Altivec version of h264_(h|v)_loop_filter_luma

patch by Graham Booker % perian A cod3r P com% with some minor fixes by me.
historic of the patch: http://trac.perian.org/ticket/113
Original thread:
Date: May 11, 2007 9:45 PM
Subject: [FFmpeg-devel] [PATCH] Altivec version of-altivec h264_h-v_loop_filter_luma

Originally committed as revision 9264 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 268 insertions, 0 deletions

diff --git a/libavcodec/ppc/h264_altivec.c b/libavcodec/ppc/h264_altivec.c
index deffce893a..8910ac986f 100644
--- a/libavcodec/ppc/h264_altivec.c
+++ b/libavcodec/ppc/h264_altivec.c
@@ -600,6 +600,272 @@ void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) {
     ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel);
 }
 
+#define transpose4x16(r0, r1, r2, r3) {      \
+    register vector unsigned char r4;        \
+    register vector unsigned char r5;        \
+    register vector unsigned char r6;        \
+    register vector unsigned char r7;        \
+                                             \
+    r4 = vec_mergeh(r0, r2);  /*0, 2 set 0*/ \
+    r5 = vec_mergel(r0, r2);  /*0, 2 set 1*/ \
+    r6 = vec_mergeh(r1, r3);  /*1, 3 set 0*/ \
+    r7 = vec_mergel(r1, r3);  /*1, 3 set 1*/ \
+                                             \
+    r0 = vec_mergeh(r4, r6);  /*all set 0*/  \
+    r1 = vec_mergel(r4, r6);  /*all set 1*/  \
+    r2 = vec_mergeh(r5, r7);  /*all set 2*/  \
+    r3 = vec_mergel(r5, r7);  /*all set 3*/  \
+}
+
+static inline void write16x4(uint8_t *dst, int dst_stride,     register vector unsigned char r0,
+                             register vector unsigned char r1, register vector unsigned char r2,
+                             register vector unsigned char r3) {
+    DECLARE_ALIGNED_16(unsigned char, result[64]);
+    uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst;
+    int int_dst_stride = dst_stride/4;
+
+    vec_st(r0, 0, result);
+    vec_st(r1, 16, result);
+    vec_st(r2, 32, result);
+    vec_st(r3, 48, result);
+    /* FIXME: there has to be a better way!!!! */
+    *dst_int = *src_int;
+    *(dst_int+   int_dst_stride) = *(src_int + 1);
+    *(dst_int+ 2*int_dst_stride) = *(src_int + 2);
+    *(dst_int+ 3*int_dst_stride) = *(src_int + 3);
+    *(dst_int+ 4*int_dst_stride) = *(src_int + 4);
+    *(dst_int+ 5*int_dst_stride) = *(src_int + 5);
+    *(dst_int+ 6*int_dst_stride) = *(src_int + 6);
+    *(dst_int+ 7*int_dst_stride) = *(src_int + 7);
+    *(dst_int+ 8*int_dst_stride) = *(src_int + 8);
+    *(dst_int+ 9*int_dst_stride) = *(src_int + 9);
+    *(dst_int+10*int_dst_stride) = *(src_int + 10);
+    *(dst_int+11*int_dst_stride) = *(src_int + 11);
+    *(dst_int+12*int_dst_stride) = *(src_int + 12);
+    *(dst_int+13*int_dst_stride) = *(src_int + 13);
+    *(dst_int+14*int_dst_stride) = *(src_int + 14);
+    *(dst_int+15*int_dst_stride) = *(src_int + 15);
+}
+
+/** \brief performs a 6x16 transpose of data in src, and stores it to dst
+    \todo FIXME: see if we can't spare some vec_lvsl() by them factorizing
+    out of unaligned_load() */
+#define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\
+    register vector unsigned char r0  = unaligned_load(0,             src);\
+    register vector unsigned char r1  = unaligned_load(   src_stride, src);\
+    register vector unsigned char r2  = unaligned_load(2* src_stride, src);\
+    register vector unsigned char r3  = unaligned_load(3* src_stride, src);\
+    register vector unsigned char r4  = unaligned_load(4* src_stride, src);\
+    register vector unsigned char r5  = unaligned_load(5* src_stride, src);\
+    register vector unsigned char r6  = unaligned_load(6* src_stride, src);\
+    register vector unsigned char r7  = unaligned_load(7* src_stride, src);\
+    register vector unsigned char r14 = unaligned_load(14*src_stride, src);\
+    register vector unsigned char r15 = unaligned_load(15*src_stride, src);\
+                                                                           \
+    r8  = unaligned_load( 8*src_stride, src);                              \
+    r9  = unaligned_load( 9*src_stride, src);                              \
+    r10 = unaligned_load(10*src_stride, src);                              \
+    r11 = unaligned_load(11*src_stride, src);                              \
+    r12 = unaligned_load(12*src_stride, src);                              \
+    r13 = unaligned_load(13*src_stride, src);                              \
+                                                                           \
+    /*Merge first pairs*/                                                  \
+    r0 = vec_mergeh(r0, r8);    /*0, 8*/                                   \
+    r1 = vec_mergeh(r1, r9);    /*1, 9*/                                   \
