generic-library/webrtc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Optimized EstCodeLpcCoef() for iSAC with intrinsics in Android-Neon platform.

Browse Source 
Cycles of the whole iSAC codec was reduced by 7.9%, measured by offline file test, with time() function.

Bit exact.

** Code style cleanup is not considered in this CL. **
Review URL: https://webrtc-codereview.appspot.com/1069004

git-svn-id: http://webrtc.googlecode.com/svn/trunk@3643 4adac7df-926f-26a2-2b94-8c16560cd09d
This commit is contained in:
kma@webrtc.org 2013-03-09 00:38:14 +00:00
parent 5d3ced5df0
commit 23da8622c0
5 changed files with 530 additions and 265 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

508
webrtc/modules/audio_coding/codecs/isac/fix/source/entropy_coding.c
View File

@ -27,6 +27,40 @@
#include "settings.h" #include "settings.h"
#include "signal_processing_library.h" #include "signal_processing_library.h"
/*
* Eenumerations for arguments to functions WebRtcIsacfix_MatrixProduct1()
* and WebRtcIsacfix_MatrixProduct2().
*/
enum matrix_index_factor {
kTIndexFactor1 = 1,
kTIndexFactor2 = 2,
kTIndexFactor3 = SUBFRAMES,
kTIndexFactor4 = LPC_SHAPE_ORDER
};
enum matrix_index_step {
kTIndexStep1 = 1,
kTIndexStep2 = SUBFRAMES,
kTIndexStep3 = LPC_SHAPE_ORDER
};
enum matrixprod_loop_count {
kTLoopCount1 = SUBFRAMES,
kTLoopCount2 = 2,
kTLoopCount3 = LPC_SHAPE_ORDER
};
enum matrix1_shift_value {
kTMatrix1_shift0 = 0,
kTMatrix1_shift1 = 1,
kTMatrix1_shift5 = 5
};
enum matrixprod_init_case {
kTInitCase0 = 0,
kTInitCase1 = 1
};
/* /*
This function implements the fix-point correspondant function to lrint. This function implements the fix-point correspondant function to lrint.
@ -775,6 +809,115 @@ static void Lar2polyFix(WebRtc_Word32 *larsQ17,
} }
} }
/*
Function WebRtcIsacfix_MatrixProduct1C() does one form of matrix multiplication.
It first shifts input data of one matrix, determines the right indexes for the
two matrixes, multiply them, and write the results into an output buffer.
Note that two factors (or, multipliers) determine the initialization values of
the variable |matrix1_index| in the code. The relationship is
|matrix1_index| = |matrix1_index_factor1| * |matrix1_index_factor2|, where
|matrix1_index_factor1| is given by the argument while |matrix1_index_factor2|
is determined by the value of argument |matrix1_index_init_case|;
|matrix1_index_factor2| is the value of the outmost loop counter j (when
|matrix1_index_init_case| is 0), or the value of the middle loop counter k (when
|matrix1_index_init_case| is non-zero).
|matrix0_index| is determined the same way.
Arguments:
matrix0[]: matrix0 data in Q15 domain.
matrix1[]: matrix1 data.
matrix_product[]: output data (matrix product).
matrix1_index_factor1: The first of two factors determining the
initialization value of matrix1_index.
matrix0_index_factor1: The first of two factors determining the
initialization value of matrix0_index.
matrix1_index_init_case: Case number for selecting the second of two
factors determining the initialization value
of matrix1_index and matrix0_index.
matrix1_index_step: Incremental step for matrix1_index.
matrix0_index_step: Incremental step for matrix0_index.
inner_loop_count: Maximum count of the inner loop.
mid_loop_count: Maximum count of the intermediate loop.
shift: Left shift value for matrix1.
*/
void WebRtcIsacfix_MatrixProduct1C(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix1_index_factor1,
const int matrix0_index_factor1,
const int matrix1_index_init_case,
const int matrix1_index_step,
const int matrix0_index_step,
const int inner_loop_count,
const int mid_loop_count,
const int shift) {
int j = 0, k = 0, n = 0;
int matrix0_index = 0, matrix1_index = 0, matrix_prod_index = 0;
int* matrix0_index_factor2 = &k;
int* matrix1_index_factor2 = &j;
if (matrix1_index_init_case != 0) {
matrix0_index_factor2 = &j;
matrix1_index_factor2 = &k;
}
for (j = 0; j < SUBFRAMES; j++) {
matrix_prod_index = mid_loop_count * j;
for (k = 0; k < mid_loop_count; k++) {
int32_t sum32 = 0;
matrix0_index = matrix0_index_factor1 * (*matrix0_index_factor2);
matrix1_index = matrix1_index_factor1 * (*matrix1_index_factor2);
for (n = 0; n < inner_loop_count; n++) {
sum32 += (WEBRTC_SPL_MUL_16_32_RSFT16(matrix0[matrix0_index],
matrix1[matrix1_index] << shift));
matrix0_index += matrix0_index_step;
matrix1_index += matrix1_index_step;
}
matrix_product[matrix_prod_index] = sum32;
matrix_prod_index++;
}
}
}
/*
Function WebRtcIsacfix_MatrixProduct2C() returns the product of two matrixes,
one of which has two columns. It first has to determine the correct index of
the first matrix before doing the actual element multiplication.
Arguments:
matrix0[]: A matrix in Q15 domain.
matrix1[]: A matrix in Q21 domain.
matrix_product[]: Output data in Q17 domain.
matrix0_index_factor: A factor determining the initialization value
of matrix0_index.
matrix0_index_step: Incremental step for matrix0_index.
