Reorganize x86 SSE intrinsics code.

Enable x86 intrinsics when building in floating-point mode.
Support SSE as an arch value.
Use RTCD to conditionally enable existing floating-point Celt SSE code.
Call functions directly (without RTCD) when their architecture can be presumed.
Use SSE4.1 intrinsics optimized code for Silk even in floating-point mode.

3 files changed, 83 insertions, 7 deletions

diff --git a/silk/x86/SigProc_FIX_sse.h b/silk/x86/SigProc_FIX_sse.h
index 9a0e0964..61efa8da 100644
--- a/silk/x86/SigProc_FIX_sse.h
+++ b/silk/x86/SigProc_FIX_sse.h
@@ -45,6 +45,12 @@ void silk_burg_modified_sse4_1(
     int                         arch                /* I    Run-time architecture                                       */
 );
 
+#if defined(OPUS_X86_PRESUME_SSE4_1)
+#define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \
+    ((void)(arch), silk_burg_modified_sse4_1(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch))
+
+#else
+
 extern void (*const SILK_BURG_MODIFIED_IMPL[OPUS_ARCHMASK + 1])(
     opus_int32                  *res_nrg,           /* O    Residual energy                                             */
     opus_int                    *res_nrg_Q,         /* O    Residual energy Q value                                     */
@@ -59,12 +65,22 @@ extern void (*const SILK_BURG_MODIFIED_IMPL[OPUS_ARCHMASK + 1])(
 #  define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \
     ((*SILK_BURG_MODIFIED_IMPL[(arch) & OPUS_ARCHMASK])(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch))
 
+#endif
+
 opus_int64 silk_inner_prod16_aligned_64_sse4_1(
     const opus_int16 *inVec1,
     const opus_int16 *inVec2,
     const opus_int   len
 );
 
+
+#if defined(OPUS_X86_PRESUME_SSE4_1)
+
+#define silk_inner_prod16_aligned_64(inVec1, inVec2, len, arch) \
+    ((void)(arch),silk_inner_prod16_aligned_64_sse4_1(inVec1, inVec2, len))
+
+#else
+
 extern opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[OPUS_ARCHMASK + 1])(
                     const opus_int16 *inVec1,
                     const opus_int16 *inVec2,
@@ -75,3 +91,4 @@ extern opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[OPUS_ARCHMASK + 1])(
 
 #endif
 #endif
+#endif
diff --git a/silk/x86/main_sse.h b/silk/x86/main_sse.h
index f970632c..afd5ec26 100644
--- a/silk/x86/main_sse.h
+++ b/silk/x86/main_sse.h
@@ -50,6 +50,15 @@ void silk_VQ_WMat_EC_sse4_1(
     opus_int                    L                               /* I    number of vectors in codebook               */
 );
 
+#if defined OPUS_X86_PRESUME_SSE4_1
+
+#define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \
+                          mu_Q9, max_gain_Q7, L, arch) \
+    ((void)(arch),silk_VQ_WMat_EC_sse4_1(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \
+                          mu_Q9, max_gain_Q7, L))
+
+#else
+
 extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])(
     opus_int8                   *ind,                           /* O    index of best codebook vector               */
     opus_int32                  *rate_dist_Q14,                 /* O    best weighted quant error + mu * rate       */
@@ -69,6 +78,8 @@ extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])(
     ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \
                           mu_Q9, max_gain_Q7, L))
 
+#endif
+
 #  define OVERRIDE_silk_NSQ
 
 void silk_NSQ_sse4_1(
@@ -89,6 +100,15 @@ void silk_NSQ_sse4_1(
     const opus_int              LTP_scale_Q14                               /* I    LTP state scaling               */
 );
 
+#if defined OPUS_X86_PRESUME_SSE4_1
+
+#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))
+
+#else
+
 extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])(
     const silk_encoder_state    *psEncC,                                    /* I/O  Encoder State                   */
     silk_nsq_state              *NSQ,                                       /* I/O  NSQ state                       */
@@ -112,6 +132,8 @@ extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])(
     ((*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))
 
+#endif
+
 #  define OVERRIDE_silk_NSQ_del_dec
 
 void silk_NSQ_del_dec_sse4_1(
@@ -132,6 +154,15 @@ void silk_NSQ_del_dec_sse4_1(
     const opus_int              LTP_scale_Q14                               /* I    LTP state scaling               */
 );
 
+#if defined OPUS_X86_PRESUME_SSE4_1
+
+#define silk_NSQ_del_dec(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_del_dec_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))
+
+#else
+
 extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])(
     const silk_encoder_state    *psEncC,                                    /* I/O  Encoder State                   */
     silk_nsq_state              *NSQ,                                       /* I/O  NSQ state                       */
@@ -155,6 +186,8 @@ extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])(
     ((*SILK_NSQ_DEL_DEC_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))
 
+#endif
+
 void silk_noise_shape_quantizer(
     silk_nsq_state      *NSQ,                   /* I/O  NSQ state                       */
     opus_int            signalType,             /* I    Signal type                     */
@@ -192,6 +225,11 @@ opus_int silk_VAD_GetSA_Q8_sse4_1(
     const opus_int16   pIn[]
 );
 
+#if defined(OPUS_X86_PRESUME_SSE4_1)
+#define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn))
+
+#else
+
 #  define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \
      ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn))
 
