Audio: Volume: Add IPC4 native Q1.31 mode and optimize

This patch adds Q1.31 gain to existing options Q8.16, Q8.23.
The Q1.31 x Q1.31 or Q1.31 x Q1.15 to multiplications are
directly supported by HiFi instructions without need for
additional shifts. Also for HiFi5 the most parallel
multiplications are available for suitable Q1.31 product.

In HiFi5 version s24 and s32 peakvolume and gain the single
multiply instruction AE_MULF32S_HH() or AE_MULF32S_LL() is
changed to quad multiply AE_MULF2P32X4RS().

In HiFi4 and HiFi3 version s24 and s32 the single multiply is
changed to dual multiply AE_MULFP32X2RS().

The MCPS saving in HiFi3/4/5 builds for gain and peakvolume
with 16/24/32 bits stereo 48 kHz formats varies between 0.5%
and 6.1%. At smallest the saving is 0.02 MCPS and at largest
0.2 MCPS.

The savings are quite small because the volume gain computation
has been already optimized and most of the load has happens in
SOF modules framework instead of gain multiplications.

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

Author: singalsu

src/audio/volume/volume.c

@@ -222,7 +222,7 @@ static inline int32_t volume_linear_ramp(struct vol_data *cd, int32_t ramp_time,
 	if (!cd->initial_ramp)
 		return cd->tvolume[channel];
 
-	return cd->rvolume[channel] + ramp_time * cd->ramp_coef[channel];
+	return sat_int32((int64_t)ramp_time * cd->ramp_coef[channel] + cd->rvolume[channel]);
 }
 #endif
 
@@ -469,7 +469,7 @@ static void set_linear_ramp_coef(struct vol_data *cd, int chan, bool constant_ra
 	 * be some accumulated error in ramp time the longer
 	 * the ramp and the smaller the transition is.
 	 */
-	coef = (2 * coef / cd->initial_ramp + 1) >> 1;
+	coef = ((int64_t)coef * 2 / cd->initial_ramp + 1) >> 1;
 
 	/* Scale coefficient by 1/8, round */
 	coef = ((coef >> 2) + 1) >> 1;

src/audio/volume/volume.h

@@ -43,17 +43,24 @@ struct sof_ipc_ctrl_value_chan;
 //** \brief Volume gain Qx.y */
 #define COMP_VOLUME_Q8_16 1
 
-//** \brief Volume gain Qx.y integer x number of bits including sign bit. */
-#define VOL_QXY_X 8
-
-//** \brief Volume gain Qx.y fractional y number of bits. */
-#define VOL_QXY_Y 16
-
 #else
 
 //** \brief Volume gain Qx.y */
-#define COMP_VOLUME_Q1_23 1
+#define COMP_VOLUME_Q1_31	1
+
+#endif
 
+#if COMP_VOLUME_Q1_31
+/** \brief Volume gain Qx.y integer x number of bits including sign bit.
+ * With Q8.23 format the gain range is -138.47 to +42.14 dB.
+ */
+#define VOL_QXY_X 1
+
+//** \brief Volume gain Qx.y fractional y number of bits. */
+#define VOL_QXY_Y 31
+
+#define VOLUME_Q17_47_SHIFT 0
+#elif COMP_VOLUME_Q1_23
 /** \brief Volume gain Qx.y integer x number of bits including sign bit.
  * With Q8.23 format the gain range is -138.47 to +42.14 dB.
  */
@@ -62,6 +69,19 @@ struct sof_ipc_ctrl_value_chan;
 //** \brief Volume gain Qx.y fractional y number of bits. */
 #define VOL_QXY_Y 23
 
+#define VOLUME_Q17_47_SHIFT 8
+
+#elif COMP_VOLUME_Q8_16
+//** \brief Volume gain Qx.y integer x number of bits including sign bit. */
+#define VOL_QXY_X 8
+
+//** \brief Volume gain Qx.y fractional y number of bits. */
+#define VOL_QXY_Y 16
+
+#define VOLUME_Q17_47_SHIFT 15
+
+#else
+#error "Need CONFIG_COMP_VOLUME_Qx_y"
 #endif
 
