVX or FMA4 in x86-32 for 6 or more channels due to stack alignment issues
+    %if HAVE_AVX
+    %if ARCH_X86_64 || %%i < 6
+    INIT_YMM avx
+    %else
+    INIT_XMM avx
+    %endif
+    MIX_3_8_TO_1_2_FLT %%i, 1, fltp
+    MIX_3_8_TO_1_2_FLT %%i, 2, fltp
+    INIT_XMM avx
+    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
+    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
+    %endif
+    %if HAVE_FMA4
+    %if ARCH_X86_64 || %%i < 6
+    INIT_YMM fma4
+    %else
+    INIT_XMM fma4
+    %endif
+    MIX_3_8_TO_1_2_FLT %%i, 1, fltp
+    MIX_3_8_TO_1_2_FLT %%i, 2, fltp
+    INIT_XMM fma4
+    MIX_3_8_TO_1_2_FLT %%i, 1, s16p
+    MIX_3_8_TO_1_2_FLT %%i, 2, s16p
+    %endif
+    %assign %%i %%i+1
+%endrep
+%endmacro
+
+MIX_3_8_TO_1_2_FLT_FUNCS

libavresample/x86/audio_mix_init.c

@@ -47,6 +47,129 @@ extern void ff_mix_1_to_2_s16p_flt_sse4(int16_t **src, float **matrix, int len,
 extern void ff_mix_1_to_2_s16p_flt_avx (int16_t **src, float **matrix, int len,
                                         int out_ch, int in_ch);
 
