From 923efad6dd5da1dcef0e53c8e6cac318abe032ed Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Marek=20Piku=C5=82a?= <m.pikula@partner.samsung.com>
Date: Thu, 20 Feb 2025 16:59:09 +0100
Subject: [PATCH] Streamline implementation overrides
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Decrease code repetition for arch-specific function overrides.
Signed-off-by: Marek PikuĊa <m.pikula@partner.samsung.com>
---
silk/NSQ.c | 19 +------
silk/VQ_WMat_EC.c | 15 +-----
silk/fixed/encode_frame_FIX.c | 9 ++--
silk/float/wrappers_FLP.c | 4 +-
silk/main.h | 48 +++---------------
silk/main_overrides.h | 88 ++++++++++++++++++++++++++++++++
silk/quant_LTP_gains.c | 4 +-
silk/x86/NSQ_sse4_1.c | 18 +------
silk/x86/VQ_WMat_EC_sse4_1.c | 15 +-----
silk/x86/main_sse.h | 96 ++++-------------------------------
silk/x86/x86_silk_map.c | 33 +-----------
silk_headers.mk | 1 +
12 files changed, 121 insertions(+), 229 deletions(-)
create mode 100644 silk/main_overrides.h
diff --git a/silk/NSQ.c b/silk/NSQ.c
index 1caa829bb..b6529f4cf 100644
--- a/silk/NSQ.c
+++ b/silk/NSQ.c
@@ -73,24 +73,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer(
);
#endif
-void silk_NSQ_c
-(
- const silk_encoder_state *psEncC, /* I Encoder State */
- silk_nsq_state *NSQ, /* I/O NSQ state */
- SideInfoIndices *psIndices, /* I/O Quantization Indices */
- const opus_int16 x16[], /* I Input */
- opus_int8 pulses[], /* O Quantized pulse signal */
- const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */
- const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */
- const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */
- const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */
- const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */
- const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */
- const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */
- const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */
- const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */
- const opus_int LTP_scale_Q14 /* I LTP state scaling */
-)
+SILK_NSQ_DECL(c)
{
opus_int k, lag, start_idx, LSF_interpolation_flag;
const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13;
diff --git a/silk/VQ_WMat_EC.c b/silk/VQ_WMat_EC.c
index 245a7e4b0..acdd1d75a 100644
--- a/silk/VQ_WMat_EC.c
+++ b/silk/VQ_WMat_EC.c
@@ -32,20 +32,7 @@ POSSIBILITY OF SUCH DAMAGE.
#include "main.h"
/* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */
-void silk_VQ_WMat_EC_c(
- opus_int8 *ind, /* O index of best codebook vector */
- opus_int32 *res_nrg_Q15, /* O best residual energy */
- opus_int32 *rate_dist_Q8, /* O best total bitrate */
- opus_int *gain_Q7, /* O sum of absolute LTP coefficients */
- const opus_int32 *XX_Q17, /* I correlation matrix */
- const opus_int32 *xX_Q17, /* I correlation vector */
- const opus_int8 *cb_Q7, /* I codebook */
- const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */
- const opus_uint8 *cl_Q5, /* I code length for each codebook vector */
