/**
\brief Test case: TC_CoreSimd_Part9
\details
- Check Packing Halfword:
__PKHBT
__PKHTB
*/
void TC_CoreSimd_Pack16 (void) {
#if ((defined (__ARM_ARCH_7EM__ ) && (__ARM_ARCH_7EM__ == 1)) || \
(defined (__ARM_FEATURE_DSP) && (__ARM_FEATURE_DSP == 1)) )
volatile uint32_t op1_u32, op2_u32;
volatile uint32_t res_u32;
/* --- __PKHBT Test ---------------------------------------------- */
op1_u32 = 0x00000111;
op2_u32 = 0x22200000;
res_u32 = __PKHBT(op1_u32, op2_u32, 0);
ASSERT_TRUE(res_u32 == 0x22200111);
op1_u32 = 0x00000111;
op2_u32 = 0x22200000;
res_u32 = __PKHBT(op1_u32, op2_u32, 4);
ASSERT_TRUE(res_u32 == 0x22000111);
/* --- __PKHTB Test ---------------------------------------------- */
op1_u32 = 0x11100000;
op2_u32 = 0x00000222;
res_u32 = __PKHTB(op1_u32, op2_u32, 0);
ASSERT_TRUE(res_u32 == 0x11100222);
op1_u32 = 0x11100000;
op2_u32 = 0x00000222;
res_u32 = __PKHTB(op1_u32, op2_u32, 4);
ASSERT_TRUE(res_u32 == 0x11100022);
#endif
}
void arm_pid_init_q15(
arm_pid_instance_q15 * S,
int32_t resetStateFlag)
{
#ifndef ARM_MATH_CM0_FAMILY
/* Run the below code for Cortex-M4 and Cortex-M3 */
/* Derived coefficient A0 */
S->A0 = __QADD16(__QADD16(S->Kp, S->Ki), S->Kd);
/* Derived coefficients and pack into A1 */
#ifndef ARM_MATH_BIG_ENDIAN
S->A1 = __PKHBT(-__QADD16(__QADD16(S->Kd, S->Kd), S->Kp), S->Kd, 16);
#else
S->A1 = __PKHBT(S->Kd, -__QADD16(__QADD16(S->Kd, S->Kd), S->Kp), 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
/* Check whether state needs reset or not */
if(resetStateFlag)
{
/* Clear the state buffer. The size will be always 3 samples */
memset(S->state, 0, 3u * sizeof(q15_t));
}
#else
/* Run the below code for Cortex-M0 */
q31_t temp; /*to store the sum */
/* Derived coefficient A0 */
temp = S->Kp + S->Ki + S->Kd;
S->A0 = (q15_t) __SSAT(temp, 16);
/* Derived coefficients and pack into A1 */
temp = -(S->Kd + S->Kd + S->Kp);
S->A1 = (q15_t) __SSAT(temp, 16);
S->A2 = S->Kd;
/* Check whether state needs reset or not */
if(resetStateFlag)
{
/* Clear the state buffer. The size will be always 3 samples */
memset(S->state, 0, 3u * sizeof(q15_t));
}
#endif /* #ifndef ARM_MATH_CM0_FAMILY */
}
buffer_c16_t Complex8DecimateBy2CIC3::execute(const buffer_c8_t& src, const buffer_c16_t& dst) {
/* Decimates by two using a non-recursive third-order CIC filter.
*/
/* CIC filter (decimating by two):
* D_I0 = i3 * 1 + i2 * 3 + i1 * 3 + i0 * 1
* D_Q0 = q3 * 1 + q2 * 3 + q1 * 3 + q0 * 1
*
* D_I1 = i5 * 1 + i4 * 3 + i3 * 3 + i2 * 1
* D_Q1 = q5 * 1 + q4 * 3 + q3 * 3 + q2 * 1
*/
uint32_t i1_i0 = _i1_i0;
uint32_t q1_q0 = _q1_q0;
/* 3:1 Scaled by 32 to normalize output to +/-32768-ish. */
constexpr uint32_t scale_factor = 32;
constexpr uint32_t k_3_1 = 0x00030001 * scale_factor;
uint32_t* src_p = reinterpret_cast<uint32_t*>(&src.p[0]);
uint32_t* const src_end = reinterpret_cast<uint32_t*>(&src.p[src.count]);
uint32_t* dst_p = reinterpret_cast<uint32_t*>(&dst.p[0]);
while(src_p < src_end) {
const uint32_t q3_i3_q2_i2 = *(src_p++); // 3
const uint32_t q5_i5_q4_i4 = *(src_p++);
const uint32_t d_i0_partial = __SMUAD(k_3_1, i1_i0); // 1: = 3 * i1 + 1 * i0
const uint32_t i3_i2 = __SXTB16(q3_i3_q2_i2, 0); // 1: (q3_i3_q2_i2 ror 0)[23:16]:(q3_i3_q2_i2 ror 0)[7:0]
const uint32_t d_i0 = __SMLADX(k_3_1, i3_i2, d_i0_partial); // 1: + 3 * i2 + 1 * i3
const uint32_t d_q0_partial = __SMUAD(k_3_1, q1_q0); // 1: = 3 * q1 * 1 * q0
const uint32_t q3_q2 = __SXTB16(q3_i3_q2_i2, 8); // 1: (q3_i3_q2_i2 ror 8)[23:16]:(q3_i3_q2_i2 ror 8)[7:0]
const uint32_t d_q0 = __SMLADX(k_3_1, q3_q2, d_q0_partial); // 1: + 3 * q2 + 1 * q3
const uint32_t d_q0_i0 = __PKHBT(d_i0, d_q0, 16); // 1: (Rm<<16)[31:16]:Rn[15:0]
const uint32_t d_i1_partial = __SMUAD(k_3_1, i3_i2); // 1: = 3 * i3 + 1 * i2
const uint32_t i5_i4 = __SXTB16(q5_i5_q4_i4, 0); // 1: (q5_i5_q4_i4 ror 0)[23:16]:(q5_i5_q4_i4 ror 0)[7:0]
const uint32_t d_i1 = __SMLADX(k_3_1, i5_i4, d_i1_partial); // 1: + 1 * i5 + 3 * i4
const uint32_t d_q1_partial = __SMUAD(k_3_1, q3_q2); // 1: = 3 * q3 * 1 * q2
const uint32_t q5_q4 = __SXTB16(q5_i5_q4_i4, 8); // 1: (q5_i5_q4_i4 ror 8)[23:16]:(q5_i5_q4_i4 ror 8)[7:0]
const uint32_t d_q1 = __SMLADX(k_3_1, q5_q4, d_q1_partial); // 1: + 1 * q5 + 3 * q4
