Alpha: fix pix_abs16

Originally committed as revision 16675 to svn://svn.ffmpeg.org/ffmpeg/trunk

2 files changed, 55 insertions, 62 deletions

diff --git a/libavcodec/alpha/dsputil_alpha.c b/libavcodec/alpha/dsputil_alpha.c
index 74f5217a72..34e892c5a8 100644
--- a/libavcodec/alpha/dsputil_alpha.c
+++ b/libavcodec/alpha/dsputil_alpha.c
@@ -42,7 +42,7 @@ void get_pixels_mvi(DCTELEM *restrict block,
 void diff_pixels_mvi(DCTELEM *block, const uint8_t *s1, const uint8_t *s2,
                      int stride);
 int pix_abs8x8_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h);
-int pix_abs16x16_mvi_asm(uint8_t *pix1, uint8_t *pix2, int line_size);
+int pix_abs16x16_mvi_asm(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h);
 int pix_abs16x16_x2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h);
 int pix_abs16x16_y2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h);
 int pix_abs16x16_xy2_mvi(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h);
@@ -287,11 +287,6 @@ void put_pixels16_axp_asm(uint8_t *block, const uint8_t *pixels,
     put_pixels_axp_asm(block + 8, pixels + 8, line_size, h);
 }
 
-static int sad16x16_mvi(void *s, uint8_t *a, uint8_t *b, int stride)
-{
-    return pix_abs16x16_mvi_asm(a, b, stride);
-}
-
 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx)
 {
     c->put_pixels_tab[0][0] = put_pixels16_axp_asm;
@@ -343,10 +338,9 @@ void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx)
 
         c->get_pixels       = get_pixels_mvi;
         c->diff_pixels      = diff_pixels_mvi;
-        c->sad[0]           = sad16x16_mvi;
+        c->sad[0]           = pix_abs16x16_mvi_asm;
         c->sad[1]           = pix_abs8x8_mvi;
-//        c->pix_abs[0][0]    = pix_abs16x16_mvi_asm; //FIXME function arguments for the asm must be fixed
-        c->pix_abs[0][0]    = sad16x16_mvi;
+        c->pix_abs[0][0]    = pix_abs16x16_mvi_asm;
         c->pix_abs[1][0]    = pix_abs8x8_mvi;
         c->pix_abs[0][1]    = pix_abs16x16_x2_mvi;
         c->pix_abs[0][2]    = pix_abs16x16_y2_mvi;
diff --git a/libavcodec/alpha/motion_est_mvi_asm.S b/libavcodec/alpha/motion_est_mvi_asm.S
index db1e30d424..2a08b07b0d 100644
--- a/libavcodec/alpha/motion_est_mvi_asm.S
+++ b/libavcodec/alpha/motion_est_mvi_asm.S
@@ -60,9 +60,8 @@ pix_abs16x16_mvi_asm:
         jsr     AT, (AT), _mcount
 #endif
 
-        and     a1, 7, t0
+        and     a2, 7, t0
         clr     v0
-        lda     a3, 16
         beq     t0, $aligned
         .align 4
 $unaligned:
@@ -86,80 +85,80 @@ $unaligned:
            td:  error right  */
 
         /* load line 0 */
-        ldq_u   t0, 0(a1)       # left_u
-        ldq_u   t1, 8(a1)       # mid
-        ldq_u   t2, 16(a1)      # right_u
-        ldq     t3, 0(a0)       # ref left
-        ldq     t4, 8(a0)       # ref right
-        addq    a0, a2, a0      # pix1
-        addq    a1, a2, a1      # pix2
+        ldq_u   t0, 0(a2)       # left_u
+        ldq_u   t1, 8(a2)       # mid
+        ldq_u   t2, 16(a2)      # right_u
+        ldq     t3, 0(a1)       # ref left
+        ldq     t4, 8(a1)       # ref right
+        addq    a1, a3, a1      # pix1
+        addq    a2, a3, a2      # pix2
         /* load line 1 */
-        ldq_u   t5, 0(a1)       # left_u
-        ldq_u   t6, 8(a1)       # mid
-        ldq_u   t7, 16(a1)      # right_u
-        ldq     t8, 0(a0)       # ref left
-        ldq     t9, 8(a0)       # ref right
-        addq    a0, a2, a0      # pix1
-        addq    a1, a2, a1      # pix2
+        ldq_u   t5, 0(a2)       # left_u
+        ldq_u   t6, 8(a2)       # mid
+        ldq_u   t7, 16(a2)      # right_u
+        ldq     t8, 0(a1)       # ref left
+        ldq     t9, 8(a1)       # ref right
+        addq    a1, a3, a1      # pix1
+        addq    a2, a3, a2      # pix2
         /* calc line 0 */
-        extql   t0, a1, t0      # left lo
-        extqh   t1, a1, ta      # left hi
-        extql   t1, a1, tb      # right lo
+        extql   t0, a2, t0      # left lo
+        extqh   t1, a2, ta      # left hi
+        extql   t1, a2, tb      # right lo
         or      t0, ta, t0      # left
-        extqh   t2, a1, t2      # right hi
+        extqh   t2, a2, t2      # right hi
         perr    t3, t0, tc      # error left
         or      t2, tb, t2      # right
         perr    t4, t2, td      # error right
         addq    v0, tc, v0      # add error left
         addq    v0, td, v0      # add error left
         /* calc line 1 */
-        extql   t5, a1, t5      # left lo
-        extqh   t6, a1, ta      # left hi
-        extql   t6, a1, tb      # right lo
+        extql   t5, a2, t5      # left lo
+        extqh   t6, a2, ta      # left hi
+        extql   t6, a2, tb      # right lo
         or      t5, ta, t5      # left
-        extqh   t7, a1, t7      # right hi
+        extqh   t7, a2, t7      # right hi
         perr    t8, t5, tc      # error left
         or      t7, tb, t7      # right
         perr    t9, t7, td      # error right
         addq    v0, tc, v0      # add error left
         addq    v0, td, v0      # add error left
         /* loop */
-        subq    a3,  2, a3      # h -= 2
-        bne     a3, $unaligned
+        subq    a4,  2, a4      # h -= 2
+        bne     a4, $unaligned
         ret
 
