Merge remote-tracking branch 'qatar/master'

* qatar/master:
  get_bits: remove x86 inline asm in A32 bitstream reader
  doc: Remove outdated information about our issue tracker
  avidec: Factor out the sync fucntionality.
  fate-aac: Expand coverage.
  ac3dsp: add x86-optimized versions of ac3dsp.extract_exponents().
  ac3dsp: simplify extract_exponents() now that it does not need to do clipping.
  ac3enc: clip coefficients after MDCT.
  ac3enc: add int32_t array clipping function to DSPUtil, including x86 versions.
  swscale: for >8bit scaling, read in native bit-depth.
  matroskadec: matroska_read_seek after after EBML_STOP leads to failure.
  doxygen: fix usage of @file directive in libavutil/{dict,file}.h
  doxygen: Help doxygen parser to understand the DECLARE_ALIGNED and offsetof macros

Conflicts:
	doc/issue_tracker.txt
	libavformat/avidec.c
	libavutil/dict.h
	libswscale/swscale.c
	libswscale/utils.c
	tests/ref/lavfi/pixfmts_scale

Merged-by: Michael Niedermayer <michaelni@gmx.at>

12 files changed, 318 insertions, 25 deletions

diff --git a/libavcodec/ac3dsp.c b/libavcodec/ac3dsp.c
index 619addc..96bd123 100644
--- a/libavcodec/ac3dsp.c
+++ b/libavcodec/ac3dsp.c
@@ -164,21 +164,8 @@ static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs)
     int i;
 
     for (i = 0; i < nb_coefs; i++) {
-        int e;
         int v = abs(coef[i]);
-        if (v == 0)
-            e = 24;
-        else {
-            e = 23 - av_log2(v);
-            if (e >= 24) {
-                e = 24;
-                coef[i] = 0;
-            } else if (e < 0) {
-                e = 0;
-                coef[i] = av_clip(coef[i], -16777215, 16777215);
-            }
-        }
-        exp[i] = e;
+        exp[i] = v ? 23 - av_log2(v) : 24;
     }
 }
 
diff --git a/libavcodec/ac3enc.h b/libavcodec/ac3enc.h
index 5f5d2c2..be2767a 100644
--- a/libavcodec/ac3enc.h
+++ b/libavcodec/ac3enc.h
@@ -50,12 +50,16 @@
 #if CONFIG_AC3ENC_FLOAT
 #define AC3_NAME(x) ff_ac3_float_ ## x
 #define MAC_COEF(d,a,b) ((d)+=(a)*(b))
+#define COEF_MIN (-16777215.0/16777216.0)
+#define COEF_MAX ( 16777215.0/16777216.0)
 typedef float SampleType;
 typedef float CoefType;
 typedef float CoefSumType;
 #else
 #define AC3_NAME(x) ff_ac3_fixed_ ## x
 #define MAC_COEF(d,a,b) MAC64(d,a,b)
+#define COEF_MIN -16777215
+#define COEF_MAX  16777215
 typedef int16_t SampleType;
 typedef int32_t CoefType;
 typedef int64_t CoefSumType;
diff --git a/libavcodec/ac3enc_fixed.c b/libavcodec/ac3enc_fixed.c
index b189609..cbe92e1 100644
--- a/libavcodec/ac3enc_fixed.c
+++ b/libavcodec/ac3enc_fixed.c
@@ -104,6 +104,15 @@ static void scale_coefficients(AC3EncodeContext *s)
 }
 
 
+/**
+ * Clip MDCT coefficients to allowable range.
+ */
+static void clip_coefficients(DSPContext *dsp, int32_t *coef, unsigned int len)
+{
+    dsp->vector_clip_int32(coef, coef, COEF_MIN, COEF_MAX, len);
+}
+
+
 static av_cold int ac3_fixed_encode_init(AVCodecContext *avctx)
 {
     AC3EncodeContext *s = avctx->priv_data;
diff --git a/libavcodec/ac3enc_float.c b/libavcodec/ac3enc_float.c
index 7d01b18..e21b99d 100644
--- a/libavcodec/ac3enc_float.c
+++ b/libavcodec/ac3enc_float.c
@@ -111,6 +111,15 @@ static void scale_coefficients(AC3EncodeContext *s)
 }
 
