Commit 0cb7af79 authored by jm's avatar jm

Blackfin assembly optimizations.


git-svn-id: http://svn.xiph.org/trunk/speex@9357 0101bb08-14d6-0310-b084-bc0e0c8e3800
parent b5bc6a8e
Copyright 2002-2004 Xiph.org Foundation, Jean-Marc Valin, David Rowe, EpicGames
Copyright 2002-2005
Xiph.org Foundation
Jean-Marc Valin
David Rowe
EpicGames
Analog Devices
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
......
......@@ -54,25 +54,30 @@ noinst_HEADERS = lsp.h \
filters.h \
stack_alloc.h \
vq.h \
vq_arm4.h \
vq_sse.h \
vq_arm4.h \
vq_bfin.h \
modes.h \
sb_celp.h \
vbr.h \
misc.h \
ltp_sse.h \
ltp_arm4.h \
ltp_bfin.h \
filters_sse.h \
filters_arm4.h \
filters_bfin.h \
math_approx.h \
smallft.h \
arch.h \
fixed_arm4.h \
fixed_arm5e.h \
fixed_bfin.h \
fixed_debug.h \
fixed_generic.h \
cb_search_sse.h \
cb_search_arm4.h
cb_search_arm4.h \
cb_search_bfin.h
libspeex_la_LDFLAGS = -version-info @SPEEX_LT_CURRENT@:@SPEEX_LT_REVISION@:@SPEEX_LT_AGE@
......
/* Copyright (C) 2005 Analog Devices
Author: Jean-Marc Valin
File: cb_search_bfin.h
Fixed codebook functions (Blackfin 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.
*/
#define OVERRIDE_COMPUTE_WEIGHTED_CODEBOOK
void compute_weighted_codebook(const signed char *shape_cb, const spx_word16_t *r, spx_word16_t *resp, spx_word16_t *resp2, spx_word32_t *E, int shape_cb_size, int subvect_size, char *stack)
{
int i;
for (i=0;i<shape_cb_size;i++)
{
__asm__ __volatile__ (
"P0 = %0;\n\t"
"LC0 = P0;\n\t"
"P1 = %1;\n\t"
"P2 = %2;\n\t"
"P3 = %3;\n\t"
"P0 = 1;\n\t"
"L0 = 0;\n\t"
"L1 = 0;\n\t"
"R2 = 0;\n\t"
"A1 = 0;\n\t"
"LOOP outter%= LC0;\n\t"
"LOOP_BEGIN outter%=;\n\t"
"A0 = 0;\n\t"
"P4 = P1;\n\t"
"I1 = P2;\n\t"
"R0 = B[P4++] (X) || R1.L = W[I1--];\n\t"
"LOOP inner%= LC1 = P0;\n\t"
"LOOP_BEGIN inner%=;\n\t"
"LOOP_END inner%=;\n\t"
"A0 += R0.L*R1.L (IS) || R0 = B[P4++] (X) || R1.L = W[I1--];\n\t" /* Is there a danger of reading too far? */
"R0 = A0;\n\t"
"R0 >>>= 11;\n\t"
"A1 += R0.L*R0.L (IS);\n\t"
"W[P3++] = R0;\n\t"
"P0 += 1;\n\t"
"LOOP_END outter%=;\n\t"
"P2 += 2;\n\t"
"P4 = %4;\n\t"
"R1 = A1;\n\t"
"[P4] = R1;\n\t"
:
: "m" (subvect_size), "m" (shape_cb), "m" (r), "m" (resp), "m" (E)
: "P0", "P1", "P2", "P3", "P4", "R0", "R1", "R2", "R3", "R4", "R5", "R6", "R7", "I0", "I1", "L0", "L1", "B0", "B1", "A0", "A1", "memory"
);
shape_cb += subvect_size;
resp += subvect_size;
E++;
}
}
/* Copyright (C) 2005 Analog Devices
Author: Jean-Marc Valin
File: filters_bfin.h
Various analysis/synthesis filters (Blackfin 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.
