Math: FIR: Copy HiFi5 filter functions from HiFi3 version

This patch contains only minimal changes to build this with
HiFi5 toolchain. Changed SOF_USE_HIFI() / SOF_USE_MIN_HIFI()
macros usage and included hifi5 header.

Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>

Author: singalsu

src/math/CMakeLists.txt

@@ -29,7 +29,7 @@ add_local_sources_ifdef(CONFIG_POWER_FIXED sof power.c)
 
 add_local_sources_ifdef(CONFIG_BINARY_LOGARITHM_FIXED sof base2log.c)
 
-add_local_sources_ifdef(CONFIG_MATH_FIR sof fir_generic.c fir_hifi2ep.c fir_hifi3.c)
+add_local_sources_ifdef(CONFIG_MATH_FIR sof fir_generic.c fir_hifi2ep.c fir_hifi3.c fir_hifi5.c)
 
 if(CONFIG_MATH_FFT)
 	add_subdirectory(fft)

src/math/fir_hifi3.c

@@ -7,7 +7,7 @@
 #include <sof/math/fir_config.h>
 #include <sof/common.h>
 
-#if SOF_USE_MIN_HIFI(3, FILTER)
+#if SOF_USE_HIFI(3, FILTER) || SOF_USE_HIFI(4, FILTER)
 