*/
void WebRtcIsacfix_MatrixProduct2C(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix0_index_factor,
const int matrix0_index_step) {
int j = 0, n = 0;
int matrix1_index = 0, matrix0_index = 0, matrix_prod_index = 0;
for (j = 0; j < SUBFRAMES; j++) {
int32_t sum32 = 0, sum32_2 = 0;
matrix1_index = 0;
matrix0_index = matrix0_index_factor * j;
for (n = SUBFRAMES; n > 0; n--) {
sum32 += (WEBRTC_SPL_MUL_16_32_RSFT16(matrix0[matrix0_index],
matrix1[matrix1_index]));
sum32_2 += (WEBRTC_SPL_MUL_16_32_RSFT16(matrix0[matrix0_index],
matrix1[matrix1_index + 1]));
matrix1_index += 2;
matrix0_index += matrix0_index_step;
}
matrix_product[matrix_prod_index] = sum32 >> 3;
matrix_product[matrix_prod_index + 1] = sum32_2 >> 3;
matrix_prod_index += 2;
}
}
int WebRtcIsacfix_DecodeLpc(WebRtc_Word32 *gain_lo_hiQ17, int WebRtcIsacfix_DecodeLpc(WebRtc_Word32 *gain_lo_hiQ17,
WebRtc_Word16 *LPCCoef_loQ15, WebRtc_Word16 *LPCCoef_loQ15,
WebRtc_Word16 *LPCCoef_hiQ15, WebRtc_Word16 *LPCCoef_hiQ15,
@ -801,7 +944,7 @@ int WebRtcIsacfix_DecodeLpcCoef(Bitstr_dec *streamdata,
{ {
int j, k, n; int j, k, n;
int err; int err;
WebRtc_Word16 pos, pos2, posg, poss, offsg, offss, offs2; WebRtc_Word16 pos, pos2, posg, poss;
WebRtc_Word16 gainpos; WebRtc_Word16 gainpos;
WebRtc_Word16 model; WebRtc_Word16 model;
WebRtc_Word16 index_QQ[KLT_ORDER_SHAPE]; WebRtc_Word16 index_QQ[KLT_ORDER_SHAPE];
@ -842,31 +985,17 @@ int WebRtcIsacfix_DecodeLpcCoef(Bitstr_dec *streamdata,
/* inverse KLT */ /* inverse KLT */
/* left transform */ // Transpose matrix! /* left transform */ // Transpose matrix!
offsg = 0; WebRtcIsacfix_MatrixProduct1(WebRtcIsacfix_kT1GainQ15[model], tmpcoeffs_gQ17,
offss = 0; tmpcoeffs2_gQ21, kTIndexFactor2, kTIndexFactor2,
posg = 0; kTInitCase0, kTIndexStep1, kTIndexStep1,
kTLoopCount2, kTLoopCount2, kTMatrix1_shift5);
poss = 0; poss = 0;
for (j=0; j<SUBFRAMES; j++) { for (j=0; j<SUBFRAMES; j++) {
offs2 = 0;
for (k=0; k<2; k++) {
sumQQ = 0;
pos = offsg;
pos2 = offs2;
for (n=0; n<2; n++) {
sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1GainQ15[model][pos2], tmpcoeffs_gQ17[pos]<<5)); // (Q15*Q17)>>(16-5) = Q21
pos++;
pos2++;
}
tmpcoeffs2_gQ21[posg] = sumQQ; //Q21
posg++;
offs2 += 2;
}
offs2 = 0;
for (k=0; k<LPC_SHAPE_ORDER; k++) { for (k=0; k<LPC_SHAPE_ORDER; k++) {
sumQQ = 0; sumQQ = 0;
pos = offss; pos = LPC_SHAPE_ORDER * j;
pos2 = offs2; pos2 = LPC_SHAPE_ORDER * k;
for (n=0; n<LPC_SHAPE_ORDER; n++) { for (n=0; n<LPC_SHAPE_ORDER; n++) {
sumQQ += WEBRTC_SPL_MUL_16_16_RSFT(tmpcoeffs_sQ10[pos], WebRtcIsacfix_kT1ShapeQ15[model][pos2], 7); // (Q10*Q15)>>7 = Q18 sumQQ += WEBRTC_SPL_MUL_16_16_RSFT(tmpcoeffs_sQ10[pos], WebRtcIsacfix_kT1ShapeQ15[model][pos2], 7); // (Q10*Q15)>>7 = Q18
pos++; pos++;
@ -874,48 +1003,16 @@ int WebRtcIsacfix_DecodeLpcCoef(Bitstr_dec *streamdata,
} }
tmpcoeffs2_sQ18[poss] = sumQQ; //Q18 tmpcoeffs2_sQ18[poss] = sumQQ; //Q18
poss++; poss++;
offs2 += LPC_SHAPE_ORDER;
} }
offsg += 2;
offss += LPC_SHAPE_ORDER;
} }
/* right transform */ // Transpose matrix /* right transform */ // Transpose matrix
offsg = 0; WebRtcIsacfix_MatrixProduct2(WebRtcIsacfix_kT2GainQ15[0], tmpcoeffs2_gQ21,
offss = 0; tmpcoeffs_gQ17, kTIndexFactor1, kTIndexStep2);
posg = 0; WebRtcIsacfix_MatrixProduct1(WebRtcIsacfix_kT2ShapeQ15[model],
poss = 0; tmpcoeffs2_sQ18, tmpcoeffs_sQ17, kTIndexFactor1, kTIndexFactor1,
for (j=0; j<SUBFRAMES; j++) { kTInitCase1, kTIndexStep3, kTIndexStep2, kTLoopCount1, kTLoopCount3,
posg = offsg; kTMatrix1_shift0);
for (k=0; k<2; k++) {
sumQQ = 0;
pos = k;
pos2 = j;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += WEBRTC_SPL_LSHIFT_W32(WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2GainQ15[model][pos2], tmpcoeffs2_gQ21[pos]), 1); // (Q15*Q21)>>(16-1) = Q21
pos += 2;
pos2 += SUBFRAMES;
}
tmpcoeffs_gQ17[posg] = WEBRTC_SPL_RSHIFT_W32(sumQQ, 4);
posg++;
}
poss = offss;
for (k=0; k<LPC_SHAPE_ORDER; k++) {
sumQQ = 0;
pos = k;
pos2 = j;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2ShapeQ15[model][pos2], tmpcoeffs2_sQ18[pos])); // (Q15*Q18)>>16 = Q17
