Changeset 285683 in webkit

Timestamp:

Nov 11, 2021, 3:48:09 PM (5 years ago)

Author:

Alan Coon

Message:

Cherry-pick r285509. rdar://problem/83381842

Fix wrong frame count of CARingBuffer in SpeechRecognitionRemoteRealtimeMediaSource
​https://bugs.webkit.org/show_bug.cgi?id=232863
<rdar://83381842>

Reviewed by Youenn Fablet.

SpeechRecognitionRealtimeMediaSourceManager::Source uses shared ring buffer to pass audio data to
SpeechRecognitionRemoteRealtimeMediaSource. We used to ask CARingBuffer in
SpeechRecognitionRealtimeMediaSourceManager::Source to allocate with m_numberOfFrames and send m_numberOfFrames
to SpeechRecognitionRemoteRealtimeMediaSource (so SpeechRecognitionRemoteRealtimeMediaSource can create a
corresponding CARingBuffer). This is wrong because CARingBuffer::allocate() rounds up frameCount to power of
two, which means m_numberOfFrames may be not the number used. We should get the actual frameCount in the
setStorage callback of SharedRingBufferStorage, and pass that value to SpeechRecognitionRemoteRealtimeMediaSource.

Manually tested.

• UIProcess/SpeechRecognitionRemoteRealtimeMediaSource.cpp: (WebKit::SpeechRecognitionRemoteRealtimeMediaSource::setStorage):
• WebProcess/Speech/SpeechRecognitionRealtimeMediaSourceManager.cpp: (WebKit::SpeechRecognitionRealtimeMediaSourceManager::Source::Source): (WebKit::SpeechRecognitionRealtimeMediaSourceManager::Source::storageChanged):

git-svn-id: ​https://svn.webkit.org/repository/webkit/trunk@285509 268f45cc-cd09-0410-ab3c-d52691b4dbfc

Location:

branches/safari-612.3.6.0-branch/Source/WebKit

Files:

• ChangeLog (modified) (1 diff)
• UIProcess/SpeechRecognitionRemoteRealtimeMediaSource.cpp (modified) (1 diff)
• WebProcess/Speech/SpeechRecognitionRealtimeMediaSourceManager.cpp (modified) (5 diffs)

branches/safari-612.3.6.0-branch/Source/WebKit/ChangeLog

@@ +1 @@
+2021-11-11  Kocsen Chung  <kocsen_chung@apple.com>
+
+        Cherry-pick r285509. rdar://problem/83381842
+
+    Fix wrong frame count of CARingBuffer in SpeechRecognitionRemoteRealtimeMediaSource
+    https://bugs.webkit.org/show_bug.cgi?id=232863
+    <rdar://83381842>
+    
+    Reviewed by Youenn Fablet.
+    
+    SpeechRecognitionRealtimeMediaSourceManager::Source uses shared ring buffer to pass audio data to
+    SpeechRecognitionRemoteRealtimeMediaSource. We used to ask CARingBuffer in
+    SpeechRecognitionRealtimeMediaSourceManager::Source to allocate with m_numberOfFrames and send m_numberOfFrames
+    to SpeechRecognitionRemoteRealtimeMediaSource (so SpeechRecognitionRemoteRealtimeMediaSource can create a
+    corresponding CARingBuffer). This is wrong because CARingBuffer::allocate() rounds up frameCount to power of
+    two, which means m_numberOfFrames may be not the number used. We should get the actual frameCount in the
+    setStorage callback of SharedRingBufferStorage, and pass that value to SpeechRecognitionRemoteRealtimeMediaSource.
+    
+    Manually tested.
+    
+    * UIProcess/SpeechRecognitionRemoteRealtimeMediaSource.cpp:
+    (WebKit::SpeechRecognitionRemoteRealtimeMediaSource::setStorage):
+    * WebProcess/Speech/SpeechRecognitionRealtimeMediaSourceManager.cpp:
+    (WebKit::SpeechRecognitionRealtimeMediaSourceManager::Source::Source):
+    (WebKit::SpeechRecognitionRealtimeMediaSourceManager::Source::storageChanged):
+    
+    git-svn-id: https://svn.webkit.org/repository/webkit/trunk@285509 268f45cc-cd09-0410-ab3c-d52691b4dbfc
+
+    2021-11-09  Sihui Liu  <sihui_liu@apple.com>
+
+            Fix wrong frame count of CARingBuffer in SpeechRecognitionRemoteRealtimeMediaSource
+            https://bugs.webkit.org/show_bug.cgi?id=232863
+            <rdar://83381842>
+
+            Reviewed by Youenn Fablet.
+
+            SpeechRecognitionRealtimeMediaSourceManager::Source uses shared ring buffer to pass audio data to
+            SpeechRecognitionRemoteRealtimeMediaSource. We used to ask CARingBuffer in
+            SpeechRecognitionRealtimeMediaSourceManager::Source to allocate with m_numberOfFrames and send m_numberOfFrames
+            to SpeechRecognitionRemoteRealtimeMediaSource (so SpeechRecognitionRemoteRealtimeMediaSource can create a
+            corresponding CARingBuffer). This is wrong because CARingBuffer::allocate() rounds up frameCount to power of
+            two, which means m_numberOfFrames may be not the number used. We should get the actual frameCount in the
+            setStorage callback of SharedRingBufferStorage, and pass that value to SpeechRecognitionRemoteRealtimeMediaSource.
+
+            Manually tested.
+
+            * UIProcess/SpeechRecognitionRemoteRealtimeMediaSource.cpp:
+            (WebKit::SpeechRecognitionRemoteRealtimeMediaSource::setStorage):
+            * WebProcess/Speech/SpeechRecognitionRealtimeMediaSourceManager.cpp:
+            (WebKit::SpeechRecognitionRealtimeMediaSourceManager::Source::Source):
+            (WebKit::SpeechRecognitionRealtimeMediaSourceManager::Source::storageChanged):
+
 2021-11-08  Kocsen Chung  <kocsen_chung@apple.com>
 