*/
#include <stdio.h>
#define OVERRIDE_NORMALIZE16
int normalize16(const spx_sig_t *x, spx_word16_t *y, spx_sig_t max_scale, int len)
{
spx_sig_t max_val=1;
int sig_shift;
__asm__
(
"%0 = 0;\n\t"
"I0 = %1;\n\t"
"L0 = 0;\n\t"
"LOOP norm_max%= LC0 = %2;\n\t"
"LOOP_BEGIN norm_max%=;\n\t"
"R1 = [I0++];\n\t"
"R1 = ABS R1;\n\t"
"LOOP_END norm_max%=;\n\t"
"%0 = MAX(%0, R1);\n\t"
: "=d" (max_val)
: "a" (x), "a" (len)
: "R1"
);
sig_shift=0;
while (max_val>max_scale)
{
sig_shift++;
max_val >>= 1;
}
__asm__ __volatile__
(
"I0 = %0;\n\t"
"L0 = 0;\n\t"
"I1 = %1;\n\t"
"L1 = 0;\n\t"
"R0 = [I0++];\n\t"
"LOOP norm_shift%= LC0 = %3 >> 1;\n\t"
"LOOP_BEGIN norm_shift%=;\n\t"
"R1 = ASHIFT R0 by %2.L || R2 = [I0++];\n\t"
"R3 = ASHIFT R2 by %2.L || R0 = [I0++];\n\t"
"R3 = PACK(R3.L, R1.L);\n\t"
"LOOP_END norm_shift%=;\n\t"
"[I1++] = R3;\n\t"
: : "a" (x), "a" (y), "d" (-sig_shift), "a" (len)
: "I0", "L0", "I1", "L1", "R0", "R1", "R2", "R3", "memory"
);
return sig_shift;
}
#define OVERRIDE_FILTER_MEM2
void filter_mem2(const spx_sig_t *_x, const spx_coef_t *num, const spx_coef_t *den, spx_sig_t *_y, int N, int ord, spx_mem_t *mem)
{
int i,j;
spx_word16_t x[N],y[N];
spx_word16_t *xx, *yy;
xx = x;
yy = y;
__asm__ __volatile__
(
"R0 = %7;\n\t" /*ord */
"P0 = %4;\n\t"
"I0 = P0;\n\t"
"B0 = P0;\n\t"
"L0 = 0;\n\t"
"P1 = %5;\n\t"
"I1 = P1;\n\t"
"B1 = P1;\n\t"
"L1 = 0;\n\t"
"P2 = %0;\n\t"
"P3 = %1;\n\t"
"I2 = P2;\n\t"
"I3 = P3;\n\t"
"L2 = 0;\n\t"
"L3 = 0;\n\t"
"P4 = %8;\n\t"
"P0 = %2;\n\t"
"P1 = %3;\n\t"
"R1 = [P4++];\n\t"
"R1 <<= 1;\n\t"
"R2 = [P0++];\n\t"
"R1 = R1 + R2;\n\t"
"[P1++] = R1;\n\t"
"R1 <<= 2;\n\t"
"W[P3] = R1.H;\n\t"
"R2 <<= 2;\n\t"
"W[P2] = R2.H;\n\t"
"R0 += -1;\n\t"
"R3 = 0;\n\t"
"LC0 = R0;\n\t"
"LOOP filter_start%= LC0;\n\t"
"LOOP_BEGIN filter_start%=;\n\t"
"R3 += 1;\n\t"
"LC1 = R3;\n\t"
"R1 = [P4++];\n\t"
"A1 = R1;\n\t"
"I0 = B0;\n\t"
"I1 = B1;\n\t"
"I2 = P2;\n\t"
"I3 = P3;\n\t"
"P2 += 2;\n\t"
"P3 += 2;\n\t"
"LOOP filter_start_inner%= LC1;\n\t"
"LOOP_BEGIN filter_start_inner%=;\n\t"
"R4.L = W[I0++];\n\t"
"R5.L = W[I2--];\n\t"
"A1 += R4.L*R5.L (IS);\n\t"
"R4.L = W[I1++];\n\t"
"R5.L = W[I3--];\n\t"
"A1 -= R4.L*R5.L (IS);\n\t"
"LOOP_END filter_start_inner%=;\n\t"
"nop;nop;nop;nop;\n\t"
"R1 = A1;\n\t"
"R1 <<= 1;\n\t"
"R2 = [P0++];\n\t"
"R1 = R1 + R2;\n\t"
"[P1++] = R1;\n\t"
"R1 <<= 2;\n\t"
"W[P3] = R1.H;\n\t"
"R2 <<= 2;\n\t"
"W[P2] = R2.H;\n\t"
"LOOP_END filter_start%=;\n\t"
"nop;nop;nop;nop;\n\t"
"R0 = %7;\n\t"
"R0 <<= 1;\n\t"
"I0 = B0;\n\t"
"I1 = B1;\n\t"
"L0 = R0;\n\t"
"L1 = R0;\n\t"
"R0 = %7;\n\t"
"R2 = %6;\n\t"
"R2 = R2 - R0;\n\t"
"R4.L = W[I0++];\n\t"
"R6.L = W[I1++];\n\t"
"LC0 = R2;\n\t"
"LOOP filter_mid%= LC0;\n\t"
"LOOP_BEGIN filter_mid%=;\n\t"
"LC1 = R0;\n\t"