const uint32_t d_q1_i1 = __PKHBT(d_i1, d_q1, 16); // 1: (Rm<<16)[31:16]:Rn[15:0]
*(dst_p++) = d_q0_i0; // 3
*(dst_p++) = d_q1_i1;
i1_i0 = i5_i4;
q1_q0 = q5_q4;
}
_i1_i0 = i1_i0;
_q1_q0 = q1_q0;
return { dst.p, src.count / 2, src.sampling_rate / 2 };
}
buffer_s16_t FM::execute(
const buffer_c16_t& src,
const buffer_s16_t& dst
) {
auto z = z_;
const auto src_p = src.p;
const auto src_end = &src.p[src.count];
auto dst_p = dst.p;
while(src_p < src_end) {
const auto s0 = *__SIMD32(src_p)++;
const auto s1 = *__SIMD32(src_p)++;
const auto t0 = multiply_conjugate_s16_s32(s0, z);
const auto t1 = multiply_conjugate_s16_s32(s1, s0);
z = s1;
const int32_t theta0_int = angle_approx_0deg27(t0) * ks16;
const int32_t theta0_sat = __SSAT(theta0_int, 16);
const int32_t theta1_int = angle_approx_0deg27(t1) * ks16;
const int32_t theta1_sat = __SSAT(theta1_int, 16);
*__SIMD32(dst_p)++ = __PKHBT(
theta0_sat,
theta1_sat,
16
);
}
z_ = z;
return { dst.p, src.count, src.sampling_rate };
}
void arm_negate_q15(
q15_t * pSrc,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* loop counter */
q15_t in1, in2; /* Temporary variables */
/*loop Unrolling */
blkCnt = blockSize >> 2u;
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
** a second loop below computes the remaining 1 to 3 samples. */
while(blkCnt > 0u)
{
/* C = ~A */
/* Read two inputs */
in1 = *pSrc++;
in2 = *pSrc++;
/* Negate and then store the results in the destination buffer by packing. */
*__SIMD32(pDst)++ = __PKHBT(__SSAT(-in1, 16), __SSAT(-in2, 16), 16);
in1 = *pSrc++;
in2 = *pSrc++;
*__SIMD32(pDst)++ = __PKHBT(__SSAT(-in1, 16), __SSAT(-in2, 16), 16);
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x4u;
while(blkCnt > 0u)
{
/* C = ~A */
/* Negate and then store the result in the destination buffer. */
*pDst++ = __SSAT(-*pSrc++, 16);
/* Decrement the loop counter */
blkCnt--;
}
}
void arm_fill_q15(
q15_t value,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* loop counter */
#ifndef ARM_MATH_CM0_FAMILY
/* Run the below code for Cortex-M4 and Cortex-M3 */
q31_t packedValue; /* value packed to 32 bits */
/*loop Unrolling */
blkCnt = blockSize >> 2u;
/* Packing two 16 bit values to 32 bit value in order to use SIMD */
packedValue = __PKHBT(value, value, 16u);
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
** a second loop below computes the remaining 1 to 3 samples. */
while(blkCnt > 0u)
{
/* C = value */
/* Fill the value in the destination buffer */
*__SIMD32(pDst)++ = packedValue;
*__SIMD32(pDst)++ = packedValue;
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x4u;
#else
/* Run the below code for Cortex-M0 */
/* Loop over blockSize number of values */
blkCnt = blockSize;
#endif /* #ifndef ARM_MATH_CM0_FAMILY */
while(blkCnt > 0u)
{
/* C = value */
/* Fill the value in the destination buffer */
*pDst++ = value;
/* Decrement the loop counter */
blkCnt--;
}
}
void arm_fill_q15(
q15_t value,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* Loop counter */
#if defined (ARM_MATH_LOOPUNROLL)
q31_t packedValue; /* value packed to 32 bits */
/* Packing two 16 bit values to 32 bit value in order to use SIMD */
packedValue = __PKHBT(value, value, 16U);
/* Loop unrolling: Compute 4 outputs at a time */
blkCnt = blockSize >> 2U;
while (blkCnt > 0U)
{
/* C = value */
/* fill 2 times 2 samples at a time */
write_q15x2_ia (&pDst, packedValue);
write_q15x2_ia (&pDst, packedValue);
/* Decrement loop counter */
blkCnt--;
}
/* Loop unrolling: Compute remaining outputs */
blkCnt = blockSize % 0x4U;
#else
/* Initialize blkCnt with number of samples */
blkCnt = blockSize;
#endif /* #if defined (ARM_MATH_LOOPUNROLL) */
while (blkCnt > 0U)
{
/* C = value */
/* Fill value in destination buffer */
*pDst++ = value;
/* Decrement loop counter */
blkCnt--;
}
}
void arm_fill_q15(
q15_t value,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* loop counter */
q31_t packedValue; /* value packed to 32 bits */
/*loop Unrolling */
blkCnt = blockSize >> 3u;
/* Packing two 16 bit values to 32 bit value in order to use SIMD */
packedValue = __PKHBT(value, value, 16u);
/* First part of the processing with loop unrolling. Compute 8 outputs at a time.