pos += LPC_SHAPE_ORDER;
pos2 += SUBFRAMES;
}
tmpcoeffs_sQ17[poss] = sumQQ;
poss++;
}
offsg += 2;
offss += LPC_SHAPE_ORDER;
}
/* scaling, mean addition, and gain restoration */ /* scaling, mean addition, and gain restoration */
gainpos = 0; gainpos = 0;
@ -968,7 +1065,7 @@ static int EstCodeLpcCoef(WebRtc_Word32 *LPCCoefQ17,
transcode_obj *transcodingParam) { transcode_obj *transcodingParam) {
int j, k, n; int j, k, n;
WebRtc_Word16 posQQ, pos2QQ, gainpos; WebRtc_Word16 posQQ, pos2QQ, gainpos;
WebRtc_Word16 pos, pos2, poss, posg, offsg, offss, offs2; WebRtc_Word16 pos, poss, posg, offsg;
WebRtc_Word16 index_gQQ[KLT_ORDER_GAIN], index_sQQ[KLT_ORDER_SHAPE]; WebRtc_Word16 index_gQQ[KLT_ORDER_GAIN], index_sQQ[KLT_ORDER_SHAPE];
WebRtc_Word16 index_ovr_gQQ[KLT_ORDER_GAIN], index_ovr_sQQ[KLT_ORDER_SHAPE]; WebRtc_Word16 index_ovr_gQQ[KLT_ORDER_GAIN], index_ovr_sQQ[KLT_ORDER_SHAPE];
WebRtc_Word32 BitsQQ; WebRtc_Word32 BitsQQ;
@ -1034,73 +1131,38 @@ static int EstCodeLpcCoef(WebRtc_Word32 *LPCCoefQ17,
/* left transform */ /* left transform */
offsg = 0; offsg = 0;
offss = 0; posg = 0;
for (j=0; j<SUBFRAMES; j++) { for (j=0; j<SUBFRAMES; j++) {
posg = offsg; // Q21 = Q6 * Q15
for (k=0; k<2; k++) { sumQQ = WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg],
sumQQ = 0; WebRtcIsacfix_kT1GainQ15[0][0]);
pos = offsg; sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg + 1],
pos2 = k; WebRtcIsacfix_kT1GainQ15[0][2]);
for (n=0; n<2; n++) { tmpcoeffs2_gQ21[posg] = sumQQ;
sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[pos], WebRtcIsacfix_kT1GainQ15[0][pos2]); //Q21 = Q6*Q15 posg++;
pos++;
pos2 += 2; // Q21 = Q6 * Q15
} sumQQ = WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg],
tmpcoeffs2_gQ21[posg] = sumQQ; WebRtcIsacfix_kT1GainQ15[0][1]);
posg++; sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg + 1],
} WebRtcIsacfix_kT1GainQ15[0][3]);
poss = offss; tmpcoeffs2_gQ21[posg] = sumQQ;
for (k=0; k<LPC_SHAPE_ORDER; k++) { posg++;
sumQQ = 0;
pos = offss;
pos2 = k;
for (n=0; n<LPC_SHAPE_ORDER; n++) {
sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1ShapeQ15[0][pos2], tmpcoeffs_sQ17[pos]<<1)); // (Q15*Q17)>>(16-1) = Q17
pos++;
pos2 += LPC_SHAPE_ORDER;
}
tmpcoeffs2_sQ17[poss] = sumQQ; //Q17
poss++;
}
offsg += 2; offsg += 2;
offss += LPC_SHAPE_ORDER;
} }
WebRtcIsacfix_MatrixProduct1(WebRtcIsacfix_kT1ShapeQ15[0], tmpcoeffs_sQ17,
tmpcoeffs2_sQ17, kTIndexFactor4, kTIndexFactor1, kTInitCase0,
kTIndexStep1, kTIndexStep3, kTLoopCount3, kTLoopCount3, kTMatrix1_shift1);
/* right transform */ /* right transform */
offsg = 0; WebRtcIsacfix_MatrixProduct2(WebRtcIsacfix_kT2GainQ15[0], tmpcoeffs2_gQ21,
offss = 0; tmpcoeffs_gQ17, kTIndexFactor3, kTIndexStep1);
offs2 = 0;
for (j=0; j<SUBFRAMES; j++) { WebRtcIsacfix_MatrixProduct1(WebRtcIsacfix_kT2ShapeQ15[0], tmpcoeffs2_sQ17,
posg = offsg; tmpcoeffs_sQ17, kTIndexFactor1, kTIndexFactor3, kTInitCase1, kTIndexStep3,
for (k=0; k<2; k++) { kTIndexStep1, kTLoopCount1, kTLoopCount3, kTMatrix1_shift1);
sumQQ = 0;
pos = k;
pos2 = offs2;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += WEBRTC_SPL_LSHIFT_W32(WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2GainQ15[0][pos2], tmpcoeffs2_gQ21[pos]), 1); // (Q15*Q21)>>(16-1) = Q21
pos += 2;
pos2++;
}
tmpcoeffs_gQ17[posg] = WEBRTC_SPL_RSHIFT_W32(sumQQ, 4);
posg++;
}
poss = offss;
for (k=0; k<LPC_SHAPE_ORDER; k++) {
sumQQ = 0;
pos = k;
pos2 = offs2;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2ShapeQ15[0][pos2], tmpcoeffs2_sQ17[pos]<<1)); // (Q15*Q17)>>(16-1) = Q17
pos += LPC_SHAPE_ORDER;
pos2++;
}
tmpcoeffs_sQ17[poss] = sumQQ;
poss++;
}
offs2 += SUBFRAMES;
offsg += 2;
offss += LPC_SHAPE_ORDER;
}
/* quantize coefficients */ /* quantize coefficients */
@ -1191,47 +1253,14 @@ static int EstCodeLpcCoef(WebRtc_Word32 *LPCCoefQ17,
/* inverse KLT */ /* inverse KLT */
/* left transform */ // Transpose matrix! /* left transform */ // Transpose matrix!