- const opus_int subfr_len, /* I number of samples per subframe */
- const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
- const opus_int L /* I number of vectors in codebook */
-)
+SILK_VQ_WMAT_EC_DECL(c)
{
opus_int k, gain_tmp_Q7;
const opus_int8 *cb_row_Q7;
diff --git a/silk/fixed/encode_frame_FIX.c b/silk/fixed/encode_frame_FIX.c
index 7c83360ba..4600422ef 100644
--- a/silk/fixed/encode_frame_FIX.c
+++ b/silk/fixed/encode_frame_FIX.c
@@ -206,10 +206,9 @@ opus_int silk_encode_frame_FIX(
sEncCtrl.Tilt_Q14, sEncCtrl.LF_shp_Q14, sEncCtrl.Gains_Q16, sEncCtrl.pitchL, sEncCtrl.Lambda_Q10, sEncCtrl.LTP_scale_Q14,
psEnc->sCmn.arch );
} else {
- silk_NSQ( &psEnc->sCmn, &psEnc->sCmn.sNSQ, &psEnc->sCmn.indices, x_frame, psEnc->sCmn.pulses,
+ silk_NSQ( psEnc->sCmn.arch, &psEnc->sCmn, &psEnc->sCmn.sNSQ, &psEnc->sCmn.indices, x_frame, psEnc->sCmn.pulses,
sEncCtrl.PredCoef_Q12[ 0 ], sEncCtrl.LTPCoef_Q14, sEncCtrl.AR_Q13, sEncCtrl.HarmShapeGain_Q14,
- sEncCtrl.Tilt_Q14, sEncCtrl.LF_shp_Q14, sEncCtrl.Gains_Q16, sEncCtrl.pitchL, sEncCtrl.Lambda_Q10, sEncCtrl.LTP_scale_Q14,
- psEnc->sCmn.arch);
+ sEncCtrl.Tilt_Q14, sEncCtrl.LF_shp_Q14, sEncCtrl.Gains_Q16, sEncCtrl.pitchL, sEncCtrl.Lambda_Q10, sEncCtrl.LTP_scale_Q14 );
}
if ( iter == maxIter && !found_lower ) {
@@ -432,10 +431,10 @@ static OPUS_INLINE void silk_LBRR_encode_FIX(
psEncCtrl->AR_Q13, psEncCtrl->HarmShapeGain_Q14, psEncCtrl->Tilt_Q14, psEncCtrl->LF_shp_Q14,
psEncCtrl->Gains_Q16, psEncCtrl->pitchL, psEncCtrl->Lambda_Q10, psEncCtrl->LTP_scale_Q14, psEnc->sCmn.arch );
} else {
- silk_NSQ( &psEnc->sCmn, &sNSQ_LBRR, psIndices_LBRR, x16,
+ silk_NSQ( psEnc->sCmn.arch, &psEnc->sCmn, &sNSQ_LBRR, psIndices_LBRR, x16,
psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], psEncCtrl->PredCoef_Q12[ 0 ], psEncCtrl->LTPCoef_Q14,
psEncCtrl->AR_Q13, psEncCtrl->HarmShapeGain_Q14, psEncCtrl->Tilt_Q14, psEncCtrl->LF_shp_Q14,
- psEncCtrl->Gains_Q16, psEncCtrl->pitchL, psEncCtrl->Lambda_Q10, psEncCtrl->LTP_scale_Q14, psEnc->sCmn.arch );
+ psEncCtrl->Gains_Q16, psEncCtrl->pitchL, psEncCtrl->Lambda_Q10, psEncCtrl->LTP_scale_Q14 );
}
/* Restore original gains */
diff --git a/silk/float/wrappers_FLP.c b/silk/float/wrappers_FLP.c
index c0c183e35..4c8de4dc3 100644
--- a/silk/float/wrappers_FLP.c
+++ b/silk/float/wrappers_FLP.c
@@ -164,8 +164,8 @@ void silk_NSQ_wrapper_FLP(
silk_NSQ_del_dec( &psEnc->sCmn, psNSQ, psIndices, x16, pulses, PredCoef_Q12[ 0 ], LTPCoef_Q14,
AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14, psEnc->sCmn.arch );
} else {
- silk_NSQ( &psEnc->sCmn, psNSQ, psIndices, x16, pulses, PredCoef_Q12[ 0 ], LTPCoef_Q14,
- AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14, psEnc->sCmn.arch );
+ silk_NSQ( psEnc->sCmn.arch, &psEnc->sCmn, psNSQ, psIndices, x16, pulses, PredCoef_Q12[ 0 ], LTPCoef_Q14,
+ AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14 );
}
}
diff --git a/silk/main.h b/silk/main.h
index cd576d8cc..abfab5b07 100644
--- a/silk/main.h
+++ b/silk/main.h
@@ -37,6 +37,7 @@ POSSIBILITY OF SUCH DAMAGE.