 
+/**
+ * Clip MDCT coefficients to allowable range.
+ */
+static void clip_coefficients(DSPContext *dsp, float *coef, unsigned int len)
+{
+    dsp->vector_clipf(coef, coef, COEF_MIN, COEF_MAX, len);
+}
+
+
 #if CONFIG_AC3_ENCODER
 AVCodec ff_ac3_float_encoder = {
     "ac3_float",
diff --git a/libavcodec/ac3enc_template.c b/libavcodec/ac3enc_template.c
index 85eea54..c7243c7 100644
--- a/libavcodec/ac3enc_template.c
+++ b/libavcodec/ac3enc_template.c
@@ -41,6 +41,8 @@ static void apply_window(DSPContext *dsp, SampleType *output,
 
 static int normalize_samples(AC3EncodeContext *s);
 
+static void clip_coefficients(DSPContext *dsp, CoefType *coef, unsigned int len);
+
 
 int AC3_NAME(allocate_sample_buffers)(AC3EncodeContext *s)
 {
@@ -171,8 +173,8 @@ static void apply_channel_coupling(AC3EncodeContext *s)
                 cpl_coef[i] += ch_coef[i];
         }
 
-        /* coefficients must be clipped to +/- 1.0 in order to be encoded */
-        s->dsp.vector_clipf(cpl_coef, cpl_coef, -1.0f, 1.0f, num_cpl_coefs);
+        /* coefficients must be clipped in order to be encoded */
+        clip_coefficients(&s->dsp, cpl_coef, num_cpl_coefs);
 
         /* scale coupling coefficients from float to 24-bit fixed-point */
         s->ac3dsp.float_to_fixed24(&block->fixed_coef[CPL_CH][cpl_start],
@@ -300,6 +302,7 @@ static void apply_channel_coupling(AC3EncodeContext *s)
         if (!block->cpl_in_use || !block->new_cpl_coords)
             continue;
 
+        clip_coefficients(&s->dsp, cpl_coords[blk][1], s->fbw_channels * 16);
         s->ac3dsp.float_to_fixed24(fixed_cpl_coords[blk][1],
                                    cpl_coords[blk][1],
                                    s->fbw_channels * 16);
@@ -433,7 +436,11 @@ int AC3_NAME(encode_frame)(AVCodecContext *avctx, unsigned char *frame,
 
     apply_mdct(s);
 
-    scale_coefficients(s);
+    if (s->fixed_point)
+        scale_coefficients(s);
+
+    clip_coefficients(&s->dsp, s->blocks[0].mdct_coef[1],
+                      AC3_MAX_COEFS * AC3_MAX_BLOCKS * s->channels);
 
     s->cpl_on = s->cpl_enabled;
     ff_ac3_compute_coupling_strategy(s);
@@ -443,6 +450,9 @@ int AC3_NAME(encode_frame)(AVCodecContext *avctx, unsigned char *frame,
 
     compute_rematrixing_strategy(s);
 
+    if (!s->fixed_point)
+        scale_coefficients(s);
+
     ff_ac3_apply_rematrixing(s);
 
     ff_ac3_process_exponents(s);
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 0e596b1..bfbe12e 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -2664,6 +2664,22 @@ static void apply_window_int16_c(int16_t *output, const int16_t *input,
     }
 }
 
