Coldfire assembler implementation of hybrid_filter for libtta. Speeds up decoding on h300 by 4.2MHz. Set svn properties.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@27404 a1c6a512-1295-4272-9138-f99709370657

4 files changed, 172 insertions, 1 deletions

diff --git a/apps/codecs/libtta/SOURCES b/apps/codecs/libtta/SOURCES
index 35f2660dd7..0a8f1171eb 100644
--- a/apps/codecs/libtta/SOURCES
+++ b/apps/codecs/libtta/SOURCES
@@ -2,3 +2,6 @@ ttadec.c
 #ifdef CPU_ARM
 filter_arm.S
 #endif
+#ifdef CPU_COLDFIRE
+filter_coldfire.S
+#endif
diff --git a/apps/codecs/libtta/filter.h b/apps/codecs/libtta/filter.h
index 6eef6dcf42..228757b9a0 100644
--- a/apps/codecs/libtta/filter.h
+++ b/apps/codecs/libtta/filter.h
@@ -42,7 +42,7 @@
 ///////// Filter Settings //////////
 static int flt_set[3] = {10, 9, 10};
 
-#ifdef CPU_ARM
+#if defined(CPU_ARM) || defined(CPU_COLDFIRE)
 int hybrid_filter(fltst *fs, int *in); /* implements in filter_arm.S */
 