offss = 0; WebRtcIsacfix_MatrixProduct1(WebRtcIsacfix_kT1ShapeQ15[0], tmpcoeffs_sQ17,
poss = 0; tmpcoeffs2_sQ17, kTIndexFactor4, kTIndexFactor4, kTInitCase0,
for (j=0; j<SUBFRAMES; j++) { kTIndexStep1, kTIndexStep1, kTLoopCount3, kTLoopCount3, kTMatrix1_shift1);
offs2 = 0;
for (k=0; k<LPC_SHAPE_ORDER; k++) {
sumQQ = 0;
pos = offss;
pos2 = offs2;
for (n=0; n<LPC_SHAPE_ORDER; n++) {
sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1ShapeQ15[0][pos2], tmpcoeffs_sQ17[pos]<<1)); // (Q15*Q17)>>(16-1) = Q17
pos++;
pos2++;
}
tmpcoeffs2_sQ17[poss] = sumQQ;
poss++;
offs2 += LPC_SHAPE_ORDER;
}
offss += LPC_SHAPE_ORDER;
}
/* right transform */ // Transpose matrix /* right transform */ // Transpose matrix
offss = 0; WebRtcIsacfix_MatrixProduct1(WebRtcIsacfix_kT2ShapeQ15[0], tmpcoeffs2_sQ17,
poss = 0; tmpcoeffs_sQ17, kTIndexFactor1, kTIndexFactor1, kTInitCase1, kTIndexStep3,
for (j=0; j<SUBFRAMES; j++) { kTIndexStep2, kTLoopCount1, kTLoopCount3, kTMatrix1_shift1);
poss = offss;
for (k=0; k<LPC_SHAPE_ORDER; k++) {
sumQQ = 0;
pos = k;
pos2 = j;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2ShapeQ15[0][pos2], tmpcoeffs2_sQ17[pos]<<1)); // (Q15*Q17)>>(16-1) = Q17
pos += LPC_SHAPE_ORDER;
pos2 += SUBFRAMES;
}
tmpcoeffs_sQ17[poss] = sumQQ;
poss++;
}
offss += LPC_SHAPE_ORDER;
}
/* scaling, mean addition, and gain restoration */ /* scaling, mean addition, and gain restoration */
poss = 0;pos=0; poss = 0;pos=0;
@ -1266,42 +1295,26 @@ static int EstCodeLpcCoef(WebRtc_Word32 *LPCCoefQ17,
offsg = 0; offsg = 0;
posg = 0; posg = 0;
for (j=0; j<SUBFRAMES; j++) { for (j=0; j<SUBFRAMES; j++) {
offs2 = 0; // (Q15 * Q17) >> (16 - 1) = Q17; Q17 << 4 = Q21.
for (k=0; k<2; k++) { sumQQ = (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1GainQ15[0][0],
sumQQ = 0; tmpcoeffs_gQ17[offsg]) << 1);
pos = offsg; sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1GainQ15[0][1],
pos2 = offs2; tmpcoeffs_gQ17[offsg + 1]) << 1);
for (n=0; n<2; n++) { tmpcoeffs2_gQ21[posg] = WEBRTC_SPL_LSHIFT_W32(sumQQ, 4);
sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1GainQ15[0][pos2], tmpcoeffs_gQ17[pos])<<1); // (Q15*Q17)>>(16-1) = Q17 posg++;
pos++;
pos2++; sumQQ = (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1GainQ15[0][2],
} tmpcoeffs_gQ17[offsg]) << 1);
tmpcoeffs2_gQ21[posg] = WEBRTC_SPL_LSHIFT_W32(sumQQ, 4); //Q17<<4 = Q21 sumQQ += (WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT1GainQ15[0][3],
posg++; tmpcoeffs_gQ17[offsg + 1]) << 1);
offs2 += 2; tmpcoeffs2_gQ21[posg] = WEBRTC_SPL_LSHIFT_W32(sumQQ, 4);
} posg++;
offsg += 2; offsg += 2;
} }
/* right transform */ // Transpose matrix /* right transform */ // Transpose matrix
offsg = 0; WebRtcIsacfix_MatrixProduct2(WebRtcIsacfix_kT2GainQ15[0], tmpcoeffs2_gQ21,
posg = 0; tmpcoeffs_gQ17, kTIndexFactor1, kTIndexStep2);
for (j=0; j<SUBFRAMES; j++) {
posg = offsg;
for (k=0; k<2; k++) {
sumQQ = 0;
pos = k;
pos2 = j;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += WEBRTC_SPL_LSHIFT_W32(WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2GainQ15[0][pos2], tmpcoeffs2_gQ21[pos]), 1); // (Q15*Q21)>>(16-1) = Q21
pos += 2;
pos2 += SUBFRAMES;
}
tmpcoeffs_gQ17[posg] = WEBRTC_SPL_RSHIFT_W32(sumQQ, 4);
posg++;
}
offsg += 2;
}
/* scaling, mean addition, and gain restoration */ /* scaling, mean addition, and gain restoration */
posg = 0; posg = 0;
@ -1323,9 +1336,9 @@ static int EstCodeLpcCoef(WebRtc_Word32 *LPCCoefQ17,
int WebRtcIsacfix_EstCodeLpcGain(WebRtc_Word32 *gain_lo_hiQ17, int WebRtcIsacfix_EstCodeLpcGain(WebRtc_Word32 *gain_lo_hiQ17,
Bitstr_enc *streamdata, Bitstr_enc *streamdata,
ISAC_SaveEncData_t* encData) { ISAC_SaveEncData_t* encData) {
int j, k, n; int j, k;
WebRtc_Word16 posQQ, pos2QQ, gainpos; WebRtc_Word16 posQQ, pos2QQ, gainpos;
WebRtc_Word16 pos, pos2, posg, offsg, offs2; WebRtc_Word16 posg;
WebRtc_Word16 index_gQQ[KLT_ORDER_GAIN]; WebRtc_Word16 index_gQQ[KLT_ORDER_GAIN];
WebRtc_Word16 tmpcoeffs_gQ6[KLT_ORDER_GAIN]; WebRtc_Word16 tmpcoeffs_gQ6[KLT_ORDER_GAIN];
@ -1343,7 +1356,7 @@ int WebRtcIsacfix_EstCodeLpcGain(WebRtc_Word32 *gain_lo_hiQ17,
} }
/* log gains, mean removal and scaling */ /* log gains, mean removal and scaling */
posg = 0; pos = 0; gainpos = 0; posg = 0; gainpos = 0;
for (k=0; k<SUBFRAMES; k++) { for (k=0; k<SUBFRAMES; k++) {
/* log gains */ /* log gains */