+static void vector_clip_int32_c(int32_t *dst, const int32_t *src, int32_t min,
+                                int32_t max, unsigned int len)
+{
+    do {
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        *dst++ = av_clip(*src++, min, max);
+        len -= 8;
+    } while (len > 0);
+}
+
 #define W0 2048
 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */
 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */
@@ -3106,6 +3122,7 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
     c->scalarproduct_int16 = scalarproduct_int16_c;
     c->scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_c;
     c->apply_window_int16 = apply_window_int16_c;
+    c->vector_clip_int32 = vector_clip_int32_c;
     c->scalarproduct_float = scalarproduct_float_c;
     c->butterflies_float = butterflies_float_c;
     c->vector_fmul_scalar = vector_fmul_scalar_c;
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index f2054a4..401a87a 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -553,6 +553,22 @@ typedef struct DSPContext {
     void (*apply_window_int16)(int16_t *output, const int16_t *input,
                                const int16_t *window, unsigned int len);
 
+    /**
+     * Clip each element in an array of int32_t to a given minimum and maximum value.
+     * @param dst  destination array
+     *             constraints: 16-byte aligned
+     * @param src  source array
+     *             constraints: 16-byte aligned
+     * @param min  minimum value
+     *             constraints: must in the the range [-(1<<24), 1<<24]
+     * @param max  maximum value
+     *             constraints: must in the the range [-(1<<24), 1<<24]
+     * @param len  number of elements in the array
+     *             constraints: multiple of 32 greater than zero
+     */
+    void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
+                              int32_t max, unsigned int len);
+
     /* rv30 functions */
     qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
     qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
diff --git a/libavcodec/get_bits.h b/libavcodec/get_bits.h
index 3b09dfd..96d33b3 100644
--- a/libavcodec/get_bits.h
+++ b/libavcodec/get_bits.h
@@ -201,19 +201,11 @@ static inline void skip_bits_long(GetBitContext *s, int n){
         }                                                               \
     } while (0)
 
-#if ARCH_X86
-#   define SKIP_CACHE(name, gb, num)                            \
-    __asm__("shldl %2, %1, %0          \n\t"                    \
-            "shll  %2, %1              \n\t"                    \
-            : "+r" (name##_cache0), "+r" (name##_cache1)        \
-            : "Ic" ((uint8_t)(num)))
-#else
 #   define SKIP_CACHE(name, gb, num) do {               \
         name##_cache0 <<= (num);                        \
         name##_cache0 |= NEG_USR32(name##_cache1,num);  \
         name##_cache1 <<= (num);                        \
     } while (0)
-#endif
 
 #   define SKIP_COUNTER(name, gb, num) name##_bit_count += (num)
 
diff --git a/libavcodec/x86/ac3dsp.asm b/libavcodec/x86/ac3dsp.asm
index 99c5df3..8c958a1 100644
--- a/libavcodec/x86/ac3dsp.asm
+++ b/libavcodec/x86/ac3dsp.asm
@@ -32,6 +32,11 @@ cextern ac3_bap_bits
 pw_bap_mul1: dw 21846, 21846, 0, 32768, 21846, 21846, 0, 32768
 pw_bap_mul2: dw 5, 7, 0, 7, 5, 7, 0, 7
 
+; used in ff_ac3_extract_exponents()
+pd_1:   times 4 dd 1
+pd_151: times 4 dd 151
+pb_shuf_4dwb: db 0, 4, 8, 12
+
 SECTION .text
 
