Description: Provide endianness converters before writing or after reading WAV
Author: Nicholas Guriev <guriev-ns@ya.ru>
Forwarded: https://github.com/desktop-app/tg_owt/pull/46
Last-Update: Thu, 28 Jan 2021 00:41:22 +0300

--- a/src/common_audio/wav_file.cc
+++ b/src/common_audio/wav_file.cc
@@ -10,6 +10,7 @@
 
 #include "common_audio/wav_file.h"
 
+#include <byteswap.h>
 #include <errno.h>
 
 #include <algorithm>
@@ -34,6 +35,38 @@ bool FormatSupported(WavFormat format) {
          format == WavFormat::kWavFormatIeeeFloat;
 }
 
+template <typename T>
+void TranslateEndianness(T* destination, const T* source, size_t length) {
+  static_assert(sizeof(T) == 2 || sizeof(T) == 4 || sizeof(T) == 8,
+                "no converter, use integral types");
+  if (sizeof(T) == 2) {
+    const uint16_t* src = reinterpret_cast<const uint16_t*>(source);
+    uint16_t* dst = reinterpret_cast<uint16_t*>(destination);
+    for (size_t index = 0; index < length; index++) {
+      dst[index] = bswap_16(src[index]);
+    }
+  }
+  if (sizeof(T) == 4) {
+    const uint32_t* src = reinterpret_cast<const uint32_t*>(source);
+    uint32_t* dst = reinterpret_cast<uint32_t*>(destination);
+    for (size_t index = 0; index < length; index++) {
+      dst[index] = bswap_32(src[index]);
+    }
+  }
+  if (sizeof(T) == 8) {
+    const uint64_t* src = reinterpret_cast<const uint64_t*>(source);
+    uint64_t* dst = reinterpret_cast<uint64_t*>(destination);
+    for (size_t index = 0; index < length; index++) {
+      dst[index] = bswap_64(src[index]);
+    }
+  }
+}
+
+template <typename T>
+void TranslateEndianness(T* buffer, size_t length) {
+  TranslateEndianness(buffer, buffer, length);
+}
+
 // Doesn't take ownership of the file handle and won't close it.
 class WavHeaderFileReader : public WavHeaderReader {
  public:
@@ -89,10 +122,6 @@ void WavReader::Reset() {
 
 size_t WavReader::ReadSamples(const size_t num_samples,
                               int16_t* const samples) {
-#ifndef WEBRTC_ARCH_LITTLE_ENDIAN
-#error "Need to convert samples to big-endian when reading from WAV file"
-#endif
-
   size_t num_samples_left_to_read = num_samples;
   size_t next_chunk_start = 0;
   while (num_samples_left_to_read > 0 && num_unread_samples_ > 0) {
@@ -105,6 +134,9 @@ size_t WavReader::ReadSamples(const size
       num_bytes_read = file_.Read(samples_to_convert.data(),
                                   chunk_size * sizeof(samples_to_convert[0]));
       num_samples_read = num_bytes_read / sizeof(samples_to_convert[0]);
+#ifdef WEBRTC_ARCH_BIG_ENDIAN
+      TranslateEndianness(samples_to_convert.data(), num_samples_read);
+#endif
 
       for (size_t j = 0; j < num_samples_read; ++j) {
         samples[next_chunk_start + j] = FloatToS16(samples_to_convert[j]);
@@ -114,6 +146,10 @@ size_t WavReader::ReadSamples(const size
       num_bytes_read = file_.Read(&samples[next_chunk_start],
                                   chunk_size * sizeof(samples[0]));
       num_samples_read = num_bytes_read / sizeof(samples[0]);
+
+#ifdef WEBRTC_ARCH_BIG_ENDIAN
+      TranslateEndianness(&samples[next_chunk_start], num_samples_read);
+#endif
     }
     RTC_CHECK(num_samples_read == 0 || (num_bytes_read % num_samples_read) == 0)
         << "Corrupt file: file ended in the middle of a sample.";
@@ -129,10 +165,6 @@ size_t WavReader::ReadSamples(const size
 }
 
