SPC Codec: Refactor for CPU and clean up some things.

CPU optimization gets its own files in which to fill-in optimizable
routines.

Some pointless #if 0's for profiling need removal. Those macros are
empty if not profiling.

Force some functions that are undesirable to be force-inlined by the
compiler to be not inlined.

Change-Id: Ia7b7e45380d7efb20c9b1a4d52e05db3ef6bbaab

6 files changed, 830 insertions, 0 deletions

diff --git a/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv4.c b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv4.c
new file mode 100644
index 0000000..7eacc3b
--- /dev/null
+++ b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv4.c
@@ -0,0 +1,253 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2007-2010 Michael Sevakis (jhMikeS)
+ *
+ * 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.
+ *
+ ****************************************************************************/
+#if !SPC_NOINTERP
+
+#define SPC_GAUSSIAN_FAST_INTERP
+static inline int gaussian_fast_interp( int16_t const* samples,
+                                        int32_t position,
+                                        int16_t const* fwd,
+                                        int16_t const* rev )
+{
+    int output;
+    int t0, t1, t2, t3;
+
+    asm volatile (
+    "ldrsh   %[t0], [%[samp]]             \n"
+    "ldrsh   %[t2], [%[fwd]]              \n"
+    "ldrsh   %[t1], [%[samp], #2]         \n"
+    "ldrsh   %[t3], [%[fwd], #2]          \n"
+    "mul     %[out], %[t0], %[t2]         \n" /* out= fwd[0]*samp[0] */
+    "ldrsh   %[t0], [%[samp], #4]         \n"
+    "ldrsh   %[t2], [%[rev], #2]          \n"
+    "mla     %[out], %[t1], %[t3], %[out] \n" /* out+=fwd[1]*samp[1] */
+    "ldrsh   %[t1], [%[samp], #6]         \n"
+    "ldrsh   %[t3], [%[rev]]              \n"
+    "mla     %[out], %[t0], %[t2], %[out] \n" /* out+=rev[1]*samp[2] */
+    "mla     %[out], %[t1], %[t3], %[out] \n" /* out+=rev[0]*samp[3] */
+    : [out]"=&r"(output),
+      [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3)
+    : [fwd]"r"(fwd), [rev]"r"(rev),
+      [samp]"r"(samples + (position >> 12)));
+
+    return output;
+}
+
+#define SPC_GAUSSIAN_FAST_AMP
+static inline int gaussian_fast_amp( struct voice_t* voice, int output,
+                                     int* amp_0, int* amp_1 )
+{
+    int t0;
+
+    asm volatile (
+    "mov    %[t0], %[out], asr #11  \n"
+    "mul    %[out], %[t0], %[envx]  \n"
+    : [out]"+r"(output), [t0]"=&r"(t0)
+    : [envx]"r"((int) voice->envx));
+
+    asm volatile (
+    "mov    %[out], %[out], asr #11 \n"
+    "mul    %[a0], %[out], %[v0]    \n"
+    "mul    %[a1], %[out], %[v1]    \n"
+    : [out]"+r"(output),
+      [a0]"=&r"(*amp_0), [a1]"=r"(*amp_1)
+    : [v0]"r"((int) voice->volume [0]),
+      [v1]"r"((int) voice->volume [1]));
+
+    return output;
+}
+
+#define SPC_GAUSSIAN_SLOW_INTERP
+static inline int gaussian_slow_interp( int16_t const* samples,
+                                        int32_t position,
+                                        int16_t const* fwd,
+                                        int16_t const* rev )
+{
+    int output;
+    int t0, t1, t2, t3;
+
+    asm volatile (
+    "ldrsh   %[t0], [%[samp]]              \n"
+    "ldrsh   %[t2], [%[fwd]]               \n"
+    "ldrsh   %[t1], [%[samp], #2]          \n"
+    "ldrsh   %[t3], [%[fwd], #2]           \n"
+    "mul     %[out], %[t2], %[t0]          \n" /* fwd[0]*samp[0] */
