bfin: Use more sensible file names
authorDiego Biurrun <diego@biurrun.de>
Fri, 14 Feb 2014 10:41:49 +0000 (11:41 +0100)
committerDiego Biurrun <diego@biurrun.de>
Thu, 13 Mar 2014 15:15:53 +0000 (08:15 -0700)
Drop non-informative _bfin suffix, rename some files for consistency with
other architectures, rename others to reflect their content.

15 files changed:
libavcodec/bfin/Makefile
libavcodec/bfin/dsputil.S [new file with mode: 0644]
libavcodec/bfin/dsputil_bfin.c [deleted file]
libavcodec/bfin/dsputil_init.c [new file with mode: 0644]
libavcodec/bfin/hpel_pixels_bfin.S [deleted file]
libavcodec/bfin/hpeldsp_bfin.c [deleted file]
libavcodec/bfin/hpeldsp_bfin.h [deleted file]
libavcodec/bfin/hpeldsp_init.c [new file with mode: 0644]
libavcodec/bfin/pixels.S [new file with mode: 0644]
libavcodec/bfin/pixels.h [new file with mode: 0644]
libavcodec/bfin/pixels_bfin.S [deleted file]
libavcodec/bfin/vp3_bfin.c [deleted file]
libavcodec/bfin/vp3_idct_bfin.S [deleted file]
libavcodec/bfin/vp3dsp.S [new file with mode: 0644]
libavcodec/bfin/vp3dsp_init.c [new file with mode: 0644]

index 4fa940b..193f8e6 100644 (file)
@@ -1,10 +1,10 @@
-OBJS += bfin/dsputil_bfin.o                                             \
+OBJS += bfin/dsputil_init.o                                             \
+        bfin/dsputil.o                                                  \
         bfin/fdct_bfin.o                                                \
-        bfin/hpel_pixels_bfin.o                                         \
         bfin/hpel_pixels_no_rnd.o                                       \
         bfin/idct_bfin.o                                                \
-        bfin/pixels_bfin.o                                              \
+        bfin/pixels.o                                                   \
 
-OBJS-$(CONFIG_HPELDSP)                  += bfin/hpeldsp_bfin.o
-OBJS-$(CONFIG_VP3DSP)                   += bfin/vp3_bfin.o              \
-                                           bfin/vp3_idct_bfin.o
+OBJS-$(CONFIG_HPELDSP)                  += bfin/hpeldsp_init.o
+OBJS-$(CONFIG_VP3DSP)                   += bfin/vp3dsp_init.o           \
+                                           bfin/vp3dsp.o
diff --git a/libavcodec/bfin/dsputil.S b/libavcodec/bfin/dsputil.S
new file mode 100644 (file)
index 0000000..e19455b
--- /dev/null
@@ -0,0 +1,382 @@
+/*
+ * Blackfin Pixel Operations
+ * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/bfin/asm.h"
+
+DEFUN(put_pixels_clamped,mL1,
+        (int16_t *block, uint8_t *dest, int line_size)):
+    [--SP] = (R7:4);
+    R4 = 0;
+    R5.l = 0x00ff;
+    R5.h = 0x00ff;
+    I0 = R0;         // block
+    I1 = R1;         // dest
+    R2 += -4;        // line_size
+    M1 = R2;
+    P0 = 8;
+    R0 = [I0++];
+    R1 = [I0++];
+    R2 = MAX(R0, R4) (V);
+    LSETUP (ppc$0,ppc$1) LC0=P0;
+ppc$0: R2 = MIN(R2, R5) (V);
+       R3 = MAX(R1, R4) (V);
+       R3 = MIN(R3, R5) (V)      || R0 = [I0++];
+       R6 = BYTEPACK (R2,R3)     || R1 = [I0++];
+       R2 = MAX(R0, R4) (V)      || [I1++] = R6;
+       R2 = MIN(R2, R5) (V);
+       R3 = MAX(R1, R4) (V);
+       R3 = MIN(R3, R5) (V)      || R0 = [I0++];
+       R6 = BYTEPACK (R2,R3)     || R1 = [I0++];
+ppc$1: R2 = Max(R0, R4) (V)      || [I1++M1] = R6;
+
+    (R7:4) = [SP++];
+    RTS;
+DEFUN_END(put_pixels_clamped)
+
+DEFUN(add_pixels_clamped,mL1,
+        (int16_t *block, uint8_t *dest, int line_size)):
+    [-- SP] = (R7:4);
+    R4 = 0;
+    I0 = 0;
+    R2 += -4;        // line_size
+    M0 = R2;
+    I1 = R1;         // dest
+    I3 = R0;         // block
+    I2 = R1;         // dest
+    P0 = 8;
+    M3 = 2;
+    R0 = [I3++]  || R2 = [I1];
+    R2 = R2 << 8                      || R0.H = W[I3--]  || R3 = [I1++];
+    R3 = R3 >> 8                      || R1.L = W[I3]    || I3 += 4;
+    R6 = BYTEOP3P(R1:0, R3:2) (LO)    || R1.H = W[I3++]  || R2 = [I1];
+
+    LSETUP(apc$2,apc$3) LC1 = P0;
+apc$2: R7 = BYTEOP3P(R1:0, R3:2) (HI, R) || R0 = [I3++]     || R3 = [I1++M0];
+       R2 = R2 << 8                      || R0.H = W[I3--];
+       R3 = R3 >> 8                      || R1.L = W[I3]    || I3 += 4;
+       R6 = R6 + R7 (S)                  || R1.H = W[I3];
+       R6 = BYTEOP3P(R1:0, R3:2) (LO)    || I3+=M3          || [I2++]=R6;
+       R7 = BYTEOP3P(R1:0, R3:2) (HI, R) || R0 = [I3++]     || R2 = [I1];
+       R2 = R2 << 8                      || R0.H = W[I3--]  || R3 = [I1++];
+       R3 = R3 >> 8                      || R1.L = W[I3]    || I3 += 4;
+       R6 = R6 + R7 (S)                  || R1.H = W[I3++];
+apc$3: R6 = BYTEOP3P(R1:0, R3:2) (LO)    || [I2++M0] = R6   || R2 = [I1];
+
+    (R7:4) = [SP++];
+    RTS;
+DEFUN_END(add_pixels_clamped)
+
+DEFUN(diff_pixels,mL1,
+       (int16_t *block, uint8_t *s1, uint8_t *s2, int stride)):
+        link 0;
+        [--sp] = (r7:4);
+        p0=8;
+        i3=r0;        // block
+        i0=r1;        // s1
+        i1=r2;        // s2
+        r2=[fp+20];   // stride
+        r2+=-8;
+        m0=r2;
+
+
+        LSETUP(.LS0,.LE0) LC0=P0;
+        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
+
+.LS0:   DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
+        (R5,R4) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++M0] || R2 = [I1++M0];
+        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R0 = [I0++]   || [I3++] = R4;
+        DISALGNEXCPT                       || R2 = [I1++]   || [I3++] = R5;
+        [i3++]=r6;
+.LE0:  [i3++]=r7;
+
+        (r7:4) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(diff_pixels)
+
+/*
+    for (i = 0; i < 16; i++) {
+        for (j = 0; j < 16; j++) {
+          sum += pix[j];
+        }
+        pix += line_size;
+    }
+*/
+DEFUN(pix_sum,mL1,
+        (uint8_t *p, int stride)):
+        link 0;
+        [--sp] = (r7:4);
+        p0=8;
+        i0=r0;        // s1
+        i1=r0;
+        m1=r1;
+        r1=r1+r1;
+        r1+=-16;       // stride
+        m0=r1;
+        i1+=m1;
+
+        r6=0;
+
+        LSETUP(LS$PS,LE$PS) LC0=P0;
+        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
+
+LS$PS:  DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
+        (R5,R4) = BYTEOP16P (R3:2,R1:0)    || R0 = [I0++]   || R2 = [I1++];
+        r6=r6+|+r5;
+        r6=r6+|+r4;
+        (R5,R4) = BYTEOP16P (R3:2,R1:0) (R)|| R1 = [I0++]   || R3 = [I1++];
+        r6=r6+|+r5;
+        r6=r6+|+r4;
+        (R5,R4) = BYTEOP16P (R3:2,R1:0)    || R0 = [I0++m0] || R2 = [I1++m0];
+        r6=r6+|+r5;
+        r6=r6+|+r4;
+        (R5,R4) = BYTEOP16P (R3:2,R1:0) (R)|| R0 = [I0++]   || R2 = [I1++];
+        r6=r6+|+r5;
+LE$PS:  r6=r6+|+r4;
+        r0.l=r6.l+r6.h;
+        r0.h=0;
+
+        (r7:4) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(pix_sum)
+
+
+DEFUN(get_pixels,mL1,
+        (int16_t *restrict block, const uint8_t *pixels, int line_size)):
+        [--sp] = (r7:4);
+        i3=r0;        // dest
+        i0=r1;        // src0
+        p0=8;
+        r2+=-8;
+        m0=r2;
+        LSETUP(gp8$0,gp8$1) LC0=P0;
+
+        DISALGNEXCPT                   || R0 = [I0++];
+        DISALGNEXCPT                   || R1 = [I0++];
+
+gp8$0:  (R7,R6) = byteunpack R1:0      || R0 = [I0++M0];
+        (R5,R4) = byteunpack R1:0 (R)  || R0 = [I0++]    || [I3++]=R6;
+        DISALGNEXCPT                   || R1 = [I0++]    || [I3++]=R7;
+        [I3++]=R4;
+gp8$1:  [I3++]=R5
+
+
+        (r7:4) = [sp++];
+        RTS;
+DEFUN_END(get_pixels)
+
+
+/* sad = sad16x16 (ubyte *mb, ubyte *refwin, srcwidth, refwinwidth, h) */
+/* 91 cycles */
+DEFUN(z_sad16x16,mL1,
+        (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h)):
+        link 0;
+        I0 = R0;
+        I1 = R1;
+
+        A1 = A0 = 0;
+        R0 = [sp+20]; // rwidth
+        P2 = [sp+24]; // height
+        R3 = 16;
+        R0 = R0 - R3;
+        R3 = R2 - R3;
+        M1 = R0;
+        M0 = R3;
+
+        DISALGNEXCPT         || R0 = [I0++]    || R2 = [I1++];
+        LSETUP (s$16, e$16) LC0=P2;
+s$16:   DISALGNEXCPT         || R1 = [I0++]    || R3 = [I1++];
+        SAA (R1:0,R3:2)      || R0 = [I0++]    || R2 = [I1++];
+        SAA (R1:0,R3:2) (R)  || R1 = [I0++]    || R3 = [I1++];
+        SAA (R1:0,R3:2)      || R0 = [I0++M0]  || R2 = [I1++M1];
+e$16:   SAA (R1:0,R3:2) (R)  || R0 = [I0++]    || R2 = [I1++];
+
+        R3=A1.L+A1.H,  R2=A0.L+A0.H ;
+        R0 = R2 + R3 ;
+        unlink;
+        RTS;
+DEFUN_END(z_sad16x16)
+
+/* sad = sad8x8 (ubyte *mb, ubyte *refwin, int srcwidth, int refwinwidth, int h) */
+/* 36 cycles */
+DEFUN(z_sad8x8,mL1,
+        (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h)):
+        I0 = R0;
+        I1 = R1;
+
+        A1 = A0 = 0;
+        r0 = [sp+12]; // rwidth
+        P2 = [sp+16]; //height
+        R3 = 8;
+        R0 = R0 - R3;
+        R3 = R2 - R3;
+        M0 = R3;
+        M1 = R0;
+
+        LSETUP (s$8, e$8) LC0=P2;
+        DISALGNEXCPT         || R0 = [I0++]   || R2 = [I1++];
+        DISALGNEXCPT         || R1 = [I0++]   || R3 = [I1++];
+s$8:    SAA (R1:0,R3:2)      || R0 = [I0++M0] || R2 = [I1++M1];
+        SAA (R1:0,R3:2) (R)  || R0 = [I0++]   || R2 = [I1++];
+e$8:    DISALGNEXCPT         || R1 = [I0++]   || R3 = [I1++];
+
+        R3=A1.L+A1.H,  R2=A0.L+A0.H ;
+        R0 = R2 + R3 ;
+        RTS;
+DEFUN_END(z_sad8x8)
+
+DEFUN(pix_norm1,mL1,
+        (uint8_t * pix, int line_size)):
+        [--SP]=(R7:4,P5:3);
+
+        // Fetch the input arguments.
+        P1 = R0;  // pix
+        P0 = R1;  // line_size
+        P5 = 16;  // loop ctr.
+        P0 -= P5;
+        M0 = P0;  // M0 = line_size-16;
+        // Now for the real work.