branches/safari-612.3.6.0-branch/Source/WebKit/UIProcess/SpeechRecognitionRemoteRealtimeMediaSource.cpp

@@ -78 +78 @@
 void SpeechRecognitionRemoteRealtimeMediaSource::setStorage(const SharedMemory::Handle& handle, const WebCore::CAAudioStreamDescription& description, uint64_t numberOfFrames)
 {
+    if (!numberOfFrames) {
+        m_ringBuffer = nullptr;
+        m_buffer = nullptr;
+        return;
+    }
+
     m_description = description;
-
     m_ringBuffer = WebCore::CARingBuffer::adoptStorage(makeUniqueRef<ReadOnlySharedRingBufferStorage>(handle), description, numberOfFrames).moveToUniquePtr();
     m_buffer = makeUnique<WebCore::WebAudioBufferList>(description);

branches/safari-612.3.6.0-branch/Source/WebKit/WebProcess/Speech/SpeechRecognitionRealtimeMediaSourceManager.cpp

@@ -64 +64 @@
         , m_connection(WTFMove(connection))
 #if PLATFORM(COCOA)
-        , m_ringBuffer(makeUniqueRef<SharedRingBufferStorage>(std::bind(&Source::storageChanged, this, std::placeholders::_1)))
+        , m_ringBuffer(makeUniqueRef<SharedRingBufferStorage>(std::bind(&Source::storageChanged, this, std::placeholders::_1, std::placeholders::_2, std::placeholders::_3)))
 #endif
     {
@@ -114 +114 @@
             ASSERT(description.platformDescription().type == PlatformDescription::CAAudioStreamBasicType);
             m_description = *WTF::get<const AudioStreamBasicDescription*>(description.platformDescription().description);
-
-            m_numberOfFrames = m_description.sampleRate() * 2;
-            m_ringBuffer.allocate(m_description.streamDescription(), m_numberOfFrames);
+            m_ringBuffer.allocate(m_description.streamDescription(), m_description.sampleRate() * 2);
         }
 
@@ -132 +130 @@
 #if PLATFORM(COCOA)
 
-    void storageChanged(SharedMemory* storage)
+    void storageChanged(SharedMemory* storage, const WebCore::CAAudioStreamDescription& description, size_t numberOfFrames)
     {
         DisableMallocRestrictionsForCurrentThreadScope scope;
@@ -143 +141 @@
         uint64_t dataSize = 0;
 #endif
-        m_connection->send(Messages::SpeechRecognitionRemoteRealtimeMediaSourceManager::SetStorage(m_identifier, SharedMemory::IPCHandle { WTFMove(handle),  dataSize }, m_description, m_numberOfFrames), 0);
+        m_connection->send(Messages::SpeechRecognitionRemoteRealtimeMediaSourceManager::SetStorage(m_identifier, SharedMemory::IPCHandle { WTFMove(handle),  dataSize }, description, numberOfFrames), 0);
     }
 
@@ -163 +161 @@
 
 #if PLATFORM(COCOA)
-    uint64_t m_numberOfFrames { 0 };
     CARingBuffer m_ringBuffer;
     CAAudioStreamDescription m_description { };