Altivec version of avg_no_rnd_vc1_chroma_mc8

Originally committed as revision 18522 to svn://svn.ffmpeg.org/ffmpeg/trunk

2 files changed, 98 insertions, 127 deletions

diff --git a/libavcodec/ppc/h264_altivec.c b/libavcodec/ppc/h264_altivec.c
index be02786e6e..29a9c7315a 100644
--- a/libavcodec/ppc/h264_altivec.c
+++ b/libavcodec/ppc/h264_altivec.c
@@ -31,6 +31,7 @@
 
 #define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
 #define PREFIX_h264_chroma_mc8_altivec         put_h264_chroma_mc8_altivec
+#define PREFIX_no_rnd_vc1_chroma_mc8_altivec   put_no_rnd_vc1_chroma_mc8_altivec
 #define PREFIX_h264_chroma_mc8_num             altivec_put_h264_chroma_mc8_num
 #define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
@@ -41,6 +42,7 @@
 #include "h264_template_altivec.c"
 #undef OP_U8_ALTIVEC
 #undef PREFIX_h264_chroma_mc8_altivec
+#undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
 #undef PREFIX_h264_chroma_mc8_num
 #undef PREFIX_h264_qpel16_h_lowpass_altivec
 #undef PREFIX_h264_qpel16_h_lowpass_num
@@ -51,6 +53,7 @@
 
 #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
 #define PREFIX_h264_chroma_mc8_altivec         avg_h264_chroma_mc8_altivec
+#define PREFIX_no_rnd_vc1_chroma_mc8_altivec   avg_no_rnd_vc1_chroma_mc8_altivec
 #define PREFIX_h264_chroma_mc8_num             altivec_avg_h264_chroma_mc8_num
 #define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
 #define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
@@ -61,6 +64,7 @@
 #include "h264_template_altivec.c"
 #undef OP_U8_ALTIVEC
 #undef PREFIX_h264_chroma_mc8_altivec
+#undef PREFIX_no_rnd_vc1_chroma_mc8_altivec
 #undef PREFIX_h264_chroma_mc8_num
 #undef PREFIX_h264_qpel16_h_lowpass_altivec
 #undef PREFIX_h264_qpel16_h_lowpass_num
@@ -179,129 +183,6 @@ static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint
     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
 }\
 