+    r2 = vec_mergeh(r2, r10);   /*2,10*/                                   \
+    r3 = vec_mergeh(r3, r11);   /*3,11*/                                   \
+    r4 = vec_mergeh(r4, r12);   /*4,12*/                                   \
+    r5 = vec_mergeh(r5, r13);   /*5,13*/                                   \
+    r6 = vec_mergeh(r6, r14);   /*6,14*/                                   \
+    r7 = vec_mergeh(r7, r15);   /*7,15*/                                   \
+                                                                           \
+    /*Merge second pairs*/                                                 \
+    r8  = vec_mergeh(r0, r4);   /*0,4, 8,12 set 0*/                        \
+    r9  = vec_mergel(r0, r4);   /*0,4, 8,12 set 1*/                        \
+    r10 = vec_mergeh(r1, r5);   /*1,5, 9,13 set 0*/                        \
+    r11 = vec_mergel(r1, r5);   /*1,5, 9,13 set 1*/                        \
+    r12 = vec_mergeh(r2, r6);   /*2,6,10,14 set 0*/                        \
+    r13 = vec_mergel(r2, r6);   /*2,6,10,14 set 1*/                        \
+    r14 = vec_mergeh(r3, r7);   /*3,7,11,15 set 0*/                        \
+    r15 = vec_mergel(r3, r7);   /*3,7,11,15 set 1*/                        \
+                                                                           \
+    /*Third merge*/                                                        \
+    r0 = vec_mergeh(r8, r12);   /*0,2,4,6,8,10,12,14 set 0*/               \
+    r1 = vec_mergel(r8, r12);   /*0,2,4,6,8,10,12,14 set 1*/               \
+    r2 = vec_mergeh(r9, r13);   /*0,2,4,6,8,10,12,14 set 2*/               \
+    r4 = vec_mergeh(r10, r14);  /*1,3,5,7,9,11,13,15 set 0*/               \
+    r5 = vec_mergel(r10, r14);  /*1,3,5,7,9,11,13,15 set 1*/               \
+    r6 = vec_mergeh(r11, r15);  /*1,3,5,7,9,11,13,15 set 2*/               \
+    /* Don't need to compute 3 and 7*/                                     \
+                                                                           \
+    /*Final merge*/                                                        \
+    r8  = vec_mergeh(r0, r4);   /*all set 0*/                              \
+    r9  = vec_mergel(r0, r4);   /*all set 1*/                              \
+    r10 = vec_mergeh(r1, r5);   /*all set 2*/                              \
+    r11 = vec_mergel(r1, r5);   /*all set 3*/                              \
+    r12 = vec_mergeh(r2, r6);   /*all set 4*/                              \
+    r13 = vec_mergel(r2, r6);   /*all set 5*/                              \
+    /* Don't need to compute 14 and 15*/                                   \
+                                                                           \
+}
+
+// out: o = |x-y| < a
+static inline vector unsigned char diff_lt_altivec ( register vector unsigned char x,
+                                                     register vector unsigned char y,
+                                                     register vector unsigned char a) {
+
+    register vector unsigned char diff = vec_subs(x, y);
+    register vector unsigned char diffneg = vec_subs(y, x);
+    register vector unsigned char o = vec_or(diff, diffneg); /* |x-y| */
+    o = vec_cmplt(o, a);
+    return o;
+}
+
+static inline vector unsigned char h264_deblock_mask ( register vector unsigned char p0,
+                                                       register vector unsigned char p1,
+                                                       register vector unsigned char q0,
+                                                       register vector unsigned char q1,
+                                                       register vector unsigned char alpha,
+                                                       register vector unsigned char beta) {
+
+    register vector unsigned char mask;
+    register vector unsigned char tempmask;
+
+    mask = diff_lt_altivec(p0, q0, alpha);
+    tempmask = diff_lt_altivec(p1, p0, beta);
+    mask = vec_and(mask, tempmask);
+    tempmask = diff_lt_altivec(q1, q0, beta);
+    mask = vec_and(mask, tempmask);
+
+    return mask;
+}
+
+// out: p1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
+#define h264_deblock_q1(p0, p1, p2, q0, tc0) {                     \
+                                                                   \
+    register vector unsigned char average = vec_avg(p0, q0);       \
+    register vector unsigned char temp;                            \
+    register vector unsigned char uncliped;                        \
+    register vector unsigned char ones;                            \
+    register vector unsigned char max;                             \
+    register vector unsigned char min;                             \
