277 lines
8.0 KiB
C
277 lines
8.0 KiB
C
|
#include "riscv_const_structs.h"
|
||
|
|
#include "ref.h"
|
||
|
|
|
||
|
|
void ref_rfft_f32(riscv_rfft_instance_f32 *S, float32_t *pSrc, float32_t *pDst)
|
||
|
|
{
|
||
|
|
uint32_t i;
|
||
|
|
|
||
|
|
if (S->ifftFlagR) {
|
||
|
|
for (i = 0; i < S->fftLenReal * 2; i++) {
|
||
|
|
pDst[i] = pSrc[i];
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
pDst[2 * i + 0] = pSrc[i];
|
||
|
|
pDst[2 * i + 1] = 0.0f;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
switch (S->fftLenReal) {
|
||
|
|
case 128:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len128, pDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 512:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len512, pDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 2048:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len2048, pDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 8192:
|
||
|
|
ref_cfft_f32(&ref_cfft_sR_f32_len8192, pDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
|
||
|
|
if (S->ifftFlagR) {
|
||
|
|
// throw away the imaginary part which should be all zeros
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
pDst[i] = pDst[2 * i];
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
void ref_rfft_fast_f32(riscv_rfft_fast_instance_f32 *S, float32_t *p,
|
||
|
|
float32_t *pOut, uint8_t ifftFlag)
|
||
|
|
{
|
||
|
|
uint32_t i, j;
|
||
|
|
|
||
|
|
if (ifftFlag) {
|
||
|
|
for (i = 0; i < S->fftLenRFFT; i++) {
|
||
|
|
pOut[i] = p[i];
|
||
|
|
}
|
||
|
|
// unpack first sample's complex part into middle sample's real part
|
||
|
|
pOut[S->fftLenRFFT] = pOut[1];
|
||
|
|
pOut[S->fftLenRFFT + 1] = 0;
|
||
|
|
pOut[1] = 0;
|
||
|
|
j = 4;
|
||
|
|
for (i = S->fftLenRFFT / 2 + 1; i < S->fftLenRFFT; i++) {
|
||
|
|
pOut[2 * i + 0] = p[2 * i + 0 - j];
|
||
|
|
pOut[2 * i + 1] = -p[2 * i + 1 - j];
|
||
|
|
j += 4;
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
for (i = 0; i < S->fftLenRFFT; i++) {
|
||
|
|
pOut[2 * i + 0] = p[i];
|
||
|
|
pOut[2 * i + 1] = 0.0f;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
switch (S->fftLenRFFT) {
|
||
|
|
case 32:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len32, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 64:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len64, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 128:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len128, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 256:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len256, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 512:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len512, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 1024:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len1024, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 2048:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len2048, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 4096:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len4096, pOut, ifftFlag, 1);
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
|
||
|
|
if (ifftFlag) {
|
||
|
|
// throw away the imaginary part which should be all zeros
|
||
|
|
for (i = 0; i < S->fftLenRFFT; i++) {
|
||
|
|
pOut[i] = pOut[2 * i];
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
// pack last sample's real part into first sample's complex part
|
||
|
|
pOut[1] = pOut[S->fftLenRFFT];
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
void ref_rfft_q31(const riscv_rfft_instance_q31 *S, q31_t *pSrc, q31_t *pDst)
|
||
|
|
{
|
||
|
|
uint32_t i;
|
||
|
|
// float32_t *fDst = (float32_t*)pDst;
|
||
|
|
float32_t fDst[S->fftLenReal * 2];
|
||
|
|
riscv_q31_to_float(pSrc, fDst, S->fftLenReal * 2);
|
||
|
|
|
||
|
|
if (S->ifftFlagR) {
|
||
|
|
for (i = 0; i < S->fftLenReal * 2; i++) {
|
||
|
|
fDst[i] = (float32_t)pSrc[i] / 2147483648.0f;
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
fDst[2 * i + 0] = (float32_t)pSrc[i] / 2147483648.0f;
|
||
|
|
fDst[2 * i + 1] = 0.0f;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
switch (S->fftLenReal) {
|
||
|
|
case 32:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len32, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 64:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len64, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 128:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len128, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 256:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len256, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 512:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len512, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 1024:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len1024, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 2048:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len2048, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 4096:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len4096, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 8192:
|
||
|
|
ref_cfft_f32(&ref_cfft_sR_f32_len8192, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
|
||
|
|
if (S->ifftFlagR) {
|
||
|
|
// throw away the imaginary part which should be all zeros
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
// read the float data, scale up for q31, cast to q31
|
||
|
|
pDst[i] = (q31_t)(fDst[2 * i] * 2147483648.0f);
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
// read the float data, scale up for q31, cast to q31
|
||
|
|
pDst[i] =
|
||
|
|
(q31_t)(fDst[i] * 2147483648.0f / (float32_t)S->fftLenReal);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
void ref_rfft_q15(const riscv_rfft_instance_q15 *S, q15_t *pSrc, q15_t *pDst)
|
||
|
|
{
|
||
|
|
uint32_t i;
|
||
|
|
// float32_t *fDst = (float32_t*)pDst;
|
||
|
|
float32_t fDst[S->fftLenReal * 2];
|
||
|
|
riscv_q15_to_float(pSrc, fDst, S->fftLenReal * 2);
|
||
|
|
|
||
|
|
if (S->ifftFlagR) {
|
||
|
|
for (i = 0; i < S->fftLenReal * 2; i++) {
|
||
|
|
fDst[i] = (float32_t)pSrc[i] / 32768.0f;
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
// read the q15 data, cast to float, scale down for float
|
||
|
|
fDst[2 * i + 0] = (float32_t)pSrc[i] / 32768.0f;
|
||
|
|
fDst[2 * i + 1] = 0.0f;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
switch (S->fftLenReal) {
|
||
|
|
case 32:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len32, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 64:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len64, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 128:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len128, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 256:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len256, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 512:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len512, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 1024:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len1024, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 2048:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len2048, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 4096:
|
||
|
|
ref_cfft_f32(&riscv_cfft_sR_f32_len4096, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
|
||
|
|
case 8192:
|
||
|
|
ref_cfft_f32(&ref_cfft_sR_f32_len8192, fDst, S->ifftFlagR,
|
||
|
|
S->bitReverseFlagR);
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
|
||
|
|
if (S->ifftFlagR) {
|
||
|
|
// throw away the imaginary part which should be all zeros
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
pDst[i] = (q15_t)(fDst[2 * i] * 32768.0f);
|
||
|
|
}
|
||
|
|
} else {
|
||
|
|
for (i = 0; i < S->fftLenReal; i++) {
|
||
|
|
pDst[i] = (q15_t)(fDst[i] * 32768.0f / (float32_t)S->fftLenReal);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|