+    "ldrsh   %[t2], [%[rev], #2]           \n"
+    "mul     %[t0], %[t3], %[t1]           \n" /* fwd[1]*samp[1] */
+    "ldrsh   %[t1], [%[samp], #4]          \n"
+    "mov     %[out], %[out], asr #12       \n"
+    "ldrsh   %[t3], [%[rev]]               \n"
+    "mul     %[t2], %[t1], %[t2]           \n" /* rev[1]*samp[2] */
+    "ldrsh   %[t1], [%[samp], #6]          \n"
+    "add     %[t0], %[out], %[t0], asr #12 \n"
+    "mul     %[t3], %[t1], %[t3]           \n" /* rev[0]*samp[3] */
+    "add     %[t2], %[t0], %[t2], asr #12  \n"
+    "mov     %[t2], %[t2], lsl #17         \n"
+    "mov     %[t3], %[t3], asr #12         \n"
+    "mov     %[t3], %[t3], asl #1          \n"
+    "add     %[out], %[t3], %[t2], asr #16 \n"
+    : [out]"=&r"(output),
+      [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=&r"(t3)
+    : [fwd]"r"(fwd), [rev]"r"(rev),
+      [samp]"r"(samples + (position >> 12)));
+
+    return CLAMP16( output );
+}
+
+#define SPC_GAUSSIAN_SLOW_AMP
+static inline int gaussian_slow_amp( struct voice_t* voice, int output,
+                                     int* amp_0, int* amp_1 )
+{
+    int t0;
+
+    asm volatile (
+    "mul     %[t0], %[out], %[envx]"
+    : [t0]"=r"(t0)
+    : [out]"r"(output), [envx]"r"((int) voice->envx));
+    asm volatile (
+    "mov     %[t0], %[t0], asr #11 \n"
+    "bic     %[t0], %[t0], #0x1    \n"
+    "mul     %[a0], %[t0], %[v0]   \n"
+    "mul     %[a1], %[t0], %[v1]   \n"
+    : [t0]"+r"(t0),
+      [a0]"=&r"(*amp_0), [a1]"=r"(*amp_1)
+    : [v0]"r"((int) voice->volume [0]),
+      [v1]"r"((int) voice->volume [1]));
+
+    return t0;
+}
+
+#else /* SPC_NOINTERP */
+
+#define SPC_LINEAR_INTERP
+static inline int linear_interp( int16_t const* samples, int32_t position )
+{
+    int output = (int) samples;
+    int y1;
+
+    asm volatile(
+    "mov    %[y1], %[f], lsr #12        \n"
+    "eor    %[f], %[f], %[y1], lsl #12  \n" 
+    "add    %[y1], %[y0], %[y1], lsl #1 \n"
+    "ldrsh  %[y0], [%[y1], #2]          \n"
+    "ldrsh  %[y1], [%[y1], #4]          \n"
+    "sub    %[y1], %[y1], %[y0]         \n"
+    "mul    %[f], %[y1], %[f]           \n"
+    "add    %[y0], %[y0], %[f], asr #12 \n"
+    : [f]"+r"(position), [y0]"+r"(output), [y1]"=&r"(y1));
+
+    return output;
+}
+
+#define SPC_LINEAR_AMP
+static inline int linear_amp( struct voice_t* voice, int output,
+                              int* amp_0, int* amp_1 )
+{
+    int t0;
+
+    asm volatile(
+    "mul    %[t0], %[out], %[envx]"
+    : [t0]"=&r"(t0)
+    : [out]"r"(output), [envx]"r"(voice->envx));
+    asm volatile(
+    "mov    %[t0], %[t0], asr #11 \n"
+    "mul    %[a1], %[t0], %[v1]   \n"
+    "mul    %[a0], %[t0], %[v0]   \n"
+    : [t0]"+r"(t0),
+      [a0]"=&r"(*amp_0), [a1]"=&r"(*amp_1)
+    : [v0]"r"((int) voice->volume [0]),
+      [v1]"r"((int) voice->volume [1]));
+
+    return t0;
+}
+
+#endif /* !SPC_NOINTERP */
+
+
+#if !SPC_NOECHO
+
+#define SPC_DSP_ECHO_APPLY
+
+/* Echo filter history */
+static int32_t fir_buf[FIR_BUF_CNT] IBSS_ATTR_SPC
+    __attribute__(( aligned(FIR_BUF_ALIGN*1) ));
+
+static inline void echo_init( struct Spc_Dsp* this )
+{
+    this->fir.ptr = fir_buf;
+    ci->memset( fir_buf, 0, sizeof fir_buf );
+}
+
+static inline void echo_apply( struct Spc_Dsp* this, uint8_t *echo_ptr,
+                               int* out_0, int* out_1 )
+{
+    int t0 = GET_LE16SA( echo_ptr     );
+    int t1 = GET_LE16SA( echo_ptr + 2 );
+
+    /* Keep last 8 samples */