-/* this code assume that stride % 16 == 0 */
-void put_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
-   DECLARE_ALIGNED_16(signed int, ABCD[4]) =
-                        {((8 - x) * (8 - y)),
-                         ((    x) * (8 - y)),
-                         ((8 - x) * (    y)),
-                         ((    x) * (    y))};
-    register int i;
-    vec_u8 fperm;
-    const vec_s32 vABCD = vec_ld(0, ABCD);
-    const vec_s16 vA = vec_splat((vec_s16)vABCD, 1);
-    const vec_s16 vB = vec_splat((vec_s16)vABCD, 3);
-    const vec_s16 vC = vec_splat((vec_s16)vABCD, 5);
-    const vec_s16 vD = vec_splat((vec_s16)vABCD, 7);
-    LOAD_ZERO;
-    const vec_s16 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
-    const vec_u16 v6us  = vec_splat_u16(6);
-    register int loadSecond     = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
-    register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
-
-    vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1;
-    vec_u8 vsrc0uc, vsrc1uc;
-    vec_s16 vsrc0ssH, vsrc1ssH;
-    vec_u8 vsrcCuc, vsrc2uc, vsrc3uc;
-    vec_s16 vsrc2ssH, vsrc3ssH, psum;
-    vec_u8 vdst, ppsum, fsum;
-
-    if (((unsigned long)dst) % 16 == 0) {
-        fperm = (vec_u8){0x10, 0x11, 0x12, 0x13,
-                         0x14, 0x15, 0x16, 0x17,
-                         0x08, 0x09, 0x0A, 0x0B,
-                         0x0C, 0x0D, 0x0E, 0x0F};
-    } else {
-        fperm = (vec_u8){0x00, 0x01, 0x02, 0x03,
-                         0x04, 0x05, 0x06, 0x07,
-                         0x18, 0x19, 0x1A, 0x1B,
-                         0x1C, 0x1D, 0x1E, 0x1F};
-    }
-
-    vsrcAuc = vec_ld(0, src);
-
-    if (loadSecond)
-        vsrcBuc = vec_ld(16, src);
-    vsrcperm0 = vec_lvsl(0, src);
-    vsrcperm1 = vec_lvsl(1, src);
-
-    vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
-    if (reallyBadAlign)
-        vsrc1uc = vsrcBuc;
-    else
-        vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
-
-    vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc0uc);
-    vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc1uc);
-
-    if (!loadSecond) {// -> !reallyBadAlign
-        for (i = 0 ; i < h ; i++) {
-
-
-            vsrcCuc = vec_ld(stride + 0, src);
-
-            vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
-            vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
-
-            vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc2uc);
-            vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc3uc);
-
-            psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
-            psum = vec_mladd(vB, vsrc1ssH, psum);
-            psum = vec_mladd(vC, vsrc2ssH, psum);
-            psum = vec_mladd(vD, vsrc3ssH, psum);
-            psum = vec_add(v28ss, psum);
-            psum = vec_sra(psum, v6us);
-
-            vdst = vec_ld(0, dst);
-            ppsum = (vec_u8)vec_packsu(psum, psum);
-            fsum = vec_perm(vdst, ppsum, fperm);
-
-            vec_st(fsum, 0, dst);
-
-            vsrc0ssH = vsrc2ssH;
-            vsrc1ssH = vsrc3ssH;
-
-            dst += stride;
-            src += stride;
-        }
-    } else {
-        vec_u8 vsrcDuc;
-        for (i = 0 ; i < h ; i++) {
-            vsrcCuc = vec_ld(stride + 0, src);
-            vsrcDuc = vec_ld(stride + 16, src);
-
-            vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
-            if (reallyBadAlign)
-                vsrc3uc = vsrcDuc;
-            else
-                vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
-
-            vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc2uc);
-            vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc3uc);
-
-            psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
-            psum = vec_mladd(vB, vsrc1ssH, psum);
-            psum = vec_mladd(vC, vsrc2ssH, psum);
-            psum = vec_mladd(vD, vsrc3ssH, psum);
-            psum = vec_add(v28ss, psum);
-            psum = vec_sr(psum, v6us);
-
-            vdst = vec_ld(0, dst);
-            ppsum = (vec_u8)vec_pack(psum, psum);
-            fsum = vec_perm(vdst, ppsum, fperm);
-
-            vec_st(fsum, 0, dst);
-
-            vsrc0ssH = vsrc2ssH;
-            vsrc1ssH = vsrc3ssH;
-
-            dst += stride;
-            src += stride;
-        }
-    }
-}
-
 static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
                                     const uint8_t * src2, int dst_stride,
                                     int src_stride1, int h)
@@ -1092,6 +973,7 @@ void dsputil_h264_init_ppc(DSPContext* c, AVCodecContext *avctx) {
         c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
         c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
         c->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_altivec;
+        c->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_altivec;
         c->h264_idct_add = ff_h264_idct_add_altivec;
         c->h264_idct_add8 = ff_h264_idct_add8_altivec;
         c->h264_idct_add16 = ff_h264_idct_add16_altivec;
diff --git a/libavcodec/ppc/h264_template_altivec.c b/libavcodec/ppc/h264_template_altivec.c
index b41024a03e..954fd16bd9 100644
--- a/libavcodec/ppc/h264_template_altivec.c
+++ b/libavcodec/ppc/h264_template_altivec.c
@@ -27,14 +27,15 @@
 
 /* this code assume that stride % 16 == 0 */
 
-#define CHROMA_MC8_ALTIVEC_CORE \
+#define CHROMA_MC8_ALTIVEC_CORE(BIAS1, BIAS2) \
         vsrc2ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc2uc);\
         vsrc3ssH = (vec_s16)vec_mergeh(zero_u8v,(vec_u8)vsrc3uc);\
 \
-        psum = vec_mladd(vA, vsrc0ssH, v32ss);\
+        psum = vec_mladd(vA, vsrc0ssH, BIAS1);\
         psum = vec_mladd(vB, vsrc1ssH, psum);\
         psum = vec_mladd(vC, vsrc2ssH, psum);\
         psum = vec_mladd(vD, vsrc3ssH, psum);\
+        psum = BIAS2(psum);\
         psum = vec_sr(psum, v6us);\
 \
         vdst = vec_ld(0, dst);\
@@ -71,6 +72,9 @@
         dst += stride;\
         src += stride;
 
+#define noop(a) a
+#define add28(a) vec_add(v28ss, a)
+
 void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
                                     int stride, int h, int x, int y) {
   POWERPC_PERF_DECLARE(PREFIX_h264_chroma_mc8_num, 1);
@@ -136,7 +140,7 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
                 vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
                 vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
 