@ -1369,44 +1382,27 @@ int WebRtcIsacfix_EstCodeLpcGain(WebRtc_Word32 *gain_lo_hiQ17,
/* KLT */ /* KLT */
/* left transform */ /* left transform */
offsg = 0; posg = 0;
for (j=0; j<SUBFRAMES; j++) { for (j=0; j<SUBFRAMES; j++) {
posg = offsg; // Q21 = Q6 * Q15
for (k=0; k<2; k++) { sumQQ = WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[j * 2],
sumQQ = 0; WebRtcIsacfix_kT1GainQ15[0][0]);
pos = offsg; sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[j * 2 + 1],
pos2 = k; WebRtcIsacfix_kT1GainQ15[0][2]);
for (n=0; n<2; n++) { tmpcoeffs2_gQ21[posg] = sumQQ;
sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[pos], WebRtcIsacfix_kT1GainQ15[0][pos2]); //Q21 = Q6*Q15 posg++;
pos++;
pos2 += 2; sumQQ = WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[j * 2],
} WebRtcIsacfix_kT1GainQ15[0][1]);
sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[j * 2 + 1],
WebRtcIsacfix_kT1GainQ15[0][3]);
tmpcoeffs2_gQ21[posg] = sumQQ; tmpcoeffs2_gQ21[posg] = sumQQ;
posg++; posg++;
}
offsg += 2;
} }
/* right transform */ /* right transform */
offsg = 0; WebRtcIsacfix_MatrixProduct2(WebRtcIsacfix_kT2GainQ15[0], tmpcoeffs2_gQ21,
offs2 = 0; tmpcoeffs_gQ17, kTIndexFactor3, kTIndexStep1);
for (j=0; j<SUBFRAMES; j++) {
posg = offsg;
for (k=0; k<2; k++) {
sumQQ = 0;
pos = k;
pos2 = offs2;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += WEBRTC_SPL_LSHIFT_W32(WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2GainQ15[0][pos2], tmpcoeffs2_gQ21[pos]), 1); // (Q15*Q21)>>(16-1) = Q21
pos += 2;
pos2++;
}
tmpcoeffs_gQ17[posg] = WEBRTC_SPL_RSHIFT_W32(sumQQ, 4);
posg++;
}
offsg += 2;
offs2 += SUBFRAMES;
}
/* quantize coefficients */ /* quantize coefficients */
@ -1454,7 +1450,8 @@ int WebRtcIsacfix_EncodeLpc(WebRtc_Word32 *gain_lo_hiQ17,
Poly2LarFix(LPCCoef_loQ15, ORDERLO, LPCCoef_hiQ15, ORDERHI, SUBFRAMES, larsQ17); Poly2LarFix(LPCCoef_loQ15, ORDERLO, LPCCoef_hiQ15, ORDERHI, SUBFRAMES, larsQ17);
status = EstCodeLpcCoef(larsQ17, gain_lo_hiQ17, model, sizeQ11, streamdata, encData, transcodeParam); status = EstCodeLpcCoef(larsQ17, gain_lo_hiQ17, model, sizeQ11,
streamdata, encData, transcodeParam);
if (status < 0) { if (status < 0) {
return (status); return (status);
} }
@ -1978,9 +1975,9 @@ int WebRtcIsacfix_EncodeReceiveBandwidth(WebRtc_Word16 *BWno, Bitstr_enc *stream
/* estimate codel length of LPC Coef */ /* estimate codel length of LPC Coef */
void WebRtcIsacfix_TranscodeLpcCoef(WebRtc_Word32 *gain_lo_hiQ17, void WebRtcIsacfix_TranscodeLpcCoef(WebRtc_Word32 *gain_lo_hiQ17,
WebRtc_Word16 *index_gQQ) { WebRtc_Word16 *index_gQQ) {
int j, k, n; int j, k;
WebRtc_Word16 posQQ, pos2QQ; WebRtc_Word16 posQQ, pos2QQ;
WebRtc_Word16 pos, pos2, posg, offsg, offs2, gainpos; WebRtc_Word16 posg, offsg, gainpos;
WebRtc_Word32 tmpcoeffs_gQ6[KLT_ORDER_GAIN]; WebRtc_Word32 tmpcoeffs_gQ6[KLT_ORDER_GAIN];
WebRtc_Word32 tmpcoeffs_gQ17[KLT_ORDER_GAIN]; WebRtc_Word32 tmpcoeffs_gQ17[KLT_ORDER_GAIN];
WebRtc_Word32 tmpcoeffs2_gQ21[KLT_ORDER_GAIN]; WebRtc_Word32 tmpcoeffs2_gQ21[KLT_ORDER_GAIN];
@ -1988,7 +1985,7 @@ void WebRtcIsacfix_TranscodeLpcCoef(WebRtc_Word32 *gain_lo_hiQ17,
/* log gains, mean removal and scaling */ /* log gains, mean removal and scaling */
posg = 0;pos=0; gainpos=0; posg = 0; gainpos=0;
for (k=0; k<SUBFRAMES; k++) { for (k=0; k<SUBFRAMES; k++) {
/* log gains */ /* log gains */
@ -2017,43 +2014,26 @@ void WebRtcIsacfix_TranscodeLpcCoef(WebRtc_Word32 *gain_lo_hiQ17,
/* left transform */ /* left transform */
offsg = 0; offsg = 0;
for (j=0; j<SUBFRAMES; j++) { for (j=0; j<SUBFRAMES; j++) {
posg = offsg; // Q21 = Q6 * Q15
for (k=0; k<2; k++) { sumQQ = WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg],
sumQQ = 0; WebRtcIsacfix_kT1GainQ15[0][0]);
pos = offsg; sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg + 1],
pos2 = k; WebRtcIsacfix_kT1GainQ15[0][2]);
for (n=0; n<2; n++) {
sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[pos], WebRtcIsacfix_kT1GainQ15[0][pos2]); //Q21 = Q6*Q15
pos++;
pos2 += 2;
}
tmpcoeffs2_gQ21[posg] = sumQQ; tmpcoeffs2_gQ21[posg] = sumQQ;
posg++; posg++;
}
offsg += 2; // Q21 = Q6 * Q15
sumQQ = WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg],
WebRtcIsacfix_kT1GainQ15[0][1]);
sumQQ += WEBRTC_SPL_MUL_16_16(tmpcoeffs_gQ6[offsg + 1],
WebRtcIsacfix_kT1GainQ15[0][3]);