+    int32_t *fir_ptr;
+    asm volatile (
+    "add    %[p], %[t_p], #8         \n"
+    "bic    %[t_p], %[p], %[mask]    \n"
+    "str    %[t0], [%[p], #-8]       \n"
+    "str    %[t1], [%[p], #-4]       \n"
+    /* duplicate at +8 eliminates wrap checking below */
+    "str    %[t0], [%[p], #56]       \n"
+    "str    %[t1], [%[p], #60]       \n"
+    : [p]"=&r"(fir_ptr), [t_p]"+r"(this->fir.ptr)
+    : [t0]"r"(t0), [t1]"r"(t1), [mask]"i"(~FIR_BUF_MASK));
+
+    int32_t *fir_coeff = this->fir.coeff;
+
+    asm volatile (
+    "ldmia  %[c]!, { r0-r1 }         \n"
+    "ldmia  %[p]!, { r4-r5 }         \n"
+    "mul    %[acc0],     r0, %[acc0] \n"
+    "mul    %[acc1],     r0, %[acc1] \n"
+    "mla    %[acc0], r4, r1, %[acc0] \n"
+    "mla    %[acc1], r5, r1, %[acc1] \n"
+    "ldmia  %[c]!, { r0-r1 }         \n"
+    "ldmia  %[p]!, { r2-r5 }         \n"
+    "mla    %[acc0], r2, r0, %[acc0] \n"
+    "mla    %[acc1], r3, r0, %[acc1] \n"
+    "mla    %[acc0], r4, r1, %[acc0] \n"
+    "mla    %[acc1], r5, r1, %[acc1] \n"
+    "ldmia  %[c]!, { r0-r1 }         \n"
+    "ldmia  %[p]!, { r2-r5 }         \n"
+    "mla    %[acc0], r2, r0, %[acc0] \n"
+    "mla    %[acc1], r3, r0, %[acc1] \n"
+    "mla    %[acc0], r4, r1, %[acc0] \n"
+    "mla    %[acc1], r5, r1, %[acc1] \n"
+    "ldmia  %[c]!, { r0-r1 }         \n"
+    "ldmia  %[p]!, { r2-r5 }         \n"
+    "mla    %[acc0], r2, r0, %[acc0] \n"
+    "mla    %[acc1], r3, r0, %[acc1] \n"
+    "mla    %[acc0], r4, r1, %[acc0] \n"
+    "mla    %[acc1], r5, r1, %[acc1] \n"
+    : [acc0]"+r"(t0), [acc1]"+r"(t1),
+      [p]"+r"(fir_ptr), [c]"+r"(fir_coeff)
+    :
+    : "r0", "r1", "r2", "r3", "r4", "r5");
+
+    *out_0 = t0;
+    *out_1 = t1;
+}
+
+#endif /* SPC_NOECHO */
diff --git a/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv4.h b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv4.h
new file mode 100644
index 0000000..c9985e1
--- /dev/null
+++ b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv4.h
@@ -0,0 +1,45 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2007-2010 Michael Sevakis (jhMikeS)
+ *
+ * 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.
+ *
+ ****************************************************************************/
+#if !SPC_NOECHO
+
+#define SPC_DSP_ECHO_APPLY
+
+enum
+{
+    FIR_BUF_CNT   = FIR_BUF_HALF * 2 * 2,
+    FIR_BUF_SIZE  = FIR_BUF_CNT * sizeof ( int32_t ),
+    FIR_BUF_ALIGN = FIR_BUF_SIZE,
+    FIR_BUF_MASK  = ~((FIR_BUF_ALIGN / 2) | (sizeof ( int32_t ) * 2 - 1))
+};
+
+/* Echo filter structure embedded in struct Spc_Dsp */
+struct echo_filter
+{
+   /* fir_buf [i + 8] == fir_buf [i], to avoid wrap checking in FIR code */
+    int32_t* ptr;
+    /* FIR history is interleaved with guard to eliminate wrap checking
+     * when convolving.
+     * |LL|RR|LL|RR|LL|RR|LL|RR|LL|RR|LL|RR|LL|RR|LL|RR|...
+     * |--|--|--|--|--|--|--|--|--|--|--|--|--|--|--|--| */
+    /* copy of echo FIR constants as int32_t, for faster access */
+    int32_t coeff [VOICE_COUNT]; 
+};
+#endif /* SPC_NOECHO */
diff --git a/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv6.c b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv6.c
new file mode 100644
index 0000000..2e3de87
--- /dev/null
+++ b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv6.c
@@ -0,0 +1,244 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2010 Michael Sevakis (jhMikeS)
+ *
+ * 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.