** a second loop below computes the remaining 1 to 7 samples. */
while(blkCnt > 0u)
{
/* C = value */
/* Fill the value in the destination buffer */
_SIMD32_OFFSET(pDst) = packedValue;
_SIMD32_OFFSET(pDst + 2) = packedValue;
_SIMD32_OFFSET(pDst + 4) = packedValue;
_SIMD32_OFFSET(pDst + 6) = packedValue;
pDst += 8u;
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 8, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x8u;
while(blkCnt > 0u)
{
/* C = value */
/* Fill the value in the destination buffer */
*pDst++ = value;
/* Decrement the loop counter */
blkCnt--;
}
}
void arm_biquad_cascade_df1_fast_q15(
const arm_biquad_casd_df1_inst_q15 * S,
q15_t * pSrc,
q15_t * pDst,
uint32_t blockSize)
{
q15_t *pIn = pSrc; /* Source pointer */
q15_t *pOut = pDst; /* Destination pointer */
q31_t in; /* Temporary variable to hold input value */
q31_t out; /* Temporary variable to hold output value */
q31_t b0; /* Temporary variable to hold bo value */
q31_t b1, a1; /* Filter coefficients */
q31_t state_in, state_out; /* Filter state variables */
q31_t acc0; /* Accumulator */
int32_t shift = (int32_t) (15 - S->postShift); /* Post shift */
q15_t *pState = S->pState; /* State pointer */
q15_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */
q31_t *pState_q31; /* 32-bit state pointer for SIMD implementation */
uint32_t sample, stage = S->numStages; /* Stage loop counter */
do {
/* Initialize state pointer of type q31 */
pState_q31 = (q31_t *) (pState);
/* Read the b0 and 0 coefficients using SIMD */
b0 = *__SIMD32(pCoeffs)++;
/* Read the b1 and b2 coefficients using SIMD */
b1 = *__SIMD32(pCoeffs)++;
/* Read the a1 and a2 coefficients using SIMD */
a1 = *__SIMD32(pCoeffs)++;
/* Read the input state values from the state buffer: x[n-1], x[n-2] */
state_in = (q31_t) (*pState_q31++);
/* Read the output state values from the state buffer: y[n-1], y[n-2] */
state_out = (q31_t) (*pState_q31);
/* Apply loop unrolling and compute 2 output values simultaneously. */
/* The variables acc0 ... acc3 hold output values that are being computed:
*
* acc0 = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2]
* acc0 = b0 * x[n] + b1 * x[n-1] + b2 * x[n-2] + a1 * y[n-1] + a2 * y[n-2]
*/
sample = blockSize >> 1u;
/* First part of the processing with loop unrolling. Compute 2 outputs at a time.
** a second loop below computes the remaining 1 sample. */
while(sample > 0u) {
/* Read the input */
in = *__SIMD32(pIn)++;
/* out = b0 * x[n] + 0 * 0 */
out = __SMUAD(b0, in);
/* acc0 = b1 * x[n-1] + acc0 += b2 * x[n-2] + out */
acc0 = __SMLAD(b1, state_in, out);
/* acc0 += a1 * y[n-1] + acc0 += a2 * y[n-2] */
acc0 = __SMLAD(a1, state_out, acc0);
/* The result is converted from 3.29 to 1.31 and then saturation is applied */
out = __SSAT((acc0 >> shift), 16);
/* Every time after the output is computed state should be updated. */
/* The states should be updated as: */
/* Xn2 = Xn1 */
/* Xn1 = Xn */
/* Yn2 = Yn1 */
/* Yn1 = acc0 */
/* x[n-N], x[n-N-1] are packed together to make state_in of type q31 */
/* y[n-N], y[n-N-1] are packed together to make state_out of type q31 */
#ifndef ARM_MATH_BIG_ENDIAN
state_in = __PKHBT(in, state_in, 16);
state_out = __PKHBT(out, state_out, 16);
#else
state_in = __PKHBT(state_in >> 16, (in >> 16), 16);
state_out = __PKHBT(state_out >> 16, (out), 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
/* out = b0 * x[n] + 0 * 0 */
out = __SMUADX(b0, in);
/* acc0 = b1 * x[n-1] + acc0 += b2 * x[n-2] + out */
acc0 = __SMLAD(b1, state_in, out);
/* acc0 += a1 * y[n-1] + acc0 += a2 * y[n-2] */
acc0 = __SMLAD(a1, state_out, acc0);
/* The result is converted from 3.29 to 1.31 and then saturation is applied */
out = __SSAT((acc0 >> shift), 16);
/* Store the output in the destination buffer. */
#ifndef ARM_MATH_BIG_ENDIAN
*__SIMD32(pOut)++ = __PKHBT(state_out, out, 16);
#else
*__SIMD32(pOut)++ = __PKHBT(out, state_out >> 16, 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
/* Every time after the output is computed state should be updated. */
/* The states should be updated as: */
/* Xn2 = Xn1 */
/* Xn1 = Xn */
/* Yn2 = Yn1 */
/* Yn1 = acc0 */
/* x[n-N], x[n-N-1] are packed together to make state_in of type q31 */
/* y[n-N], y[n-N-1] are packed together to make state_out of type q31 */
#ifndef ARM_MATH_BIG_ENDIAN
state_in = __PKHBT(in >> 16, state_in, 16);
state_out = __PKHBT(out, state_out, 16);
#else
state_in = __PKHBT(state_in >> 16, in, 16);
state_out = __PKHBT(state_out >> 16, out, 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
/* Decrement the loop counter */
sample--;
}
/* If the blockSize is not a multiple of 2, compute any remaining output samples here.
