dengzhihao
/
CMSIS-DSP
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

CMSIS-DSP: Corrected tail management bug in MVE version of arm_fir_f32.c

Browse Source
pull/19/head
GuentherMartin 6 years ago
parent 78d7e66bc9
commit f51e95f5a8
1 changed files with 69 additions and 67 deletions
Show Stats Download Patch File Download Diff File

136
Source/FilteringFunctions/arm_fir_f32.c
Unescape Escape View File

@ -188,7 +188,9 @@ static void arm_fir_f32_1_4_mve(const arm_fir_instance_f32 * S, const float32_t
blkCnt--; blkCnt--;
} }
blkCnt = blockSize & 3; blkCnt = blockSize & 3;
if (blkCnt > 0U)
{ {
mve_pred16_t p0 = vctp32q(blkCnt); mve_pred16_t p0 = vctp32q(blkCnt);
@ -217,6 +219,7 @@ static void arm_fir_f32_1_4_mve(const arm_fir_instance_f32 * S, const float32_t
*/ */
pTempSrc = &S->pState[blockSize]; pTempSrc = &S->pState[blockSize];
pTempDest = S->pState; pTempDest = S->pState;
blkCnt = numTaps >> 2; blkCnt = numTaps >> 2;
while (blkCnt > 0U) while (blkCnt > 0U)
{ {
@ -225,6 +228,7 @@ static void arm_fir_f32_1_4_mve(const arm_fir_instance_f32 * S, const float32_t
pTempDest += 4; pTempDest += 4;
blkCnt--; blkCnt--;
} }
blkCnt = numTaps & 3; blkCnt = numTaps & 3;
if (blkCnt > 0U) if (blkCnt > 0U)
{ {
@ -234,7 +238,6 @@ static void arm_fir_f32_1_4_mve(const arm_fir_instance_f32 * S, const float32_t
} }
static void arm_fir_f32_5_8_mve(const arm_fir_instance_f32 * S, const float32_t * pSrc, float32_t * pDst, uint32_t blockSize) static void arm_fir_f32_5_8_mve(const arm_fir_instance_f32 * S, const float32_t * pSrc, float32_t * pDst, uint32_t blockSize)
{ {
float32_t *pState = S->pState; /* State pointer */ float32_t *pState = S->pState; /* State pointer */
@ -314,9 +317,8 @@ static void arm_fir_f32_5_8_mve(const arm_fir_instance_f32 * S, const float32_t
} }
blkCnt = blockSize & 3; blkCnt = blockSize & 3;
if (blkCnt > 0) if (blkCnt > 0U)
{ {
mve_pred16_t p0 = vctp32q(blkCnt); mve_pred16_t p0 = vctp32q(blkCnt);
vstrwq_p_f32(pStateCur, vld1q(pTempSrc),p0); vstrwq_p_f32(pStateCur, vld1q(pTempSrc),p0);
@ -364,6 +366,7 @@ static void arm_fir_f32_5_8_mve(const arm_fir_instance_f32 * S, const float32_t
pTempDest += 4; pTempDest += 4;
blkCnt--; blkCnt--;
} }
blkCnt = numTaps & 3; blkCnt = numTaps & 3;
if (blkCnt > 0U) if (blkCnt > 0U)
{ {
@ -372,6 +375,7 @@ static void arm_fir_f32_5_8_mve(const arm_fir_instance_f32 * S, const float32_t
} }
} }
void arm_fir_f32( void arm_fir_f32(
const arm_fir_instance_f32 * S, const arm_fir_instance_f32 * S,
const float32_t * pSrc, const float32_t * pSrc,
@ -413,9 +417,6 @@ uint32_t blockSize)
} }
} }
if (blockSize >= 8) if (blockSize >= 8)
{ {
/* /*
@ -432,7 +433,7 @@ uint32_t blockSize)
{ {
int32_t i; int32_t i;
const float32_t *pCoeffsCur = pCoeffs; const float32_t *pCoeffsCur = pCoeffs;
c0 = *pCoeffsCur++; c0 = *pCoeffsCur++;
c1 = *pCoeffsCur++; c1 = *pCoeffsCur++;
c2 = *pCoeffsCur++; c2 = *pCoeffsCur++;
@ -441,37 +442,37 @@ uint32_t blockSize)
c5 = *pCoeffsCur++; c5 = *pCoeffsCur++;
c6 = *pCoeffsCur++; c6 = *pCoeffsCur++;
c7 = *pCoeffsCur++; c7 = *pCoeffsCur++;
vst1q(pStateCur, vld1q(pTempSrc)); vst1q(pStateCur, vld1q(pTempSrc));
pStateCur += 4; pStateCur += 4;
pTempSrc += 4; pTempSrc += 4;
vecIn0 = vld1q(pSamples); vecIn0 = vld1q(pSamples);
vecAcc0 = vmulq(vecIn0, c0); vecAcc0 = vmulq(vecIn0, c0);