+ *
+ ****************************************************************************/
+#if !SPC_NOINTERP
+
+#define SPC_GAUSSIAN_FAST_INTERP
+static inline int gaussian_fast_interp( int16_t const* samples,
+                                        int32_t position,
+                                        int16_t const* fwd,
+                                        int16_t const* rev )
+{
+    int output;
+    int t0, t1, t2, t3;
+
+    asm volatile (
+    /* NOTE: often-unaligned accesses */
+    "ldr      %[t0], [%[samp]]             \n" /* t0=i0i1          */
+    "ldr      %[t2], [%[fwd]]              \n" /* t2=f0f1          */
+    "ldr      %[t1], [%[samp], #4]         \n" /* t1=i2i3          */
+    "ldr      %[t3], [%[rev]]              \n" /* t3=r0r1          */
+    "smuad    %[out], %[t0], %[t2]         \n" /* out=f0*i0+f1*i1  */
+    "smladx   %[out], %[t1], %[t3], %[out] \n" /* out+=r1*i2+r0*i3 */
+    : [out]"=r"(output),
+      [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=r"(t3)
+    : [fwd]"r"(fwd), [rev]"r"(rev),
+      [samp]"r"(samples + (position >> 12)));
+
+    return output;
+}
+
+#define SPC_GAUSSIAN_FAST_AMP
+static inline int gaussian_fast_amp( struct voice_t* voice, int output,
+                                     int* amp_0, int* amp_1 )
+{
+    int t0;
+
+    asm volatile (
+    "mov      %[t0], %[out], asr #(11-5) \n" /* To do >> 16 below */
+    "mul      %[out], %[t0], %[envx]     \n"
+    : [out]"+r"(output), [t0]"=&r"(t0)
+    : [envx]"r"((int) voice->envx));
+
+    asm volatile (
+    "smulwb   %[a0], %[out], %[v0]       \n" /* amp * vol >> 16 */
+    "smulwb   %[a1], %[out], %[v1]       \n"
+    : [a0]"=&r"(*amp_0), [a1]"=r"(*amp_1)
+    : [out]"r"(output),
+      [v0]"r"(voice->volume [0]),
+      [v1]"r"(voice->volume [1]));
+
+    return output >> 5; /* 'output' still 5 bits too big */
+}
+
+#define SPC_GAUSSIAN_SLOW_INTERP
+static inline int gaussian_slow_interp( int16_t const* samples,
+                                        int32_t position,
+                                        int16_t const* fwd,
+                                        int16_t const* rev )
+{
+    int output;
+    int t0, t1, t2, t3;
+
+    asm volatile (
+    /* NOTE: often-unaligned accesses */
+    "ldr      %[t0], [%[samp]]              \n" /* t0=i0i1   */
+    "ldr      %[t2], [%[fwd]]               \n" /* t2=f0f1   */
+    "ldr      %[t1], [%[samp], #4]          \n" /* t1=i2i3   */
+    "ldr      %[t3], [%[rev]]               \n" /* t3=f2f3   */
+    "smulbb   %[out], %[t0], %[t2]          \n" /* out=f0*i0 */
+    "smultt   %[t0],  %[t0], %[t2]          \n" /* t0=f1*i1  */
+    "smulbt   %[t2],  %[t1], %[t3]          \n" /* t2=r1*i2  */
+    "smultb   %[t3],  %[t1], %[t3]          \n" /* t3=r0*i3  */
+    : [out]"=r"(output),
+      [t0]"=&r"(t0), [t1]"=&r"(t1), [t2]"=&r"(t2), [t3]"=r"(t3)
+    : [fwd]"r"(fwd), [rev]"r"(rev),
+      [samp]"r"(samples + (position >> 12)));
+
+    asm volatile (
+    "mov      %[out], %[out], asr #12       \n"
+    "add      %[t0], %[out], %[t0], asr #12 \n"
+    "add      %[t2], %[t0], %[t2], asr #12  \n"
+    "pkhbt    %[t0], %[t2], %[t3], asl #4   \n" /* t3[31:16], t2[15:0] */
+    "sadd16   %[t0], %[t0], %[t0]           \n" /* t3[31:16]*2, t2[15:0]*2 */
+    "qsubaddx %[out], %[t0], %[t0]          \n" /* out[15:0]=
+                                                 * sat16(t3[31:16]+t2[15:0]) */
+    : [out]"+r"(output),
+      [t0]"+r"(t0), [t2]"+r"(t2), [t3]"+r"(t3));
+
+    /* output will be sign-extended in next step */
+    return output;
+}
+
+#define SPC_GAUSSIAN_SLOW_AMP
+static inline int gaussian_slow_amp( struct voice_t* voice, int output,
+                                     int* amp_0, int* amp_1 )
