Revert "Removed some gcc workarounds for Helium that are no more needed." (0e8b8ec8) · Commits · Wenxi XU / Cmsis Dsp

Source/FilteringFunctions/arm_biquad_cascade_stereo_df2T_f16.c

+6 −1

Original line number	Diff line number	Diff line
		@@ -45,7 +45,12 @@
		@param[out] pDst points to the block of output data
		@param[in] blockSize number of samples to process
		*/
		#if (defined(ARM_MATH_MVE_FLOAT16) && defined(ARM_MATH_HELIUM_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE)

		#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) && defined(ARM_DSP_BUILT_WITH_GCC)
		#pragma GCC warning "Scalar version of arm_biquad_cascade_stereo_df2T_f16 built. Helium version has build issues with gcc."
		#endif

		#if (defined(ARM_MATH_MVE_FLOAT16) && defined(ARM_MATH_HELIUM_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE) && !defined(ARM_DSP_BUILT_WITH_GCC)
		ARM_DSP_ATTRIBUTE void arm_biquad_cascade_stereo_df2T_f16(
		const arm_biquad_cascade_stereo_df2T_instance_f16 * S,
		const float16_t * pSrc,

+5 −1

Original line number	Diff line number	Diff line
		@@ -56,7 +56,11 @@
		Refer to \ref arm_correlate_opt_q7() for a faster implementation of this function.
		*/

		#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
		#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) && defined(ARM_DSP_BUILT_WITH_GCC)
		#pragma GCC warning "Scalar version of arm_correlate_q7 built. Helium version has build issues with gcc."
		#endif

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

		#include "arm_vec_filtering.h"

+5 −1

Original line number	Diff line number	Diff line
		@@ -51,7 +51,11 @@
		- \ref ARM_MATH_SIZE_MISMATCH : Matrix size check failed
		*/

		#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
		#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) && defined(ARM_DSP_BUILT_WITH_GCC)
		#pragma GCC warning "Scalar version of arm_mat_cmplx_mult_f16 built. Helium version has build issues with gcc."
		#endif

		#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) && !defined(ARM_DSP_BUILT_WITH_GCC)

		#include "arm_helium_utils.h"

+6 −1

Original line number	Diff line number	Diff line
		@@ -45,7 +45,12 @@
		@param[out] pIndex index of maximum value returned here
		*/

		#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
		#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) && defined(ARM_DSP_BUILT_WITH_GCC)
		#pragma GCC warning "Scalar version of arm_absmax_q7 built. Helium version has build issues with gcc."
		#endif


		#if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE) && !defined(ARM_DSP_BUILT_WITH_GCC)

		#include <stdint.h>
		#include "arm_helium_utils.h"

+10 −1

Original line number	Diff line number	Diff line
		@@ -159,6 +159,11 @@ ARM_DSP_ATTRIBUTE void arm_rfft_q15(
		#include "arm_helium_utils.h"
		#include "arm_vec_fft.h"

		#if defined(ARM_DSP_BUILT_WITH_GCC)
		#define MVE_CMPLX_MULT_FX_AxB_S16(A,B) vqdmladhxq_s16(vqdmlsdhq_s16((__typeof(A))vuninitializedq_s16(), A, B), A, B)
		#define MVE_CMPLX_MULT_FX_AxConjB_S16(A,B) vqdmladhq_s16(vqdmlsdhxq_s16((__typeof(A))vuninitializedq_s16(), A, B), A, B)

		#endif

		ARM_DSP_ATTRIBUTE void arm_split_rfft_q15(
		q15_t * pSrc,
		@@ -200,9 +205,13 @@ ARM_DSP_ATTRIBUTE void arm_split_rfft_q15(
		q15x8_t coefA = vldrhq_gather_shifted_offset_s16(pCoefAb, offsetCoef);
		q15x8_t coefB = vldrhq_gather_shifted_offset_s16(pCoefBb, offsetCoef);


		#if defined(ARM_DSP_BUILT_WITH_GCC)
		q15x8_t out = vhaddq_s16(MVE_CMPLX_MULT_FX_AxB_S16(in1, coefA),
		MVE_CMPLX_MULT_FX_AxConjB_S16(coefB, in2));
		#else
		q15x8_t out = vhaddq_s16(MVE_CMPLX_MULT_FX_AxB(in1, coefA, q15x8_t),
		MVE_CMPLX_MULT_FX_AxConjB(coefB, in2, q15x8_t));
		#endif
		vst1q_s16(pOut1, out);
		pOut1 += 8;