 #else
diff --git a/apps/codecs/libtta/filter_coldfire.S b/apps/codecs/libtta/filter_coldfire.S
new file mode 100644
index 0000000000..3950eb52e6
--- /dev/null
+++ b/apps/codecs/libtta/filter_coldfire.S
@@ -0,0 +1,164 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2010 Nils Wallménius
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
+ * KIND, either express or implied.
+ *
+ ****************************************************************************/
+
+#include "config.h"
+
+/*
+ * The following is an assembler optimised version of
+ * void hybrid_filter(fltst *fs, int *in)
+ */
+
+#if defined(USE_IRAM)
+    .section .icode
+#else
+    .text
+#endif
+    .align 2
+    .global     hybrid_filter
+    .type       hybrid_filter, @function
+
+hybrid_filter:
+    lea.l    (-8*4, %sp), %sp
+    movem.l  %d2-%d7/%a2-%a3, (%sp)       | save some registers
+    move.l   (8*4+4, %sp), %a0            | a0 = fs
+    movem.l  (%a0), %d4-%d5               | d4 = fs->index, d5 = fs->error
+
+    lea.l    (%a0, %d4.l*4), %a2
+    lea.l    (148, %a2), %a1              | a1 = fs->dl + fs->index (*pA)
+    lea.l    (52, %a2), %a2               | a2 = fs->dx + fs->index (*pM)
+
+    move.l   (%a1)+, %a3                  | load one value from *pA (needed in every case)
+    movem.l  (20, %a0), %d0-%d3           | load 4 values from *pB
+
+    tst.l    %d5
+    blt      .hf_negative
+    bgt      .hf_positive
+
+                                          | fs->error == 0
+    mac.l    %d0, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d1, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d2, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d3, %a3, (%a1)+, %d4, %acc0
+    movem.l  (4*4+20, %a0), %d0-%d3       | load 4 values from *pB
+    bra      0f
+
+    .hf_negative:                         | fs->error < 0
+    movem.l  (%a2), %d4-%d7               | load 4 values from *pM
+    sub.l    %d4, %d0
+    sub.l    %d5, %d1
+    sub.l    %d6, %d2
+    sub.l    %d7, %d3
+    movem.l  %d0-%d3, (20, %a0)
+    mac.l    %d0, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d1, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d2, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d3, %a3, (%a1)+, %d4, %acc0
+
+    movem.l  (4*4+20, %a0), %d0-%d3       | load 4 values from *pB
+    movem.l  (4*4, %a2), %d5-%d7/%a3      | load 4 values from *pM
+    sub.l    %d5, %d0
+    sub.l    %d6, %d1
+    sub.l    %d7, %d2
+    sub.l    %a3, %d3
+    movem.l  %d0-%d3, (4*4+20, %a0)
+    bra      0f
+
+    .hf_positive:                         | fs->error > 0
+    movem.l  (%a2), %d4-%d7               | load 4 values from *pM
+    add.l    %d4, %d0
+    add.l    %d5, %d1
+    add.l    %d6, %d2
+    add.l    %d7, %d3
+    movem.l  %d0-%d3, (20, %a0)
+    mac.l    %d0, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d1, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d2, %a3, (%a1)+, %a3, %acc0
+    mac.l    %d3, %a3, (%a1)+, %d4, %acc0
+
+    movem.l  (4*4+20, %a0), %d0-%d3       | load 4 values from *pB
+    movem.l  (4*4, %a2), %d5-%d7/%a3      | load 4 values from *pM
+    add.l    %d5, %d0
+    add.l    %d6, %d1
+    add.l    %d7, %d2
+    add.l    %a3, %d3
+    movem.l  %d0-%d3, (4*4+20, %a0)
+
+    0:
+
+    mac.l    %d0, %d4, (%a1)+, %d5, %acc0 | common macro block
+    mac.l    %d1, %d5, (%a1)+, %d6, %acc0
+    mac.l    %d2, %d6, (%a1),  %d7, %acc0
+    mac.l    %d3, %d7, %acc0
+
+    move.l   (8*4+8, %sp), %a3            | a3 = in
+    move.l   (%a3), %d3
+    move.l   %d3, (4, %a0)                | fs->error = *in
+    movclr.l %acc0, %d0                   | d0 = sum
+    movem.l  (8,  %a0), %d1-%d2
+    add.l    %d1, %d0                     | sum +=  fs->round
+    asr.l    %d2, %d0                     | sum >>= fs->shift
+
+    add.l    %d0, %d3
+    move.l   %d3, (%a3)                   | *in += (sum >> fs->shift)
+
+    move.l   %d3, ( 1*4, %a1)
+    sub.l    %d7, %d3
+    move.l   %d3, ( 0*4, %a1)
+    sub.l    %d6, %d3
+    move.l   %d3, (-1*4, %a1)
+    sub.l    %d5, %d3
+    move.l   %d3, (-2*4, %a1)
+
+    moveq    #30,%d0
+    asr.l    %d0,%d7
+    asr.l    %d0,%d6
+    asr.l    %d0,%d5
+    asr.l    %d0,%d4
+
+    moveq    #1,%d0
+    or.l     %d0,%d7
+    or.l     %d0,%d6
+    or.l     %d0,%d5
+    or.l     %d0,%d4
+
+    lsl.l    #2,%d7
+    lsl.l    #1,%d6
+    lsl.l    #1,%d5
+    movem.l  %d4-%d7, (8*4-3*4,%a2)       | store to *pM
+
+    move.l   (%a0), %d0
+    addq.l   #1, %d0
+    cmp.l    #16, %d0                     | ++fs->index == 16 ?
+    bne      1f
+
+    movem.l  (16*4+148, %a0), %d0-%d7
+    movem.l  %d0-%d7, (148, %a0)
+    movem.l  (16*4+52, %a0), %d0-%d7
+    movem.l  %d0-%d7, (52, %a0)
+    clr.l    %d0                          | fs->index = 0
+    1:
+
+    move.l   %d0, (%a0)
+
+    movem.l  (%sp), %d2-%d7/%a2-%a3       | restore stacked regs
+    lea.l    (8*4, %sp), %sp
+    rts
+
diff --git a/apps/codecs/libtta/ttadec.c b/apps/codecs/libtta/ttadec.c
index 2ff2d24da9..9d53a327f2 100644
--- a/apps/codecs/libtta/ttadec.c
+++ b/apps/codecs/libtta/ttadec.c
@@ -392,6 +392,10 @@ int player_init (tta_info *info) {
     unsigned int data_offset;
     unsigned int st_size;
 
+#ifdef CPU_COLDFIRE
+    coldfire_set_macsr(0); /* signed integer mode */
+#endif
+
     ttainfo = info;
 
     framelen = 0;