** No loop unrolling is used. */
if((blockSize & 0x1u) != 0u) {
/* Read the input */
in = *pIn++;
/* out = b0 * x[n] + 0 * 0 */
#ifndef ARM_MATH_BIG_ENDIAN
out = __SMUAD(b0, in);
#else
out = __SMUADX(b0, in);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
/* acc0 = b1 * x[n-1] + acc0 += b2 * x[n-2] + out */
acc0 = __SMLAD(b1, state_in, out);
/* acc0 += a1 * y[n-1] + acc0 += a2 * y[n-2] */
acc0 = __SMLAD(a1, state_out, acc0);
/* The result is converted from 3.29 to 1.31 and then saturation is applied */
out = __SSAT((acc0 >> shift), 16);
/* Store the output in the destination buffer. */
*pOut++ = (q15_t) out;
/* Every time after the output is computed state should be updated. */
/* The states should be updated as: */
/* Xn2 = Xn1 */
/* Xn1 = Xn */
/* Yn2 = Yn1 */
/* Yn1 = acc0 */
/* x[n-N], x[n-N-1] are packed together to make state_in of type q31 */
/* y[n-N], y[n-N-1] are packed together to make state_out of type q31 */
#ifndef ARM_MATH_BIG_ENDIAN
state_in = __PKHBT(in, state_in, 16);
state_out = __PKHBT(out, state_out, 16);
#else
state_in = __PKHBT(state_in >> 16, in, 16);
state_out = __PKHBT(state_out >> 16, out, 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
}
/* The first stage goes from the input buffer to the output buffer. */
/* Subsequent (numStages - 1) occur in-place in the output buffer */
pIn = pDst;
/* Reset the output pointer */
pOut = pDst;
/* Store the updated state variables back into the state array */
*__SIMD32(pState)++ = state_in;
*__SIMD32(pState)++ = state_out;
/* Decrement the loop counter */
stage--;
} while(stage > 0u);
void arm_abs_q15(
q15_t * pSrc,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* loop counter */
#ifndef ARM_MATH_CM0
/* Run the below code for Cortex-M4 and Cortex-M3 */
q15_t in1; /* Input value1 */
q15_t in2; /* Input value2 */
/*loop Unrolling */
blkCnt = blockSize >> 2u;
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
** a second loop below computes the remaining 1 to 3 samples. */
while(blkCnt > 0u)
{
/* C = |A| */
/* Read two inputs */
in1 = *pSrc++;
in2 = *pSrc++;
/* Store the Absolute result in the destination buffer by packing the two values, in a single cycle */
#ifndef ARM_MATH_BIG_ENDIAN
*__SIMD32(pDst)++ =
__PKHBT(((in1 > 0) ? in1 : __QSUB16(0, in1)),
((in2 > 0) ? in2 : __QSUB16(0, in2)), 16);
#else
*__SIMD32(pDst)++ =
__PKHBT(((in2 > 0) ? in2 : __QSUB16(0, in2)),
((in1 > 0) ? in1 : __QSUB16(0, in1)), 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
in1 = *pSrc++;
in2 = *pSrc++;
#ifndef ARM_MATH_BIG_ENDIAN
*__SIMD32(pDst)++ =
__PKHBT(((in1 > 0) ? in1 : __QSUB16(0, in1)),
((in2 > 0) ? in2 : __QSUB16(0, in2)), 16);
#else
*__SIMD32(pDst)++ =
__PKHBT(((in2 > 0) ? in2 : __QSUB16(0, in2)),
((in1 > 0) ? in1 : __QSUB16(0, in1)), 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x4u;
while(blkCnt > 0u)
{
/* C = |A| */
/* Read the input */
in1 = *pSrc++;
/* Calculate absolute value of input and then store the result in the destination buffer. */
*pDst++ = (in1 > 0) ? in1 : __QSUB16(0, in1);
/* Decrement the loop counter */
blkCnt--;
}
#else
/* Run the below code for Cortex-M0 */
q15_t in; /* Temporary input variable */
/* Initialize blkCnt with number of samples */
blkCnt = blockSize;
while(blkCnt > 0u)
{
/* C = |A| */
/* Read the input */
in = *pSrc++;
/* Calculate absolute value of input and then store the result in the destination buffer. */
*pDst++ = (in > 0) ? in : ((in == (q15_t) 0x8000) ? 0x7fff : -in);
/* Decrement the loop counter */
blkCnt--;
}
#endif /* #ifndef ARM_MATH_CM0 */
}
void arm_q7_to_q15(
q7_t * pSrc,
q15_t * pDst,
uint32_t blockSize)
{
q7_t *pIn = pSrc; /* Src pointer */
uint32_t blkCnt; /* loop counter */
#ifndef ARM_MATH_CM0_FAMILY
q31_t in;
q31_t in1, in2;
q31_t out1, out2;
/* Run the below code for Cortex-M4 and Cortex-M3 */
/*loop Unrolling */
blkCnt = blockSize >> 2u;
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
** a second loop below computes the remaining 1 to 3 samples. */
while(blkCnt > 0u)
{
/* C = (q15_t) A << 8 */
/* convert from q7 to q15 and then store the results in the destination buffer */
in = *__SIMD32(pIn)++;
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in1 = __SXTB16(__ROR(in, 8));
/* extend remainig two q7_t values to q15_t values */
in2 = __SXTB16(in);
in1 = in1 << 8u;
in2 = in2 << 8u;
in1 = in1 & 0xFF00FF00;
in2 = in2 & 0xFF00FF00;
#ifndef ARM_MATH_BIG_ENDIAN
out2 = __PKHTB(in1, in2, 16);
out1 = __PKHBT(in2, in1, 16);
#else
out1 = __PKHTB(in1, in2, 16);
out2 = __PKHBT(in2, in1, 16);
#endif
*__SIMD32(pDst)++ = out1;
*__SIMD32(pDst)++ = out2;
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x4u;
#else
/* Run the below code for Cortex-M0 */
/* Loop over blockSize number of values */
blkCnt = blockSize;
#endif /* #ifndef ARM_MATH_CM0_FAMILY */
while(blkCnt > 0u)
{
/* C = (q15_t) A << 8 */
/* convert from q7 to q15 and then store the results in the destination buffer */
*pDst++ = (q15_t) * pIn++ << 8;
/* Decrement the loop counter */
blkCnt--;
}
}
buffer_c16_t FIRAndDecimateComplex::execute(
const buffer_c16_t& src,
const buffer_c16_t& dst
) {
/* int16_t input (sample count "n" must be multiple of decimation_factor)
* -> int16_t output, decimated by decimation_factor.
