git.sesse.net Git - casparcg/blob - core/mixer/audio/audio_util.h

   1 #pragma once\r
   2 \r
   3 #include <vector>\r
   4 \r
   5 #include <stdint.h>\r
   6 \r
   7 #include <tbb/cache_aligned_allocator.h>\r
   8 \r
   9 namespace caspar { namespace core {\r
  10 \r
  11 // NOTE: Input data pointer should be larger than input.size() to allow sse to read beyond\r
  12 template<typename T>\r
  13 static std::vector<int16_t, tbb::cache_aligned_allocator<int16_t>> audio_32_to_16_sse(const T& audio_data)\r
  14 {       \r
  15         auto size                = std::distance(std::begin(audio_data), std::end(audio_data));\r
  16         auto input32     = &(*std::begin(audio_data));\r
  17         auto output16    = std::vector<int16_t, tbb::cache_aligned_allocator<int16_t>>(size);\r
  18 \r
  19         auto input128    = reinterpret_cast<const __m128i*>(input32);\r
  20         auto output128   = reinterpret_cast<__m128i*>(output16.data());\r
  21 \r
  22         for(int n = 0; n < size/8; ++n)         \r
  23         {\r
  24                 auto xmm0 = _mm_srai_epi32(_mm_load_si128(input128++), 16);\r
  25                 auto xmm1 = _mm_srai_epi32(_mm_load_si128(input128++), 16);\r
  26                 auto xmm3 = _mm_packs_epi32(xmm0, xmm1);\r
  27                 _mm_store_si128(output128++, xmm3);\r
  28         }\r
  29 \r
  30         for(int n = size/8; n < size; ++n)\r
  31                 output16[n] = input32[n] >> 16;\r
  32 \r
  33         return output16;\r
  34 }\r
  35 \r
  36 template<typename T>\r
  37 static std::vector<int8_t, tbb::cache_aligned_allocator<int8_t>> audio_32_to_24(const T& audio_data)\r
  38 {       \r
  39         auto size                = std::distance(std::begin(audio_data), std::end(audio_data));\r
  40         auto input8              = reinterpret_cast<const int8_t*>(&(*std::begin(audio_data)));\r
  41         auto output8     = std::vector<int8_t, tbb::cache_aligned_allocator<int8_t>>();\r
  42                         \r
  43         output8.reserve(size*3);\r
  44         for(int n = 0; n < size; ++n)\r
  45         {\r
  46                 output8.push_back(input8[n*4+1]);\r
  47                 output8.push_back(input8[n*4+2]);\r
  48                 output8.push_back(input8[n*4+3]);\r
  49         }\r
  50 \r
  51         return output8;\r
  52 }\r
  53 \r
  54 template<typename T>\r
  55 static std::vector<int8_t, tbb::cache_aligned_allocator<int8_t>> audio_32_to_16(const T& audio_data)\r
  56 {       \r
  57         auto size                = std::distance(std::begin(audio_data), std::end(audio_data));\r
  58         auto input8              = reinterpret_cast<const int8_t*>(&(*std::begin(audio_data)));\r
  59         auto output8     = std::vector<int8_t, tbb::cache_aligned_allocator<int8_t>>();\r
  60                         \r
  61         output8.reserve(size*2);\r
  62         for(int n = 0; n < size; ++n)\r
  63         {\r
  64                 output8.push_back(input8[n*4+2]);\r
  65                 output8.push_back(input8[n*4+3]);\r
  66         }\r
  67 \r
  68         return output8;\r
  69 }\r
  70 \r
  71 }}