+{
+    asm volatile (
+    "smulbb   %[out], %[out], %[envx]"
+    : [out]"+r"(output)
+    : [envx]"r"(voice->envx));
+
+    asm volatile (
+    "mov      %[out], %[out], asr #11 \n"
+    "bic      %[out], %[out], #0x1    \n"
+    "smulbb   %[amp_0], %[out], %[v0] \n"
+    "smulbb   %[amp_1], %[out], %[v1] \n"
+    : [out]"+r"(output),
+      [amp_0]"=&r"(*amp_0), [amp_1]"=r"(*amp_1)
+    : [v0]"r"(voice->volume[0]), [v1]"r"(voice->volume[1]));
+
+    return output;
+}
+
+#endif /* !SPC_NOINTERP */
+
+#if !SPC_NOECHO
+
+#define SPC_DSP_ECHO_APPLY
+
+/* Echo filter history */
+static int32_t fir_buf[FIR_BUF_CNT] IBSS_ATTR_SPC
+    __attribute__(( aligned(FIR_BUF_ALIGN*1) ));
+
+static inline void echo_init( struct Spc_Dsp* this )
+{
+    this->fir.ptr = fir_buf;
+    ci->memset( fir_buf, 0, sizeof fir_buf );
+}
+
+static inline void echo_apply(struct Spc_Dsp* this,
+        uint8_t* const echo_ptr, int* out_0, int* out_1)
+{
+    /* Keep last 8 samples */
+    int32_t* fir_ptr;
+    int t0;
+    asm volatile (
+    "ldr      %[t0], [%[ep]]              \n"
+    "add      %[p], %[t_p], #4            \n"
+    "bic      %[t_p], %[p], %[mask]       \n"
+    "str      %[t0], [%[p], #-4]          \n"
+    /* duplicate at +8 eliminates wrap checking below */
+    "str    %[t0], [%[p], #28]            \n"
+    : [p]"=&r"(fir_ptr), [t_p]"+r"(this->fir.ptr),
+      [t0]"=&r"(t0)
+    : [ep]"r"(echo_ptr), [mask]"i"(~FIR_BUF_MASK));
+
+    int32_t* fir_coeff = (int32_t *)this->fir.coeff;
+
+    asm volatile (                          /* L0R0 = acc0          */
+    "ldmia    %[p]!, { r2-r5 }            \n" /* L1R1-L4R4 = r2-r5    */
+    "ldmia    %[c]!, { r0-r1 }            \n" /* C0C1-C2C3 = r0-r1    */
+    "pkhbt    %[acc0], %[t0], r2, asl #16 \n" /* L0R0,L1R1->L0L1,R0R1 */
+    "pkhtb    r2, r2, %[t0], asr #16      \n"
+    "smuad    %[acc0], %[acc0], r0        \n" /* acc0=L0*C0+L1*C1     */
+    "smuad    %[acc1], r2, r0             \n" /* acc1=R0*C0+R1*C1     */
+    "pkhbt    %[t0], r3, r4, asl #16      \n" /* L2R2,L3R3->L2L3,R2R3 */
+    "pkhtb    r4, r4, r3, asr #16         \n"
+    "smlad    %[acc0], %[t0], r1, %[acc0] \n" /* acc0+=L2*C2+L3*C3    */
+    "smlad    %[acc1], r4, r1, %[acc1]    \n" /* acc1+=R2*C2+R3*C3    */
+    "ldmia    %[p], { r2-r4 }             \n" /* L5R5-L7R7 = r2-r4    */
+    "ldmia    %[c], { r0-r1 }             \n" /* C4C5-C6C7 = r0-r1    */
+    "pkhbt    %[t0], r5, r2, asl #16      \n" /* L4R4,L5R5->L4L5,R4R5 */
+    "pkhtb    r2, r2, r5, asr #16         \n"
+    "smlad    %[acc0], %[t0], r0, %[acc0] \n" /* acc0+=L4*C4+L5*C5    */
+    "smlad    %[acc1], r2, r0, %[acc1]    \n" /* acc1+=R4*C4+R5*C5    */
+    "pkhbt    %[t0], r3, r4, asl #16      \n" /* L6R6,L7R7->L6L7,R6R7 */
+    "pkhtb    r4, r4, r3, asr #16         \n"
+    "smlad    %[acc0], %[t0], r1, %[acc0] \n" /* acc0+=L6*C6+L7*C7    */
+    "smlad    %[acc1], r4, r1, %[acc1]    \n" /* acc1+=R6*C6+R7*C7    */
+    : [t0]"+r"(t0), [acc0]"=&r"(*out_0), [acc1]"=&r"(*out_1),
+      [p]"+r"(fir_ptr), [c]"+r"(fir_coeff)
+    :
+    : "r0", "r1", "r2", "r3", "r4", "r5");
+}
+
+#define SPC_DSP_ECHO_FEEDBACK
+static inline void echo_feedback(struct Spc_Dsp* this, uint8_t* echo_ptr,
+                                 int echo_0, int echo_1, int fb_0, int fb_1)
+{
+    int e0, e1;
+    asm volatile (
+    "mov      %[e0], %[ei0], asl #7        \n"
+    "mov      %[e1], %[ei1], asl #7        \n"
+    "mla      %[e0], %[fb0], %[ef], %[e0]  \n"
+    "mla      %[e1], %[fb1], %[ef], %[e1]  \n"
+    : [e0]"=&r"(e0), [e1]"=&r"(e1)
+    : [ei0]"r"(echo_0), [ei1]"r"(echo_1),
+      [fb0]"r"(fb_0), [fb1]"r"(fb_1),
+      [ef]"r"((int)this->r.g.echo_feedback));
+    asm volatile (