 size_t WavReader::ReadSamples(const size_t num_samples, float* const samples) {
-#ifndef WEBRTC_ARCH_LITTLE_ENDIAN
-#error "Need to convert samples to big-endian when reading from WAV file"
-#endif
-
   size_t num_samples_left_to_read = num_samples;
   size_t next_chunk_start = 0;
   while (num_samples_left_to_read > 0 && num_unread_samples_ > 0) {
@@ -145,6 +177,9 @@ size_t WavReader::ReadSamples(const size
       num_bytes_read = file_.Read(samples_to_convert.data(),
                                   chunk_size * sizeof(samples_to_convert[0]));
       num_samples_read = num_bytes_read / sizeof(samples_to_convert[0]);
+#ifdef WEBRTC_ARCH_BIG_ENDIAN
+      TranslateEndianness(samples_to_convert.data(), num_samples_read);
+#endif
 
       for (size_t j = 0; j < num_samples_read; ++j) {
         samples[next_chunk_start + j] =
@@ -155,6 +190,9 @@ size_t WavReader::ReadSamples(const size
       num_bytes_read = file_.Read(&samples[next_chunk_start],
                                   chunk_size * sizeof(samples[0]));
       num_samples_read = num_bytes_read / sizeof(samples[0]);
+#ifdef WEBRTC_ARCH_BIG_ENDIAN
+      TranslateEndianness(&samples[next_chunk_start], num_samples_read);
+#endif
 
       for (size_t j = 0; j < num_samples_read; ++j) {
         samples[next_chunk_start + j] =
@@ -213,24 +251,32 @@ WavWriter::WavWriter(FileWrapper file,
 }
 
 void WavWriter::WriteSamples(const int16_t* samples, size_t num_samples) {
-#ifndef WEBRTC_ARCH_LITTLE_ENDIAN
-#error "Need to convert samples to little-endian when writing to WAV file"
-#endif
-
   for (size_t i = 0; i < num_samples; i += kMaxChunksize) {
     const size_t num_remaining_samples = num_samples - i;
     const size_t num_samples_to_write =
         std::min(kMaxChunksize, num_remaining_samples);
 
     if (format_ == WavFormat::kWavFormatPcm) {
+#ifndef WEBRTC_ARCH_BIG_ENDIAN
       RTC_CHECK(
           file_.Write(&samples[i], num_samples_to_write * sizeof(samples[0])));
+#else
+      std::array<int16_t, kMaxChunksize> converted_samples;
+      TranslateEndianness(converted_samples.data(), &samples[i],
+                          num_samples_to_write);
+      RTC_CHECK(
+          file_.Write(converted_samples.data(),
+                      num_samples_to_write * sizeof(converted_samples[0])));
+#endif
     } else {
       RTC_CHECK_EQ(format_, WavFormat::kWavFormatIeeeFloat);
       std::array<float, kMaxChunksize> converted_samples;
       for (size_t j = 0; j < num_samples_to_write; ++j) {
         converted_samples[j] = S16ToFloat(samples[i + j]);
       }
+#ifdef WEBRTC_ARCH_BIG_ENDIAN
+      TranslateEndianness(converted_samples.data(), num_samples_to_write);
+#endif
       RTC_CHECK(
           file_.Write(converted_samples.data(),
                       num_samples_to_write * sizeof(converted_samples[0])));
@@ -243,10 +289,6 @@ void WavWriter::WriteSamples(const int16
 }
 
 void WavWriter::WriteSamples(const float* samples, size_t num_samples) {
-#ifndef WEBRTC_ARCH_LITTLE_ENDIAN
-#error "Need to convert samples to little-endian when writing to WAV file"
-#endif
-
   for (size_t i = 0; i < num_samples; i += kMaxChunksize) {
     const size_t num_remaining_samples = num_samples - i;
     const size_t num_samples_to_write =
@@ -257,6 +299,9 @@ void WavWriter::WriteSamples(const float
       for (size_t j = 0; j < num_samples_to_write; ++j) {
         converted_samples[j] = FloatS16ToS16(samples[i + j]);
       }
+#ifdef WEBRTC_ARCH_BIG_ENDIAN
+      TranslateEndianness(converted_samples.data(), num_samples_to_write);
+#endif
       RTC_CHECK(
           file_.Write(converted_samples.data(),
                       num_samples_to_write * sizeof(converted_samples[0])));
@@ -266,6 +311,9 @@ void WavWriter::WriteSamples(const float
       for (size_t j = 0; j < num_samples_to_write; ++j) {
         converted_samples[j] = FloatS16ToFloat(samples[i + j]);
       }
+#ifdef WEBRTC_ARCH_BIG_ENDIAN
+      TranslateEndianness(converted_samples.data(), num_samples_to_write);
+#endif
       RTC_CHECK(
           file_.Write(converted_samples.data(),
                       num_samples_to_write * sizeof(converted_samples[0])));
--- a/src/common_audio/wav_header.cc
+++ b/src/common_audio/wav_header.cc
@@ -14,6 +14,8 @@
 