+#define DEFINE_MIX_3_8_TO_1_2(chan)                                         \
+extern void ff_mix_ ## chan ## _to_1_fltp_flt_sse(float **src,              \
+                                                  float **matrix, int len,  \
+                                                  int out_ch, int in_ch);   \
+extern void ff_mix_ ## chan ## _to_2_fltp_flt_sse(float **src,              \
+                                                  float **matrix, int len,  \
+                                                  int out_ch, int in_ch);   \
+                                                                            \
+extern void ff_mix_ ## chan ## _to_1_s16p_flt_sse2(int16_t **src,           \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);  \
+extern void ff_mix_ ## chan ## _to_2_s16p_flt_sse2(int16_t **src,           \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);  \
+                                                                            \
+extern void ff_mix_ ## chan ## _to_1_s16p_flt_sse4(int16_t **src,           \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);  \
+extern void ff_mix_ ## chan ## _to_2_s16p_flt_sse4(int16_t **src,           \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);  \
+                                                                            \
+extern void ff_mix_ ## chan ## _to_1_fltp_flt_avx(float **src,              \
+                                                  float **matrix, int len,  \
+                                                  int out_ch, int in_ch);   \
+extern void ff_mix_ ## chan ## _to_2_fltp_flt_avx(float **src,              \
+                                                  float **matrix, int len,  \
+                                                  int out_ch, int in_ch);   \
+                                                                            \
+extern void ff_mix_ ## chan ## _to_1_s16p_flt_avx(int16_t **src,            \
+                                                  float **matrix, int len,  \
+                                                  int out_ch, int in_ch);   \
+extern void ff_mix_ ## chan ## _to_2_s16p_flt_avx(int16_t **src,            \
+                                                  float **matrix, int len,  \
+                                                  int out_ch, int in_ch);   \
+                                                                            \
+extern void ff_mix_ ## chan ## _to_1_fltp_flt_fma4(float **src,             \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);  \
+extern void ff_mix_ ## chan ## _to_2_fltp_flt_fma4(float **src,             \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);  \
+                                                                            \
+extern void ff_mix_ ## chan ## _to_1_s16p_flt_fma4(int16_t **src,           \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);  \
+extern void ff_mix_ ## chan ## _to_2_s16p_flt_fma4(int16_t **src,           \
+                                                   float **matrix, int len, \
+                                                   int out_ch, int in_ch);
+
+DEFINE_MIX_3_8_TO_1_2(3)
+DEFINE_MIX_3_8_TO_1_2(4)
+DEFINE_MIX_3_8_TO_1_2(5)
+DEFINE_MIX_3_8_TO_1_2(6)
+DEFINE_MIX_3_8_TO_1_2(7)
+DEFINE_MIX_3_8_TO_1_2(8)
+
+#define SET_MIX_3_8_TO_1_2(chan)                                            \
+    if (mm_flags & AV_CPU_FLAG_SSE && HAVE_SSE) {                           \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 1, 16, 4, "SSE",                        \
+                              ff_mix_ ## chan ## _to_1_fltp_flt_sse);       \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 2, 16, 4, "SSE",                        \
+                              ff_mix_## chan ##_to_2_fltp_flt_sse);         \
+    }                                                                       \
+    if (mm_flags & AV_CPU_FLAG_SSE2 && HAVE_SSE) {                          \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 1, 16, 8, "SSE2",                       \
+                              ff_mix_ ## chan ## _to_1_s16p_flt_sse2);      \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 2, 16, 8, "SSE2",                       \
+                              ff_mix_ ## chan ## _to_2_s16p_flt_sse2);      \
+    }                                                                       \
+    if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {                          \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 1, 16, 8, "SSE4",                       \
+                              ff_mix_ ## chan ## _to_1_s16p_flt_sse4);      \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 2, 16, 8, "SSE4",                       \
+                              ff_mix_ ## chan ## _to_2_s16p_flt_sse4);      \
+    }                                                                       \
+    if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {                           \
+        int ptr_align = 32;                                                 \
+        int smp_align = 8;                                                  \
+        if (ARCH_X86_32 || chan >= 6) {                                     \
+            ptr_align = 16;                                                 \
+            smp_align = 4;                                                  \
+        }                                                                   \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 1, ptr_align, smp_align, "AVX",         \
+                              ff_mix_ ## chan ## _to_1_fltp_flt_avx);       \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 2, ptr_align, smp_align, "AVX",         \
+                              ff_mix_ ## chan ## _to_2_fltp_flt_avx);       \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 1, 16, 8, "AVX",                        \
+                              ff_mix_ ## chan ## _to_1_s16p_flt_avx);       \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 2, 16, 8, "AVX",                        \
+                              ff_mix_ ## chan ## _to_2_s16p_flt_avx);       \
+    }                                                                       \
+    if (mm_flags & AV_CPU_FLAG_FMA4 && HAVE_FMA4) {                         \
+        int ptr_align = 32;                                                 \
+        int smp_align = 8;                                                  \
+        if (ARCH_X86_32 || chan >= 6) {                                     \
+            ptr_align = 16;                                                 \
+            smp_align = 4;                                                  \
+        }                                                                   \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 1, ptr_align, smp_align, "FMA4",        \
+                              ff_mix_ ## chan ## _to_1_fltp_flt_fma4);      \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_FLTP, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 2, ptr_align, smp_align, "FMA4",        \
+                              ff_mix_ ## chan ## _to_2_fltp_flt_fma4);      \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 1, 16, 8, "FMA4",                       \
+                              ff_mix_ ## chan ## _to_1_s16p_flt_fma4);      \
+        ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,\
+                              chan, 2, 16, 8, "FMA4",                       \
+                              ff_mix_ ## chan ## _to_2_s16p_flt_fma4);      \
+    }
+
 av_cold void ff_audio_mix_init_x86(AudioMix *am)
 {
 #if HAVE_YASM
@@ -80,5 +203,12 @@ av_cold void ff_audio_mix_init_x86(AudioMix *am)
         ff_audio_mix_set_func(am, AV_SAMPLE_FMT_S16P, AV_MIX_COEFF_TYPE_FLT,
                               1, 2, 16, 8, "AVX", ff_mix_1_to_2_s16p_flt_avx);
     }
+
+    SET_MIX_3_8_TO_1_2(3)
+    SET_MIX_3_8_TO_1_2(4)
+    SET_MIX_3_8_TO_1_2(5)
+    SET_MIX_3_8_TO_1_2(6)
+    SET_MIX_3_8_TO_1_2(7)
+    SET_MIX_3_8_TO_1_2(8)
 #endif
 }

libavresample/x86/util.asm

@@ -26,7 +26,8 @@
     pmovsxwd     m%1, m%1
     SWAP %1, %2
 %else
-    punpckhwd    m%2, m%1
+    mova         m%2, m%1
+    punpckhwd    m%2, m%2
     punpcklwd    m%1, m%1
     psrad        m%2, 16
     psrad        m%1, 16