vecIn0 = vld1q(&pSamples[1]); vecIn0 = vld1q(&pSamples[1]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c1); vecAcc0 = vfmaq(vecAcc0, vecIn0, c1);
vecIn0 = vld1q(&pSamples[2]); vecIn0 = vld1q(&pSamples[2]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c2); vecAcc0 = vfmaq(vecAcc0, vecIn0, c2);
vecIn0 = vld1q(&pSamples[3]); vecIn0 = vld1q(&pSamples[3]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c3); vecAcc0 = vfmaq(vecAcc0, vecIn0, c3);
vecIn0 = vld1q(&pSamples[4]); vecIn0 = vld1q(&pSamples[4]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c4); vecAcc0 = vfmaq(vecAcc0, vecIn0, c4);
vecIn0 = vld1q(&pSamples[5]); vecIn0 = vld1q(&pSamples[5]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c5); vecAcc0 = vfmaq(vecAcc0, vecIn0, c5);
vecIn0 = vld1q(&pSamples[6]); vecIn0 = vld1q(&pSamples[6]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c6); vecAcc0 = vfmaq(vecAcc0, vecIn0, c6);
vecIn0 = vld1q(&pSamples[7]); vecIn0 = vld1q(&pSamples[7]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c7); vecAcc0 = vfmaq(vecAcc0, vecIn0, c7);
pSamples += 8; pSamples += 8;
numCnt = ((int32_t)numTaps - 8) / 8; numCnt = ((int32_t)numTaps - 8) / 8;
for (i = 0; i < numCnt; i++) for (i = 0; i < numCnt; i++)
@ -484,31 +485,31 @@ uint32_t blockSize)
c5 = *pCoeffsCur++; c5 = *pCoeffsCur++;
c6 = *pCoeffsCur++; c6 = *pCoeffsCur++;
c7 = *pCoeffsCur++; c7 = *pCoeffsCur++;
vecIn0 = vld1q(pSamples); vecIn0 = vld1q(pSamples);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c0); vecAcc0 = vfmaq(vecAcc0, vecIn0, c0);
vecIn0 = vld1q(&pSamples[1]); vecIn0 = vld1q(&pSamples[1]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c1); vecAcc0 = vfmaq(vecAcc0, vecIn0, c1);
vecIn0 = vld1q(&pSamples[2]); vecIn0 = vld1q(&pSamples[2]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c2); vecAcc0 = vfmaq(vecAcc0, vecIn0, c2);
vecIn0 = vld1q(&pSamples[3]); vecIn0 = vld1q(&pSamples[3]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c3); vecAcc0 = vfmaq(vecAcc0, vecIn0, c3);
vecIn0 = vld1q(&pSamples[4]); vecIn0 = vld1q(&pSamples[4]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c4); vecAcc0 = vfmaq(vecAcc0, vecIn0, c4);
vecIn0 = vld1q(&pSamples[5]); vecIn0 = vld1q(&pSamples[5]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c5); vecAcc0 = vfmaq(vecAcc0, vecIn0, c5);
vecIn0 = vld1q(&pSamples[6]); vecIn0 = vld1q(&pSamples[6]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c6); vecAcc0 = vfmaq(vecAcc0, vecIn0, c6);
vecIn0 = vld1q(&pSamples[7]); vecIn0 = vld1q(&pSamples[7]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c7); vecAcc0 = vfmaq(vecAcc0, vecIn0, c7);
pSamples += 8; pSamples += 8;
} }
@ -523,24 +524,25 @@ uint32_t blockSize)
numCnt --; numCnt --;
} }
vst1q(pOutput, vecAcc0); vst1q(pOutput, vecAcc0);
pOutput += 4; pOutput += 4;
pSamples = pSamples - numTaps + 4; pSamples = pSamples - numTaps + 4;
blkCnt--; blkCnt--;
} }
blkCnt = blockSize & 3; blkCnt = blockSize & 3;
if (blkCnt > 0U)
{ {
mve_pred16_t p0 = vctp32q(blkCnt); mve_pred16_t p0 = vctp32q(blkCnt);
int32_t i; int32_t i;
const float32_t *pCoeffsCur = pCoeffs; const float32_t *pCoeffsCur = pCoeffs;
vst1q(pStateCur, vld1q(pTempSrc)); vst1q(pStateCur, vld1q(pTempSrc));
pStateCur += 4; pStateCur += 4;
pTempSrc += 4; pTempSrc += 4;
