[opus] Update opus to 1.5.2

opus/celt/arm/armcpu.c

@@ -96,7 +96,7 @@ static OPUS_INLINE opus_uint32 opus_cpu_capabilities(void){
 /* Linux based */
 #include <stdio.h>
 
-opus_uint32 opus_cpu_capabilities(void)
+static opus_uint32 opus_cpu_capabilities(void)
 {
   opus_uint32 flags = 0;
   FILE *cpuinfo;
@@ -169,7 +169,7 @@ opus_uint32 opus_cpu_capabilities(void)
 #include <sys/types.h>
 #include <sys/sysctl.h>
 
-opus_uint32 opus_cpu_capabilities(void)
+static opus_uint32 opus_cpu_capabilities(void)
 {
   opus_uint32 flags = 0;
 
@@ -191,6 +191,54 @@ opus_uint32 opus_cpu_capabilities(void)
   return flags;
 }
 
+#elif defined(__FreeBSD__)
+#include <sys/auxv.h>
+
+static opus_uint32 opus_cpu_capabilities(void)
+{
+  long hwcap = 0;
+  opus_uint32 flags = 0;
+
+# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \
+ || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+  /* FreeBSD requires armv6+, which always supports media instructions */
+  flags |= OPUS_CPU_ARM_MEDIA_FLAG;
+# endif
+
+  elf_aux_info(AT_HWCAP, &hwcap, sizeof hwcap);
+
+# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_MEDIA) \
+ || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+#  ifdef HWCAP_EDSP
+  if (hwcap & HWCAP_EDSP)
+    flags |= OPUS_CPU_ARM_EDSP_FLAG;
+#  endif
+
+#  if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+#   ifdef HWCAP_NEON
+  if (hwcap & HWCAP_NEON)
+    flags |= OPUS_CPU_ARM_NEON_FLAG;
+#   elif defined(HWCAP_ASIMD)
+  if (hwcap & HWCAP_ASIMD)
+    flags |= OPUS_CPU_ARM_NEON_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_EDSP_FLAG;
+#   endif
+#  endif
+#  if defined(OPUS_ARM_MAY_HAVE_DOTPROD) && defined(HWCAP_ASIMDDP)
+  if (hwcap & HWCAP_ASIMDDP)
+    flags |= OPUS_CPU_ARM_DOTPROD_FLAG;
+#  endif
+# endif
+
+#if defined(OPUS_ARM_PRESUME_AARCH64_NEON_INTR)
+    flags |= OPUS_CPU_ARM_EDSP_FLAG | OPUS_CPU_ARM_MEDIA_FLAG | OPUS_CPU_ARM_NEON_FLAG;
+# if defined(OPUS_ARM_PRESUME_DOTPROD)
+    flags |= OPUS_CPU_ARM_DOTPROD_FLAG;
+# endif
+#endif
+
+  return (flags);
+}
+
 #else
 /* The feature registers which can tell us what the processor supports are
  * accessible in priveleged modes only, so we can't have a general user-space

opus/celt/x86/x86cpu.c

@@ -41,7 +41,7 @@
   (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \
   (defined(OPUS_X86_MAY_HAVE_AVX2) && !defined(OPUS_X86_PRESUME_AVX2)))
 
-#if defined(_WIN32)
+#if defined(_MSC_VER)
 
 #include <intrin.h>
 static _inline void cpuid(unsigned int CPUInfo[4], unsigned int InfoType)

opus/celt/x86/x86cpu.h

@@ -68,8 +68,22 @@ int opus_select_arch(void);
   Use this to work around those restrictions (which should hopefully all get
    optimized to a single MOVD instruction).
   GCC implemented _mm_loadu_si32() since GCC 11; HOWEVER, there is a bug!
-  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99754 */
-#  if !defined(_MSC_VER) && !OPUS_GNUC_PREREQ(11,3) && !(defined(__clang__) && (__clang_major__ >= 8))
+  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99754
+  LLVM implemented _mm_loadu_si32() since Clang 8.0, however the
+   __clang_major__ version number macro is unreliable, as vendors
+   (specifically, Apple) will use different numbering schemes than upstream.
+  Clang's advice is "use feature detection", but they do not provide feature
+   detection support for specific SIMD functions.
+  We follow the approach from the SIMDe project and instead detect unrelated
+   features that should be available in the version we want (see
+   <https://github.com/simd-everywhere/simde/blob/master/simde/simde-detect-clang.h>).*/
+#  if defined(__clang__)
+#   if __has_warning("-Wextra-semi-stmt") || \
+ __has_builtin(__builtin_rotateleft32)
+#    define OPUS_CLANG_8 (1)
+#   endif
+#  endif
+#  if !defined(_MSC_VER) && !OPUS_GNUC_PREREQ(11,3) && !defined(OPUS_CLANG_8)
 #   include <string.h>
 #   include <emmintrin.h>
 