#include "debug.h"
#include "entenc.h"
#include "entdec.h"
+#include "main_overrides.h"
#if defined(OPUS_X86_MAY_HAVE_SSE4_1)
#include "x86/main_sse.h"
@@ -222,53 +223,18 @@ void silk_quant_LTP_gains(
);
/* Entropy constrained matrix-weighted VQ, for a single input data vector */
-void silk_VQ_WMat_EC_c(
- opus_int8 *ind, /* O index of best codebook vector */
- opus_int32 *res_nrg_Q15, /* O best residual energy */
- opus_int32 *rate_dist_Q8, /* O best total bitrate */
- opus_int *gain_Q7, /* O sum of absolute LTP coefficients */
- const opus_int32 *XX_Q17, /* I correlation matrix */
- const opus_int32 *xX_Q17, /* I correlation vector */
- const opus_int8 *cb_Q7, /* I codebook */
- const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */
- const opus_uint8 *cl_Q5, /* I code length for each codebook vector */
- const opus_int subfr_len, /* I number of samples per subframe */
- const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
- const opus_int L /* I number of vectors in codebook */
-);
-
-#if !defined(OVERRIDE_silk_VQ_WMat_EC)
-#define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, subfr_len, max_gain_Q7, L, arch) \
- ((void)(arch),silk_VQ_WMat_EC_c(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, subfr_len, max_gain_Q7, L))
+SILK_VQ_WMAT_EC_DECL(c);
+#ifndef silk_VQ_WMat_EC
+#define silk_VQ_WMat_EC(...) OVERRIDE_IMPL_SINGLE(silk_VQ_WMat_EC, c, __VA_ARGS__)
#endif
/************************************/
/* Noise shaping quantization (NSQ) */
/************************************/
-void silk_NSQ_c(
- const silk_encoder_state *psEncC, /* I Encoder State */
- silk_nsq_state *NSQ, /* I/O NSQ state */
- SideInfoIndices *psIndices, /* I/O Quantization Indices */
- const opus_int16 x16[], /* I Input */
- opus_int8 pulses[], /* O Quantized pulse signal */
- const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */
- const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */
- const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */
- const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */
- const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */
- const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */
- const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */
- const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */
- const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */
- const opus_int LTP_scale_Q14 /* I LTP state scaling */
-);
-
-#if !defined(OVERRIDE_silk_NSQ)
-#define silk_NSQ(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \
- HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \
- ((void)(arch),silk_NSQ_c(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \
- HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14))
+SILK_NSQ_DECL(c);
+#ifndef silk_NSQ
+#define silk_NSQ(...) OVERRIDE_IMPL_SINGLE(silk_NSQ, c, __VA_ARGS__)
#endif
/* Noise shaping using delayed decision */
diff --git a/silk/main_overrides.h b/silk/main_overrides.h
new file mode 100644
index 000000000..7b1143664
--- /dev/null
+++ b/silk/main_overrides.h
@@ -0,0 +1,88 @@
+/* Copyright (c) 2025, Samsung R&D Institute Poland
+ Written by Marek Pikula
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+
+ - Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+
+ - Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+ OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef SILK_MAIN_OVERRIDES_H
+#define SILK_MAIN_OVERRIDES_H
+
+/** Override implementation with a single function. */
+#define OVERRIDE_IMPL_SINGLE(func, impl, arch, ...) \
+ ((void)(arch), func ## _ ## impl(__VA_ARGS__))
+
+/** Override implementation with a multiple arch-specific functions.
+ *
+ * It uses an array declared with OVERRIDE_IMPL_ARRAY_DECL().
+ */
+#define OVERRIDE_IMPL_ARRAY(func, arch, ...) \
+ ((*OVERRIDE_MAP_ ## func[(arch) & OPUS_ARCHMASK])(__VA_ARGS__))
+
+/** Declare a mapping array for use with OVERRIDE_IMPL_ARRAY(). */
+#define OVERRIDE_IMPL_ARRAY_DECL(func) \
+ const func ## _t OVERRIDE_MAP_ ## func[ OPUS_ARCHMASK + 1 ]
+
+
+// Function declarations.