c0 = *pCoeffsCur++; c0 = *pCoeffsCur++;
c1 = *pCoeffsCur++; c1 = *pCoeffsCur++;
c2 = *pCoeffsCur++; c2 = *pCoeffsCur++;
@ -549,33 +551,33 @@ uint32_t blockSize)
c5 = *pCoeffsCur++; c5 = *pCoeffsCur++;
c6 = *pCoeffsCur++; c6 = *pCoeffsCur++;
c7 = *pCoeffsCur++; c7 = *pCoeffsCur++;
vecIn0 = vld1q(pSamples); vecIn0 = vld1q(pSamples);
vecAcc0 = vmulq(vecIn0, c0); vecAcc0 = vmulq(vecIn0, c0);
vecIn0 = vld1q(&pSamples[1]); vecIn0 = vld1q(&pSamples[1]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c1); vecAcc0 = vfmaq(vecAcc0, vecIn0, c1);
vecIn0 = vld1q(&pSamples[2]); vecIn0 = vld1q(&pSamples[2]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c2); vecAcc0 = vfmaq(vecAcc0, vecIn0, c2);
vecIn0 = vld1q(&pSamples[3]); vecIn0 = vld1q(&pSamples[3]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c3); vecAcc0 = vfmaq(vecAcc0, vecIn0, c3);
vecIn0 = vld1q(&pSamples[4]); vecIn0 = vld1q(&pSamples[4]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c4); vecAcc0 = vfmaq(vecAcc0, vecIn0, c4);
vecIn0 = vld1q(&pSamples[5]); vecIn0 = vld1q(&pSamples[5]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c5); vecAcc0 = vfmaq(vecAcc0, vecIn0, c5);
vecIn0 = vld1q(&pSamples[6]); vecIn0 = vld1q(&pSamples[6]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c6); vecAcc0 = vfmaq(vecAcc0, vecIn0, c6);
vecIn0 = vld1q(&pSamples[7]); vecIn0 = vld1q(&pSamples[7]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c7); vecAcc0 = vfmaq(vecAcc0, vecIn0, c7);
pSamples += 8; pSamples += 8;
numCnt = ((int32_t)numTaps - 8) / 8; numCnt = ((int32_t)numTaps - 8) / 8;
for (i = 0; i < numCnt; i++) for (i = 0; i < numCnt; i++)
@ -588,31 +590,31 @@ uint32_t blockSize)
c5 = *pCoeffsCur++; c5 = *pCoeffsCur++;
c6 = *pCoeffsCur++; c6 = *pCoeffsCur++;
c7 = *pCoeffsCur++; c7 = *pCoeffsCur++;
vecIn0 = vld1q(pSamples); vecIn0 = vld1q(pSamples);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c0); vecAcc0 = vfmaq(vecAcc0, vecIn0, c0);
vecIn0 = vld1q(&pSamples[1]); vecIn0 = vld1q(&pSamples[1]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c1); vecAcc0 = vfmaq(vecAcc0, vecIn0, c1);
vecIn0 = vld1q(&pSamples[2]); vecIn0 = vld1q(&pSamples[2]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c2); vecAcc0 = vfmaq(vecAcc0, vecIn0, c2);
vecIn0 = vld1q(&pSamples[3]); vecIn0 = vld1q(&pSamples[3]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c3); vecAcc0 = vfmaq(vecAcc0, vecIn0, c3);
vecIn0 = vld1q(&pSamples[4]); vecIn0 = vld1q(&pSamples[4]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c4); vecAcc0 = vfmaq(vecAcc0, vecIn0, c4);
vecIn0 = vld1q(&pSamples[5]); vecIn0 = vld1q(&pSamples[5]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c5); vecAcc0 = vfmaq(vecAcc0, vecIn0, c5);
vecIn0 = vld1q(&pSamples[6]); vecIn0 = vld1q(&pSamples[6]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c6); vecAcc0 = vfmaq(vecAcc0, vecIn0, c6);
vecIn0 = vld1q(&pSamples[7]); vecIn0 = vld1q(&pSamples[7]);
vecAcc0 = vfmaq(vecAcc0, vecIn0, c7); vecAcc0 = vfmaq(vecAcc0, vecIn0, c7);
pSamples += 8; pSamples += 8;
} }
@ -627,8 +629,7 @@ uint32_t blockSize)
numCnt --; numCnt --;
} }
vstrwq_p_f32(pOutput, vecAcc0, p0); vstrwq_p_f32(pOutput, vecAcc0, p0);
} }
} }
@ -641,40 +642,40 @@ uint32_t blockSize)
