int main ()
{
int v;
long long vll;
v = _add2 (a, b);
if (v != 0x1000f000)
abort ();
v = _sub2 (a, b);
if (v != 0x9000b000)
abort ();
v = _sub2 (b, a);
if (v != 0x70005000)
abort ();
v = _add4 (a4, b4);
if (v != 0x10f02000)
abort ();
v = _sub4 (a4, b4);
if (v != 0x90b04000)
abort ();
v = _saddu4 (a4, c4);
if (v != 0xfff050ff)
abort ();
v = _sadd2 (a, b);
if (v != 0x1000f000)
abort ();
v = _sadd2 (a, c);
if (v != 0x7fff8000)
abort ();
v = _ssub2 (a, b);
if (v != 0x7fffb000)
abort ();
v = _ssub2 (b, a);
if (v != 0x80005000)
abort ();
vll = _smpy2ll (a, b);
if (vll != 0xd8000000f4000000ll)
abort ();
vll = _smpy2ll (d, d);
if (vll != 0x7fffffff00000002ll)
abort ();
v = _avg2 (b, e);
if (v != 0x08002001)
abort ();
v = _avgu4 (d4, e4);
if (v != 0x88102980)
abort ();
v = _abs2 (a);
if (v != 0x50003000)
abort ();
v = _abs2 (f);
if (v != 0x40007fff)
abort ();
return 0;
}
void DSP_QMFA_process(DSP_QMFA_bank_t *QMFA_bank_obj)
{
sint32 i,j;
uint32 * restrict filter = (uint32 *)QMFA_bank_obj->flt_ptr;
uint32 * restrict data_ptr = (uint32 *)(QMFA_bank_obj->data_in_buffer + 0); /*input */
uint32 L = QMFA_bank_obj->blk_len/2;
uint32 M = FILT_LEN/2;
uint32 * restrict hist_rd_ptr0 = (uint32 *)QMFA_bank_obj->history[0];
uint32 * restrict hist_rd_ptr1 = (uint32 *)QMFA_bank_obj->history[1];
uint32 * restrict hist_wr_ptr0 = (uint32 *)QMFA_bank_obj->history[0];
uint32 * restrict hist_wr_ptr1 = (uint32 *)QMFA_bank_obj->history[1];
uint32 * restrict data_out_lo_ptr = (uint32 *)QMFA_bank_obj->data_out_LO_ptr;
uint32 * restrict data_out_hi_ptr = (uint32 *)QMFA_bank_obj->data_out_HI_ptr;
for(i=0;i<L;++i)
{
/*Iterations 0 to (L-1) produce the current block of output*/
register uint32 o0,o1,out0,out1;
register __int40_t accum0l,accum1l,accum0r,accum1r; /*accumulators for polyphase filter commponent 0 and 1*/
register long long temp;
accum0l = 0;accum0r = 0;
accum1l = 0;accum1r = 0;
uint32 *dptr = data_ptr + 2*i + 1;
uint32 *fptr = filter;
#pragma MUST_ITERATE( 1)
for(j=0;j<QMFA_bank_obj->iter_count[i];++j)
{
/* Limit for j is beginning at the rightmost overlap
position to the leftmost overlap position*/
register uint32 d0,d1;
register uint32 filt_coef10 = *fptr++; /*Loads the decimated filter coeficients*/
register uint32 filt_coef0 = _packh2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
register uint32 filt_coef1 = _pack2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
d1 = *dptr--;//data_ptr1[(2*(i-j))]; /*load the data right to left with 1:2 split*/
d0 = *dptr--;//data_ptr0[(2*(i-j))]; /*load the data right to left with 1:2 split*/
temp = _mpy2ll(d0, filt_coef0);
accum0l = _lsadd(_loll(temp), accum0l); accum0r = _lsadd(_hill(temp), accum0r);
temp = _mpy2ll(d1, filt_coef1);
accum1l = _lsadd(_loll(temp), accum1l); accum1r = _lsadd(_hill(temp), accum1r);
}
o0 = _pack2( _sat((accum0r + 16384)>>15), _sat((accum0l + 16384)>>15));
o1 = _pack2( _sat((accum1r + 16384)>>15), _sat((accum1l + 16384)>>15));
o0 = _add2(o0, hist_rd_ptr0[i]); /*add history sample filter 0 overlap add*/