+        A1 = A0 = 0;
+        lsetup(_pix_norm1_blkfn_loopStart, _pix_norm1_blkfn_loopEnd) LC1 = P5;
+        I0 = P1;
+        DISALGNEXCPT || r0 = [i0++];
+
+_pix_norm1_blkfn_loopStart:
+        // following unpacks pix1[0..15] pix1+line_size[0..15]
+        DISALGNEXCPT || r1 = [i0++];
+
+        (r5, r4) = byteunpack r1:0 || r0 = [i0++];
+        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
+        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
+        (r5, r4) = byteunpack r1:0(r) || r1 = [i0++];
+        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
+        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
+        (r5, r4) = byteunpack r1:0 || r0 = [i0++M0];
+        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
+        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
+        (r5, r4) = byteunpack r1:0(r) || r0 = [i0++];
+        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
+_pix_norm1_blkfn_loopEnd:
+        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
+
+
+// Clean up at the end:
+        R2 = A0, R3 = A1;
+        R0 = R2 + R3 (S);
+
+        (R7:4,P5:3)=[SP++];
+
+        RTS;
+DEFUN_END(pix_norm1)
+
+DEFUN(sse4,mL1,
+        (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)):
+        link 0;
+        [--sp] = (r7:6);
+        p0=[fp+24];   // h
+        i0=r1;        // pix1
+        i1=r2;        // pix2
+        r2=[fp+20];   // line_size
+        r2+=-4;
+        m0=r2;
+
+        a0=a1=0;
+        LSETUP(.S40,.E40) LC0=P0;
+        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
+
+.S40:   DISALGNEXCPT                       || R1 = [I0++M0] || R3 = [I1++M0];
+        (R7,R6) = BYTEOP16M (R1:0,R3:2);
+        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
+.E40:   a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
+        a0 += a1;
+        r0 = a0;
+
+        (r7:6) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(sse4)
+
+DEFUN(sse8,mL1,
+        (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)):
+        link 0;
+        [--sp] = (r7:6);
+        p0=[fp+24];   // h
+        i0=r1;        // pix1
+        i1=r2;        // pix2
+        r2=[fp+20];   // line_size
+        r2+=-8;
+        m0=r2;
+
+        a0=a1=0;
+        LSETUP(.S80,.E80) LC0=P0;
+        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
+
+.S80:   DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
+        (R7,R6) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++M0] || R2 = [I1++M0];
+        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
+        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
+        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R0 = [I0++]   || R2 = [I1++];
+        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
+.E80:   a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
+        a0 += a1;
+        r0 = a0;
+
+        (r7:6) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(sse8)
+
+DEFUN(sse16,mL1,
+        (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)):
+        link 0;
+        [--sp] = (r7:6);
+        p0=[fp+24];   // h
+        i0=r1;        // pix1
+        i1=r2;        // pix2
+        r2=[fp+20];   // line_size
+        r2+=-16;
+        m0=r2;
+
+        a0=a1=0;
+        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
+        LSETUP(.S160,.E160) LC0=P0;
+
+.S160:  DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
+        (R7,R6) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++]   || R2 = [I1++];
+        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
+        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
+        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R1 = [I0++]   || R3 = [I1++];
+        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
+        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
+        (R7,R6) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++M0] || R2 = [I1++M0];
+        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
+        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
+        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R0 = [I0++]   || R2 = [I1++];
+        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
+.E160:  a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
+        a0 += a1;
+        r0 = a0;
+
+        (r7:6) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(sse16)
diff --git a/libavcodec/bfin/dsputil_bfin.c b/libavcodec/bfin/dsputil_bfin.c
deleted file mode 100644 (file)
index 7a214a3..0000000
+++ /dev/null
@@ -1,196 +0,0 @@
-/*
- * BlackFin DSPUTILS
- *
- * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
- * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include <stdint.h>
-
-#include "config.h"
-#include "libavutil/attributes.h"
-#include "libavutil/bfin/attributes.h"
-#include "libavcodec/avcodec.h"
-#include "libavcodec/dsputil.h"
-
-void ff_bfin_idct(int16_t *block) attribute_l1_text;
-void ff_bfin_fdct(int16_t *block) attribute_l1_text;
-
-void ff_bfin_add_pixels_clamped(const int16_t *block, uint8_t *dest,
-                                int line_size) attribute_l1_text;
-void ff_bfin_put_pixels_clamped(const int16_t *block, uint8_t *dest,
-                                int line_size) attribute_l1_text;
-
-void ff_bfin_diff_pixels(int16_t *block, const uint8_t *s1, const uint8_t *s2,
-                         int stride)  attribute_l1_text;
-void ff_bfin_get_pixels(int16_t *restrict block, const uint8_t *pixels,
-                        int line_size) attribute_l1_text;
-
-int ff_bfin_pix_norm1(uint8_t *pix, int line_size) attribute_l1_text;
-int ff_bfin_pix_sum(uint8_t *p, int stride) attribute_l1_text;
-
-int ff_bfin_z_sad8x8(uint8_t *blk1, uint8_t *blk2, int dsz,
-                     int line_size, int h) attribute_l1_text;
-int ff_bfin_z_sad16x16(uint8_t *blk1, uint8_t *blk2, int dsz,
-                       int line_size, int h) attribute_l1_text;
-
-int ff_bfin_sse4(void *v, uint8_t *pix1, uint8_t *pix2,
-                 int line_size, int h) attribute_l1_text;
-int ff_bfin_sse8(void *v, uint8_t *pix1, uint8_t *pix2,
-                 int line_size, int h) attribute_l1_text;
-int ff_bfin_sse16(void *v, uint8_t *pix1, uint8_t *pix2,
-                  int line_size, int h) attribute_l1_text;
-
-static void bfin_idct_add(uint8_t *dest, int line_size, int16_t *block)
-{
-    ff_bfin_idct(block);
-    ff_bfin_add_pixels_clamped(block, dest, line_size);
-}
-
-static void bfin_idct_put(uint8_t *dest, int line_size, int16_t *block)
-{
-    ff_bfin_idct(block);
-    ff_bfin_put_pixels_clamped(block, dest, line_size);
-}
-
-static void bfin_clear_blocks(int16_t *blocks)
-{
-    // This is just a simple memset.
-    __asm__ ("P0=192; "
-             "I0=%0;  "
-             "R0=0;   "
-             "LSETUP(clear_blocks_blkfn_lab,clear_blocks_blkfn_lab)LC0=P0;"
-             "clear_blocks_blkfn_lab:"
-             "[I0++]=R0;"
-             ::"a" (blocks) : "P0", "I0", "R0");
-}
-
-static int bfin_pix_abs16(void *c, uint8_t *blk1, uint8_t *blk2,
-                          int line_size, int h)
-{
-    return ff_bfin_z_sad16x16(blk1, blk2, line_size, line_size, h);
-}
-
-static uint8_t vtmp_blk[256] attribute_l1_data_b;
-
-static int bfin_pix_abs16_x2(void *c, uint8_t *blk1, uint8_t *blk2,
-                             int line_size, int h)
-{
-    ff_bfin_put_pixels16uc(vtmp_blk, blk2, blk2 + 1, 16, line_size, h);
-    return ff_bfin_z_sad16x16(blk1, vtmp_blk, line_size, 16, h);
-}
-
-static int bfin_pix_abs16_y2(void *c, uint8_t *blk1, uint8_t *blk2,
-                             int line_size, int h)
-{
-    ff_bfin_put_pixels16uc(vtmp_blk, blk2, blk2 + line_size, 16, line_size, h);
-    return ff_bfin_z_sad16x16(blk1, vtmp_blk, line_size, 16, h);
-}
-
-static int bfin_pix_abs16_xy2(void *c, uint8_t *blk1, uint8_t *blk2,
-                              int line_size, int h)
-{
-    ff_bfin_z_put_pixels16_xy2(vtmp_blk, blk2, 16, line_size, h);
-    return ff_bfin_z_sad16x16(blk1, vtmp_blk, line_size, 16, h);
-}
-
-static int bfin_pix_abs8(void *c, uint8_t *blk1, uint8_t *blk2,
-                         int line_size, int h)
-{
-    return ff_bfin_z_sad8x8(blk1, blk2, line_size, line_size, h);
-}
-
-static int bfin_pix_abs8_x2(void *c, uint8_t *blk1, uint8_t *blk2,
-                            int line_size, int h)
-{
-    ff_bfin_put_pixels8uc(vtmp_blk, blk2, blk2 + 1, 8, line_size, h);
-    return ff_bfin_z_sad8x8(blk1, vtmp_blk, line_size, 8, h);
-}
-
-static int bfin_pix_abs8_y2(void *c, uint8_t *blk1, uint8_t *blk2,
-                            int line_size, int h)
-{
-    ff_bfin_put_pixels8uc(vtmp_blk, blk2, blk2 + line_size, 8, line_size, h);
-    return ff_bfin_z_sad8x8(blk1, vtmp_blk, line_size, 8, h);
-}
-
-static int bfin_pix_abs8_xy2(void *c, uint8_t *blk1, uint8_t *blk2,
-                             int line_size, int h)
-{
-    ff_bfin_z_put_pixels8_xy2(vtmp_blk, blk2, 8, line_size, h);
-    return ff_bfin_z_sad8x8(blk1, vtmp_blk, line_size, 8, h);
-}
-
-/*
- * decoder optimization
- * start on 2/11 100 frames of 352x240@25 compiled with no optimization -g debugging
- * 9.824s ~ 2.44x off
- * 6.360s ~ 1.58x off with -O2
- * 5.740s ~ 1.43x off with idcts
- *
- * 2.64s    2/20 same sman.mp4 decode only
- */
-
-av_cold void ff_dsputil_init_bfin(DSPContext *c, AVCodecContext *avctx)
-{
-    const int high_bit_depth = avctx->bits_per_raw_sample > 8;
-
-    c->diff_pixels = ff_bfin_diff_pixels;
-
-    c->put_pixels_clamped = ff_bfin_put_pixels_clamped;
-    c->add_pixels_clamped = ff_bfin_add_pixels_clamped;
-
-    if (!high_bit_depth)
-        c->get_pixels = ff_bfin_get_pixels;
-
-    c->clear_blocks = bfin_clear_blocks;
-
-    c->pix_sum   = ff_bfin_pix_sum;
-    c->pix_norm1 = ff_bfin_pix_norm1;
-
-    c->sad[0] = bfin_pix_abs16;
-    c->sad[1] = bfin_pix_abs8;
-
-    /* TODO [0] 16  [1] 8 */
-    c->pix_abs[0][0] = bfin_pix_abs16;
-    c->pix_abs[0][1] = bfin_pix_abs16_x2;
-    c->pix_abs[0][2] = bfin_pix_abs16_y2;
-    c->pix_abs[0][3] = bfin_pix_abs16_xy2;
-
-    c->pix_abs[1][0] = bfin_pix_abs8;
-    c->pix_abs[1][1] = bfin_pix_abs8_x2;
-    c->pix_abs[1][2] = bfin_pix_abs8_y2;
-    c->pix_abs[1][3] = bfin_pix_abs8_xy2;
-
-    c->sse[0] = ff_bfin_sse16;
-    c->sse[1] = ff_bfin_sse8;
-    c->sse[2] = ff_bfin_sse4;
-
-    if (avctx->bits_per_raw_sample <= 8) {
-        if (avctx->dct_algo == FF_DCT_AUTO)
-            c->fdct = ff_bfin_fdct;
-
-        if (avctx->idct_algo == FF_IDCT_AUTO) {
-            c->idct_permutation_type = FF_NO_IDCT_PERM;
-            c->idct                  = ff_bfin_idct;
-            c->idct_add              = bfin_idct_add;
-            c->idct_put              = bfin_idct_put;
-        }
-    }
-}
diff --git a/libavcodec/bfin/dsputil_init.c b/libavcodec/bfin/dsputil_init.c
new file mode 100644 (file)
index 0000000..7a214a3
--- /dev/null
@@ -0,0 +1,196 @@
+/*
+ * BlackFin DSPUTILS
+ *
+ * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
+ * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "config.h"
+#include "libavutil/attributes.h"
+#include "libavutil/bfin/attributes.h"
+#include "libavcodec/avcodec.h"
+#include "libavcodec/dsputil.h"
+
+void ff_bfin_idct(int16_t *block) attribute_l1_text;
+void ff_bfin_fdct(int16_t *block) attribute_l1_text;
+
+void ff_bfin_add_pixels_clamped(const int16_t *block, uint8_t *dest,
+                                int line_size) attribute_l1_text;
+void ff_bfin_put_pixels_clamped(const int16_t *block, uint8_t *dest,
+                                int line_size) attribute_l1_text;
+
+void ff_bfin_diff_pixels(int16_t *block, const uint8_t *s1, const uint8_t *s2,
+                         int stride)  attribute_l1_text;
+void ff_bfin_get_pixels(int16_t *restrict block, const uint8_t *pixels,
+                        int line_size) attribute_l1_text;
+
+int ff_bfin_pix_norm1(uint8_t *pix, int line_size) attribute_l1_text;
+int ff_bfin_pix_sum(uint8_t *p, int stride) attribute_l1_text;
+
+int ff_bfin_z_sad8x8(uint8_t *blk1, uint8_t *blk2, int dsz,
+                     int line_size, int h) attribute_l1_text;
+int ff_bfin_z_sad16x16(uint8_t *blk1, uint8_t *blk2, int dsz,
+                       int line_size, int h) attribute_l1_text;
+
+int ff_bfin_sse4(void *v, uint8_t *pix1, uint8_t *pix2,
+                 int line_size, int h) attribute_l1_text;
+int ff_bfin_sse8(void *v, uint8_t *pix1, uint8_t *pix2,
+                 int line_size, int h) attribute_l1_text;
+int ff_bfin_sse16(void *v, uint8_t *pix1, uint8_t *pix2,
+                  int line_size, int h) attribute_l1_text;
+
+static void bfin_idct_add(uint8_t *dest, int line_size, int16_t *block)
+{
+    ff_bfin_idct(block);
+    ff_bfin_add_pixels_clamped(block, dest, line_size);
+}
+
+static void bfin_idct_put(uint8_t *dest, int line_size, int16_t *block)
+{
+    ff_bfin_idct(block);
+    ff_bfin_put_pixels_clamped(block, dest, line_size);
+}
+
+static void bfin_clear_blocks(int16_t *blocks)
+{
+    // This is just a simple memset.