uint32_t numTaps = S->numTaps; /* Number of filter coefficients in the filter */ uint32_t numTaps = S->numTaps; /* Number of filter coefficients in the filter */
uint32_t i, blkCnt; /* Loop counters */ uint32_t i, blkCnt; /* Loop counters */
pStateCurnt = &(S->pState[(numTaps - 1U)]); pStateCurnt = &(S->pState[(numTaps - 1U)]);
blkCnt = blockSize; blkCnt = blockSize;
while (blkCnt > 0U) while (blkCnt > 0U)
{ {
/* Copy one sample at a time into state buffer */ /* Copy one sample at a time into state buffer */
*pStateCurnt++ = *pSrc++; *pStateCurnt++ = *pSrc++;
/* Set the accumulator to zero */ /* Set the accumulator to zero */
acc0 = 0.0f; acc0 = 0.0f;
/* Initialize state pointer */ /* Initialize state pointer */
px = pState; px = pState;
/* Initialize Coefficient pointer */ /* Initialize Coefficient pointer */
pb = pCoeffs; pb = pCoeffs;
i = numTaps; i = numTaps;
/* Perform the multiply-accumulates */ /* Perform the multiply-accumulates */
while (i > 0U) while (i > 0U)
{ {
/* acc = b[numTaps-1] * x[n-numTaps-1] + b[numTaps-2] * x[n-numTaps-2] + b[numTaps-3] * x[n-numTaps-3] +...+ b[0] * x[0] */ /* acc = b[numTaps-1] * x[n-numTaps-1] + b[numTaps-2] * x[n-numTaps-2] + b[numTaps-3] * x[n-numTaps-3] +...+ b[0] * x[0] */
acc0 += *px++ * *pb++; acc0 += *px++ * *pb++;
i--; i--;
} }
/* Store result in destination buffer. */ /* Store result in destination buffer. */
*pDst++ = acc0; *pDst++ = acc0;
/* Advance state pointer by 1 for the next sample */ /* Advance state pointer by 1 for the next sample */
pState = pState + 1U; pState = pState + 1U;
/* Decrement loop counter */ /* Decrement loop counter */
blkCnt--; blkCnt--;
} }
@ -694,6 +695,7 @@ uint32_t blockSize)
pTempDest += 4; pTempDest += 4;
blkCnt--; blkCnt--;
} }
blkCnt = numTaps & 3; blkCnt = numTaps & 3;
if (blkCnt > 0U) if (blkCnt > 0U)
{ {
@ -773,7 +775,7 @@ uint32_t blockSize)
xa = vextq_f32(x0,x1,1); xa = vextq_f32(x0,x1,1);
xb = vextq_f32(x1,x2,1); xb = vextq_f32(x1,x2,1);
accv0 = vmlaq_n_f32(accv0,xa,vgetq_lane_f32(b, 1)); accv0 = vmlaq_n_f32(accv0,xa,vgetq_lane_f32(b, 1));
accv1 = vmlaq_n_f32(accv1,xb,vgetq_lane_f32(b, 1)); accv1 = vmlaq_n_f32(accv1,xb,vgetq_lane_f32(b, 1));
@ -785,7 +787,7 @@ uint32_t blockSize)
xa = vextq_f32(x0,x1,3); xa = vextq_f32(x0,x1,3);
xb = vextq_f32(x1,x2,3); xb = vextq_f32(x1,x2,3);
accv0 = vmlaq_n_f32(accv0,xa,vgetq_lane_f32(b, 3)); accv0 = vmlaq_n_f32(accv0,xa,vgetq_lane_f32(b, 3));
accv1 = vmlaq_n_f32(accv1,xb,vgetq_lane_f32(b, 3)); accv1 = vmlaq_n_f32(accv1,xb,vgetq_lane_f32(b, 3));
@ -821,7 +823,7 @@ uint32_t blockSize)
xa = vextq_f32(x0,x1,2); xa = vextq_f32(x0,x1,2);
xb = vextq_f32(x1,x2,2); xb = vextq_f32(x1,x2,2);
accv0 = vmlaq_n_f32(accv0,xa,*pb); accv0 = vmlaq_n_f32(accv0,xa,*pb);
accv1 = vmlaq_n_f32(accv1,xb,*pb); accv1 = vmlaq_n_f32(accv1,xb,*pb);
@ -836,7 +838,7 @@ uint32_t blockSize)
xa = vextq_f32(x0,x1,1); xa = vextq_f32(x0,x1,1);
xb = vextq_f32(x1,x2,1); xb = vextq_f32(x1,x2,1);
accv0 = vmlaq_n_f32(accv0,xa,*pb); accv0 = vmlaq_n_f32(accv0,xa,*pb);
accv1 = vmlaq_n_f32(accv1,xb,*pb); accv1 = vmlaq_n_f32(accv1,xb,*pb);
@ -844,7 +846,7 @@ uint32_t blockSize)
break; break;
case 1: case 1:
{ {
accv0 = vmlaq_n_f32(accv0,x0,*pb); accv0 = vmlaq_n_f32(accv0,x0,*pb);
accv1 = vmlaq_n_f32(accv1,x1,*pb); accv1 = vmlaq_n_f32(accv1,x1,*pb);

Write Preview
Loading…
Cancel
Save