+                                                                   \
+    temp = vec_xor(average, p2);                                   \
+    average = vec_avg(average, p2);     /*avg(p2, avg(p0, q0)) */  \
+    ones = vec_splat_u8(1);                                        \
+    temp = vec_and(temp, ones);         /*(p2^avg(p0, q0)) & 1 */  \
+    uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */\
+    max = vec_adds(p1, tc0);                                       \
+    min = vec_subs(p1, tc0);                                       \
+    p1 = vec_max(min, uncliped);                                   \
+    p1 = vec_min(max, p1);                                         \
+}
+
+#define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) {                                           \
+                                                                                                  \
+    const vec_u8_t A0v = (vec_u8_t) AVV(0xA0,0xA0,0xA0,0xA0,0xA0,0xA0,0xA0,0xA0,                  \
+                                        0xA0,0xA0,0xA0,0xA0,0xA0,0xA0,0xA0,0xA0);                 \
+                                                                                                  \
+    register vector unsigned char pq0bit = vec_xor(p0,q0);                                        \
+    register vector unsigned char temp;                                                           \
+    register vector unsigned char q1minus;                                                        \
+    register vector unsigned char p0minus;                                                        \
+    register vector unsigned char stage1;                                                         \
+    register vector unsigned char stage2;                                                         \
+    register vector unsigned char vec160;                                                         \
+    register vector unsigned char delta;                                                          \
+    register vector unsigned char deltaneg;                                                       \
+                                                                                                  \
+    temp = vec_cmpeq(p0, p0);                                                                     \
+    q1minus = vec_xor(temp, q1);               /* 255 - q1 */                                     \
+    stage1 = vec_avg(p1, q1minus);             /* (p1 - q1 + 256)>>1 */                           \
+    stage2 = vec_sr(stage1, vec_splat_u8(1));  /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */     \
+    p0minus = vec_xor(temp, p0);               /* 255 - p0 */                                     \
+    stage1 = vec_avg(q0, p0minus);             /* (q0 - p0 + 256)>>1 */                           \
+    pq0bit = vec_and(pq0bit, vec_splat_u8(1));                                                    \
+    stage2 = vec_avg(stage2, pq0bit);          /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
+    stage2 = vec_adds(stage2, stage1);         /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */  \
+    vec160 = vec_ld(0, &A0v);                                                                     \
+    deltaneg = vec_subs(vec160, stage2);       /* -d */                                           \
+    delta = vec_subs(stage2, vec160);          /* d */                                            \
+    deltaneg = vec_min(tc0masked, deltaneg);                                                      \
+    delta = vec_min(tc0masked, delta);                                                            \
+    p0 = vec_subs(p0, deltaneg);                                                                  \
+    q0 = vec_subs(q0, delta);                                                                     \
+    p0 = vec_adds(p0, delta);                                                                     \
+    q0 = vec_adds(q0, deltaneg);                                                                  \
+}
+
+#define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) {            \
+    DECLARE_ALIGNED_16(unsigned char, temp[16]);                                             \
+    register vector unsigned char alphavec;                                                  \
+    register vector unsigned char betavec;                                                   \
+    register vector unsigned char mask;                                                      \
+    register vector unsigned char p1mask;                                                    \
+    register vector unsigned char q1mask;                                                    \
+    register vector char tc0vec;                                                             \