+    __asm__ ("P0=192; "
+             "I0=%0;  "
+             "R0=0;   "
+             "LSETUP(clear_blocks_blkfn_lab,clear_blocks_blkfn_lab)LC0=P0;"
+             "clear_blocks_blkfn_lab:"
+             "[I0++]=R0;"
+             ::"a" (blocks) : "P0", "I0", "R0");
+}
+
+static int bfin_pix_abs16(void *c, uint8_t *blk1, uint8_t *blk2,
+                          int line_size, int h)
+{
+    return ff_bfin_z_sad16x16(blk1, blk2, line_size, line_size, h);
+}
+
+static uint8_t vtmp_blk[256] attribute_l1_data_b;
+
+static int bfin_pix_abs16_x2(void *c, uint8_t *blk1, uint8_t *blk2,
+                             int line_size, int h)
+{
+    ff_bfin_put_pixels16uc(vtmp_blk, blk2, blk2 + 1, 16, line_size, h);
+    return ff_bfin_z_sad16x16(blk1, vtmp_blk, line_size, 16, h);
+}
+
+static int bfin_pix_abs16_y2(void *c, uint8_t *blk1, uint8_t *blk2,
+                             int line_size, int h)
+{
+    ff_bfin_put_pixels16uc(vtmp_blk, blk2, blk2 + line_size, 16, line_size, h);
+    return ff_bfin_z_sad16x16(blk1, vtmp_blk, line_size, 16, h);
+}
+
+static int bfin_pix_abs16_xy2(void *c, uint8_t *blk1, uint8_t *blk2,
+                              int line_size, int h)
+{
+    ff_bfin_z_put_pixels16_xy2(vtmp_blk, blk2, 16, line_size, h);
+    return ff_bfin_z_sad16x16(blk1, vtmp_blk, line_size, 16, h);
+}
+
+static int bfin_pix_abs8(void *c, uint8_t *blk1, uint8_t *blk2,
+                         int line_size, int h)
+{
+    return ff_bfin_z_sad8x8(blk1, blk2, line_size, line_size, h);
+}
+
+static int bfin_pix_abs8_x2(void *c, uint8_t *blk1, uint8_t *blk2,
+                            int line_size, int h)
+{
+    ff_bfin_put_pixels8uc(vtmp_blk, blk2, blk2 + 1, 8, line_size, h);
+    return ff_bfin_z_sad8x8(blk1, vtmp_blk, line_size, 8, h);
+}
+
+static int bfin_pix_abs8_y2(void *c, uint8_t *blk1, uint8_t *blk2,
+                            int line_size, int h)
+{
+    ff_bfin_put_pixels8uc(vtmp_blk, blk2, blk2 + line_size, 8, line_size, h);
+    return ff_bfin_z_sad8x8(blk1, vtmp_blk, line_size, 8, h);
+}
+
+static int bfin_pix_abs8_xy2(void *c, uint8_t *blk1, uint8_t *blk2,
+                             int line_size, int h)
+{
+    ff_bfin_z_put_pixels8_xy2(vtmp_blk, blk2, 8, line_size, h);
+    return ff_bfin_z_sad8x8(blk1, vtmp_blk, line_size, 8, h);
+}
+
+/*
+ * decoder optimization
+ * start on 2/11 100 frames of 352x240@25 compiled with no optimization -g debugging
+ * 9.824s ~ 2.44x off
+ * 6.360s ~ 1.58x off with -O2
+ * 5.740s ~ 1.43x off with idcts
+ *
+ * 2.64s    2/20 same sman.mp4 decode only
+ */
+
+av_cold void ff_dsputil_init_bfin(DSPContext *c, AVCodecContext *avctx)
+{
+    const int high_bit_depth = avctx->bits_per_raw_sample > 8;
+
+    c->diff_pixels = ff_bfin_diff_pixels;
+
+    c->put_pixels_clamped = ff_bfin_put_pixels_clamped;
+    c->add_pixels_clamped = ff_bfin_add_pixels_clamped;
+
+    if (!high_bit_depth)
+        c->get_pixels = ff_bfin_get_pixels;
+
+    c->clear_blocks = bfin_clear_blocks;
+
+    c->pix_sum   = ff_bfin_pix_sum;
+    c->pix_norm1 = ff_bfin_pix_norm1;
+
+    c->sad[0] = bfin_pix_abs16;
+    c->sad[1] = bfin_pix_abs8;
+
+    /* TODO [0] 16  [1] 8 */
+    c->pix_abs[0][0] = bfin_pix_abs16;
+    c->pix_abs[0][1] = bfin_pix_abs16_x2;
+    c->pix_abs[0][2] = bfin_pix_abs16_y2;
+    c->pix_abs[0][3] = bfin_pix_abs16_xy2;
+
+    c->pix_abs[1][0] = bfin_pix_abs8;
+    c->pix_abs[1][1] = bfin_pix_abs8_x2;
+    c->pix_abs[1][2] = bfin_pix_abs8_y2;
+    c->pix_abs[1][3] = bfin_pix_abs8_xy2;
+
+    c->sse[0] = ff_bfin_sse16;
+    c->sse[1] = ff_bfin_sse8;
+    c->sse[2] = ff_bfin_sse4;
+
+    if (avctx->bits_per_raw_sample <= 8) {
+        if (avctx->dct_algo == FF_DCT_AUTO)
+            c->fdct = ff_bfin_fdct;
+
+        if (avctx->idct_algo == FF_IDCT_AUTO) {
+            c->idct_permutation_type = FF_NO_IDCT_PERM;
+            c->idct                  = ff_bfin_idct;
+            c->idct_add              = bfin_idct_add;
+            c->idct_put              = bfin_idct_put;
+        }
+    }
+}
diff --git a/libavcodec/bfin/hpel_pixels_bfin.S b/libavcodec/bfin/hpel_pixels_bfin.S
deleted file mode 100644 (file)
index 289aa2c..0000000
+++ /dev/null
@@ -1,207 +0,0 @@
-/*
- * Blackfin Pixel Operations
- * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include "libavutil/bfin/asm.h"
-
-/*
-  motion compensation
-  primitives
-
-     * Halfpel motion compensation with rounding (a+b+1)>>1.
-     * This is an array[4][4] of motion compensation funcions for 4
-     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
-     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
-     * @param block destination where the result is stored
-     * @param pixels source
-     * @param line_size number of bytes in a horizontal line of block
-     * @param h height
-
-*/
-
-DEFUN(put_pixels8uc,mL1,
-        (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
-                 int dest_size, int line_size, int h)):
-        i3=r0;        // dest
-        i0=r1;        // src0
-        i1=r2;        // src1
-        r0=[sp+12];   // dest_size
-        r2=[sp+16];   // line_size
-        p0=[sp+20];   // h
-        [--sp] = (r7:6);
-        r0+=-4;
-        m3=r0;
-        r2+=-8;
-        m0=r2;
-        LSETUP(pp8$0,pp8$1) LC0=P0;
-        DISALGNEXCPT                || R0 = [I0++]  || R2  =[I1++];
-
-pp8$0:  DISALGNEXCPT                || R1 = [I0++]  || R3  =[I1++];
-        R6 = BYTEOP1P(R1:0,R3:2)    || R0 = [I0++M0]|| R2  =[I1++M0];
-        R7 = BYTEOP1P(R1:0,R3:2)(R) || R0 = [I0++]  || [I3++] = R6 ;
-pp8$1:  DISALGNEXCPT                || R2 = [I1++]  || [I3++M3] = R7;
-
-        (r7:6) = [sp++];
-        RTS;
-DEFUN_END(put_pixels8uc)
-
-DEFUN(put_pixels16uc,mL1,
-        (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
-                 int dest_size, int line_size, int h)):
-        link 0;
-        [--sp] = (r7:6);
-        i3=r0;        // dest
-        i0=r1;        // src0
-        i1=r2;        // src1
-        r0=[fp+20];   // dest_size
-        r2=[fp+24];   // line_size
-        p0=[fp+28];   // h
-
-
-        r0+=-12;
-        m3=r0;        // line_size
-        r2+=-16;
-        m0=r2;
-
-        LSETUP(pp16$0,pp16$1) LC0=P0;
-         DISALGNEXCPT                || R0 = [I0++]   || R2  =[I1++];
-
-pp16$0:  DISALGNEXCPT                || R1 = [I0++]   || R3  =[I1++];
-         R6 = BYTEOP1P(R1:0,R3:2)    || R0 = [I0++]   || R2  =[I1++];
-         R7 = BYTEOP1P(R1:0,R3:2)(R) || R1 = [I0++]   || R3  =[I1++];
-         [I3++] = R6;
-         R6 = BYTEOP1P(R1:0,R3:2)    || R0 = [I0++M0] || R2  =[I1++M0];
-         R7 = BYTEOP1P(R1:0,R3:2)(R) || R0 = [I0++]   || [I3++] = R7 ;
-         [I3++] = R6;
-pp16$1:  DISALGNEXCPT                || R2 = [I1++]   || [I3++M3] = R7;
-
-        (r7:6) = [sp++];
-        unlink;
-        RTS;
-DEFUN_END(put_pixels16uc)
-
-DEFUN(z_put_pixels16_xy2,mL1,
-        (uint8_t *block, const uint8_t *s0,
-                 int dest_size, int line_size, int h)):
-        link 0;
-        [--sp] = (r7:4);
-        i3=r0;        // dest
-        i0=r1;        // src0--> pixels
-        i1=r1;        // src1--> pixels + line_size
-        r2+=-12;
-        m2=r2;        // m2=dest_width-4
-        r2=[fp+20];
-        m3=r2;        // line_size
-        p0=[fp+24];   // h
-        r2+=-16;
-        i1+=m3;       /* src1 + line_size */
-        m0=r2;        /* line-size - 20 */
-
-        B0 = I0;
-        B1 = I1;
-        B3 = I3;
-
-        DISALGNEXCPT                       || R0 = [I0++] || R2  =[I1++];
-
-        LSETUP(LS$16E,LE$16E) LC0=P0;
-LS$16E: DISALGNEXCPT                       || R1 = [I0++] || R3  =[I1++];
-        R4 = BYTEOP2P (R3:2,R1:0) (RNDL)   || R0 = [I0++] || R2  =[I1++];
-        R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R1 = [I0++] || [I3++] = R4 ;
-        DISALGNEXCPT                       || R3 = [I1++] || [I3++] = R5;
-        R4 = BYTEOP2P (R3:2,R1:0) (RNDL)   || R0 = [I0++M0]|| R2  = [I1++M0];
-        R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R0 = [I0++] || [I3++] = R4 ;
-LE$16E: DISALGNEXCPT                       || R2 = [I1++] || [I3++M2] = R5;
-
-        M1 = 1;
-        I3 = B3;
-        I1 = B1;
-        I0 = B0;
-
-        I0 += M1;
-        I1 += M1;
-
-        DISALGNEXCPT                       || R0 = [I0++] || R2  =[I1++];
-        LSETUP(LS$16O,LE$16O) LC0=P0;
-LS$16O: DISALGNEXCPT                       || R1 = [I0++] || R3  =[I1++];
-        R4 = BYTEOP2P (R3:2,R1:0) (RNDH)   || R0 = [I0++] || R2  =[I1++];
-        R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R1 = [I0++] || R6  =[I3++];
-        R4 = R4 +|+ R6                       || R7 = [I3--];
-        R5 = R5 +|+ R7                       || [I3++] = R4;
-        DISALGNEXCPT                       || R3  =[I1++] || [I3++] = R5;
-        R4 = BYTEOP2P (R3:2,R1:0) (RNDH)   || R0 = [I0++M0]|| R2  = [I1++M0];
-        R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R0 = [I0++] || R6 = [I3++];
-        R4 = R4 +|+ R6                       || R7 = [I3--];
-        R5 = R5 +|+ R7                       || [I3++] = R4;
-LE$16O: DISALGNEXCPT                       || R2 = [I1++] || [I3++M2] = R5;
-
-        (r7:4) = [sp++];
-        unlink;
-        rts;
-DEFUN_END(z_put_pixels16_xy2)
-
-DEFUN(z_put_pixels8_xy2,mL1,
-        (uint8_t *block, const uint8_t *s0,
-                 int dest_size, int line_size, int h)):
-        link 0;
-        [--sp] = (r7:4);
-        i3=r0;        // dest
-        i0=r1;        // src0--> pixels
-        i1=r1;        // src1--> pixels + line_size
-        r2+=-4;
-        m2=r2;        // m2=dest_width-4
-        r2=[fp+20];
-        m3=r2;        // line_size
-        p0=[fp+24];   // h
-        r2+=-8;
-        i1+=m3;       /* src1 + line_size */
-        m0=r2;        /* line-size - 20 */
-
-        b0 = I0;
-        b1 = I1;
-        b3 = I3;
-
-        LSETUP(LS$8E,LE$8E) LC0=P0;
-        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
-LS$8E:  DISALGNEXCPT                       || R1 = [I0++]   || R3  =[I1++];
-        R4 = BYTEOP2P (R3:2,R1:0) (RNDL)   || R0 = [I0++M0] || R2  =[I1++M0];
-        R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R0 = [I0++]   || [I3++] = R4 ;
-LE$8E:  DISALGNEXCPT                       || R2 = [I1++]   || [I3++M2] = R5;
-
-        M1 = 1;
-        I3 = b3;
-        I1 = b1;
-        I0 = b0;
-
-        I0 += M1;
-        I1 += M1;
-
-        LSETUP(LS$8O,LE$8O) LC0=P0;
-        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
-LS$8O:  DISALGNEXCPT                       || R1 = [I0++]   || R3  =[I1++];
-        R4 = BYTEOP2P (R3:2,R1:0) (RNDH)   || R0 = [I0++M0] || R2  =[I1++M0];
-        R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R0 = [I0++]   || R6  =[I3++];
-        R4 = R4 +|+ R6                                      || R7 = [I3--];
-        R5 = R5 +|+ R7                                      || [I3++] = R4;
-LE$8O:  DISALGNEXCPT                       || R2  =[I1++]   || [I3++M2] = R5;
-
-        (r7:4) = [sp++];
-        unlink;
-        rts;
-DEFUN_END(z_put_pixels8_xy2)
diff --git a/libavcodec/bfin/hpeldsp_bfin.c b/libavcodec/bfin/hpeldsp_bfin.c
deleted file mode 100644 (file)
index 263f5be..0000000
+++ /dev/null
@@ -1,146 +0,0 @@
-/*
- * BlackFin halfpel functions