+
+/* Entropy constrained matrix-weighted VQ, for a single input data vector */
+#define SILK_VQ_WMAT_EC_DECL(impl, ...) \
+ void (__VA_ARGS__ silk_VQ_WMat_EC_ ## impl)( \
+ opus_int8 *ind, /* O index of best codebook vector */ \
+ opus_int32 *res_nrg_Q15, /* O best residual energy */ \
+ opus_int32 *rate_dist_Q8, /* O best total bitrate */ \
+ opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ \
+ const opus_int32 *XX_Q17, /* I correlation matrix */ \
+ const opus_int32 *xX_Q17, /* I correlation vector */ \
+ const opus_int8 *cb_Q7, /* I codebook */ \
+ const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ \
+ const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ \
+ const opus_int subfr_len, /* I number of samples per subframe */ \
+ const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ \
+ const opus_int L /* I number of vectors in codebook */ \
+ )
+typedef SILK_VQ_WMAT_EC_DECL(t, *const);
+
+/* Noise shaping quantization (NSQ) */
+#define SILK_NSQ_DECL(impl, ...) \
+ void (__VA_ARGS__ silk_NSQ_ ## impl)( \
+ const silk_encoder_state *psEncC, /* I Encoder State */ \
+ silk_nsq_state *NSQ, /* I/O NSQ state */ \
+ SideInfoIndices *psIndices, /* I/O Quantization Indices */ \
+ const opus_int16 x16[], /* I Input */ \
+ opus_int8 pulses[], /* O Quantized pulse signal */ \
+ const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ \
+ const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ \
+ const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ \
+ const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ \
+ const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ \
+ const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ \
+ const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ \
+ const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ \
+ const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ \
+ const opus_int LTP_scale_Q14 /* I LTP state scaling */ \
+ )
+typedef SILK_NSQ_DECL(t, *const);
+
+#endif // SILK_MAIN_OVERRIDES_H
diff --git a/silk/quant_LTP_gains.c b/silk/quant_LTP_gains.c
index d6b8eff8d..8a7e812ea 100644
--- a/silk/quant_LTP_gains.c
+++ b/silk/quant_LTP_gains.c
@@ -82,6 +82,7 @@ void silk_quant_LTP_gains(
max_gain_Q7 = silk_log2lin( ( SILK_FIX_CONST( MAX_SUM_LOG_GAIN_DB / 6.0, 7 ) - sum_log_gain_tmp_Q7 )
+ SILK_FIX_CONST( 7, 7 ) ) - gain_safety;
silk_VQ_WMat_EC(
+ arch, /* I Run-time architecture */
&temp_idx[ j ], /* O index of best codebook vector */
&res_nrg_Q15_subfr, /* O residual energy */
&rate_dist_Q7_subfr, /* O best weighted quantization error + mu * rate */
@@ -93,8 +94,7 @@ void silk_quant_LTP_gains(
cl_ptr_Q5, /* I code length for each codebook vector */
subfr_len, /* I number of samples per subframe */
max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
- cbk_size, /* I number of vectors in codebook */
- arch /* I Run-time architecture */
+ cbk_size /* I number of vectors in codebook */
);
res_nrg_Q15 = silk_ADD_POS_SAT32( res_nrg_Q15, res_nrg_Q15_subfr );
diff --git a/silk/x86/NSQ_sse4_1.c b/silk/x86/NSQ_sse4_1.c
index 3c9aca7ba..7a0170bff 100644
--- a/silk/x86/NSQ_sse4_1.c
+++ b/silk/x86/NSQ_sse4_1.c
@@ -71,23 +71,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1(
opus_int32 table[][4] /* I */
);
-void silk_NSQ_sse4_1(
- const silk_encoder_state *psEncC, /* I Encoder State */
- silk_nsq_state *NSQ, /* I/O NSQ state */
- SideInfoIndices *psIndices, /* I/O Quantization Indices */
- const opus_int16 x16[], /* I Input */
- opus_int8 pulses[], /* O Quantized pulse signal */
- const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */
- const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */
- const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */
- const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */
- const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */
- const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */
- const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */
- const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */
- const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */
- const opus_int LTP_scale_Q14 /* I LTP state scaling */
-)
+SILK_NSQ_DECL(sse4_1)
{
opus_int k, lag, start_idx, LSF_interpolation_flag;
const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13;
diff --git a/silk/x86/VQ_WMat_EC_sse4_1.c b/silk/x86/VQ_WMat_EC_sse4_1.c
index df4626b60..012f0a38a 100644
--- a/silk/x86/VQ_WMat_EC_sse4_1.c