o1 = _add2(o1, hist_rd_ptr1[i]); /*add history sample filter 1 overlap add*/
out0 = _add2(o0,o1) ;
data_out_lo_ptr[i] = out0; /*out bank0 is out0 + out1*/
out1 = _sub2(o0,o1) ;
data_out_hi_ptr[i] = out1; /*out bank1 is out0 - out1*/
}
for(i=L;i<L+M-1;++i)
{
/*Iterations L to (L + (M-1) -1) produce the
history for overlap add for next block*/
register uint32 o0,o1;
register __int40_t accum0l,accum1l,accum0r,accum1r; /*accumulators for polyphase filter commponent 0 and 1*/
register long long temp;
accum0l = 0;accum0r = 0;
accum1l = 0;accum1r = 0;
uint32 *dptr = data_ptr + 2*(L-1) + 1;
uint32 *fptr = filter + (i-L+1);
#pragma MUST_ITERATE( 1)
for(j=0;j<QMFA_bank_obj->iter_count[i];++j)
{
/*same logic for j starting at the
rightmost point of overlap to leftmost*/
register sint32 d0,d1;
register sint32 filt_coef10 = *fptr++;; /*Loads the decimated filter coeficients*/
register sint32 filt_coef0 = _packh2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
register sint32 filt_coef1 = _pack2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
d1 = *dptr--; /*load the data right to left with 1:2 split*/
d0 = *dptr--; /*load the data right to left with 1:2 split*/
temp = _mpy2ll(d0, filt_coef0);
accum0l = _lsadd(_loll(temp), accum0l); accum0r = _lsadd(_hill(temp), accum0r);
temp = _mpy2ll(d1, filt_coef1);
accum1l = _lsadd(_loll(temp), accum1l); accum1r = _lsadd(_hill(temp), accum1r);
}
o0 = _pack2( _sat((accum0r + 16384)>>15), _sat((accum0l + 16384)>>15));
o1 = _pack2( _sat((accum1r + 16384)>>15), _sat((accum1l + 16384)>>15));
o0 = _add2(o0, hist_rd_ptr0[i]); /*add history sample filter 0 overlap add*/
o1 = _add2(o1, hist_rd_ptr1[i]); /*add history sample filter 1 overlap add*/
hist_wr_ptr0[(i-L)] = o0; /* write out overlap add history history filter 0*/
hist_wr_ptr1[(i-L)] = o1; /* write out overlap add history history filter 1*/
}
return;
}
void chroma_sample_interpolation_TI(unsigned char image_Cb [RESTRICT], unsigned char image_Cr [RESTRICT]
, unsigned char refPicLXCb[RESTRICT], unsigned char refPicLXCr[RESTRICT]
, const short xFracl, const short yFracl, const short PicWidthSamples,const short stride)
{
unsigned char* pucCbPtrA = refPicLXCb;
unsigned char* pucCbPtrB = refPicLXCb + 1;
unsigned char* pucCbPtrC = refPicLXCb + PicWidthSamples;
unsigned char* pucCbPtrD = refPicLXCb + PicWidthSamples + 1;
unsigned char* pucCrPtrE = refPicLXCr;
unsigned char* pucCrPtrF = refPicLXCr + 1;
unsigned char* pucCrPtrG = refPicLXCr + PicWidthSamples;
unsigned char* pucCrPtrH = refPicLXCr + PicWidthSamples + 1;
unsigned char* pucOutputCbPtr = image_Cb;
unsigned char* pucOutputCrPtr = image_Cr;
unsigned int uiTmp1,uiTmp2;
unsigned int ui1_1,ui1_2,ui2_1,ui2_2,res_1,res_2,res_3,res_4;
unsigned int tmpend1_1,tmpend1_2,tmpend2_1,tmpend2_2;
unsigned int uiA,uiB,uiC,uiD;
unsigned int uiE,uiF,uiG,uiH;
unsigned int uicst = xFracl * yFracl;
uiTmp1 = _pack2(uicst,uicst);
uiTmp2 = (_pack2(xFracl,yFracl)) << 3;
uiTmp2 = _sub2(uiTmp2,uiTmp1);
uiTmp1 = (uicst) + ((uicst - ((xFracl + yFracl) <<3) + 64) << 16);