+    "ssat     %[e0], #16, %[e0], asr #14   \n"
+    "ssat     %[e1], #16, %[e1], asr #14   \n"
+    "pkhbt    %[e0], %[e0], %[e1], lsl #16 \n"
+    "str      %[e0], [%[ep]]               \n"
+    : [e0]"+r"(e0), [e1]"+r"(e1)
+    : [ep]"r"((int32_t *)echo_ptr));
+}
+
+#define SPC_DSP_GENERATE_OUTPUT
+static inline void echo_output( struct Spc_Dsp* this, int global_muting,
+    int global_vol_0, int global_vol_1, int chans_0, int chans_1, 
+    int fb_0, int fb_1, int* out_0, int* out_1 )
+{
+    int t0, t1;
+
+    asm volatile (
+    "mul      %[t0], %[gv0], %[ch0]       \n"
+    "mul      %[t1], %[gv1], %[ch1]       \n"
+    : [t0]"=&r"(t0), [t1]"=r"(t1)
+    : [gv0]"r"(global_vol_0), [gv1]"r"(global_vol_1),
+      [ch0]"r"(chans_0), [ch1]"r"(chans_1));
+    asm volatile (
+    "mla      %[t0], %[i0], %[ev0], %[t0] \n"
+    "mla      %[t1], %[i1], %[ev1], %[t1] \n"
+    : [t0]"+r"(t0), [t1]"+r"(t1)
+    : [i0]"r"(fb_0), [i1]"r"(fb_1),
+      [ev0]"r"((int)this->r.g.echo_volume_0),
+      [ev1]"r"((int)this->r.g.echo_volume_1));
+    asm volatile (
+    "mov      %[o0], %[t0], asr %[gm]     \n"
+    "mov      %[o1], %[t1], asr %[gm]     \n"
+    : [o0]"=&r"(*out_0), [o1]"=r"(*out_1)
+    : [t0]"r"(t0), [t1]"r"(t1), 
+      [gm]"r"(global_muting));
+}
+
+#endif /* SPC_NOECHO */
diff --git a/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv6.h b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv6.h
new file mode 100644
index 0000000..a36d816
--- /dev/null
+++ b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_armv6.h
@@ -0,0 +1,45 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2010 Michael Sevakis (jhMikeS)
+ *
+ * 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.
+ *
+ ****************************************************************************/
+#if !SPC_NOECHO
+
+#define SPC_DSP_ECHO_APPLY
+
+enum
+{
+    FIR_BUF_CNT   = FIR_BUF_HALF * 2,
+    FIR_BUF_SIZE  = FIR_BUF_CNT * sizeof ( int32_t ),
+    FIR_BUF_ALIGN = FIR_BUF_SIZE,
+    FIR_BUF_MASK  = ~((FIR_BUF_ALIGN / 2) | (sizeof ( int32_t ) - 1))
+};
+
+/* Echo filter structure embedded in struct Spc_Dsp */
+struct echo_filter
+{   /* fir_buf [i + 8] == fir_buf [i], to avoid wrap checking in FIR code */
+    int32_t* ptr;
+    /* FIR history is interleaved with guard to eliminate wrap checking
+     * when convolving.
+     * |LR|LR|LR|LR|LR|LR|LR|LR|--|--|--|--|--|--|--|--| */
+    /* copy of echo FIR constants as int16_t, loaded as int32 for
+     * halfword, packed multiples */
+    int16_t coeff [VOICE_COUNT];
+};
+
+#endif /* SPC_NOECHO */
diff --git a/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c
new file mode 100644
index 0000000..b0d14d1
--- /dev/null
+++ b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c
@@ -0,0 +1,198 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2007 Michael Sevakis (jhMikeS)
+ *
+ * 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.
+ *
+ ****************************************************************************/
+#if SPC_NOINTERP
+
+#define SPC_LINEAR_INTERP
+static inline int linear_interp( int16_t const* samples, int32_t position )
+{
+    uint32_t f = position;
+    int32_t y0, y1;
+
+    /**
+     * output = y0 + f*y1 - f*y0
+     */
+    asm volatile (
+    "move.l   %[f], %[y1]               \n" /* separate frac and whole */
+    "and.l    #0xfff, %[f]              \n"
+    "asr.l    %[sh], %[y1]              \n"
+    "move.l   2(%[s], %[y1].l*2), %[y1] \n" /* y0=upper, y1=lower  */
+    "mac.w    %[f]l, %[y1]l, %%acc0     \n" /* %acc0 = f*y1 */
+    "msac.w   %[f]l, %[y1]u, %%acc0     \n" /* %acc0 -= f*y0 */
+    "swap     %[y1]                     \n" /* separate out y0 and sign extend */
+    "movea.w  %[y1], %[y0]              \n"