"A1 = 0;\n\t"
"I2 = P2;\n\t"
"I3 = P3;\n\t"
"P2 += 2;\n\t"
"P3 += 2;\n\t"
"R5.L = W[I2--];\n\t"
"R7.L = W[I3--];\n\t"
"LOOP filter_mid_inner%= LC1;\n\t"
"LOOP_BEGIN filter_mid_inner%=;\n\t"
"A1 += R4.L*R5.L (IS) || R4.L = W[I0++] || R5.L = W[I2--];\n\t"
"LOOP_END filter_mid_inner%=;\n\t"
"A1 -= R6.L*R7.L (IS) || R6.L = W[I1++] || R7.L = W[I3--];\n\t"
"R1 = A1;\n\t"
"R1 <<= 1;\n\t"
"R2 = [P0++];\n\t"
"R1 = R1 + R2;\n\t"
"[P1++] = R1;\n\t"
"R1 <<= 2;\n\t"
"W[P3] = R1.H;\n\t"
"R2 <<= 2;\n\t"
"W[P2] = R2.H;\n\t"
"LOOP_END filter_mid%=;\n\t"
"nop;nop;nop;nop;\n\t"
: : "m" (xx), "m" (yy), "m" (_x), "m" (_y), "m" (num), "m" (den), "m" (N), "m" (ord), "m" (mem)
: "R0", "R1", "R2", "R3", "R4", "R5", "R7", "P0", "P1", "P2", "P3", "P4", "B0", "B1", "I0", "I1", "I2", "I3", "L0", "L1", "L2", "L3", "memory"
);
for (i=0;i<ord;i++)
{
spx_mem_t m = 0;
for (j=0;j<ord-i;j++)
{
m = MAC16_16(m, x[N-1-j], num[j+i]);
m = MAC16_16(m, -y[N-1-j], den[j+i]);
}
mem[i] = m;
}
}
#define OVERRIDE_IIR_MEM2
void iir_mem2(const spx_sig_t *x, const spx_coef_t *den, spx_sig_t *y, int N, int ord, spx_mem_t *mem)
{
int i;
spx_coef_t num2[12];
spx_coef_t *num;
num = (spx_coef_t*)((((int)num2)+4)&0xfffffffc);
for (i=0;i<10;i++)
num[i] = 0;
filter_mem2(x, num, den, y, N, ord, mem);
}
#define OVERRIDE_FIR_MEM2
void fir_mem2(const spx_sig_t *x, const spx_coef_t *num, spx_sig_t *y, int N, int ord, spx_mem_t *mem)
{
int i;
spx_coef_t den2[12];
spx_coef_t *den;
den = (spx_coef_t*)((((int)den2)+4)&0xfffffffc);
for (i=0;i<10;i++)
den[i] = 0;
filter_mem2(x, num, den, y, N, ord, mem);
}
#define min(a,b) ((a)<(b) ? (a):(b))
#define OVERRIDE_COMPUTE_IMPULSE_RESPONSE
void compute_impulse_response(const spx_coef_t *ak, const spx_coef_t *awk1, const spx_coef_t *awk2, spx_word16_t *y, int N, int ord, char *stack)
{
int i,j;
VARDECL(spx_word16_t *ytmp);
ALLOC(ytmp, N, spx_word16_t);
y[0] = LPC_SCALING;
for (i=0;i<ord;i++)
y[i+1] = awk1[i];
i++;
for (;i<N;i++)
y[i] = 0;
for (i=0;i<N;i++)
{
spx_word32_t yi = SHL32(EXTEND32(y[i]),LPC_SHIFT);
spx_word32_t yi2 = 0;
for (j=0;j<min(i,ord);j++)
{
yi = MAC16_16(yi, awk2[j], -ytmp[i-j-1]);
yi2 = MAC16_16(yi2, ak[j], -y[i-j-1]);
}
ytmp[i] = EXTRACT16(SHR32(yi,LPC_SHIFT));
yi2 = ADD32(yi2,SHL32(yi,1));
y[i] = EXTRACT16(SHR32(yi2,LPC_SHIFT));
}
}
#if 0 /* Equivalent C function for filter_mem2 */
void filter_mem2(const spx_sig_t *_x, const spx_coef_t *num, const spx_coef_t *den, spx_sig_t *_y, int N, int ord, spx_mem_t *mem)
{
int i,j;
spx_word16_t xi,yi,nyi;
spx_word16_t x[N],y[N];
spx_word16_t *xx, *yy;
xx = x;
yy = y;
for (i=0;i<N;i++)
{
x[i] = EXTRACT16(SHR32(_x[i],SIG_SHIFT));
}
for (i=0;i<ord;i++)
{
spx_word32_t yi = mem[i];
for (j=0;j<i;j++)
{