tmpcoeffs2_gQ21[posg] = sumQQ;
posg++;
} }
/* right transform */ /* right transform */
offsg = 0; WebRtcIsacfix_MatrixProduct2(WebRtcIsacfix_kT2GainQ15[0], tmpcoeffs2_gQ21,
offs2 = 0; tmpcoeffs_gQ17, kTIndexFactor3, kTIndexStep1);
for (j=0; j<SUBFRAMES; j++) {
posg = offsg;
for (k=0; k<2; k++) {
sumQQ = 0;
pos = k;
pos2 = offs2;
for (n=0; n<SUBFRAMES; n++) {
sumQQ += WEBRTC_SPL_LSHIFT_W32(WEBRTC_SPL_MUL_16_32_RSFT16(WebRtcIsacfix_kT2GainQ15[0][pos2], tmpcoeffs2_gQ21[pos]), 1); // (Q15*Q21)>>(16-1) = Q21
pos += 2;
pos2++;
}
tmpcoeffs_gQ17[posg] = WEBRTC_SPL_RSHIFT_W32(sumQQ, 4);
posg++;
}
offsg += 2;
offs2 += SUBFRAMES;
}
/* quantize coefficients */ /* quantize coefficients */
for (k=0; k<KLT_ORDER_GAIN; k++) //ATTN: ok? for (k=0; k<KLT_ORDER_GAIN; k++) //ATTN: ok?

60
webrtc/modules/audio_coding/codecs/isac/fix/source/entropy_coding.h
View File

@ -108,4 +108,62 @@ int WebRtcIsacfix_EncodeReceiveBandwidth(WebRtc_Word16 *BWno,
void WebRtcIsacfix_TranscodeLpcCoef(WebRtc_Word32 *tmpcoeffs_gQ6, void WebRtcIsacfix_TranscodeLpcCoef(WebRtc_Word32 *tmpcoeffs_gQ6,
WebRtc_Word16 *index_gQQ); WebRtc_Word16 *index_gQQ);
#endif /* WEBRTC_MODULES_AUDIO_CODING_CODECS_ISAC_FIX_SOURCE_ENTROPY_CODING_H_ */ // Pointer functions for LPC transforms.
typedef void (*MatrixProduct1)(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix1_index_factor1,
const int matrix0_index_factor1,
const int matrix1_index_init_case,
const int matrix1_index_step,
const int matrix0_index_step,
const int inner_loop_count,
const int mid_loop_count,
const int shift);
typedef void (*MatrixProduct2)(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix0_index_factor,
const int matrix0_index_step);
extern MatrixProduct1 WebRtcIsacfix_MatrixProduct1;
extern MatrixProduct2 WebRtcIsacfix_MatrixProduct2;
void WebRtcIsacfix_MatrixProduct1C(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix1_index_factor1,
const int matrix0_index_factor1,
const int matrix1_index_init_case,
const int matrix1_index_step,
const int matrix0_index_step,
const int inner_loop_count,
const int mid_loop_count,
const int shift);
void WebRtcIsacfix_MatrixProduct2C(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix0_index_factor,
const int matrix0_index_step);
#if (defined WEBRTC_DETECT_ARM_NEON) || (defined WEBRTC_ARCH_ARM_NEON)
void WebRtcIsacfix_MatrixProduct1Neon(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix1_index_factor1,
const int matrix0_index_factor1,
const int matrix1_index_init_case,
const int matrix1_index_step,
const int matrix0_index_step,
const int inner_loop_count,
const int mid_loop_count,
const int shift);
void WebRtcIsacfix_MatrixProduct2Neon(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix0_index_factor,
const int matrix0_index_step);
#endif
#endif // WEBRTC_MODULES_AUDIO_CODING_CODECS_ISAC_FIX_SOURCE_ENTROPY_CODING_H_

220
webrtc/modules/audio_coding/codecs/isac/fix/source/entropy_coding_neon.c Normal file
View File

@ -0,0 +1,220 @@
/*
* Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/* This file contains WebRtcIsacfix_MatrixProduct1Neon() and
* WebRtcIsacfix_MatrixProduct2Neon() for ARM Neon platform. API's are in
* entropy_coding.c. Results are bit exact with the c code for
* generic platforms.
*/
#include "entropy_coding.h"
#include <arm_neon.h>
#include <assert.h>
#include <stddef.h>
#include "signal_processing_library.h"
void WebRtcIsacfix_MatrixProduct1Neon(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix1_index_factor1,
const int matrix0_index_factor1,
const int matrix1_index_init_case,
const int matrix1_index_step,
const int matrix0_index_step,
const int inner_loop_count,
const int mid_loop_count,
const int shift) {
int j = 0, k = 0, n = 0;
int matrix1_index = 0, matrix0_index = 0, matrix_prod_index = 0;
int* matrix1_index_factor2 = &j;
int* matrix0_index_factor2 = &k;
if (matrix1_index_init_case != 0) {
matrix1_index_factor2 = &k;
matrix0_index_factor2 = &j;
}
int32x4_t shift32x4 = vdupq_n_s32(shift);
int32x2_t shift32x2 = vdup_n_s32(shift);
assert(inner_loop_count % 2 == 0);
assert(mid_loop_count % 2 == 0);
if (matrix1_index_init_case != 0 && matrix1_index_factor1 == 1) {
for (j = 0; j < SUBFRAMES; j++) {
matrix_prod_index = mid_loop_count * j;
for (k = 0; k < (mid_loop_count >> 2) << 2; k += 4) {
// Initialize sum_32x4 to zeros.
