/**
\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_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 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 };
}
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--;
}
}