 /**
@@ -92,16 +112,22 @@ struct sof_ipc_ctrl_value_chan;
  */
 #define PEAK_16S_32C_ADJUST 16
 
-/**
- * \brief Volume maximum value.
- * TODO: This should be 1 << (VOL_QX_BITS + VOL_QY_BITS - 1) - 1 but
- * the current volume code cannot handle the full Q1.16 range correctly.
- */
+#if COMP_VOLUME_Q1_31
+/** \brief Volume maximum value. */
+#define VOL_MAX		INT32_MAX
+
+/** \brief Volume 0dB value. */
+#define VOL_ZERO_DB	INT32_MAX
+
+#else
+/** \brief Volume maximum value. */
 #define VOL_MAX		((1 << (VOL_QXY_X + VOL_QXY_Y - 1)) - 1)
 
 /** \brief Volume 0dB value. */
 #define VOL_ZERO_DB	BIT(VOL_QXY_Y)
 
+#endif
+
 /** \brief Volume minimum value. */
 #define VOL_MIN		0
 

src/audio/volume/volume_hifi3.c

@@ -114,15 +114,16 @@ static void vol_s24_to_s24_s32(struct processing_module *mod, struct input_strea
 			/* Multiply the input sample */
 #if COMP_VOLUME_Q8_16
 			out_sample = AE_MULFP32X2RS(AE_SLAI32S(volume, 7), AE_SLAI32(in_sample, 8));
+			out_sample = AE_SLAI32S(out_sample, 8);
 #elif COMP_VOLUME_Q1_23
 			out_sample = AE_MULFP32X2RS(volume, AE_SLAI32(in_sample, 8));
-#else
-#error "Need CONFIG_COMP_VOLUME_Qx_y"
+			out_sample = AE_SLAI32S(out_sample, 8);
+#elif COMP_VOLUME_Q1_31
+			out_sample = AE_MULFP32X2RS(volume, AE_SLAI32(in_sample, 8));
 #endif
 
-			/* Shift for S24_LE */
-			out_sample = AE_SLAI32S(out_sample, 8);
-			out_sample = AE_SRAI32(out_sample, 8);
+			/* Shift and round for S24_LE */
+			out_sample = AE_SRAI32R(out_sample, 8);
 
 			/* Store the output sample */
 			AE_SA32X2_IP(out_sample, outu, out);
@@ -204,8 +205,6 @@ static void vol_s32_to_s24_s32(struct processing_module *mod, struct input_strea
 	ae_f32x2 out_sample;
 	ae_f32x2 volume;
 	int i, n, m;
-	ae_f64 mult0;
-	ae_f64 mult1;
 	ae_f32x2 *buf;
 	ae_f32x2 *buf_end;
 	ae_f32x2 *vol;
@@ -249,23 +248,19 @@ static void vol_s32_to_s24_s32(struct processing_module *mod, struct input_strea
 
 			/* Load the input sample */
 			AE_LA32X2_IP(in_sample, inu, in);
-
-#if COMP_VOLUME_Q8_16
-			/* Q8.16 x Q1.31 << 1 -> Q9.48 */
-			mult0 = AE_MULF32S_HH(volume, in_sample);
-			mult0 = AE_SRAI64(mult0, 1);			/* Q9.47 */
-			mult1 = AE_MULF32S_LL(volume, in_sample);
-			mult1 = AE_SRAI64(mult1, 1);
-			out_sample = AE_ROUND32X2F48SSYM(mult0, mult1);	/* Q9.47 -> Q1.31 */
-#elif COMP_VOLUME_Q1_23
-			/* Q1.23 x Q1.31 << 1 -> Q2.55 */
-			mult0 = AE_MULF32S_HH(volume, in_sample);
-			mult0 = AE_SRAI64(mult0, 8);			/* Q2.47 */
-			mult1 = AE_MULF32S_LL(volume, in_sample);
-			mult1 = AE_SRAI64(mult1, 8);
-			out_sample = AE_ROUND32X2F48SSYM(mult0, mult1);	/* Q2.47 -> Q1.31 */
+#if COMP_VOLUME_Q1_31
+			out_sample = AE_MULFP32X2RS(volume, in_sample);
 #else
-#error "Need CONFIG_COMP_VOLUME_Qx_y"
+			/* With Q1.31 x Q1.31 -> Q17.47 HiFi multiplications the result is
+			 * Q8.16 x Q1.31 << 1 >> 16 -> Q9.32, shift left by 15 for Q17.47
+			 * Q1.23 x Q1.31 << 1 >> 16 -> Q2.39, shift left by 8 for Q17.47
+			 */
+			ae_f64 mult0 = AE_MULF32R_HH(volume, in_sample);
+			ae_f64 mult1 = AE_MULF32R_LL(volume, in_sample);
+
+			mult0 = AE_SLAI64(mult0, VOLUME_Q17_47_SHIFT);
+			mult1 = AE_SLAI64(mult1, VOLUME_Q17_47_SHIFT);
+			out_sample = AE_ROUND32X2F48SSYM(mult0, mult1);	/* Q2.47 -> Q1.31 */
 #endif
 			AE_SA32X2_IP(out_sample, outu, out);
 		}
@@ -389,25 +384,28 @@ static void vol_s16_to_s16(struct processing_module *mod, struct input_stream_bu
 			/* load second two volume gain */
 			AE_L32X2_XC(volume1, vol, inc);
 
