diff --git a/Include/dsp/basic_math_functions.h b/Include/dsp/basic_math_functions.h index dcc1f2a5..49aa1e00 100755 --- a/Include/dsp/basic_math_functions.h +++ b/Include/dsp/basic_math_functions.h @@ -1,8 +1,8 @@ /****************************************************************************** * @file basic_math_functions.h * @brief Public header file for CMSIS DSP Library - * @version V1.10.0 - * @date 08 July 2021 + * @version V1.11.1 + * @date 14 July 2022 * Target Processor: Cortex-M and Cortex-A cores ******************************************************************************/ /* diff --git a/Include/dsp/basic_math_functions_f16.h b/Include/dsp/basic_math_functions_f16.h index de9f58ba..12e5a328 100755 --- a/Include/dsp/basic_math_functions_f16.h +++ b/Include/dsp/basic_math_functions_f16.h @@ -1,8 +1,8 @@ /****************************************************************************** * @file basic_math_functions_f16.h * @brief Public header file for CMSIS DSP Library - * @version V1.10.0 - * @date 08 July 2021 + * @version V1.10.1 + * @date 14 July 2022 * Target Processor: Cortex-M and Cortex-A cores ******************************************************************************/ /* @@ -159,6 +159,8 @@ void arm_clip_f16(const float16_t * pSrc, float16_t high, uint32_t numSamples); + + #endif /* defined(ARM_FLOAT16_SUPPORTED)*/ #ifdef __cplusplus diff --git a/Include/dsp/statistics_functions.h b/Include/dsp/statistics_functions.h index c5224c6d..4d308b31 100755 --- a/Include/dsp/statistics_functions.h +++ b/Include/dsp/statistics_functions.h @@ -1,8 +1,8 @@ /****************************************************************************** * @file statistics_functions.h * @brief Public header file for CMSIS DSP Library - * @version V1.10.0 - * @date 08 July 2021 + * @version V1.10.1 + * @date 14 July 2022 * Target Processor: Cortex-M and Cortex-A cores ******************************************************************************/ /* @@ -970,6 +970,67 @@ void arm_mse_f64( uint32_t blockSize, float64_t * pResult); + +/** + * @brief Accumulation value of a floating-point vector. + * @param[in] pSrc is input pointer + * @param[in] blockSize is the number of samples to process + * @param[out] pResult is output value. + */ + +void arm_accumulate_f32( +const float32_t * pSrc, + uint32_t blockSize, + float32_t * pResult); + +/** + * @brief Accumulation value of a floating-point vector. + * @param[in] pSrc is input pointer + * @param[in] blockSize is the number of samples to process + * @param[out] pResult is output value. + */ + +void arm_accumulate_f64( +const float64_t * pSrc, + uint32_t blockSize, + float64_t * pResult); + +/** + * @brief Accumulation value of a Q7 vector. + * @param[in] pSrc is input pointer + * @param[in] blockSize is the number of samples to process + * @param[out] pResult is output value. + */ + +void arm_accumulate_q7( +const q7_t * pSrc, + uint32_t blockSize, + q7_t * pResult); + +/** + * @brief Accumulation value of a Q15 vector. + * @param[in] pSrc is input pointer + * @param[in] blockSize is the number of samples to process + * @param[out] pResult is output value. + */ + +void arm_accumulate_q15( +const q15_t * pSrc, + uint32_t blockSize, + q15_t * pResult); + +/** + * @brief Accumulation value of a Q31 vector. + * @param[in] pSrc is input pointer + * @param[in] blockSize is the number of samples to process + * @param[out] pResult is output value. + */ + +void arm_accumulate_q31( +const q31_t * pSrc, + uint32_t blockSize, + q31_t * pResult); + #ifdef __cplusplus } #endif diff --git a/Include/dsp/statistics_functions_f16.h b/Include/dsp/statistics_functions_f16.h index 124c0a09..33b12c4b 100755 --- a/Include/dsp/statistics_functions_f16.h +++ b/Include/dsp/statistics_functions_f16.h @@ -1,8 +1,8 @@ /****************************************************************************** * @file statistics_functions_f16.h * @brief Public header file for CMSIS DSP Library - * @version V1.10.0 - * @date 08 July 2021 + * @version V1.10.1 + * @date 