ac3enc: add int32_t array clipping function to DSPUtil, including x86 versions.

libavcodec/dsputil.c

@@ -2676,6 +2676,22 @@ 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) */
@@ -3122,6 +3138,7 @@ 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;

libavcodec/dsputil.h

@@ -555,6 +555,22 @@ 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];

libavcodec/x86/dsputil_mmx.c

@@ -2429,6 +2429,15 @@ 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();
@@ -2570,6 +2579,8 @@ 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) {
@@ -2855,6 +2866,11 @@ 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 {
@@ -2880,6 +2896,13 @@ 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) {

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