* taps are normalized to 1 << 16 == 1.0.
*/
const auto output_sampling_rate = src.sampling_rate / decimation_factor_;
const size_t output_samples = src.count / decimation_factor_;
sample_t* dst_p = dst.p;
const buffer_c16_t result { dst.p, output_samples, output_sampling_rate };
const sample_t* src_p = src.p;
size_t outer_count = output_samples;
while(outer_count > 0) {
/* Put new samples into delay buffer */
auto z_new_p = &samples_[taps_count_ - decimation_factor_];
for(size_t i=0; i<decimation_factor_; i++) {
*__SIMD32(z_new_p)++ = *__SIMD32(src_p)++;
}
size_t loop_count = taps_count_ / 8;
auto t_p = &taps_reversed_[0];
auto z_p = &samples_[0];
int64_t t_real = 0;
int64_t t_imag = 0;
while(loop_count > 0) {
const auto tap0 = *__SIMD32(t_p)++;
const auto sample0 = *__SIMD32(z_p)++;
const auto tap1 = *__SIMD32(t_p)++;
const auto sample1 = *__SIMD32(z_p)++;
t_real = __SMLSLD(sample0, tap0, t_real);
t_imag = __SMLALDX(sample0, tap0, t_imag);
t_real = __SMLSLD(sample1, tap1, t_real);
t_imag = __SMLALDX(sample1, tap1, t_imag);
const auto tap2 = *__SIMD32(t_p)++;
const auto sample2 = *__SIMD32(z_p)++;
const auto tap3 = *__SIMD32(t_p)++;
const auto sample3 = *__SIMD32(z_p)++;
t_real = __SMLSLD(sample2, tap2, t_real);
t_imag = __SMLALDX(sample2, tap2, t_imag);
t_real = __SMLSLD(sample3, tap3, t_real);
t_imag = __SMLALDX(sample3, tap3, t_imag);
const auto tap4 = *__SIMD32(t_p)++;
const auto sample4 = *__SIMD32(z_p)++;
const auto tap5 = *__SIMD32(t_p)++;
const auto sample5 = *__SIMD32(z_p)++;
t_real = __SMLSLD(sample4, tap4, t_real);
t_imag = __SMLALDX(sample4, tap4, t_imag);
t_real = __SMLSLD(sample5, tap5, t_real);
t_imag = __SMLALDX(sample5, tap5, t_imag);
const auto tap6 = *__SIMD32(t_p)++;
const auto sample6 = *__SIMD32(z_p)++;
const auto tap7 = *__SIMD32(t_p)++;
const auto sample7 = *__SIMD32(z_p)++;
t_real = __SMLSLD(sample6, tap6, t_real);
t_imag = __SMLALDX(sample6, tap6, t_imag);
t_real = __SMLSLD(sample7, tap7, t_real);
t_imag = __SMLALDX(sample7, tap7, t_imag);
loop_count--;
}
/* TODO: Re-evaluate whether saturation is performed, normalization,
* all that jazz.