yi = MAC16_16(yi, num[j], x[i-j-1]);
yi = MAC16_16(yi, den[j], -y[i-j-1]);
}
_y[i] = ADD32(_x[i],SHL32(yi,1));
y[i] = EXTRACT16(SHR32(_y[i],SIG_SHIFT));
}
for (i=ord;i<N;i++)
{
spx_word32_t yi = 0;
for (j=0;j<ord;j++)
{
yi = MAC16_16(yi, num[j], x[i-j-1]);
yi = MAC16_16(yi, den[j], -y[i-j-1]);
}
_y[i] = ADD32(_x[i],SHL32(yi,1));
y[i] = EXTRACT16(SHR32(_y[i],SIG_SHIFT));
}
for (i=0;i<ord;i++)
{
spx_mem_t m = 0;
for (j=0;j<ord-i;j++)
{
m = MAC16_16(m, x[N-1-j], num[j+i]);
m = MAC16_16(m, -y[N-1-j], den[j+i]);
}
mem[i] = m;
}
}
#endif
/* Copyright (C) 2005 Analog Devices
Author: Jean-Marc Valin */
/**
@file fixed_bfin.h
@brief Blackfin fixed-point operations
*/
/*
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.
*/
#ifndef FIXED_BFIN_H
#define FIXED_BFIN_H
#undef DIV32_16
static inline spx_word16_t DIV32_16(spx_word32_t a, spx_word16_t b)
{
spx_word32_t res, bb;
bb = b;
__asm__ (
"P0 = 15;\n\t"
"R0 = %1;\n\t"
"R1 = %2;\n\t"
"R0 <<= 1;\n\t"
"DIVS (R0, R1);\n\t"
"LOOP divide%= LC0 = P0;\n\t"
"LOOP_BEGIN divide%=;\n\t"
"LOOP_END divide%=;\n\t"
"DIVQ (R0, R1);\n\t"
"R0 = R0.L;\n\t"
"%0 = R0;\n\t"
: "=m" (res)
: "m" (a), "m" (bb)
: "P0", "R0", "R1", "cc");
return res;
}
#undef MULT16_32_Q15
static inline spx_word32_t MULT16_32_Q15(spx_word16_t a, spx_word32_t b)
{
spx_word32_t res;
__asm__
(
"%1 <<= 1;\n\t"
"A1 = %2.L*%1.L (M,IS);\n\t"
"A1 = A1 >>> 16;\n\t"
"R1 = (A1 += %2.L*%1.H) (IS);\n\t"
"%0 = R1;\n\t"
: "=&d" (res), "=&d" (b)
: "d" (a), "1" (b)
: "A1", "R1"
);
return res;
}
#undef MAC16_32_Q15
static inline spx_word32_t MAC16_32_Q15(spx_word32_t c, spx_word16_t a, spx_word32_t b)
{
spx_word32_t res;
__asm__
(
"%1 <<= 1;\n\t"
"A1 = %2.L*%1.L (M,IS);\n\t"
"A1 = A1 >>> 16;\n\t"
"R1 = (A1 += %2.L*%1.H) (IS);\n\t"
"%0 = R1 + %4;\n\t"
: "=&d" (res), "=&d" (b)
: "d" (a), "1" (b), "d" (c)
: "A1", "R1"
);
return res;
}
#undef MULT16_32_Q14
static inline spx_word32_t MULT16_32_Q14(spx_word16_t a, spx_word32_t b)
{
spx_word32_t res;
__asm__
(
"%2 <<= 2;\n\t"
"A1 = %1.L*%2.L (M,IS);\n\t"
"A1 = A1 >>> 16;\n\t"
"R1 = (A1 += %1.L*%2.H) (IS);\n\t"
"%0 = R1;\n\t"
: "=d" (res), "=d" (a), "=d" (b)
: "1" (a), "2" (b)
: "A1", "R1"
);
return res;
}
#undef MAC16_32_Q14
static inline spx_word32_t MAC16_32_Q14(spx_word32_t c, spx_word16_t a, spx_word32_t b)
{
spx_word32_t res;
__asm__
(
"%1 <<= 2;\n\t"
"A1 = %2.L*%1.L (M,IS);\n\t"
"A1 = A1 >>> 16;\n\t"
"R1 = (A1 += %2.L*%1.H) (IS);\n\t"
"%0 = R1 + %4;\n\t"
: "=&d" (res), "=&d" (b)
: "d" (a), "1" (b), "d" (c)
: "A1", "R1"
);
return res;
}
#endif
/* Copyright (C) 2005 Analog Devices
Author: Jean-Marc Valin
File: ltp_bfin.h
Long-Term Prediction functions (Blackfin 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.
