diff --git a/browser/app/profile/firefox.js b/browser/app/profile/firefox.js index f0605a5ab0c5..522b708f72aa 100644 --- a/browser/app/profile/firefox.js +++ b/browser/app/profile/firefox.js @@ -1683,8 +1683,6 @@ pref("media.gmp-manager.certs.1.commonName", "aus4.mozilla.org"); pref("media.gmp-manager.certs.2.issuerName", "CN=Thawte SSL CA,O=\"Thawte, Inc.\",C=US"); pref("media.gmp-manager.certs.2.commonName", "aus4.mozilla.org"); -// Delete HTTP cache v2 data of users that didn't opt-in manually -pref("browser.cache.auto_delete_cache_version", 1); // Play with different values of the decay time and get telemetry, // 0 means to randomize (and persist) the experiment value in users' profiles, // -1 means no experiment is run and we use the preferred value for frecency (6h) diff --git a/browser/components/loop/standalone/content/css/webapp.css b/browser/components/loop/standalone/content/css/webapp.css index e58f5d8e6a55..0916d62c55c5 100644 --- a/browser/components/loop/standalone/content/css/webapp.css +++ b/browser/components/loop/standalone/content/css/webapp.css @@ -92,7 +92,7 @@ header { margin: 0 auto; height: 30px; background-size: contain; - background-image: url("../../content/shared/img/mozilla-logo.png"); + background-image: url("../shared/img/mozilla-logo.png"); background-repeat: no-repeat; } @@ -106,7 +106,7 @@ header { width: 100px; height: 100px; margin: 1rem auto; - background-image: url("../../content/shared/img/firefox-logo.png"); + background-image: url("../shared/img/firefox-logo.png"); background-size: cover; background-repeat: no-repeat; } diff --git a/browser/devtools/framework/toolbox.js b/browser/devtools/framework/toolbox.js index 1d3c7deb4635..b4bb85dba1dd 100644 --- a/browser/devtools/framework/toolbox.js +++ b/browser/devtools/framework/toolbox.js @@ -1358,8 +1358,10 @@ Toolbox.prototype = { // Destroying the walker and inspector fronts outstanding.push(this.destroyInspector().then(() => { // Removing buttons - this._pickerButton.removeEventListener("command", this._togglePicker, false); - this._pickerButton = null; + if (this._pickerButton) { + this._pickerButton.removeEventListener("command", this._togglePicker, false); + this._pickerButton = null; + } })); // Remove the host UI diff --git a/browser/devtools/main.js b/browser/devtools/main.js index e26efc1927db..748d5da0d6ff 100644 --- a/browser/devtools/main.js +++ b/browser/devtools/main.js @@ -277,7 +277,9 @@ Tools.jsprofiler = { inMenu: true, isTargetSupported: function (target) { - return !target.isAddon; + // Hide the profiler when debugging devices pre bug 1046394, + // that don't expose profiler actor in content processes. + return !target.isAddon && (!target.isApp || target.form.profilerActor); }, build: function (frame, target) { diff --git a/browser/devtools/profiler/test/browser.ini b/browser/devtools/profiler/test/browser.ini index 31e4f0817a20..4ef4e289a7b4 100644 --- a/browser/devtools/profiler/test/browser.ini +++ b/browser/devtools/profiler/test/browser.ini @@ -93,6 +93,7 @@ skip-if = true # Bug 1047124 [browser_profiler_tree-abstract-01.js] [browser_profiler_tree-abstract-02.js] [browser_profiler_tree-abstract-03.js] +[browser_profiler_tree-abstract-04.js] [browser_profiler_tree-frame-node.js] [browser_profiler_tree-model-01.js] [browser_profiler_tree-model-02.js] diff --git a/browser/devtools/profiler/test/browser_profiler_tree-abstract-04.js b/browser/devtools/profiler/test/browser_profiler_tree-abstract-04.js new file mode 100644 index 000000000000..c94b16185aef --- /dev/null +++ b/browser/devtools/profiler/test/browser_profiler_tree-abstract-04.js @@ -0,0 +1,68 @@ +/* Any copyright is dedicated to the Public Domain. + http://creativecommons.org/publicdomain/zero/1.0/ */ + +/** + * Tests that the treeview expander arrow doesn't react to dblclick events. + */ + +let { AbstractTreeItem } = Cu.import("resource:///modules/devtools/AbstractTreeItem.jsm", {}); +let { Heritage } = Cu.import("resource:///modules/devtools/ViewHelpers.jsm", {}); + +let test = Task.async(function*() { + let container = document.createElement("vbox"); + gBrowser.selectedBrowser.parentNode.appendChild(container); + + // Populate the tree and test the root item... + + let treeRoot = new MyCustomTreeItem(gDataSrc, { parent: null }); + treeRoot.attachTo(container); + + let originalTreeRootExpanded = treeRoot.expanded; + info("Double clicking on the root item arrow and waiting for focus event."); + let receivedFocusEvent = treeRoot.once("focus"); + EventUtils.sendMouseEvent({ type: "dblclick" }, treeRoot.target.querySelector(".arrow")); + + yield receivedFocusEvent; + is(treeRoot.expanded, originalTreeRootExpanded, + "A double click on the arrow was ignored."); + + container.remove(); + finish(); +}); + +function MyCustomTreeItem(dataSrc, properties) { + AbstractTreeItem.call(this, properties); + this.itemDataSrc = dataSrc; +} + +MyCustomTreeItem.prototype = Heritage.extend(AbstractTreeItem.prototype, { + _displaySelf: function(document, arrowNode) { + let node = document.createElement("hbox"); + node.MozMarginStart = (this.level * 10) + "px"; + node.appendChild(arrowNode); + node.appendChild(document.createTextNode(this.itemDataSrc.label)); + return node; + }, + _populateSelf: function(children) { + for (let childDataSrc of this.itemDataSrc.children) { + children.push(new MyCustomTreeItem(childDataSrc, { + parent: this, + level: this.level + 1 + })); + } + } +}); + +let gDataSrc = { + label: "root", + children: [{ + label: "foo", + children: [] + }, { + label: "bar", + children: [{ + label: "baz", + children: [] + }] + }] +}; diff --git a/browser/devtools/profiler/utils/shared.js b/browser/devtools/profiler/utils/shared.js index 597168cc33f8..b2caee61fcbb 100644 --- a/browser/devtools/profiler/utils/shared.js +++ b/browser/devtools/profiler/utils/shared.js @@ -88,6 +88,12 @@ ProfilerConnection.prototype = { if (this._target.chrome) { this._profiler = this._target.form.profilerActor; } + // Or when we are debugging content processes, we already have the tab + // specific one. Use it immediately. + else if (this._target.form && this._target.form.profilerActor) { + this._profiler = this._target.form.profilerActor; + yield this._registerEventNotifications(); + } // Check if we already have a grip to the `listTabs` response object // and, if we do, use it to get to the profiler actor. else if (this._target.root) { diff --git a/browser/devtools/shared/widgets/AbstractTreeItem.jsm b/browser/devtools/shared/widgets/AbstractTreeItem.jsm index 4f3cdfe1664c..da1bf2f185a6 100644 --- a/browser/devtools/shared/widgets/AbstractTreeItem.jsm +++ b/browser/devtools/shared/widgets/AbstractTreeItem.jsm @@ -427,7 +427,12 @@ AbstractTreeItem.prototype = { * Handler for the "dblclick" event on the element displaying this tree item. */ _onDoubleClick: function(e) { - this._onArrowClick(e); + // Ignore dblclick on the arrow as it has already recived and handled two + // click events. + if (!e.target.classList.contains("arrow")) { + this._onArrowClick(e); + } + this.focus(); }, diff --git a/browser/themes/linux/devtools/commandline.css b/browser/themes/linux/devtools/commandline.css index faa8e05db513..06d7512aeefc 100644 --- a/browser/themes/linux/devtools/commandline.css +++ b/browser/themes/linux/devtools/commandline.css @@ -136,7 +136,7 @@ .gcli-menu-more { font-size: 80%; - text-align: right; + text-align: end; -moz-padding-end: 8px; } diff --git a/browser/themes/linux/devtools/webconsole_networkpanel.css b/browser/themes/linux/devtools/webconsole_networkpanel.css index 10d3e49a4150..3803fe3efd6d 100644 --- a/browser/themes/linux/devtools/webconsole_networkpanel.css +++ b/browser/themes/linux/devtools/webconsole_networkpanel.css @@ -50,7 +50,7 @@ h1 .info { font-weight: bold; color: #000; white-space: nowrap; - text-align: right; + text-align: end; vertical-align: top; width: 10%; } diff --git a/browser/themes/osx/devtools/commandline.css b/browser/themes/osx/devtools/commandline.css index faa8e05db513..06d7512aeefc 100644 --- a/browser/themes/osx/devtools/commandline.css +++ b/browser/themes/osx/devtools/commandline.css @@ -136,7 +136,7 @@ .gcli-menu-more { font-size: 80%; - text-align: right; + text-align: end; -moz-padding-end: 8px; } diff --git a/browser/themes/osx/devtools/webconsole_networkpanel.css b/browser/themes/osx/devtools/webconsole_networkpanel.css index 44340b78d213..ec23925f8542 100644 --- a/browser/themes/osx/devtools/webconsole_networkpanel.css +++ b/browser/themes/osx/devtools/webconsole_networkpanel.css @@ -51,7 +51,7 @@ h1 .info { padding-right: 4px; color: #000; white-space: nowrap; - text-align: right; + text-align: end; vertical-align: top; width: 10%; } diff --git a/browser/themes/shared/devtools/app-manager/device.css b/browser/themes/shared/devtools/app-manager/device.css index 4141ef5f9906..55364d1be64e 100644 --- a/browser/themes/shared/devtools/app-manager/device.css +++ b/browser/themes/shared/devtools/app-manager/device.css @@ -190,7 +190,7 @@ button { } .permission-table-header > div:first-child { - text-align: left; + text-align: start; padding-left: 10px; flex-basis: 30%; } @@ -228,7 +228,7 @@ button { } .permission > div:first-child { - text-align: left; + text-align: start; padding: 3px 10px; flex-basis: 30%; font-weight: bold; @@ -277,7 +277,7 @@ button { display: flex; padding: 15px 10px; display: block; - text-align: left; + text-align: start; flex-grow: 1; } diff --git a/browser/themes/shared/devtools/common.css b/browser/themes/shared/devtools/common.css index 05f522fc1cb1..ec747f6bf008 100644 --- a/browser/themes/shared/devtools/common.css +++ b/browser/themes/shared/devtools/common.css @@ -105,7 +105,7 @@ } .devtools-autocomplete-listbox > richlistitem > .autocomplete-count { - text-align: right; + text-align: end; } /* Rest of the dark and light theme */ diff --git a/browser/themes/shared/devtools/ruleview.css b/browser/themes/shared/devtools/ruleview.css index d901e8905150..01f663e2490d 100644 --- a/browser/themes/shared/devtools/ruleview.css +++ b/browser/themes/shared/devtools/ruleview.css @@ -8,7 +8,7 @@ .ruleview-rule-source { -moz-padding-start: 5px; - text-align: right; + text-align: end; float: right; -moz-user-select: none; } diff --git a/browser/themes/windows/devtools/commandline.css b/browser/themes/windows/devtools/commandline.css index faa8e05db513..06d7512aeefc 100644 --- a/browser/themes/windows/devtools/commandline.css +++ b/browser/themes/windows/devtools/commandline.css @@ -136,7 +136,7 @@ .gcli-menu-more { font-size: 80%; - text-align: right; + text-align: end; -moz-padding-end: 8px; } diff --git a/browser/themes/windows/devtools/webconsole_networkpanel.css b/browser/themes/windows/devtools/webconsole_networkpanel.css index d66232fe9add..a7cc36e5167c 100644 --- a/browser/themes/windows/devtools/webconsole_networkpanel.css +++ b/browser/themes/windows/devtools/webconsole_networkpanel.css @@ -51,7 +51,7 @@ h1 .info { padding-right: 4px; color: #000; white-space: nowrap; - text-align: right; + text-align: end; vertical-align: top; width: 10%; } diff --git a/build/automation.py.in b/build/automation.py.in index 1c426441f54b..38bddb52db41 100644 --- a/build/automation.py.in +++ b/build/automation.py.in @@ -819,7 +819,7 @@ class Automation(object): xrePath = None, certPath = None, debuggerInfo = None, symbolsPath = None, timeout = -1, maxTime = None, onLaunch = None, - webapprtChrome = False, screenshotOnFail=False, testPath=None, bisectChunk=None): + detectShutdownLeaks = False, screenshotOnFail=False, testPath=None, bisectChunk=None): """ Run the app, log the duration it took to execute, return the status code. Kills the app if it runs for longer than |maxTime| seconds, or outputs nothing for |timeout| seconds. diff --git a/build/automationutils.py b/build/automationutils.py index 98c0f3e9293a..a6b6a55ee8d8 100644 --- a/build/automationutils.py +++ b/build/automationutils.py @@ -606,6 +606,8 @@ class ShutdownLeaks(object): self._logWindow(line) elif line[2:10] == "DOCSHELL": self._logDocShell(line) + elif line.startswith("TEST-START | Shutdown"): + self.seenShutdown = True elif message['action'] == 'test_start': fileName = message['test'].replace("chrome://mochitests/content/browser/", "") self.currentTest = {"fileName": fileName, "windows": set(), "docShells": set()} @@ -614,10 +616,11 @@ class ShutdownLeaks(object): if self.currentTest and (self.currentTest["windows"] or self.currentTest["docShells"]): self.tests.append(self.currentTest) self.currentTest = None - elif message['action'] == 'suite_end': - self.seenShutdown = True def process(self): + if not self.seenShutdown: + self.logger("TEST-UNEXPECTED-FAIL | ShutdownLeaks | process() called before end of test suite") + for test in self._parseLeakingTests(): for url, count in self._zipLeakedWindows(test["leakedWindows"]): self.logger("TEST-UNEXPECTED-FAIL | %s | leaked %d window(s) until shutdown [url = %s]" % (test["fileName"], count, url)) diff --git a/build/gyp.mozbuild b/build/gyp.mozbuild index 85f5a2d5aac1..5b835a3eebb2 100644 --- a/build/gyp.mozbuild +++ b/build/gyp.mozbuild @@ -51,7 +51,7 @@ gyp_vars = { # codec enable/disables: 'include_g711': 1, 'include_opus': 1, - 'include_g722': 0, + 'include_g722': 1, 'include_ilbc': 0, 'include_isac': 0, 'include_pcm16b': 1, diff --git a/content/html/content/test/test_iframe_sandbox_general.html b/content/html/content/test/test_iframe_sandbox_general.html index 3552266396ab..203ae6de5f08 100644 --- a/content/html/content/test/test_iframe_sandbox_general.html +++ b/content/html/content/test/test_iframe_sandbox_general.html @@ -16,6 +16,7 @@ Implement HTML5 sandbox attribute for IFRAMEs - general tests SimpleTest.expectAssertions(0, 1); SimpleTest.waitForExplicitFinish(); +SimpleTest.requestCompleteLog(); // a postMessage handler that is used by sandboxed iframes without // 'allow-same-origin' to communicate pass/fail back to this main page. diff --git a/content/media/AudioCompactor.h b/content/media/AudioCompactor.h index 86de490835e4..d695eb6fc43d 100644 --- a/content/media/AudioCompactor.h +++ b/content/media/AudioCompactor.h @@ -58,7 +58,8 @@ public: duration.value(), framesCopied, buffer.forget(), - aChannels)); + aChannels, + aSampleRate)); // Remove the frames we just pushed into the queue and loop if there is // more to be done. diff --git a/content/media/MediaData.h b/content/media/MediaData.h index 6ff32ded570b..cd03a667b2fb 100644 --- a/content/media/MediaData.h +++ b/content/media/MediaData.h @@ -71,10 +71,12 @@ public: int64_t aDuration, uint32_t aFrames, AudioDataValue* aData, - uint32_t aChannels) + uint32_t aChannels, + uint32_t aRate) : MediaData(AUDIO_SAMPLES, aOffset, aTime, aDuration) , mFrames(aFrames) , mChannels(aChannels) + , mRate(aRate) , mAudioData(aData) { MOZ_COUNT_CTOR(AudioData); @@ -92,6 +94,7 @@ public: const uint32_t mFrames; const uint32_t mChannels; + const uint32_t mRate; // At least one of mAudioBuffer/mAudioData must be non-null. // mChannels channels, each with mFrames frames nsRefPtr mAudioBuffer; diff --git a/content/media/MediaDecoderStateMachine.cpp b/content/media/MediaDecoderStateMachine.cpp index c74b3bdd7012..c52b8a87fd80 100644 --- a/content/media/MediaDecoderStateMachine.cpp +++ b/content/media/MediaDecoderStateMachine.cpp @@ -2802,7 +2802,8 @@ MediaDecoderStateMachine::DropAudioUpToSeekTarget(AudioData* aSample) duration.value(), frames, audioData.forget(), - channels)); + channels, + audio->mRate)); AudioQueue().PushFront(data.forget()); return NS_OK; diff --git a/content/media/apple/AppleMP3Reader.cpp b/content/media/apple/AppleMP3Reader.cpp index a713b9155b0b..73401d73e5d1 100644 --- a/content/media/apple/AppleMP3Reader.cpp +++ b/content/media/apple/AppleMP3Reader.cpp @@ -260,7 +260,7 @@ AppleMP3Reader::AudioSampleCallback(UInt32 aNumBytes, AudioData *audio = new AudioData(mDecoder->GetResource()->Tell(), time, duration, numFrames, reinterpret_cast(decoded.forget()), - mAudioChannels); + mAudioChannels, mAudioSampleRate); mAudioQueue.Push(audio); mCurrentAudioFrame += numFrames; diff --git a/content/media/encoder/EncodedFrameContainer.h b/content/media/encoder/EncodedFrameContainer.h index 55f46f3ec81f..4aff28bacec3 100644 --- a/content/media/encoder/EncodedFrameContainer.h +++ b/content/media/encoder/EncodedFrameContainer.h @@ -62,10 +62,9 @@ public: AMR_AUDIO_FRAME, UNKNOWN // FrameType not set }; - nsresult SwapInFrameData(nsTArray& aData) + void SwapInFrameData(nsTArray& aData) { mFrameData.SwapElements(aData); - return NS_OK; } nsresult SwapOutFrameData(nsTArray& aData) { diff --git a/content/media/encoder/OmxTrackEncoder.cpp b/content/media/encoder/OmxTrackEncoder.cpp index 5cc06205582c..008b02756a33 100644 --- a/content/media/encoder/OmxTrackEncoder.cpp +++ b/content/media/encoder/OmxTrackEncoder.cpp @@ -143,8 +143,7 @@ OmxVideoTrackEncoder::GetEncodedTrack(EncodedFrameContainer& aData) videoData->SetFrameType((outFlags & OMXCodecWrapper::BUFFER_SYNC_FRAME) ? EncodedFrame::AVC_I_FRAME : EncodedFrame::AVC_P_FRAME); } - rv = videoData->SwapInFrameData(buffer); - NS_ENSURE_SUCCESS(rv, rv); + videoData->SwapInFrameData(buffer); videoData->SetTimeStamp(outTimeStampUs); aData.AppendEncodedFrame(videoData); } @@ -187,8 +186,7 @@ OmxAudioTrackEncoder::AppendEncodedFrames(EncodedFrameContainer& aContainer) MOZ_ASSERT(false, "audio codec not supported"); } audiodata->SetTimeStamp(outTimeUs); - rv = audiodata->SwapInFrameData(frameData); - NS_ENSURE_SUCCESS(rv, rv); + audiodata->SwapInFrameData(frameData); aContainer.AppendEncodedFrame(audiodata); } diff --git a/content/media/encoder/VP8TrackEncoder.cpp b/content/media/encoder/VP8TrackEncoder.cpp index b51a8934b9b7..c36ae8bc08a0 100644 --- a/content/media/encoder/VP8TrackEncoder.cpp +++ b/content/media/encoder/VP8TrackEncoder.cpp @@ -173,7 +173,6 @@ VP8TrackEncoder::GetEncodedPartitions(EncodedFrameContainer& aData) vpx_codec_iter_t iter = nullptr; EncodedFrame::FrameType frameType = EncodedFrame::VP8_P_FRAME; nsTArray frameData; - nsresult rv; const vpx_codec_cx_pkt_t *pkt = nullptr; while ((pkt = vpx_codec_get_cx_data(mVPXContext, &iter)) != nullptr) { switch (pkt->kind) { @@ -212,8 +211,7 @@ VP8TrackEncoder::GetEncodedPartitions(EncodedFrameContainer& aData) videoData->SetDuration( (uint64_t)FramesToUsecs(pkt->data.frame.duration, mTrackRate).value()); } - rv = videoData->SwapInFrameData(frameData); - NS_ENSURE_SUCCESS(rv, rv); + videoData->SwapInFrameData(frameData); VP8LOG("GetEncodedPartitions TimeStamp %lld Duration %lld\n", videoData->GetTimeStamp(), videoData->GetDuration()); VP8LOG("frameType %d\n", videoData->GetFrameType()); diff --git a/content/media/fmp4/BlankDecoderModule.cpp b/content/media/fmp4/BlankDecoderModule.cpp index 74efce486aa0..637d8fdf2673 100644 --- a/content/media/fmp4/BlankDecoderModule.cpp +++ b/content/media/fmp4/BlankDecoderModule.cpp @@ -193,7 +193,8 @@ public: aDuration, uint32_t(frames.value()), samples, - mChannelCount); + mChannelCount, + mSampleRate); } private: diff --git a/content/media/fmp4/MP4Reader.cpp b/content/media/fmp4/MP4Reader.cpp index f60a9983f433..5f2f536ffa45 100644 --- a/content/media/fmp4/MP4Reader.cpp +++ b/content/media/fmp4/MP4Reader.cpp @@ -381,6 +381,10 @@ MP4Reader::ReadMetadata(MediaInfo* aInfo, NS_ENSURE_TRUE(mAudio.mDecoder != nullptr, NS_ERROR_FAILURE); nsresult rv = mAudio.mDecoder->Init(); NS_ENSURE_SUCCESS(rv, rv); + + // Decode one audio frame to detect potentially incorrect channels count or + // sampling rate from demuxer. + Decode(kAudio); } if (HasVideo()) { @@ -585,7 +589,15 @@ MP4Reader::Output(TrackType aTrack, MediaData* aSample) switch (aTrack) { case kAudio: { MOZ_ASSERT(aSample->mType == MediaData::AUDIO_SAMPLES); - AudioQueue().Push(static_cast(aSample)); + AudioData* audioData = static_cast(aSample); + AudioQueue().Push(audioData); + if (audioData->mChannels != mInfo.mAudio.mChannels || + audioData->mRate != mInfo.mAudio.mRate) { + LOG("MP4Reader::Output change of sampling rate:%d->%d", + mInfo.mAudio.mRate, audioData->mRate); + mInfo.mAudio.mRate = audioData->mRate; + mInfo.mAudio.mChannels = audioData->mChannels; + } break; } case kVideo: { diff --git a/content/media/fmp4/PlatformDecoderModule.cpp b/content/media/fmp4/PlatformDecoderModule.cpp index 2f594bfcf48d..c19b03bceedc 100644 --- a/content/media/fmp4/PlatformDecoderModule.cpp +++ b/content/media/fmp4/PlatformDecoderModule.cpp @@ -133,7 +133,10 @@ PlatformDecoderModule::Create() #endif #ifdef MOZ_FFMPEG if (sFFmpegDecoderEnabled) { - return FFmpegRuntimeLinker::CreateDecoderModule(); + nsAutoPtr m(FFmpegRuntimeLinker::CreateDecoderModule()); + if (m) { + return m.forget(); + } } #endif #ifdef MOZ_APPLEMEDIA diff --git a/content/media/fmp4/apple/AppleATDecoder.cpp b/content/media/fmp4/apple/AppleATDecoder.cpp index 4642b189cf24..f96f102988e5 100644 --- a/content/media/fmp4/apple/AppleATDecoder.cpp +++ b/content/media/fmp4/apple/AppleATDecoder.cpp @@ -240,7 +240,7 @@ AppleATDecoder::SampleCallback(uint32_t aNumBytes, AudioBufferList decBuffer; decBuffer.mNumberBuffers = 1; - decBuffer.mBuffers[0].mNumberChannels = mConfig.channel_count; + decBuffer.mBuffers[0].mNumberChannels = mOutputFormat.mChannelsPerFrame; decBuffer.mBuffers[0].mDataByteSize = decodedSize; decBuffer.mBuffers[0].mData = decoded.get(); @@ -271,7 +271,9 @@ AppleATDecoder::SampleCallback(uint32_t aNumBytes, break; } - const int rate = mConfig.samples_per_second; + const int rate = mOutputFormat.mSampleRate; + const int channels = mOutputFormat.mChannelsPerFrame; + int64_t time = FramesToUsecs(mCurrentAudioFrame, rate).value(); int64_t duration = FramesToUsecs(numFrames, rate).value(); @@ -281,7 +283,7 @@ AppleATDecoder::SampleCallback(uint32_t aNumBytes, AudioData *audio = new AudioData(mSamplePosition, time, duration, numFrames, reinterpret_cast(decoded.forget()), - rate); + channels, rate); mCallback->Output(audio); mHaveOutput = true; @@ -299,30 +301,30 @@ AppleATDecoder::SampleCallback(uint32_t aNumBytes, void AppleATDecoder::SetupDecoder() { - AudioStreamBasicDescription inputFormat, outputFormat; + AudioStreamBasicDescription inputFormat; // Fill in the input format description from the stream. AppleUtils::GetProperty(mStream, kAudioFileStreamProperty_DataFormat, &inputFormat); // Fill in the output format manually. - PodZero(&outputFormat); - outputFormat.mFormatID = kAudioFormatLinearPCM; - outputFormat.mSampleRate = inputFormat.mSampleRate; - outputFormat.mChannelsPerFrame = inputFormat.mChannelsPerFrame; + PodZero(&mOutputFormat); + mOutputFormat.mFormatID = kAudioFormatLinearPCM; + mOutputFormat.mSampleRate = inputFormat.mSampleRate; + mOutputFormat.mChannelsPerFrame = inputFormat.mChannelsPerFrame; #if defined(MOZ_SAMPLE_TYPE_FLOAT32) - outputFormat.mBitsPerChannel = 32; - outputFormat.mFormatFlags = + mOutputFormat.mBitsPerChannel = 32; + mOutputFormat.mFormatFlags = kLinearPCMFormatFlagIsFloat | 0; #else # error Unknown audio sample type #endif // Set up the decoder so it gives us one sample per frame - outputFormat.mFramesPerPacket = 1; - outputFormat.mBytesPerPacket = outputFormat.mBytesPerFrame - = outputFormat.mChannelsPerFrame * outputFormat.mBitsPerChannel / 8; + mOutputFormat.mFramesPerPacket = 1; + mOutputFormat.mBytesPerPacket = mOutputFormat.mBytesPerFrame + = mOutputFormat.mChannelsPerFrame * mOutputFormat.mBitsPerChannel / 8; - OSStatus rv = AudioConverterNew(&inputFormat, &outputFormat, &mConverter); + OSStatus