 #include "common_audio/wav_header.h"
 
+#include <endian.h>
+
 #include <cstring>
 #include <limits>
 #include <string>
@@ -26,10 +28,6 @@
 namespace webrtc {
 namespace {
 
-#ifndef WEBRTC_ARCH_LITTLE_ENDIAN
-#error "Code not working properly for big endian platforms."
-#endif
-
 #pragma pack(2)
 struct ChunkHeader {
   uint32_t ID;
@@ -174,6 +172,8 @@ bool FindWaveChunk(ChunkHeader* chunk_he
     if (readable->Read(chunk_header, sizeof(*chunk_header)) !=
         sizeof(*chunk_header))
       return false;  // EOF.
+    chunk_header->Size = le32toh(chunk_header->Size);
+
     if (ReadFourCC(chunk_header->ID) == sought_chunk_id)
       return true;  // Sought chunk found.
     // Ignore current chunk by skipping its payload.
@@ -187,6 +187,13 @@ bool ReadFmtChunkData(FmtPcmSubchunk* fm
   if (readable->Read(&(fmt_subchunk->AudioFormat), kFmtPcmSubchunkSize) !=
       kFmtPcmSubchunkSize)
     return false;
+  fmt_subchunk->AudioFormat = le16toh(fmt_subchunk->AudioFormat);
+  fmt_subchunk->NumChannels = le16toh(fmt_subchunk->NumChannels);
+  fmt_subchunk->SampleRate = le32toh(fmt_subchunk->SampleRate);
+  fmt_subchunk->ByteRate = le32toh(fmt_subchunk->ByteRate);
+  fmt_subchunk->BlockAlign = le16toh(fmt_subchunk->BlockAlign);
+  fmt_subchunk->BitsPerSample = le16toh(fmt_subchunk->BitsPerSample);
+
   const uint32_t fmt_size = fmt_subchunk->header.Size;
   if (fmt_size != kFmtPcmSubchunkSize) {
     // There is an optional two-byte extension field permitted to be present
@@ -214,19 +221,22 @@ void WritePcmWavHeader(size_t num_channe
   auto header = rtc::MsanUninitialized<WavHeaderPcm>({});
   const size_t bytes_in_payload = bytes_per_sample * num_samples;
 
-  header.riff.header.ID = PackFourCC('R', 'I', 'F', 'F');
-  header.riff.header.Size = RiffChunkSize(bytes_in_payload, *header_size);
-  header.riff.Format = PackFourCC('W', 'A', 'V', 'E');
-  header.fmt.header.ID = PackFourCC('f', 'm', 't', ' ');
-  header.fmt.header.Size = kFmtPcmSubchunkSize;
-  header.fmt.AudioFormat = MapWavFormatToHeaderField(WavFormat::kWavFormatPcm);
-  header.fmt.NumChannels = static_cast<uint16_t>(num_channels);
-  header.fmt.SampleRate = sample_rate;
-  header.fmt.ByteRate = ByteRate(num_channels, sample_rate, bytes_per_sample);
-  header.fmt.BlockAlign = BlockAlign(num_channels, bytes_per_sample);
-  header.fmt.BitsPerSample = static_cast<uint16_t>(8 * bytes_per_sample);
-  header.data.header.ID = PackFourCC('d', 'a', 't', 'a');
-  header.data.header.Size = static_cast<uint32_t>(bytes_in_payload);
+  header.riff.header.ID = htole32(PackFourCC('R', 'I', 'F', 'F'));
+  header.riff.header.Size =
+      htole32(RiffChunkSize(bytes_in_payload, *header_size));
+  header.riff.Format = htole32(PackFourCC('W', 'A', 'V', 'E'));
+  header.fmt.header.ID = htole32(PackFourCC('f', 'm', 't', ' '));
+  header.fmt.header.Size = htole32(kFmtPcmSubchunkSize);
+  header.fmt.AudioFormat =
+      htole16(MapWavFormatToHeaderField(WavFormat::kWavFormatPcm));
+  header.fmt.NumChannels = htole16(num_channels);
+  header.fmt.SampleRate = htole32(sample_rate);
+  header.fmt.ByteRate =
+      htole32(ByteRate(num_channels, sample_rate, bytes_per_sample));
+  header.fmt.BlockAlign = htole16(BlockAlign(num_channels, bytes_per_sample));
+  header.fmt.BitsPerSample = htole16(8 * bytes_per_sample);
+  header.data.header.ID = htole32(PackFourCC('d', 'a', 't', 'a'));
+  header.data.header.Size = htole32(bytes_in_payload);
 