@@ -201,6 +239,8 @@ extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])(
 
 #  define OVERRIDE_silk_warped_LPC_analysis_filter_FIX
 
+#endif
+
 void silk_warped_LPC_analysis_filter_FIX_sse4_1(
           opus_int32            state[],                    /* I/O  State [order + 1]                   */
           opus_int32            res_Q2[],                   /* O    Residual signal [length]            */
@@ -211,6 +251,12 @@ void silk_warped_LPC_analysis_filter_FIX_sse4_1(
     const opus_int              order                       /* I    Filter order (even)                 */
 );
 
+#if defined(OPUS_X86_PRESUME_SSE4_1)
+#define silk_warped_LPC_analysis_filter_FIX(state, res_Q2, coef_Q13, input, lambda_Q16, length, order, arch) \
+    ((void)(arch),silk_warped_LPC_analysis_filter_FIX_c(state, res_Q2, coef_Q13, input, lambda_Q16, length, order))
+
+#else
+
 extern void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[OPUS_ARCHMASK + 1])(
           opus_int32            state[],                    /* I/O  State [order + 1]                   */
           opus_int32            res_Q2[],                   /* O    Residual signal [length]            */
@@ -224,5 +270,7 @@ extern void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[OPUS_ARCHMASK + 1])
 #  define silk_warped_LPC_analysis_filter_FIX(state, res_Q2, coef_Q13, input, lambda_Q16, length, order, arch) \
     ((*SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[(arch) & OPUS_ARCHMASK])(state, res_Q2, coef_Q13, input, lambda_Q16, length, order))
 
+#endif
+
 # endif
 #endif
diff --git a/silk/x86/x86_silk_map.c b/silk/x86/x86_silk_map.c
index 6747d101..ad9fef2a 100644
--- a/silk/x86/x86_silk_map.c
+++ b/silk/x86/x86_silk_map.c
@@ -35,6 +35,10 @@
 #include "pitch.h"
 #include "main.h"
 
+#if !defined(OPUS_X86_PRESUME_SSE4_1)
+
+#if defined(FIXED_POINT)
+
 opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[ OPUS_ARCHMASK + 1 ] )(
     const opus_int16 *inVec1,
     const opus_int16 *inVec2,
@@ -42,18 +46,20 @@ opus_int64 (*const SILK_INNER_PROD16_ALIGNED_64_IMPL[ OPUS_ARCHMASK + 1 ] )(
 ) = {
   silk_inner_prod16_aligned_64_c,                  /* non-sse */
   silk_inner_prod16_aligned_64_c,
+  silk_inner_prod16_aligned_64_c,
   MAY_HAVE_SSE4_1( silk_inner_prod16_aligned_64 ), /* sse4.1 */
-  NULL
 };
 
+#endif
+
 opus_int (*const SILK_VAD_GETSA_Q8_IMPL[ OPUS_ARCHMASK + 1 ] )(
     silk_encoder_state *psEncC,
     const opus_int16   pIn[]
 ) = {
   silk_VAD_GetSA_Q8_c,                  /* non-sse */
   silk_VAD_GetSA_Q8_c,
+  silk_VAD_GetSA_Q8_c,
   MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ), /* sse4.1 */
-  NULL
 };
 
 void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )(
@@ -75,8 +81,8 @@ void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )(
 ) = {
   silk_NSQ_c,                  /* non-sse */
   silk_NSQ_c,
+  silk_NSQ_c,
   MAY_HAVE_SSE4_1( silk_NSQ ), /* sse4.1 */
-  NULL
 };
 
 void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )(
@@ -94,8 +100,8 @@ void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )(
 ) = {
   silk_VQ_WMat_EC_c,                  /* non-sse */
   silk_VQ_WMat_EC_c,
+  silk_VQ_WMat_EC_c,
   MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ), /* sse4.1 */
-  NULL
 };
 
 void (*const SILK_NSQ_DEL_DEC_IMPL[ OPUS_ARCHMASK + 1 ] )(
@@ -117,10 +123,12 @@ void (*const SILK_NSQ_DEL_DEC_IMPL[ OPUS_ARCHMASK + 1 ] )(
 ) = {
   silk_NSQ_del_dec_c,                  /* non-sse */
   silk_NSQ_del_dec_c,
+  silk_NSQ_del_dec_c,
   MAY_HAVE_SSE4_1( silk_NSQ_del_dec ), /* sse4.1 */
-  NULL
 };
 
+#if defined(FIXED_POINT)
+
 void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[ OPUS_ARCHMASK + 1 ] )(
     opus_int32                  state[],                    /* I/O  State [order + 1]                   */
     opus_int32                  res_Q2[],                   /* O    Residual signal [length]            */
@@ -132,8 +140,8 @@ void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[ OPUS_ARCHMASK + 1 ] )(
 ) = {
   silk_warped_LPC_analysis_filter_FIX_c,                  /* non-sse */
   silk_warped_LPC_analysis_filter_FIX_c,
+  silk_warped_LPC_analysis_filter_FIX_c,
   MAY_HAVE_SSE4_1( silk_warped_LPC_analysis_filter_FIX ), /* sse4.1 */
-  NULL
 };
 
 void (*const SILK_BURG_MODIFIED_IMPL[ OPUS_ARCHMASK + 1 ] )(
@@ -149,6 +157,9 @@ void (*const SILK_BURG_MODIFIED_IMPL[ OPUS_ARCHMASK + 1 ] )(
 ) = {
   silk_burg_modified_c,                  /* non-sse */
   silk_burg_modified_c,
+  silk_burg_modified_c,
   MAY_HAVE_SSE4_1( silk_burg_modified ), /* sse4.1 */
-  NULL
 };
+
+#endif
+#endif