         .align 4
 $aligned:
         /* load line 0 */
-        ldq     t0, 0(a1)       # left
-        ldq     t1, 8(a1)       # right
-        addq    a1, a2, a1      # pix2
-        ldq     t2, 0(a0)       # ref left
-        ldq     t3, 8(a0)       # ref right
-        addq    a0, a2, a0      # pix1
+        ldq     t0, 0(a2)       # left
+        ldq     t1, 8(a2)       # right
+        addq    a2, a3, a2      # pix2
+        ldq     t2, 0(a1)       # ref left
+        ldq     t3, 8(a1)       # ref right
+        addq    a1, a3, a1      # pix1
         /* load line 1 */
-        ldq     t4, 0(a1)       # left
-        ldq     t5, 8(a1)       # right
-        addq    a1, a2, a1      # pix2
-        ldq     t6, 0(a0)       # ref left
-        ldq     t7, 8(a0)       # ref right
-        addq    a0, a2, a0      # pix1
+        ldq     t4, 0(a2)       # left
+        ldq     t5, 8(a2)       # right
+        addq    a2, a3, a2      # pix2
+        ldq     t6, 0(a1)       # ref left
+        ldq     t7, 8(a1)       # ref right
+        addq    a1, a3, a1      # pix1
         /* load line 2 */
-        ldq     t8, 0(a1)       # left
-        ldq     t9, 8(a1)       # right
-        addq    a1, a2, a1      # pix2
-        ldq     ta, 0(a0)       # ref left
-        ldq     tb, 8(a0)       # ref right
-        addq    a0, a2, a0      # pix1
+        ldq     t8, 0(a2)       # left
+        ldq     t9, 8(a2)       # right
+        addq    a2, a3, a2      # pix2
+        ldq     ta, 0(a1)       # ref left
+        ldq     tb, 8(a1)       # ref right
+        addq    a1, a3, a1      # pix1
         /* load line 3 */
-        ldq     tc, 0(a1)       # left
-        ldq     td, 8(a1)       # right
-        addq    a1, a2, a1      # pix2
-        ldq     te, 0(a0)       # ref left
-        ldq     tf, 8(a0)       # ref right
+        ldq     tc, 0(a2)       # left
+        ldq     td, 8(a2)       # right
+        addq    a2, a3, a2      # pix2
+        ldq     te, 0(a1)       # ref left
+        ldq     a0, 8(a1)       # ref right
         /* calc line 0 */
         perr    t0, t2, t0      # error left
-        addq    a0, a2, a0      # pix1
+        addq    a1, a3, a1      # pix1
         perr    t1, t3, t1      # error right
         addq    v0, t0, v0      # add error left
         /* calc line 1 */
@@ -175,11 +174,11 @@ $aligned:
         /* calc line 3 */
         perr    tc, te, t0      # error left
         addq    v0, t1, v0      # add error right
-        perr    td, tf, t1      # error right
+        perr    td, a0, t1      # error right
         addq    v0, t0, v0      # add error left
         addq    v0, t1, v0      # add error right
         /* loop */
-        subq    a3,  4, a3      # h -= 4
-        bne     a3, $aligned
+        subq    a4,  4, a4      # h -= 4
+        bne     a4, $aligned
         ret
         .end pix_abs16x16_mvi_asm