iir_mem16() in assembler for Coldfire for a decent performance boost. Add EMAC init in nb_celp.c, since all modes need this as a base.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@15274 a1c6a512-1295-4272-9138-f99709370657

5 files changed, 178 insertions, 0 deletions

diff --git a/apps/codecs/libspeex/SOURCES b/apps/codecs/libspeex/SOURCES
index df73565734..a5d4318282 100644
--- a/apps/codecs/libspeex/SOURCES
+++ b/apps/codecs/libspeex/SOURCES
@@ -42,3 +42,6 @@ vbr.c
 vorbis_psy.c
 vq.c
 window.c
+#ifdef CPU_COLDFIRE
+filters_cf.S
+#endif
diff --git a/apps/codecs/libspeex/config-speex.h b/apps/codecs/libspeex/config-speex.h
index b7385eee02..31ebe80741 100644
--- a/apps/codecs/libspeex/config-speex.h
+++ b/apps/codecs/libspeex/config-speex.h
@@ -11,6 +11,11 @@
 #define ARM4_ASM
 #endif
 
+/* Make use of Coldfire assembly optimizations */
+#if defined(CPU_COLDFIRE)
+#define COLDFIRE_ASM
+#endif
+
 /* Make use of Blackfin assembly optimizations */
 /* #undef BFIN_ASM */
 