+++ b/silk/x86/VQ_WMat_EC_sse4_1.c
@@ -36,20 +36,7 @@
#include "celt/x86/x86cpu.h"
/* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */
-void silk_VQ_WMat_EC_sse4_1(
- opus_int8 *ind, /* O index of best codebook vector */
- opus_int32 *res_nrg_Q15, /* O best residual energy */
- opus_int32 *rate_dist_Q8, /* O best total bitrate */
- opus_int *gain_Q7, /* O sum of absolute LTP coefficients */
- const opus_int32 *XX_Q17, /* I correlation matrix */
- const opus_int32 *xX_Q17, /* I correlation vector */
- const opus_int8 *cb_Q7, /* I codebook */
- const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */
- const opus_uint8 *cl_Q5, /* I code length for each codebook vector */
- const opus_int subfr_len, /* I number of samples per subframe */
- const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
- const opus_int L /* I number of vectors in codebook */
-)
+SILK_VQ_WMAT_EC_DECL(sse4_1)
{
opus_int k, gain_tmp_Q7;
const opus_int8 *cb_row_Q7;
diff --git a/silk/x86/main_sse.h b/silk/x86/main_sse.h
index b254d53e7..057eb1eb7 100644
--- a/silk/x86/main_sse.h
+++ b/silk/x86/main_sse.h
@@ -32,107 +32,33 @@
# include "config.h"
# endif
+#include "main_overrides.h"
+
# if defined(OPUS_X86_MAY_HAVE_SSE4_1)
-void silk_VQ_WMat_EC_sse4_1(
- opus_int8 *ind, /* O index of best codebook vector */
- opus_int32 *res_nrg_Q15, /* O best residual energy */
- opus_int32 *rate_dist_Q8, /* O best total bitrate */
- opus_int *gain_Q7, /* O sum of absolute LTP coefficients */
- const opus_int32 *XX_Q17, /* I correlation matrix */
- const opus_int32 *xX_Q17, /* I correlation vector */
- const opus_int8 *cb_Q7, /* I codebook */
- const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */
- const opus_uint8 *cl_Q5, /* I code length for each codebook vector */
- const opus_int subfr_len, /* I number of samples per subframe */
- const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
- const opus_int L /* I number of vectors in codebook */
-);
+SILK_VQ_WMAT_EC_DECL(sse4_1);
-# if defined OPUS_X86_PRESUME_SSE4_1
+# if defined(OPUS_X86_PRESUME_SSE4_1)
-# define OVERRIDE_silk_VQ_WMat_EC
-# define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \
- subfr_len, max_gain_Q7, L, arch) \
- ((void)(arch),silk_VQ_WMat_EC_sse4_1(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \
- subfr_len, max_gain_Q7, L))
+# define silk_VQ_WMat_EC(...) OVERRIDE_IMPL_SINGLE(silk_VQ_WMat_EC, sse4_1, __VA_ARGS__)
# elif defined(OPUS_HAVE_RTCD)
-extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])(
- opus_int8 *ind, /* O index of best codebook vector */
- opus_int32 *res_nrg_Q15, /* O best residual energy */
- opus_int32 *rate_dist_Q8, /* O best total bitrate */
- opus_int *gain_Q7, /* O sum of absolute LTP coefficients */
- const opus_int32 *XX_Q17, /* I correlation matrix */
- const opus_int32 *xX_Q17, /* I correlation vector */
- const opus_int8 *cb_Q7, /* I codebook */
- const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */
- const opus_uint8 *cl_Q5, /* I code length for each codebook vector */
- const opus_int subfr_len, /* I number of samples per subframe */
- const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
- const opus_int L /* I number of vectors in codebook */
-);
-
-# define OVERRIDE_silk_VQ_WMat_EC
-# define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \
- subfr_len, max_gain_Q7, L, arch) \
- ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \
- subfr_len, max_gain_Q7, L))
+extern OVERRIDE_IMPL_ARRAY_DECL(silk_VQ_WMat_EC);
+# define silk_VQ_WMat_EC(...) OVERRIDE_IMPL_ARRAY(silk_VQ_WMat_EC, __VA_ARGS__)
# endif
-void silk_NSQ_sse4_1(
- const silk_encoder_state *psEncC, /* I Encoder State */
- silk_nsq_state *NSQ, /* I/O NSQ state */
- SideInfoIndices *psIndices, /* I/O Quantization Indices */
- const opus_int16 x16[], /* I Input */
- opus_int8 pulses[], /* O Quantized pulse signal */
- const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */
- const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */
- const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */
- const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */
- const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */
- const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */
- const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */
- const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */
- const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */
- const opus_int LTP_scale_Q14 /* I LTP state scaling */
-);
+SILK_NSQ_DECL(sse4_1);
# if defined OPUS_X86_PRESUME_SSE4_1
-# define OVERRIDE_silk_NSQ
-# define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
- HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \
- ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
- HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14))
+# define silk_NSQ(...) OVERRIDE_IMPL_SINGLE(silk_NSQ, sse4_1, __VA_ARGS__)
# elif defined(OPUS_HAVE_RTCD)
-extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])(
- const silk_encoder_state *psEncC, /* I Encoder State */
- silk_nsq_state *NSQ, /* I/O NSQ state */
- SideInfoIndices *psIndices, /* I/O Quantization Indices */
- const opus_int16 x16[], /* I Input */
- opus_int8 pulses[], /* O Quantized pulse signal */
- const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */
- const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */
- const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */
- const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */
- const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */
- const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */
- const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */
- const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */
- const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */
- const opus_int LTP_scale_Q14 /* I LTP state scaling */
-);
-
-# define OVERRIDE_silk_NSQ
-# define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
- HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \
- ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
- HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14))
+extern OVERRIDE_IMPL_ARRAY_DECL(silk_NSQ);
+# define silk_NSQ(...) OVERRIDE_IMPL_ARRAY(silk_NSQ, __VA_ARGS__)
# endif
diff --git a/silk/x86/x86_silk_map.c b/silk/x86/x86_silk_map.c
index 39ad75276..eb4d728d6 100644
--- a/silk/x86/x86_silk_map.c
+++ b/silk/x86/x86_silk_map.c
@@ -69,23 +69,7 @@ opus_int (*const SILK_VAD_GETSA_Q8_IMPL[ OPUS_ARCHMASK + 1 ] )(
MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ) /* avx */
};
-void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )(
- const silk_encoder_state *psEncC, /* I Encoder State */
- silk_nsq_state *NSQ, /* I/O NSQ state */
- SideInfoIndices *psIndices, /* I/O Quantization Indices */
- const opus_int16 x16[], /* I Input */
- opus_int8 pulses[], /* O Quantized pulse signal */
- const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */
- const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */
- const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */
- const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */
- const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */
- const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */
- const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */
- const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */
- const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */
- const opus_int LTP_scale_Q14 /* I LTP state scaling */
-) = {
+OVERRIDE_IMPL_ARRAY_DECL(silk_NSQ) = {
silk_NSQ_c, /* non-sse */
silk_NSQ_c,
silk_NSQ_c,
@@ -93,20 +77,7 @@ void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )(
MAY_HAVE_SSE4_1( silk_NSQ ) /* avx */
};
-void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )(
- opus_int8 *ind, /* O index of best codebook vector */
- opus_int32 *res_nrg_Q15, /* O best residual energy */
- opus_int32 *rate_dist_Q8, /* O best total bitrate */
- opus_int *gain_Q7, /* O sum of absolute LTP coefficients */
- const opus_int32 *XX_Q17, /* I correlation matrix */
- const opus_int32 *xX_Q17, /* I correlation vector */
- const opus_int8 *cb_Q7, /* I codebook */
- const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */
- const opus_uint8 *cl_Q5, /* I code length for each codebook vector */
- const opus_int subfr_len, /* I number of samples per subframe */
- const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
- const opus_int L /* I number of vectors in codebook */
-) = {
+OVERRIDE_IMPL_ARRAY_DECL(silk_VQ_WMat_EC) = {
silk_VQ_WMat_EC_c, /* non-sse */
silk_VQ_WMat_EC_c,
silk_VQ_WMat_EC_c,
diff --git a/silk_headers.mk b/silk_headers.mk
index 2588067c7..7888c1af4 100644
--- a/silk_headers.mk
+++ b/silk_headers.mk
@@ -6,6 +6,7 @@ silk/API.h \
silk/typedef.h \
silk/define.h \
silk/main.h \
+silk/main_overrides.h \
silk/x86/main_sse.h \
silk/PLC.h \
silk/structs.h \