 ;-----------------------------------------------------------------------------
@@ -346,3 +351,100 @@ cglobal ac3_compute_mantissa_size_sse2, 1,2,4, mant_cnt, sum
     movd       eax, m0
     add        eax, sumd
     RET
+
+;------------------------------------------------------------------------------
+; void ff_ac3_extract_exponents(uint8_t *exp, int32_t *coef, int nb_coefs)
+;------------------------------------------------------------------------------
+
+%macro PABSD_MMX 2 ; src/dst, tmp
+    pxor     %2, %2
+    pcmpgtd  %2, %1
+    pxor     %1, %2
+    psubd    %1, %2
+%endmacro
+
+%macro PABSD_SSSE3 1-2 ; src/dst, unused
+    pabsd    %1, %1
+%endmacro
+
+%ifdef HAVE_AMD3DNOW
+INIT_MMX
+cglobal ac3_extract_exponents_3dnow, 3,3,0, exp, coef, len
+    add      expq, lenq
+    lea     coefq, [coefq+4*lenq]
+    neg      lenq
+    movq       m3, [pd_1]
+    movq       m4, [pd_151]
+.loop:
+    movq       m0, [coefq+4*lenq  ]
+    movq       m1, [coefq+4*lenq+8]
+    PABSD_MMX  m0, m2
+    PABSD_MMX  m1, m2
+    pslld      m0, 1
+    por        m0, m3
+    pi2fd      m2, m0
+    psrld      m2, 23
+    movq       m0, m4
+    psubd      m0, m2
+    pslld      m1, 1
+    por        m1, m3
+    pi2fd      m2, m1
+    psrld      m2, 23
+    movq       m1, m4
+    psubd      m1, m2
+    packssdw   m0, m0
+    packuswb   m0, m0
+    packssdw   m1, m1
+    packuswb   m1, m1
+    punpcklwd  m0, m1
+    movd  [expq+lenq], m0
+    add      lenq, 4
+    jl .loop
+    REP_RET
+%endif
+
+%macro AC3_EXTRACT_EXPONENTS 1
+cglobal ac3_extract_exponents_%1, 3,3,5, exp, coef, len
+    add     expq, lenq
+    lea    coefq, [coefq+4*lenq]
+    neg     lenq
+    mova      m2, [pd_1]
+    mova      m3, [pd_151]
+%ifidn %1, ssse3 ;
+    movd      m4, [pb_shuf_4dwb]
+%endif
+.loop:
+    ; move 4 32-bit coefs to xmm0
+    mova      m0, [coefq+4*lenq]
+    ; absolute value
+    PABSD     m0, m1
+    ; convert to float and extract exponents
+    pslld     m0, 1
+    por       m0, m2
+    cvtdq2ps  m1, m0
+    psrld     m1, 23
+    mova      m0, m3
+    psubd     m0, m1
+    ; move the lowest byte in each of 4 dwords to the low dword
+%ifidn %1, ssse3
+    pshufb    m0, m4
+%else
+    packssdw  m0, m0
+    packuswb  m0, m0
+%endif
+    movd  [expq+lenq], m0
+
+    add     lenq, 4
+    jl .loop
+    REP_RET
+%endmacro
+
+%ifdef HAVE_SSE
+INIT_XMM
+%define PABSD PABSD_MMX
+AC3_EXTRACT_EXPONENTS sse2
+%ifdef HAVE_SSSE3
+%define PABSD PABSD_SSSE3
+AC3_EXTRACT_EXPONENTS ssse3
+%endif
+%endif
diff --git a/libavcodec/x86/ac3dsp_mmx.c b/libavcodec/x86/ac3dsp_mmx.c
index e853b88..3127570 100644
--- a/libavcodec/x86/ac3dsp_mmx.c
+++ b/libavcodec/x86/ac3dsp_mmx.c
@@ -44,6 +44,10 @@ extern void ff_float_to_fixed24_sse2 (int32_t *dst, const float *src, unsigned i
 
 extern int ff_ac3_compute_mantissa_size_sse2(uint16_t mant_cnt[6][16]);
 