int32x4_t sum_32x4 = veorq_s32(sum_32x4, sum_32x4);
matrix1_index = k;
matrix0_index = matrix0_index_factor1 * j;
for (n = 0; n < inner_loop_count; n++) {
int32x4_t matrix0_32x4 =
vdupq_n_s32((int32_t)(matrix0[matrix0_index]) << 15);
int32x4_t matrix1_32x4 =
vshlq_s32(vld1q_s32(&matrix1[matrix1_index]), shift32x4);
int32x4_t multi_32x4 = vqdmulhq_s32(matrix0_32x4, matrix1_32x4);
sum_32x4 = vqaddq_s32(sum_32x4, multi_32x4);
matrix1_index += matrix1_index_step;
matrix0_index += matrix0_index_step;
}
vst1q_s32(&matrix_product[matrix_prod_index], sum_32x4);
matrix_prod_index += 4;
}
if (mid_loop_count % 4 > 1) {
// Initialize sum_32x2 to zeros.
int32x2_t sum_32x2 = veor_s32(sum_32x2, sum_32x2);
matrix1_index = k;
k += 2;
matrix0_index = matrix0_index_factor1 * j;
for (n = 0; n < inner_loop_count; n++) {
int32x2_t matrix0_32x2 =
vdup_n_s32((int32_t)(matrix0[matrix0_index]) << 15);
int32x2_t matrix1_32x2 =
vshl_s32(vld1_s32(&matrix1[matrix1_index]), shift32x2);
int32x2_t multi_32x2 = vqdmulh_s32(matrix0_32x2, matrix1_32x2);
sum_32x2 = vqadd_s32(sum_32x2, multi_32x2);
matrix1_index += matrix1_index_step;
matrix0_index += matrix0_index_step;
}
vst1_s32(&matrix_product[matrix_prod_index], sum_32x2);
matrix_prod_index += 2;
}
}
}
else if (matrix1_index_init_case == 0 && matrix0_index_factor1 == 1) {
for (j = 0; j < SUBFRAMES; j++) {
matrix_prod_index = mid_loop_count * j;
for (k = 0; k < (mid_loop_count >> 2) << 2; k += 4) {
// Initialize sum_32x4 to zeros.
int32x4_t sum_32x4 = veorq_s32(sum_32x4, sum_32x4);
matrix1_index = matrix1_index_factor1 * j;
matrix0_index = k;
for (n = 0; n < inner_loop_count; n++) {
int32x4_t matrix1_32x4 = vdupq_n_s32(matrix1[matrix1_index] << shift);
int32x4_t matrix0_32x4 =
vshll_n_s16(vld1_s16(&matrix0[matrix0_index]), 15);
int32x4_t multi_32x4 = vqdmulhq_s32(matrix0_32x4, matrix1_32x4);
sum_32x4 = vqaddq_s32(sum_32x4, multi_32x4);
matrix1_index += matrix1_index_step;
matrix0_index += matrix0_index_step;
}
vst1q_s32(&matrix_product[matrix_prod_index], sum_32x4);
matrix_prod_index += 4;
}
if (mid_loop_count % 4 > 1) {
// Initialize sum_32x2 to zeros.
int32x2_t sum_32x2 = veor_s32(sum_32x2, sum_32x2);
matrix1_index = matrix1_index_factor1 * j;
matrix0_index = k;
for (n = 0; n < inner_loop_count; n++) {
int32x2_t multi_32x2;
int32x2_t matrix1_32x2 = vdup_n_s32(matrix1[matrix1_index] << shift);
int32x2_t matrix0_32x2 =
vset_lane_s32((int32_t)matrix0[matrix0_index], matrix0_32x2, 0);
matrix0_32x2 = vset_lane_s32((int32_t)matrix0[matrix0_index + 1],
matrix0_32x2, 1);
matrix0_32x2 = vshl_n_s32(matrix0_32x2, 15);
multi_32x2 = vqdmulh_s32(matrix1_32x2, matrix0_32x2);
sum_32x2 = vqadd_s32(sum_32x2, multi_32x2);
matrix1_index += matrix1_index_step;
matrix0_index += matrix0_index_step;
}
vst1_s32(&matrix_product[matrix_prod_index], sum_32x2);
matrix_prod_index += 2;
}
}
}
else if (matrix1_index_init_case == 0 &&
matrix1_index_step == 1 &&
matrix0_index_step == 1) {
for (j = 0; j < SUBFRAMES; j++) {
matrix_prod_index = mid_loop_count * j;
for (k = 0; k < mid_loop_count; k++) {
int32x2_t sum_32x2;
// Initialize sum_32x4 to zeros.