- *
- * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
- * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include <stddef.h>
-#include <stdint.h>
-
-#include "libavutil/attributes.h"
-#include "libavutil/bfin/attributes.h"
-#include "libavcodec/hpeldsp.h"
-#include "hpeldsp_bfin.h"
-
-void ff_bfin_put_pixels8uc_no_rnd(uint8_t *block, const uint8_t *s0,
-                                  const uint8_t *s1, int line_size,
-                                  int h) attribute_l1_text;
-void ff_bfin_put_pixels16uc_no_rnd(uint8_t *block, const uint8_t *s0,
-                                   const uint8_t *s1, int line_size,
-                                   int h) attribute_l1_text;
-
-static void bfin_put_pixels8(uint8_t *block, const uint8_t *pixels,
-                             ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels8uc(block, pixels, pixels, line_size, line_size, h);
-}
-
-static void bfin_put_pixels8_x2(uint8_t *block, const uint8_t *pixels,
-                                ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels8uc(block, pixels, pixels + 1, line_size, line_size, h);
-}
-
-static void bfin_put_pixels8_y2(uint8_t *block, const uint8_t *pixels,
-                                ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels8uc(block, pixels, pixels + line_size,
-                          line_size, line_size, h);
-}
-
-static void bfin_put_pixels8_xy2(uint8_t *block, const uint8_t *s0,
-                                 ptrdiff_t line_size, int h)
-{
-    ff_bfin_z_put_pixels8_xy2(block, s0, line_size, line_size, h);
-}
-
-static void bfin_put_pixels16(uint8_t *block, const uint8_t *pixels,
-                              ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels16uc(block, pixels, pixels, line_size, line_size, h);
-}
-
-static void bfin_put_pixels16_x2(uint8_t *block, const uint8_t *pixels,
-                                 ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels16uc(block, pixels, pixels + 1, line_size, line_size, h);
-}
-
-static void bfin_put_pixels16_y2(uint8_t *block, const uint8_t *pixels,
-                                 ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels16uc(block, pixels, pixels + line_size,
-                           line_size, line_size, h);
-}
-
-static void bfin_put_pixels16_xy2(uint8_t *block, const uint8_t *s0,
-                                  ptrdiff_t line_size, int h)
-{
-    ff_bfin_z_put_pixels16_xy2(block, s0, line_size, line_size, h);
-}
-
-static void bfin_put_pixels8_no_rnd(uint8_t *block, const uint8_t *pixels,
-                                    ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels8uc_no_rnd(block, pixels, pixels, line_size, h);
-}
-
-static void bfin_put_pixels8_x2_no_rnd(uint8_t *block, const uint8_t *pixels,
-                                       ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels8uc_no_rnd(block, pixels, pixels + 1, line_size, h);
-}
-
-static void bfin_put_pixels8_y2_no_rnd(uint8_t *block, const uint8_t *pixels,
-                                       ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels8uc_no_rnd(block, pixels, pixels + line_size,
-                                 line_size, h);
-}
-
-static void bfin_put_pixels16_no_rnd(uint8_t *block, const uint8_t *pixels,
-                                     ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels16uc_no_rnd(block, pixels, pixels, line_size, h);
-}
-
-static void bfin_put_pixels16_x2_no_rnd(uint8_t *block, const uint8_t *pixels,
-                                        ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels16uc_no_rnd(block, pixels, pixels + 1, line_size, h);
-}
-
-static void bfin_put_pixels16_y2_no_rnd(uint8_t *block, const uint8_t *pixels,
-                                        ptrdiff_t line_size, int h)
-{
-    ff_bfin_put_pixels16uc_no_rnd(block, pixels, pixels + line_size,
-                                  line_size, h);
-}
-
-av_cold void ff_hpeldsp_init_bfin(HpelDSPContext *c, int flags)
-{
-    c->put_pixels_tab[0][0] = bfin_put_pixels16;
-    c->put_pixels_tab[0][1] = bfin_put_pixels16_x2;
-    c->put_pixels_tab[0][2] = bfin_put_pixels16_y2;
-    c->put_pixels_tab[0][3] = bfin_put_pixels16_xy2;
-
-    c->put_pixels_tab[1][0] = bfin_put_pixels8;
-    c->put_pixels_tab[1][1] = bfin_put_pixels8_x2;
-    c->put_pixels_tab[1][2] = bfin_put_pixels8_y2;
-    c->put_pixels_tab[1][3] = bfin_put_pixels8_xy2;
-
-    c->put_no_rnd_pixels_tab[1][0] = bfin_put_pixels8_no_rnd;
-    c->put_no_rnd_pixels_tab[1][1] = bfin_put_pixels8_x2_no_rnd;
-    c->put_no_rnd_pixels_tab[1][2] = bfin_put_pixels8_y2_no_rnd;
-
-    c->put_no_rnd_pixels_tab[0][0] = bfin_put_pixels16_no_rnd;
-    c->put_no_rnd_pixels_tab[0][1] = bfin_put_pixels16_x2_no_rnd;
-    c->put_no_rnd_pixels_tab[0][2] = bfin_put_pixels16_y2_no_rnd;
-}
diff --git a/libavcodec/bfin/hpeldsp_bfin.h b/libavcodec/bfin/hpeldsp_bfin.h
deleted file mode 100644 (file)
index b02561c..0000000
+++ /dev/null
@@ -1,40 +0,0 @@
-/*
- * Copyright (C) 2007 Marc Hoffman <mmh@pleasantst.com>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#ifndef AVCODEC_BFIN_HPELDSP_BFIN_H
-#define AVCODEC_BFIN_HPELDSP_BFIN_H
-
-#include <stdint.h>
-
-#include "libavutil/bfin/attributes.h"
-
-void ff_bfin_z_put_pixels16_xy2(uint8_t *block, const uint8_t *s0,
-                                int dest_size, int line_size, int h) attribute_l1_text;
-void ff_bfin_z_put_pixels8_xy2(uint8_t *block, const uint8_t *s0,
-                               int dest_size, int line_size, int h) attribute_l1_text;
-
-void ff_bfin_put_pixels8uc(uint8_t *block, const uint8_t *s0,
-                           const uint8_t *s1, int dest_size, int line_size,
-                           int h) attribute_l1_text;
-void ff_bfin_put_pixels16uc(uint8_t *block, const uint8_t *s0,
-                            const uint8_t *s1, int dest_size, int line_size,
-                            int h) attribute_l1_text;
-
-#endif /* AVCODEC_BFIN_HPELDSP_BFIN_H */
diff --git a/libavcodec/bfin/hpeldsp_init.c b/libavcodec/bfin/hpeldsp_init.c
new file mode 100644 (file)
index 0000000..9eb45e0
--- /dev/null
@@ -0,0 +1,146 @@
+/*
+ * BlackFin halfpel functions
+ *
+ * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
+ * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include "libavutil/attributes.h"
+#include "libavutil/bfin/attributes.h"
+#include "libavcodec/hpeldsp.h"
+#include "pixels.h"
+
+void ff_bfin_put_pixels8uc_no_rnd(uint8_t *block, const uint8_t *s0,
+                                  const uint8_t *s1, int line_size,
+                                  int h) attribute_l1_text;
+void ff_bfin_put_pixels16uc_no_rnd(uint8_t *block, const uint8_t *s0,
+                                   const uint8_t *s1, int line_size,
+                                   int h) attribute_l1_text;
+
+static void bfin_put_pixels8(uint8_t *block, const uint8_t *pixels,
+                             ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels8uc(block, pixels, pixels, line_size, line_size, h);
+}
+
+static void bfin_put_pixels8_x2(uint8_t *block, const uint8_t *pixels,
+                                ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels8uc(block, pixels, pixels + 1, line_size, line_size, h);
+}
+
+static void bfin_put_pixels8_y2(uint8_t *block, const uint8_t *pixels,
+                                ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels8uc(block, pixels, pixels + line_size,
+                          line_size, line_size, h);
+}
+
+static void bfin_put_pixels8_xy2(uint8_t *block, const uint8_t *s0,
+                                 ptrdiff_t line_size, int h)
+{
+    ff_bfin_z_put_pixels8_xy2(block, s0, line_size, line_size, h);
+}
+
+static void bfin_put_pixels16(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels16uc(block, pixels, pixels, line_size, line_size, h);
+}
+
+static void bfin_put_pixels16_x2(uint8_t *block, const uint8_t *pixels,
+                                 ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels16uc(block, pixels, pixels + 1, line_size, line_size, h);
+}
+
+static void bfin_put_pixels16_y2(uint8_t *block, const uint8_t *pixels,
+                                 ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels16uc(block, pixels, pixels + line_size,
+                           line_size, line_size, h);
+}
+
+static void bfin_put_pixels16_xy2(uint8_t *block, const uint8_t *s0,
+                                  ptrdiff_t line_size, int h)
+{
+    ff_bfin_z_put_pixels16_xy2(block, s0, line_size, line_size, h);
+}
+
+static void bfin_put_pixels8_no_rnd(uint8_t *block, const uint8_t *pixels,
+                                    ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels8uc_no_rnd(block, pixels, pixels, line_size, h);
+}
+
+static void bfin_put_pixels8_x2_no_rnd(uint8_t *block, const uint8_t *pixels,
+                                       ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels8uc_no_rnd(block, pixels, pixels + 1, line_size, h);
+}
+
+static void bfin_put_pixels8_y2_no_rnd(uint8_t *block, const uint8_t *pixels,
+                                       ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels8uc_no_rnd(block, pixels, pixels + line_size,
+                                 line_size, h);
+}
+
+static void bfin_put_pixels16_no_rnd(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels16uc_no_rnd(block, pixels, pixels, line_size, h);
+}
+
+static void bfin_put_pixels16_x2_no_rnd(uint8_t *block, const uint8_t *pixels,
+                                        ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels16uc_no_rnd(block, pixels, pixels + 1, line_size, h);
+}
+
+static void bfin_put_pixels16_y2_no_rnd(uint8_t *block, const uint8_t *pixels,
+                                        ptrdiff_t line_size, int h)
+{
+    ff_bfin_put_pixels16uc_no_rnd(block, pixels, pixels + line_size,
+                                  line_size, h);
+}
+
+av_cold void ff_hpeldsp_init_bfin(HpelDSPContext *c, int flags)
+{
+    c->put_pixels_tab[0][0] = bfin_put_pixels16;
+    c->put_pixels_tab[0][1] = bfin_put_pixels16_x2;
+    c->put_pixels_tab[0][2] = bfin_put_pixels16_y2;
+    c->put_pixels_tab[0][3] = bfin_put_pixels16_xy2;
+
+    c->put_pixels_tab[1][0] = bfin_put_pixels8;
+    c->put_pixels_tab[1][1] = bfin_put_pixels8_x2;
+    c->put_pixels_tab[1][2] = bfin_put_pixels8_y2;
+    c->put_pixels_tab[1][3] = bfin_put_pixels8_xy2;
+
+    c->put_no_rnd_pixels_tab[1][0] = bfin_put_pixels8_no_rnd;
+    c->put_no_rnd_pixels_tab[1][1] = bfin_put_pixels8_x2_no_rnd;
+    c->put_no_rnd_pixels_tab[1][2] = bfin_put_pixels8_y2_no_rnd;
+
+    c->put_no_rnd_pixels_tab[0][0] = bfin_put_pixels16_no_rnd;
+    c->put_no_rnd_pixels_tab[0][1] = bfin_put_pixels16_x2_no_rnd;
+    c->put_no_rnd_pixels_tab[0][2] = bfin_put_pixels16_y2_no_rnd;
+}
diff --git a/libavcodec/bfin/pixels.S b/libavcodec/bfin/pixels.S
new file mode 100644 (file)
index 0000000..289aa2c
--- /dev/null
@@ -0,0 +1,207 @@
+/*
+ * Blackfin Pixel Operations
+ * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/bfin/asm.h"
+
+/*
+  motion compensation
+  primitives
+
+     * Halfpel motion compensation with rounding (a+b+1)>>1.