+extern void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_coefs);
+extern void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs);
+extern void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs);
+
 av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
 {
     int mm_flags = av_get_cpu_flags();
@@ -56,6 +60,7 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
         c->ac3_rshift_int32 = ff_ac3_rshift_int32_mmx;
     }
     if (mm_flags & AV_CPU_FLAG_3DNOW && HAVE_AMD3DNOW) {
+        c->extract_exponents = ff_ac3_extract_exponents_3dnow;
         if (!bit_exact) {
             c->float_to_fixed24 = ff_float_to_fixed24_3dnow;
         }
@@ -72,6 +77,7 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
         c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_sse2;
         c->float_to_fixed24 = ff_float_to_fixed24_sse2;
         c->compute_mantissa_size = ff_ac3_compute_mantissa_size_sse2;
+        c->extract_exponents = ff_ac3_extract_exponents_sse2;
         if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
             c->ac3_lshift_int16 = ff_ac3_lshift_int16_sse2;
             c->ac3_rshift_int32 = ff_ac3_rshift_int32_sse2;
@@ -79,6 +85,9 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact)
     }
     if (mm_flags & AV_CPU_FLAG_SSSE3 && HAVE_SSSE3) {
         c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_ssse3;
+        if (!(mm_flags & AV_CPU_FLAG_ATOM)) {
+            c->extract_exponents = ff_ac3_extract_exponents_ssse3;
+        }
     }
 #endif
 }
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 78cad4c..80bb6cd 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -2333,6 +2333,15 @@ int  ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, int w, i
 
 float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
 
+void ff_vector_clip_int32_mmx     (int32_t *dst, const int32_t *src, int32_t min,
+                                   int32_t max, unsigned int len);
+void ff_vector_clip_int32_sse2    (int32_t *dst, const int32_t *src, int32_t min,
+                                   int32_t max, unsigned int len);
+void ff_vector_clip_int32_sse2_int(int32_t *dst, const int32_t *src, int32_t min,
+                                   int32_t max, unsigned int len);
+void ff_vector_clip_int32_sse41   (int32_t *dst, const int32_t *src, int32_t min,
+                                   int32_t max, unsigned int len);
+
 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 {
     int mm_flags = av_get_cpu_flags();
@@ -2473,6 +2482,8 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 
         c->put_rv40_chroma_pixels_tab[0]= ff_put_rv40_chroma_mc8_mmx;
         c->put_rv40_chroma_pixels_tab[1]= ff_put_rv40_chroma_mc4_mmx;
+
+        c->vector_clip_int32 = ff_vector_clip_int32_mmx;
 #endif
 