 #include <sof/audio/buffer.h>
 #include <sof/math/fir_hifi3.h>

src/math/fir_hifi5.c

@@ -0,0 +1,256 @@
+// SPDX-License-Identifier: BSD-3-Clause
+//
+// Copyright(c) 2017-2025 Intel Corporation.
+//
+// Author: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
+
+#include <sof/math/fir_config.h>
+#include <sof/common.h>
+
+#if SOF_USE_MIN_HIFI(5, FILTER)
+
+#include <sof/audio/buffer.h>
+#include <sof/math/fir_hifi3.h>
+#include <user/fir.h>
+#include <xtensa/config/defs.h>
+#include <xtensa/tie/xt_hifi5.h>
+#include <rtos/symbol.h>
+#include <errno.h>
+#include <stddef.h>
+#include <stdint.h>
+
+/*
+ * EQ FIR algorithm code
+ */
+
+void fir_reset(struct fir_state_32x16 *fir)
+{
+	fir->taps = 0;
+	fir->length = 0;
+	fir->out_shift = 0;
+	fir->coef = NULL;
+	/* There may need to know the beginning of dynamic allocation after
+	 * reset so omitting setting also fir->delay to NULL.
+	 */
+}
+EXPORT_SYMBOL(fir_reset);
+
+int fir_delay_size(struct sof_fir_coef_data *config)
+{
+	/* Check FIR tap count for implementation specific constraints */
+	if (config->length > SOF_FIR_MAX_LENGTH || config->length < 4)
+		return -EINVAL;
+
+	/* The optimization requires the tap count to be multiple of four */
+	if (config->length & 0x3)
+		return -EINVAL;
+
+	/* The dual sample version needs one more delay entry. To preserve
+	 * align for 64 bits need to add two.
+	 */
+	return (config->length + 2) * sizeof(int32_t);
+}
+EXPORT_SYMBOL(fir_delay_size);
+
+int fir_init_coef(struct fir_state_32x16 *fir,
+		  struct sof_fir_coef_data *config)
+{
+	/* The length is taps plus two since the filter computes two
+	 * samples per call. Length plus one would be minimum but the add
+	 * must be even. The even length is needed for 64 bit loads from delay
+	 * lines with 32 bit samples.
+	 */
+	fir->taps = (int)config->length;
+	fir->length = fir->taps + 2;
+	fir->out_shift = (int)config->out_shift;
+	fir->coef = (ae_f16x4 *)&config->coef[0];
+	return 0;
+}
+EXPORT_SYMBOL(fir_init_coef);
+
+void fir_init_delay(struct fir_state_32x16 *fir, int32_t **data)
+{
+	fir->delay = (ae_int32 *)*data;
+	fir->delay_end = fir->delay + fir->length;
+	fir->rwp = (ae_int32 *)(fir->delay + fir->length - 1);
+	*data += fir->length; /* Point to next delay line start */
+}
+EXPORT_SYMBOL(fir_init_delay);
+
+void fir_get_lrshifts(struct fir_state_32x16 *fir, int *lshift,
+		      int *rshift)
+{
+	*lshift = (fir->out_shift < 0) ? -fir->out_shift : 0;
+	*rshift = (fir->out_shift > 0) ? fir->out_shift : 0;
+}
+EXPORT_SYMBOL(fir_get_lrshifts);
+
+/* HiFi EP has the follow number of reqisters that should not be exceeded
+ * 4x 56 bit registers in register file Q
+ * 8x 48 bit registers in register file P
+ */
+
+void fir_32x16(struct fir_state_32x16 *fir, ae_int32 x, ae_int32 *y, int shift)
+{
+	/* This function uses
+	 * 1x 56 bit registers Q,
+	 * 4x 48 bit registers P
+	 * 3x integers
+	 * 2x address pointers,
+	 */
+	ae_f64 a;
+	ae_valign u;
+	ae_f32x2 data2;
+	ae_f16x4 coefs;
+	ae_f32x2 d0;
+	ae_f32x2 d1;
+	int i;
+	ae_int32 *dp = fir->rwp;
+	ae_int16x4 *coefp = (ae_int16x4 *)fir->coef;
+	const int taps_div_4 = fir->taps >> 2;
+	const int inc = sizeof(int32_t);
+
+	/* Bypass samples if taps count is zero. */
+	if (!taps_div_4) {
+		*y = x;
+		return;
+	}
+
+	/* Write sample to delay */
+	AE_S32_L_XC(x, fir->rwp, -sizeof(int32_t));
+
+	/* Prime the coefficients stream */
+	u = AE_LA64_PP(coefp);
+
+	/* Note: If the next function is converted to handle two samples
+	 * per call the data load can be done with single instruction
+	 * AE_LP24X2F_C(data2, dp, sizeof(ae_p24x2f));
+	 */