+     * This is an array[4][4] of motion compensation funcions for 4
+     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
+     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
+     * @param block destination where the result is stored
+     * @param pixels source
+     * @param line_size number of bytes in a horizontal line of block
+     * @param h height
+
+*/
+
+DEFUN(put_pixels8uc,mL1,
+        (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
+                 int dest_size, int line_size, int h)):
+        i3=r0;        // dest
+        i0=r1;        // src0
+        i1=r2;        // src1
+        r0=[sp+12];   // dest_size
+        r2=[sp+16];   // line_size
+        p0=[sp+20];   // h
+        [--sp] = (r7:6);
+        r0+=-4;
+        m3=r0;
+        r2+=-8;
+        m0=r2;
+        LSETUP(pp8$0,pp8$1) LC0=P0;
+        DISALGNEXCPT                || R0 = [I0++]  || R2  =[I1++];
+
+pp8$0:  DISALGNEXCPT                || R1 = [I0++]  || R3  =[I1++];
+        R6 = BYTEOP1P(R1:0,R3:2)    || R0 = [I0++M0]|| R2  =[I1++M0];
+        R7 = BYTEOP1P(R1:0,R3:2)(R) || R0 = [I0++]  || [I3++] = R6 ;
+pp8$1:  DISALGNEXCPT                || R2 = [I1++]  || [I3++M3] = R7;
+
+        (r7:6) = [sp++];
+        RTS;
+DEFUN_END(put_pixels8uc)
+
+DEFUN(put_pixels16uc,mL1,
+        (uint8_t *block, const uint8_t *s0, const uint8_t *s1,
+                 int dest_size, int line_size, int h)):
+        link 0;
+        [--sp] = (r7:6);
+        i3=r0;        // dest
+        i0=r1;        // src0
+        i1=r2;        // src1
+        r0=[fp+20];   // dest_size
+        r2=[fp+24];   // line_size
+        p0=[fp+28];   // h
+
+
+        r0+=-12;
+        m3=r0;        // line_size
+        r2+=-16;
+        m0=r2;
+
+        LSETUP(pp16$0,pp16$1) LC0=P0;
+         DISALGNEXCPT                || R0 = [I0++]   || R2  =[I1++];
+
+pp16$0:  DISALGNEXCPT                || R1 = [I0++]   || R3  =[I1++];
+         R6 = BYTEOP1P(R1:0,R3:2)    || R0 = [I0++]   || R2  =[I1++];
+         R7 = BYTEOP1P(R1:0,R3:2)(R) || R1 = [I0++]   || R3  =[I1++];
+         [I3++] = R6;
+         R6 = BYTEOP1P(R1:0,R3:2)    || R0 = [I0++M0] || R2  =[I1++M0];
+         R7 = BYTEOP1P(R1:0,R3:2)(R) || R0 = [I0++]   || [I3++] = R7 ;
+         [I3++] = R6;
+pp16$1:  DISALGNEXCPT                || R2 = [I1++]   || [I3++M3] = R7;
+
+        (r7:6) = [sp++];
+        unlink;
+        RTS;
+DEFUN_END(put_pixels16uc)
+
+DEFUN(z_put_pixels16_xy2,mL1,
+        (uint8_t *block, const uint8_t *s0,
+                 int dest_size, int line_size, int h)):
+        link 0;
+        [--sp] = (r7:4);
+        i3=r0;        // dest
+        i0=r1;        // src0--> pixels
+        i1=r1;        // src1--> pixels + line_size
+        r2+=-12;
+        m2=r2;        // m2=dest_width-4
+        r2=[fp+20];
+        m3=r2;        // line_size
+        p0=[fp+24];   // h
+        r2+=-16;
+        i1+=m3;       /* src1 + line_size */
+        m0=r2;        /* line-size - 20 */
+
+        B0 = I0;
+        B1 = I1;
+        B3 = I3;
+
+        DISALGNEXCPT                       || R0 = [I0++] || R2  =[I1++];
+
+        LSETUP(LS$16E,LE$16E) LC0=P0;
+LS$16E: DISALGNEXCPT                       || R1 = [I0++] || R3  =[I1++];
+        R4 = BYTEOP2P (R3:2,R1:0) (RNDL)   || R0 = [I0++] || R2  =[I1++];
+        R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R1 = [I0++] || [I3++] = R4 ;
+        DISALGNEXCPT                       || R3 = [I1++] || [I3++] = R5;
+        R4 = BYTEOP2P (R3:2,R1:0) (RNDL)   || R0 = [I0++M0]|| R2  = [I1++M0];
+        R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R0 = [I0++] || [I3++] = R4 ;
+LE$16E: DISALGNEXCPT                       || R2 = [I1++] || [I3++M2] = R5;
+
+        M1 = 1;
+        I3 = B3;
+        I1 = B1;
+        I0 = B0;
+
+        I0 += M1;
+        I1 += M1;
+
+        DISALGNEXCPT                       || R0 = [I0++] || R2  =[I1++];
+        LSETUP(LS$16O,LE$16O) LC0=P0;
+LS$16O: DISALGNEXCPT                       || R1 = [I0++] || R3  =[I1++];
+        R4 = BYTEOP2P (R3:2,R1:0) (RNDH)   || R0 = [I0++] || R2  =[I1++];
+        R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R1 = [I0++] || R6  =[I3++];
+        R4 = R4 +|+ R6                       || R7 = [I3--];
+        R5 = R5 +|+ R7                       || [I3++] = R4;
+        DISALGNEXCPT                       || R3  =[I1++] || [I3++] = R5;
+        R4 = BYTEOP2P (R3:2,R1:0) (RNDH)   || R0 = [I0++M0]|| R2  = [I1++M0];
+        R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R0 = [I0++] || R6 = [I3++];
+        R4 = R4 +|+ R6                       || R7 = [I3--];
+        R5 = R5 +|+ R7                       || [I3++] = R4;
+LE$16O: DISALGNEXCPT                       || R2 = [I1++] || [I3++M2] = R5;
+
+        (r7:4) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(z_put_pixels16_xy2)
+
+DEFUN(z_put_pixels8_xy2,mL1,
+        (uint8_t *block, const uint8_t *s0,
+                 int dest_size, int line_size, int h)):
+        link 0;
+        [--sp] = (r7:4);
+        i3=r0;        // dest
+        i0=r1;        // src0--> pixels
+        i1=r1;        // src1--> pixels + line_size
+        r2+=-4;
+        m2=r2;        // m2=dest_width-4
+        r2=[fp+20];
+        m3=r2;        // line_size
+        p0=[fp+24];   // h
+        r2+=-8;
+        i1+=m3;       /* src1 + line_size */
+        m0=r2;        /* line-size - 20 */
+
+        b0 = I0;
+        b1 = I1;
+        b3 = I3;
+
+        LSETUP(LS$8E,LE$8E) LC0=P0;
+        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
+LS$8E:  DISALGNEXCPT                       || R1 = [I0++]   || R3  =[I1++];
+        R4 = BYTEOP2P (R3:2,R1:0) (RNDL)   || R0 = [I0++M0] || R2  =[I1++M0];
+        R5 = BYTEOP2P (R3:2,R1:0) (RNDL,R) || R0 = [I0++]   || [I3++] = R4 ;
+LE$8E:  DISALGNEXCPT                       || R2 = [I1++]   || [I3++M2] = R5;
+
+        M1 = 1;
+        I3 = b3;
+        I1 = b1;
+        I0 = b0;
+
+        I0 += M1;
+        I1 += M1;
+
+        LSETUP(LS$8O,LE$8O) LC0=P0;
+        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
+LS$8O:  DISALGNEXCPT                       || R1 = [I0++]   || R3  =[I1++];
+        R4 = BYTEOP2P (R3:2,R1:0) (RNDH)   || R0 = [I0++M0] || R2  =[I1++M0];
+        R5 = BYTEOP2P (R3:2,R1:0) (RNDH,R) || R0 = [I0++]   || R6  =[I3++];
+        R4 = R4 +|+ R6                                      || R7 = [I3--];
+        R5 = R5 +|+ R7                                      || [I3++] = R4;
+LE$8O:  DISALGNEXCPT                       || R2  =[I1++]   || [I3++M2] = R5;
+
+        (r7:4) = [sp++];
+        unlink;
+        rts;
+DEFUN_END(z_put_pixels8_xy2)
diff --git a/libavcodec/bfin/pixels.h b/libavcodec/bfin/pixels.h
new file mode 100644 (file)
index 0000000..0934429
--- /dev/null
@@ -0,0 +1,40 @@
+/*
+ * Copyright (C) 2007 Marc Hoffman <mmh@pleasantst.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_BFIN_PIXELS_H
+#define AVCODEC_BFIN_PIXELS_H
+
+#include <stdint.h>
+
+#include "libavutil/bfin/attributes.h"
+
+void ff_bfin_z_put_pixels16_xy2(uint8_t *block, const uint8_t *s0,
+                                int dest_size, int line_size, int h) attribute_l1_text;
+void ff_bfin_z_put_pixels8_xy2(uint8_t *block, const uint8_t *s0,
+                               int dest_size, int line_size, int h) attribute_l1_text;
+
+void ff_bfin_put_pixels8uc(uint8_t *block, const uint8_t *s0,
+                           const uint8_t *s1, int dest_size, int line_size,
+                           int h) attribute_l1_text;
+void ff_bfin_put_pixels16uc(uint8_t *block, const uint8_t *s0,
+                            const uint8_t *s1, int dest_size, int line_size,
+                            int h) attribute_l1_text;
+
+#endif /* AVCODEC_BFIN_PIXELS_H */
diff --git a/libavcodec/bfin/pixels_bfin.S b/libavcodec/bfin/pixels_bfin.S
deleted file mode 100644 (file)
index e19455b..0000000
+++ /dev/null
@@ -1,382 +0,0 @@
-/*
- * Blackfin Pixel Operations
- * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include "libavutil/bfin/asm.h"
-
-DEFUN(put_pixels_clamped,mL1,
-        (int16_t *block, uint8_t *dest, int line_size)):
-    [--SP] = (R7:4);
-    R4 = 0;
-    R5.l = 0x00ff;
-    R5.h = 0x00ff;
-    I0 = R0;         // block
-    I1 = R1;         // dest
-    R2 += -4;        // line_size
-    M1 = R2;
-    P0 = 8;
-    R0 = [I0++];
-    R1 = [I0++];
-    R2 = MAX(R0, R4) (V);
-    LSETUP (ppc$0,ppc$1) LC0=P0;
-ppc$0: R2 = MIN(R2, R5) (V);
-       R3 = MAX(R1, R4) (V);
-       R3 = MIN(R3, R5) (V)      || R0 = [I0++];
-       R6 = BYTEPACK (R2,R3)     || R1 = [I0++];
-       R2 = MAX(R0, R4) (V)      || [I1++] = R6;
-       R2 = MIN(R2, R5) (V);
-       R3 = MAX(R1, R4) (V);
-       R3 = MIN(R3, R5) (V)      || R0 = [I0++];
-       R6 = BYTEPACK (R2,R3)     || R1 = [I0++];
-ppc$1: R2 = Max(R0, R4) (V)      || [I1++M1] = R6;
-
-    (R7:4) = [SP++];
-    RTS;
-DEFUN_END(put_pixels_clamped)
-
-DEFUN(add_pixels_clamped,mL1,
-        (int16_t *block, uint8_t *dest, int line_size)):
-    [-- SP] = (R7:4);
-    R4 = 0;
-    I0 = 0;
-    R2 += -4;        // line_size
-    M0 = R2;
-    I1 = R1;         // dest
-    I3 = R0;         // block
-    I2 = R1;         // dest
-    P0 = 8;
-    M3 = 2;
-    R0 = [I3++]  || R2 = [I1];
-    R2 = R2 << 8                      || R0.H = W[I3--]  || R3 = [I1++];
-    R3 = R3 >> 8                      || R1.L = W[I3]    || I3 += 4;
-    R6 = BYTEOP3P(R1:0, R3:2) (LO)    || R1.H = W[I3++]  || R2 = [I1];
-
-    LSETUP(apc$2,apc$3) LC1 = P0;
-apc$2: R7 = BYTEOP3P(R1:0, R3:2) (HI, R) || R0 = [I3++]     || R3 = [I1++M0];
-       R2 = R2 << 8                      || R0.H = W[I3--];
-       R3 = R3 >> 8                      || R1.L = W[I3]    || I3 += 4;
-       R6 = R6 + R7 (S)                  || R1.H = W[I3];
-       R6 = BYTEOP3P(R1:0, R3:2) (LO)    || I3+=M3          || [I2++]=R6;
-       R7 = BYTEOP3P(R1:0, R3:2) (HI, R) || R0 = [I3++]     || R2 = [I1];
-       R2 = R2 << 8                      || R0.H = W[I3--]  || R3 = [I1++];
-       R3 = R3 >> 8                      || R1.L = W[I3]    || I3 += 4;
-       R6 = R6 + R7 (S)                  || R1.H = W[I3++];
-apc$3: R6 = BYTEOP3P(R1:0, R3:2) (LO)    || [I2++M0] = R6   || R2 = [I1];
-
-    (R7:4) = [SP++];
-    RTS;
-DEFUN_END(add_pixels_clamped)
-
-DEFUN(diff_pixels,mL1,
-       (int16_t *block, uint8_t *s1, uint8_t *s2, int stride)):
-        link 0;
-        [--sp] = (r7:4);
-        p0=8;
-        i3=r0;        // block
-        i0=r1;        // s1
-        i1=r2;        // s2
-        r2=[fp+20];   // stride
-        r2+=-8;
-        m0=r2;
-
-
-        LSETUP(.LS0,.LE0) LC0=P0;