rv = AudioConverterNew(&inputFormat, &mOutputFormat, &mConverter); if (rv) { LOG("Error %d constructing AudioConverter", rv); mConverter = nullptr; diff --git a/content/media/fmp4/apple/AppleATDecoder.h b/content/media/fmp4/apple/AppleATDecoder.h index fba3e38892a3..b236a1974d96 100644 --- a/content/media/fmp4/apple/AppleATDecoder.h +++ b/content/media/fmp4/apple/AppleATDecoder.h @@ -52,6 +52,7 @@ private: uint64_t mCurrentAudioFrame; int64_t mSamplePosition; bool mHaveOutput; + AudioStreamBasicDescription mOutputFormat; void SetupDecoder(); void SubmitSample(nsAutoPtr aSample); diff --git a/content/media/fmp4/eme/EMEAACDecoder.cpp b/content/media/fmp4/eme/EMEAACDecoder.cpp index db3efbbbcf21..fa32b5e1395a 100644 --- a/content/media/fmp4/eme/EMEAACDecoder.cpp +++ b/content/media/fmp4/eme/EMEAACDecoder.cpp @@ -180,7 +180,8 @@ EMEAACDecoder::Decoded(const nsTArray& aPCM, duration.value(), numFrames, audioData.forget(), - aChannels)); + aChannels, + aRate)); #ifdef LOG_SAMPLE_DECODE LOG("Decoded audio sample! timestamp=%lld duration=%lld currentLength=%u", diff --git a/content/media/fmp4/ffmpeg/FFmpegAACDecoder.cpp b/content/media/fmp4/ffmpeg/FFmpegAACDecoder.cpp index 8d7cbede7763..4c224b5583e2 100644 --- a/content/media/fmp4/ffmpeg/FFmpegAACDecoder.cpp +++ b/content/media/fmp4/ffmpeg/FFmpegAACDecoder.cpp @@ -36,9 +36,6 @@ FFmpegAACDecoder::Init() static AudioDataValue* CopyAndPackAudio(AVFrame* aFrame, uint32_t aNumChannels, uint32_t aNumSamples) { - // These are the only two valid AAC packet sizes. - NS_ASSERTION(aNumSamples == 960 || aNumSamples == 1024, - "Should have exactly one AAC audio packet."); MOZ_ASSERT(aNumChannels <= MAX_CHANNELS); nsAutoArrayPtr audio( @@ -93,13 +90,14 @@ FFmpegAACDecoder::DecodePacket(MP4Sample* aSample) "Only one audio packet should be received at a time."); uint32_t numChannels = mCodecContext->channels; + uint32_t samplingRate = mCodecContext->sample_rate; nsAutoArrayPtr audio( CopyAndPackAudio(mFrame, numChannels, mFrame->nb_samples)); nsAutoPtr data( new AudioData(packet.pos, aSample->composition_timestamp, aSample->duration, - mFrame->nb_samples, audio.forget(), numChannels)); + mFrame->nb_samples, audio.forget(), numChannels, samplingRate)); mCallback->Output(data.forget()); diff --git a/content/media/fmp4/ffmpeg/FFmpegRuntimeLinker.cpp b/content/media/fmp4/ffmpeg/FFmpegRuntimeLinker.cpp index fe322268c056..357bdd1cd43e 100644 --- a/content/media/fmp4/ffmpeg/FFmpegRuntimeLinker.cpp +++ b/content/media/fmp4/ffmpeg/FFmpegRuntimeLinker.cpp @@ -99,6 +99,9 @@ FFmpegRuntimeLinker::Bind(const char* aLibName, uint32_t Version) /* static */ PlatformDecoderModule* FFmpegRuntimeLinker::CreateDecoderModule() { + if (!Link()) { + return nullptr; + } PlatformDecoderModule* module = sLib->Factory(); return module; } diff --git a/content/media/fmp4/gonk/GonkAudioDecoderManager.cpp b/content/media/fmp4/gonk/GonkAudioDecoderManager.cpp index 5365393c1d9f..ec6f816962ce 100644 --- a/content/media/fmp4/gonk/GonkAudioDecoderManager.cpp +++ b/content/media/fmp4/gonk/GonkAudioDecoderManager.cpp @@ -115,8 +115,13 @@ GonkAudioDecoderManager::CreateAudioData(int64_t aStreamOffset, AudioData **v) { if (!duration.isValid()) { return NS_ERROR_UNEXPECTED; } - *v = new AudioData(aStreamOffset, timeUs, duration.value(), frames, buffer.forget(), - mAudioChannels); + *v = new AudioData(aStreamOffset, + timeUs, + duration.value(), + frames, + buffer.forget(), + mAudioChannels, + mAudioRate); ReleaseAudioBuffer(); return NS_OK; } diff --git a/content/media/fmp4/wmf/WMFAudioMFTManager.cpp b/content/media/fmp4/wmf/WMFAudioMFTManager.cpp index 3a9dbdca196d..6412845515ec 100644 --- a/content/media/fmp4/wmf/WMFAudioMFTManager.cpp +++ b/content/media/fmp4/wmf/WMFAudioMFTManager.cpp @@ -263,7 +263,8 @@ WMFAudioMFTManager::Output(int64_t aStreamOffset, duration, numFrames, audioData.forget(), - mAudioChannels); + mAudioChannels, + mAudioRate); #ifdef LOG_SAMPLE_DECODE LOG("Decoded audio sample! timestamp=%lld duration=%lld currentLength=%u", diff --git a/content/media/ogg/OggReader.cpp b/content/media/ogg/OggReader.cpp index 89f8ca9faa94..ccc65e08f8f0 100644 --- a/content/media/ogg/OggReader.cpp +++ b/content/media/ogg/OggReader.cpp @@ -434,7 +434,8 @@ nsresult OggReader::DecodeVorbis(ogg_packet* aPacket) { duration, frames, buffer.forget(), - channels)); + channels, + mVorbisState->mInfo.rate)); mDecodedAudioFrames += frames; @@ -550,7 +551,8 @@ nsresult OggReader::DecodeOpus(ogg_packet* aPacket) { endTime - startTime, frames, buffer.forget(), - channels)); + channels, + mOpusState->mRate)); mDecodedAudioFrames += frames; diff --git a/content/media/wave/WaveReader.cpp b/content/media/wave/WaveReader.cpp index 89feb2a527a7..a68dd80456d6 100644 --- a/content/media/wave/WaveReader.cpp +++ b/content/media/wave/WaveReader.cpp @@ -243,7 +243,8 @@ bool WaveReader::DecodeAudioData() static_cast(readSizeTime * USECS_PER_S), static_cast(frames), sampleBuffer.forget(), - mChannels)); + mChannels, + mSampleRate)); return true; } diff --git a/content/media/webm/WebMReader.cpp b/content/media/webm/WebMReader.cpp index 30071c134f8c..589b31ab2d3d 100644 --- a/content/media/webm/WebMReader.cpp +++ b/content/media/webm/WebMReader.cpp @@ -613,11 +613,12 @@ bool WebMReader::DecodeAudioPacket(nestegg_packet* aPacket, int64_t aOffset) total_frames += frames; AudioQueue().Push(new AudioData(aOffset, - time.value(), - duration.value(), - frames, - buffer.forget(), - mChannels)); + time.value(), + duration.value(), + frames, + buffer.forget(), + mChannels, + rate)); mAudioFrames += frames; if (vorbis_synthesis_read(&mVorbisDsp, frames) != 0) { return false; @@ -738,11 +739,12 @@ bool WebMReader::DecodeAudioPacket(nestegg_packet* aPacket, int64_t aOffset) return false; }; AudioQueue().Push(new AudioData(mDecoder->GetResource()->Tell(), - time.value(), - duration.value(), - frames, - buffer.forget(), - mChannels)); + time.value(), + duration.value(), + frames, + buffer.forget(), + mChannels, + rate)); mAudioFrames += frames; #else diff --git a/content/media/wmf/WMFReader.cpp b/content/media/wmf/WMFReader.cpp index 2b7d5e95f339..1958a7a86a31 100644 --- a/content/media/wmf/WMFReader.cpp +++ b/content/media/wmf/WMFReader.cpp @@ -650,7 +650,8 @@ WMFReader::DecodeAudioData() duration, numFrames, pcmSamples.forget(), - mAudioChannels)); + mAudioChannels, + mAudioRate)); #ifdef LOG_SAMPLE_DECODE DECODER_LOG("Decoded audio sample! timestamp=%lld duration=%lld currentLength=%u", diff --git a/dom/base/URL.cpp b/dom/base/URL.cpp index 7a73c265e7e4..354ad29412f5 100644 --- a/dom/base/URL.cpp +++ b/dom/base/URL.cpp @@ -203,6 +203,16 @@ URL::RevokeObjectURL(const GlobalObject& aGlobal, const nsAString& aURL) } } +nsIPrincipal* +URL::GetPrincipalFromURL(const GlobalObject& aGlobal, const nsAString& aURL, + ErrorResult& aRv) +{ + MOZ_ASSERT(nsContentUtils::IsCallerChrome()); + + NS_LossyConvertUTF16toASCII asciiurl(aURL); + return nsHostObjectProtocolHandler::GetDataEntryPrincipal(asciiurl); +} + void URL::GetHref(nsString& aHref, ErrorResult& aRv) const { diff --git a/dom/base/URL.h b/dom/base/URL.h index 23cfa85e0d8b..6f5315976865 100644 --- a/dom/base/URL.h +++ b/dom/base/URL.h @@ -12,6 +12,7 @@ #include "nsString.h" class nsIDOMBlob; +class nsIPrincipal; class nsISupports; class nsIURI; @@ -68,6 +69,9 @@ public: ErrorResult& aError); static void RevokeObjectURL(const GlobalObject& aGlobal, const nsAString& aURL); + static nsIPrincipal* GetPrincipalFromURL(const GlobalObject& aGlobal, + const nsAString& aURL, + ErrorResult& aError); void GetHref(nsString& aHref, ErrorResult& aRv) const; diff --git a/dom/base/nsGlobalWindow.cpp b/dom/base/nsGlobalWindow.cpp index 44e5e3e9432f..d2f63f8cef5a 100644 --- a/dom/base/nsGlobalWindow.cpp +++ b/dom/base/nsGlobalWindow.cpp @@ -2676,9 +2676,15 @@ nsGlobalWindow::SetNewDocument(nsIDocument* aDocument, if (!aState) { if (reUseInnerWindow) { + if (newInnerWindow->mDoc != aDocument) { newInnerWindow->mDoc = aDocument; + // The storage objects contain the URL of the window. We have to + // recreate them when the innerWindow is reused. + newInnerWindow->mLocalStorage = nullptr; + newInnerWindow->mSessionStorage = nullptr; + if (newInnerWindow->IsDOMBinding()) { WindowBinding::ClearCachedDocumentValue(cx, newInnerWindow); } else { @@ -11460,11 +11466,15 @@ nsGlobalWindow::Observe(nsISupports* aSubject, const char* aTopic, // Clone the storage event included in the observer notification. We want // to dispatch clones rather than the original event. + ErrorResult error; nsRefPtr newEvent = CloneStorageEvent(fireMozStorageChanged ? NS_LITERAL_STRING("MozStorageChanged") : NS_LITERAL_STRING("storage"), - event); + event, error); + if (error.Failed()) { + return error.ErrorCode(); + } newEvent->SetTrusted(true); @@ -11560,7 +11570,8 @@ nsGlobalWindow::Observe(nsISupports* aSubject, const char* aTopic, already_AddRefed nsGlobalWindow::CloneStorageEvent(const nsAString& aType, - const nsRefPtr& aEvent) + const nsRefPtr& aEvent, + ErrorResult& aRv) { MOZ_ASSERT(IsInnerWindow()); @@ -11572,7 +11583,26 @@ nsGlobalWindow::CloneStorageEvent(const nsAString& aType, aEvent->GetOldValue(dict.mOldValue); aEvent->GetNewValue(dict.mNewValue); aEvent->GetUrl(dict.mUrl); - dict.mStorageArea = aEvent->GetStorageArea(); + + nsRefPtr storageArea = aEvent->GetStorageArea(); + MOZ_ASSERT(storageArea); + + nsRefPtr storage; + if (storageArea->GetType() == DOMStorage::LocalStorage) { + storage = GetLocalStorage(aRv); + } else { + MOZ_ASSERT(storageArea->GetType() == DOMStorage::SessionStorage); + storage = GetSessionStorage(aRv); + } + + if (aRv.Failed() || !storage) { + return nullptr; + } + + MOZ_ASSERT(storage); + MOZ_ASSERT(storage->IsForkOf(storageArea)); + + dict.mStorageArea = storage; nsRefPtr event = StorageEvent::Constructor(this, aType, dict); return event.forget(); diff --git a/dom/base/nsGlobalWindow.h b/dom/base/nsGlobalWindow.h index d187f63d75b7..850564da70de 100644 --- a/dom/base/nsGlobalWindow.h +++ b/dom/base/nsGlobalWindow.h @@ -1386,7 +1386,8 @@ protected: // Inner windows only. already_AddRefed CloneStorageEvent(const nsAString& aType, - const nsRefPtr& aEvent); + const nsRefPtr& aEvent, + mozilla::ErrorResult& aRv); // Outer windows only. nsDOMWindowList* GetWindowList(); diff --git a/dom/base/test/file_url.jsm b/dom/base/test/file_url.jsm index 5a83b798b685..ec0a5a4f7b3c 100644 --- a/dom/base/test/file_url.jsm +++ b/dom/base/test/file_url.jsm @@ -13,6 +13,10 @@ this.checkFromJSM = function checkFromJSM(ok, is) { var url = URL.createObjectURL(blob); ok(url, "URL is created!"); + var p = URL.getPrincipalFromURL(url); + ok(p, "Principal exists."); + ok(p instanceof Components.interfaces.nsIPrincipal, "Principal is a nsIPrincipal"); + URL.revokeObjectURL(url); ok(true, "URL is revoked"); } diff --git a/dom/base/test/mochitest.ini b/dom/base/test/mochitest.ini index 8b8ac1b030ce..f7bef0e3e278 100644 --- a/dom/base/test/mochitest.ini +++ b/dom/base/test/mochitest.ini @@ -24,6 +24,7 @@ skip-if = buildapp == 'mulet' [test_bug989665.html] [test_bug999456.html] [test_bug1022229.html] +[test_bug1043106.html] [test_clearTimeoutIntervalNoArg.html] [test_consoleEmptyStack.html] [test_constructor-assignment.html] diff --git a/dom/base/test/test_bug1043106.html b/dom/base/test/test_bug1043106.html new file mode 100644 index 000000000000..d55893acb1d0 --- /dev/null +++ b/dom/base/test/test_bug1043106.html @@ -0,0 +1,43 @@ + + + + + + Test for Bug 1043106 + + + + + Mozilla Bug 1043106 + + + + diff --git a/dom/base/test/test_url.html b/dom/base/test/test_url.html index 7b7e8ba4bd63..09af8cf32dee 100644 --- a/dom/base/test/test_url.html +++ b/dom/base/test/test_url.html @@ -319,5 +319,19 @@ url.hostname = "2001::1"; is(url.hostname, "localhost", "Setting bad hostname fails"); + + diff --git a/dom/events/IMEContentObserver.cpp b/dom/events/IMEContentObserver.cpp index aa015168b959..6501239f8c9e 100644 --- a/dom/events/IMEContentObserver.cpp +++ b/dom/events/IMEContentObserver.cpp @@ -7,6 +7,7 @@ #include "ContentEventHandler.h" #include "IMEContentObserver.h" #include "mozilla/AsyncEventDispatcher.h" +#include "mozilla/AutoRestore.h" #include "mozilla/EventStateManager.h" #include "mozilla/IMEStateManager.h" #include "mozilla/TextComposition.h" @@ -87,6 +88,7 @@ IMEContentObserver::IMEContentObserver() , mIsSelectionChangeEventPending(false) , mSelectionChangeCausedOnlyByComposition(false) , mIsPositionChangeEventPending(false) + , mIsFlushingPendingNotifications(false) { #ifdef DEBUG TestMergingTextChangeData(); @@ -429,12 +431,14 @@ class TextChangeEvent : public nsRunnable { public: TextChangeEvent(IMEContentObserver* aDispatcher, - const IMEContentObserver::TextChangeData& aData) + IMEContentObserver::TextChangeData& aData) : mDispatcher(aDispatcher) , mData(aData) { MOZ_ASSERT(mDispatcher); MOZ_ASSERT(mData.mStored); + // Reset mStored because this now consumes the data. + aData.mStored = false; } NS_IMETHOD Run() @@ -962,27 +966,73 @@ IMEContentObserver::MaybeNotifyIMEOfPositionChange() FlushMergeableNotifications(); } +class AsyncMergeableNotificationsFlusher : public nsRunnable +{ +public: + AsyncMergeableNotificationsFlusher(IMEContentObserver* aIMEContentObserver) + : mIMEContentObserver(aIMEContentObserver) + { + MOZ_ASSERT(mIMEContentObserver); + } + + NS_IMETHOD Run() + { + mIMEContentObserver->FlushMergeableNotifications(); + return NS_OK; + } + +private: + nsRefPtr mIMEContentObserver; +}; + void IMEContentObserver::FlushMergeableNotifications() { + // If we're in handling an edit action, this method will be called later. + // If this is already detached from the widget, this doesn't need to notify + // anything. if (mIsEditorInTransaction || !mWidget) { return; } + // Notifying something may cause nested call of this method. For example, + // when somebody notified one of the notifications may dispatch query content + // event. Then, it causes flushing layout which may cause another layout + // change notification. + + if (mIsFlushingPendingNotifications) { + // So, if this is already called, this should do nothing. + return; + } + + AutoRestore flusing(mIsFlushingPendingNotifications); + mIsFlushingPendingNotifications = true; + + // NOTE: Reset each pending flag because sending notification may cause + // another change. + if (mTextChangeData.mStored) { nsContentUtils::AddScriptRunner(new TextChangeEvent(this, mTextChangeData)); - mTextChangeData.mStored = false; } if (mIsSelectionChangeEventPending) { + mIsSelectionChangeEventPending = false; nsContentUtils::AddScriptRunner( new SelectionChangeEvent(this, mSelectionChangeCausedOnlyByComposition)); - mIsSelectionChangeEventPending = false; } if (mIsPositionChangeEventPending) { - nsContentUtils::AddScriptRunner(new PositionChangeEvent(this)); mIsPositionChangeEventPending = false; + nsContentUtils::AddScriptRunner(new PositionChangeEvent(this)); + } + + // If notifications may cause new change, we should notify them now. + if (mTextChangeData.mStored || + mIsSelectionChangeEventPending || + mIsPositionChangeEventPending) { + nsRefPtr asyncFlusher = + new AsyncMergeableNotificationsFlusher(this); + NS_DispatchToCurrentThread(asyncFlusher); } } diff --git a/dom/events/IMEContentObserver.h b/dom/events/IMEContentObserver.h index 50220e7adab5..7f1b1e684299 100644 --- a/dom/events/IMEContentObserver.h +++ b/dom/events/IMEContentObserver.h @@ -38,6 +38,8 @@ class IMEContentObserver MOZ_FINAL : public nsISelectionListener , public nsSupportsWeakReference , public nsIEditorObserver { + friend class AsyncMergeableNotificationsFlusher; + public: IMEContentObserver(); @@ -219,6 +221,7 @@ private: bool mIsSelectionChangeEventPending; bool mSelectionChangeCausedOnlyByComposition; bool mIsPositionChangeEventPending; + bool mIsFlushingPendingNotifications; }; } // namespace mozilla diff --git a/dom/src/storage/DOMStorage.h b/dom/src/storage/DOMStorage.h index ae79969adde4..b9ffcce23cd0 100644 --- a/dom/src/storage/DOMStorage.h +++ b/dom/src/storage/DOMStorage.h @@ -126,6 +126,12 @@ public: bool IsPrivate() const { return mIsPrivate; } bool IsSessionOnly() const { return mIsSessionOnly; } + bool IsForkOf(const DOMStorage* aOther) const + { + MOZ_ASSERT(aOther); + return mCache == aOther->mCache; + } + private: ~DOMStorage(); diff --git a/dom/webidl/URL.webidl b/dom/webidl/URL.webidl index 042fcbae6a80..66e7d0496034 100644 --- a/dom/webidl/URL.webidl +++ b/dom/webidl/URL.webidl @@ -38,3 +38,9 @@ partial interface URL { [Throws] static DOMString? createObjectURL(MediaSource source, optional objectURLOptions options); }; + +// mozilla extensions +partial interface URL { + [Throws, ChromeOnly] + static Principal getPrincipalFromURL(DOMString blobURL); +}; diff --git a/dom/workers/URL.cpp b/dom/workers/URL.cpp index 9a44abb0fc8a..cc937ae48554 100644 --- a/dom/workers/URL.cpp +++ b/dom/workers/URL.cpp @@ -893,6 +893,16 @@ URL::RevokeObjectURL(const GlobalObject& aGlobal, const nsAString& aUrl) } } +// static +nsIPrincipal* +URL::GetPrincipalFromURL(const GlobalObject& aGlobal, const nsAString& aUrl, + ErrorResult& aRv) +{ + // This method is not implemented in workers. + aRv.Throw(NS_ERROR_NOT_IMPLEMENTED); + return nullptr; +} + void URL::URLSearchParamsUpdated(URLSearchParams* aSearchParams) { diff --git a/dom/workers/URL.h b/dom/workers/URL.h index 7a7c47559828..10897194c44a 100644 --- a/dom/workers/URL.h +++ b/dom/workers/URL.h @@ -13,6 +13,8 @@ #include "mozilla/dom/BindingDeclarations.h" #include "mozilla/dom/URLSearchParams.h" +class nsIPrincipal; + namespace mozilla { namespace dom { struct objectURLOptions; @@ -67,6 +69,10 @@ public: static void RevokeObjectURL(const GlobalObject& aGlobal, const nsAString& aUrl); + static nsIPrincipal* GetPrincipalFromURL(const GlobalObject& aGlobal, + const nsAString& aURL, + ErrorResult& aError); + void GetHref(nsString& aHref, ErrorResult& aRv) const; void SetHref(const nsAString& aHref, ErrorResult& aRv); diff --git a/gfx/skia/trunk/src/opts/SkBlitRow_opts_arm.cpp.orig b/gfx/skia/trunk/src/opts/SkBlitRow_opts_arm.cpp.orig deleted file mode 100644 index 34b8564723cd..000000000000 --- a/gfx/skia/trunk/src/opts/SkBlitRow_opts_arm.cpp.orig +++ /dev/null @@ -1,380 +0,0 @@ -/* - * Copyright 2012 The Android Open Source Project - * - * Use of this source code is governed by a BSD-style license that can be - * found in the LICENSE file. - */ - -#include "SkBlitRow.h" -#include "SkColorPriv.h" -#include "SkDither.h" -#include "SkMathPriv.h" -#include "SkUtils.h" -#include "SkUtilsArm.h" - -// Define USE_NEON_CODE to indicate that we need to build NEON routines -#define USE_NEON_CODE (!SK_ARM_NEON_IS_NONE) - -// Define USE_ARM_CODE to indicate that we need to build ARM routines -#define USE_ARM_CODE (!SK_ARM_NEON_IS_ALWAYS) - -#if USE_NEON_CODE - #include "SkBlitRow_opts_arm_neon.h" -#endif - -#if USE_ARM_CODE - -static void S32A_D565_Opaque(uint16_t* SK_RESTRICT dst, - const SkPMColor* SK_RESTRICT src, int count, - U8CPU alpha, int /*x*/, int /*y*/) { - SkASSERT(255 == alpha); - - asm volatile ( - "1: \n\t" - "ldr r3, [%[src]], #4 \n\t" - "cmp r3, #0xff000000 \n\t" - "blo 2f \n\t" - "and r4, r3, #0x0000f8 \n\t" - "and r5, r3, #0x00fc00 \n\t" - "and r6, r3, #0xf80000 \n\t" - "pld [r1, #32] \n\t" - "lsl r3, r4, #8 \n\t" - "orr r3, r3, r5, lsr #5 \n\t" - "orr r3, r3, r6, lsr #19 \n\t" - "subs %[count], %[count], #1 \n\t" - "strh r3, [%[dst]], #2 \n\t" - "bne 1b \n\t" - "b 4f \n\t" - "2: \n\t" - "lsrs r7, r3, #24 \n\t" - "beq 3f \n\t" - "ldrh r4, [%[dst]] \n\t" - "rsb r7, r7, #255 \n\t" - "and r6, r4, #0x001f \n\t" -#if SK_ARM_ARCH == 6 - "lsl r5, r4, #21 \n\t" - "lsr r5, r5, #26 \n\t" -#else - "ubfx r5, r4, #5, #6 \n\t" -#endif - "pld [r0, #16] \n\t" - "lsr r4, r4, #11 \n\t" -#ifdef SK_ARM_HAS_EDSP - "smulbb r6, r6, r7 \n\t" - "smulbb r5, r5, r7 \n\t" - "smulbb r4, r4, r7 \n\t" -#else - "mul r6, r6, r7 \n\t" - "mul r5, r5, r7 \n\t" - "mul r4, r4, r7 \n\t" -#endif - "uxtb r7, r3, ROR #16 \n\t" - "uxtb ip, r3, ROR #8 \n\t" - "and r3, r3, #0xff \n\t" - "add r6, r6, #16 \n\t" - "add r5, r5, #32 \n\t" - "add r4, r4, #16 \n\t" - "add r6, r6, r6, lsr #5 \n\t" - "add r5, r5, r5, lsr #6 \n\t" - "add r4, r4, r4, lsr #5 \n\t" - "add r6, r7, r6, lsr #5 \n\t" - "add r5, ip, r5, lsr #6 \n\t" - "add r4, r3, r4, lsr #5 \n\t" - "lsr r6, r6, #3 \n\t" - "and r5, r5, #0xfc \n\t" - "and r4, r4, #0xf8 \n\t" - "orr r6, r6, r5, lsl #3 \n\t" - "orr r4, r6, r4, lsl #8 \n\t" - "strh r4, [%[dst]], #2 \n\t" - "pld [r1, #32] \n\t" - "subs %[count], %[count], #1 \n\t" - "bne 1b \n\t" - "b 4f \n\t" - "3: \n\t" - "subs %[count], %[count], #1 \n\t" - "add %[dst], %[dst], #2 \n\t" - "bne 1b \n\t" - "4: \n\t" - : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) - : - : "memory", "cc", "r3", "r4", "r5", "r6", "r7", "ip" - ); -} - -static void S32A_Opaque_BlitRow32_arm(SkPMColor* SK_RESTRICT dst, - const SkPMColor* SK_RESTRICT src, - int count, U8CPU alpha) { - - SkASSERT(255 == alpha); - - asm volatile ( - "cmp %[count], #0 \n\t" /* comparing count with 0 */ - "beq 3f \n\t" /* if zero exit */ - - "mov ip, #0xff \n\t" /* load the 0xff mask in ip */ - "orr ip, ip, ip, lsl #16 \n\t" /* convert it to 0xff00ff in ip */ - - "cmp %[count], #2 \n\t" /* compare count with 2 */ - "blt 2f \n\t" /* if less than 2 -> single loop */ - - /* Double Loop */ - "1: \n\t" /* */ - "ldm %[src]!, {r5,r6} \n\t" /* load the src(s) at r5-r6 */ - "ldm %[dst], {r7,r8} \n\t" /* loading dst(s) into r7-r8 */ - "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */ - - /* ----------- */ - "and r9, ip, r7 \n\t" /* r9 = br masked by ip */ - "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */ - "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */ - - "mul r9, r9, r4 \n\t" /* br = br * scale */ - "mul r10, r10, r4 \n\t" /* ag = ag * scale */ - "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */ - - "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */ - "lsr r4, r6, #24 \n\t" /* extracting the alpha from source and storing it to r4 */ - "orr r7, r9, r10 \n\t" /* br | ag*/ - - "add r7, r5, r7 \n\t" /* dst = src + calc dest(r7) */ - "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 255 -> r4=scale */ - - /* ----------- */ - "and r9, ip, r8 \n\t" /* r9 = br masked by ip */ - - "and r10, ip, r8, lsr #8 \n\t" /* r10 = ag masked by ip */ - "mul r9, r9, r4 \n\t" /* br = br * scale */ - "sub %[count], %[count], #2 \n\t" - "mul r10, r10, r4 \n\t" /* ag = ag * scale */ - - "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */ - "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */ - "cmp %[count], #1 \n\t" /* comparing count with 1 */ - "orr r8, r9, r10 \n\t" /* br | ag */ - - "add r8, r6, r8 \n\t" /* dst = src + calc dest(r8) */ - - /* ----------------- */ - "stm %[dst]!, {r7,r8} \n\t" /* *dst = r7, increment dst by two (each times 4) */ - /* ----------------- */ - - "bgt 1b \n\t" /* if greater than 1 -> reloop */ - "blt 3f \n\t" /* if less than 1 -> exit */ - - /* Single Loop */ - "2: \n\t" /* */ - "ldr r5, [%[src]], #4 \n\t" /* load the src pointer into r5 r5=src */ - "ldr r7, [%[dst]] \n\t" /* loading dst into r7 */ - "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */ - - /* ----------- */ - "and r9, ip, r7 \n\t" /* r9 = br masked by ip */ - "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */ - - "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */ - "mul r9, r9, r4 \n\t" /* br = br * scale */ - "mul r10, r10, r4 \n\t" /* ag = ag * scale */ - "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */ - - "and r10, r10, ip, lsl #8 \n\t" /* mask ag */ - "orr r7, r9, r10 \n\t" /* br | ag */ - - "add r7, r5, r7 \n\t" /* *dst = src + calc dest(r7) */ - - /* ----------------- */ - "str r7, [%[dst]], #4 \n\t" /* *dst = r7, increment dst by one (times 4) */ - /* ----------------- */ - - "3: \n\t" /* */ - : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) - : - : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "ip", "memory" - ); -} - -/* - * ARM asm version of S32A_Blend_BlitRow32 - */ -void S32A_Blend_BlitRow32_arm(SkPMColor* SK_RESTRICT dst, - const SkPMColor* SK_RESTRICT src, - int count, U8CPU alpha) { - asm volatile ( - "cmp %[count], #0 \n\t" /* comparing count with 0 */ - "beq 3f \n\t" /* if zero exit */ - - "mov r12, #0xff \n\t" /* load the 0xff mask in r12 */ - "orr r12, r12, r12, lsl #16 \n\t" /* convert it to 0xff00ff in r12 */ - - /* src1,2_scale */ - "add %[alpha], %[alpha], #1 \n\t" /* loading %[alpha]=src_scale=alpha+1 */ - - "cmp %[count], #2 \n\t" /* comparing count with 2 */ - "blt 2f \n\t" /* if less than 2 -> single loop */ - - /* Double Loop */ - "1: \n\t" /* */ - "ldm %[src]!, {r5, r6} \n\t" /* loading src pointers into r5 and r6 */ - "ldm %[dst], {r7, r8} \n\t" /* loading dst pointers into r7 and r8 */ - - /* dst1_scale and dst2_scale*/ - "lsr r9, r5, #24 \n\t" /* src >> 24 */ - "lsr r10, r6, #24 \n\t" /* src >> 24 */ -#ifdef SK_ARM_HAS_EDSP - "smulbb r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */ - "smulbb r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */ -#else - "mul r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */ - "mul r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */ -#endif - "lsr r9, r9, #8 \n\t" /* r9 >> 8 */ - "lsr r10, r10, #8 \n\t" /* r10 >> 8 */ - "rsb r9, r9, #256 \n\t" /* dst1_scale = r9 = 255 - r9 + 1 */ - "rsb r10, r10, #256 \n\t" /* dst2_scale = r10 = 255 - r10 + 1 */ - - /* ---------------------- */ - - /* src1, src1_scale */ - "and r11, r12, r5, lsr #8 \n\t" /* ag = r11 = r5 masked by r12 lsr by #8 */ - "and r4, r12, r5 \n\t" /* rb = r4 = r5 masked by r12 */ - "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */ - "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */ - "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ - "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ - "orr r5, r11, r4 \n\t" /* r5 = (src1, src_scale) */ - - /* dst1, dst1_scale */ - "and r11, r12, r7, lsr #8 \n\t" /* ag = r11 = r7 masked by r12 lsr by #8 */ - "and r4, r12, r7 \n\t" /* rb = r4 = r7 masked by r12 */ - "mul r11, r11, r9 \n\t" /* ag = r11 times dst_scale (r9) */ - "mul r4, r4, r9 \n\t" /* rb = r4 times dst_scale (r9) */ - "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ - "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ - "orr r9, r11, r4 \n\t" /* r9 = (dst1, dst_scale) */ - - /* ---------------------- */ - "add r9, r5, r9 \n\t" /* *dst = src plus dst both scaled */ - /* ---------------------- */ - - /* ====================== */ - - /* src2, src2_scale */ - "and r11, r12, r6, lsr #8 \n\t" /* ag = r11 = r6 masked by r12 lsr by #8 */ - "and r4, r12, r6 \n\t" /* rb = r4 = r6 masked by r12 */ - "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */ - "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */ - "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ - "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ - "orr r6, r11, r4 \n\t" /* r6 = (src2, src_scale) */ - - /* dst2, dst2_scale */ - "and r11, r12, r8, lsr #8 \n\t" /* ag = r11 = r8 masked by r12 lsr by #8 */ - "and r4, r12, r8 \n\t" /* rb = r4 = r8 masked by r12 */ - "mul r11, r11, r10 \n\t" /* ag = r11 times dst_scale (r10) */ - "mul r4, r4, r10 \n\t" /* rb = r4 times dst_scale (r6) */ - "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ - "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ - "orr r10, r11, r4 \n\t" /* r10 = (dst2, dst_scale) */ - - "sub %[count], %[count], #2 \n\t" /* decrease count by 2 */ - /* ---------------------- */ - "add r10, r6, r10 \n\t" /* *dst = src plus dst both scaled */ - /* ---------------------- */ - "cmp %[count], #1 \n\t" /* compare count with 1 */ - /* ----------------- */ - "stm %[dst]!, {r9, r10} \n\t" /* copy r9 and r10 to r7 and r8 respectively */ - /* ----------------- */ - - "bgt 1b \n\t" /* if %[count] greater than 1 reloop */ - "blt 3f \n\t" /* if %[count] less than 1 exit */ - /* else get into the single loop */ - /* Single Loop */ - "2: \n\t" /* */ - "ldr r5, [%[src]], #4 \n\t" /* loading src pointer into r5: r5=src */ - "ldr r7, [%[dst]] \n\t" /* loading dst pointer into r7: r7=dst */ - - "lsr r6, r5, #24 \n\t" /* src >> 24 */ - "and r8, r12, r5, lsr #8 \n\t" /* ag = r8 = r5 masked by r12 lsr by #8 */ -#ifdef SK_ARM_HAS_EDSP - "smulbb r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */ -#else - "mul r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */ -#endif - "and r9, r12, r5 \n\t" /* rb = r9 = r5 masked by r12 */ - "lsr r6, r6, #8 \n\t" /* r6 >> 8 */ - "mul r8, r8, %[alpha] \n\t" /* ag = r8 times scale */ - "rsb r6, r6, #256 \n\t" /* r6 = 255 - r6 + 1 */ - - /* src, src_scale */ - "mul r9, r9, %[alpha] \n\t" /* rb = r9 times scale */ - "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ - "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */ - "orr r10, r8, r9 \n\t" /* r10 = (scr, src_scale) */ - - /* dst, dst_scale */ - "and r8, r12, r7, lsr #8 \n\t" /* ag = r8 = r7 masked by r12 lsr by #8 */ - "and r9, r12, r7 \n\t" /* rb = r9 = r7 masked by r12 */ - "mul r8, r8, r6 \n\t" /* ag = r8 times scale (r6) */ - "mul r9, r9, r6 \n\t" /* rb = r9 times scale (r6) */ - "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ - "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */ - "orr r7, r8, r9 \n\t" /* r7 = (dst, dst_scale) */ - - "add r10, r7, r10 \n\t" /* *dst = src plus dst both scaled */ - - /* ----------------- */ - "str r10, [%[dst]], #4 \n\t" /* *dst = r10, postincrement dst by one (times 4) */ - /* ----------------- */ - - "3: \n\t" /* */ - : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count), [alpha] "+r" (alpha) - : - : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11", "r12", "memory" - ); - -} - -/////////////////////////////////////////////////////////////////////////////// - -static const SkBlitRow::Proc sk_blitrow_platform_565_procs_arm[] = { - // no dither - // NOTE: For the functions below, we don't have a special version - // that assumes that each source pixel is opaque. But our S32A is - // still faster than the default, so use it. - S32A_D565_Opaque, // S32_D565_Opaque - NULL, // S32_D565_Blend - S32A_D565_Opaque, // S32A_D565_Opaque - NULL, // S32A_D565_Blend - - // dither - NULL, // S32_D565_Opaque_Dither - NULL, // S32_D565_Blend_Dither - NULL, // S32A_D565_Opaque_Dither - NULL, // S32A_D565_Blend_Dither -}; - -static const SkBlitRow::Proc32 sk_blitrow_platform_32_procs_arm[] = { - NULL, // S32_Opaque, - NULL, // S32_Blend, - S32A_Opaque_BlitRow32_arm, // S32A_Opaque, - S32A_Blend_BlitRow32_arm // S32A_Blend -}; - -#endif // USE_ARM_CODE - -SkBlitRow::Proc SkBlitRow::PlatformProcs565(unsigned flags) { - return SK_ARM_NEON_WRAP(sk_blitrow_platform_565_procs_arm)[flags]; -} - -SkBlitRow::Proc32 SkBlitRow::PlatformProcs32(unsigned flags) { - return SK_ARM_NEON_WRAP(sk_blitrow_platform_32_procs_arm)[flags]; -} - -/////////////////////////////////////////////////////////////////////////////// -#define Color32_arm NULL -SkBlitRow::ColorProc SkBlitRow::PlatformColorProc() { - return SK_ARM_NEON_WRAP(Color32_arm); -} - -SkBlitRow::ColorRectProc PlatformColorRectProcFactory() { - return NULL; -} diff --git a/gfx/skia/trunk/src/pathops/SkLineParameters.h.orig b/gfx/skia/trunk/src/pathops/SkLineParameters.h.orig deleted file mode 100644 index 92343c691bd2..000000000000 --- a/gfx/skia/trunk/src/pathops/SkLineParameters.h.orig +++ /dev/null @@ -1,175 +0,0 @@ -/* - * Copyright 2012 Google Inc. - * - * Use of this source code is governed by a BSD-style license that can be - * found in the LICENSE file. - */ -#include "SkPathOpsCubic.h" -#include "SkPathOpsLine.h" -#include "SkPathOpsQuad.h" - -// Sources -// computer-aided design - volume 22 number 9 november 1990 pp 538 - 549 -// online at http://cagd.cs.byu.edu/~tom/papers/bezclip.pdf - -// This turns a line segment into a parameterized line, of the form -// ax + by + c = 0 -// When a^2 + b^2 == 1, the line is normalized. -// The distance to the line for (x, y) is d(x,y) = ax + by + c -// -// Note that the distances below are not necessarily normalized. To get the true -// distance, it's necessary to either call normalize() after xxxEndPoints(), or -// divide the result of xxxDistance() by sqrt(normalSquared()) - -class SkLineParameters { -public: - - bool cubicEndPoints(const SkDCubic& pts) { - int endIndex = 1; - cubicEndPoints(pts, 0, endIndex); - if (dy() != 0) { - return true; - } - if (dx() == 0) { - cubicEndPoints(pts, 0, ++endIndex); - SkASSERT(endIndex == 2); - if (dy() != 0) { - return true; - } - if (dx() == 0) { - cubicEndPoints(pts, 0, ++endIndex); // line - SkASSERT(endIndex == 3); - return false; - } - } - // FIXME: after switching to round sort, remove bumping fA - if (dx() < 0) { // only worry about y bias when breaking cw/ccw tie - return true; - } - // if cubic tangent is on x axis, look at next control point to break tie - // control point may be approximate, so it must move significantly to account for error - if (NotAlmostEqualUlps(pts[0].fY, pts[++endIndex].fY)) { - if (pts[0].fY > pts[endIndex].fY) { - fA = DBL_EPSILON; // push it from 0 to slightly negative (y() returns -a) - } - return true; - } - if (endIndex == 3) { - return true; - } - SkASSERT(endIndex == 2); - if (pts[0].fY > pts[3].fY) { - fA = DBL_EPSILON; // push it from 0 to slightly negative (y() returns -a) - } - return true; - } - - void cubicEndPoints(const SkDCubic& pts, int s, int e) { - fA = pts[s].fY - pts[e].fY; - fB = pts[e].fX - pts[s].fX; - fC = pts[s].fX * pts[e].fY - pts[e].fX * pts[s].fY; - } - - double cubicPart(const SkDCubic& part) { - cubicEndPoints(part); - if (part[0] == part[1] || ((const SkDLine& ) part[0]).nearRay(part[2])) { - return pointDistance(part[3]); - } - return pointDistance(part[2]); - } - - void lineEndPoints(const SkDLine& pts) { - fA = pts[0].fY - pts[1].fY; - fB = pts[1].fX - pts[0].fX; - fC = pts[0].fX * pts[1].fY - pts[1].fX * pts[0].fY; - } - - bool quadEndPoints(const SkDQuad& pts) { - quadEndPoints(pts, 0, 1); - if (dy() != 0) { - return true; - } - if (dx() == 0) { - quadEndPoints(pts, 0, 2); - return false; - } - if (dx() < 0) { // only worry about y bias when breaking cw/ccw tie - return true; - } - // FIXME: after switching to round sort, remove this - if (pts[0].fY > pts[2].fY) { - fA = DBL_EPSILON; - } - return true; - } - - void quadEndPoints(const SkDQuad& pts, int s, int e) { - fA = pts[s].fY - pts[e].fY; - fB = pts[e].fX - pts[s].fX; - fC = pts[s].fX * pts[e].fY - pts[e].fX * pts[s].fY; - } - - double quadPart(const SkDQuad& part) { - quadEndPoints(part); - return pointDistance(part[2]); - } - - double normalSquared() const { - return fA * fA + fB * fB; - } - - bool normalize() { - double normal = sqrt(normalSquared()); - if (approximately_zero(normal)) { - fA = fB = fC = 0; - return false; - } - double reciprocal = 1 / normal; - fA *= reciprocal; - fB *= reciprocal; - fC *= reciprocal; - return true; - } - - void cubicDistanceY(const SkDCubic& pts, SkDCubic& distance) const { - double oneThird = 1 / 3.0; - for (int index = 0; index < 4; ++index) { - distance[index].fX = index * oneThird; - distance[index].fY = fA * pts[index].fX + fB * pts[index].fY + fC; - } - } - - void quadDistanceY(const SkDQuad& pts, SkDQuad& distance) const { - double oneHalf = 1 / 2.0; - for (int index = 0; index < 3; ++index) { - distance[index].fX = index * oneHalf; - distance[index].fY = fA * pts[index].fX + fB * pts[index].fY + fC; - } - } - - double controlPtDistance(const SkDCubic& pts, int index) const { - SkASSERT(index == 1 || index == 2); - return fA * pts[index].fX + fB * pts[index].fY + fC; - } - - double controlPtDistance(const SkDQuad& pts) const { - return fA * pts[1].fX + fB * pts[1].fY + fC; - } - - double pointDistance(const SkDPoint& pt) const { - return fA * pt.fX + fB * pt.fY + fC; - } - - double dx() const { - return fB; - } - - double dy() const { - return -fA; - } - -private: - double fA; - double fB; - double fC; -}; diff --git a/gfx/skia/trunk/src/pathops/SkLineParameters.h.rej b/gfx/skia/trunk/src/pathops/SkLineParameters.h.rej deleted file mode 100644 index 9374f1f34c4a..000000000000 --- a/gfx/skia/trunk/src/pathops/SkLineParameters.h.rej +++ /dev/null @@ -1,13 +0,0 @@ ---- src/pathops/SkLineParameters.h -+++ src/pathops/SkLineParameters.h -@@ -168,8 +172,10 @@ - return -a; - } - - private: - double a; - double b; - double c; - }; -+ -+#endif diff --git a/js/src/assembler/assembler/MacroAssembler.h b/js/src/assembler/assembler/MacroAssembler.h index bfcc47eca387..0d86a42ad2e4 100644 --- a/js/src/assembler/assembler/MacroAssembler.h +++ b/js/src/assembler/assembler/MacroAssembler.h @@ -39,13 +39,8 @@ #include "jit/none/BaseMacroAssembler-none.h" namespace JSC { typedef MacroAssemblerNone MacroAssembler; } -#elif WTF_CPU_ARM_THUMB2 -#include "assembler/assembler/MacroAssemblerARMv7.h" -namespace JSC { typedef MacroAssemblerARMv7 MacroAssembler; } - -#elif WTF_CPU_ARM_TRADITIONAL -#include "assembler/assembler/MacroAssemblerARM.h" -namespace JSC { typedef MacroAssemblerARM MacroAssembler; } +#elif JS_CODEGEN_ARM +// Merged with the jit backend support. #elif WTF_CPU_MIPS #include "assembler/assembler/MacroAssemblerMIPS.h" diff --git a/js/src/assembler/assembler/MacroAssemblerARM.cpp b/js/src/assembler/assembler/MacroAssemblerARM.cpp deleted file mode 100644 index 91c5591d2e85..000000000000 --- a/js/src/assembler/assembler/MacroAssemblerARM.cpp +++ /dev/null @@ -1,100 +0,0 @@ -/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- - * vim: set ts=8 sts=4 et sw=4 tw=99: - * - * ***** BEGIN LICENSE BLOCK ***** - * Copyright (C) 2009 University of Szeged - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * - * THIS SOFTWARE IS PROVIDED BY UNIVERSITY OF SZEGED ``AS IS'' AND ANY - * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE - * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR - * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL UNIVERSITY OF SZEGED OR - * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY - * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * ***** END LICENSE BLOCK ***** */ - -#include "assembler/wtf/Platform.h" - -#if ENABLE_ASSEMBLER && WTF_CPU_ARM_TRADITIONAL - -#include "assembler/assembler/MacroAssemblerARM.h" - -#if (WTF_OS_LINUX || WTF_OS_ANDROID) && !defined(JS_ARM_SIMULATOR) -#include -#include -#include -#include -#include -#include - -// lame check for kernel version -// see bug 586550 -#include -#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28) -#include -#else -#include -#endif - -#endif - -namespace JSC { - -static bool isVFPPresent() -{ -#ifdef JS_ARM_SIMULATOR - return true; -#else -#if WTF_OS_LINUX - int fd = open("/proc/self/auxv", O_RDONLY); - if (fd > 0) { - Elf32_auxv_t aux; - while (read(fd, &aux, sizeof(Elf32_auxv_t))) { - if (aux.a_type == AT_HWCAP) { - close(fd); - return aux.a_un.a_val & HWCAP_VFP; - } - } - close(fd); - } -#endif - -#if defined(__GNUC__) && defined(__VFP_FP__) - return true; -#endif - -#ifdef WTF_OS_ANDROID - FILE *fp = fopen("/proc/cpuinfo", "r"); - if (!fp) - return false; - - char buf[1024]; - fread(buf, sizeof(char), sizeof(buf), fp); - fclose(fp); - if (strstr(buf, "vfp")) - return true; -#endif - return false; -#endif // JS_ARM_SIMULATOR -} - -const bool MacroAssemblerARM::s_isVFPPresent = isVFPPresent(); - -} - -#endif // ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) diff --git a/js/src/assembler/assembler/MacroAssemblerARM.h b/js/src/assembler/assembler/MacroAssemblerARM.h deleted file mode 100644 index d46047706d8c..000000000000 --- a/js/src/assembler/assembler/MacroAssemblerARM.h +++ /dev/null @@ -1,51 +0,0 @@ -/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- - * vim: set ts=8 sts=4 et sw=4 tw=99: - * - * ***** BEGIN LICENSE BLOCK ***** - * Copyright (C) 2008 Apple Inc. - * Copyright (C) 2009, 2010 University of Szeged - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * - * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY - * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE - * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR - * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR - * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY - * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * ***** END LICENSE BLOCK ***** */ - -#ifndef assembler_assembler_MacroAssemblerARM_h -#define assembler_assembler_MacroAssemblerARM_h - -#include "assembler/wtf/Platform.h" - -#if ENABLE_ASSEMBLER && WTF_CPU_ARM_TRADITIONAL - -namespace JSC { - -class MacroAssemblerARM { -public: - static bool supportsFloatingPoint() { return s_isVFPPresent; } - static const bool s_isVFPPresent; -}; - -} - -#endif // ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) - -#endif /* assembler_assembler_MacroAssemblerARM_h */ diff --git a/js/src/assembler/assembler/MacroAssemblerARMv7.h b/js/src/assembler/assembler/MacroAssemblerARMv7.h deleted file mode 100644 index 881e4bd64e91..000000000000 --- a/js/src/assembler/assembler/MacroAssemblerARMv7.h +++ /dev/null @@ -1,49 +0,0 @@ -/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- - * vim: set ts=8 sts=4 et sw=4 tw=99: - * - * ***** BEGIN LICENSE BLOCK ***** - * Copyright (C) 2009 Apple Inc. All rights reserved. - * Copyright (C) 2010 University of Szeged - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * - * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY - * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE - * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR - * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR - * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY - * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * ***** END LICENSE BLOCK ***** */ - -#ifndef assembler_assembler_MacroAssemblerARMv7_h -#define assembler_assembler_MacroAssemblerARMv7_h - -#include "assembler/wtf/Platform.h" - -#if ENABLE(ASSEMBLER) - -namespace JSC { - -class MacroAssemblerARMv7 { -public: - static bool supportsFloatingPoint() { return true; } -}; - -} // namespace JSC - -#endif // ENABLE(ASSEMBLER) - -#endif /* assembler_assembler_MacroAssemblerARMv7_h */ diff --git a/js/src/assembler/assembler/X86Assembler.h b/js/src/assembler/assembler/X86Assembler.h index 767bafc3d867..032b99c64e3c 100644 --- a/js/src/assembler/assembler/X86Assembler.h +++ b/js/src/assembler/assembler/X86Assembler.h @@ -298,12 +298,16 @@ private: OP2_UCOMISD_VsdWsd = 0x2E, OP2_MOVMSKPD_EdVd = 0x50, OP2_ADDSD_VsdWsd = 0x58, + OP2_ADDPS_VpsWps = 0x58, OP2_MULSD_VsdWsd = 0x59, + OP2_MULPS_VpsWps = 0x59, OP2_CVTSS2SD_VsdEd = 0x5A, OP2_CVTSD2SS_VsdEd = 0x5A, OP2_SUBSD_VsdWsd = 0x5C, + OP2_SUBPS_VpsWps = 0x5C, OP2_MINSD_VsdWsd = 0x5D, OP2_DIVSD_VsdWsd = 0x5E, + OP2_DIVPS_VpsWps = 0x5E, OP2_MAXSD_VsdWsd = 0x5F, OP2_SQRTSD_VsdWsd = 0x51, OP2_SQRTSS_VssWss = 0x51, @@ -328,7 +332,10 @@ private: OP2_MOVZX_GvEw = 0xB7, OP2_XADD_EvGv = 0xC1, OP2_PEXTRW_GdUdIb = 0xC5, - OP2_SHUFPS_VpsWpsIb = 0xC6 + OP2_SHUFPS_VpsWpsIb = 0xC6, + OP2_PXORDQ_VdqWdq = 0xEF, + OP2_PSUBD_VdqWdq = 0xFA, + OP2_PADDD_VdqWdq = 0xFE } TwoByteOpcodeID; typedef enum { @@ -661,6 +668,124 @@ public: m_formatter.twoByteOp(OP2_XADD_EvGv, srcdest, base, index, scale, offset); } + void paddd_rr(XMMRegisterID src, XMMRegisterID dst) + { + spew("paddd %s, %s", nameFPReg(src), nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PADDD_VdqWdq, (RegisterID)dst, (RegisterID)src); + } + void paddd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + spew("paddd %s0x%x(%s), %s", + PRETTY_PRINT_OFFSET(offset), nameIReg(base), nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PADDD_VdqWdq, (RegisterID)dst, base, offset); + } + void paddd_mr(const void* address, XMMRegisterID dst) + { + spew("paddd %p, %s", + address, nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PADDD_VdqWdq, (RegisterID)dst, address); + } + + void psubd_rr(XMMRegisterID src, XMMRegisterID dst) + { + spew("psubd %s, %s", nameFPReg(src), nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PSUBD_VdqWdq, (RegisterID)dst, (RegisterID)src); + } + void psubd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + spew("psubd %s0x%x(%s), %s", + PRETTY_PRINT_OFFSET(offset), nameIReg(base), nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PSUBD_VdqWdq, (RegisterID)dst, base, offset); + } + void psubd_mr(const void* address, XMMRegisterID dst) + { + spew("psubd %p, %s", + address, nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PSUBD_VdqWdq, (RegisterID)dst, address); + } + + void addps_rr(XMMRegisterID src, XMMRegisterID dst) + { + spew("addps %s, %s", + nameFPReg(src), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_ADDPS_VpsWps, (RegisterID)dst, (RegisterID)src); + } + void addps_mr(int offset, RegisterID base, XMMRegisterID dst) + { + spew("addps %s0x%x(%s), %s", + PRETTY_PRINT_OFFSET(offset), nameIReg(base), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_ADDPS_VpsWps, (RegisterID)dst, base, offset); + } + void addps_mr(const void* address, XMMRegisterID dst) + { + spew("addps %p, %s", + address, nameFPReg(dst)); + m_formatter.twoByteOp(OP2_ADDPS_VpsWps, (RegisterID)dst, address); + } + + void subps_rr(XMMRegisterID src, XMMRegisterID dst) + { + spew("subps %s, %s", + nameFPReg(src), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_SUBPS_VpsWps, (RegisterID)dst, (RegisterID)src); + } + void subps_mr(int offset, RegisterID base, XMMRegisterID dst) + { + spew("subps %s0x%x(%s), %s", + PRETTY_PRINT_OFFSET(offset), nameIReg(base), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_SUBPS_VpsWps, (RegisterID)dst, base, offset); + } + void subps_mr(const void* address, XMMRegisterID dst) + { + spew("subps %p, %s", + address, nameFPReg(dst)); + m_formatter.twoByteOp(OP2_SUBPS_VpsWps, (RegisterID)dst, address); + } + + void mulps_rr(XMMRegisterID src, XMMRegisterID dst) + { + spew("mulps %s, %s", + nameFPReg(src), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_MULPS_VpsWps, (RegisterID)dst, (RegisterID)src); + } + void mulps_mr(int offset, RegisterID base, XMMRegisterID dst) + { + spew("mulps %s0x%x(%s), %s", + PRETTY_PRINT_OFFSET(offset), nameIReg(base), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_MULPS_VpsWps, (RegisterID)dst, base, offset); + } + void mulps_mr(const void* address, XMMRegisterID dst) + { + spew("mulps %p, %s", + address, nameFPReg(dst)); + m_formatter.twoByteOp(OP2_MULPS_VpsWps, (RegisterID)dst, address); + } + + void divps_rr(XMMRegisterID src, XMMRegisterID dst) + { + spew("divps %s, %s", + nameFPReg(src), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_DIVPS_VpsWps, (RegisterID)dst, (RegisterID)src); + } + void divps_mr(int offset, RegisterID base, XMMRegisterID dst) + { + spew("divps %s0x%x(%s), %s", + PRETTY_PRINT_OFFSET(offset), nameIReg(base), nameFPReg(dst)); + m_formatter.twoByteOp(OP2_DIVPS_VpsWps, (RegisterID)dst, base, offset); + } + void divps_mr(const void* address, XMMRegisterID dst) + { + spew("divps %p, %s", + address, nameFPReg(dst)); + m_formatter.twoByteOp(OP2_DIVPS_VpsWps, (RegisterID)dst, address); + } + void andl_rr(RegisterID src, RegisterID dst) { spew("andl %s, %s", @@ -2590,6 +2715,14 @@ public: m_formatter.twoByteOp(OP2_MOVD_VdEd, (RegisterID)dst, src); } + void pxor_rr(XMMRegisterID src, XMMRegisterID dst) + { + spew("pxor %s, %s", + nameFPReg(src), nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PXORDQ_VdqWdq, (RegisterID)dst, (RegisterID)src); + } + void pshufd_irr(uint32_t mask, XMMRegisterID src, XMMRegisterID dst) { JS_ASSERT(mask < 256); @@ -2932,6 +3065,40 @@ public: m_formatter.twoByteOp(OP2_MOVAPD_VsdWsd, (RegisterID)dst, (RegisterID)src); } +#ifdef WTF_CPU_X86_64 + JmpSrc movaps_ripr(XMMRegisterID dst) + { + spew("movaps ?