libavutil/eval.c

@@ -797,11 +797,10 @@ int main(int argc, char **argv)
         av_expr_parse_and_eval(&d, *expr,
                                const_names, const_values,
                                NULL, NULL, NULL, NULL, NULL, 0, NULL);
-        if(isnan(d)){
+        if (isnan(d))
             printf("'%s' -> nan\n\n", *expr);
-        }else{
+        else
             printf("'%s' -> %f\n\n", *expr, d);
-        }
     }
 
     av_expr_parse_and_eval(&d, "1+(5-2)^(3-1)+1/2+sin(PI)-max(-2.2,-3.1)",

libavutil/x86/float_dsp.asm

@@ -42,12 +42,7 @@ ALIGN 16
 
     sub       lenq, 2*mmsize
     jge       .loop
-%if mmsize == 32
-    vzeroupper
-    RET
-%else
     REP_RET
-%endif
 %endmacro
 
 INIT_XMM sse
@@ -88,12 +83,7 @@ cglobal vector_fmac_scalar, 4,4,3, dst, src, mul, len
     mova  [dstq+lenq+mmsize], m2
     sub    lenq, 2*mmsize
     jge .loop
-%if mmsize == 32
-    vzeroupper
-    RET
-%else
     REP_RET
-%endif
 %endmacro
 
 INIT_XMM sse

libavutil/x86/x86inc.asm

@@ -392,11 +392,14 @@ DECLARE_REG 14, R15, R15D, R15W, R15B, 120
 %macro RET 0
     WIN64_RESTORE_XMM_INTERNAL rsp
     POP_IF_USED 14, 13, 12, 11, 10, 9, 8, 7
+%if mmsize == 32
+    vzeroupper
+%endif
     ret
 %endmacro
 
 %macro REP_RET 0
-    %if regs_used > 7 || xmm_regs_used > 6
+    %if regs_used > 7 || xmm_regs_used > 6 || mmsize == 32
         RET
     %else
         rep ret
@@ -433,11 +436,14 @@ DECLARE_REG 14, R15, R15D, R15W, R15B, 72
 
 %macro RET 0
     POP_IF_USED 14, 13, 12, 11, 10, 9
+%if mmsize == 32
+    vzeroupper
+%endif
     ret
 %endmacro
 
 %macro REP_RET 0
-    %if regs_used > 9
+    %if regs_used > 9 || mmsize == 32
         RET
     %else
         rep ret
@@ -479,11 +485,14 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
 
 %macro RET 0
     POP_IF_USED 6, 5, 4, 3
+%if mmsize == 32
+    vzeroupper
+%endif
     ret
 %endmacro
 
 %macro REP_RET 0
-    %if regs_used > 3
+    %if regs_used > 3 || mmsize == 32
         RET
     %else
         rep ret
@@ -1126,16 +1135,22 @@ AVX_INSTR pfmul, 1, 0, 1
 %undef j
 
 %macro FMA_INSTR 3
-    %macro %1 4-7 %1, %2, %3
-        %if cpuflag(xop)
-            v%5 %1, %2, %3, %4
+    %macro %1 5-8 %1, %2, %3
+        %if cpuflag(xop) || cpuflag(fma4)
+            v%6 %1, %2, %3, %4
         %else
-            %6 %1, %2, %3
-            %7 %1, %4
+            %ifidn %1, %4
+                %7 %5, %2, %3
+                %8 %1, %4, %5
+            %else
+                %7 %1, %2, %3
+                %8 %1, %4
+            %endif
         %endif
     %endmacro
 %endmacro
 
+FMA_INSTR  fmaddps,   mulps, addps
 FMA_INSTR  pmacsdd,  pmulld, paddd
 FMA_INSTR  pmacsww,  pmullw, paddw
 FMA_INSTR pmadcswd, pmaddwd, paddd

tests/Makefile

@@ -15,9 +15,6 @@ ffservertest: ffserver$(EXESUF) tests/vsynth1/00.pgm tests/data/asynth1.sw
 
 OBJDIRS += tests/data tests/vsynth1
 
-# Required due to missing automatic dependency tracking for HOSTOBJS.
-tests/rotozoom.o tests/videogen.o: tests/utils.c
-
 tests/vsynth1/00.pgm: tests/videogen$(HOSTEXESUF) | tests/vsynth1
 	$(M)./$< 'tests/vsynth1/'