-        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
-
-.LS0:   DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
-        (R5,R4) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++M0] || R2 = [I1++M0];
-        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R0 = [I0++]   || [I3++] = R4;
-        DISALGNEXCPT                       || R2 = [I1++]   || [I3++] = R5;
-        [i3++]=r6;
-.LE0:  [i3++]=r7;
-
-        (r7:4) = [sp++];
-        unlink;
-        rts;
-DEFUN_END(diff_pixels)
-
-/*
-    for (i = 0; i < 16; i++) {
-        for (j = 0; j < 16; j++) {
-          sum += pix[j];
-        }
-        pix += line_size;
-    }
-*/
-DEFUN(pix_sum,mL1,
-        (uint8_t *p, int stride)):
-        link 0;
-        [--sp] = (r7:4);
-        p0=8;
-        i0=r0;        // s1
-        i1=r0;
-        m1=r1;
-        r1=r1+r1;
-        r1+=-16;       // stride
-        m0=r1;
-        i1+=m1;
-
-        r6=0;
-
-        LSETUP(LS$PS,LE$PS) LC0=P0;
-        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
-
-LS$PS:  DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
-        (R5,R4) = BYTEOP16P (R3:2,R1:0)    || R0 = [I0++]   || R2 = [I1++];
-        r6=r6+|+r5;
-        r6=r6+|+r4;
-        (R5,R4) = BYTEOP16P (R3:2,R1:0) (R)|| R1 = [I0++]   || R3 = [I1++];
-        r6=r6+|+r5;
-        r6=r6+|+r4;
-        (R5,R4) = BYTEOP16P (R3:2,R1:0)    || R0 = [I0++m0] || R2 = [I1++m0];
-        r6=r6+|+r5;
-        r6=r6+|+r4;
-        (R5,R4) = BYTEOP16P (R3:2,R1:0) (R)|| R0 = [I0++]   || R2 = [I1++];
-        r6=r6+|+r5;
-LE$PS:  r6=r6+|+r4;
-        r0.l=r6.l+r6.h;
-        r0.h=0;
-
-        (r7:4) = [sp++];
-        unlink;
-        rts;
-DEFUN_END(pix_sum)
-
-
-DEFUN(get_pixels,mL1,
-        (int16_t *restrict block, const uint8_t *pixels, int line_size)):
-        [--sp] = (r7:4);
-        i3=r0;        // dest
-        i0=r1;        // src0
-        p0=8;
-        r2+=-8;
-        m0=r2;
-        LSETUP(gp8$0,gp8$1) LC0=P0;
-
-        DISALGNEXCPT                   || R0 = [I0++];
-        DISALGNEXCPT                   || R1 = [I0++];
-
-gp8$0:  (R7,R6) = byteunpack R1:0      || R0 = [I0++M0];
-        (R5,R4) = byteunpack R1:0 (R)  || R0 = [I0++]    || [I3++]=R6;
-        DISALGNEXCPT                   || R1 = [I0++]    || [I3++]=R7;
-        [I3++]=R4;
-gp8$1:  [I3++]=R5
-
-
-        (r7:4) = [sp++];
-        RTS;
-DEFUN_END(get_pixels)
-
-
-/* sad = sad16x16 (ubyte *mb, ubyte *refwin, srcwidth, refwinwidth, h) */
-/* 91 cycles */
-DEFUN(z_sad16x16,mL1,
-        (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h)):
-        link 0;
-        I0 = R0;
-        I1 = R1;
-
-        A1 = A0 = 0;
-        R0 = [sp+20]; // rwidth
-        P2 = [sp+24]; // height
-        R3 = 16;
-        R0 = R0 - R3;
-        R3 = R2 - R3;
-        M1 = R0;
-        M0 = R3;
-
-        DISALGNEXCPT         || R0 = [I0++]    || R2 = [I1++];
-        LSETUP (s$16, e$16) LC0=P2;
-s$16:   DISALGNEXCPT         || R1 = [I0++]    || R3 = [I1++];
-        SAA (R1:0,R3:2)      || R0 = [I0++]    || R2 = [I1++];
-        SAA (R1:0,R3:2) (R)  || R1 = [I0++]    || R3 = [I1++];
-        SAA (R1:0,R3:2)      || R0 = [I0++M0]  || R2 = [I1++M1];
-e$16:   SAA (R1:0,R3:2) (R)  || R0 = [I0++]    || R2 = [I1++];
-
-        R3=A1.L+A1.H,  R2=A0.L+A0.H ;
-        R0 = R2 + R3 ;
-        unlink;
-        RTS;
-DEFUN_END(z_sad16x16)
-
-/* sad = sad8x8 (ubyte *mb, ubyte *refwin, int srcwidth, int refwinwidth, int h) */
-/* 36 cycles */
-DEFUN(z_sad8x8,mL1,
-        (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h)):
-        I0 = R0;
-        I1 = R1;
-
-        A1 = A0 = 0;
-        r0 = [sp+12]; // rwidth
-        P2 = [sp+16]; //height
-        R3 = 8;
-        R0 = R0 - R3;
-        R3 = R2 - R3;
-        M0 = R3;
-        M1 = R0;
-
-        LSETUP (s$8, e$8) LC0=P2;
-        DISALGNEXCPT         || R0 = [I0++]   || R2 = [I1++];
-        DISALGNEXCPT         || R1 = [I0++]   || R3 = [I1++];
-s$8:    SAA (R1:0,R3:2)      || R0 = [I0++M0] || R2 = [I1++M1];
-        SAA (R1:0,R3:2) (R)  || R0 = [I0++]   || R2 = [I1++];
-e$8:    DISALGNEXCPT         || R1 = [I0++]   || R3 = [I1++];
-
-        R3=A1.L+A1.H,  R2=A0.L+A0.H ;
-        R0 = R2 + R3 ;
-        RTS;
-DEFUN_END(z_sad8x8)
-
-DEFUN(pix_norm1,mL1,
-        (uint8_t * pix, int line_size)):
-        [--SP]=(R7:4,P5:3);
-
-        // Fetch the input arguments.
-        P1 = R0;  // pix
-        P0 = R1;  // line_size
-        P5 = 16;  // loop ctr.
-        P0 -= P5;
-        M0 = P0;  // M0 = line_size-16;
-        // Now for the real work.
-        A1 = A0 = 0;
-        lsetup(_pix_norm1_blkfn_loopStart, _pix_norm1_blkfn_loopEnd) LC1 = P5;
-        I0 = P1;
-        DISALGNEXCPT || r0 = [i0++];
-
-_pix_norm1_blkfn_loopStart:
-        // following unpacks pix1[0..15] pix1+line_size[0..15]
-        DISALGNEXCPT || r1 = [i0++];
-
-        (r5, r4) = byteunpack r1:0 || r0 = [i0++];
-        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
-        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
-        (r5, r4) = byteunpack r1:0(r) || r1 = [i0++];
-        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
-        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
-        (r5, r4) = byteunpack r1:0 || r0 = [i0++M0];
-        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
-        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
-        (r5, r4) = byteunpack r1:0(r) || r0 = [i0++];
-        a1 += r5.h * r5.h, a0 += r5.l * r5.l (is);
-_pix_norm1_blkfn_loopEnd:
-        a1 += r4.h * r4.h, a0 += r4.l * r4.l (is);
-
-
-// Clean up at the end:
-        R2 = A0, R3 = A1;
-        R0 = R2 + R3 (S);
-
-        (R7:4,P5:3)=[SP++];
-
-        RTS;
-DEFUN_END(pix_norm1)
-
-DEFUN(sse4,mL1,
-        (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)):
-        link 0;
-        [--sp] = (r7:6);
-        p0=[fp+24];   // h
-        i0=r1;        // pix1
-        i1=r2;        // pix2
-        r2=[fp+20];   // line_size
-        r2+=-4;
-        m0=r2;
-
-        a0=a1=0;
-        LSETUP(.S40,.E40) LC0=P0;
-        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
-
-.S40:   DISALGNEXCPT                       || R1 = [I0++M0] || R3 = [I1++M0];
-        (R7,R6) = BYTEOP16M (R1:0,R3:2);
-        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
-.E40:   a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
-        a0 += a1;
-        r0 = a0;
-
-        (r7:6) = [sp++];
-        unlink;
-        rts;
-DEFUN_END(sse4)
-
-DEFUN(sse8,mL1,
-        (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)):
-        link 0;
-        [--sp] = (r7:6);
-        p0=[fp+24];   // h
-        i0=r1;        // pix1
-        i1=r2;        // pix2
-        r2=[fp+20];   // line_size
-        r2+=-8;
-        m0=r2;
-
-        a0=a1=0;
-        LSETUP(.S80,.E80) LC0=P0;
-        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
-
-.S80:   DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
-        (R7,R6) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++M0] || R2 = [I1++M0];
-        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
-        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
-        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R0 = [I0++]   || R2 = [I1++];
-        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
-.E80:   a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
-        a0 += a1;
-        r0 = a0;
-
-        (r7:6) = [sp++];
-        unlink;
-        rts;
-DEFUN_END(sse8)
-
-DEFUN(sse16,mL1,
-        (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)):
-        link 0;
-        [--sp] = (r7:6);
-        p0=[fp+24];   // h
-        i0=r1;        // pix1
-        i1=r2;        // pix2
-        r2=[fp+20];   // line_size
-        r2+=-16;
-        m0=r2;
-
-        a0=a1=0;
-        DISALGNEXCPT                       || R0 = [I0++]   || R2  =[I1++];
-        LSETUP(.S160,.E160) LC0=P0;
-
-.S160:  DISALGNEXCPT                       || R1 = [I0++]   || R3 = [I1++];
-        (R7,R6) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++]   || R2 = [I1++];
-        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
-        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
-        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R1 = [I0++]   || R3 = [I1++];
-        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
-        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
-        (R7,R6) = BYTEOP16M (R1:0,R3:2)    || R0 = [I0++M0] || R2 = [I1++M0];
-        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
-        a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
-        (R7,R6) = BYTEOP16M (R1:0,R3:2) (R)|| R0 = [I0++]   || R2 = [I1++];
-        a0 += r7.l * r7.l, a1 += r7.h * r7.h (is);
-.E160:  a0 += r6.l * r6.l, a1 += r6.h * r6.h (is);
-        a0 += a1;
-        r0 = a0;
-
-        (r7:6) = [sp++];
-        unlink;
-        rts;
-DEFUN_END(sse16)
diff --git a/libavcodec/bfin/vp3_bfin.c b/libavcodec/bfin/vp3_bfin.c
deleted file mode 100644 (file)
index ae7ce93..0000000
+++ /dev/null
@@ -1,66 +0,0 @@
-/*
- * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-
-#include <stdint.h>
-#include <string.h>
-
-#include "libavutil/attributes.h"
-#include "libavcodec/avcodec.h"
-#include "libavcodec/vp3dsp.h"
-#include "libavcodec/dsputil.h"
-
-void ff_bfin_vp3_idct(int16_t *block);
-
-/* Intra iDCT offset 128 */
-static void bfin_vp3_idct_put(uint8_t *dest, int line_size, int16_t *block)
-{
-    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP + 128;
-    int i,j;
-
-    ff_bfin_vp3_idct (block);
-
-    for (i=0;i<8;i++)
-        for (j=0;j<8;j++)
-            dest[line_size*i + j] = cm[block[j*8 + i]];
-
-    memset(block, 0, 128);
-}
-
-/* Inter iDCT */
-static void bfin_vp3_idct_add(uint8_t *dest, int line_size, int16_t *block)
-{
-    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
-    int i, j;
-
-    ff_bfin_vp3_idct (block);
-    for (i = 0; i < 8; i++)
-        for (j = 0; j < 8; j++)
-            dest[line_size*i + j] = cm[dest[line_size*i + j] + block[j*8 + i]];
-
-    memset(block, 0, 128);
-}
-
-av_cold void ff_vp3dsp_init_bfin(VP3DSPContext *c, int flags)
-{
-    if (!(flags & CODEC_FLAG_BITEXACT)) {
-        c->idct_add = bfin_vp3_idct_add;
-        c->idct_put = bfin_vp3_idct_put;
-    }
-}
diff --git a/libavcodec/bfin/vp3_idct_bfin.S b/libavcodec/bfin/vp3_idct_bfin.S
deleted file mode 100644 (file)
index 615cae9..0000000
+++ /dev/null
@@ -1,273 +0,0 @@
-/*
- * vp3_idct BlackFin
- *
- * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
- *
- * This file is part of Libav.