14 July 2022 * Target Processor: Cortex-M and Cortex-A cores ******************************************************************************/ /* @@ -258,6 +258,19 @@ void arm_mse_f16( uint32_t blockSize, float16_t * pResult); + +/** + * @brief Sum value of a floating-point vector. + * @param[in] pSrc is input pointer + * @param[in] blockSize is the number of samples to process + * @param[out] pResult is output value. + */ + void arm_accumulate_f16( + const float16_t * pSrc, + uint32_t blockSize, + float16_t * pResult); + + #endif /*defined(ARM_FLOAT16_SUPPORTED)*/ #ifdef __cplusplus } diff --git a/Source/BasicMathFunctions/BasicMathFunctionsF16.c b/Source/BasicMathFunctions/BasicMathFunctionsF16.c index f185db48..c175a499 100755 --- a/Source/BasicMathFunctions/BasicMathFunctionsF16.c +++ b/Source/BasicMathFunctions/BasicMathFunctionsF16.c @@ -35,3 +35,4 @@ #include "arm_scale_f16.c" #include "arm_sub_f16.c" #include "arm_clip_f16.c" + diff --git a/Source/StatisticsFunctions/StatisticsFunctions.c b/Source/StatisticsFunctions/StatisticsFunctions.c index 54469928..00f3c90a 100644 --- a/Source/StatisticsFunctions/StatisticsFunctions.c +++ b/Source/StatisticsFunctions/StatisticsFunctions.c @@ -3,8 +3,8 @@ * Title: StatisticsFunctions.c * Description: Combination of all statistics function source files. * - * $Date: 16. March 2020 - * $Revision: V1.1.0 + * $Date: 14 July 2022 + * $Revision: V1.1.1 * * Target Processor: Cortex-M cores * -------------------------------------------------------------------- */ @@ -98,3 +98,9 @@ #include "arm_mse_q31.c" #include "arm_mse_f32.c" #include "arm_mse_f64.c" +#include "arm_accumulate_f32.c" +#include "arm_accumulate_f64.c" +#include "arm_accumulate_q7.c" +#include "arm_accumulate_q15.c" +#include "arm_accumulate_q31.c" + diff --git a/Source/StatisticsFunctions/StatisticsFunctionsF16.c b/Source/StatisticsFunctions/StatisticsFunctionsF16.c index 510746e6..734c83ff 100755 --- a/Source/StatisticsFunctions/StatisticsFunctionsF16.c +++ b/Source/StatisticsFunctions/StatisticsFunctionsF16.c @@ -3,8 +3,8 @@ * Title: StatisticsFunctions.c * Description: Combination of all statistics function source files. * - * $Date: 16. March 2020 - * $Revision: V1.1.0 + * $Date: 14 July 2022 + * $Revision: V1.1.1 * * Target Processor: Cortex-M cores * -------------------------------------------------------------------- */ @@ -44,3 +44,4 @@ #include "arm_absmax_no_idx_f16.c" #include "arm_absmin_no_idx_f16.c" #include "arm_mse_f16.c" +#include "arm_accumulate_f16.c" diff --git a/Source/StatisticsFunctions/arm_accumulate_f16.c b/Source/StatisticsFunctions/arm_accumulate_f16.c new file mode 100755 index 00000000..9ed7c66c --- /dev/null +++ b/Source/StatisticsFunctions/arm_accumulate_f16.c @@ -0,0 +1,152 @@ +/* ---------------------------------------------------------------------- + * Project: CMSIS DSP Library + * Title: arm_accumulate_f16.c + * Description: accumulation value of a floating-point vector + * + * $Date: 14 July 2022 + * $Revision: V1.0.0 + * + * Target Processor: Cortex-M and Cortex-A cores + * -------------------------------------------------------------------- */ +/* + * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved. + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "dsp/statistics_functions_f16.h" + +#if defined(ARM_FLOAT16_SUPPORTED) + + +/** + @ingroup groupStats + */ + +/** + @defgroup Accumulation + + Calculates the accumulation of the input vector. Sum is defined as the addition of the elements in the vector. + The underlying algorithm is used: + + 
+ Result = (pSrc[0] + pSrc[1] + pSrc[2] + ... + pSrc[blockSize-1]);
+
+ + There are separate functions for floating-point, Q31, Q15, and Q7 data types. + */ + +/** + @addtogroup Accumulation + @{ + */ + +/** + @brief accumulate value of a floating-point vector. + @param[in] pSrc points to the input vector. + @param[in] blockSize number of samples in input vector. + @param[out] pResult sum value returned here. + @return none + */ +#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) + +#include "arm_helium_utils.h" + +void arm_accumulate_f16( + const float16_t * pSrc, + uint32_t blockSize, + float16_t * pResult) +{ + int32_t blkCnt; /* loop counters */ + f16x8_t vecSrc; + f16x8_t sumVec = vdupq_n_f16(0.0f16); + + blkCnt = blockSize; + do { + mve_pred16_t p = vctp16q(blkCnt); + + vecSrc = vldrhq_z_f16((float16_t const *) pSrc, p); + sumVec = vaddq_m_f16(sumVec, sumVec, vecSrc, p); + + blkCnt -= 8; + pSrc += 8; + } + while (blkCnt > 0); + + *pResult = vecAddAcrossF16Mve(sumVec); +} + + +#else + +void arm_accumulate_f16( + const float16_t * pSrc, + uint32_t blockSize, + float16_t * pResult) +{ + uint32_t blkCnt; /* Loop counter */ + float16_t sum = 0.0f; /* Temporary result storage */ + +#if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE) + + /* Loop unrolling: Compute 4 outputs at a time */ + blkCnt = blockSize >> 2U; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += (_Float16)*pSrc++; + + sum += (_Float16)*pSrc++; + + sum += (_Float16)*pSrc++; + + sum += (_Float16)*pSrc++; + + /* Decrement the 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 = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += (_Float16)*pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store result to destination */ + *pResult = sum ; +} +#endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */ + +/** + @} end of Accumulation group + */ + +#endif /* #if defined(ARM_FLOAT16_SUPPORTED) */ + diff --git a/Source/StatisticsFunctions/arm_accumulate_f32.c b/Source/StatisticsFunctions/arm_accumulate_f32.c new file mode 100644 index 00000000..f967aac5 --- /dev/null +++ b/Source/StatisticsFunctions/arm_accumulate_f32.c @@ -0,0 +1,198 @@ +/* ---------------------------------------------------------------------- + * Project: CMSIS DSP Library + * Title: arm_accumulate_f32.c + * Description: Sum value of a floating-point vector + * + * $Date: 14 July 2022 + * $Revision: V1.0.0 + * + * Target Processor: Cortex-M and Cortex-A cores + * -------------------------------------------------------------------- */ +/* + * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved. + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "dsp/statistics_functions.h" + +/** + @ingroup groupStats + */ + + +/** + @addtogroup Accumulation + @{ + */ + +/** + @brief Accumulation value of a floating-point vector. + @param[in] pSrc points to the input vector. + @param[in] blockSize number of samples in input vector. + @param[out] pResult sum value returned here. + @return none + */ +#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) + +#include "arm_helium_utils.h" + +void arm_accumulate_f32( + const float32_t * pSrc, + uint32_t blockSize, + float32_t * pResult) +{ + uint32_t blkCnt; /* loop counters */ + f32x4_t vecSrc; + f32x4_t sumVec = vdupq_n_f32(0.0f); + float32_t sum = 0.0f; + + /* Compute 4 outputs at a time */ + blkCnt = blockSize >> 2U; + while (blkCnt > 0U) + { + vecSrc = vldrwq_f32(pSrc); + sumVec = vaddq_f32(sumVec, vecSrc); + + blkCnt --; + pSrc += 4; + } + + sum = vecAddAcrossF32Mve(sumVec); + + /* Tail */ + blkCnt = blockSize & 0x3; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + *pResult = sum; +} + + +#else +#if defined(ARM_MATH_NEON_EXPERIMENTAL) && !defined(ARM_MATH_AUTOVECTORIZE) +void arm_accumulate_f32( + const float32_t * pSrc, + uint32_t blockSize, + float32_t * pResult) +{ + float32_t sum = 0.0f; /* Temporary result storage */ + float32x4_t sumV = vdupq_n_f32(0.0f); /* Temporary result storage */ + float32x2_t sumV2; + + uint32_t blkCnt; /* Loop counter */ + + float32x4_t inV; + + blkCnt = blockSize >> 2U; + + /* Compute 4 outputs at a time. + ** a second loop below computes the remaining 1 to 3 samples. */ + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + inV = vld1q_f32(pSrc); + sumV = vaddq_f32(sumV, inV); + + pSrc += 4; + /* Decrement the loop counter */ + blkCnt--; + } + + sumV2 = vpadd_f32(vget_low_f32(sumV),vget_high_f32(sumV)); + sum = vget_lane_f32(sumV2, 0) + vget_lane_f32(sumV2, 1); + + /* If the blockSize is not a multiple of 4, compute any remaining output samples here. + ** No loop unrolling is used. */ + blkCnt = blockSize & 3; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement the loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store the result to the destination */ + *pResult = sum; +} +#else +void arm_accumulate_f32( + const float32_t * pSrc, + uint32_t blockSize, + float32_t * pResult) +{ + uint32_t blkCnt; /* Loop counter */ + float32_t sum = 0.0f; /* Temporary result storage */ + +#if defined (ARM_MATH_LOOPUNROLL) && !defined(ARM_MATH_AUTOVECTORIZE) + + /* Loop unrolling: Compute 4 outputs at a time */ + blkCnt = blockSize >> 2U; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + sum += *pSrc++; + + sum += *pSrc++; + + sum += *pSrc++; + + /* Decrement the 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 = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store result to destination */ + *pResult = sum ; +} +#endif /* #if defined(ARM_MATH_NEON) */ +#endif /* defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE) */ + +/** + @} end of Accumulation group + */ diff --git a/Source/StatisticsFunctions/arm_accumulate_f64.c b/Source/StatisticsFunctions/arm_accumulate_f64.c new file mode 100644 index 00000000..4cda86b3 --- /dev/null +++ b/Source/StatisticsFunctions/arm_accumulate_f64.c @@ -0,0 +1,127 @@ +/* ---------------------------------------------------------------------- + * Project: CMSIS DSP Library + * Title: arm_accumulate_f64.c + * Description: Accumulation value of a floating-point vector + * + * $Date: 14 July 2022 + * $Revision: V1.0.0 + * + * Target Processor: Cortex-M and Cortex-A cores + * -------------------------------------------------------------------- */ +/* + * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved. + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "dsp/basic_math_functions.h" + +/** + @ingroup groupStats + */ + + +/** + @addtogroup Accumulation + @{ + */ + +/** + @brief Accumulation value of a floating-point vector. + @param[in] pSrc points to the input vector. + @param[in] blockSize number of samples in input vector. + @param[out] pResult sum value returned here. + @return none + */ +#if defined(ARM_MATH_NEON) +void arm_accumulate_f64( + const float64_t * pSrc, + uint32_t blockSize, + float64_t * pResult) +{ + uint32_t blkCnt; /* Loop counter */ + + /*Neon buffers*/ + float64x2_t vSum = vdupq_n_f64(0.0f); + float64x2_t afterLoad ; + + float64_t sum = 0.; /* Temporary result storage */ + + /* Initialize blkCnt with number of samples */ + blkCnt = blockSize >> 1U; + + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + + afterLoad = vld1q_f64(pSrc); + vSum = vaddq_f64(vSum, afterLoad); + + /* Decrement loop counter */ + blkCnt--; + + pSrc += 2; + } + sum = vaddvq_f64(vSum); + + /* Tail */ + blkCnt = blockSize & 1 ; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store result to destination */ + *pResult = sum; +} +#else +void arm_accumulate_f64( + const float64_t * pSrc, + uint32_t blockSize, + float64_t * pResult) +{ + uint32_t blkCnt; /* Loop counter */ + float64_t sum = 0.; /* Temporary result storage */ + + /* Initialize blkCnt with number of samples */ + blkCnt = blockSize; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store result to destination */ + *pResult = sum; +} + +#endif + + +/** + @} end of Accumulation group + */ diff --git a/Source/StatisticsFunctions/arm_accumulate_q15.c b/Source/StatisticsFunctions/arm_accumulate_q15.c new file mode 100644 index 00000000..7396c673 --- /dev/null +++ b/Source/StatisticsFunctions/arm_accumulate_q15.c @@ -0,0 +1,156 @@ +/* ---------------------------------------------------------------------- + * Project: CMSIS DSP Library + * Title: arm_accumulate_q15.c + * Description: Accumulation value of a Q15 vector + * + * $Date: 14 July 2022 + * $Revision: V1.0.0 + * + * Target Processor: Cortex-M and Cortex-A cores + * -------------------------------------------------------------------- */ +/* + * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved. + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "dsp/statistics_functions.h" + +/** + @ingroup groupStats + */ + +/** + @addtogroup Accumulation + @{ + */ + +/** + @brief Accumulation value of a Q15 vector. + @param[in] pSrc points to the input vector + @param[in] blockSize number of samples in input vector + @param[out] pResult sum value returned here + @return none + + @par Scaling and Overflow Behavior + The function is implemented using a 32-bit internal accumulator. + The input is represented in 1.15 format and is accumulated in a 32-bit + accumulator in 17.15 format. + There is no risk of internal overflow with this approach, and the + full precision of intermediate result is preserved. + Finally, the accumulator is truncated to yield a result of 1.15 format. + */ + +#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) +void arm_accumulate_q15( + const q15_t * pSrc, + uint32_t blockSize, + q15_t * pResult) +{ + uint32_t blkCnt; /* loop counters */ + q15x8_t vecSrc; + q31_t sum = 0L; + + /* Compute 8 outputs at a time */ + blkCnt = blockSize >> 3U; + while (blkCnt > 0U) + { + vecSrc = vldrhq_s16(pSrc); + /* + * sum lanes + */ + sum = vaddvaq(sum, vecSrc); + + blkCnt--; + pSrc += 8; + } + + /* Tail */ + blkCnt = blockSize & 0x7; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store the result to the destination */ + *pResult = sum; +} +#else +void arm_accumulate_q15( + const q15_t * pSrc, + uint32_t blockSize, + q15_t * pResult) +{ + uint32_t blkCnt; /* Loop counter */ + q31_t sum = 0; /* Temporary result storage */ + +#if defined (ARM_MATH_LOOPUNROLL) + q31_t in; +#endif + +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 outputs at a time */ + blkCnt = blockSize >> 2U; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + in = read_q15x2_ia (&pSrc); + sum += ((in << 16U) >> 16U); + sum += (in >> 16U); + + in = read_q15x2_ia (&pSrc); + sum += ((in << 16U) >> 16U); + sum += (in >> 16U); + + /* Decrement the 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 = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store result to destination */ + *pResult = sum; +} +#endif /* defined(ARM_MATH_MVEI) */ + +/** + @} end of Accumulation group + */ diff --git a/Source/StatisticsFunctions/arm_accumulate_q31.c b/Source/StatisticsFunctions/arm_accumulate_q31.c new file mode 100644 index 00000000..40abe5e6 --- /dev/null +++ b/Source/StatisticsFunctions/arm_accumulate_q31.c @@ -0,0 +1,149 @@ +/* ---------------------------------------------------------------------- + * Project: CMSIS DSP Library + * Title: arm_accumulate_q31.c + * Description: Accumulation value of a Q31 vector + * + * $Date: 14 July 2022 + * $Revision: V1.0.0 + * + * Target Processor: Cortex-M and Cortex-A cores + * -------------------------------------------------------------------- */ +/* + * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved. + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "dsp/statistics_functions.h" + +/** + @ingroup groupStats + */ + +/** + @addtogroup Accumulation + @{ + */ + +/** + @brief Accumulation value of a Q31 vector. + @param[in] pSrc points to the input vector + @param[in] blockSize number of samples in input vector + @param[out] pResult sum value returned here + @return none + + @par Scaling and Overflow Behavior + The function is implemented using a 64-bit internal accumulator. + The input is represented in 1.31 format and is accumulated in a 64-bit + accumulator in 33.31 format. + There is no risk of internal overflow with this approach, and the + full precision of intermediate result is preserved. + Finally, the accumulator is truncated to yield a result of 1.31 format. + */ +#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) +void arm_accumulate_q31( + const q31_t * pSrc, + uint32_t blockSize, + q31_t * pResult) +{ + uint32_t blkCnt; /* loop counters */ + q31x4_t vecSrc; + q63_t sum = 0LL; + + + /* Compute 4 outputs at a time */ + blkCnt = blockSize >> 2U; + while (blkCnt > 0U) + { + + vecSrc = vldrwq_s32(pSrc); + /* + * sum lanes + */ + sum = vaddlvaq(sum, vecSrc); + + blkCnt --; + pSrc += 4; + } + + /* Tail */ + blkCnt = blockSize & 0x3; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + blkCnt --; + } + + *pResult = sum; +} +#else +void arm_accumulate_q31( + const q31_t * pSrc, + uint32_t blockSize, + q31_t * pResult) +{ + uint32_t blkCnt; /* Loop counter */ + q63_t sum = 0; /* Temporary result storage */ + +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 outputs at a time */ + blkCnt = blockSize >> 2U; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + sum += *pSrc++; + + sum += *pSrc++; + + sum += *pSrc++; + + /* Decrement the 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 = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store result to destination */ + *pResult = sum; +} +#endif /* defined(ARM_MATH_MVEI) */ + +/** + @} end of Accumulation group + */ diff --git a/Source/StatisticsFunctions/arm_accumulate_q7.c b/Source/StatisticsFunctions/arm_accumulate_q7.c new file mode 100644 index 00000000..54e81d04 --- /dev/null +++ b/Source/StatisticsFunctions/arm_accumulate_q7.c @@ -0,0 +1,153 @@ +/* ---------------------------------------------------------------------- + * Project: CMSIS DSP Library + * Title: arm_accumulate_q7.c + * Description: Accumulation value of a Q7 vector + * + * $Date: 14 July 2022 + * $Revision: V1.0.0 + * + * Target Processor: Cortex-M and Cortex-A cores + * -------------------------------------------------------------------- */ +/* + * Copyright (C) 2010-2021 ARM Limited or its affiliates. All rights reserved. + * + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the License); you may + * not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an AS IS BASIS, WITHOUT + * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "dsp/statistics_functions.h" + +/** + @ingroup groupStats + */ + +/** + @addtogroup Accumulation + @{ + */ + +/** + @brief Accumulation value of a Q7 vector. + @param[in] pSrc points to the input vector + @param[in] blockSize number of samples in input vector + @param[out] pResult sum value returned here + @return none + + @par Scaling and Overflow Behavior + The function is implemented using a 32-bit internal accumulator. + The input is represented in 1.7 format and is accumulated in a 32-bit + accumulator in 25.7 format. + There is no risk of internal overflow with this approach, and the + full precision of intermediate result is preserved. + Finally, the accumulator is truncated to yield a result of 1.7 format. + */ + +#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) + +void arm_accumulate_q7( + const q7_t * pSrc, + uint32_t blockSize, + q7_t * pResult) +{ + uint32_t blkCnt; /* loop counters */ + q7x16_t vecSrc; + q31_t sum = 0L; + + + blkCnt = blockSize >> 4; + while (blkCnt > 0U) + { + vecSrc = vldrbq_s8(pSrc); + /* + * sum lanes + */ + sum = vaddvaq(sum, vecSrc); + + blkCnt--; + pSrc += 16; + } + + blkCnt = blockSize & 0xF; + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store the result to the destination */ + *pResult = sum; +} +#else +void arm_accumulate_q7( + const q7_t * pSrc, + uint32_t blockSize, + q7_t * pResult) +{ + uint32_t blkCnt; /* Loop counter */ + q31_t sum = 0; /* Temporary result storage */ + +#if defined (ARM_MATH_LOOPUNROLL) + q31_t in; +#endif + +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 outputs at a time */ + blkCnt = blockSize >> 2U; + + while (blkCnt > 0U) + { + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + in = read_q7x4_ia (&pSrc); + sum += ((in << 24U) >> 24U); + sum += ((in << 16U) >> 24U); + sum += ((in << 8U) >> 24U); + sum += (in >> 24U); + + /* Decrement the 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 = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + sum += *pSrc++; + + /* Decrement loop counter */ + blkCnt--; + } + + /* C = (A[0] + A[1] + A[2] + ... + A[blockSize-1]) */ + /* Store result to destination */ + *pResult = sum; +} +#endif /* defined(ARM_MATH_MVEI) */ + +/** + @} end of Accumulation group + */