*/
const int32_t r = t_real >> 16;
const int32_t i = t_imag >> 16;
const int32_t r_sat = __SSAT(r, 16);
const int32_t i_sat = __SSAT(i, 16);
*__SIMD32(dst_p)++ = __PKHBT(
r_sat,
i_sat,
16
);
/* Shift sample buffer left/down by decimation factor. */
const size_t unroll_factor = 4;
size_t shift_count = (taps_count_ - decimation_factor_) / unroll_factor;
sample_t* t = &samples_[0];
const sample_t* s = &samples_[decimation_factor_];
while(shift_count > 0) {
*__SIMD32(t)++ = *__SIMD32(s)++;
*__SIMD32(t)++ = *__SIMD32(s)++;
*__SIMD32(t)++ = *__SIMD32(s)++;
*__SIMD32(t)++ = *__SIMD32(s)++;
shift_count--;
}
shift_count = (taps_count_ - decimation_factor_) % unroll_factor;
while(shift_count > 0) {
*(t++) = *(s++);
shift_count--;
}
outer_count--;
}
return result;
}
void arm_mult_q15(
q15_t * pSrcA,
q15_t * pSrcB,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* loop counters */
q31_t inA1, inA2, inB1, inB2; /* temporary input variables */
q15_t out1, out2, out3, out4; /* temporary output variables */
q31_t mul1, mul2, mul3, mul4; /* temporary variables */
/* loop Unrolling */
blkCnt = blockSize >> 2u;
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
** a second loop below computes the remaining 1 to 3 samples. */
while(blkCnt > 0u)
{
/* read two samples at a time from sourceA */
inA1 = *__SIMD32(pSrcA)++;
/* read two samples at a time from sourceB */
inB1 = *__SIMD32(pSrcB)++;
/* read two samples at a time from sourceA */
inA2 = *__SIMD32(pSrcA)++;
/* read two samples at a time from sourceB */
inB2 = *__SIMD32(pSrcB)++;
/* multiply mul = sourceA * sourceB */
mul1 = (q31_t)((q15_t)(inA1 >> 16)* (q15_t)(inB1>> 16));
mul2 = (q31_t)((q15_t)inA1 * (q15_t)inB1);
mul3 = (q31_t)((q15_t)(inA2 >> 16)* (q15_t)(inB2>> 16));
mul4 = (q31_t)((q15_t)inA2 * (q15_t)inB2);
/* shift result by 15 to get 16 bit result */
mul1 = mul1 >> 15;
mul2 = mul2 >> 15;
mul3 = mul3 >> 15;
mul4 = mul4 >> 15;
/* saturate result to 16 bit */
#ifdef CCS
out1 = (q15_t) __SSATA(mul1, 0, 16);
out2 = (q15_t) __SSATA(mul2, 0, 16);
out3 = (q15_t) __SSATA(mul3, 0, 16);
out4 = (q15_t) __SSATA(mul4, 0, 16);
#else
out1 = (q15_t) __SSAT(mul1, 16);
out2 = (q15_t) __SSAT(mul2, 16);
out3 = (q15_t) __SSAT(mul3, 16);
out4 = (q15_t) __SSAT(mul4, 16);
#endif // #ifdef CCS
/* store the result */
#ifndef ARM_MATH_BIG_ENDIAN
*__SIMD32(pDst)++ = __PKHBT(out2, out1, 16);
*__SIMD32(pDst)++ = __PKHBT(out4, out3, 16);
#else
*__SIMD32(pDst)++ = __PKHBT(out2, out1, 16);
*__SIMD32(pDst)++ = __PKHBT(out4, out3, 16);
#endif // #ifndef ARM_MATH_BIG_ENDIAN
/* Decrement the blockSize loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x4u;
while(blkCnt > 0u)
{
/* C = A * B */
/* Multiply the inputs and store the result in the destination buffer */
#ifdef CCS
*pDst++ = (q15_t) __SSATA(((q31_t) ((*pSrcA++) * (*pSrcB++)) >> 15), 0, 16);
#else
*pDst++ = (q15_t) __SSAT((((q31_t) (*pSrcA++) * (*pSrcB++)) >> 15), 16);
#endif // #ifdef CCS
/* Decrement the blockSize loop counter */
blkCnt--;
}
}
void arm_q7_to_q15(
q7_t * pSrc,
q15_t * pDst,
uint32_t blockSize)
{
q7_t *pIn = pSrc; /* Src pointer */
uint32_t blkCnt; /* loop counter */
q31_t in;
q31_t in1, in2;
q31_t out1, out2;
q31_t and = 0xFF00FF00;
/*loop Unrolling */
blkCnt = blockSize >> 3u;
/* First part of the processing with loop unrolling. Compute 8 outputs at a time.
** a second loop below computes the remaining 1 to 7 samples. */
while(blkCnt > 0u)
{
/* C = (q15_t) A << 8 */
/* convert from q7 to q15 and then store the results in the destination buffer */
/* read 4 samples at a time */
in = *__SIMD32(pIn)++;
#ifdef CCS
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in1 = __SXTB16(in, 8);
/* extend remainig two q7_t values to q15_t values */
in2 = __SXTB16(in, 0);
#else
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in1 = __SXTB16(__ROR(in, 8));
/* extend remainig two q7_t values to q15_t values */
in2 = __SXTB16(in);
#endif /* shift in1 by 8 to convert q7_t value to q15_t value (ex: 0x00ff00ff ==> 0xff00ff00*/
in1 = in1 << 8u;
in2 = in2 << 8u;
/* read next 4 sampels */
in = *__SIMD32(pIn)++;
/* anding with 0xff00ff00 */
in1 = in1 & and;
out2 = in2 & and;
/* pack two 16 bit values */
out1 = __PKHTB(in1, out2, 16);
out2 = __PKHBT(out2, in1, 16);
#ifndef ARM_MATH_BIG_ENDIAN
/* store two q15_t samples at a time to destination */
_SIMD32_OFFSET(pDst + 2) = out1;
#ifdef CCS
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in1 = __SXTB16(in, 8);
#else
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in1 = __SXTB16(__ROR(in, 8));
#endif
/* store two q15_t samples at a time to destination */
_SIMD32_OFFSET(pDst) = out2;
#else