uicst = _packh2(uiTmp1,uiTmp2); // cst2 cst3
uiTmp1 = _pack2(uiTmp2,uiTmp1); // cst4 cst1
uicst = _spacku4(uicst,uiTmp1);
uiA = _mem2(pucCbPtrA);
uiB = _mem2(pucCbPtrB);
uiC = _mem2(pucCbPtrC);
uiD = _mem2(pucCbPtrD);
uiE = _mem2(pucCrPtrE);
uiF = _mem2(pucCrPtrF);
uiG = _mem2(pucCrPtrG);
uiH = _mem2(pucCrPtrH);
pucCbPtrA += PicWidthSamples;
pucCbPtrB += PicWidthSamples;
pucCbPtrC += PicWidthSamples;
pucCbPtrD += PicWidthSamples;
pucCrPtrE += PicWidthSamples;
pucCrPtrF += PicWidthSamples;
pucCrPtrG += PicWidthSamples;
pucCrPtrH += PicWidthSamples;
uiA += (_mem2(pucCbPtrA) << 16);
uiB += (_mem2(pucCbPtrB) << 16);
uiC += (_mem2(pucCbPtrC) << 16);
uiD += (_mem2(pucCbPtrD) << 16);
uiE += (_mem2(pucCrPtrE) << 16);
uiF += (_mem2(pucCrPtrF) << 16);
uiG += (_mem2(pucCrPtrG) << 16);
uiH += (_mem2(pucCrPtrH) << 16);
uiTmp1 = _packh4(uiA,uiB);
uiTmp2 = _packh4(uiC,uiD);
ui1_1 = _packh4(uiTmp1,uiTmp2);
ui2_1 = _packl4(uiTmp1,uiTmp2);
uiTmp1 = _packl4(uiA,uiB);
uiTmp2 = _packl4(uiC,uiD);
ui1_2 = _packh4(uiTmp1,uiTmp2);
ui2_2 = _packl4(uiTmp1,uiTmp2);
tmpend1_1 = _dotpu4(uicst,ui1_1);
tmpend1_2 = _dotpu4(uicst,ui1_2);
tmpend2_1 = _dotpu4(uicst,ui2_1);
tmpend2_2 = _dotpu4(uicst,ui2_2);
res_1 = _pack2(tmpend1_1,tmpend1_2);
res_2 = _pack2(tmpend2_1,tmpend2_2);
res_1 = _shr2(_sadd2(res_1,0x00200020),6);
res_2 = _shr2(_sadd2(res_2,0x00200020),6);
res_1 = _spacku4(0x00000000,res_1);
res_2 = _spacku4(0x00000000,res_2);
_mem2(pucOutputCbPtr) = res_2;
pucOutputCbPtr += stride;
_mem2(pucOutputCbPtr) = res_1;
uiTmp1 = _packh4(uiE,uiF);
uiTmp2 = _packh4(uiG,uiH);
ui1_1 = _packh4(uiTmp1,uiTmp2);
ui2_1 = _packl4(uiTmp1,uiTmp2);
uiTmp1 = _packl4(uiE,uiF);
uiTmp2 = _packl4(uiG,uiH);
ui1_2 = _packh4(uiTmp1,uiTmp2);
ui2_2 = _packl4(uiTmp1,uiTmp2);
tmpend1_1 = _dotpu4(uicst,ui1_1);
tmpend1_2 = _dotpu4(uicst,ui1_2);
tmpend2_1 = _dotpu4(uicst,ui2_1);
tmpend2_2 = _dotpu4(uicst,ui2_2);
res_3 = _pack2(tmpend1_1,tmpend1_2);
res_4 = _pack2(tmpend2_1,tmpend2_2);
res_3 = _shr2(_sadd2(res_3,0x00200020),6);
res_4 = _shr2(_sadd2(res_4,0x00200020),6);
res_3 = _spacku4(0x00000000,res_3);
res_4 = _spacku4(0x00000000,res_4);
_mem2(pucOutputCrPtr) = res_4;
pucOutputCrPtr += stride;
_mem2(pucOutputCrPtr) = res_3;
}
void DSP_QMFS_process(DSP_QMFS_bank_t *QMFS_bank_obj)
{
sint32 i,j;
uint32 * restrict filter = (uint32 *)QMFS_bank_obj->flt_ptr;
uint32 * restrict data_ptr0 = (uint32 *)QMFS_bank_obj->data_in_buffer_LO; /*input buffer bank0*/
uint32 * restrict data_ptr1 = (uint32 *)QMFS_bank_obj->data_in_buffer_HI; /*input buffer bank0*/
sint32 L = QMFS_bank_obj->blk_len;
sint32 M = FILT_LEN/2;
uint32 * restrict hist_rd_ptr0 = (uint32 *)QMFS_bank_obj->history[0];
uint32 * restrict hist_rd_ptr1 = (uint32 *)QMFS_bank_obj->history[1];
uint32 * restrict hist_wr_ptr0 = (uint32 *)QMFS_bank_obj->history[0];
uint32 * restrict hist_wr_ptr1 = (uint32 *)QMFS_bank_obj->history[1];
uint32 * restrict dataout_ptr = (uint32 *)QMFS_bank_obj->data_out_ptr;
for(i=0;i<L;++i)
{