- *
- * Libav is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * Libav is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- */
-/*
-   This blackfin DSP code implements an 8x8 inverse type II DCT.
-
-Prototype       : void ff_bfin_vp3_idct(int16_t *in)
-
-Registers Used  : A0, A1, R0-R7, I0-I3, B0, B2, B3, M0-M2, L0-L3, P0-P5, LC0.
-
-*/
-
-#include "libavutil/bfin/asm.h"
-
-SECTION_L1_DATA_B
-
-.align 4;
-coefs:
-.short 0x5a82;           //  C4
-.short 0x5a82;           //  C4
-.short 0x30FC;           //cos(3pi/8)  C6
-.short 0x7642;           //cos(pi/8)   C2
-.short 0x18F9;           //cos(7pi/16)
-.short 0x7D8A;           //cos(pi/16)
-.short 0x471D;           //cos(5pi/16)
-.short 0x6A6E;           //cos(3pi/16)
-.short 0x18F9;           //cos(7pi/16)
-.short 0x7D8A;           //cos(pi/16)
-
-SECTION_L1_DATA_A
-
-vtmp: .space 256
-
-#define TMP0 FP-8
-#define TMP1 FP-12
-#define TMP2 FP-16
-
-
-.text
-DEFUN(vp3_idct,mL1,
-        (int16_t *block)):
-
-/********************** Function Prologue *********************************/
-    link 16;
-    [--SP] = (R7:4, P5:3);   // Push the registers onto the stack.
-    B0 = R0;                 // Pointer to Input matrix
-    RELOC(R1, P3, coefs);    // Pointer to Coefficients
-    RELOC(R2, P3, vtmp);     // Pointer to Temporary matrix
-    B3 = R1;
-    B2 = R2;
-    L3 = 20;                // L3 is used for making the coefficient array
-                            // circular.
-                            // MUST BE RESTORED TO ZERO at function exit.
-    M1 = 16 (X);            // All these registers are initialized for
-    M3 = 8(X);              // modifying address offsets.
-
-    I0 = B0;                // I0 points to Input Element (0, 0).
-    I2 = B0;                // I2 points to Input Element (0, 0).
-    I2 += M3 || R0.H = W[I0];
-                            // Element 0 is read into R0.H
-    I1 = I2;                // I1 points to input Element (0, 6).
-    I1 += 4  || R0.L = W[I2++];
-                            // I2 points to input Element (0, 4).
-                            // Element 4 is read into R0.L.
-    P2 = 8 (X);
-    P3 = 32 (X);
-    P4 = -32 (X);
-    P5 = 98 (X);
-    R7 = 0x8000(Z);
-    I3 = B3;                // I3 points to Coefficients
-    P0 = B2;                // P0 points to array Element (0, 0) of temp
-    P1 = B2;
-    R7 = [I3++] || [TMP2]=R7;            // Coefficient C4 is read into R7.H and R7.L.
-    MNOP;
-    NOP;
-
-    /*
-     *   A1 =      Y0 * cos(pi/4)
-     *   A0 =      Y0 * cos(pi/4)
-     *   A1 = A1 + Y4 * cos(pi/4)
-     *   A0 = A0 - Y4 * cos(pi/4)
-     *   load:
-     *     R1=(Y2,Y6)
-     *     R7=(C2,C6)
-     *   res:
-     *     R3=Y0, R2=Y4
-     */
-    A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || I0+= 4       || R1.L=W[I1++];
-    R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
-
-    LSETUP (.0, .1) LC0 = P2; // perform 8 1d idcts
-
-    P2 = 112 (X);
-    P1 = P1 + P2;           // P1 points to element (7, 0) of temp buffer.
-    P2 = -94(X);
-
-.0:
-       /*
-        *   A1 =      Y2 * cos(3pi/8)
-        *   A0 =      Y2 * cos(pi/8)
-        *   A1 = A1 - Y6 * cos(pi/8)
-        *   A0 = A0 + Y6 * cos(3pi/8)
-        *      R5 = (Y1,Y7)
-        *      R7 = (C1,C7)
-        *   res:
-        *      R1=Y2, R0=Y6
-        */
-        A1=R7.L*R1.H,       A0=R7.H*R1.H (IS)        || I0+=4        || R5.H=W[I0];
-        R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS)  || R5.L=W[I1--] || R7=[I3++];
-        /*
-        *   Y0 = Y0 + Y6.
-        *   Y4 = Y4 + Y2.
-        *   Y2 = Y4 - Y2.
-        *   Y6 = Y0 - Y6.
-        *     R3 is saved
-        *     R6.l=Y3
-        * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
-        */
-        R3=R3+R0, R0=R3-R0;
-        R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
-        /*
-         *  Compute the odd portion (1,3,5,7) even is done.
-         *
-         *  Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
-         *  Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
-         *  Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
-         *  Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
-         */
-        //  R5=(Y1,Y7)  R6=(Y5,Y3)                                                   // R7=(C1,C7)
-        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || [TMP1]=R2 || R6.H=W[I2--];
-        A1-=R7.H*R5.L,       A0+=R7.L*R5.L (IS)       || I0-=4     || R7=[I3++];
-        A1+=R7.H*R6.H,       A0+=R7.L*R6.H (IS)       || I0+=M1;                     // R7=(C3,C5)
-        R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
-        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || R4=[TMP0];
-        A1+=R7.H*R5.L,       A0-=R7.L*R5.L (IS)       || I1+=M1    || R7=[I3++];     // R7=(C1,C7)
-        A1+=R7.L*R6.H,       A0-=R7.H*R6.H (IS);
-        R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
-        // R3=Y1, R2=Y7, R7=Y5, R6=Y3
-
-        /* Transpose write column. */
-        R5.H=R4+R2 (RND12);                                   // Y0=Y0+Y7
-        R5.L=R4-R2 (RND12) || R4 = [TMP1];                    // Y7=Y7-Y0
-        R2.H=R1+R7 (RND12) || W[P0++P3]=R5.H;                 // Y2=Y2+Y5 st Y0
-        R2.L=R1-R7 (RND12) || W[P1++P4]=R5.L || R7=[I3++];    // Y5=Y2-Y5 st Y7
-        R5.H=R0-R3 (RND12) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
-        R5.L=R0+R3 (RND12) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
-        R3.H=R4-R6 (RND12) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
-        R3.L=R4+R6 (RND12) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
-
-        /* pipeline loop start, + drain Y3, Y4 */
-        A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || W[P0++P2]= R3.H || R1.H = W[I0--];
-.1:     R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
-
-
-
-    I0 = B2;                // I0 points to Input Element (0, 0)
-    I2 = B2;                // I2 points to Input Element (0, 0)
-    I2 += M3 || R0.H = W[I0];
-                            // Y0 is read in R0.H
-    I1 = I2;                // I1 points to input Element (0, 6)
-    I1 += 4  || R0.L = W[I2++];
-                            // I2 points to input Element (0, 4)
-                            // Y4 is read in R0.L
-    P2 = 8 (X);
-    I3 = B3;                // I3 points to Coefficients
-    P0 = B0;                // P0 points to array Element (0, 0) for writing
-                            // output
-    P1 = B0;
-    R7 = [I3++];            // R7.H = C4 and R7.L = C4
-    NOP;
-
-    /*
-     *   A1 =      Y0 * cos(pi/4)
-     *   A0 =      Y0 * cos(pi/4)
-     *   A1 = A1 + Y4 * cos(pi/4)
-     *   A0 = A0 - Y4 * cos(pi/4)
-     *   load:
-     *     R1=(Y2,Y6)
-     *     R7=(C2,C6)
-     *   res:
-     *     R3=Y0, R2=Y4
-     */
-    A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || I0+=4        || R1.L=W[I1++];
-    R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
-
-    LSETUP (.2, .3) LC0 = P2; // peform 8 1d idcts
-    P2 = 112 (X);
-    P1 = P1 + P2;
-    P2 = -94(X);
-
-.2:
-        /*
-         *   A1 =      Y2 * cos(3pi/8)
-         *   A0 =      Y2 * cos(pi/8)
-         *   A1 = A1 - Y6 * cos(pi/8)
-         *   A0 = A0 + Y6 * cos(3pi/8)
-         *      R5 = (Y1,Y7)
-         *      R7 = (C1,C7)
-         *   res:
-         *      R1=Y2, R0=Y6
-         */
-        A1=R7.L*R1.H,       A0=R7.H*R1.H (IS)        || I0+=4        || R5.H=W[I0];
-        R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS)  || R5.L=W[I1--] || R7=[I3++];
-        /*
-        *   Y0 = Y0 + Y6.
-        *   Y4 = Y4 + Y2.
-        *   Y2 = Y4 - Y2.
-        *   Y6 = Y0 - Y6.
-        *     R3 is saved
-        *     R6.l=Y3
-        * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
-        */
-        R3=R3+R0, R0=R3-R0;
-        R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
-        /*
-         *  Compute the odd portion (1,3,5,7) even is done.
-         *
-         *  Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
-         *  Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
-         *  Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
-         *  Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
-         */
-        //  R5=(Y1,Y7)  R6=(Y5,Y3)                                                   // R7=(C1,C7)
-        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || [TMP1]=R2 || R6.H=W[I2--];
-        A1-=R7.H*R5.L,       A0+=R7.L*R5.L (IS)       || I0-=4     || R7=[I3++];
-        A1+=R7.H*R6.H,       A0+=R7.L*R6.H (IS)       || I0+=M1;                     // R7=(C3,C5)
-        R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
-        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || R4=[TMP0];
-        A1+=R7.H*R5.L,       A0-=R7.L*R5.L (IS)       || I1+=M1    || R7=[I3++];     // R7=(C1,C7)
-        A1+=R7.L*R6.H,       A0-=R7.H*R6.H (IS);
-        R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
-        // R3=Y1, R2=Y7, R7=Y5, R6=Y3
-
-        /* Transpose write column. */
-        R5.H=R4+R2 (RND20);                                   // Y0=Y0+Y7
-        R5.L=R4-R2 (RND20) || R4 = [TMP1];                    // Y7=Y7-Y0
-        R5=R5>>>2(v);
-        R2.H=R1+R7 (RND20) || W[P0++P3]=R5.H;                 // Y2=Y2+Y5 st Y0
-        R2.L=R1-R7 (RND20) || W[P1++P4]=R5.L || R7=[I3++];    // Y5=Y2-Y5 st Y7
-        R2=R2>>>2(v);
-        R5.H=R0-R3 (RND20) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
-        R5.L=R0+R3 (RND20) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
-        R5=R5>>>2(v);
-        R3.H=R4-R6 (RND20) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
-        R3.L=R4+R6 (RND20) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
-        R3=R3>>>2(v);
-        /* pipeline loop start, + drain Y3, Y4 */
-        A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || W[P0++P2]= R3.H || R1.H = W[I0--];
-.3:     R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
-
-    L3 = 0;
-    (R7:4,P5:3)=[SP++];
-    unlink;
-    RTS;
-DEFUN_END(vp3_idct)
diff --git a/libavcodec/bfin/vp3dsp.S b/libavcodec/bfin/vp3dsp.S
new file mode 100644 (file)
index 0000000..615cae9
--- /dev/null
@@ -0,0 +1,273 @@
+/*
+ * vp3_idct BlackFin
+ *
+ * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+/*
+   This blackfin DSP code implements an 8x8 inverse type II DCT.