*/
#define OVERRIDE_INNER_PROD
static spx_word32_t inner_prod(const spx_word16_t *x, const spx_word16_t *y, int len)
{
spx_word32_t sum=0;
__asm__ __volatile__ (
"P0 = %3;\n\t"
"P1 = %1;\n\t"
"P2 = %2;\n\t"
"I0 = P1;\n\t"
"I1 = P2;\n\t"
"L0 = 0;\n\t"
"L1 = 0;\n\t"
"A0 = 0;\n\t"
"R0.L = W[I0++] || R1.L = W[I1++];\n\t"
"LOOP inner%= LC0 = P0;\n\t"
"LOOP_BEGIN inner%=;\n\t"
"LOOP_END inner%=;\n\t"
"A0 += R0.L*R1.L (IS) || R0.L = W[I0++] || R1.L = W[I1++];\n\t"
"A0 = A0 >>> 6;\n\t"
"R0 = A0;\n\t"
"%0 = R0;\n\t"
: "=m" (sum)
: "m" (x), "m" (y), "m" (len)
: "P0", "P1", "P2", "R0", "R1", "A0", "I0", "I1", "L0", "L1", "R3"
);
return sum;
}
#define OVERRIDE_PITCH_XCORR
static void pitch_xcorr(const spx_word16_t *_x, const spx_word16_t *_y, spx_word32_t *corr, int len, int nb_pitch, char *stack)
{
int i;
corr += nb_pitch - 1;
__asm__ __volatile__ (
"P2 = %0;\n\t"
"I0 = P2;\n\t" /* x in I0 */
"B0 = P2;\n\t" /* x in B0 */
"R0 = %3;\n\t" /* len in R0 */
"P3 = %3;\n\t" /* len in R0 */
"P4 = %4;\n\t" /* nb_pitch in R0 */
"R1 = R0 << 1;\n\t" /* number of bytes in x */
"L0 = R1;\n\t"
"P0 = %1;\n\t"
"P1 = %2;\n\t"
"B1 = P1;\n\t"
"L1 = 0;\n\t" /*Disable looping on I1*/
"r0 = [I0++];\n\t"
"LOOP pitch%= LC0 = P4 >> 1;\n\t"
"LOOP_BEGIN pitch%=;\n\t"
"I1 = P0;\n\t"
"A1 = A0 = 0;\n\t"
"R1 = [I1++];\n\t"
"LOOP inner_prod%= LC1 = P3 >> 1;\n\t"
"LOOP_BEGIN inner_prod%=;\n\t"
"A0 += R0.L*R1.L , A1 += R0.L*R1.H (is) || R1.L = W[I1++];\n\t"
"LOOP_END inner_prod%=;\n\t"
"A0 += R0.H*R1.H , A1 += R0.H*R1.L (is) || R1.H = W[I1++] || R0 = [I0++];\n\t"
"A0 = A0 >>> 6;\n\t"
"A1 = A1 >>> 6;\n\t"
"R2 = A0, R3 = A1;\n\t"
"[P1--] = r2;\n\t"
"[P1--] = r3;\n\t"
"LOOP_END pitch%=;\n\t"
"P0 += 4;\n\t"
: : "m" (_x), "m" (_y), "m" (corr), "m" (len), "m" (nb_pitch)
: "A0", "A1", "P0", "P1", "P2", "P3", "P4", "R0", "R1", "R2", "R3", "I0", "I1", "L0", "L1", "B0", "B1", "memory"
);
}
/* Copyright (C) 2005 Analog Devices
Author: Jean-Marc Valin
File: vq_arm4.h
Blackfin-optimized vq routine
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.
*/
#define OVERRIDE_VQ_NBEST
void vq_nbest(spx_word16_t *in, const spx_word16_t *codebook, int len, int entries, spx_word32_t *E, int N, int *nbest, spx_word32_t *best_dist, char *stack)
{
int i,j,k,used;
used = 0;
for (i=0;i<entries;i++)
{
spx_word32_t dist;
__asm__
(
"%0 >>= 1;\n\t"
"A0 = %0;\n\t"
"I0 = %3;\n\t"
"L0 = 0;\n\t"