Finish up XAudio with a round of bugfixes

libaegisub/audio/provider.cpp

@@ -73,7 +73,6 @@ public:
 }
 
 namespace agi {
-	
 void AudioProvider::FillBufferInt16Mono(int16_t* buf, int64_t start, int64_t count) const {
 	if (!float_samples && bytes_per_sample == 2 && channels == 1) {
 		FillBuffer(buf, start, count);
@@ -242,7 +241,7 @@ void SaveAudioClip(AudioProvider const& provider, fs::path const& path, int star
 
 	out.write("WAVEfmt ");
 	out.write<int32_t>(16); // Size of chunk
-	out.write<int16_t>(1);  // compression format (PCM)
+	out.write<int16_t>(provider.AreSamplesFloat() ? 3 : 1);  // compression format (1: WAVE_FORMAT_PCM, 3: WAVE_FORMAT_IEEE_FLOAT)
 	out.write<int16_t>(provider.GetChannels());
 	out.write<int32_t>(provider.GetSampleRate());
 	out.write<int32_t>(provider.GetSampleRate() * provider.GetChannels() * provider.GetBytesPerSample());
@@ -262,4 +261,4 @@ void SaveAudioClip(AudioProvider const& provider, fs::path const& path, int star
 		out.write(buf);
 	}
 }
-}
+}

libaegisub/audio/provider_lock.cpp

@@ -29,6 +29,11 @@ class LockAudioProvider final : public agi::AudioProviderWrapper {
 		source->GetAudio(buf, start, count);
 	}
 
+	void FillBufferInt16Mono(int16_t *buf, int64_t start, int64_t count) const override {
+		std::unique_lock<std::mutex> lock(mutex);
+		source->GetInt16MonoAudio(buf, start, count);
+	}
+
 public:
 	LockAudioProvider(std::unique_ptr<AudioProvider> src)
 	: AudioProviderWrapper(std::move(src))

libaegisub/audio/provider_ram.cpp

@@ -46,14 +46,14 @@ public:
 		decoded_samples = 0;
 
 		try {
-			blockcache.resize((source->GetNumSamples() * source->GetBytesPerSample() + CacheBlockSize - 1) >> CacheBits);
+			blockcache.resize((num_samples * bytes_per_sample * channels + CacheBlockSize - 1) >> CacheBits);
 		}
 		catch (std::bad_alloc const&) {
 			throw AudioProviderError("Not enough memory available to cache in RAM");
 		}
 
 		decoder = std::thread([&] {
-			int64_t readsize = CacheBlockSize / source->GetBytesPerSample();
+			int64_t readsize = CacheBlockSize / bytes_per_sample / channels;
 			for (size_t i = 0; i < blockcache.size(); i++) {
 				if (cancelled) break;
 				auto actual_read = std::min<int64_t>(readsize, num_samples - i * readsize);
@@ -95,4 +95,4 @@ namespace agi {
 std::unique_ptr<AudioProvider> CreateRAMAudioProvider(std::unique_ptr<AudioProvider> src) {
 	return agi::make_unique<RAMAudioProvider>(std::move(src));
 }
-}
+}

src/audio_provider_ffmpegsource.cpp

@@ -126,8 +126,10 @@ void FFmpegSourceAudioProvider::LoadAudio(agi::fs::path const& filename) {
 
 	// reindex if the error handling mode has changed
 	FFMS_IndexErrorHandling ErrorHandling = GetErrorHandlingMode();
+#if FFMS_VERSION >= ((2 << 24) | (17 << 16) | (2 << 8) | 0)
 	if (Index && FFMS_GetErrorHandling(Index) != ErrorHandling)
 		Index = nullptr;
+#endif
 
 	// moment of truth
 	if (!Index) {
@@ -165,22 +167,23 @@ void FFmpegSourceAudioProvider::LoadAudio(agi::fs::path const& filename) {
 			throw agi::AudioProviderError("unknown or unsupported sample format");
 	}
 
+#if FFMS_VERSION >= ((2 << 24) | (17 << 16) | (4 << 8) | 0)
 	if (OPT_GET("Provider/Audio/FFmpegSource/Downmix")->GetBool()) {
-		if (channels > 2 || bytes_per_sample != 2 || float_samples) {
+		if (channels > 1 || bytes_per_sample != 2 || float_samples) {
 			std::unique_ptr<FFMS_ResampleOptions, decltype(&FFMS_DestroyResampleOptions)>
 				opt(FFMS_CreateResampleOptions(AudioSource), FFMS_DestroyResampleOptions);
-			if (channels > 2)
-				opt->ChannelLayout = FFMS_CH_FRONT_LEFT | FFMS_CH_FRONT_RIGHT;
+			opt->ChannelLayout = FFMS_CH_FRONT_CENTER;
 			opt->SampleFormat = FFMS_FMT_S16;
 
 			// Might fail if FFMS2 wasn't built with libavresample
 			if (!FFMS_SetOutputFormatA(AudioSource, opt.get(), nullptr)) {
-				channels = channels > 2 ? 2 : channels;
+				channels = 1;
 				bytes_per_sample = 2;
 				float_samples = false;
 			}
 		}
 	}
+#endif
 }
 
 }
@@ -189,4 +192,4 @@ std::unique_ptr<agi::AudioProvider> CreateFFmpegSourceAudioProvider(agi::fs::pat
 	return agi::make_unique<FFmpegSourceAudioProvider>(file, br);
 }
 
-#endif /* WITH_FFMS2 */
+#endif /* WITH_FFMS2 */

src/audio_renderer_spectrum.cpp

@@ -208,8 +208,8 @@ void AudioSpectrumRenderer::FillBlock(size_t block_index, float *block)
 	assert(block);
 
 	int64_t first_sample = (((int64_t)block_index) << derivation_dist) - ((int64_t)1 << derivation_size);
-	provider->GetAudio(&audio_scratch[0], first_sample, 2 << derivation_size);
-
+	provider->GetInt16MonoAudio(audio_scratch.data(), first_sample, 2 << derivation_size);
+	
 	// Because the FFTs used here are unnormalized DFTs, we have to compensate
 	// the possible length difference between derivation_size used in the
 	// calculations and its user-provided counterpart. Thus, the display is

src/audio_renderer_waveform.cpp

@@ -88,7 +88,7 @@ void AudioWaveformRenderer::Render(wxBitmap &bmp, int start, AudioRenderingStyle
 
 	for (int x = 0; x < rect.width; ++x)
 	{
-		provider->GetAudio(audio_buffer.get(), (int64_t)cur_sample, (int64_t)pixel_samples);
+		provider->GetInt16MonoAudio(reinterpret_cast<int16_t*>(audio_buffer.get()), (int64_t)cur_sample, (int64_t)pixel_samples);
 		cur_sample += pixel_samples;
 
 		int peak_min = 0, peak_max = 0;