+	a = AE_ZEROQ56();
+	for (i = 0; i < taps_div_4; i++) {
+		/* Load four coefficients. Coef_3 contains tap h[n],
+		 * coef_2 contains h[n+1], coef_1 contains h[n+2], and
+		 * coef_0 contains h[n+3];
+		 */
+		AE_LA16X4_IP(coefs, u, coefp);
+
+		/* Load two data samples and pack to d0 to data2_h and
+		 * d1 to data2_l.
+		 */
+		AE_L32_XC(d0, dp, inc);
+		AE_L32_XC(d1, dp, inc);
+		data2 = AE_SEL32_LL(d0, d1);
+
+		/* Accumulate
+		 * a += data2_h * coefs_3 + data2_l * coefs_2. The Q1.31
+		 * data and Q1.15 coefficients are used as 24 bits as
+		 * Q1.23 values.
+		 */
+		AE_MULAAFD32X16_H3_L2(a, data2, coefs);
+
+		/* Repeat the same for next two taps and increase coefp.
+		 * a += data2_h * coefs_1 + data2_l * coefs_0.
+		 */
+		AE_L32_XC(d0, dp, inc);
+		AE_L32_XC(d1, dp, inc);
+		data2 = AE_SEL32_LL(d0, d1);
+		AE_MULAAFD32X16_H1_L0(a, data2, coefs);
+	}
+
+	/* Do scaling shifts and store sample. */
+	a = AE_SLAA64S(a, shift);
+	AE_S32_L_I(AE_ROUND32F48SSYM(a), (ae_int32 *)y, 0);
+}
+EXPORT_SYMBOL(fir_32x16);
+
+/* HiFi EP has the follow number of reqisters that should not be exceeded
+ * 4x 56 bit registers in register file Q
+ * 8x 48 bit registers in register file P
+ */
+
+void fir_32x16_2x(struct fir_state_32x16 *fir, ae_int32 x0, ae_int32 x1,
+		  ae_int32 *y0, ae_int32 *y1, int shift)
+{
+	/* This function uses
+	 * 2x 56 bit registers Q,
+	 * 4x 48 bit registers P
+	 * 3x integers
+	 * 2x address pointers,
+	 */
+	ae_f64 a;
+	ae_f64 b;
+	ae_valign u;
+	ae_f32x2 d0;
+	ae_f32x2 d1;
+	ae_f16x4 coefs;
+	int i;
+	ae_f32x2 *dp;
+	ae_f16x4 *coefp = fir->coef;
+	const int taps_div_4 = fir->taps >> 2;
+	const int inc = 2 * sizeof(int32_t);
+
+	/* Bypass samples if taps count is zero. */
+	if (!taps_div_4) {
+		*y0 = x0;
+		*y1 = x1;
+		return;
+	}
+
+	/* Write samples to delay */
+	AE_S32_L_XC(x0, fir->rwp, -sizeof(int32_t));
+	dp = (ae_f32x2 *)fir->rwp;
+	AE_S32_L_XC(x1, fir->rwp, -sizeof(int32_t));
+
+	/* Note: If the next function is converted to handle two samples
+	 * per call the data load can be done with single instruction
+	 * AE_LP24X2F_C(data2, dp, sizeof(ae_p24x2f));
+	 */
+	a = AE_ZERO64();
+	b = AE_ZERO64();
+
+	/* Prime the coefficients stream */
+	u = AE_LA64_PP(coefp);
+
+	/* Load two data samples and pack to d0 to data2_h and
+	 * d1 to data2_l.
+	 */
+	AE_L32X2_XC(d0, dp, inc);
+	for (i = 0; i < taps_div_4; i++) {
+		/* Load four coefficients. Coef_3 contains tap h[n],
+		 * coef_2 contains h[n+1], coef_1 contains h[n+2], and
+		 * coef_0 contains h[n+3];
+		 */
+		AE_LA16X4_IP(coefs, u, coefp);
+
+		/* Load two data samples. Upper part d1_h is x[n+1] and
+		 * lower part d1_l is x[n].
+		 */
+		AE_L32X2_XC(d1, dp, inc);
+
+		/* Quad MAC (HH)
+		 * b += d0_h * coefs_3 + d0_l * coefs_2
+		 * a += d0_l * coefs_3 + d1_h * coefs_2
+		 */
+		AE_MULAFD32X16X2_FIR_HH(b, a, d0, d1, coefs);
+		d0 = d1;
+
+		/* Repeat the same for next two taps and increase coefp. */
+		AE_L32X2_XC(d1, dp, inc);
+
+		/* Quad MAC (HL)
+		 * b += d0_h * coefs_1 + d0_l * coefs_0
+		 * a += d0_l * coefs_1 + d1_h * coefs_0
+		 */
+		AE_MULAFD32X16X2_FIR_HL(b, a, d0, d1, coefs);
+		d0 = d1;
+	}
+
+	/* Do scaling shifts and store sample. */
+	b = AE_SLAA64S(b, shift);
+	a = AE_SLAA64S(a, shift);
+	AE_S32_L_I(AE_ROUND32F48SSYM(b), (ae_int32 *)y1, 0);
+	AE_S32_L_I(AE_ROUND32F48SSYM(a), (ae_int32 *)y0, 0);
+}
+EXPORT_SYMBOL(fir_32x16_2x);
+
+#endif

zephyr/CMakeLists.txt

@@ -694,6 +694,7 @@ elseif(CONFIG_MATH_FIR)
 		${SOF_MATH_PATH}/fir_generic.c
 		${SOF_MATH_PATH}/fir_hifi2ep.c
 		${SOF_MATH_PATH}/fir_hifi3.c
+		${SOF_MATH_PATH}/fir_hifi5.c
 	)
 endif()