/* store two q15_t samples at a time to destination */
_SIMD32_OFFSET(pDst) = out1;
#ifdef CCS
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in1 = __SXTB16(in, 8);
#else
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in1 = __SXTB16(__ROR(in, 8));
#endif
/* store two q15_t samples at a time to destination */
_SIMD32_OFFSET(pDst + 2) = out2;
#endif // #ifndef ARM_MATH_BIG_ENDIAN
#ifdef CCS
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in2 = __SXTB16(in, 0);
#else
/* rotatate in by 8 and extend two q7_t values to q15_t values */
in2 = __SXTB16(in);
#endif
/* shift in1 by 8 to convert q7_t value to q15_t value (ex: 0x00ff00ff ==> 0xff00ff00*/
in1 = in1 << 8u;
in2 = in2 << 8u;
/* anding with 0xff00ff00 */
out1 = in1 & and;
out2 = in2 & and;
/* pack two 16 bit values */
out1 = __PKHTB(in1, out2, 16);
out2 = __PKHBT(out2, in1, 16);
/* store two q15_t samples at a time to destination */
#ifndef ARM_MATH_BIG_ENDIAN
_SIMD32_OFFSET(pDst + 6) = out1;
_SIMD32_OFFSET(pDst + 4) = out2;
#else
_SIMD32_OFFSET(pDst + 4) = out1;
_SIMD32_OFFSET(pDst + 6) = out2;
#endif // #ifndef ARM_MATH_BIG_ENDIAN
/* incremnet destination pointer */
pDst += 8u;
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 8, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x8u;
while(blkCnt > 0u)
{
/* C = (q15_t) A << 8 */
/* convert from q7 to q15 and then store the results in the destination buffer */
*pDst++ = (q15_t) * pIn++ << 8;
/* Decrement the loop counter */
blkCnt--;
}
}
void arm_offset_q15(
q15_t * pSrc,
q15_t offset,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* loop counter */
#ifndef ARM_MATH_CM0_FAMILY
/* Run the below code for Cortex-M4 and Cortex-M3 */
q31_t offset_packed; /* Offset packed to 32 bit */
/*loop Unrolling */
blkCnt = blockSize >> 2u;
/* Offset is packed to 32 bit in order to use SIMD32 for addition */
offset_packed = __PKHBT(offset, offset, 16);
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
** a second loop below computes the remaining 1 to 3 samples. */
while(blkCnt > 0u)
{
/* C = A + offset */
/* Add offset and then store the results in the destination buffer, 2 samples at a time. */
*__SIMD32(pDst)++ = __QADD16(*__SIMD32(pSrc)++, offset_packed);
*__SIMD32(pDst)++ = __QADD16(*__SIMD32(pSrc)++, offset_packed);
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x4u;
while(blkCnt > 0u)
{
/* C = A + offset */
/* Add offset and then store the results in the destination buffer. */
*pDst++ = (q15_t) __QADD16(*pSrc++, offset);
/* Decrement the loop counter */
blkCnt--;
}
#else
/* Run the below code for Cortex-M0 */
/* Initialize blkCnt with number of samples */
blkCnt = blockSize;
while(blkCnt > 0u)
{
/* C = A + offset */
/* Add offset and then store the results in the destination buffer. */
*pDst++ = (q15_t) __SSAT(((q31_t) * pSrc++ + offset), 16);
/* Decrement the loop counter */
blkCnt--;
}
#endif /* #ifndef ARM_MATH_CM0_FAMILY */
}
void arm_negate_q15(
q15_t * pSrc,
q15_t * pDst,
uint32_t blockSize)
{
uint32_t blkCnt; /* loop counter */
#ifndef ARM_MATH_CM0
/* Run the below code for Cortex-M4 and Cortex-M3 */
q15_t in1, in2; /* Temporary variables */
/*loop Unrolling */
blkCnt = blockSize >> 2u;
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
** a second loop below computes the remaining 1 to 3 samples. */
while(blkCnt > 0u)
{
/* C = -A */
/* Read two inputs */
in1 = *pSrc++;
in2 = *pSrc++;
/* Negate and then store the results in the destination buffer by packing. */
#ifndef ARM_MATH_BIG_ENDIAN
*__SIMD32(pDst)++ = __PKHBT(__SSAT(-in1, 16), __SSAT(-in2, 16), 16);
#else
*__SIMD32(pDst)++ = __PKHBT(__SSAT(-in2, 16), __SSAT(-in1, 16), 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
in1 = *pSrc++;
in2 = *pSrc++;
#ifndef ARM_MATH_BIG_ENDIAN
*__SIMD32(pDst)++ = __PKHBT(__SSAT(-in1, 16), __SSAT(-in2, 16), 16);
#else
*__SIMD32(pDst)++ = __PKHBT(__SSAT(-in2, 16), __SSAT(-in1, 16), 16);
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
/* Decrement the loop counter */
blkCnt--;
}
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
** No loop unrolling is used. */
blkCnt = blockSize % 0x4u;
#else
/* Run the below code for Cortex-M0 */
/* Initialize blkCnt with number of samples */
blkCnt = blockSize;
#endif /* #ifndef ARM_MATH_CM0 */
while(blkCnt > 0u)
{
/* C = -A */
/* Negate and then store the result in the destination buffer. */
*pDst++ = __SSAT(-*pSrc++, 16);
/* Decrement the loop counter */
blkCnt--;
}
}
buffer_c16_t TranslateByFSOver4AndDecimateBy2CIC3::execute(const buffer_c8_t& src, const buffer_c16_t& dst) {
/* Translates incoming complex<int8_t> samples by -fs/4,
* decimates by two using a non-recursive third-order CIC filter.