opus/silk/x86/NSQ_del_dec_avx2.c

@@ -73,7 +73,6 @@ static OPUS_INLINE int verify_assumptions(const silk_encoder_state *psEncC)
 /* Intrinsics not defined on MSVC */
 #ifdef _MSC_VER
 #include <Intsafe.h>
-#define __m128i_u __m128i
 static inline int __builtin_sadd_overflow(opus_int32 a, opus_int32 b, opus_int32* res)
 {
     *res = a+b;
@@ -959,7 +958,7 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_avx2(
     {
         __m256i x = _mm256_cvtepi16_epi64(_mm_loadu_si64(&x16[i]));
         x = _mm256_slli_epi64(_mm256_mul_epi32(x, _mm256_set1_epi32(inv_gain_Q26)), 16);
-        _mm_storeu_si128((__m128i_u*)&x_sc_Q10[i], silk_cvtepi64_epi32_high(x));
+        _mm_storeu_si128((__m128i*)&x_sc_Q10[i], silk_cvtepi64_epi32_high(x));
     }
 
     /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */
@@ -985,8 +984,8 @@ static OPUS_INLINE void silk_nsq_del_dec_scale_states_avx2(
         /* Scale long-term shaping state */
         for (i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx; i+=4)
         {
-            __m128i_u* p = (__m128i_u*)&NSQ->sLTP_shp_Q14[i];
-            *p = silk_mm_smulww_epi32(*p, gain_adj_Q16);
+	    opus_int32 *p = &NSQ->sLTP_shp_Q14[i];
+            _mm_storeu_si128((__m128i*)p, silk_mm_smulww_epi32(_mm_loadu_si128((__m128i*)p), gain_adj_Q16));
         }
 
         /* Scale long-term prediction state */
@@ -1041,13 +1040,13 @@ static OPUS_INLINE void silk_LPC_analysis_filter_avx2(
         /* Allowing wrap around so that two wraps can cancel each other. The rare
            cases where the result wraps around can only be triggered by invalid streams*/
 
-        __m256i in_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i_u*)&in_ptr[-8]));
-        __m256i B_v  = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i_u*)&      B[0]));
+        __m256i in_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)&in_ptr[-8]));
+        __m256i B_v  = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)&      B[0]));
         __m256i sum = _mm256_mullo_epi32(in_v, silk_mm256_reverse_epi32(B_v));
         if (order > 10)
         {
-            in_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i_u*)&in_ptr[-16]));
-            B_v  = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i_u*)&B       [8]));
+            in_v = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)&in_ptr[-16]));
+            B_v  = _mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*)&B       [8]));
             B_v  = silk_mm256_reverse_epi32(B_v);
         }
         else

opus/src/opus_private.h

@@ -214,7 +214,7 @@ int opus_multistream_decode_native(
 
 opus_int32 opus_packet_extensions_parse(const unsigned char *data, opus_int32 len, opus_extension_data *extensions, opus_int32 *nb_extensions);
 
-opus_int32 opus_packet_extensions_generate(unsigned char *data, opus_int32 len, const opus_extension_data  *extensions, int nb_extensions, int pad);
+opus_int32 opus_packet_extensions_generate(unsigned char *data, opus_int32 len, const opus_extension_data  *extensions, opus_int32 nb_extensions, int pad);
 
 opus_int32 opus_packet_extensions_count(const unsigned char *data, opus_int32 len);
 

opus/src/repacketizer.c

@@ -155,7 +155,8 @@ opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int
    /* incorporate any extensions from the repacketizer padding */
    for (i=begin;i<end;i++)
    {
-      int frame_ext_count, j;
+      int j;
+      opus_int32 frame_ext_count;
       frame_ext_count = total_ext_count - ext_count;
       int ret = opus_packet_extensions_parse(rp->paddings[i], rp->padding_len[i],
          &all_extensions[ext_count], &frame_ext_count);