+    "movclr.l %%acc0, %[y1]             \n" /* fetch, scale down, add y0 */
+    "asr.l    %[sh], %[y1]              \n" /* output = y0 + (result >> 12) */
+    "add.l    %[y0], %[y1]              \n"
+    : [f]"+d"(f), [y0]"=&a"(y0), [y1]"=&d"(y1)
+    : [s]"a"(samples), [sh]"d"(12));
+
+    return y1;
+}
+
+#define SPC_LINEAR_AMP
+static inline int linear_amp( struct voice_t* voice, int output,
+                              int* amp_0, int* amp_1 )
+{
+    asm volatile (
+    "mac.w %[out]l, %[envx]l, %%acc0"
+    :
+    : [out]"r"(output), [envx]"r"(voice->envx));
+    asm volatile (
+    "movclr.l %%acc0, %[out]        \n"
+    "asr.l    #8, %[out]            \n"
+    "mac.l    %[v0], %[out], %%acc0 \n"
+    "mac.l    %[v1], %[out], %%acc1 \n"
+    "asr.l    #3, %[out]            \n"
+    : [out]"+r"(output)
+    : [v0]"r"((int) voice->volume [0]),
+      [v1]"r"((int) voice->volume [1]));
+    asm volatile (
+    "movclr.l %%acc0, %[a0]         \n"
+    "asr.l    #3, %[a0]             \n"
+    "movclr.l %%acc1, %[a1]         \n"
+    "asr.l    #3, %[a1]             \n"
+    : [a0]"=d"(*amp_0), [a1]"=d"(*amp_1));
+
+    return output;
+}
+
+#endif /* SPC_NOINTERP */
+
+
+#if !SPC_NOECHO
+
+#define SPC_DSP_ECHO_APPLY
+
+/* Echo filter history */
+static int32_t fir_buf[FIR_BUF_CNT] IBSS_ATTR_SPC
+    __attribute__(( aligned(FIR_BUF_ALIGN*1) ));
+
+static inline void echo_init( struct Spc_Dsp* this )
+{
+    /* Initialize mask register with the buffer address mask */
+    asm volatile ("move.l %0, %%mask" : : "i"(FIR_BUF_MASK));
+    this->fir.ptr = fir_buf;
+    this->fir.hist_ptr = &fir_buf [1];
+    ci->memset( fir_buf, 0, sizeof fir_buf );
+}
+
+static inline void echo_apply( struct Spc_Dsp* this, uint8_t* echo_ptr,
+                               int* out_0, int* out_1 )
+{
+    int t0, t1, t2;
+
+    t1 = swap_odd_even32( *(int32_t *)echo_ptr );
+
+    /* Keep last 8 samples */
+    *this->fir.ptr = t1;
+    this->fir.ptr = this->fir.hist_ptr;
+
+    asm volatile (
+    "move.l                         (%[c])  , %[t2]         \n"
+    "mac.w    %[t1]u, %[t2]u, <<,   (%[p])+&, %[t0], %%acc0 \n"
+    "mac.w    %[t1]l, %[t2]u, <<,   (%[p])& , %[t1], %%acc1 \n"
+    "mac.w    %[t0]u, %[t2]l, <<                   , %%acc0 \n"
+    "mac.w    %[t0]l, %[t2]l, <<,  4(%[c])  , %[t2], %%acc1 \n"
+    "mac.w    %[t1]u, %[t2]u, <<,  4(%[p])& , %[t0], %%acc0 \n"
+    "mac.w    %[t1]l, %[t2]u, <<,  8(%[p])& , %[t1], %%acc1 \n"
+    "mac.w    %[t0]u, %[t2]l, <<                   , %%acc0 \n"
+    "mac.w    %[t0]l, %[t2]l, <<,  8(%[c])  , %[t2], %%acc1 \n"
+    "mac.w    %[t1]u, %[t2]u, <<, 12(%[p])& , %[t0], %%acc0 \n"
+    "mac.w    %[t1]l, %[t2]u, <<, 16(%[p])& , %[t1], %%acc1 \n"
+    "mac.w    %[t0]u, %[t2]l, <<                   , %%acc0 \n"
+    "mac.w    %[t0]l, %[t2]l, <<, 12(%[c])  , %[t2], %%acc1 \n"
+    "mac.w    %[t1]u, %[t2]u, <<, 20(%[p])& , %[t0], %%acc0 \n"
+    "mac.w    %[t1]l, %[t2]u, <<                   , %%acc1 \n"
+    "mac.w    %[t0]u, %[t2]l, <<                   , %%acc0 \n"
+    "mac.w    %[t0]l, %[t2]l, <<                   , %%acc1 \n"
+    : [t0]"=&r"(t0), [t1]"+r"(t1), [t2]"=&r"(t2),
+      [p]"+a"(this->fir.hist_ptr)
+    : [c]"a"(this->fir.coeff));
+    asm volatile (
+    "movclr.l %%acc0, %[o0]             \n"
+    "movclr.l %%acc1, %[o1]             \n"
+    "mac.l    %[ev0], %[o0], >>, %%acc2 \n" /* echo volume */
+    "mac.l    %[ev1], %[o1], >>, %%acc3 \n"
+    : [o0]"=&r"(*out_0), [o1]"=&r"(*out_1)
+    : [ev0]"r"((int) this->r.g.echo_volume_0),
+      [ev1]"r"((int) this->r.g.echo_volume_1));
+}
+
+#define SPC_DSP_ECHO_FEEDBACK
+static inline void echo_feedback( struct Spc_Dsp* this, uint8_t* echo_ptr,