+			/* Load the input sample */
+			AE_LA16X4_IP(in_sample, inu, in);
+
+#if COMP_VOLUME_Q1_31
+			out_sample0 = AE_MULFP32X16X2RS_H(volume0, in_sample);
+			out_sample1 = AE_MULFP32X16X2RS_L(volume1, in_sample);
+#else
 #if COMP_VOLUME_Q8_16
-			/* Q8.16 to Q9.23 */
+			/* Shift Q8.16 to Q9.23
+			 * No need to shift Q1.23, it is OK as such
+			 */
 			volume0 = AE_SLAI32S(volume0, 7);
 			volume1 = AE_SLAI32S(volume1, 7);
-#elif COMP_VOLUME_Q1_23
-			/* No need to shift, Q1.23 is OK as such */
-#else
-#error "Need CONFIG_COMP_VOLUME_Qx_y"
 #endif
-			/* Load the input sample */
-			AE_LA16X4_IP(in_sample, inu, in);
-
 			/* Multiply the input sample */
 			out_sample0 = AE_MULFP32X16X2RS_H(volume0, in_sample);
 			out_sample1 = AE_MULFP32X16X2RS_L(volume1, in_sample);
 
 			/* Q9.23 to Q1.31 */
 			out_sample0 = AE_SLAI32S(out_sample0, 8);
 			out_sample1 = AE_SLAI32S(out_sample1, 8);
+#endif
 
 			/* store the output */
 			out_sample = AE_ROUND16X4F32SSYM(out_sample0, out_sample1);