diff --git a/apps/codecs/libspeex/filters.c b/apps/codecs/libspeex/filters.c
index a6a5f62d26..02f93a27b1 100644
--- a/apps/codecs/libspeex/filters.c
+++ b/apps/codecs/libspeex/filters.c
@@ -45,6 +45,8 @@
 #include "filters_sse.h"
 #elif defined (ARM4_ASM) || defined(ARM5E_ASM)
 #include "filters_arm4.h"
+#elif defined (COLDFIRE_ASM)
+#define OVERRIDE_IIR_MEM16
 #elif defined (BFIN_ASM)
 #include "filters_bfin.h"
 #endif
diff --git a/apps/codecs/libspeex/filters_cf.S b/apps/codecs/libspeex/filters_cf.S
new file mode 100644
index 0000000000..579af11581
--- /dev/null
+++ b/apps/codecs/libspeex/filters_cf.S
@@ -0,0 +1,165 @@
+/* Copyright (C) 2007 Thom Johansen */
+/**
+   @file filters_cf.S
+   @brief Various analysis/synthesis filters (Coldfire version)
+*/
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+   
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+   
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+   
+   - Neither the name of the Xiph.org Foundation nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
+   CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+    .text
+/* void iir_mem16(const spx_word16_t *x, const spx_coef_t *den, spx_word16_t *y, int N, int ord, spx_mem_t *mem, char *stack) */
+    .global iir_mem16
+iir_mem16:
+    lea.l    (-44, %sp), %sp
+    movem.l  %d2-%d7/%a2-%a6, (%sp)
+    movem.l  (44+4, %sp), %a3-%a5   | a3 = x, a4 = den, a5 = y
+    movem.l  (44+20, %sp), %d0/%a6  | d0 = ord, a6 = mem
+    moveq.l  #8, %d1                | Jump to correct routine based on 'ord'
+    cmp.l    %d1, %d0
+    jeq      .order_8
+    moveq.l  #10, %d1
+    cmp.l    %d1, %d0
+    jeq      .order_10
+    jra      .exit
+
+    | d0 = y[i], d1-d7, a0 = mem[0] .. mem[7]
+    | a3 = x, a4 = den, a5 = y, a6 = temp 
+.order_8:
+    movem.l  (%a6), %d1-%d7/%a0 | Fetch mem[] array
+0:
+    moveq.l  #13, %d0
+    add.l    #4096, %d1
+    asr.l    %d0, %d1           | mem[0] >> 13 with rounding
+    move.w   (%a3)+, %d0
+    ext.l    %d0
+    add.l    %d1, %d0           | Add with x[i]
+    move.l   #32768, %d1
+    add.l    %d1, %d0           | Bias result to [0..65535]
+    cmp.l    #65535, %d0        | Clip to [0..65535] range
+    jle      1f
+    spl.b    %d0                
+    ext.w    %d0
+1:
+    sub.l    %d1, %d0           | Bias clipped result back to [-32768..32767]
+    neg.l    %d0                | msac.w is bugged in gas, do this for now
+    move.w   %d0, (%a5)+        | Write result to y[i]
+    move.l   (%a4)+, %a6        | Fetch den[0] and den[1]
+    mac.w    %a6u, %d0l, %acc0
+    mac.w    %a6l, %d0l, (%a4)+, %a6, %acc1
+    mac.w    %a6u, %d0l, %acc2
+    mac.w    %a6l, %d0l, (%a4)+, %a6, %acc3
+    movclr.l %acc0, %d1
+    add.l    %d2, %d1           | mem[0] = mem[1] - den[0]*y[i]
+    movclr.l %acc1, %d2
+    add.l    %d3, %d2           | mem[1] = mem[2] - den[1]*y[i]
+    movclr.l %acc2, %d3
+    add.l    %d4, %d3           | mem[2] = mem[3] - den[2]*y[i]
+    movclr.l %acc3, %d4
+    add.l    %d5, %d4           | mem[3] = mem[4] - den[3]*y[i]
+    mac.w    %a6u, %d0l, %acc0
+    mac.w    %a6l, %d0l, (%a4)+, %a6, %acc1
+    mac.w    %a6u, %d0l, %acc2
+    mac.w    %a6l, %d0l, %acc3
+    lea.l    (-16, %a4), %a4    | wrap den pointer back to den[0]
+    movclr.l %acc0, %d5
+    add.l    %d6, %d5           | mem[4] = mem[5] - den[4]*y[i]
+    movclr.l %acc1, %d6
+    add.l    %d7, %d6           | mem[5] = mem[6] - den[5]*y[i]
+    movclr.l %acc2, %d7
+    add.l    %a0, %d7           | mem[6] = mem[7] - den[6]*y[i]
+    movclr.l %acc3, %a0         | mem[7] = -den[7]*y[i]
+    subq.l   #1, (44+16, %sp)   | Have we done all samples?
+    jne      0b
+    move.l   (44+24, %sp), %a6  | Fetch mem pointer
+    movem.l  %d1-%d7/%a0, (%a6) | Save back mem[]
+    jra     .exit
+
+    | d0 = y[i], d1-d7, a0-a2 = mem[0] .. mem[9]
+    | a3 = x, a4 = den, a5 = y, a6 = temp 
+.order_10:
+    movem.l  (%a6), %d1-%d7/%a0-%a2 | Fetch mem[] array 
+0:
+    moveq.l  #13, %d0
+    add.l    #4096, %d1
+    asr.l    %d0, %d1           | mem[0] >> 13 with rounding
+    move.w   (%a3)+, %d0
+    ext.l    %d0
+    add.l    %d1, %d0           | Add with x[i]
+    move.l   #32768, %d1
+    add.l    %d1, %d0           | Bias result to [0..65535]
+    cmp.l    #65535, %d0        | Clip to [0..65535] range
+    jle      1f
+    spl.b    %d0                
+    ext.w    %d0
+1:
+    sub.l    %d1, %d0           | Bias clipped result back to [-32768..32767]
+    neg.l    %d0                | msac.w is bugged in gas, do this for now
+    move.w   %d0, (%a5)+        | Write result to y[i]
+    move.l   (%a4)+, %a6        | Fetch den[0] and den[1]
+    mac.w    %a6u, %d0l, %acc0
+    mac.w    %a6l, %d0l, (%a4)+, %a6, %acc1
+    mac.w    %a6u, %d0l, %acc2
+    mac.w    %a6l, %d0l, (%a4)+, %a6, %acc3
+    movclr.l %acc0, %d1
+    add.l    %d2, %d1           | mem[0] = mem[1] - den[0]*y[i]
+    movclr.l %acc1, %d2
+    add.l    %d3, %d2           | mem[1] = mem[2] - den[1]*y[i]
+    movclr.l %acc2, %d3
+    add.l    %d4, %d3           | mem[2] = mem[3] - den[2]*y[i]
+    movclr.l %acc3, %d4
+    add.l    %d5, %d4           | mem[3] = mem[4] - den[3]*y[i]
+    mac.w    %a6u, %d0l, %acc0
+    mac.w    %a6l, %d0l, (%a4)+, %a6, %acc1
+    mac.w    %a6u, %d0l, %acc2
+    mac.w    %a6l, %d0l, (%a4)+, %a6, %acc3
+    lea.l    (-20, %a4), %a4    | wrap den pointer back to den[0]
+    movclr.l %acc0, %d5
+    add.l    %d6, %d5           | mem[4] = mem[5] - den[4]*y[i]
+    movclr.l %acc1, %d6
+    add.l    %d7, %d6           | mem[5] = mem[6] - den[5]*y[i]
+    movclr.l %acc2, %d7
+    add.l    %a0, %d7           | mem[6] = mem[7] - den[6]*y[i]
+    movclr.l %acc3, %a0
+    add.l    %a1, %a0           | mem[7] = mem[8] - den[7]*y[i]
+    mac.w    %a6u, %d0l, %acc0
+    mac.w    %a6l, %d0l, %acc1
+    movclr.l %acc0, %a1
+    add.l    %a2, %a1           | mem[8] = mem[9] - den[8]*y[i]
+    movclr.l %acc1, %a2         | mem[9] = -den[9]*y[i]
+
+    subq.l   #1, (44+16, %sp)   | Have we done all samples?
+    jne      0b
+    move.l   (44+24, %sp), %a6  | Fetch mem pointer
+    movem.l  %d1-%d7/%a0-%a2, (%a6) | Save back mem[]
+
+.exit:
+    movem.l (%sp), %d2-%d7/%a2-%a6
+    lea.l (44, %sp), %sp
+    rts
+
diff --git a/apps/codecs/libspeex/nb_celp.c b/apps/codecs/libspeex/nb_celp.c
index 0798ffbcf4..5d167a2aa3 100644
--- a/apps/codecs/libspeex/nb_celp.c
+++ b/apps/codecs/libspeex/nb_celp.c
@@ -1108,6 +1108,9 @@ void *nb_decoder_init(const SpeexMode *m)
    st->isWideband = 0;
    st->highpass_enabled = 1;
 
+#ifdef CPU_COLDFIRE
+   coldfire_set_macsr(0); // Integer mode
+#endif
 #ifdef ENABLE_VALGRIND
    VALGRIND_MAKE_READABLE(st, NB_DEC_STACK);
 #endif