/* iteration 0 to (L-1) produces the L outputs of the current block*/
register uint32 t0,t1,out;
register __int40_t accum0l, accum1l, accum0r, accum1r;
register long long temp;
accum0l = 0; accum0r = 0;
accum1l = 0; accum1r = 0;
uint32 *dptr0 = data_ptr0 + i ;
uint32 *dptr1 = data_ptr1 + i ;
uint32 *fptr = filter;
#pragma MUST_ITERATE( 1)
for(j=0; j<QMFS_bank_obj->iter_count[i];++j)
{
/*j indexes from the point of rightmost overlap to the leftmost overlap position */
register uint32 d0,d1;
register uint32 filt_coef10 = *fptr++; /*Loads the decimated filter coeficients*/
register uint32 filt_coef0 = _pack2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
register uint32 filt_coef1 = _packh2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
d1 = *dptr1--;//data_ptr1[(2*(i-j))]; /*load the data right to left with 1:2 split*/
d0 = *dptr0--;//data_ptr0[(2*(i-j))]; /*load the data right to left with 1:2 split*/
t0 = _add2(d0,d1);
t1 = _sub2(d0,d1);
temp = _mpy2ll(t0, filt_coef0);
accum0l = _lsadd(_loll(temp), accum0l); accum0r = _lsadd(_hill(temp), accum0r);
temp = _mpy2ll(t1, filt_coef1);
accum1l = _lsadd(_loll(temp), accum1l); accum1r = _lsadd(_hill(temp), accum1r);
}
t0 = _pack2( _sat((accum0l + 16384)>>15), _sat((accum0l + 16384)>>15));
t1 = _pack2( _sat((accum1l + 16384)>>15), _sat((accum1l + 16384)>>15));
out = _add2(t0, hist_rd_ptr0[i]);
dataout_ptr[2*i + 0] = out; /*filter 0 produces the odd output*/
out = _add2(t1, hist_rd_ptr1[i]);
dataout_ptr[2*i + 1] = out; /*filter 1 produces the even output*/
}
for(i=L;i<L+M-1;++i)
{
register uint32 t0,t1,out;
register __int40_t accum0l, accum1l, accum0r, accum1r;
register long long temp;
accum0l = 0; accum0r = 0;
accum1l = 0; accum1r = 0;
uint32 *dptr0 = data_ptr0 + (L-1) ;
uint32 *dptr1 = data_ptr1 + (L-1) ;
uint32 *fptr = filter + (i-L+1);
#pragma MUST_ITERATE( 1)
for(j=0; j<QMFS_bank_obj->iter_count[i];++j)
{
/*same logic for j starting at the
rightmost point of overlap to leftmost*/
register sint32 d0,d1;
register sint32 filt_coef10 = *fptr++;; /*Loads the decimated filter coeficients*/
register sint32 filt_coef0 = _pack2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
register sint32 filt_coef1 = _packh2(filt_coef10,filt_coef10); /*in the correct order i.e. with reversal*/
d1 = *dptr1--; /*load the data right to left with 1:2 split*/
d0 = *dptr0--; /*load the data right to left with 1:2 split*/
t0 = _add2(d0,d1);
t1 = _sub2(d0,d1);
temp = _mpy2ll(t0, filt_coef0);
accum0l = _lsadd(_loll(temp), accum0l); accum0r = _lsadd(_hill(temp), accum0r);
temp = _mpy2ll(t1, filt_coef1);
accum1l = _lsadd(_loll(temp), accum1l); accum1r = _lsadd(_hill(temp), accum1r);
}
t0 = _pack2( _sat((accum0l + 16384)>>15), _sat((accum0l + 16384)>>15));
t1 = _pack2( _sat((accum1l + 16384)>>15), _sat((accum1l + 16384)>>15));
out = _add2(t0, hist_rd_ptr0[i]);
hist_wr_ptr0[(i-L)] = out; /* write out overlap add history history filter 0*/
out = _add2(t1, hist_rd_ptr1[i]);
hist_wr_ptr1[(i-L)] = out; /* write out overlap add history history filter 1*/
}
return;
}