+                                  int echo_0, int echo_1, int fb_0, int fb_1 )
+{
+    asm volatile (
+    /* scale echo voices; saturate if overflow */
+    "mac.l    %[sh], %[e1]     , %%acc1 \n"
+    "mac.l    %[sh], %[e0]     , %%acc0 \n"
+    /* add scaled output from FIR filter */
+    "mac.l    %[fb1], %[ef], <<, %%acc1 \n"
+    "mac.l    %[fb0], %[ef], <<, %%acc0 \n"
+    :
+    : [e0]"d"(echo_0), [e1]"d"(echo_1),
+      [fb0]"r"(fb_0), [fb1]"r"(fb_1),
+      [ef]"r"((int)this->r.g.echo_feedback),
+      [sh]"r"(1 << 9));
+    /* swap and fetch feedback results */
+    int t0;
+    asm volatile(
+    "move.l   #0x00ff00ff, %[t0]        \n"
+    "movclr.l %%acc1, %[e1]             \n"
+    "swap.w   %[e1]                     \n"
+    "movclr.l %%acc0, %[e0]             \n"
+    "move.w   %[e1], %[e0]              \n"
+    "and.l    %[e0], %[t0]              \n"
+    "eor.l    %[t0], %[e0]              \n"
+    "lsl.l    #8, %[t0]                 \n"
+    "lsr.l    #8, %[e0]                 \n"
+    "or.l     %[e0], %[t0]              \n"
+    : [e0]"=&d"(echo_0), [e1]"=&d"(echo_1),
+      [t0]"=&d"(t0));
+
+    /* save final feedback into echo buffer */
+    *(int32_t *)echo_ptr = t0;
+}
+
+#define SPC_DSP_GENERATE_OUTPUT
+static inline void echo_output( struct Spc_Dsp* this, int global_muting,
+    int global_vol_0, int global_vol_1, int chans_0, int chans_1, 
+    int fb_0, int fb_1, int* out_0, int* out_1 )
+{
+    asm volatile (
+    "mac.l    %[ch0], %[gv0], %%acc2 \n" /* global volume */
+    "mac.l    %[ch1], %[gv1], %%acc3 \n"
+    :
+    : [ch0]"r"(chans_0), [gv0]"r"(global_vol_0),
+      [ch1]"r"(chans_1), [gv1]"r"(global_vol_1));
+    asm volatile (
+    "movclr.l %%acc2, %[a0]          \n" /* fetch mixed output */
+    "movclr.l %%acc3, %[a1]          \n"
+    "asr.l    %[gm],  %[a0]          \n" /* scale by global_muting shift */
+    "asr.l    %[gm],  %[a1]          \n"
+    : [a0]"=&d"(*out_0), [a1]"=&d"(*out_1)
+    : [gm]"d"(global_muting));
+
+    /* scaled echo is stored in %acc2 and %acc3 */
+    (void)this; (void)fb_0; (void)fb_1;
+}
+
+#endif /* !SPC_NOECHO */
diff --git a/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.h b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.h
new file mode 100644
index 0000000..f9aafab
--- /dev/null
+++ b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.h
@@ -0,0 +1,45 @@
+/***************************************************************************
+ *             __________               __   ___.
+ *   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
+ *   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
+ *   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
+ *   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
+ *                     \/            \/     \/    \/            \/
+ * $Id$
+ *
+ * Copyright (C) 2007 Michael Sevakis (jhMikeS)
+ *
+ * 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.
+ *
+ ****************************************************************************/
+#if !SPC_NOECHO
+
+#define SPC_DSP_ECHO_APPLY
+
+enum
+{
+    FIR_BUF_CNT   = FIR_BUF_HALF,
+    FIR_BUF_SIZE  = FIR_BUF_CNT * sizeof ( int32_t ),
+    FIR_BUF_ALIGN = FIR_BUF_SIZE * 2,
+    FIR_BUF_MASK  = ~((FIR_BUF_ALIGN / 2) | (sizeof ( int32_t ) - 1))
+};
+
+/* Echo filter structure embedded in struct Spc_Dsp */
+struct echo_filter
+{
+    /* FIR history is interleaved. Hardware handles wrapping by mask.
+     * |LR|LR|LR|LR|LR|LR|LR|LR| */
+    int32_t* ptr;
+    /* wrapped address just behind current position -
+       allows mac.w to increment and mask ptr */
+    int32_t* hist_ptr;
+    /* copy of echo FIR constants as int16_t for use with mac.w */
+    int16_t coeff [VOICE_COUNT];
+};
+#endif /* !SPC_NOECHO */