src/audio/volume/volume_hifi3_with_peakvol.c

@@ -78,18 +78,21 @@ static void vol_s24_to_s24_s32(struct processing_module *mod, struct input_strea
 				AE_L32_XP(in_sample, in, inc);
 				/* calc peak vol */
 				peak_vol = AE_MAXABS32S(in_sample, peak_vol);
-#if COMP_VOLUME_Q8_16
+
 				/* Multiply the input sample */
+#if COMP_VOLUME_Q8_16
 				out_sample = AE_MULFP32X2RS(AE_SLAI32S(volume, 7),
 							    AE_SLAI32(in_sample, 8));
+				out_sample = AE_SLAI32S(out_sample, 8);
 #elif COMP_VOLUME_Q1_23
-				out_sample = AE_MULFP32X2RS(volume, AE_SLAI32S(in_sample, 8));
-#else
-#error "Need CONFIG_COMP_VOLUME_Qx_y"
-#endif
-				/* Shift for S24_LE */
+				out_sample = AE_MULFP32X2RS(volume, AE_SLAI32(in_sample, 8));
 				out_sample = AE_SLAI32S(out_sample, 8);
-				out_sample = AE_SRAI32(out_sample, 8);
+#elif COMP_VOLUME_Q1_31
+				out_sample = AE_MULFP32X2RS(volume, AE_SLAI32(in_sample, 8));
+#endif
+
+				/* Shift and round for S24_LE */
+				out_sample = AE_SRAI32R(out_sample, 8);
 				/* Store the output sample */
 				AE_S32_L_XP(out_sample, out, inc);
 			}
@@ -182,7 +185,6 @@ static void vol_s32_to_s24_s32(struct processing_module *mod, struct input_strea
 	ae_f32x2 out_sample;
 	ae_f32x2 volume;
 	int i, n, channel, m;
-	ae_f64 mult0;
 	const int channels_count = audio_stream_get_channels(sink);
 	const int inc = sizeof(ae_f32) * channels_count;
 	int samples = channels_count * frames;
@@ -214,19 +216,20 @@ static void vol_s32_to_s24_s32(struct processing_module *mod, struct input_strea
 				AE_L32_XP(in_sample, in, inc);
 				/* calc peak vol */
 				peak_vol = AE_MAXABS32S(in_sample, peak_vol);
-#if COMP_VOLUME_Q8_16
-				/* Q8.16 x Q1.31 << 1 -> Q9.48 */
-				mult0 = AE_MULF32S_HH(volume, in_sample);
-				mult0 = AE_SRAI64(mult0, 1);			/* Q9.47 */
-				out_sample = AE_ROUND32F48SASYM(mult0);	/* Q9.47 -> Q1.31 */
-#elif COMP_VOLUME_Q1_23
-				/* Q1.23 x Q1.31 << 1 -> Q2.55 */
-				mult0 = AE_MULF32S_HH(volume, in_sample);
-				mult0 = AE_SRAI64(mult0, 8);			/* Q2.47 */
-				out_sample = AE_ROUND32F48SSYM(mult0);	/* Q2.47 -> Q1.31 */
+
+#if COMP_VOLUME_Q1_31
+				out_sample = AE_MULFP32X2RS(volume, in_sample);
 #else
-#error "Need CONFIG_COMP_VOLUME_Qx_y"
+				/* With Q1.31 x Q1.31 -> Q17.47 HiFi multiplications the result is
+				 * Q8.16 x Q1.31 << 1 >> 16 -> Q9.32, shift left by 15 for Q17.47
+				 * Q1.23 x Q1.31 << 1 >> 16 -> Q2.39, shift left by 8 for Q17.47
+				 */
+				ae_f64 mult0 = AE_MULF32R_HH(volume, in_sample);
+
+				mult0 = AE_SLAI64(mult0, VOLUME_Q17_47_SHIFT);
+				out_sample = AE_ROUND32F48SSYM(mult0);	/* Q2.47 -> Q1.31 */
 #endif
+
 				AE_S32_L_XP(out_sample, out, inc);
 			}
 			peak_vol = AE_SLAA32S(peak_vol, attenuation);
@@ -345,12 +348,10 @@ static void vol_s16_to_s16(struct processing_module *mod, struct input_stream_bu
 			/* Load volume */
 			volume = (ae_f32x2)cd->volume[channel];
 #if COMP_VOLUME_Q8_16
-			/* Q8.16 to Q9.23 */
+			/* Shift Q8.16 to Q9.23
+			 * No need to shift Q1.23, it is OK as such
+			 */
 			volume = AE_SLAI32S(volume, 7);
-#elif COMP_VOLUME_Q1_23
-			/* No need to shift, Q1.23 is OK as such */
-#else
-#error "Need CONFIG_COMP_VOLUME_Qx_y"
 #endif
 			for (i = 0; i < n; i += channels_count) {
 				/* Load the input sample */
@@ -361,8 +362,12 @@ static void vol_s16_to_s16(struct processing_module *mod, struct input_stream_bu
 				/* Multiply the input sample */
 				out_sample0 = AE_MULFP32X16X2RS_H(volume, in_sample);
 
+#if COMP_VOLUME_Q1_31
+				/* No shift need, the product is Q1.31 */
+#else
 				/* Q9.23 to Q1.31 */
 				out_sample0 = AE_SLAI32S(out_sample0, 8);
+#endif
 
 				/* store the output */
 				out_sample = AE_ROUND16X4F32SSYM(out_sample0, out_sample0);