+
+Prototype       : void ff_bfin_vp3_idct(int16_t *in)
+
+Registers Used  : A0, A1, R0-R7, I0-I3, B0, B2, B3, M0-M2, L0-L3, P0-P5, LC0.
+
+*/
+
+#include "libavutil/bfin/asm.h"
+
+SECTION_L1_DATA_B
+
+.align 4;
+coefs:
+.short 0x5a82;           //  C4
+.short 0x5a82;           //  C4
+.short 0x30FC;           //cos(3pi/8)  C6
+.short 0x7642;           //cos(pi/8)   C2
+.short 0x18F9;           //cos(7pi/16)
+.short 0x7D8A;           //cos(pi/16)
+.short 0x471D;           //cos(5pi/16)
+.short 0x6A6E;           //cos(3pi/16)
+.short 0x18F9;           //cos(7pi/16)
+.short 0x7D8A;           //cos(pi/16)
+
+SECTION_L1_DATA_A
+
+vtmp: .space 256
+
+#define TMP0 FP-8
+#define TMP1 FP-12
+#define TMP2 FP-16
+
+
+.text
+DEFUN(vp3_idct,mL1,
+        (int16_t *block)):
+
+/********************** Function Prologue *********************************/
+    link 16;
+    [--SP] = (R7:4, P5:3);   // Push the registers onto the stack.
+    B0 = R0;                 // Pointer to Input matrix
+    RELOC(R1, P3, coefs);    // Pointer to Coefficients
+    RELOC(R2, P3, vtmp);     // Pointer to Temporary matrix
+    B3 = R1;
+    B2 = R2;
+    L3 = 20;                // L3 is used for making the coefficient array
+                            // circular.
+                            // MUST BE RESTORED TO ZERO at function exit.
+    M1 = 16 (X);            // All these registers are initialized for
+    M3 = 8(X);              // modifying address offsets.
+
+    I0 = B0;                // I0 points to Input Element (0, 0).
+    I2 = B0;                // I2 points to Input Element (0, 0).
+    I2 += M3 || R0.H = W[I0];
+                            // Element 0 is read into R0.H
+    I1 = I2;                // I1 points to input Element (0, 6).
+    I1 += 4  || R0.L = W[I2++];
+                            // I2 points to input Element (0, 4).
+                            // Element 4 is read into R0.L.
+    P2 = 8 (X);
+    P3 = 32 (X);
+    P4 = -32 (X);
+    P5 = 98 (X);
+    R7 = 0x8000(Z);
+    I3 = B3;                // I3 points to Coefficients
+    P0 = B2;                // P0 points to array Element (0, 0) of temp
+    P1 = B2;
+    R7 = [I3++] || [TMP2]=R7;            // Coefficient C4 is read into R7.H and R7.L.
+    MNOP;
+    NOP;
+
+    /*
+     *   A1 =      Y0 * cos(pi/4)
+     *   A0 =      Y0 * cos(pi/4)
+     *   A1 = A1 + Y4 * cos(pi/4)
+     *   A0 = A0 - Y4 * cos(pi/4)
+     *   load:
+     *     R1=(Y2,Y6)
+     *     R7=(C2,C6)
+     *   res:
+     *     R3=Y0, R2=Y4
+     */
+    A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || I0+= 4       || R1.L=W[I1++];
+    R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
+
+    LSETUP (.0, .1) LC0 = P2; // perform 8 1d idcts
+
+    P2 = 112 (X);
+    P1 = P1 + P2;           // P1 points to element (7, 0) of temp buffer.
+    P2 = -94(X);
+
+.0:
+       /*
+        *   A1 =      Y2 * cos(3pi/8)
+        *   A0 =      Y2 * cos(pi/8)
+        *   A1 = A1 - Y6 * cos(pi/8)
+        *   A0 = A0 + Y6 * cos(3pi/8)
+        *      R5 = (Y1,Y7)
+        *      R7 = (C1,C7)
+        *   res:
+        *      R1=Y2, R0=Y6
+        */
+        A1=R7.L*R1.H,       A0=R7.H*R1.H (IS)        || I0+=4        || R5.H=W[I0];
+        R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS)  || R5.L=W[I1--] || R7=[I3++];
+        /*
+        *   Y0 = Y0 + Y6.
+        *   Y4 = Y4 + Y2.
+        *   Y2 = Y4 - Y2.
+        *   Y6 = Y0 - Y6.
+        *     R3 is saved
+        *     R6.l=Y3
+        * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
+        */
+        R3=R3+R0, R0=R3-R0;
+        R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
+        /*
+         *  Compute the odd portion (1,3,5,7) even is done.
+         *
+         *  Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
+         *  Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
+         *  Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
+         *  Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
+         */
+        //  R5=(Y1,Y7)  R6=(Y5,Y3)                                                   // R7=(C1,C7)
+        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || [TMP1]=R2 || R6.H=W[I2--];
+        A1-=R7.H*R5.L,       A0+=R7.L*R5.L (IS)       || I0-=4     || R7=[I3++];
+        A1+=R7.H*R6.H,       A0+=R7.L*R6.H (IS)       || I0+=M1;                     // R7=(C3,C5)
+        R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
+        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || R4=[TMP0];
+        A1+=R7.H*R5.L,       A0-=R7.L*R5.L (IS)       || I1+=M1    || R7=[I3++];     // R7=(C1,C7)
+        A1+=R7.L*R6.H,       A0-=R7.H*R6.H (IS);
+        R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
+        // R3=Y1, R2=Y7, R7=Y5, R6=Y3
+
+        /* Transpose write column. */
+        R5.H=R4+R2 (RND12);                                   // Y0=Y0+Y7
+        R5.L=R4-R2 (RND12) || R4 = [TMP1];                    // Y7=Y7-Y0
+        R2.H=R1+R7 (RND12) || W[P0++P3]=R5.H;                 // Y2=Y2+Y5 st Y0
+        R2.L=R1-R7 (RND12) || W[P1++P4]=R5.L || R7=[I3++];    // Y5=Y2-Y5 st Y7
+        R5.H=R0-R3 (RND12) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
+        R5.L=R0+R3 (RND12) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
+        R3.H=R4-R6 (RND12) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
+        R3.L=R4+R6 (RND12) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
+
+        /* pipeline loop start, + drain Y3, Y4 */
+        A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || W[P0++P2]= R3.H || R1.H = W[I0--];
+.1:     R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
+
+
+
+    I0 = B2;                // I0 points to Input Element (0, 0)
+    I2 = B2;                // I2 points to Input Element (0, 0)
+    I2 += M3 || R0.H = W[I0];
+                            // Y0 is read in R0.H
+    I1 = I2;                // I1 points to input Element (0, 6)
+    I1 += 4  || R0.L = W[I2++];
+                            // I2 points to input Element (0, 4)
+                            // Y4 is read in R0.L
+    P2 = 8 (X);
+    I3 = B3;                // I3 points to Coefficients
+    P0 = B0;                // P0 points to array Element (0, 0) for writing
+                            // output
+    P1 = B0;
+    R7 = [I3++];            // R7.H = C4 and R7.L = C4
+    NOP;
+
+    /*
+     *   A1 =      Y0 * cos(pi/4)
+     *   A0 =      Y0 * cos(pi/4)
+     *   A1 = A1 + Y4 * cos(pi/4)
+     *   A0 = A0 - Y4 * cos(pi/4)
+     *   load:
+     *     R1=(Y2,Y6)
+     *     R7=(C2,C6)
+     *   res:
+     *     R3=Y0, R2=Y4
+     */
+    A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || I0+=4        || R1.L=W[I1++];
+    R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
+
+    LSETUP (.2, .3) LC0 = P2; // peform 8 1d idcts
+    P2 = 112 (X);
+    P1 = P1 + P2;
+    P2 = -94(X);
+
+.2:
+        /*
+         *   A1 =      Y2 * cos(3pi/8)
+         *   A0 =      Y2 * cos(pi/8)
+         *   A1 = A1 - Y6 * cos(pi/8)
+         *   A0 = A0 + Y6 * cos(3pi/8)
+         *      R5 = (Y1,Y7)
+         *      R7 = (C1,C7)
+         *   res:
+         *      R1=Y2, R0=Y6
+         */
+        A1=R7.L*R1.H,       A0=R7.H*R1.H (IS)        || I0+=4        || R5.H=W[I0];
+        R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS)  || R5.L=W[I1--] || R7=[I3++];
+        /*
+        *   Y0 = Y0 + Y6.
+        *   Y4 = Y4 + Y2.
+        *   Y2 = Y4 - Y2.
+        *   Y6 = Y0 - Y6.
+        *     R3 is saved
+        *     R6.l=Y3
+        * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
+        */
+        R3=R3+R0, R0=R3-R0;
+        R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
+        /*
+         *  Compute the odd portion (1,3,5,7) even is done.
+         *
+         *  Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
+         *  Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
+         *  Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
+         *  Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
+         */
+        //  R5=(Y1,Y7)  R6=(Y5,Y3)                                                   // R7=(C1,C7)
+        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || [TMP1]=R2 || R6.H=W[I2--];
+        A1-=R7.H*R5.L,       A0+=R7.L*R5.L (IS)       || I0-=4     || R7=[I3++];
+        A1+=R7.H*R6.H,       A0+=R7.L*R6.H (IS)       || I0+=M1;                     // R7=(C3,C5)
+        R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
+        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || R4=[TMP0];
+        A1+=R7.H*R5.L,       A0-=R7.L*R5.L (IS)       || I1+=M1    || R7=[I3++];     // R7=(C1,C7)
+        A1+=R7.L*R6.H,       A0-=R7.H*R6.H (IS);
+        R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
+        // R3=Y1, R2=Y7, R7=Y5, R6=Y3
+
+        /* Transpose write column. */
+        R5.H=R4+R2 (RND20);                                   // Y0=Y0+Y7
+        R5.L=R4-R2 (RND20) || R4 = [TMP1];                    // Y7=Y7-Y0
+        R5=R5>>>2(v);
+        R2.H=R1+R7 (RND20) || W[P0++P3]=R5.H;                 // Y2=Y2+Y5 st Y0
+        R2.L=R1-R7 (RND20) || W[P1++P4]=R5.L || R7=[I3++];    // Y5=Y2-Y5 st Y7
+        R2=R2>>>2(v);
+        R5.H=R0-R3 (RND20) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
+        R5.L=R0+R3 (RND20) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
+        R5=R5>>>2(v);
+        R3.H=R4-R6 (RND20) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
+        R3.L=R4+R6 (RND20) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
+        R3=R3>>>2(v);
+        /* pipeline loop start, + drain Y3, Y4 */
+        A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || W[P0++P2]= R3.H || R1.H = W[I0--];
+.3:     R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
+
+    L3 = 0;
+    (R7:4,P5:3)=[SP++];
+    unlink;
+    RTS;
+DEFUN_END(vp3_idct)
diff --git a/libavcodec/bfin/vp3dsp_init.c b/libavcodec/bfin/vp3dsp_init.c
new file mode 100644 (file)
index 0000000..ae7ce93
--- /dev/null
@@ -0,0 +1,66 @@
+/*
+ * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+#include <string.h>
+
+#include "libavutil/attributes.h"
+#include "libavcodec/avcodec.h"
+#include "libavcodec/vp3dsp.h"
+#include "libavcodec/dsputil.h"
+
+void ff_bfin_vp3_idct(int16_t *block);
+
+/* Intra iDCT offset 128 */
+static void bfin_vp3_idct_put(uint8_t *dest, int line_size, int16_t *block)
+{
+    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP + 128;
+    int i,j;
+
+    ff_bfin_vp3_idct (block);
+
+    for (i=0;i<8;i++)
+        for (j=0;j<8;j++)
+            dest[line_size*i + j] = cm[block[j*8 + i]];
+
+    memset(block, 0, 128);
+}
+
+/* Inter iDCT */
+static void bfin_vp3_idct_add(uint8_t *dest, int line_size, int16_t *block)
+{
+    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
+    int i, j;
+
+    ff_bfin_vp3_idct (block);
+    for (i = 0; i < 8; i++)
+        for (j = 0; j < 8; j++)
+            dest[line_size*i + j] = cm[dest[line_size*i + j] + block[j*8 + i]];
+
+    memset(block, 0, 128);
+}
+
+av_cold void ff_vp3dsp_init_bfin(VP3DSPContext *c, int flags)
+{
+    if (!(flags & CODEC_FLAG_BITEXACT)) {
+        c->idct_add = bfin_vp3_idct_add;
+        c->idct_put = bfin_vp3_idct_put;
+    }
+}