   // Do an extra copy rather than writing everything to buf directly, since buf
   // might not be correctly aligned.
@@ -245,24 +255,26 @@ void WriteIeeeFloatWavHeader(size_t num_
   auto header = rtc::MsanUninitialized<WavHeaderIeeeFloat>({});
   const size_t bytes_in_payload = bytes_per_sample * num_samples;
 
-  header.riff.header.ID = PackFourCC('R', 'I', 'F', 'F');
-  header.riff.header.Size = RiffChunkSize(bytes_in_payload, *header_size);
-  header.riff.Format = PackFourCC('W', 'A', 'V', 'E');
-  header.fmt.header.ID = PackFourCC('f', 'm', 't', ' ');
-  header.fmt.header.Size = kFmtIeeeFloatSubchunkSize;
+  header.riff.header.ID = htole32(PackFourCC('R', 'I', 'F', 'F'));
+  header.riff.header.Size =
+      htole32(RiffChunkSize(bytes_in_payload, *header_size));
+  header.riff.Format = htole32(PackFourCC('W', 'A', 'V', 'E'));
+  header.fmt.header.ID = htole32(PackFourCC('f', 'm', 't', ' '));
+  header.fmt.header.Size = htole32(kFmtIeeeFloatSubchunkSize);
   header.fmt.AudioFormat =
-      MapWavFormatToHeaderField(WavFormat::kWavFormatIeeeFloat);
-  header.fmt.NumChannels = static_cast<uint16_t>(num_channels);
-  header.fmt.SampleRate = sample_rate;
-  header.fmt.ByteRate = ByteRate(num_channels, sample_rate, bytes_per_sample);
-  header.fmt.BlockAlign = BlockAlign(num_channels, bytes_per_sample);
-  header.fmt.BitsPerSample = static_cast<uint16_t>(8 * bytes_per_sample);
-  header.fmt.ExtensionSize = 0;
-  header.fact.header.ID = PackFourCC('f', 'a', 'c', 't');
-  header.fact.header.Size = 4;
-  header.fact.SampleLength = static_cast<uint32_t>(num_channels * num_samples);
-  header.data.header.ID = PackFourCC('d', 'a', 't', 'a');
-  header.data.header.Size = static_cast<uint32_t>(bytes_in_payload);
+      htole16(MapWavFormatToHeaderField(WavFormat::kWavFormatIeeeFloat));
+  header.fmt.NumChannels = htole16(num_channels);
+  header.fmt.SampleRate = htole32(sample_rate);
+  header.fmt.ByteRate =
+      htole32(ByteRate(num_channels, sample_rate, bytes_per_sample));
+  header.fmt.BlockAlign = htole16(BlockAlign(num_channels, bytes_per_sample));
+  header.fmt.BitsPerSample = htole16(8 * bytes_per_sample);
+  header.fmt.ExtensionSize = htole16(0);
+  header.fact.header.ID = htole32(PackFourCC('f', 'a', 'c', 't'));
+  header.fact.header.Size = htole32(4);
+  header.fact.SampleLength = htole32(num_channels * num_samples);
+  header.data.header.ID = htole32(PackFourCC('d', 'a', 't', 'a'));
+  header.data.header.Size = htole32(bytes_in_payload);
 
   // Do an extra copy rather than writing everything to buf directly, since buf
   // might not be correctly aligned.
@@ -391,6 +403,7 @@ bool ReadWavHeader(WavHeaderReader* read
     return false;
   if (ReadFourCC(header.riff.Format) != "WAVE")
     return false;
+  header.riff.header.Size = le32toh(header.riff.header.Size);
 
   // Find "fmt " and "data" chunks. While the official Wave file specification
   // does not put requirements on the chunks order, it is uncommon to find the