*/
/* Derivation of algorithm:
* Original CIC filter (decimating by two):
* D_I0 = i3 * 1 + i2 * 3 + i1 * 3 + i0 * 1
* D_Q0 = q3 * 1 + q2 * 3 + q1 * 3 + q0 * 1
*
* D_I1 = i5 * 1 + i4 * 3 + i3 * 3 + i2 * 1
* D_Q1 = q5 * 1 + q4 * 3 + q3 * 3 + q2 * 1
*
* Translate -fs/4, phased 180 degrees, accomplished by complex multiplication
* of complex length-4 sequence:
*
* Substitute:
* i0 = -i0, q0 = -q0
* i1 = -q1, q1 = i1
* i2 = i2, q2 = q2
* i3 = q3, q3 = -i3
* i4 = -i4, q4 = -q4
* i5 = -q5, q5 = i5
*
* Resulting taps (with decimation by 2, four samples in, two samples out):
* D_I0 = q3 * 1 + i2 * 3 + -q1 * 3 + -i0 * 1
* D_Q0 = -i3 * 1 + q2 * 3 + i1 * 3 + -q0 * 1
*
* D_I1 = -q5 * 1 + -i4 * 3 + q3 * 3 + i2 * 1
* D_Q1 = i5 * 1 + -q4 * 3 + -i3 * 3 + q2 * 1
*/
// 6 cycles per complex input sample, not including loop overhead.
uint32_t q1_i0 = _q1_i0;
uint32_t q0_i1 = _q0_i1;
/* 3:1 Scaled by 32 to normalize output to +/-32768-ish. */
constexpr uint32_t scale_factor = 32;
const uint32_t k_3_1 = 0x00030001 * scale_factor;
uint32_t* src_p = reinterpret_cast<uint32_t*>(&src.p[0]);
uint32_t* const src_end = reinterpret_cast<uint32_t*>(&src.p[src.count]);
uint32_t* dst_p = reinterpret_cast<uint32_t*>(&dst.p[0]);
while(src_p < src_end) {
const uint32_t q3_i3_q2_i2 = *(src_p++); // 3
const uint32_t q5_i5_q4_i4 = *(src_p++);
const uint32_t i2_i3 = __SXTB16(q3_i3_q2_i2, 16); // 1: (q3_i3_q2_i2 ror 16)[23:16]:(q3_i3_q2_i2 ror 16)[7:0]
const uint32_t q3_q2 = __SXTB16(q3_i3_q2_i2, 8); // 1: (q3_i3_q2_i2 ror 8)[23:16]:(q3_i3_q2_i2 ror 8)[7:0]
const uint32_t i2_q3 = __PKHTB(i2_i3, q3_q2, 16); // 1: Rn[31:16]:(Rm>>16)[15:0]
const uint32_t i3_q2 = __PKHBT(q3_q2, i2_i3, 16); // 1:(Rm<<16)[31:16]:Rn[15:0]
// D_I0 = 3 * (i2 - q1) + (q3 - i0)
const uint32_t i2_m_q1_q3_m_i0 = __QSUB16(i2_q3, q1_i0); // 1: Rn[31:16]-Rm[31:16]:Rn[15:0]-Rm[15:0]
const uint32_t d_i0 = __SMUAD(k_3_1, i2_m_q1_q3_m_i0); // 1: Rm[15:0]*Rs[15:0]+Rm[31:16]*Rs[31:16]
// D_Q0 = 3 * (q2 + i1) - (i3 + q0)
const uint32_t i3_p_q0_q2_p_i1 = __QADD16(i3_q2, q0_i1); // 1: Rn[31:16]+Rm[31:16]:Rn[15:0]+Rm[15:0]
const uint32_t d_q0 = __SMUSDX(i3_p_q0_q2_p_i1, k_3_1); // 1: Rm[15:0]*Rs[31:16]–Rm[31:16]*RsX[15:0]
const uint32_t d_q0_i0 = __PKHBT(d_i0, d_q0, 16); // 1: (Rm<<16)[31:16]:Rn[15:0]
const uint32_t i5_i4 = __SXTB16(q5_i5_q4_i4, 0); // 1: (q5_i5_q4_i4 ror 0)[23:16]:(q5_i5_q4_i4 ror 0)[7:0]
const uint32_t q4_q5 = __SXTB16(q5_i5_q4_i4, 24); // 1: (q5_i5_q4_i4 ror 24)[23:16]:(q5_i5_q4_i4 ror 24)[7:0]
const uint32_t q4_i5 = __PKHTB(q4_q5, i5_i4, 16); // 1: Rn[31:16]:(Rm>>16)[15:0]
const uint32_t q5_i4 = __PKHBT(i5_i4, q4_q5, 16); // 1: (Rm<<16)[31:16]:Rn[15:0]
// D_I1 = (i2 - q5) + 3 * (q3 - i4)
const uint32_t i2_m_q5_q3_m_i4 = __QSUB16(i2_q3, q5_i4); // 1: Rn[31:16]-Rm[31:16]:Rn[15:0]-Rm[15:0]
const uint32_t d_i1 = __SMUADX(i2_m_q5_q3_m_i4, k_3_1); // 1: Rm[15:0]*Rs[31:16]+Rm[31:16]*Rs[15:0]
// D_Q1 = (i5 + q2) - 3 * (q4 + i3)
const uint32_t q4_p_i3_i5_p_q2 = __QADD16(q4_i5, i3_q2); // 1: Rn[31:16]+Rm[31:16]:Rn[15:0]+Rm[15:0]
const uint32_t d_q1 = __SMUSD(k_3_1, q4_p_i3_i5_p_q2); // 1: Rm[15:0]*Rs[15:0]–Rm[31:16]*Rs[31:16]
const uint32_t d_q1_i1 = __PKHBT(d_i1, d_q1, 16); // 1: (Rm<<16)[31:16]:Rn[15:0]
*(dst_p++) = d_q0_i0; // 3
*(dst_p++) = d_q1_i1;
q1_i0 = q5_i4;
q0_i1 = q4_i5;
}
_q1_i0 = q1_i0;
_q0_i1 = q0_i1;
return { dst.p, src.count / 2, src.sampling_rate / 2 };
}