         if (mm_flags & AV_CPU_FLAG_MMX2) {
@@ -2756,6 +2767,11 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 #if HAVE_YASM
             c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
             c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
+            if (mm_flags & AV_CPU_FLAG_ATOM) {
+                c->vector_clip_int32 = ff_vector_clip_int32_sse2_int;
+            } else {
+                c->vector_clip_int32 = ff_vector_clip_int32_sse2;
+            }
             if (avctx->flags & CODEC_FLAG_BITEXACT) {
                 c->apply_window_int16 = ff_apply_window_int16_sse2_ba;
             } else {
@@ -2781,6 +2797,13 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             }
 #endif
         }
+
+        if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
+#if HAVE_YASM
+            c->vector_clip_int32 = ff_vector_clip_int32_sse41;
+#endif
+        }
+
 #if HAVE_AVX && HAVE_YASM
         if (mm_flags & AV_CPU_FLAG_AVX) {
             if (bit_depth == 10) {
diff --git a/libavcodec/x86/dsputil_yasm.asm b/libavcodec/x86/dsputil_yasm.asm
index 695aba5..1f5a4f6 100644
--- a/libavcodec/x86/dsputil_yasm.asm
+++ b/libavcodec/x86/dsputil_yasm.asm
@@ -1048,3 +1048,118 @@ emu_edge sse
 %ifdef ARCH_X86_32
 emu_edge mmx
 %endif
+
+;-----------------------------------------------------------------------------
+; void ff_vector_clip_int32(int32_t *dst, const int32_t *src, int32_t min,
+;                           int32_t max, unsigned int len)
+;-----------------------------------------------------------------------------
+
+%macro PMINSD_MMX 3 ; dst, src, tmp
+    mova      %3, %2
+    pcmpgtd   %3, %1
+    pxor      %1, %2
+    pand      %1, %3
+    pxor      %1, %2
+%endmacro
+
+%macro PMAXSD_MMX 3 ; dst, src, tmp
+    mova      %3, %1
+    pcmpgtd   %3, %2
+    pand      %1, %3
+    pandn     %3, %2
+    por       %1, %3
+%endmacro
+
+%macro CLIPD_MMX 3-4 ; src/dst, min, max, tmp
+    PMINSD_MMX %1, %3, %4
+    PMAXSD_MMX %1, %2, %4
+%endmacro
+
+%macro CLIPD_SSE2 3-4 ; src/dst, min (float), max (float), unused
+    cvtdq2ps  %1, %1
+    minps     %1, %3
+    maxps     %1, %2
+    cvtps2dq  %1, %1
+%endmacro
+
+%macro CLIPD_SSE41 3-4 ;  src/dst, min, max, unused
+    pminsd  %1, %3
+    pmaxsd  %1, %2
+%endmacro
+
+%macro SPLATD_MMX 1
+    punpckldq  %1, %1
+%endmacro
+
+%macro SPLATD_SSE2 1
+    pshufd  %1, %1, 0
+%endmacro
+
+%macro VECTOR_CLIP_INT32 4
+cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len
+%ifidn %1, sse2
+    cvtsi2ss  m4, minm
+    cvtsi2ss  m5, maxm
+%else
+    movd      m4, minm
+    movd      m5, maxm
+%endif
+    SPLATD    m4
+    SPLATD    m5
+.loop:
+%assign %%i 1
+%rep %3
+    mova      m0,  [srcq+mmsize*0*%%i]
+    mova      m1,  [srcq+mmsize*1*%%i]
+    mova      m2,  [srcq+mmsize*2*%%i]
+    mova      m3,  [srcq+mmsize*3*%%i]
+%if %4
+    mova      m7,  [srcq+mmsize*4*%%i]
+    mova      m8,  [srcq+mmsize*5*%%i]
+    mova      m9,  [srcq+mmsize*6*%%i]
+    mova      m10, [srcq+mmsize*7*%%i]
+%endif
+    CLIPD  m0,  m4, m5, m6
+    CLIPD  m1,  m4, m5, m6
+    CLIPD  m2,  m4, m5, m6
+    CLIPD  m3,  m4, m5, m6
+%if %4
+    CLIPD  m7,  m4, m5, m6
+    CLIPD  m8,  m4, m5, m6
+    CLIPD  m9,  m4, m5, m6
+    CLIPD  m10, m4, m5, m6
+%endif
+    mova  [dstq+mmsize*0*%%i], m0
+    mova  [dstq+mmsize*1*%%i], m1
+    mova  [dstq+mmsize*2*%%i], m2
+    mova  [dstq+mmsize*3*%%i], m3
+%if %4
+    mova  [dstq+mmsize*4*%%i], m7
+    mova  [dstq+mmsize*5*%%i], m8
+    mova  [dstq+mmsize*6*%%i], m9
+    mova  [dstq+mmsize*7*%%i], m10
+%endif
+%assign %%i %%i+1
+%endrep
+    add     srcq, mmsize*4*(%3+%4)
+    add     dstq, mmsize*4*(%3+%4)
+    sub     lend, mmsize*(%3+%4)
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_MMX
+%define SPLATD SPLATD_MMX
+%define CLIPD CLIPD_MMX
+VECTOR_CLIP_INT32 mmx, 0, 1, 0
+INIT_XMM
+%define SPLATD SPLATD_SSE2
+VECTOR_CLIP_INT32 sse2_int, 6, 1, 0
+%define CLIPD CLIPD_SSE2
+VECTOR_CLIP_INT32 sse2, 6, 2, 0
+%define CLIPD CLIPD_SSE41
+%ifdef m8
+VECTOR_CLIP_INT32 sse41, 11, 1, 1
+%else
+VECTOR_CLIP_INT32 sse41, 6, 1, 0
+%endif