(%%rip), %s", + nameFPReg(dst)); + m_formatter.twoByteRipOp(OP2_MOVAPS_VsdWsd, (RegisterID)dst, 0); + return JmpSrc(m_formatter.size()); + } + + JmpSrc movdqa_ripr(XMMRegisterID dst) + { + spew("movdqa ?(%%rip), %s", + nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteRipOp(OP2_MOVDQ_VdqWdq, (RegisterID)dst, 0); + return JmpSrc(m_formatter.size()); + } +#else + void movaps_mr(const void* address, XMMRegisterID dst) + { + spew("movaps %p, %s", + address, nameFPReg(dst)); + m_formatter.twoByteOp(OP2_MOVAPS_VsdWsd, (RegisterID)dst, address); + } + + void movdqa_mr(const void* address, XMMRegisterID dst) + { + spew("movdqa %p, %s", + address, nameFPReg(dst)); + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_MOVDQ_VdqWdq, (RegisterID)dst, address); + } +#endif // WTF_CPU_X86_64 + void movdqu_rm(XMMRegisterID src, int offset, RegisterID base) { spew("movdqu %s, %s0x%x(%s)", @@ -3346,6 +3513,19 @@ public: m_formatter.floatConstant(f); } + void int32x4Constant(const int32_t s[4]) + { + spew(".int32x4 (%d %d %d %d)", s[0], s[1], s[2], s[3]); + MOZ_ASSERT(m_formatter.isAligned(16)); + m_formatter.int32x4Constant(s); + } + void float32x4Constant(const float f[4]) + { + spew(".float32x4 (%f %f %f %f)", f[0], f[1], f[2], f[3]); + MOZ_ASSERT(m_formatter.isAligned(16)); + m_formatter.float32x4Constant(f); + } + void int64Constant(int64_t i) { spew(".quad %lld", (long long)i); @@ -4030,12 +4210,30 @@ private: m_buffer.putIntUnchecked(u.u32); } + void int32x4Constant(const int32_t s[4]) + { + for (size_t i = 0; i < 4; ++i) + int32Constant(s[i]); + } + + void float32x4Constant(const float s[4]) + { + for (size_t i = 0; i < 4; ++i) + floatConstant(s[i]); + } + void int64Constant(int64_t i) { m_buffer.ensureSpace(sizeof(int64_t)); m_buffer.putInt64Unchecked(i); } + void int32Constant(int32_t i) + { + m_buffer.ensureSpace(sizeof(int32_t)); + m_buffer.putIntUnchecked(i); + } + // Administrative methods: size_t size() const { return m_buffer.size(); } diff --git a/js/src/builtin/SIMD.cpp b/js/src/builtin/SIMD.cpp index 43233fc21b82..bd4fe35d7bb0 100644 --- a/js/src/builtin/SIMD.cpp +++ b/js/src/builtin/SIMD.cpp @@ -417,7 +417,7 @@ TypedObjectMemory(HandleValue v) template JSObject * -js::Create(JSContext *cx, typename V::Elem *data) +js::CreateSimd(JSContext *cx, typename V::Elem *data) { typedef typename V::Elem Elem; Rooted typeDescr(cx, &V::GetTypeDescr(*cx->global())); @@ -433,8 +433,8 @@ js::Create(JSContext *cx, typename V::Elem *data) return result; } -template JSObject *js::Create(JSContext *cx, Float32x4::Elem *data); -template JSObject *js::Create(JSContext *cx, Int32x4::Elem *data); +template JSObject *js::CreateSimd(JSContext *cx, Float32x4::Elem *data); +template JSObject *js::CreateSimd(JSContext *cx, Int32x4::Elem *data); namespace js { template @@ -608,7 +608,7 @@ CoercedFunc(JSContext *cx, unsigned argc, Value *vp) } RetElem *coercedResult = reinterpret_cast(result); - RootedObject obj(cx, Create(cx, coercedResult)); + RootedObject obj(cx, CreateSimd(cx, coercedResult)); if (!obj) return false; @@ -653,7 +653,7 @@ FuncWith(JSContext *cx, unsigned argc, Value *vp) result[i] = OpWith::apply(i, withAsBool, val[i]); } - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -712,7 +712,7 @@ FuncShuffle(JSContext *cx, unsigned argc, Value *vp) result[i] = val2[(maskArg >> (i * SELECT_SHIFT)) & SELECT_MASK]; } - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -740,7 +740,7 @@ Int32x4BinaryScalar(JSContext *cx, unsigned argc, Value *vp) for (unsigned i = 0; i < 4; i++) result[i] = Op::apply(val[i], bits); - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -764,7 +764,7 @@ FuncConvert(JSContext *cx, unsigned argc, Value *vp) for (unsigned i = 0; i < Vret::lanes; i++) result[i] = RetElem(val[i]); - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -783,7 +783,7 @@ FuncConvertBits(JSContext *cx, unsigned argc, Value *vp) return ErrorBadArgs(cx); RetElem *val = TypedObjectMemory(args[0]); - RootedObject obj(cx, Create(cx, val)); + RootedObject obj(cx, CreateSimd(cx, val)); if (!obj) return false; @@ -805,7 +805,7 @@ FuncZero(JSContext *cx, unsigned argc, Value *vp) for (unsigned i = 0; i < Vret::lanes; i++) result[i] = RetElem(0); - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -831,7 +831,7 @@ FuncSplat(JSContext *cx, unsigned argc, Value *vp) for (unsigned i = 0; i < Vret::lanes; i++) result[i] = arg; - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -854,7 +854,7 @@ Int32x4Bool(JSContext *cx, unsigned argc, Value *vp) for (unsigned i = 0; i < Int32x4::lanes; i++) result[i] = args[i].toBoolean() ? 0xFFFFFFFF : 0x0; - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -882,7 +882,7 @@ Float32x4Clamp(JSContext *cx, unsigned argc, Value *vp) result[i] = result[i] > upperLimit[i] ? upperLimit[i] : result[i]; } - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; @@ -917,7 +917,7 @@ Int32x4Select(JSContext *cx, unsigned argc, Value *vp) orInt[i] = Or::apply(tr[i], fr[i]); float *result = reinterpret_cast(orInt); - RootedObject obj(cx, Create(cx, result)); + RootedObject obj(cx, CreateSimd(cx, result)); if (!obj) return false; diff --git a/js/src/builtin/SIMD.h b/js/src/builtin/SIMD.h index a8f13d83ba92..9bebfac9b378 100644 --- a/js/src/builtin/SIMD.h +++ b/js/src/builtin/SIMD.h @@ -164,7 +164,7 @@ struct Int32x4 { }; template -JSObject *Create(JSContext *cx, typename V::Elem *data); +JSObject *CreateSimd(JSContext *cx, typename V::Elem *data); #define DECLARE_SIMD_FLOAT32X4_FUNCTION(Name, Func, Operands, Flags, MIRId) \ extern bool \ diff --git a/js/src/gc/GCRuntime.h b/js/src/gc/GCRuntime.h index 4bded5b5c09a..4a8dee8fc4ea 100644 --- a/js/src/gc/GCRuntime.h +++ b/js/src/gc/GCRuntime.h @@ -274,7 +274,8 @@ class GCRuntime bool triggerGC(JS::gcreason::Reason reason); bool triggerZoneGC(Zone *zone, JS::gcreason::Reason reason); - void maybeGC(Zone *zone); + bool maybeGC(Zone *zone); + void maybePeriodicFullGC(); void minorGC(JS::gcreason::Reason reason); void minorGC(JSContext *cx, JS::gcreason::Reason reason); void gcIfNeeded(JSContext *cx); @@ -487,7 +488,7 @@ class GCRuntime void getNextZoneGroup(); void endMarkingZoneGroup(); void beginSweepingZoneGroup(); - bool releaseObservedTypes(); + bool shouldReleaseObservedTypes(); void endSweepingZoneGroup(); bool sweepPhase(SliceBudget &sliceBudget); void endSweepPhase(JSGCInvocationKind gckind, bool lastGC); @@ -567,7 +568,6 @@ class GCRuntime bool chunkAllocationSinceLastGC; int64_t nextFullGCTime; int64_t lastGCTime; - int64_t jitReleaseTime; JSGCMode mode; @@ -589,6 +589,12 @@ class GCRuntime */ volatile uintptr_t isNeeded; + /* Incremented at the start of every major GC. */ + uint64_t majorGCNumber; + + /* The major GC number at which to release observed type information. */ + uint64_t jitReleaseNumber; + /* Incremented on every GC slice. */ uint64_t number; @@ -624,6 +630,9 @@ class GCRuntime /* Whether any sweeping will take place in the separate GC helper thread. */ bool sweepOnBackgroundThread; + /* Whether observed type information is being released in the current GC. */ + bool releaseObservedTypes; + /* Whether any black->gray edges were found during marking. */ bool foundBlackGrayEdges; diff --git a/js/src/jit-test/tests/ion/bug977966.js b/js/src/jit-test/tests/ion/bug977966.js new file mode 100644 index 000000000000..d4948d5cb342 --- /dev/null +++ b/js/src/jit-test/tests/ion/bug977966.js @@ -0,0 +1,83 @@ +setJitCompilerOption("baseline.usecount.trigger", 10); +setJitCompilerOption("ion.usecount.trigger", 20); + +function join_check() { + var lengthWasCalled = false; + var obj = {"0": "", "1": ""}; + Object.defineProperty(obj, "length", { + get : function(){ lengthWasCalled = true; return 2; }, + enumerable : true, + configurable : true + }); + + var res = Array.prototype.join.call(obj, { toString: function () { + if (lengthWasCalled) + return "good"; + else + return "bad"; + }}) + + assertEq(res, "good"); +} +function split(i) { + var x = (i + "->" + i).split("->"); + assertEq(x[0], "" + i); + return i; +} + +function join(i) { + var x = [i, i].join("->"); + assertEq(x, i + "->" + i); + return i; +} + +function split_join(i) { + var x = (i + "-" + i).split("-").join("->"); + assertEq(x, i + "->" + i); + return i; +} + +function split_join_2(i) { + var x = (i + "-" + i).split("-"); + x.push("" + i); + var res = x.join("->"); + assertEq(res, i + "->" + i + "->" + i); + return i; +} + +function resumeHere() { bailout(); } + +function split_join_3(i) { + var x = (i + "-" + i).split("-"); + resumeHere(); + var res = x.join("->"); + assertEq(res, i + "->" + i); + return i; +} + +function trip(i) { + if (i == 99) + assertEq(myjoin.arguments[1][0], "" + i) +} + +function myjoin(i, x) { + trip(i); + return x.join("->"); +} + +function split_join_4(i) { + var x = (i + "-" + i).split("-"); + var res = myjoin(i, x); + assertEq(res, i + "->" + i); + return i; +} + +for (var i = 0; i < 100; ++i) { + join_check(i); + split(i); + join(i); + split_join(i); + split_join_2(i); + split_join_3(i); + split_join_4(i); +} diff --git a/js/src/jit/BaselineCompiler.cpp b/js/src/jit/BaselineCompiler.cpp index 32e5262ff1cc..c6d204537e49 100644 --- a/js/src/jit/BaselineCompiler.cpp +++ b/js/src/jit/BaselineCompiler.cpp @@ -13,6 +13,7 @@ #include "jit/IonAnalysis.h" #include "jit/IonLinker.h" #include "jit/IonSpewer.h" +#include "jit/JitcodeMap.h" #ifdef JS_ION_PERF # include "jit/PerfSpewer.h" #endif @@ -244,6 +245,21 @@ BaselineCompiler::compile() if (script->compartment()->debugMode()) baselineScript->setDebugMode(); + // Register a native => bytecode mapping entry for this script if needed. + if (cx->runtime()->jitRuntime()->isNativeToBytecodeMapEnabled(cx->runtime())) { + IonSpew(IonSpew_Profiling, "Added JitcodeGlobalEntry for baseline script %s:%d (%p)", + script->filename(), script->lineno(), baselineScript); + JitcodeGlobalEntry::BaselineEntry entry; + entry.init(code->raw(), code->raw() + code->instructionsSize(), script); + + JitcodeGlobalTable *globalTable = cx->runtime()->jitRuntime()->getJitcodeGlobalTable(); + if (!globalTable->addEntry(entry)) + return Method_Error; + + // Mark the jitcode as having a bytecode map. + code->setHasBytecodeMap(); + } + script->setBaselineScript(cx, baselineScript); return Method_Compiled; diff --git a/js/src/jit/BaselineDebugModeOSR.cpp b/js/src/jit/BaselineDebugModeOSR.cpp index a132e1d55362..ee11e050b29d 100644 --- a/js/src/jit/BaselineDebugModeOSR.cpp +++ b/js/src/jit/BaselineDebugModeOSR.cpp @@ -9,6 +9,8 @@ #include "mozilla/DebugOnly.h" #include "jit/IonLinker.h" + +#include "jit/JitcodeMap.h" #include "jit/PerfSpewer.h" #include "jit/IonFrames-inl.h" @@ -659,6 +661,11 @@ jit::RecompileOnStackBaselineScriptsForDebugMode(JSContext *cx, JSCompartment *c MinorGC(cx->runtime(), JS::gcreason::EVICT_NURSERY); #endif + // When the profiler is enabled, we need to suppress sampling from here until + // the end of the function, since the basline jit scripts are in a state of + // flux. + AutoSuppressProfilerSampling suppressProfilerSampling(cx); + // Try to recompile all the scripts. If we encounter an error, we need to // roll back as if none of the compilations happened, so that we don't // crash. diff --git a/js/src/jit/BaselineJIT.cpp b/js/src/jit/BaselineJIT.cpp index abf08c9f9552..3b08c35b12e6 100644 --- a/js/src/jit/BaselineJIT.cpp +++ b/js/src/jit/BaselineJIT.cpp @@ -446,6 +446,7 @@ BaselineScript::Destroy(FreeOp *fop, BaselineScript *script) */ JS_ASSERT(fop->runtime()->gc.nursery.isEmpty()); #endif + fop->delete_(script); } @@ -690,6 +691,27 @@ BaselineScript::nativeCodeForPC(JSScript *script, jsbytecode *pc, PCMappingSlotI jsbytecode * BaselineScript::pcForReturnOffset(JSScript *script, uint32_t nativeOffset) +{ + return pcForNativeOffset(script, nativeOffset, true); +} + +jsbytecode * +BaselineScript::pcForReturnAddress(JSScript *script, uint8_t *nativeAddress) +{ + JS_ASSERT(script->baselineScript() == this); + JS_ASSERT(nativeAddress >= method_->raw()); + JS_ASSERT(nativeAddress < method_->raw() + method_->instructionsSize()); + return pcForReturnOffset(script, uint32_t(nativeAddress - method_->raw())); +} + +jsbytecode * +BaselineScript::pcForNativeOffset(JSScript *script, uint32_t nativeOffset) +{ + return pcForNativeOffset(script, nativeOffset, false); +} + +jsbytecode * +BaselineScript::pcForNativeOffset(JSScript *script, uint32_t nativeOffset, bool isReturn) { JS_ASSERT(script->baselineScript() == this); JS_ASSERT(nativeOffset < method_->instructionsSize()); @@ -707,14 +729,19 @@ BaselineScript::pcForReturnOffset(JSScript *script, uint32_t nativeOffset) i--; PCMappingIndexEntry &entry = pcMappingIndexEntry(i); - JS_ASSERT(nativeOffset >= entry.nativeOffset); + JS_ASSERT_IF(isReturn, nativeOffset >= entry.nativeOffset); CompactBufferReader reader(pcMappingReader(i)); jsbytecode *curPC = script->offsetToPC(entry.pcOffset); uint32_t curNativeOffset = entry.nativeOffset; JS_ASSERT(script->containsPC(curPC)); - JS_ASSERT(curNativeOffset <= nativeOffset); + JS_ASSERT_IF(isReturn, nativeOffset >= curNativeOffset); + + // In the raw native-lookup case, the native code address can occur + // before the start of ops. Associate those with bytecode offset 0. + if (!isReturn && (curNativeOffset > nativeOffset)) + return script->code(); while (true) { // If the high bit is set, the native offset relative to the @@ -723,22 +750,28 @@ BaselineScript::pcForReturnOffset(JSScript *script, uint32_t nativeOffset) if (b & 0x80) curNativeOffset += reader.readUnsigned(); - if (curNativeOffset == nativeOffset) + if (isReturn ? (nativeOffset == curNativeOffset) : (nativeOffset <= curNativeOffset)) + return curPC; + + // If this is a raw native lookup (not jsop return addresses), then + // the native address may lie in-between the last delta-entry in + // a pcMappingIndexEntry, and the next pcMappingIndexEntry. + if (!isReturn && !reader.more()) return curPC; curPC += GetBytecodeLength(curPC); } - MOZ_ASSUME_UNREACHABLE("Invalid pc"); + MOZ_ASSUME_UNREACHABLE("Bad baseline jitcode address"); } jsbytecode * -BaselineScript::pcForReturnAddress(JSScript *script, uint8_t *nativeAddress) +BaselineScript::pcForNativeAddress(JSScript *script, uint8_t *nativeAddress) { JS_ASSERT(script->baselineScript() == this); JS_ASSERT(nativeAddress >= method_->raw()); JS_ASSERT(nativeAddress < method_->raw() + method_->instructionsSize()); - return pcForReturnOffset(script, uint32_t(nativeAddress - method_->raw())); + return pcForNativeOffset(script, uint32_t(nativeAddress - method_->raw())); } void diff --git a/js/src/jit/BaselineJIT.h b/js/src/jit/BaselineJIT.h index 8086dc195ce1..7fe06e2e5ee3 100644 --- a/js/src/jit/BaselineJIT.h +++ b/js/src/jit/BaselineJIT.h @@ -315,9 +315,17 @@ struct BaselineScript void copyPCMappingEntries(const CompactBufferWriter &entries); uint8_t *nativeCodeForPC(JSScript *script, jsbytecode *pc, PCMappingSlotInfo *slotInfo = nullptr); + jsbytecode *pcForReturnOffset(JSScript *script, uint32_t nativeOffset); jsbytecode *pcForReturnAddress(JSScript *script, uint8_t *nativeAddress); + jsbytecode *pcForNativeAddress(JSScript *script, uint8_t *nativeAddress); + jsbytecode *pcForNativeOffset(JSScript *script, uint32_t nativeOffset); + + private: + jsbytecode *pcForNativeOffset(JSScript *script, uint32_t nativeOffset, bool isReturn); + + public: // Toggle debug traps (used for breakpoints and step mode) in the script. // If |pc| is nullptr, toggle traps for all ops in the script. Else, only // toggle traps at |pc|. diff --git a/js/src/jit/CodeGenerator.cpp b/js/src/jit/CodeGenerator.cpp index 39eb80d55fd8..4d497f34c25b 100644 --- a/js/src/jit/CodeGenerator.cpp +++ b/js/src/jit/CodeGenerator.cpp @@ -26,6 +26,7 @@ #include "jit/IonLinker.h" #include "jit/IonOptimizationLevels.h" #include "jit/IonSpewer.h" +#include "jit/JitcodeMap.h" #include "jit/Lowering.h" #include "jit/MIRGenerator.h" #include "jit/MoveEmitter.h" @@ -120,7 +121,7 @@ CodeGeneratorShared::addCache(LInstruction *lir, size_t cacheIndex) cache->setIdempotent(); OutOfLineUpdateCache *ool = new(alloc()) OutOfLineUpdateCache(lir, cacheIndex); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mir)) return false; // OOL-specific state depends on the type of cache. @@ -184,7 +185,7 @@ CodeGenerator::visitValueToInt32(LValueToInt32 *lir) Label fails; if (lir->mode() == LValueToInt32::TRUNCATE) { - OutOfLineCode *oolDouble = oolTruncateDouble(temp, output); + OutOfLineCode *oolDouble = oolTruncateDouble(temp, output, lir->mir()); if (!oolDouble) return false; @@ -691,7 +692,7 @@ CodeGenerator::visitTestOAndBranch(LTestOAndBranch *lir) "Objects which can't emulate undefined should have been constant-folded"); OutOfLineTestObject *ool = new(alloc()) OutOfLineTestObject(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; Label *truthy = getJumpLabelForBranch(lir->ifTruthy()); @@ -714,7 +715,7 @@ CodeGenerator::visitTestVAndBranch(LTestVAndBranch *lir) // object. if (lir->mir()->operandMightEmulateUndefined() && input->mightBeType(MIRType_Object)) { ool = new(alloc()) OutOfLineTestObject(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; } @@ -1317,7 +1318,7 @@ bool CodeGenerator::visitInterruptCheckImplicit(LInterruptCheckImplicit *lir) { OutOfLineInterruptCheckImplicit *ool = new(alloc()) OutOfLineInterruptCheckImplicit(current, lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; lir->setOolEntry(ool->entry()); @@ -1408,6 +1409,15 @@ CodeGenerator::visitStart(LStart *lir) return true; } +bool +CodeGenerator::visitPcOffset(LPcOffset *lir) +{ + if (!addNativeToBytecodeEntry(lir->mir()->trackedSite())) + return false; + + return true; +} + bool CodeGenerator::visitReturn(LReturn *lir) { @@ -2043,7 +2053,7 @@ CodeGenerator::visitPostWriteBarrierO(LPostWriteBarrierO *lir) { #ifdef JSGC_GENERATIONAL OutOfLineCallPostWriteBarrier *ool = new(alloc()) OutOfLineCallPostWriteBarrier(lir, lir->object()); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; Register temp = ToTempRegisterOrInvalid(lir->temp()); @@ -2069,7 +2079,7 @@ CodeGenerator::visitPostWriteBarrierV(LPostWriteBarrierV *lir) { #ifdef JSGC_GENERATIONAL OutOfLineCallPostWriteBarrier *ool = new(alloc()) OutOfLineCallPostWriteBarrier(lir, lir->object()); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; Register temp = ToTempRegisterOrInvalid(lir->temp()); @@ -2749,7 +2759,6 @@ CodeGenerator::visitArraySplice(LArraySplice *lir) return callVM(ArraySpliceDenseInfo, lir); } - bool CodeGenerator::visitBail(LBail *lir) { @@ -2985,7 +2994,7 @@ CodeGenerator::visitCheckOverRecursed(LCheckOverRecursed *lir) const void *limitAddr = GetIonContext()->runtime->addressOfJitStackLimit(); CheckOverRecursedFailure *ool = new(alloc()) CheckOverRecursedFailure(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; // Conditional forward (unlikely) branch to failure. @@ -3072,7 +3081,7 @@ CodeGenerator::visitCheckOverRecursedPar(LCheckOverRecursedPar *lir) // Conditional forward (unlikely) branch to failure. CheckOverRecursedFailure *ool = new(alloc()) CheckOverRecursedFailure(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; masm.branchPtr(Assembler::BelowOrEqual, StackPointer, tempReg, ool->entry()); @@ -3405,6 +3414,14 @@ CodeGenerator::generateBody() resetOsiPointRegs(iter->safepoint()); #endif + if (iter->mirRaw()) { + // Only add instructions that have a tracked inline script tree. + if (iter->mirRaw()->trackedSite().hasTree()) { + if (!addNativeToBytecodeEntry(iter->mirRaw()->trackedSite())) + return false; + } + } + if (!iter->accept(this)) return false; @@ -3540,7 +3557,7 @@ CodeGenerator::visitNewArray(LNewArray *lir) return visitNewArrayCallVM(lir); OutOfLineNewArray *ool = new(alloc()) OutOfLineNewArray(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; masm.createGCObject(objReg, tempReg, templateObject, lir->mir()->initialHeap(), ool->entry()); @@ -3710,7 +3727,7 @@ CodeGenerator::visitNewObject(LNewObject *lir) return visitNewObjectVMCall(lir); OutOfLineNewObject *ool = new(alloc()) OutOfLineNewObject(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; bool initFixedSlots = ShouldInitFixedSlots(lir, templateObject); @@ -3923,6 +3940,9 @@ bool CodeGenerator::emitAllocateGCThingPar(LInstruction *lir, Register objReg, Register cxReg, Register tempReg1, Register tempReg2, JSObject *templateObj) { + JS_ASSERT(lir->mirRaw()); + JS_ASSERT(lir->mirRaw()->isInstruction()); + gc::AllocKind allocKind = templateObj->tenuredGetAllocKind(); #ifdef JSGC_FJGENERATIONAL OutOfLineCode *ool = oolCallVM(NewGCThingParInfo, lir, @@ -3931,7 +3951,7 @@ CodeGenerator::emitAllocateGCThingPar(LInstruction *lir, Register objReg, Regist return false; #else OutOfLineNewGCThingPar *ool = new(alloc()) OutOfLineNewGCThingPar(lir, allocKind, objReg, cxReg); - if (!ool || !addOutOfLineCode(ool)) + if (!ool || !addOutOfLineCode(ool, lir->mirRaw()->toInstruction())) return false; #endif @@ -4826,7 +4846,7 @@ CodeGenerator::visitIsNullOrLikeUndefined(LIsNullOrLikeUndefined *lir) Label *notNullOrLikeUndefined; if (lir->mir()->operandMightEmulateUndefined()) { ool = new(alloc()) OutOfLineTestObjectWithLabels(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; nullOrLikeUndefined = ool->label1(); notNullOrLikeUndefined = ool->label2(); @@ -4910,7 +4930,7 @@ CodeGenerator::visitIsNullOrLikeUndefinedAndBranch(LIsNullOrLikeUndefinedAndBran OutOfLineTestObject *ool = nullptr; if (lir->cmpMir()->operandMightEmulateUndefined()) { ool = new(alloc()) OutOfLineTestObject(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->cmpMir())) return false; } @@ -4959,7 +4979,7 @@ CodeGenerator::visitEmulatesUndefined(LEmulatesUndefined *lir) MOZ_ASSERT(op == JSOP_EQ || op == JSOP_NE, "Strict equality should have been folded"); OutOfLineTestObjectWithLabels *ool = new(alloc()) OutOfLineTestObjectWithLabels(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; Label *emulatesUndefined = ool->label1(); @@ -4993,7 +5013,7 @@ CodeGenerator::visitEmulatesUndefinedAndBranch(LEmulatesUndefinedAndBranch *lir) MOZ_ASSERT(op == JSOP_EQ || op == JSOP_NE, "Strict equality should have been folded"); OutOfLineTestObject *ool = new(alloc()) OutOfLineTestObject(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->cmpMir())) return false; Label *equal; @@ -5497,7 +5517,7 @@ CodeGenerator::visitNotO(LNotO *lir) "This should be constant-folded if the object can't emulate undefined."); OutOfLineTestObjectWithLabels *ool = new(alloc()) OutOfLineTestObjectWithLabels(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; Label *ifEmulatesUndefined = ool->label1(); @@ -5536,7 +5556,7 @@ CodeGenerator::visitNotV(LNotV *lir) // object. if (lir->mir()->operandMightEmulateUndefined() && operand->mightBeType(MIRType_Object)) { ool = new(alloc()) OutOfLineTestObjectWithLabels(); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; ifTruthy = ool->label1(); ifFalsy = ool->label2(); @@ -5758,7 +5778,7 @@ bool CodeGenerator::visitStoreElementHoleT(LStoreElementHoleT *lir) { OutOfLineStoreElementHole *ool = new(alloc()) OutOfLineStoreElementHole(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; Register elements = ToRegister(lir->elements()); @@ -5783,7 +5803,7 @@ bool CodeGenerator::visitStoreElementHoleV(LStoreElementHoleV *lir) { OutOfLineStoreElementHole *ool = new(alloc()) OutOfLineStoreElementHole(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; Register elements = ToRegister(lir->elements()); @@ -6110,6 +6130,18 @@ CodeGenerator::visitArrayConcat(LArrayConcat *lir) return callVM(ArrayConcatDenseInfo, lir); } +typedef JSString *(*ArrayJoinFn)(JSContext *, HandleObject, HandleString); +static const VMFunction ArrayJoinInfo = FunctionInfo(jit::ArrayJoin); + +bool +CodeGenerator::visitArrayJoin(LArrayJoin *lir) +{ + pushArg(ToRegister(lir->separator())); + pushArg(ToRegister(lir->array())); + + return callVM(ArrayJoinInfo, lir); +} + typedef JSObject *(*GetIteratorObjectFn)(JSContext *, HandleObject, uint32_t); static const VMFunction GetIteratorObjectInfo = FunctionInfo(GetIteratorObject); @@ -6564,6 +6596,13 @@ CodeGenerator::generate() gen->info().script()->filename(), gen->info().script()->lineno()); + // Initialize native code table with an entry to the start of + // top-level script. + InlineScriptTree *tree = gen->info().inlineScriptTree(); + jsbytecode *startPC = tree->script()->code(); + if (!addNativeToBytecodeEntry(BytecodeSite(tree, startPC))) + return false; + if (!snapshots_.init()) return false; @@ -6618,22 +6657,74 @@ CodeGenerator::generate() if (!generatePrologue()) return false; + + // Reset native => bytecode map table with top-level script and startPc. + if (!addNativeToBytecodeEntry(BytecodeSite(tree, startPC))) + return false; + if (!generateBody()) return false; + + // Reset native => bytecode map table with top-level script and startPc. + if (!addNativeToBytecodeEntry(BytecodeSite(tree, startPC))) + return false; + if (!generateEpilogue()) return false; + + // Reset native => bytecode map table with top-level script and startPc. + if (!addNativeToBytecodeEntry(BytecodeSite(tree, startPC))) + return false; + if (!generateInvalidateEpilogue()) return false; #if defined(JS_ION_PERF) // Note the end of the inline code and start of the OOL code. perfSpewer_.noteEndInlineCode(masm); #endif + + // native => bytecode entries for OOL code will be added + // by CodeGeneratorShared::generateOutOfLineCode if (!generateOutOfLineCode()) return false; + // Add terminal entry. + if (!addNativeToBytecodeEntry(BytecodeSite(tree, startPC))) + return false; + + // Dump Native to bytecode entries to spew. + dumpNativeToBytecodeEntries(); + return !masm.oom(); } +struct AutoDiscardIonCode +{ + JSContext *cx; + types::RecompileInfo *recompileInfo; + IonScript *ionScript; + bool keep; + + AutoDiscardIonCode(JSContext *cx, types::RecompileInfo *recompileInfo) + : cx(cx), recompileInfo(recompileInfo), ionScript(nullptr), keep(false) {} + + ~AutoDiscardIonCode() { + if (keep) + return; + + // Use js_free instead of IonScript::Destroy: the cache list and + // backedge list are still uninitialized. + if (ionScript) + js_free(ionScript); + + recompileInfo->compilerOutput(cx->zone()->types)->invalidate(); + } + + void keepIonCode() { + keep = true; + } +}; + bool CodeGenerator::link(JSContext *cx, types::CompilerConstraintList *constraints) { @@ -6678,6 +6769,8 @@ CodeGenerator::link(JSContext *cx, types::CompilerConstraintList *constraints) if (executionMode == ParallelExecution) AddPossibleCallees(cx, graph.mir(), callTargets); + AutoDiscardIonCode discardIonCode(cx, &recompileInfo); + IonScript *ionScript = IonScript::New(cx, recompileInfo, graph.totalSlotCount(), scriptFrameSize, @@ -6687,10 +6780,9 @@ CodeGenerator::link(JSContext *cx, types::CompilerConstraintList *constraints) cacheList_.length(), runtimeData_.length(), safepoints_.size(), callTargets.length(), patchableBackedges_.length(), optimizationLevel); - if (!ionScript) { - recompileInfo.compilerOutput(cx->zone()->types)->invalidate(); + if (!ionScript) return false; - } + discardIonCode.ionScript = ionScript; // Lock the runtime against interrupt callbacks during the link. // We don't want an interrupt request to protect the code for the script @@ -6714,23 +6806,38 @@ CodeGenerator::link(JSContext *cx, types::CompilerConstraintList *constraints) JitCode *code = (executionMode == SequentialExecution) ? linker.newCodeForIonScript(cx) : linker.newCode(cx, ION_CODE); - if (!code) { - // Use js_free instead of IonScript::Destroy: the cache list and - // backedge list are still uninitialized. - js_free(ionScript); - recompileInfo.compilerOutput(cx->zone()->types)->invalidate(); + if (!code) return false; + + // Encode native to bytecode map if profiling is enabled. + if (isNativeToBytecodeMapEnabled()) { + // Generate native-to-bytecode main table. + if (!generateCompactNativeToBytecodeMap(cx, code)) + return false; + + uint8_t *ionTableAddr = ((uint8_t *) nativeToBytecodeMap_) + nativeToBytecodeTableOffset_; + JitcodeIonTable *ionTable = (JitcodeIonTable *) ionTableAddr; + + // Construct the IonEntry that will go into the global table. + JitcodeGlobalEntry::IonEntry entry; + if (!ionTable->makeIonEntry(cx, code, nativeToBytecodeScriptListLength_, + nativeToBytecodeScriptList_, entry)) + { + return false; + } + + // Add entry to the global table. + JitcodeGlobalTable *globalTable = cx->runtime()->jitRuntime()->getJitcodeGlobalTable(); + if (!globalTable->addEntry(entry)) { + // Memory may have been allocated for the entry. + entry.destroy(); + return false; + } + + // Mark the jitcode as having a bytecode map. + code->setHasBytecodeMap(); } - ionScript->setMethod(code); - ionScript->setSkipArgCheckEntryOffset(getSkipArgCheckEntryOffset()); - - // If SPS is enabled, mark IonScript as having been instrumented with SPS - if (sps_.enabled()) - ionScript->setHasSPSInstrumentation(); - - SetIonScript(script, executionMode, ionScript); - if (cx->runtime()->spsProfiler.enabled()) { const char *filename = script->filename(); if (filename == nullptr) @@ -6744,6 +6851,15 @@ CodeGenerator::link(JSContext *cx, types::CompilerConstraintList *constraints) js_free(buf); } + ionScript->setMethod(code); + ionScript->setSkipArgCheckEntryOffset(getSkipArgCheckEntryOffset()); + + // If SPS is enabled, mark IonScript as having been instrumented with SPS + if (sps_.enabled()) + ionScript->setHasSPSInstrumentation(); + + SetIonScript(script, executionMode, ionScript); + // In parallel execution mode, when we first compile a script, we // don't know that its potential callees are compiled, so set a // flag warning that the callees may not be fully compiled. @@ -6844,6 +6960,9 @@ CodeGenerator::link(JSContext *cx, types::CompilerConstraintList *constraints) if (IonScriptCounts *counts = extractScriptCounts()) script->addIonCounts(counts); + // Make sure that AutoDiscardIonCode does not free the relevant info. + discardIonCode.keepIonCode(); + return true; } @@ -6875,7 +6994,7 @@ CodeGenerator::visitUnboxFloatingPoint(LUnboxFloatingPoint *lir) // Out-of-line path to convert int32 to double or bailout // if this instruction is fallible. OutOfLineUnboxFloatingPoint *ool = new(alloc()) OutOfLineUnboxFloatingPoint(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; FloatRegister resultReg = ToFloatRegister(result); @@ -7673,7 +7792,7 @@ CodeGenerator::visitTypeOfV(LTypeOfV *lir) // The input may be a callable object (result is "function") or may // emulate undefined (result is "undefined"). Use an OOL path. ool = new(alloc()) OutOfLineTypeOfV(lir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mir())) return false; masm.branchTestObject(Assembler::Equal, tag, ool->entry()); diff --git a/js/src/jit/CodeGenerator.h b/js/src/jit/CodeGenerator.h index 034fa4ac90c9..3f3319da2dcc 100644 --- a/js/src/jit/CodeGenerator.h +++ b/js/src/jit/CodeGenerator.h @@ -68,6 +68,7 @@ class CodeGenerator : public CodeGeneratorSpecific bool visitParameter(LParameter *lir); bool visitCallee(LCallee *lir); bool visitStart(LStart *lir); + bool visitPcOffset(LPcOffset *lir); bool visitReturn(LReturn *ret); bool visitDefVar(LDefVar *lir); bool visitDefFun(LDefFun *lir); @@ -251,6 +252,7 @@ class CodeGenerator : public CodeGeneratorSpecific bool visitArrayPushV(LArrayPushV *lir); bool visitArrayPushT(LArrayPushT *lir); bool visitArrayConcat(LArrayConcat *lir); + bool visitArrayJoin(LArrayJoin *lir); bool visitLoadTypedArrayElement(LLoadTypedArrayElement *lir); bool visitLoadTypedArrayElementHole(LLoadTypedArrayElementHole *lir); bool visitStoreTypedArrayElement(LStoreTypedArrayElement *lir); diff --git a/js/src/jit/CompactBuffer.h b/js/src/jit/CompactBuffer.h index 9c9e07158be3..b12fce286cfa 100644 --- a/js/src/jit/CompactBuffer.h +++ b/js/src/jit/CompactBuffer.h @@ -68,6 +68,11 @@ class CompactBufferReader uint32_t b1 = readByte(); return b0 | (b1 << 8); } + uint32_t readNativeEndianUint32_t() { + // Must be at 4-byte boundary + JS_ASSERT(uintptr_t(buffer_) % sizeof(uint32_t) == 0); + return *reinterpret_cast(buffer_); + } uint32_t readUnsigned() { return readVariableLength(); } @@ -93,6 +98,10 @@ class CompactBufferReader MOZ_ASSERT(start < end_); MOZ_ASSERT(buffer_ < end_); } + + const uint8_t *currentPosition() const { + return buffer_; + } }; class CompactBufferWriter @@ -140,6 +149,15 @@ class CompactBufferWriter writeByte(value & 0xFF); writeByte(value >> 8); } + void writeNativeEndianUint32_t(uint32_t value) { + // Must be at 4-byte boundary + JS_ASSERT(length() % sizeof(uint32_t) == 0); + writeFixedUint32_t(0); + if (oom()) + return; + uint8_t *endPtr = buffer() + length(); + reinterpret_cast(endPtr)[-1] = value; + } size_t length() const { return buffer_.length(); } diff --git a/js/src/jit/CompileInfo.h b/js/src/jit/CompileInfo.h index e6250b17f708..a9eea178eafc 100644 --- a/js/src/jit/CompileInfo.h +++ b/js/src/jit/CompileInfo.h @@ -83,6 +83,9 @@ class InlineScriptTree { bool isOutermostCaller() const { return caller_ == nullptr; } + bool hasCaller() const { + return caller_ != nullptr; + } InlineScriptTree *outermostCaller() { if (isOutermostCaller()) return this; @@ -97,12 +100,27 @@ class InlineScriptTree { return script_; } - InlineScriptTree *children() const { + bool hasChildren() const { + return children_ != nullptr; + } + InlineScriptTree *firstChild() const { + JS_ASSERT(hasChildren()); return children_; } + + bool hasNextCallee() const { + return nextCallee_ != nullptr; + } InlineScriptTree *nextCallee() const { + JS_ASSERT(hasNextCallee()); return nextCallee_; } + + unsigned depth() const { + if (isOutermostCaller()) + return 1; + return 1 + caller_->depth(); + } }; class BytecodeSite { @@ -119,7 +137,14 @@ class BytecodeSite { BytecodeSite(InlineScriptTree *tree, jsbytecode *pc) : tree_(tree), pc_(pc) - {} + { + JS_ASSERT(tree_ != nullptr); + JS_ASSERT(pc_ != nullptr); + } + + bool hasTree() const { + return tree_ != nullptr; + } InlineScriptTree *tree() const { return tree_; @@ -128,6 +153,10 @@ class BytecodeSite { jsbytecode *pc() const { return pc_; } + + JSScript *script() const { + return tree_ ? tree_->script() : nullptr; + } }; diff --git a/js/src/jit/Ion.cpp b/js/src/jit/Ion.cpp index a00824ddd323..a5d8c44f1849 100644 --- a/js/src/jit/Ion.cpp +++ b/js/src/jit/Ion.cpp @@ -27,6 +27,7 @@ #include "jit/IonBuilder.h" #include "jit/IonOptimizationLevels.h" #include "jit/IonSpewer.h" +#include "jit/JitcodeMap.h" #include "jit/JitCommon.h" #include "jit/JitCompartment.h" #include "jit/LICM.h" @@ -165,7 +166,8 @@ JitRuntime::JitRuntime() functionWrappers_(nullptr), osrTempData_(nullptr), ionCodeProtected_(false), - ionReturnOverride_(MagicValue(JS_ARG_POISON)) + ionReturnOverride_(MagicValue(JS_ARG_POISON)), + jitcodeGlobalTable_(nullptr) { } @@ -177,6 +179,10 @@ JitRuntime::~JitRuntime() // Note: The interrupt lock is not taken here, as JitRuntime is only // destroyed along with its containing JSRuntime. js_delete(ionAlloc_); + + // By this point, the jitcode global table should be empty. + JS_ASSERT_IF(jitcodeGlobalTable_, jitcodeGlobalTable_->empty()); + js_delete(jitcodeGlobalTable_); } bool @@ -289,6 +295,10 @@ JitRuntime::initialize(JSContext *cx) return false; } + jitcodeGlobalTable_ = cx->new_(); + if (!jitcodeGlobalTable_) + return false; + return true; } @@ -761,6 +771,12 @@ JitCode::finalize(FreeOp *fop) // to read the contents of the pool we are releasing references in. JS_ASSERT(fop->runtime()->currentThreadOwnsInterruptLock()); + // If this jitcode has a bytecode map, de-register it. + if (hasBytecodeMap_) { + JS_ASSERT(fop->runtime()->jitRuntime()->hasJitcodeGlobalTable()); + fop->runtime()->jitRuntime()->getJitcodeGlobalTable()->removeEntry(raw()); + } + // Buffer can be freed at any time hereafter. Catch use-after-free bugs. // Don't do this if the Ion code is protected, as the signal handler will // deadlock trying to reacquire the interrupt lock. diff --git a/js/src/jit/IonBuilder.cpp b/js/src/jit/IonBuilder.cpp index 17bb9cebde04..c2642e4aba7c 100644 --- a/js/src/jit/IonBuilder.cpp +++ b/js/src/jit/IonBuilder.cpp @@ -1277,6 +1277,9 @@ IonBuilder::traverseBytecode() } #endif + if (isNativeToBytecodeMapEnabled()) + current->add(MPcOffset::New(alloc())); + // Nothing in inspectOpcode() is allowed to advance the pc. JSOp op = JSOp(*pc); if (!inspectOpcode(op)) diff --git a/js/src/jit/IonBuilder.h b/js/src/jit/IonBuilder.h index 46cf3198e78f..e1f23366faeb 100644 --- a/js/src/jit/IonBuilder.h +++ b/js/src/jit/IonBuilder.h @@ -675,6 +675,7 @@ class IonBuilder : public MIRGenerator InliningStatus inlineArrayPopShift(CallInfo &callInfo, MArrayPopShift::Mode mode); InliningStatus inlineArrayPush(CallInfo &callInfo); InliningStatus inlineArrayConcat(CallInfo &callInfo); + InliningStatus inlineArrayJoin(CallInfo &callInfo); InliningStatus inlineArraySplice(CallInfo &callInfo); // Math natives. diff --git a/js/src/jit/IonCaches.cpp b/js/src/jit/IonCaches.cpp index 8bff6005ae83..a752ffccf1ba 100644 --- a/js/src/jit/IonCaches.cpp +++ b/js/src/jit/IonCaches.cpp @@ -16,6 +16,7 @@ #include "jit/Ion.h" #include "jit/IonLinker.h" #include "jit/IonSpewer.h" +#include "jit/JitcodeMap.h" #include "jit/Lowering.h" #ifdef JS_ION_PERF # include "jit/PerfSpewer.h" @@ -429,6 +430,22 @@ IonCache::linkAndAttachStub(JSContext *cx, MacroAssembler &masm, StubAttacher &a attachStub(masm, attacher, code); + // Add entry to native => bytecode mapping for this stub if needed. + if (cx->runtime()->jitRuntime()->isNativeToBytecodeMapEnabled(cx->runtime())) { + JitcodeGlobalEntry::IonCacheEntry entry; + entry.init(code->raw(), code->raw() + code->instructionsSize(), rejoinAddress()); + + // Add entry to the global table. + JitcodeGlobalTable *globalTable = cx->runtime()->jitRuntime()->getJitcodeGlobalTable(); + if (!globalTable->addEntry(entry)) { + entry.destroy(); + return false; + } + + // Mark the jitcode as having a bytecode map. + code->setHasBytecodeMap(); + } + return true; } diff --git a/js/src/jit/IonCaches.h b/js/src/jit/IonCaches.h index a0a7868207d3..551b990be1bb 100644 --- a/js/src/jit/IonCaches.h +++ b/js/src/jit/IonCaches.h @@ -213,6 +213,9 @@ class IonCache profilerLeavePc_ = pc; } + // Get the address at which IC rejoins the mainline jitcode. + virtual void *rejoinAddress() = 0; + virtual void emitInitialJump(MacroAssembler &masm, AddCacheState &addState) = 0; virtual void bindInitialJump(MacroAssembler &masm, AddCacheState &addState) = 0; virtual void updateBaseAddress(JitCode *code, MacroAssembler &masm); @@ -398,6 +401,10 @@ class RepatchIonCache : public IonCache // Update the labels once the code is finalized. void updateBaseAddress(JitCode *code, MacroAssembler &masm); + + virtual void *rejoinAddress() MOZ_OVERRIDE { + return rejoinLabel().raw(); + } }; // @@ -496,6 +503,10 @@ class DispatchIonCache : public IonCache // Fix up the first stub pointer once the code is finalized. void updateBaseAddress(JitCode *code, MacroAssembler &masm); + + virtual void *rejoinAddress() MOZ_OVERRIDE { + return rejoinLabel_.raw(); + } }; // Define the cache kind and pre-declare data structures used for calling inline diff --git a/js/src/jit/IonCode.h b/js/src/jit/IonCode.h index 204fec3f2ad8..c64388e4cf78 100644 --- a/js/src/jit/IonCode.h +++ b/js/src/jit/IonCode.h @@ -44,6 +44,8 @@ class JitCode : public gc::BarrieredCell uint8_t kind_ : 3; // jit::CodeKind, for the memory reporters. bool invalidated_ : 1; // Whether the code object has been invalidated. // This is necessary to prevent GC tracing. + bool hasBytecodeMap_ : 1; // Whether the code object has been registered with + // native=>bytecode mapping tables. #if JS_BITS_PER_WORD == 32 // Ensure JitCode is gc::Cell aligned. @@ -66,7 +68,8 @@ class JitCode : public gc::BarrieredCell preBarrierTableBytes_(0), headerSize_(headerSize), kind_(kind), - invalidated_(false) + invalidated_(false), + hasBytecodeMap_(false) { MOZ_ASSERT(CodeKind(kind_) == kind); MOZ_ASSERT(headerSize_ == headerSize); @@ -89,6 +92,9 @@ class JitCode : public gc::BarrieredCell uint8_t *raw() const { return code_; } + uint8_t *rawEnd() const { + return code_ + insnSize_; + } size_t instructionsSize() const { return insnSize_; } @@ -98,6 +104,10 @@ class JitCode : public gc::BarrieredCell invalidated_ = true; } + void setHasBytecodeMap() { + hasBytecodeMap_ = true; + } + void togglePreBarriers(bool enabled); // If this JitCode object has been, effectively, corrupted due to diff --git a/js/src/jit/IonFrames.cpp b/js/src/jit/IonFrames.cpp index cca0f1880662..e22b7177b06c 100644 --- a/js/src/jit/IonFrames.cpp +++ b/js/src/jit/IonFrames.cpp @@ -19,6 +19,7 @@ #include "jit/Ion.h" #include "jit/IonMacroAssembler.h" #include "jit/IonSpewer.h" +#include "jit/JitcodeMap.h" #include "jit/JitCompartment.h" #include "jit/ParallelFunctions.h" #include "jit/PcScriptCache.h" @@ -112,6 +113,7 @@ JitFrameIterator::JitFrameIterator(IonJSFrameLayout *fp, ExecutionMode mode) mode_(mode), kind_(Kind_FrameIterator) { + verifyReturnAddressUsingNativeToBytecodeMap(); } IonBailoutIterator * @@ -328,6 +330,9 @@ JitFrameIterator::operator++() type_ = JitFrame_BaselineStub; returnAddressToFp_ = current()->returnAddress(); current_ = prev; + + verifyReturnAddressUsingNativeToBytecodeMap(); + return *this; } @@ -2229,6 +2234,80 @@ JitFrameIterator::dump() const fputc('\n', stderr); } +#ifdef DEBUG +bool +JitFrameIterator::verifyReturnAddressUsingNativeToBytecodeMap() +{ + JS_ASSERT(returnAddressToFp_ != nullptr); + + // Only handle Ion frames for now. + if (type_ != JitFrame_IonJS && type_ != JitFrame_BaselineJS) + return true; + + JSRuntime *rt = js::TlsPerThreadData.get()->runtimeIfOnOwnerThread(); + + // Don't verify on non-main-thread. + if (!rt) + return true; + + // Don't verify if sampling is being suppressed. + if (!rt->isProfilerSamplingEnabled()) + return true; + + if (rt->isHeapMinorCollecting()) + return true; + + JitRuntime *jitrt = rt->jitRuntime(); + + // Look up and print bytecode info for the native address. + JitcodeGlobalEntry entry; + if (!jitrt->getJitcodeGlobalTable()->lookup(returnAddressToFp_, &entry)) + return true; + + IonSpew(IonSpew_Profiling, "Found nativeToBytecode entry for %p: %p - %p", + returnAddressToFp_, entry.nativeStartAddr(), entry.nativeEndAddr()); + + JitcodeGlobalEntry::BytecodeLocationVector location; + uint32_t depth = UINT32_MAX; + if (!entry.callStackAtAddr(rt, returnAddressToFp_, location, &depth)) + return false; + JS_ASSERT(depth > 0 && depth != UINT32_MAX); + JS_ASSERT(location.length() == depth); + + IonSpew(IonSpew_Profiling, "Found bytecode location of depth %d:", depth); + for (size_t i = 0; i < location.length(); i++) { + IonSpew(IonSpew_Profiling, " %s:%d - %d", + location[i].script->filename(), location[i].script->lineno(), + (int) (location[i].pc - location[i].script->code())); + } + + if (type_ == JitFrame_IonJS) { + // Create an InlineFrameIterator here and verify the mapped info against the iterator info. + InlineFrameIterator inlineFrames(GetJSContextFromJitCode(), this); + for (size_t idx = 0; idx < location.length(); idx++) { + JS_ASSERT(idx < location.length()); + JS_ASSERT_IF(idx < location.length() - 1, inlineFrames.more()); + + IonSpew(IonSpew_Profiling, "Match %d: ION %s:%d(%d) vs N2B %s:%d(%d)", + (int)idx, + inlineFrames.script()->filename(), + inlineFrames.script()->lineno(), + inlineFrames.pc() - inlineFrames.script()->code(), + location[idx].script->filename(), + location[idx].script->lineno(), + location[idx].pc - location[idx].script->code()); + + JS_ASSERT(inlineFrames.script() == location[idx].script); + + if (inlineFrames.more()) + ++inlineFrames; + } + } + + return true; +} +#endif // DEBUG + IonJSFrameLayout * InvalidationBailoutStack::fp() const { diff --git a/js/src/jit/IonSpewer.cpp b/js/src/jit/IonSpewer.cpp index 3fba076e19fd..db0ef3253b92 100644 --- a/js/src/jit/IonSpewer.cpp +++ b/js/src/jit/IonSpewer.cpp @@ -250,6 +250,7 @@ jit::CheckLogging() " range Range Analysis\n" " unroll Loop unrolling\n" " logs C1 and JSON visualization logging\n" + " profiling Profiling-related information\n" " all Everything\n" "\n" " bl-aborts Baseline compiler abort messages\n" @@ -304,6 +305,8 @@ jit::CheckLogging() EnableChannel(IonSpew_CacheFlush); if (ContainsFlag(env, "logs")) EnableIonDebugLogging(); + if (ContainsFlag(env, "profiling")) + EnableChannel(IonSpew_Profiling); if (ContainsFlag(env, "all")) LoggingBits = uint32_t(-1); diff --git a/js/src/jit/IonSpewer.h b/js/src/jit/IonSpewer.h index 867003737e7c..26fd83f28ed8 100644 --- a/js/src/jit/IonSpewer.h +++ b/js/src/jit/IonSpewer.h @@ -56,6 +56,8 @@ namespace jit { _(Safepoints) \ /* Debug info about Pools*/ \ _(Pools) \ + /* Profiling-related information */ \ + _(Profiling) \ /* Debug info about the I$ */ \ _(CacheFlush) \ \ diff --git a/js/src/jit/IonTypes.h b/js/src/jit/IonTypes.h index f5aaf8c4f7ff..4bc547573aae 100644 --- a/js/src/jit/IonTypes.h +++ b/js/src/jit/IonTypes.h @@ -7,6 +7,7 @@ #ifndef jit_IonTypes_h #define jit_IonTypes_h +#include "mozilla/HashFunctions.h" #include "mozilla/TypedEnum.h" #include "jstypes.h" @@ -232,6 +233,111 @@ static const uint32_t VECTOR_SCALE_BITS = 2; static const uint32_t VECTOR_SCALE_SHIFT = ELEMENT_TYPE_BITS + ELEMENT_TYPE_SHIFT; static const uint32_t VECTOR_SCALE_MASK = (1 << VECTOR_SCALE_BITS) - 1; +class SimdConstant { + public: + enum Type { + Int32x4, + Float32x4, + Undefined = -1 + }; + + private: + Type type_; + union { + int32_t i32x4[4]; + float f32x4[4]; + } u; + + bool defined() const { + return type_ != Undefined; + } + + void fillInt32x4(int32_t x, int32_t y, int32_t z, int32_t w) + { + type_ = Int32x4; + u.i32x4[0] = x; + u.i32x4[1] = y; + u.i32x4[2] = z; + u.i32x4[3] = w; + } + + void fillFloat32x4(float x, float y, float z, float w) + { + type_ = Float32x4; + u.f32x4[0] = x; + u.f32x4[1] = y; + u.f32x4[2] = z; + u.f32x4[3] = w; + } + + public: + // Doesn't have a default constructor, as it would prevent it from being + // included in unions. + + static SimdConstant CreateX4(int32_t x, int32_t