+    register vector unsigned char finaltc0;                                                  \
+    register vector unsigned char tc0masked;                                                 \
+                                                                                             \
+    temp[0] = alpha;                                                                         \
+    temp[1] = beta;                                                                          \
+    alphavec = vec_ld(0, temp);                                                              \
+    betavec = vec_splat(alphavec, 0x1);                                                      \
+    alphavec = vec_splat(alphavec, 0x0);                                                     \
+    mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */            \
+                                                                                             \
+    *((int *)temp) = *((int *)tc0);                                                          \
+    tc0vec = vec_ld(0, temp);                                                                \
+    tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
+    tc0vec = vec_mergeh(tc0vec, tc0vec);                                                     \
+    mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_u8(-1)));  /* if tc0[i] >= 0 */         \
+    finaltc0 = vec_and(tc0vec, mask);                           /*tc = tc0[i]*/              \
+                                                                                             \
+    p1mask = diff_lt_altivec(p2, p0, betavec);                                               \
+    p1mask = vec_and(p1mask, mask);                             /* if( |p2 - p0| < beta) */  \
+    tc0masked = vec_and(p1mask, tc0vec);                                                     \
+    finaltc0 = vec_sub(finaltc0, p1mask);                       /* tc++ */                   \
+    h264_deblock_q1(p0, p1, p2, q0, tc0masked);                                              \
+    /*end if*/                                                                               \
+                                                                                             \
+    q1mask = diff_lt_altivec(q2, q0, betavec);                                               \
+    q1mask = vec_and(q1mask, mask);                             /* if ( |q2 - q0| < beta ) */\
+    tc0masked = vec_and(q1mask, tc0vec);                                                     \
+    finaltc0 = vec_sub(finaltc0, q1mask);                       /* tc++ */                   \
+    h264_deblock_q1(p0, q1, q2, q0, tc0masked);                                              \
+    /*end if*/                                                                               \
+                                                                                             \
+    h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0);                                            \
+}
+
+static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
+
+    if((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) {
+        register vector unsigned char p2 = vec_ld(-3*stride, pix);
+        register vector unsigned char p1 = vec_ld(-2*stride, pix);
+        register vector unsigned char p0 = vec_ld(-1*stride, pix);
+        register vector unsigned char q0 = vec_ld(0, pix);
+        register vector unsigned char q1 = vec_ld(stride, pix);
+        register vector unsigned char q2 = vec_ld(2*stride, pix);
+        h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0);
+        vec_st(p1, -2*stride, pix);
+        vec_st(p0, -1*stride, pix);
+        vec_st(q0, 0, pix);
+        vec_st(q1, stride, pix);
+    }
+}
+
+static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) {
+
+    register vector unsigned char line0, line1, line2, line3, line4, line5;
+    if((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0)
+        return;
+    readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5);
+    h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0);
+    transpose4x16(line1, line2, line3, line4);
+    write16x4(pix-2, stride, line1, line2, line3, line4);
+}
+
 void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
 
 #ifdef HAVE_ALTIVEC
@@ -609,6 +875,8 @@ void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
     c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
     c->h264_idct_add = ff_h264_idct_add_altivec;
     c->h264_idct8_add = ff_h264_idct8_add_altivec;
+    c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec;
+    c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec;
 
 #define dspfunc(PFX, IDX, NUM) \
     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \