Corrected issue #124 (c35c8d2d) · Commits · Wenxi XU / Cmsis Dsp

Source/FilteringFunctions/arm_conv_q15.c

+14 −2

Original line number	Diff line number	Diff line
		@@ -58,10 +58,13 @@
		@remark
		Refer to \ref arm_conv_opt_q15() for a faster implementation of this function using scratch buffers.
		*/


		#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
		#include "arm_helium_utils.h"
		#include "arm_vec_filtering.h"


		void arm_conv_q15(
		const q15_t * pSrcA,
		uint32_t srcALen,
		@@ -82,6 +85,7 @@ void arm_conv_q15(
		int32_t block1, block2, block3;



		uint16x8_t decrIdxVec = vddupq_u16(7, 1);


		@@ -107,6 +111,7 @@ void arm_conv_q15(
		block2 = srcALen - srcBLen + 1;
		block3 = srcBLen - 1;


		pA = pIn1;
		pB = pIn2 - 7;

		@@ -192,7 +197,7 @@ void arm_conv_q15(
		pA++;
		}

		for (i = block3; i >= 1; i -= 2)
		for (i = block3; i >= 2; i -= 2)
		{
		uint32_t count = i;
		int64_t acc0 = 0LL;
		@@ -206,7 +211,7 @@ void arm_conv_q15(
		*pDst++ = (q15_t) acc1;
		pA += 2;
		}
		for (; i >= 1; i--)
		for (; i > 0; i--)
		{
		uint32_t count = i;
		int64_t acc = 0LL;
		@@ -218,6 +223,8 @@ void arm_conv_q15(
		*pDst++ = (q15_t) acc;
		pA++;
		}


		}
		#else
		void arm_conv_q15(
		@@ -241,6 +248,8 @@ void arm_conv_q15(
		uint32_t blockSize1, blockSize2, blockSize3; /* Loop counters */
		uint32_t j, k, count, blkCnt; /* Loop counters */



		/* The algorithm implementation is based on the lengths of the inputs. */
		/* srcB is always made to slide across srcA. */
		/* So srcBLen is always considered as shorter or equal to srcALen */
		@@ -280,6 +289,9 @@ void arm_conv_q15(
		blockSize1 = srcBLen - 1U;
		blockSize2 = srcALen - (srcBLen - 1U);




		/* --------------------------
		* Initializations of stage1
		* -------------------------*/

+1 −1

Original line number	Diff line number	Diff line
		@@ -84,6 +84,7 @@ void arm_conv_q31(
		uint32_t vddupStartIdx = 3;
		uint32x4_t decrIdxVec = vddupq_u32(vddupStartIdx, 1);


		if (srcALen < srcBLen)
		{
		/*
		@@ -219,7 +220,6 @@ void arm_conv_q31(
		*pDst++ = (q31_t) acc;
		pA++;
		}

		}

		#else

+4 −2

Original line number	Diff line number	Diff line
		@@ -55,6 +55,7 @@
		@remark
		Refer to \ref arm_conv_opt_q7() for a faster implementation of this function.
		*/

		#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
		#include "arm_helium_utils.h"

		@@ -186,7 +187,7 @@ void arm_conv_q7(
		pA++;
		}

		for (i = block3; i >= 1; i -= 2)
		for (i = block3; i >= 2; i -= 2)
		{
		uint32_t count = i;
		int32_t acc0 = 0;
		@@ -200,7 +201,7 @@ void arm_conv_q7(
		*pDst++ = (q7_t) acc1;
		pA += 2;
		}
		for (; i >= 1; i--)
		for (; i > 0; i--)
		{
		uint32_t count = i;
		int32_t acc = 0;
		@@ -212,6 +213,7 @@ void arm_conv_q7(
		*pDst++ = (q7_t) acc;
		pA++;
		}

		}

		#else

+1 −0

Original line number	Diff line number	Diff line
		@@ -42,6 +42,7 @@ a double precision computation.
		inpB, inputB.nbSamples(),
		outp);

		ASSERT_EMPTY_TAIL(output);
		ASSERT_SNR(ref,output,(q15_t)SNR_THRESHOLD);
		ASSERT_NEAR_EQ(ref,output,ABS_ERROR_Q15);

+1 −0

Original line number	Diff line number	Diff line
		@@ -70,6 +70,7 @@ For tests of the error value of the Levinson Durbin algorithm
		inpB, inputB.nbSamples(),
		outp);

		ASSERT_EMPTY_TAIL(output);
		ASSERT_SNR(ref,output,(q31_t)SNR_THRESHOLD);
		ASSERT_NEAR_EQ(ref,output,ABS_ERROR_Q31);