shithub: opus

Download patch

ref: df637713aac51cb6c14ebfe07fad00819a99401f
parent: c143b72c4c321b5907cbe839079efd3b2455ac9a
author: Jean-Marc Valin <jmvalin@amazon.com>
date: Mon Nov 27 17:57:32 EST 2023

Fixes for ARMv7/AArch32

1) Enable asm/intrinsics even for floating-point
2) Make sure ARMv8 asimd enables EDSP/MEDIA/Neon
3) Add dotp architecture to rtcd table since AArch *can* have dotp

--- a/celt/arm/armcpu.c
+++ b/celt/arm/armcpu.c
@@ -127,6 +127,9 @@
         p = strstr(buf, " neon");
         if(p != NULL && (p[5] == ' ' || p[5] == '\n'))
           flags |= OPUS_CPU_ARM_NEON_FLAG;
+        p = strstr(buf, " asimd");
+        if(p != NULL && (p[6] == ' ' || p[6] == '\n'))
+          flags |= OPUS_CPU_ARM_NEON_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_EDSP_FLAG;
 #  endif
 #  if defined(OPUS_ARM_MAY_HAVE_DOTPROD)
         p = strstr(buf, " asimddp");
--- a/configure.ac
+++ b/configure.ac
@@ -220,7 +220,7 @@
     case $host_cpu in
       arm*)
         dnl Currently we only have asm for fixed-point
-        AS_IF([test "$enable_float" != "yes"],[
+        #AS_IF([test "$enable_float" != "yes"],[
             cpu_arm=yes
             AC_DEFINE([OPUS_ARM_ASM], [],  [Make use of ARM asm optimization])
             AS_GCC_INLINE_ASSEMBLY(
@@ -366,7 +366,7 @@
                   [*** ARM assembly requires perl -- disabling optimizations])
                 asm_optimization="(missing perl dependency for ARM)"
             ])
-        ])
+        #])
         ;;
     esac
 ],[
--- a/silk/arm/arm_silk_map.c
+++ b/silk/arm/arm_silk_map.c
@@ -49,6 +49,7 @@
       silk_biquad_alt_stride2_c,    /* EDSP */
       silk_biquad_alt_stride2_c,    /* Media */
       silk_biquad_alt_stride2_neon, /* Neon */
+      silk_biquad_alt_stride2_neon, /* dotprod */
 };
 
 opus_int32 (*const SILK_LPC_INVERSE_PRED_GAIN_IMPL[OPUS_ARCHMASK + 1])( /* O   Returns inverse prediction gain in energy domain, Q30        */
@@ -59,6 +60,7 @@
       silk_LPC_inverse_pred_gain_c,    /* EDSP */
       silk_LPC_inverse_pred_gain_c,    /* Media */
       silk_LPC_inverse_pred_gain_neon, /* Neon */
+      silk_LPC_inverse_pred_gain_neon, /* dotprod */
 };
 
 void  (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])(
@@ -82,6 +84,7 @@
       silk_NSQ_del_dec_c,    /* EDSP */
       silk_NSQ_del_dec_c,    /* Media */
       silk_NSQ_del_dec_neon, /* Neon */
+      silk_NSQ_del_dec_neon, /* dotprod */
 };
 
 /*There is no table for silk_noise_shape_quantizer_short_prediction because the
@@ -97,6 +100,7 @@
   silk_NSQ_noise_shape_feedback_loop_c,    /* EDSP */
   silk_NSQ_noise_shape_feedback_loop_c,    /* Media */
   silk_NSQ_noise_shape_feedback_loop_neon, /* NEON */
+  silk_NSQ_noise_shape_feedback_loop_neon, /* dotprod */
 };
 
 # endif
@@ -116,6 +120,7 @@
       silk_warped_autocorrelation_FIX_c,    /* EDSP */
       silk_warped_autocorrelation_FIX_c,    /* Media */
       silk_warped_autocorrelation_FIX_neon, /* Neon */
+      silk_warped_autocorrelation_FIX_neon, /* dotprod */
 };
 
 # endif
--