int32x4_t sum_32x4 = veorq_s32(sum_32x4, sum_32x4);
matrix1_index = matrix1_index_factor1 * j;
matrix0_index = matrix0_index_factor1 * k;
for (n = 0; n < (inner_loop_count >> 2) << 2; n += 4) {
int32x4_t matrix1_32x4 =
vshlq_s32(vld1q_s32(&matrix1[matrix1_index]), shift32x4);
int32x4_t matrix0_32x4 =
vshll_n_s16(vld1_s16(&matrix0[matrix0_index]), 15);
int32x4_t multi_32x4 = vqdmulhq_s32(matrix0_32x4, matrix1_32x4);
sum_32x4 = vqaddq_s32(sum_32x4, multi_32x4);
matrix1_index += 4;
matrix0_index += 4;
}
sum_32x2 = vqadd_s32(vget_low_s32(sum_32x4), vget_high_s32(sum_32x4));
if (inner_loop_count % 4 > 1) {
int32x2_t multi_32x2;
int32x2_t matrix1_32x2 =
vshl_s32(vld1_s32(&matrix1[matrix1_index]), shift32x2);
int32x2_t matrix0_32x2 =
vset_lane_s32((int32_t)matrix0[matrix0_index], matrix0_32x2, 0);
matrix0_32x2 = vset_lane_s32((int32_t)matrix0[matrix0_index + 1],
matrix0_32x2, 1);
matrix0_32x2 = vshl_n_s32(matrix0_32x2, 15);
multi_32x2 = vqdmulh_s32(matrix1_32x2, matrix0_32x2);
sum_32x2 = vqadd_s32(sum_32x2, multi_32x2);
}
sum_32x2 = vpadd_s32(sum_32x2, sum_32x2);
vst1_lane_s32(&matrix_product[matrix_prod_index], sum_32x2, 0);
matrix_prod_index++;
}
}
}
else {
for (j = 0; j < SUBFRAMES; j++) {
matrix_prod_index = mid_loop_count * j;
for (k=0; k < mid_loop_count; k++) {
int32_t sum32 = 0;
matrix1_index = matrix1_index_factor1 * (*matrix1_index_factor2);
matrix0_index = matrix0_index_factor1 * (*matrix0_index_factor2);
for (n = 0; n < inner_loop_count; n++) {
sum32 += (WEBRTC_SPL_MUL_16_32_RSFT16(matrix0[matrix0_index],
matrix1[matrix1_index] << shift));
matrix1_index += matrix1_index_step;
matrix0_index += matrix0_index_step;
}
matrix_product[matrix_prod_index] = sum32;
matrix_prod_index++;
}
}
}
}
void WebRtcIsacfix_MatrixProduct2Neon(const int16_t matrix0[],
const int32_t matrix1[],
int32_t matrix_product[],
const int matrix0_index_factor,
const int matrix0_index_step) {
int j = 0, n = 0;
int matrix1_index = 0, matrix0_index = 0, matrix_prod_index = 0;
for (j = 0; j < SUBFRAMES; j++) {
// Initialize sum_32x2 to zeros.
int32x2_t sum_32x2 = veor_s32(sum_32x2, sum_32x2);
matrix1_index = 0;
matrix0_index = matrix0_index_factor * j;
for (n = SUBFRAMES; n > 0; n--) {
int32x2_t matrix0_32x2 =
vdup_n_s32((int32_t)(matrix0[matrix0_index]) << 15);
int32x2_t matrix1_32x2 = vld1_s32(&matrix1[matrix1_index]);
int32x2_t multi_32x2 = vqdmulh_s32(matrix0_32x2, matrix1_32x2);
sum_32x2 = vqadd_s32(sum_32x2, multi_32x2);
matrix1_index += 2;
matrix0_index += matrix0_index_step;
}
sum_32x2 = vshr_n_s32(sum_32x2, 3);
vst1_s32(&matrix_product[matrix_prod_index], sum_32x2);
matrix_prod_index += 2;
}
}

6
webrtc/modules/audio_coding/codecs/isac/fix/source/isacfix.c
View File

@ -31,6 +31,8 @@
FilterMaLoopFix WebRtcIsacfix_FilterMaLoopFix; FilterMaLoopFix WebRtcIsacfix_FilterMaLoopFix;
Spec2Time WebRtcIsacfix_Spec2Time; Spec2Time WebRtcIsacfix_Spec2Time;
Time2Spec WebRtcIsacfix_Time2Spec; Time2Spec WebRtcIsacfix_Time2Spec;
MatrixProduct1 WebRtcIsacfix_MatrixProduct1;
MatrixProduct2 WebRtcIsacfix_MatrixProduct2;
/************************************************************************** /**************************************************************************
* WebRtcIsacfix_AssignSize(...) * WebRtcIsacfix_AssignSize(...)
@ -192,6 +194,8 @@ static void WebRtcIsacfix_InitNeon(void) {
WebRtcIsacfix_CalculateResidualEnergyNeon; WebRtcIsacfix_CalculateResidualEnergyNeon;
WebRtcIsacfix_AllpassFilter2FixDec16 = WebRtcIsacfix_AllpassFilter2FixDec16 =
WebRtcIsacfix_AllpassFilter2FixDec16Neon; WebRtcIsacfix_AllpassFilter2FixDec16Neon;
WebRtcIsacfix_MatrixProduct1 = WebRtcIsacfix_MatrixProduct1Neon;
WebRtcIsacfix_MatrixProduct2 = WebRtcIsacfix_MatrixProduct2Neon;
} }
#endif #endif
@ -281,6 +285,8 @@ WebRtc_Word16 WebRtcIsacfix_EncoderInit(ISACFIX_MainStruct *ISAC_main_inst,
WebRtcIsacfix_AllpassFilter2FixDec16 = WebRtcIsacfix_AllpassFilter2FixDec16C; WebRtcIsacfix_AllpassFilter2FixDec16 = WebRtcIsacfix_AllpassFilter2FixDec16C;
WebRtcIsacfix_Time2Spec = WebRtcIsacfix_Time2SpecC; WebRtcIsacfix_Time2Spec = WebRtcIsacfix_Time2SpecC;
WebRtcIsacfix_Spec2Time = WebRtcIsacfix_Spec2TimeC; WebRtcIsacfix_Spec2Time = WebRtcIsacfix_Spec2TimeC;
WebRtcIsacfix_MatrixProduct1 = WebRtcIsacfix_MatrixProduct1C;
WebRtcIsacfix_MatrixProduct2 = WebRtcIsacfix_MatrixProduct2C ;
#ifdef WEBRTC_DETECT_ARM_NEON #ifdef WEBRTC_DETECT_ARM_NEON
if ((WebRtc_GetCPUFeaturesARM() & kCPUFeatureNEON) != 0) { if ((WebRtc_GetCPUFeaturesARM() & kCPUFeatureNEON) != 0) {

1
webrtc/modules/audio_coding/codecs/isac/fix/source/isacfix.gypi
View File

@ -97,6 +97,7 @@
'<(webrtc_root)/common_audio/common_audio.gyp:signal_processing', '<(webrtc_root)/common_audio/common_audio.gyp:signal_processing',
], ],
'sources': [ 'sources': [
'entropy_coding_neon.c',
'filterbanks_neon.S', 'filterbanks_neon.S',
'filters_neon.S', 'filters_neon.S',
'lattice_neon.S', 'lattice_neon.S',