-                CHROMA_MC8_ALTIVEC_CORE
+                CHROMA_MC8_ALTIVEC_CORE(v32ss, noop)
             }
         } else {
             vec_u8 vsrcDuc;
@@ -149,7 +153,7 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
                 else
                     vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
 
-                CHROMA_MC8_ALTIVEC_CORE
+                CHROMA_MC8_ALTIVEC_CORE(v32ss, noop)
             }
         }
     } else {
@@ -202,6 +206,91 @@ void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src,
     POWERPC_PERF_STOP_COUNT(PREFIX_h264_chroma_mc8_num, 1);
 }
 
+/* this code assume that stride % 16 == 0 */
+void PREFIX_no_rnd_vc1_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
+   DECLARE_ALIGNED_16(signed int, ABCD[4]) =
+                        {((8 - x) * (8 - y)),
+                         ((    x) * (8 - y)),
+                         ((8 - x) * (    y)),
+                         ((    x) * (    y))};
+    register int i;
+    vec_u8 fperm;
+    const vec_s32 vABCD = vec_ld(0, ABCD);
+    const vec_s16 vA = vec_splat((vec_s16)vABCD, 1);
+    const vec_s16 vB = vec_splat((vec_s16)vABCD, 3);
+    const vec_s16 vC = vec_splat((vec_s16)vABCD, 5);
+    const vec_s16 vD = vec_splat((vec_s16)vABCD, 7);
+    LOAD_ZERO;
+    const vec_s16 v28ss = vec_sub(vec_sl(vec_splat_s16(1),vec_splat_u16(5)),vec_splat_s16(4));
+    const vec_u16 v6us  = vec_splat_u16(6);
+    register int loadSecond     = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
+    register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;
+
+    vec_u8 vsrcAuc, av_uninit(vsrcBuc), vsrcperm0, vsrcperm1;
+    vec_u8 vsrc0uc, vsrc1uc;
+    vec_s16 vsrc0ssH, vsrc1ssH;
+    vec_u8 vsrcCuc, vsrc2uc, vsrc3uc;
+    vec_s16 vsrc2ssH, vsrc3ssH, psum;
+    vec_u8 vdst, ppsum, vfdst, fsum;
+
+    if (((unsigned long)dst) % 16 == 0) {
+        fperm = (vec_u8){0x10, 0x11, 0x12, 0x13,
+                         0x14, 0x15, 0x16, 0x17,
+                         0x08, 0x09, 0x0A, 0x0B,
+                         0x0C, 0x0D, 0x0E, 0x0F};
+    } else {
+        fperm = (vec_u8){0x00, 0x01, 0x02, 0x03,
+                         0x04, 0x05, 0x06, 0x07,
+                         0x18, 0x19, 0x1A, 0x1B,
+                         0x1C, 0x1D, 0x1E, 0x1F};
+    }
+
+    vsrcAuc = vec_ld(0, src);
+
+    if (loadSecond)
+        vsrcBuc = vec_ld(16, src);
+    vsrcperm0 = vec_lvsl(0, src);
+    vsrcperm1 = vec_lvsl(1, src);
+
+    vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
+    if (reallyBadAlign)
+        vsrc1uc = vsrcBuc;
+    else
+        vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);
+
+    vsrc0ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc0uc);
+    vsrc1ssH = (vec_s16)vec_mergeh(zero_u8v, (vec_u8)vsrc1uc);
+
+    if (!loadSecond) {// -> !reallyBadAlign
+        for (i = 0 ; i < h ; i++) {
+
+
+            vsrcCuc = vec_ld(stride + 0, src);
+
+            vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
+            vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);
+
+            CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28)
+        }
+    } else {
+        vec_u8 vsrcDuc;
+        for (i = 0 ; i < h ; i++) {
+            vsrcCuc = vec_ld(stride + 0, src);
+            vsrcDuc = vec_ld(stride + 16, src);
+
+            vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
+            if (reallyBadAlign)
+                vsrc3uc = vsrcDuc;
+            else
+                vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);
+
+            CHROMA_MC8_ALTIVEC_CORE(vec_splat_s16(0), add28)
+        }
+    }
+}
+
+#undef noop
+#undef add28
 #undef CHROMA_MC8_ALTIVEC_CORE
 
 /* this code assume stride % 16 == 0 */