y, int32_t z, int32_t w) { + SimdConstant cst; + cst.fillInt32x4(x, y, z, w); + return cst; + } + static SimdConstant CreateX4(int32_t *array) { + SimdConstant cst; + cst.fillInt32x4(array[0], array[1], array[2], array[3]); + return cst; + } + static SimdConstant CreateX4(float x, float y, float z, float w) { + SimdConstant cst; + cst.fillFloat32x4(x, y, z, w); + return cst; + } + static SimdConstant CreateX4(float *array) { + SimdConstant cst; + cst.fillFloat32x4(array[0], array[1], array[2], array[3]); + return cst; + } + + uint32_t length() const { + JS_ASSERT(defined()); + switch(type_) { + case Int32x4: + case Float32x4: + return 4; + case Undefined: + break; + } + MOZ_CRASH("Unexpected SIMD kind"); + } + + Type type() const { + JS_ASSERT(defined()); + return type_; + } + + const int32_t *asInt32x4() const { + JS_ASSERT(defined() && type_ == Int32x4); + return u.i32x4; + } + const float *asFloat32x4() const { + JS_ASSERT(defined() && type_ == Float32x4); + return u.f32x4; + } + + bool operator==(const SimdConstant &rhs) const { + JS_ASSERT(defined() && rhs.defined()); + if (type() != rhs.type()) + return false; + return memcmp(&u, &rhs.u, sizeof(u)) == 0; + } + + // SimdConstant is a HashPolicy + typedef SimdConstant Lookup; + static HashNumber hash(const SimdConstant &val) { + return mozilla::HashBytes(&val.u, sizeof(SimdConstant)); + } + static bool match(const SimdConstant &lhs, const SimdConstant &rhs) { + return lhs == rhs; + } +}; + // The ordering of this enumeration is important: Anything < Value is a // specialized type. Furthermore, anything < String has trivial conversion to // a number. diff --git a/js/src/jit/JitCompartment.h b/js/src/jit/JitCompartment.h index 6a2f1df27cd2..abede15b6013 100644 --- a/js/src/jit/JitCompartment.h +++ b/js/src/jit/JitCompartment.h @@ -56,6 +56,7 @@ typedef void (*EnterJitCode)(void *code, unsigned argc, Value *argv, Interpreter size_t numStackValues, Value *vp); class IonBuilder; +class JitcodeGlobalTable; // ICStubSpace is an abstraction for allocation policy and storage for stub data. // There are two kinds of stubs: optimized stubs and fallback stubs (the latter @@ -232,6 +233,9 @@ class JitRuntime // their callee. js::Value ionReturnOverride_; + // Global table of jitcode native address => bytecode address mappings. + JitcodeGlobalTable *jitcodeGlobalTable_; + private: JitCode *generateExceptionTailStub(JSContext *cx); JitCode *generateBailoutTailStub(JSContext *cx); @@ -381,6 +385,23 @@ class JitRuntime JS_ASSERT(!v.isMagic()); ionReturnOverride_ = v; } + + bool hasJitcodeGlobalTable() const { + return jitcodeGlobalTable_ != nullptr; + } + + JitcodeGlobalTable *getJitcodeGlobalTable() { + JS_ASSERT(hasJitcodeGlobalTable()); + return jitcodeGlobalTable_; + } + + bool isNativeToBytecodeMapEnabled(JSRuntime *rt) { +#ifdef DEBUG + return true; +#else // DEBUG + return rt->spsProfiler.enabled(); +#endif // DEBUG + } }; class JitZone diff --git a/js/src/jit/JitFrameIterator.h b/js/src/jit/JitFrameIterator.h index 13ae5fa60c12..f20fe53bc218 100644 --- a/js/src/jit/JitFrameIterator.h +++ b/js/src/jit/JitFrameIterator.h @@ -258,6 +258,12 @@ class JitFrameIterator void dump() const; inline BaselineFrame *baselineFrame() const; + +#ifdef DEBUG + bool verifyReturnAddressUsingNativeToBytecodeMap(); +#else + inline bool verifyReturnAddressUsingNativeToBytecodeMap() { return true; } +#endif }; class IonJSFrameLayout; diff --git a/js/src/jit/JitcodeMap.cpp b/js/src/jit/JitcodeMap.cpp new file mode 100644 index 000000000000..f6348a221603 --- /dev/null +++ b/js/src/jit/JitcodeMap.cpp @@ -0,0 +1,702 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "jit/JitcodeMap.h" + +#include "mozilla/DebugOnly.h" +#include "jit/BaselineJIT.h" +#include "jit/IonSpewer.h" + +#include "js/Vector.h" + +namespace js { +namespace jit { + +bool +JitcodeGlobalEntry::IonEntry::callStackAtAddr(JSRuntime *rt, void *ptr, + BytecodeLocationVector &results, + uint32_t *depth) const +{ + JS_ASSERT(containsPointer(ptr)); + uint32_t ptrOffset = reinterpret_cast(ptr) - + reinterpret_cast(nativeStartAddr()); + + uint32_t regionIdx = regionTable()->findRegionEntry(ptrOffset); + JS_ASSERT(regionIdx < regionTable()->numRegions()); + + JitcodeRegionEntry region = regionTable()->regionEntry(regionIdx); + *depth = region.scriptDepth(); + + JitcodeRegionEntry::ScriptPcIterator locationIter = region.scriptPcIterator(); + JS_ASSERT(locationIter.hasMore()); + bool first = true; + while (locationIter.hasMore()) { + uint32_t scriptIdx, pcOffset; + locationIter.readNext(&scriptIdx, &pcOffset); + // For the first entry pushed (innermost frame), the pcOffset is obtained + // from the delta-run encodings. + if (first) { + pcOffset = region.findPcOffset(ptrOffset, pcOffset); + first = false; + } + JSScript *script = getScript(scriptIdx); + jsbytecode *pc = script->offsetToPC(pcOffset); + if (!results.append(BytecodeLocation(script, pc))) + return false; + } + + return true; +} + +void +JitcodeGlobalEntry::IonEntry::destroy() +{ + // The region table is stored at the tail of the compacted data, + // which means the start of the region table is a pointer to + // the _middle_ of the memory space allocated for it. + // + // When freeing it, obtain the payload start pointer first. + if (regionTable_) + js_free((void*) (regionTable_->payloadStart())); + regionTable_ = nullptr; + + // Single tag is just pointer-to-jsscript, no memory to free. + ScriptListTag tag = scriptListTag(); + if (tag > Single) + js_free(scriptListPointer()); + scriptList_ = 0; +} + +bool +JitcodeGlobalEntry::BaselineEntry::callStackAtAddr(JSRuntime *rt, void *ptr, + BytecodeLocationVector &results, + uint32_t *depth) const +{ + JS_ASSERT(containsPointer(ptr)); + JS_ASSERT(script_->hasBaselineScript()); + + jsbytecode *pc = script_->baselineScript()->pcForNativeAddress(script_, (uint8_t*) ptr); + if (!results.append(BytecodeLocation(script_, pc))) + return false; + + *depth = 1; + + return true; +} + +bool +JitcodeGlobalEntry::IonCacheEntry::callStackAtAddr(JSRuntime *rt, void *ptr, + BytecodeLocationVector &results, + uint32_t *depth) const +{ + JS_ASSERT(containsPointer(ptr)); + + // There must exist an entry for the rejoin addr if this entry exists. + JitRuntime *jitrt = rt->jitRuntime(); + JitcodeGlobalEntry entry; + jitrt->getJitcodeGlobalTable()->lookupInfallible(rejoinAddr(), &entry); + JS_ASSERT(entry.isIon()); + + return entry.callStackAtAddr(rt, rejoinAddr(), results, depth); +} + + +static int ComparePointers(const void *a, const void *b) { + const uint8_t *a_ptr = reinterpret_cast(a); + const uint8_t *b_ptr = reinterpret_cast(b); + if (a_ptr < b_ptr) + return -1; + if (a_ptr > b_ptr) + return 1; + return 0; +} + +/* static */ int +JitcodeGlobalEntry::compare(const JitcodeGlobalEntry &ent1, const JitcodeGlobalEntry &ent2) +{ + // Both parts of compare cannot be a query. + JS_ASSERT(!(ent1.isQuery() && ent2.isQuery())); + + // Ensure no overlaps for non-query lookups. + JS_ASSERT_IF(!ent1.isQuery() && !ent2.isQuery(), !ent1.overlapsWith(ent2)); + + return ComparePointers(ent1.nativeStartAddr(), ent2.nativeStartAddr()); +} + +bool +JitcodeGlobalTable::lookup(void *ptr, JitcodeGlobalEntry *result) +{ + JS_ASSERT(result); + + // Construct a JitcodeGlobalEntry::Query to do the lookup + JitcodeGlobalEntry query = JitcodeGlobalEntry::MakeQuery(ptr); + return tree_.contains(query, result); +} + +void +JitcodeGlobalTable::lookupInfallible(void *ptr, JitcodeGlobalEntry *result) +{ + mozilla::DebugOnly success = lookup(ptr, result); + JS_ASSERT(success); +} + +bool +JitcodeGlobalTable::addEntry(const JitcodeGlobalEntry &entry) +{ + // Should only add Main entries for now. + JS_ASSERT(entry.isIon() || entry.isBaseline() || entry.isIonCache()); + return tree_.insert(entry); +} + +void +JitcodeGlobalTable::removeEntry(void *startAddr) +{ + JitcodeGlobalEntry query = JitcodeGlobalEntry::MakeQuery(startAddr); + tree_.remove(query); +} + +/* static */ void +JitcodeRegionEntry::WriteHead(CompactBufferWriter &writer, + uint32_t nativeOffset, uint8_t scriptDepth) +{ + writer.writeUnsigned(nativeOffset); + writer.writeByte(scriptDepth); +} + +/* static */ void +JitcodeRegionEntry::ReadHead(CompactBufferReader &reader, + uint32_t *nativeOffset, uint8_t *scriptDepth) +{ + *nativeOffset = reader.readUnsigned(); + *scriptDepth = reader.readByte(); +} + +/* static */ void +JitcodeRegionEntry::WriteScriptPc(CompactBufferWriter &writer, + uint32_t scriptIdx, uint32_t pcOffset) +{ + writer.writeUnsigned(scriptIdx); + writer.writeUnsigned(pcOffset); +} + +/* static */ void +JitcodeRegionEntry::ReadScriptPc(CompactBufferReader &reader, + uint32_t *scriptIdx, uint32_t *pcOffset) +{ + *scriptIdx = reader.readUnsigned(); + *pcOffset = reader.readUnsigned(); +} + +/* static */ void +JitcodeRegionEntry::WriteDelta(CompactBufferWriter &writer, + uint32_t nativeDelta, int32_t pcDelta) +{ + if (pcDelta >= 0) { + // 1 and 2-byte formats possible. + + // NNNN-BBB0 + if (pcDelta <= ENC1_PC_DELTA_MAX && nativeDelta <= ENC1_NATIVE_DELTA_MAX) { + uint8_t encVal = ENC1_MASK_VAL | (pcDelta << ENC1_PC_DELTA_SHIFT) | + (nativeDelta << ENC1_NATIVE_DELTA_SHIFT); + writer.writeByte(encVal); + return; + } + + // NNNN-NNNN BBBB-BB01 + if (pcDelta <= ENC2_PC_DELTA_MAX && nativeDelta <= ENC2_NATIVE_DELTA_MAX) { + uint16_t encVal = ENC2_MASK_VAL | (pcDelta << ENC2_PC_DELTA_SHIFT) | + (nativeDelta << ENC2_NATIVE_DELTA_SHIFT); + writer.writeByte(encVal & 0xff); + writer.writeByte((encVal >> 8) & 0xff); + return; + } + } + + // NNNN-NNNN NNNB-BBBB BBBB-B011 + if (pcDelta >= ENC3_PC_DELTA_MIN && pcDelta <= ENC3_PC_DELTA_MAX && + nativeDelta <= ENC3_NATIVE_DELTA_MAX) + { + uint32_t encVal = ENC3_MASK_VAL | + ((pcDelta << ENC3_PC_DELTA_SHIFT) & ENC3_PC_DELTA_MASK) | + (nativeDelta << ENC3_NATIVE_DELTA_SHIFT); + writer.writeByte(encVal & 0xff); + writer.writeByte((encVal >> 8) & 0xff); + writer.writeByte((encVal >> 16) & 0xff); + return; + } + + // NNNN-NNNN NNNN-NNNN BBBB-BBBB BBBB-B111 + if (pcDelta >= ENC4_PC_DELTA_MIN && pcDelta <= ENC4_PC_DELTA_MAX && + nativeDelta <= ENC4_NATIVE_DELTA_MAX) + { + uint32_t encVal = ENC4_MASK_VAL | + ((pcDelta << ENC4_PC_DELTA_SHIFT) & ENC4_PC_DELTA_MASK) | + (nativeDelta << ENC4_NATIVE_DELTA_SHIFT); + writer.writeByte(encVal & 0xff); + writer.writeByte((encVal >> 8) & 0xff); + writer.writeByte((encVal >> 16) & 0xff); + writer.writeByte((encVal >> 24) & 0xff); + return; + } + + // Should never get here. + MOZ_CRASH("pcDelta/nativeDelta values are too large to encode."); +} + +/* static */ void +JitcodeRegionEntry::ReadDelta(CompactBufferReader &reader, + uint32_t *nativeDelta, int32_t *pcDelta) +{ + // NB: + // It's possible to get nativeDeltas with value 0 in two cases: + // + // 1. The last region's run. This is because the region table's start + // must be 4-byte aligned, and we must insert padding bytes to align the + // payload section before emitting the table. + // + // 2. A zero-offset nativeDelta with a negative pcDelta. + // + // So if nativeDelta is zero, then pcDelta must be <= 0. + + // NNNN-BBB0 + const uint32_t firstByte = reader.readByte(); + if ((firstByte & ENC1_MASK) == ENC1_MASK_VAL) { + uint32_t encVal = firstByte; + *nativeDelta = encVal >> ENC1_NATIVE_DELTA_SHIFT; + *pcDelta = (encVal & ENC1_PC_DELTA_MASK) >> ENC1_PC_DELTA_SHIFT; + JS_ASSERT_IF(*nativeDelta == 0, *pcDelta <= 0); + return; + } + + // NNNN-NNNN BBBB-BB01 + const uint32_t secondByte = reader.readByte(); + if ((firstByte & ENC2_MASK) == ENC2_MASK_VAL) { + uint32_t encVal = firstByte | secondByte << 8; + *nativeDelta = encVal >> ENC2_NATIVE_DELTA_SHIFT; + *pcDelta = (encVal & ENC2_PC_DELTA_MASK) >> ENC2_PC_DELTA_SHIFT; + JS_ASSERT(*pcDelta != 0); + JS_ASSERT_IF(*nativeDelta == 0, *pcDelta <= 0); + return; + } + + // NNNN-NNNN NNNB-BBBB BBBB-B011 + const uint32_t thirdByte = reader.readByte(); + if ((firstByte & ENC3_MASK) == ENC3_MASK_VAL) { + uint32_t encVal = firstByte | secondByte << 8 | thirdByte << 16; + *nativeDelta = encVal >> ENC3_NATIVE_DELTA_SHIFT; + + uint32_t pcDeltaU = (encVal & ENC3_PC_DELTA_MASK) >> ENC3_PC_DELTA_SHIFT; + // Fix sign if necessary. + if (pcDeltaU > ENC3_PC_DELTA_MAX) + pcDeltaU |= ~ENC3_PC_DELTA_MAX; + *pcDelta = pcDeltaU; + JS_ASSERT(*pcDelta != 0); + JS_ASSERT_IF(*nativeDelta == 0, *pcDelta <= 0); + return; + } + + // NNNN-NNNN NNNN-NNNN BBBB-BBBB BBBB-B111 + JS_ASSERT((firstByte & ENC4_MASK) == ENC4_MASK_VAL); + const uint32_t fourthByte = reader.readByte(); + uint32_t encVal = firstByte | secondByte << 8 | thirdByte << 16 | fourthByte << 24; + *nativeDelta = encVal >> ENC4_NATIVE_DELTA_SHIFT; + + uint32_t pcDeltaU = (encVal & ENC4_PC_DELTA_MASK) >> ENC4_PC_DELTA_SHIFT; + // fix sign if necessary + if (pcDeltaU > ENC4_PC_DELTA_MAX) + pcDeltaU |= ~ENC4_PC_DELTA_MAX; + *pcDelta = pcDeltaU; + + JS_ASSERT(*pcDelta != 0); + JS_ASSERT_IF(*nativeDelta == 0, *pcDelta <= 0); +} + +/* static */ uint32_t +JitcodeRegionEntry::ExpectedRunLength(const CodeGeneratorShared::NativeToBytecode *entry, + const CodeGeneratorShared::NativeToBytecode *end) +{ + JS_ASSERT(entry < end); + + // We always use the first entry, so runLength starts at 1 + uint32_t runLength = 1; + + uint32_t curNativeOffset = entry->nativeOffset.offset(); + uint32_t curBytecodeOffset = entry->tree->script()->pcToOffset(entry->pc); + + for (auto nextEntry = entry + 1; nextEntry != end; nextEntry += 1) { + // If the next run moves to a different inline site, stop the run. + if (nextEntry->tree != entry->tree) + break; + + uint32_t nextNativeOffset = nextEntry->nativeOffset.offset(); + uint32_t nextBytecodeOffset = nextEntry->tree->script()->pcToOffset(nextEntry->pc); + JS_ASSERT(nextNativeOffset >= curNativeOffset); + + uint32_t nativeDelta = nextNativeOffset - curNativeOffset; + int32_t bytecodeDelta = int32_t(nextBytecodeOffset) - int32_t(curBytecodeOffset); + + // If deltas are too large (very unlikely), stop the run. + if (!IsDeltaEncodeable(nativeDelta, bytecodeDelta)) + break; + + runLength++; + + // If the run has grown to its maximum length, stop the run. + if (runLength == MAX_RUN_LENGTH) + break; + + curNativeOffset = nextNativeOffset; + curBytecodeOffset = nextBytecodeOffset; + } + + return runLength; +} + +struct JitcodeMapBufferWriteSpewer +{ +#ifdef DEBUG + CompactBufferWriter *writer; + uint32_t startPos; + + static const uint32_t DumpMaxBytes = 50; + + JitcodeMapBufferWriteSpewer(CompactBufferWriter &w) + : writer(&w), startPos(writer->length()) + {} + + void spewAndAdvance(const char *name) { + uint32_t curPos = writer->length(); + const uint8_t *start = writer->buffer() + startPos; + const uint8_t *end = writer->buffer() + curPos; + const char *MAP = "0123456789ABCDEF"; + uint32_t bytes = end - start; + + char buffer[DumpMaxBytes * 3]; + for (uint32_t i = 0; i < bytes; i++) { + buffer[i*3] = MAP[(start[i] >> 4) & 0xf]; + buffer[i*3 + 1] = MAP[(start[i] >> 0) & 0xf]; + buffer[i*3 + 2] = ' '; + } + if (bytes >= DumpMaxBytes) + buffer[DumpMaxBytes*3 - 1] = '\0'; + else + buffer[bytes*3 - 1] = '\0'; + + IonSpew(IonSpew_Profiling, "%s@%d[%d bytes] - %s", name, int(startPos), int(bytes), buffer); + + // Move to the end of the current buffer. + startPos = writer->length(); + } +#else // !DEBUG + JitcodeMapBufferWriteSpewer(CompactBufferWriter &w) {} + void spewAndAdvance(const char *name) {} +#endif // DEBUG +}; + +// Write a run, starting at the given NativeToBytecode entry, into the given buffer writer. +/* static */ bool +JitcodeRegionEntry::WriteRun(CompactBufferWriter &writer, + JSScript **scriptList, uint32_t scriptListSize, + uint32_t runLength, const CodeGeneratorShared::NativeToBytecode *entry) +{ + JS_ASSERT(runLength > 0); + JS_ASSERT(runLength <= MAX_RUN_LENGTH); + + // Calculate script depth. + JS_ASSERT(entry->tree->depth() <= 0xff); + uint8_t scriptDepth = entry->tree->depth(); + uint32_t regionNativeOffset = entry->nativeOffset.offset(); + + JitcodeMapBufferWriteSpewer spewer(writer); + + // Write the head info. + IonSpew(IonSpew_Profiling, " Head Info: nativeOffset=%d scriptDepth=%d", + int(regionNativeOffset), int(scriptDepth)); + WriteHead(writer, regionNativeOffset, scriptDepth); + spewer.spewAndAdvance(" "); + + // Write each script/pc pair. + { + InlineScriptTree *curTree = entry->tree; + jsbytecode *curPc = entry->pc; + for (uint8_t i = 0; i < scriptDepth; i++) { + // Find the index of the script within the list. + // NB: scriptList is guaranteed to contain curTree->script() + uint32_t scriptIdx = 0; + for (; scriptIdx < scriptListSize; scriptIdx++) { + if (scriptList[scriptIdx] == curTree->script()) + break; + } + JS_ASSERT(scriptIdx < scriptListSize); + + uint32_t pcOffset = curTree->script()->pcToOffset(curPc); + + IonSpew(IonSpew_Profiling, " Script/PC %d: scriptIdx=%d pcOffset=%d", + int(i), int(scriptIdx), int(pcOffset)); + WriteScriptPc(writer, scriptIdx, pcOffset); + spewer.spewAndAdvance(" "); + + JS_ASSERT_IF(i < scriptDepth - 1, curTree->hasCaller()); + curPc = curTree->callerPc(); + curTree = curTree->caller(); + } + } + + // Start writing runs. + uint32_t curNativeOffset = entry->nativeOffset.offset(); + uint32_t curBytecodeOffset = entry->tree->script()->pcToOffset(entry->pc); + + IonSpew(IonSpew_Profiling, " Writing Delta Run from nativeOffset=%d bytecodeOffset=%d", + int(curNativeOffset), int(curBytecodeOffset)); + + // Skip first entry because it is implicit in the header. Start at subsequent entry. + for (uint32_t i = 1; i < runLength; i++) { + JS_ASSERT(entry[i].tree == entry->tree); + + uint32_t nextNativeOffset = entry[i].nativeOffset.offset(); + uint32_t nextBytecodeOffset = entry[i].tree->script()->pcToOffset(entry[i].pc); + JS_ASSERT(nextNativeOffset >= curNativeOffset); + + uint32_t nativeDelta = nextNativeOffset - curNativeOffset; + int32_t bytecodeDelta = int32_t(nextBytecodeOffset) - int32_t(curBytecodeOffset); + JS_ASSERT(IsDeltaEncodeable(nativeDelta, bytecodeDelta)); + + IonSpew(IonSpew_Profiling, " RunEntry native: %d-%d [%d] bytecode: %d-%d [%d]", + int(curNativeOffset), int(nextNativeOffset), int(nativeDelta), + int(curBytecodeOffset), int(nextBytecodeOffset), int(bytecodeDelta)); + WriteDelta(writer, nativeDelta, bytecodeDelta); + + // Spew the bytecode in these ranges. + if (curBytecodeOffset < nextBytecodeOffset) { + IonSpewStart(IonSpew_Profiling, " OPS: "); + uint32_t curBc = curBytecodeOffset; + while (curBc < nextBytecodeOffset) { + jsbytecode *pc = entry[i].tree->script()->offsetToPC(curBc); + JSOp op = JSOp(*pc); + IonSpewCont(IonSpew_Profiling, "%s ", js_CodeName[op]); + curBc += GetBytecodeLength(pc); + } + IonSpewFin(IonSpew_Profiling); + } + spewer.spewAndAdvance(" "); + + curNativeOffset = nextNativeOffset; + curBytecodeOffset = nextBytecodeOffset; + } + + if (writer.oom()) + return false; + + return true; +} + +void +JitcodeRegionEntry::unpack() +{ + CompactBufferReader reader(data_, end_); + ReadHead(reader, &nativeOffset_, &scriptDepth_); + JS_ASSERT(scriptDepth_ > 0); + + scriptPcStack_ = reader.currentPosition(); + // Skip past script/pc stack + for (unsigned i = 0; i < scriptDepth_; i++) { + uint32_t scriptIdx, pcOffset; + ReadScriptPc(reader, &scriptIdx, &pcOffset); + } + + deltaRun_ = reader.currentPosition(); +} + +uint32_t +JitcodeRegionEntry::findPcOffset(uint32_t queryNativeOffset, uint32_t startPcOffset) const +{ + DeltaIterator iter = deltaIterator(); + uint32_t curNativeOffset = nativeOffset(); + uint32_t curPcOffset = startPcOffset; + while (iter.hasMore()) { + uint32_t nativeDelta; + int32_t pcDelta; + iter.readNext(&nativeDelta, &pcDelta); + + // The start address of the next delta-run entry is counted towards + // the current delta-run entry, because return addresses should + // associate with the bytecode op prior (the call) not the op after. + if (queryNativeOffset <= curNativeOffset + nativeDelta) + break; + curNativeOffset += nativeDelta; + curPcOffset += pcDelta; + } + return curPcOffset; +} + +bool +JitcodeIonTable::makeIonEntry(JSContext *cx, JitCode *code, + uint32_t numScripts, JSScript **scripts, + JitcodeGlobalEntry::IonEntry &out) +{ + typedef JitcodeGlobalEntry::IonEntry::SizedScriptList SizedScriptList; + + JS_ASSERT(numScripts > 0); + + if (numScripts == 1) { + out.init(code->raw(), code->raw() + code->instructionsSize(), scripts[0], this); + return true; + } + + if (numScripts < uint32_t(JitcodeGlobalEntry::IonEntry::Multi)) { + out.init(code->raw(), code->raw() + code->instructionsSize(), numScripts, scripts, this); + return true; + } + + // Create SizedScriptList + void *mem = cx->malloc_(SizedScriptList::AllocSizeFor(numScripts)); + if (!mem) + return false; + SizedScriptList *scriptList = new (mem) SizedScriptList(numScripts, scripts); + out.init(code->raw(), code->raw() + code->instructionsSize(), scriptList, this); + return true; +} + +uint32_t +JitcodeIonTable::findRegionEntry(uint32_t nativeOffset) const +{ + static const uint32_t LINEAR_SEARCH_THRESHOLD = 8; + uint32_t regions = numRegions(); + JS_ASSERT(regions > 0); + + // For small region lists, just search linearly. + if (regions <= LINEAR_SEARCH_THRESHOLD) { + JitcodeRegionEntry previousEntry = regionEntry(0); + for (uint32_t i = 1; i < regions; i++) { + JitcodeRegionEntry nextEntry = regionEntry(i); + JS_ASSERT(nextEntry.nativeOffset() >= previousEntry.nativeOffset()); + + // See note in binary-search code below about why we use '<=' here instead of + // '<'. Short explanation: regions are closed at their ending addresses, + // and open at their starting addresses. + if (nativeOffset <= nextEntry.nativeOffset()) + return i-1; + + previousEntry = nextEntry; + } + // If nothing found, assume it falls within last region. + return regions - 1; + } + + // For larger ones, binary search the region table. + uint32_t idx = 0; + uint32_t count = regions; + while (count > 1) { + uint32_t step = count/2; + uint32_t mid = idx + step; + JitcodeRegionEntry midEntry = regionEntry(mid); + + // A region memory range is closed at its ending address, not starting + // address. This is because the return address for calls must associate + // with the call's bytecode PC, not the PC of the bytecode operator after + // the call. + // + // So a query is < an entry if the query nativeOffset is <= the start address + // of the entry, and a query is >= an entry if the query nativeOffset is > the + // start address of an entry. + if (nativeOffset <= midEntry.nativeOffset()) { + // Target entry is below midEntry. + count = step; + } else { // if (nativeOffset > midEntry.nativeOffset()) + // Target entry is at midEntry or above. + idx = mid; + count -= step; + } + } + return idx; +} + +/* static */ bool +JitcodeIonTable::WriteIonTable(CompactBufferWriter &writer, + JSScript **scriptList, uint32_t scriptListSize, + const CodeGeneratorShared::NativeToBytecode *start, + const CodeGeneratorShared::NativeToBytecode *end, + uint32_t *tableOffsetOut, uint32_t *numRegionsOut) +{ + JS_ASSERT(tableOffsetOut != nullptr); + JS_ASSERT(numRegionsOut != nullptr); + JS_ASSERT(writer.length() == 0); + JS_ASSERT(scriptListSize > 0); + + IonSpew(IonSpew_Profiling, "Writing native to bytecode map for %s:%d (%d entries)", + scriptList[0]->filename(), scriptList[0]->lineno(), + int(end - start)); + + IonSpew(IonSpew_Profiling, " ScriptList of size %d", int(scriptListSize)); + for (uint32_t i = 0; i < scriptListSize; i++) { + IonSpew(IonSpew_Profiling, " Script %d - %s:%d", + int(i), scriptList[i]->filename(), int(scriptList[i]->lineno())); + } + + // Write out runs first. Keep a vector tracking the positive offsets from payload + // start to the run. + const CodeGeneratorShared::NativeToBytecode *curEntry = start; + js::Vector runOffsets; + + while (curEntry != end) { + // Calculate the length of the next run. + uint32_t runLength = JitcodeRegionEntry::ExpectedRunLength(curEntry, end); + JS_ASSERT(runLength > 0); + JS_ASSERT(runLength <= (end - curEntry)); + IonSpew(IonSpew_Profiling, " Run at entry %d, length %d, buffer offset %d", + int(curEntry - start), int(runLength), int(writer.length())); + + // Store the offset of the run. + if (!runOffsets.append(writer.length())) + return false; + + // Encode the run. + if (!JitcodeRegionEntry::WriteRun(writer, scriptList, scriptListSize, runLength, curEntry)) + return false; + + curEntry += runLength; + } + + // Done encoding regions. About to start table. Ensure we are aligned to 4 bytes + // since table is composed of uint32_t values. + uint32_t padding = sizeof(uint32_t) - (writer.length() % sizeof(uint32_t)); + if (padding == sizeof(uint32_t)) + padding = 0; + IonSpew(IonSpew_Profiling, " Padding %d bytes after run @%d", + int(padding), int(writer.length())); + for (uint32_t i = 0; i < padding; i++) + writer.writeByte(0); + + // Now at start of table. + uint32_t tableOffset = writer.length(); + + // The table being written at this point will be accessed directly via uint32_t + // pointers, so all writes below use native endianness. + + // Write out numRegions + IonSpew(IonSpew_Profiling, " Writing numRuns=%d", int(runOffsets.length())); + writer.writeNativeEndianUint32_t(runOffsets.length()); + + // Write out region offset table. The offsets in |runOffsets| are currently forward + // offsets from the beginning of the buffer. We convert them to backwards offsets + // from the start of the table before writing them into their table entries. + for (uint32_t i = 0; i < runOffsets.length(); i++) { + IonSpew(IonSpew_Profiling, " Run %d offset=%d backOffset=%d @%d", + int(i), int(runOffsets[i]), int(tableOffset - runOffsets[i]), int(writer.length())); + writer.writeNativeEndianUint32_t(tableOffset - runOffsets[i]); + } + + if (writer.oom()) + return false; + + *tableOffsetOut = tableOffset; + *numRegionsOut = runOffsets.length(); + return true; +} + + +} // namespace jit +} // namespace js diff --git a/js/src/jit/JitcodeMap.h b/js/src/jit/JitcodeMap.h new file mode 100644 index 000000000000..fcdfd696c2b5 --- /dev/null +++ b/js/src/jit/JitcodeMap.h @@ -0,0 +1,865 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_JitcodeMap_h +#define jit_JitcodeMap_h + +#include "ds/SplayTree.h" +#include "jit/CompactBuffer.h" +#include "jit/CompileInfo.h" +#include "jit/shared/CodeGenerator-shared.h" + +namespace js { +namespace jit { + +/* + * The Ion jitcode map implements tables to allow mapping from addresses in ion jitcode + * to the list of (JSScript *, jsbytecode *) pairs that are implicitly active in the frame at + * that point in the native code. + * + * To represent this information efficiently, a multi-level table is used. + * + * At the top level, a global splay-tree of JitcodeGlobalEntry describings the mapping for + * each individual IonCode script generated by compiles. The entries are ordered by their + * nativeStartAddr. + * + * Every entry in the table is of fixed size, but there are different entry types, + * distinguished by the kind field. + */ + +class JitcodeIonTable; +class JitcodeRegionEntry; + +class JitcodeGlobalEntry +{ + public: + enum Kind { + INVALID = 0, + Ion, + Baseline, + IonCache, + Query, + LIMIT + }; + JS_STATIC_ASSERT(LIMIT <= 8); + + struct BytecodeLocation { + JSScript *script; + jsbytecode *pc; + BytecodeLocation(JSScript *script, jsbytecode *pc) : script(script), pc(pc) {} + }; + typedef Vector BytecodeLocationVector; + + struct BaseEntry + { + void *nativeStartAddr_; + void *nativeEndAddr_; + Kind kind_; + + void init() { + nativeStartAddr_ = nullptr; + nativeEndAddr_ = nullptr; + kind_ = INVALID; + } + + void init(Kind kind, void *nativeStartAddr, void *nativeEndAddr) { + JS_ASSERT(nativeStartAddr); + JS_ASSERT(nativeEndAddr); + JS_ASSERT(kind > INVALID && kind < LIMIT); + nativeStartAddr_ = nativeStartAddr; + nativeEndAddr_ = nativeEndAddr; + kind_ = kind; + } + + Kind kind() const { + return kind_; + } + void *nativeStartAddr() const { + return nativeStartAddr_; + } + void *nativeEndAddr() const { + return nativeEndAddr_; + } + + bool startsBelowPointer(void *ptr) const { + return ((uint8_t *)nativeStartAddr()) <= ((uint8_t *) ptr); + } + bool endsAbovePointer(void *ptr) const { + return ((uint8_t *)nativeEndAddr()) > ((uint8_t *) ptr); + } + bool containsPointer(void *ptr) const { + return startsBelowPointer(ptr) && endsAbovePointer(ptr); + } + }; + + struct IonEntry : public BaseEntry + { + uintptr_t scriptList_; + + // regionTable_ points to the start of the region table within the + // packed map for compile represented by this entry. Since the + // region table occurs at the tail of the memory region, this pointer + // points somewhere inside the region memory space, and not to the start + // of the memory space. + JitcodeIonTable *regionTable_; + + static const unsigned LowBits = 3; + static const uintptr_t LowMask = (uintptr_t(1) << LowBits) - 1; + + enum ScriptListTag { + Single = 0, + Multi = 7 + }; + + struct SizedScriptList { + uint32_t size; + JSScript *scripts[0]; + SizedScriptList(uint32_t sz, JSScript **scr) : size(sz) { + for (uint32_t i = 0; i < size; i++) + scripts[i] = scr[i]; + } + + static uint32_t AllocSizeFor(uint32_t nscripts) { + return sizeof(SizedScriptList) + (nscripts * sizeof(JSScript *)); + } + }; + + void init(void *nativeStartAddr, void *nativeEndAddr, + JSScript *script, JitcodeIonTable *regionTable) + { + JS_ASSERT((uintptr_t(script) & LowMask) == 0); + JS_ASSERT(script); + JS_ASSERT(regionTable); + BaseEntry::init(Ion, nativeStartAddr, nativeEndAddr); + scriptList_ = uintptr_t(script); + regionTable_ = regionTable; + } + + void init(void *nativeStartAddr, void *nativeEndAddr, + unsigned numScripts, JSScript **scripts, JitcodeIonTable *regionTable) + { + JS_ASSERT((uintptr_t(scripts) & LowMask) == 0); + JS_ASSERT(numScripts >= 1); + JS_ASSERT(numScripts <= 6); + JS_ASSERT(scripts); + JS_ASSERT(regionTable); + BaseEntry::init(Ion, nativeStartAddr, nativeEndAddr); + scriptList_ = uintptr_t(scripts) | numScripts; + regionTable_ = regionTable; + } + + void init(void *nativeStartAddr, void *nativeEndAddr, + SizedScriptList *scripts, JitcodeIonTable *regionTable) + { + JS_ASSERT((uintptr_t(scripts) & LowMask) == 0); + JS_ASSERT(scripts->size > 6); + JS_ASSERT(scripts); + JS_ASSERT(regionTable); + + BaseEntry::init(Ion, nativeStartAddr, nativeEndAddr); + scriptList_ = uintptr_t(scripts) | uintptr_t(Multi); + regionTable_ = regionTable; + } + + ScriptListTag scriptListTag() const { + return static_cast(scriptList_ & LowMask); + } + void *scriptListPointer() const { + return reinterpret_cast(scriptList_ & ~LowMask); + } + + JSScript *singleScript() const { + JS_ASSERT(scriptListTag() == Single); + return reinterpret_cast(scriptListPointer()); + } + JSScript **rawScriptArray() const { + JS_ASSERT(scriptListTag() < Multi); + return reinterpret_cast(scriptListPointer()); + } + SizedScriptList *sizedScriptList() const { + JS_ASSERT(scriptListTag() == Multi); + return reinterpret_cast(scriptListPointer()); + } + + unsigned numScripts() const { + ScriptListTag tag = scriptListTag(); + if (tag == Single) + return 1; + + if (tag < Multi) { + JS_ASSERT(int(tag) >= 2); + return static_cast(tag); + } + + return sizedScriptList()->size; + } + + JSScript *getScript(unsigned idx) const { + JS_ASSERT(idx < numScripts()); + + ScriptListTag tag = scriptListTag(); + + if (tag == Single) + return singleScript(); + + if (tag < Multi) { + JS_ASSERT(int(tag) >= 2); + return rawScriptArray()[idx]; + } + + return sizedScriptList()->scripts[idx]; + } + + void destroy(); + + JitcodeIonTable *regionTable() const { + return regionTable_; + } + + int scriptIndex(JSScript *script) const { + unsigned count = numScripts(); + for (unsigned i = 0; i < count; i++) { + if (getScript(i) == script) + return i; + } + return -1; + } + + bool callStackAtAddr(JSRuntime *rt, void *ptr, BytecodeLocationVector &results, + uint32_t *depth) const; + }; + + struct BaselineEntry : public BaseEntry + { + JSScript *script_; + + void init(void *nativeStartAddr, void *nativeEndAddr, JSScript *script) + { + JS_ASSERT(script != nullptr); + BaseEntry::init(Baseline, nativeStartAddr, nativeEndAddr); + script_ = script; + } + + JSScript *script() const { + return script_; + } + + void destroy() {} + + bool callStackAtAddr(JSRuntime *rt, void *ptr, BytecodeLocationVector &results, + uint32_t *depth) const; + }; + + struct IonCacheEntry : public BaseEntry + { + void *rejoinAddr_; + + void init(void *nativeStartAddr, void *nativeEndAddr, void *rejoinAddr) + { + JS_ASSERT(rejoinAddr != nullptr); + BaseEntry::init(IonCache, nativeStartAddr, nativeEndAddr); + rejoinAddr_ = rejoinAddr; + } + + void *rejoinAddr() const { + return rejoinAddr_; + } + + void destroy() {} + + bool callStackAtAddr(JSRuntime *rt, void *ptr, BytecodeLocationVector &results, + uint32_t *depth) const; + }; + + // QueryEntry is never stored in the table, just used for queries + // where an instance of JitcodeGlobalEntry is required to do tree + // lookups. + struct QueryEntry : public BaseEntry + { + void init(void *addr) { + BaseEntry::init(Query, addr, addr); + } + uint8_t *addr() const { + return reinterpret_cast(nativeStartAddr()); + } + void destroy() {} + }; + + private: + union { + // Shadowing BaseEntry instance to allow access to base fields + // and type extraction. + BaseEntry base_; + + // The most common entry type: describing jitcode generated by + // Ion main-line code. + IonEntry ion_; + + // Baseline jitcode. + BaselineEntry baseline_; + + // IonCache stubs. + IonCacheEntry ionCache_; + + // When doing queries on the SplayTree for particular addresses, + // the query addresses are representd using a QueryEntry. + QueryEntry query_; + }; + + public: + JitcodeGlobalEntry() { + base_.init(); + } + + JitcodeGlobalEntry(const IonEntry &ion) { + ion_ = ion; + } + + JitcodeGlobalEntry(const BaselineEntry &baseline) { + baseline_ = baseline; + } + + JitcodeGlobalEntry(const IonCacheEntry &ionCache) { + ionCache_ = ionCache; + } + + JitcodeGlobalEntry(const QueryEntry &query) { + query_ = query; + } + + static JitcodeGlobalEntry MakeQuery(void *ptr) { + QueryEntry query; + query.init(ptr); + return JitcodeGlobalEntry(query); + } + + void destroy() { + switch (kind()) { + case Ion: + ionEntry().destroy(); + break; + case Baseline: + baselineEntry().destroy(); + break; + case IonCache: + ionCacheEntry().destroy(); + break; + case Query: + queryEntry().destroy(); + break; + default: + MOZ_ASSUME_UNREACHABLE("Invalid JitcodeGlobalEntry kind."); + } + } + + void *nativeStartAddr() const { + return base_.nativeStartAddr(); + } + void *nativeEndAddr() const { + return base_.nativeEndAddr(); + } + + bool startsBelowPointer(void *ptr) const { + return base_.startsBelowPointer(ptr); + } + bool endsAbovePointer(void *ptr) const { + return base_.endsAbovePointer(ptr); + } + bool containsPointer(void *ptr) const { + return base_.containsPointer(ptr); + } + + bool overlapsWith(const JitcodeGlobalEntry &entry) const { + // Catch full containment of |entry| within |this|, and partial overlaps. + if (containsPointer(entry.nativeStartAddr()) || containsPointer(entry.nativeEndAddr())) + return true; + + // Catch full containment of |this| within |entry|. + if (startsBelowPointer(entry.nativeEndAddr()) && endsAbovePointer(entry.nativeStartAddr())) + return true; + + return false; + } + + Kind kind() const { + return base_.kind(); + } + + bool isIon() const { + return kind() == Ion; + } + bool isBaseline() const { + return kind() == Baseline; + } + bool isIonCache() const { + return kind() == IonCache; + } + bool isQuery() const { + return kind() == Query; + } + + IonEntry &ionEntry() { + JS_ASSERT(isIon()); + return ion_; + } + BaselineEntry &baselineEntry() { + JS_ASSERT(isBaseline()); + return baseline_; + } + IonCacheEntry &ionCacheEntry() { + JS_ASSERT(isIonCache()); + return ionCache_; + } + QueryEntry &queryEntry() { + JS_ASSERT(isQuery()); + return query_; + } + + const IonEntry &ionEntry() const { + JS_ASSERT(isIon()); + return ion_; + } + const BaselineEntry &baselineEntry() const { + JS_ASSERT(isBaseline()); + return baseline_; + } + const IonCacheEntry &ionCacheEntry() const { + JS_ASSERT(isIonCache()); + return ionCache_; + } + const QueryEntry &queryEntry() const { + JS_ASSERT(isQuery()); + return query_; + } + + // Read the inline call stack at a given point in the native code and append into + // the given vector. Innermost (script,pc) pair will be appended first, and + // outermost appended last. + // + // Returns false on memory failure. + bool callStackAtAddr(JSRuntime *rt, void *ptr, BytecodeLocationVector &results, + uint32_t *depth) const + { + switch (kind()) { + case Ion: + return ionEntry().callStackAtAddr(rt, ptr, results, depth); + case Baseline: + return baselineEntry().callStackAtAddr(rt, ptr, results, depth); + case IonCache: + return ionCacheEntry().callStackAtAddr(rt, ptr, results, depth); + default: + MOZ_ASSUME_UNREACHABLE("Invalid JitcodeGlobalEntry kind."); + } + return false; + } + + // Figure out the number of the (JSScript *, jsbytecode *) pairs that are active + // at this location. + uint32_t lookupInlineCallDepth(void *ptr); + + // Compare two global entries. + static int compare(const JitcodeGlobalEntry &ent1, const JitcodeGlobalEntry &ent2); +}; + +/* + * Global table of JitcodeGlobalEntry values sorted by native address range. + */ +class JitcodeGlobalTable +{ + public: + typedef SplayTree EntryTree; + + typedef Vector EntryVector; + + private: + static const size_t LIFO_CHUNK_SIZE = 16 * 1024; + LifoAlloc treeAlloc_; + EntryTree tree_; + EntryVector entries_; + + public: + JitcodeGlobalTable() : treeAlloc_(LIFO_CHUNK_SIZE), tree_(&treeAlloc_), entries_() {} + ~JitcodeGlobalTable() {} + + bool empty() const { + return tree_.empty(); + } + + bool lookup(void *ptr, JitcodeGlobalEntry *result); + void lookupInfallible(void *ptr, JitcodeGlobalEntry *result); + + bool addEntry(const JitcodeGlobalEntry::IonEntry &entry) { + return addEntry(JitcodeGlobalEntry(entry)); + } + bool addEntry(const JitcodeGlobalEntry::BaselineEntry &entry) { + return addEntry(JitcodeGlobalEntry(entry)); + } + bool addEntry(const JitcodeGlobalEntry::IonCacheEntry &entry) { + return addEntry(JitcodeGlobalEntry(entry)); + } + + void removeEntry(void *startAddr); + + private: + bool addEntry(const JitcodeGlobalEntry &entry); +}; + + +/* + * Container class for main jitcode table. + * The Region table's memory is structured as follows: + * + * +------------------------------------------------+ | + * | Region 1 Run | | + * |------------------------------------------------| | + * | Region 2 Run | | + * | | | + * | | | + * |------------------------------------------------| | + * | Region 3 Run | | + * | | | + * |------------------------------------------------| |-- Payload + * | | | + * | ... | | + * | | | + * |------------------------------------------------| | + * | Region M Run | | + * | | | + * +================================================+ <- RegionTable pointer points here + * | uint23_t numRegions = M | | + * +------------------------------------------------+ | + * | Region 1 | | + * | uint32_t entryOffset = size(Payload) | | + * +------------------------------------------------+ | + * | | |-- Table + * | ... | | + * | | | + * +------------------------------------------------+ | + * | Region M | | + * | uint32_t entryOffset | | + * +------------------------------------------------+ | + * + * The region table is composed of two sections: a tail section that contains a table of + * fixed-size entries containing offsets into the the head section, and a head section that + * holds a sequence of variable-sized runs. The table in the tail section serves to + * locate the variable-length encoded structures in the head section. + * + * The entryOffsets in the table indicate the bytes offset to subtract from the regionTable + * pointer to arrive at the encoded region in the payload. + * + * + * Variable-length entries in payload + * ---------------------------------- + * The entryOffsets in the region table's fixed-sized entries refer to a location within the + * variable-length payload section. This location contains a compactly encoded "run" of + * mappings. + * + * Each run starts by describing the offset within the native code it starts at, and the + * sequence of (JSScript *, jsbytecode *) pairs active at that site. Following that, there + * are a number of variable-length entries encoding (nativeOffsetDelta, bytecodeOffsetDelta) + * pairs for the run. + * + * VarUint32 nativeOffset; + * - The offset from nativeStartAddr in the global table entry at which + * the jitcode for this region starts. + * + * Uint8_t scriptDepth; + * - The depth of inlined scripts for this region. + * + * List inlineScriptPcStack; + * - We encode (2 * scriptDepth) VarUint32s here. Each pair of uint32s are taken + * as an index into the scriptList in the global table entry, and a pcOffset + * respectively. + * + * List deltaRun; + * - The rest of the entry is a deltaRun that stores a series of variable-length + * encoded NativeAndBytecodeDelta datums. + */ +class JitcodeRegionEntry +{ + private: + static const unsigned MAX_RUN_LENGTH = 100; + + public: + static void WriteHead(CompactBufferWriter &writer, + uint32_t nativeOffset, uint8_t scriptDepth); + static void ReadHead(CompactBufferReader &reader, + uint32_t *nativeOffset, uint8_t *scriptDepth); + + static void WriteScriptPc(CompactBufferWriter &writer, uint32_t scriptIdx, uint32_t pcOffset); + static void ReadScriptPc(CompactBufferReader &reader, uint32_t *scriptIdx, uint32_t *pcOffset); + + static void WriteDelta(CompactBufferWriter &writer, uint32_t nativeDelta, int32_t pcDelta); + static void ReadDelta(CompactBufferReader &reader, uint32_t *nativeDelta, int32_t *pcDelta); + + // Given a pointer into an array of NativeToBytecode (and a pointer to the end of the array), + // compute the number of entries that would be consume by outputting a run starting + // at this one. + static uint32_t ExpectedRunLength(const CodeGeneratorShared::NativeToBytecode *entry, + const CodeGeneratorShared::NativeToBytecode *end); + + // Write a run, starting at the given NativeToBytecode entry, into the given buffer writer. + static bool WriteRun(CompactBufferWriter &writer, + JSScript **scriptList, uint32_t scriptListSize, + uint32_t runLength, const CodeGeneratorShared::NativeToBytecode *entry); + + // Delta Run entry formats are encoded little-endian: + // + // byte 0 + // NNNN-BBB0 + // Single byte format. nativeDelta in [0, 15], pcDelta in [0, 7] + // + static const uint32_t ENC1_MASK = 0x1; + static const uint32_t ENC1_MASK_VAL = 0x0; + + static const uint32_t ENC1_NATIVE_DELTA_MAX = 0xf; + static const unsigned ENC1_NATIVE_DELTA_SHIFT = 4; + + static const uint32_t ENC1_PC_DELTA_MASK = 0x0e; + static const int32_t ENC1_PC_DELTA_MAX = 0x7; + static const unsigned ENC1_PC_DELTA_SHIFT = 1; + + // byte 1 byte 0 + // NNNN-NNNN BBBB-BB01 + // Two-byte format. nativeDelta in [0, 255], pcDelta in [0, 63] + // + static const uint32_t ENC2_MASK = 0x3; + static const uint32_t ENC2_MASK_VAL = 0x1; + + static const uint32_t ENC2_NATIVE_DELTA_MAX = 0xff; + static const unsigned ENC2_NATIVE_DELTA_SHIFT = 8; + + static const uint32_t ENC2_PC_DELTA_MASK = 0x00fc; + static const int32_t ENC2_PC_DELTA_MAX = 0x3f; + static const unsigned ENC2_PC_DELTA_SHIFT = 2; + + // byte 2 byte 1 byte 0 + // NNNN-NNNN NNNB-BBBB BBBB-B011 + // Three-byte format. nativeDelta in [0, 2047], pcDelta in [-512, 511] + // + static const uint32_t ENC3_MASK = 0x7; + static const uint32_t ENC3_MASK_VAL = 0x3; + + static const uint32_t ENC3_NATIVE_DELTA_MAX = 0x7ff; + static const unsigned ENC3_NATIVE_DELTA_SHIFT = 13; + + static const uint32_t ENC3_PC_DELTA_MASK = 0x001ff8; + static const int32_t ENC3_PC_DELTA_MAX = 0x1ff; + static const int32_t ENC3_PC_DELTA_MIN = -ENC3_PC_DELTA_MAX - 1; + static const unsigned ENC3_PC_DELTA_SHIFT = 3; + + // byte 3 byte 2 byte 1 byte 0 + // NNNN-NNNN NNNN-NNNN BBBB-BBBB BBBB-B111 + // Three-byte format. nativeDelta in [0, 65535], pcDelta in [-4096, 4095] + static const uint32_t ENC4_MASK = 0x7; + static const uint32_t ENC4_MASK_VAL = 0x7; + + static const uint32_t ENC4_NATIVE_DELTA_MAX = 0xffff; + static const unsigned ENC4_NATIVE_DELTA_SHIFT = 16; + + static const uint32_t ENC4_PC_DELTA_MASK = 0x0000fff8; + static const int32_t ENC4_PC_DELTA_MAX = 0xfff; + static const int32_t ENC4_PC_DELTA_MIN = -ENC4_PC_DELTA_MAX - 1; + static const unsigned ENC4_PC_DELTA_SHIFT = 3; + + static bool IsDeltaEncodeable(uint32_t nativeDelta, int32_t pcDelta) { + return (nativeDelta <= ENC4_NATIVE_DELTA_MAX) && + (pcDelta >= ENC4_PC_DELTA_MIN) && (pcDelta <= ENC4_PC_DELTA_MAX); + } + + private: + const uint8_t *data_; + const uint8_t *end_; + + // Unpacked state from jitcode entry. + uint32_t nativeOffset_; + uint8_t scriptDepth_; + const uint8_t *scriptPcStack_; + const uint8_t *deltaRun_; + + void unpack(); + + public: + JitcodeRegionEntry(const uint8_t *data, const uint8_t *end) + : data_(data), end_(end), + nativeOffset_(0), scriptDepth_(0), + scriptPcStack_(nullptr), deltaRun_(nullptr) + { + JS_ASSERT(data_ < end_); + unpack(); + JS_ASSERT(scriptPcStack_ < end_); + JS_ASSERT(deltaRun_ <= end_); + } + + uint32_t nativeOffset() const { + return nativeOffset_; + } + uint32_t scriptDepth() const { + return scriptDepth_; + } + + class ScriptPcIterator + { + private: + uint32_t count_; + const uint8_t *start_; + const uint8_t *end_; + + uint32_t idx_; + const uint8_t *cur_; + + public: + ScriptPcIterator(uint32_t count, const uint8_t *start, const uint8_t *end) + : count_(count), start_(start), end_(end), idx_(0), cur_(start_) + {} + + bool hasMore() const + { + JS_ASSERT((idx_ == count_) == (cur_ == end_)); + JS_ASSERT((idx_ < count_) == (cur_ < end_)); + return cur_ < end_; + } + + void readNext(uint32_t *scriptIdxOut, uint32_t *pcOffsetOut) + { + JS_ASSERT(scriptIdxOut); + JS_ASSERT(pcOffsetOut); + JS_ASSERT(hasMore()); + + CompactBufferReader reader(cur_, end_); + ReadScriptPc(reader, scriptIdxOut, pcOffsetOut); + + cur_ = reader.currentPosition(); + JS_ASSERT(cur_ <= end_); + + idx_++; + JS_ASSERT_IF(idx_ == count_, cur_ == end_); + } + + void reset() { + idx_ = 0; + cur_ = start_; + } + }; + + ScriptPcIterator scriptPcIterator() const { + // End of script+pc sequence is the start of the delta run. + return ScriptPcIterator(scriptDepth_, scriptPcStack_, deltaRun_); + } + + class DeltaIterator { + private: + const uint8_t *start_; + const uint8_t *end_; + const uint8_t *cur_; + + public: + DeltaIterator(const uint8_t *start, const uint8_t *end) + : start_(start), end_(end), cur_(start) + {} + + bool hasMore() const + { + JS_ASSERT(cur_ <= end_); + return cur_ < end_; + } + + void readNext(uint32_t *nativeDeltaOut, int32_t *pcDeltaOut) + { + JS_ASSERT(nativeDeltaOut != nullptr); + JS_ASSERT(pcDeltaOut != nullptr); + + JS_ASSERT(hasMore()); + + CompactBufferReader reader(cur_, end_); + ReadDelta(reader, nativeDeltaOut, pcDeltaOut); + + cur_ = reader.currentPosition(); + JS_ASSERT(cur_ <= end_); + } + + void reset() { + cur_ = start_; + } + }; + DeltaIterator deltaIterator() const { + return DeltaIterator(deltaRun_, end_); + } + + uint32_t findPcOffset(uint32_t queryNativeOffset, uint32_t startPcOffset) const; +}; + +class JitcodeIonTable +{ + private: + /* Variable length payload section "below" here. */ + uint32_t numRegions_; + uint32_t regionOffsets_[0]; + + const uint8_t *payloadEnd() const { + return reinterpret_cast(this); + } + + public: + JitcodeIonTable(uint32_t numRegions) + : numRegions_(numRegions) + { + for (uint32_t i = 0; i < numRegions; i++) + regionOffsets_[i] = 0; + } + + bool makeIonEntry(JSContext *cx, JitCode *code, uint32_t numScripts, JSScript **scripts, + JitcodeGlobalEntry::IonEntry &out); + + uint32_t numRegions() const { + return numRegions_; + } + + uint32_t regionOffset(uint32_t regionIndex) const { + JS_ASSERT(regionIndex < numRegions()); + return regionOffsets_[regionIndex]; + } + + JitcodeRegionEntry regionEntry(uint32_t regionIndex) const { + const uint8_t *regionStart = payloadEnd() - regionOffset(regionIndex); + const uint8_t *regionEnd = payloadEnd(); + if (regionIndex < numRegions_ - 1) + regionEnd -= regionOffset(regionIndex + 1); + return JitcodeRegionEntry(regionStart, regionEnd); + } + + bool regionContainsOffset(uint32_t regionIndex, uint32_t nativeOffset) { + JS_ASSERT(regionIndex < numRegions()); + + JitcodeRegionEntry ent = regionEntry(regionIndex); + if (nativeOffset < ent.nativeOffset()) + return false; + + if (regionIndex == numRegions_ - 1) + return true; + + return nativeOffset < regionEntry(regionIndex + 1).nativeOffset(); + } + + uint32_t findRegionEntry(uint32_t offset) const; + + const uint8_t *payloadStart() const { + // The beginning of the payload the beginning of the first region are the same. + return payloadEnd() - regionOffset(0); + } + + static bool WriteIonTable(CompactBufferWriter &writer, + JSScript **scriptList, uint32_t scriptListSize, + const CodeGeneratorShared::NativeToBytecode *start, + const CodeGeneratorShared::NativeToBytecode *end, + uint32_t *tableOffsetOut, uint32_t *numRegionsOut); +}; + + +} // namespace jit +} // namespace js + +#endif /* jit_JitcodeMap_h */ diff --git a/js/src/jit/LIR-Common.h b/js/src/jit/LIR-Common.h index 37c7e0c3bd89..0d4387c69e63 100644 --- a/js/src/jit/LIR-Common.h +++ b/js/src/jit/LIR-Common.h @@ -128,6 +128,25 @@ class LMoveGroup : public LInstructionHelper<0, 0, 0> } }; +// Constructs a SIMD value with 4 components (e.g. int32x4, float32x4). +class LSimdValueX4 : public LInstructionHelper<1, 4, 0> +{ + public: + LIR_HEADER(SimdValueX4) + LSimdValueX4(const LAllocation &x, const LAllocation &y, + const LAllocation &z, const LAllocation &w) + { + setOperand(0, x); + setOperand(1, y); + setOperand(2, z); + setOperand(3, w); + } + + MSimdValueX4 *mir() const { + return mir_->toSimdValueX4(); + } +}; + // Extracts an element from a given SIMD int32x4 lane. class LSimdExtractElementI : public LInstructionHelper<1, 1, 0> { @@ -166,6 +185,42 @@ class LSimdExtractElementF : public LInstructionHelper<1, 1, 0> } }; +// Binary SIMD arithmetic operation between two SIMD operands +class LSimdBinaryArith : public LInstructionHelper<1, 2, 0> +{ + public: + LSimdBinaryArith() {} + + const LAllocation *lhs() { + return getOperand(0); + } + const LAllocation *rhs() { + return getOperand(1); + } + MSimdBinaryArith::Operation operation() const { + return mir_->toSimdBinaryArith()->operation(); + } + const char *extraName() const { + return MSimdBinaryArith::OperationName(operation()); + } +}; + +// Binary SIMD arithmetic operation between two Int32x4 operands +class LSimdBinaryArithIx4 : public LSimdBinaryArith +{ + public: + LIR_HEADER(SimdBinaryArithIx4); + LSimdBinaryArithIx4() : LSimdBinaryArith() {} +}; + +// Binary SIMD arithmetic operation between two Float32x4 operands +class LSimdBinaryArithFx4 : public LSimdBinaryArith +{ + public: + LIR_HEADER(SimdBinaryArithFx4); + LSimdBinaryArithFx4() : LSimdBinaryArith() {} +}; + // Constant 32-bit integer. class LInteger : public LInstructionHelper<1, 0, 0> { @@ -253,6 +308,26 @@ class LFloat32 : public LInstructionHelper<1, 0, 0> } }; +// Constant SIMD int32x4 +class LInt32x4 : public LInstructionHelper<1, 0, 0> +{ + public: + LIR_HEADER(Int32x4); + + explicit LInt32x4() {} + const SimdConstant &getValue() const { return mir_->toSimdConstant()->value(); } +}; + +// Constant SIMD float32x4 +class LFloat32x4 : public LInstructionHelper<1, 0, 0> +{ + public: + LIR_HEADER(Float32x4); + + explicit LFloat32x4() {} + const SimdConstant &getValue() const { return mir_->toSimdConstant()->value(); } +}; + // A constant Value. class LValue : public LInstructionHelper { @@ -725,6 +800,10 @@ class LCheckOverRecursed : public LInstructionHelper<0, 0, 0> LCheckOverRecursed() { } + + MCheckOverRecursed *mir() const { + return mir_->toCheckOverRecursed(); + } }; class LCheckOverRecursedPar : public LInstructionHelper<0, 1, 1> @@ -744,6 +823,10 @@ class LCheckOverRecursedPar : public LInstructionHelper<0, 1, 1> const LDefinition *getTempReg() { return getTemp(0); } + + MCheckOverRecursedPar *mir() const { + return mir_->toCheckOverRecursedPar(); + } }; class LAsmJSInterruptCheck : public LInstructionHelper<0, 0, 1> @@ -804,6 +887,9 @@ class LInterruptCheckImplicit : public LInstructionHelper<0, 0, 0> void setOolEntry(Label *oolEntry) { oolEntry_ = oolEntry; } + MInterruptCheck *mir() const { + return mir_->toInterruptCheck(); + } }; class LInterruptCheckPar : public LInstructionHelper<0, 1, 1> @@ -823,6 +909,9 @@ class LInterruptCheckPar : public LInstructionHelper<0, 1, 1> const LDefinition *getTempReg() { return getTemp(0); } + MInterruptCheckPar *mir() const { + return mir_->toInterruptCheckPar(); + } }; class LDefVar : public LCallInstructionHelper<0, 1, 0> @@ -2767,6 +2856,10 @@ class LAddI : public LBinaryMath<0> void setRecoversInput() { recoversInput_ = true; } + + MAdd *mir() const { + return mir_->toAdd(); + } }; // Subtracts two integers, returning an integer value. @@ -2791,6 +2884,9 @@ class LSubI : public LBinaryMath<0> void setRecoversInput() { recoversInput_ = true; } + MSub *mir() const { + return mir_->toSub(); + } }; // Performs an add, sub, mul, or div on two double values. @@ -3133,6 +3229,9 @@ class LValueToInt32 : public LInstructionHelper<1, BOX_PIECES, 2> JS_ASSERT(mode_ == TRUNCATE); return mir_->toTruncateToInt32(); } + MInstruction *mir() const { + return mir_->toInstruction(); + } }; // Convert a double to an int32. @@ -3187,6 +3286,10 @@ class LTruncateDToInt32 : public LInstructionHelper<1, 1, 1> const LDefinition *tempFloat() { return getTemp(0); } + + MTruncateToInt32 *mir() const { + return mir_->toTruncateToInt32(); + } }; // Convert a float32 to a truncated int32. @@ -3205,6 +3308,10 @@ class LTruncateFToInt32 : public LInstructionHelper<1, 1, 1> const LDefinition *tempFloat() { return getTemp(0); } + + MTruncateToInt32 *mir() const { + return mir_->toTruncateToInt32(); + } }; // Convert a boolean value to a string. @@ -3287,6 +3394,17 @@ class LStart : public LInstructionHelper<0, 0, 0> LIR_HEADER(Start) }; +// No-op instruction that prints nativeOffset, script, pcOffset during codegen. +class LPcOffset : public LInstructionHelper<0, 0, 0> +{ + public: + LIR_HEADER(PcOffset) + + const MPcOffset *mir() const { + return mir_->toPcOffset(); + } +}; + // Passed the BaselineFrame address in the OsrFrameReg by SideCannon(). // Forwards this object to the LOsrValues for Value materialization. class LOsrEntry : public LInstructionHelper<1, 0, 0> @@ -4267,6 +4385,27 @@ class LArrayConcat : public LCallInstructionHelper<1, 2, 2> } }; +class LArrayJoin : public LCallInstructionHelper<1, 2, 0> +{ + public: + LIR_HEADER(ArrayJoin) + + LArrayJoin(const LAllocation &array, const LAllocation &sep) { + setOperand(0, array); + setOperand(1, sep); + } + + const MArrayJoin *mir() const { + return mir_->toArrayJoin(); + } + const LAllocation *array() { + return getOperand(0); + } + const LAllocation *separator() { + return getOperand(1); + } +}; + // Load a typed value from a typed array's elements vector. class LLoadTypedArrayElement : public LInstructionHelper<1, 2, 1> { diff --git a/js/src/jit/LOpcodes.h b/js/src/jit/LOpcodes.h index f16b6126c94f..ddd52db342ce 100644 --- a/js/src/jit/LOpcodes.h +++ b/js/src/jit/LOpcodes.h @@ -16,8 +16,13 @@ _(Pointer) \ _(Double) \ _(Float32) \ + _(SimdValueX4) \ + _(Int32x4) \ + _(Float32x4) \ _(SimdExtractElementI) \ _(SimdExtractElementF) \ + _(SimdBinaryArithIx4) \ + _(SimdBinaryArithFx4) \ _(Value) \ _(CloneLiteral) \ _(Parameter) \ @@ -151,6 +156,7 @@ _(DoubleToString) \ _(ValueToString) \ _(Start) \ + _(PcOffset) \ _(OsrEntry) \ _(OsrValue) \ _(OsrScopeChain) \ @@ -200,6 +206,7 @@ _(ArrayPushV) \ _(ArrayPushT) \ _(ArrayConcat) \ + _(ArrayJoin) \ _(StoreElementHoleV) \ _(StoreElementHoleT) \ _(LoadTypedArrayElement) \ diff --git a/js/src/jit/Lowering.cpp b/js/src/jit/Lowering.cpp index 28a8563eb685..4c88da4a235c 100644 --- a/js/src/jit/Lowering.cpp +++ b/js/src/jit/Lowering.cpp @@ -1674,6 +1674,13 @@ LIRGenerator::visitStart(MStart *start) return add(lir); } +bool +LIRGenerator::visitPcOffset(MPcOffset *pcOffset) +{ + LPcOffset *lir = new(alloc()) LPcOffset; + return add(lir, pcOffset); +} + bool LIRGenerator::visitNop(MNop *nop) { @@ -2758,6 +2765,18 @@ LIRGenerator::visitArrayConcat(MArrayConcat *ins) return defineReturn(lir, ins) && assignSafepoint(lir, ins); } +bool +LIRGenerator::visitArrayJoin(MArrayJoin *ins) +{ + JS_ASSERT(ins->type() == MIRType_String); + JS_ASSERT(ins->array()->type() == MIRType_Object); + JS_ASSERT(ins->sep()->type() == MIRType_String); + + LArrayJoin *lir = new(alloc()) LArrayJoin(useRegisterAtStart(ins->array()), + useRegisterAtStart(ins->sep())); + return defineReturn(lir, ins) && assignSafepoint(lir, ins); +} + bool LIRGenerator::visitStringSplit(MStringSplit *ins) { @@ -3638,6 +3657,31 @@ LIRGenerator::visitRecompileCheck(MRecompileCheck *ins) return assignSafepoint(lir, ins); } +bool +LIRGenerator::visitSimdValueX4(MSimdValueX4 *ins) +{ + LAllocation x = useRegisterAtStart(ins->getOperand(0)); + LAllocation y = useRegisterAtStart(ins->getOperand(1)); + LAllocation z = useRegisterAtStart(ins->getOperand(2)); + LAllocation w = useRegisterAtStart(ins->getOperand(3)); + + return define(new(alloc()) LSimdValueX4(x, y, z, w), ins); +} + +bool +LIRGenerator::visitSimdConstant(MSimdConstant *ins) +{ + JS_ASSERT(IsSimdType(ins->type())); + + if (ins->type() == MIRType_Int32x4) + return define(new(alloc()) LInt32x4(), ins); + if (ins->type() == MIRType_Float32x4) + return define(new(alloc()) LFloat32x4(), ins); + + MOZ_ASSUME_UNREACHABLE("Unknown SIMD kind when generating constant"); + return false; +} + bool LIRGenerator::visitSimdExtractElement(MSimdExtractElement *ins) { @@ -3660,6 +3704,25 @@ LIRGenerator::visitSimdExtractElement(MSimdExtractElement *ins) return false; } +bool +LIRGenerator::visitSimdBinaryArith(MSimdBinaryArith *ins) +{ + JS_ASSERT(IsSimdType(ins->type())); + + if (ins->type() == MIRType_Int32x4) { + LSimdBinaryArithIx4 *add = new(alloc()) LSimdBinaryArithIx4(); + return lowerForFPU(add, ins, ins->lhs(), ins->rhs()); + } + + if (ins->type() == MIRType_Float32x4) { + LSimdBinaryArithFx4 *add = new(alloc()) LSimdBinaryArithFx4(); + return lowerForFPU(add, ins, ins->lhs(), ins->rhs()); + } + + MOZ_ASSUME_UNREACHABLE("Unknown SIMD kind when adding values"); + return false; +} + static void SpewResumePoint(MBasicBlock *block, MInstruction *ins, MResumePoint *resumePoint) { diff --git a/js/src/jit/Lowering.h b/js/src/jit/Lowering.h index 4cb3ea77d46c..a3715c8bf63c 100644 --- a/js/src/jit/Lowering.h +++ b/js/src/jit/Lowering.h @@ -141,6 +141,7 @@ class LIRGenerator : public LIRGeneratorSpecific bool visitFromCharCode(MFromCharCode *ins); bool visitStringSplit(MStringSplit *ins); bool visitStart(MStart *start); + bool visitPcOffset(MPcOffset *pcOffset); bool visitOsrEntry(MOsrEntry *entry); bool visitNop(MNop *nop); bool visitLimitedTruncate(MLimitedTruncate *nop); @@ -199,6 +200,7 @@ class LIRGenerator : public LIRGeneratorSpecific bool visitArrayPopShift(MArrayPopShift *ins); bool visitArrayPush(MArrayPush *ins); bool visitArrayConcat(MArrayConcat *ins); + bool visitArrayJoin(MArrayJoin *ins); bool visitLoadTypedArrayElement(MLoadTypedArrayElement *ins); bool visitLoadTypedArrayElementHole(MLoadTypedArrayElementHole *ins); bool visitLoadTypedArrayElementStatic(MLoadTypedArrayElementStatic *ins); @@ -264,6 +266,9 @@ class LIRGenerator : public LIRGeneratorSpecific bool visitGetDOMMember(MGetDOMMember *ins); bool visitRecompileCheck(MRecompileCheck *ins); bool visitSimdExtractElement(MSimdExtractElement *ins); + bool visitSimdBinaryArith(MSimdBinaryArith *ins); + bool visitSimdValueX4(MSimdValueX4 *ins); + bool visitSimdConstant(MSimdConstant *ins); }; } // namespace jit diff --git a/js/src/jit/MCallOptimize.cpp b/js/src/jit/MCallOptimize.cpp index 0d987e35eac4..2cffdbaaac80 100644 --- a/js/src/jit/MCallOptimize.cpp +++ b/js/src/jit/MCallOptimize.cpp @@ -42,6 +42,8 @@ IonBuilder::inlineNativeCall(CallInfo &callInfo, JSFunction *target) return inlineArrayPush(callInfo); if (native == js::array_concat) return inlineArrayConcat(callInfo); + if (native == js::array_join) + return inlineArrayJoin(callInfo); if (native == js::array_splice) return inlineArraySplice(callInfo); @@ -477,6 +479,29 @@ IonBuilder::inlineArraySplice(CallInfo &callInfo) return InliningStatus_Inlined; } +IonBuilder::InliningStatus +IonBuilder::inlineArrayJoin(CallInfo &callInfo) +{ + if (callInfo.argc() != 1 || callInfo.constructing()) + return InliningStatus_Error; + + if (getInlineReturnType() != MIRType_String) + return InliningStatus_Error; + if (callInfo.thisArg()->type() != MIRType_Object) + return InliningStatus_Error; + if (callInfo.getArg(0)->type() != MIRType_String) + return InliningStatus_Error; + + callInfo.setImplicitlyUsedUnchecked(); + + MArrayJoin *ins = MArrayJoin::New(alloc(), callInfo.thisArg(), callInfo.getArg(0)); + + current->add(ins); + current->push(ins); + + return InliningStatus_Inlined; +} + IonBuilder::InliningStatus IonBuilder::inlineArrayPush(CallInfo &callInfo) { diff --git a/js/src/jit/MIR.cpp b/js/src/jit/MIR.cpp index 05eaa8e24b60..e2b1efaf1271 100644 --- a/js/src/jit/MIR.cpp +++ b/js/src/jit/MIR.cpp @@ -30,6 +30,7 @@ using namespace js::jit; using mozilla::NumbersAreIdentical; using mozilla::IsFloat32Representable; using mozilla::Maybe; +using mozilla::DebugOnly; #ifdef DEBUG size_t MUse::index() const @@ -588,6 +589,39 @@ MConstant::canProduceFloat32() const return true; } +MDefinition* +MSimdValueX4::foldsTo(TempAllocator &alloc) +{ + DebugOnly scalarType = SimdTypeToScalarType(type()); + for (size_t i = 0; i < 4; ++i) { + MDefinition *op = getOperand(i); + if (!op->isConstant()) + return this; + JS_ASSERT(op->type() == scalarType); + } + + SimdConstant cst; + switch (type()) { + case MIRType_Int32x4: { + int32_t a[4]; + for (size_t i = 0; i < 4; ++i) + a[i] = getOperand(i)->toConstant()->value().toInt32(); + cst = SimdConstant::CreateX4(a); + break; + } + case MIRType_Float32x4: { + float a[4]; + for (size_t i = 0; i < 4; ++i) + a[i] = getOperand(i)->toConstant()->value().toNumber(); + cst = SimdConstant::CreateX4(a); + break; + } + default: MOZ_ASSUME_UNREACHABLE("unexpected type in MSimdValueX4::foldsTo"); + } + + return MSimdConstant::New(alloc, cst, type()); +} + MCloneLiteral * MCloneLiteral::New(TempAllocator &alloc, MDefinition *obj) { @@ -3421,6 +3455,32 @@ MBoundsCheck::foldsTo(TempAllocator &alloc) return this; } +MDefinition * +MArrayJoin::foldsTo(TempAllocator &alloc) { + MDefinition *arr = array(); + + if (!arr->isStringSplit()) + return this; + + this->setRecoveredOnBailout(); + if (arr->hasLiveDefUses()) { + this->setNotRecoveredOnBailout(); + return this; + } + + // We're replacing foo.split(bar).join(baz) by + // foo.replace(bar, baz). MStringSplit could be recovered by + // a bailout. As we are removing its last use, and its result + // could be captured by a resume point, this MStringSplit will + // be executed on the bailout path. + MDefinition *string = arr->toStringSplit()->string(); + MDefinition *pattern = arr->toStringSplit()->separator(); + MDefinition *replacement = sep(); + + setNotRecoveredOnBailout(); + return MStringReplace::New(alloc, string, pattern, replacement); +} + bool jit::ElementAccessIsDenseNative(MDefinition *obj, MDefinition *id) { diff --git a/js/src/jit/MIR.h b/js/src/jit/MIR.h index f0b95f43f06a..6f3e43760e19 100644 --- a/js/src/jit/MIR.h +++ b/js/src/jit/MIR.h @@ -13,6 +13,7 @@ #define jit_MIR_h #include "mozilla/Array.h" +#include "mozilla/DebugOnly.h" #include "jit/CompilerRoot.h" #include "jit/FixedList.h" @@ -1094,6 +1095,24 @@ class MStart : public MNullaryInstruction } }; +class MPcOffset : public MNullaryInstruction +{ + private: + MPcOffset() { + setGuard(); + } + + public: + INSTRUCTION_HEADER(PcOffset) + static MPcOffset *New(TempAllocator &alloc) { + return new(alloc) MPcOffset(); + } + + AliasSet getAliasSet() const { + return AliasSet::None(); + } +}; + // Instruction marking on entrypoint for on-stack replacement. // OSR may occur at loop headers (at JSOP_TRACE). // There is at most one MOsrEntry per MIRGraph. @@ -1233,6 +1252,77 @@ class MConstant : public MNullaryInstruction ALLOW_CLONE(MConstant) }; +// Generic constructor of SIMD valuesX4. +class MSimdValueX4 : public MQuaternaryInstruction +{ + protected: + MSimdValueX4(MIRType type, MDefinition *x, MDefinition *y, MDefinition *z, MDefinition *w) + : MQuaternaryInstruction(x, y, z, w) + { + JS_ASSERT(IsSimdType(type)); + mozilla::DebugOnly scalarType = SimdTypeToScalarType(type); + JS_ASSERT(scalarType == x->type()); + JS_ASSERT(scalarType == y->type()); + JS_ASSERT(scalarType == z->type()); + JS_ASSERT(scalarType == w->type()); + + setMovable(); + setResultType(type); + } + + public: + INSTRUCTION_HEADER(SimdValueX4) + + static MSimdValueX4 *New(TempAllocator &alloc, MIRType type, MDefinition *x, + MDefinition *y, MDefinition *z, MDefinition *w) + { + return new(alloc) MSimdValueX4(type, x, y, z, w); + } + + AliasSet getAliasSet() const { + return AliasSet::None(); + } + + bool congruentTo(const MDefinition *ins) const { + return congruentIfOperandsEqual(ins); + } + + MDefinition *foldsTo(TempAllocator &alloc); +}; + +// A constant SIMD value. +class MSimdConstant : public MNullaryInstruction +{ + SimdConstant value_; + + protected: + MSimdConstant(const SimdConstant &v, MIRType type) : value_(v) { + JS_ASSERT(IsSimdType(type)); + setResultType(type); + setMovable(); + } + + public: + INSTRUCTION_HEADER(SimdConstant); + static MSimdConstant *New(TempAllocator &alloc, const SimdConstant &v, MIRType type) { + return new(alloc) MSimdConstant(v, type); + } + + bool congruentTo(const MDefinition *ins) const { + if (!ins->isSimdConstant()) + return false; + return value() == ins->toSimdConstant()->value(); + } + + const SimdConstant &value() const { + return value_; + } + + AliasSet getAliasSet() const { + return AliasSet::None(); + } +}; + // Extracts a lane element from a given vector type, given by its lane symbol. class MSimdExtractElement : public MUnaryInstruction { @@ -1274,6 +1364,63 @@ class MSimdExtractElement : public MUnaryInstruction } }; +class MSimdBinaryArith : public MBinaryInstruction +{ + public: + enum Operation { + Add, + Sub, + Mul, + Div + }; + + static const char* OperationName(Operation op) { + switch (op) { + case Add: return "Add"; + case Sub: return "Sub"; + case Mul: return "Mul"; + case Div: return "Div"; + } + MOZ_ASSUME_UNREACHABLE("unexpected operation"); + } + + private: + Operation operation_; + + MSimdBinaryArith(MDefinition *left, MDefinition *right, Operation op, MIRType type) + : MBinaryInstruction(left, right), operation_(op) + { + JS_ASSERT_IF(type == MIRType_Int32x4, op == Add || op == Sub); + JS_ASSERT(IsSimdType(type)); + JS_ASSERT(left->type() == right->type()); + JS_ASSERT(left->type() == type); + setResultType(type); + setMovable(); + if (op == Add || op == Mul) + setCommutative(); + } + + public: + INSTRUCTION_HEADER(SimdBinaryArith); + static MSimdBinaryArith *NewAsmJS(TempAllocator &alloc, MDefinition *left, MDefinition *right, + Operation op, MIRType t) + { + return new(alloc) MSimdBinaryArith(left, right, op, t); + } + + AliasSet getAliasSet() const { + return AliasSet::None(); + } + + Operation operation() const { return operation_; } + + bool congruentTo(const MDefinition *ins) const { + if (!binaryCongruentTo(ins)) + return false; + return operation_ == ins->toSimdBinaryArith()->operation(); + } +}; + // Deep clone a constant JSObject. class MCloneLiteral : public MUnaryInstruction, @@ -7166,6 +7313,39 @@ class MArrayConcat } }; +class MArrayJoin + : public MBinaryInstruction, + public MixPolicy, StringPolicy<1> > +{ + MArrayJoin(MDefinition *array, MDefinition *sep) + : MBinaryInstruction(array, sep) + { + setResultType(MIRType_String); + } + public: + INSTRUCTION_HEADER(ArrayJoin) + static MArrayJoin *New(TempAllocator &alloc, MDefinition *array, MDefinition *sep) + { + return new (alloc) MArrayJoin(array, sep); + } + TypePolicy *typePolicy() { + return this; + } + MDefinition *array() const { + return getOperand(0); + } + MDefinition *sep() const { + return getOperand(1); + } + bool possiblyCalls() const { + return true; + } + virtual AliasSet getAliasSet() const { + return AliasSet::Load(AliasSet::Element | AliasSet::ObjectFields); + } + MDefinition *foldsTo(TempAllocator &alloc); +}; + class MLoadTypedArrayElement : public MBinaryInstruction { diff --git a/js/src/jit/MIRGenerator.h b/js/src/jit/MIRGenerator.h index 587983b2bbb4..a6503914563a 100644 --- a/js/src/jit/MIRGenerator.h +++ b/js/src/jit/MIRGenerator.h @@ -76,7 +76,21 @@ class MIRGenerator } bool instrumentedProfiling() { - return GetIonContext()->runtime->spsProfiler().enabled(); + if (!instrumentedProfilingIsCached_) { + instrumentedProfiling_ = GetIonContext()->runtime->spsProfiler().enabled(); + instrumentedProfilingIsCached_ = true; + } + return instrumentedProfiling_; + } + + bool isNativeToBytecodeMapEnabled() { + if (compilingAsmJS()) + return false; +#ifdef DEBUG + return true; +#else + return instrumentedProfiling(); +#endif } // Whether the main thread is trying to cancel this build. @@ -167,6 +181,9 @@ class MIRGenerator // slots is not compatible with that. bool modifiesFrameArguments_; + bool instrumentedProfiling_; + bool instrumentedProfilingIsCached_; + #if defined(JS_ION_PERF) AsmJSPerfSpewer asmJSPerfSpewer_; diff --git a/js/src/jit/MIRGraph.cpp b/js/src/jit/MIRGraph.cpp index c201402160ee..6fc60a706865 100644 --- a/js/src/jit/MIRGraph.cpp +++ b/js/src/jit/MIRGraph.cpp @@ -34,6 +34,8 @@ MIRGenerator::MIRGenerator(CompileCompartment *compartment, const JitCompileOpti usesSimdCached_(false), minAsmJSHeapLength_(AsmJSAllocationGranularity), modifiesFrameArguments_(false), + instrumentedProfiling_(false), + instrumentedProfilingIsCached_(false), options(options) { } diff --git a/js/src/jit/MOpcodes.h b/js/src/jit/MOpcodes.h index bd6a5bbc101e..3de0eb3aaf99 100644 --- a/js/src/jit/MOpcodes.h +++ b/js/src/jit/MOpcodes.h @@ -12,7 +12,10 @@ namespace jit { #define MIR_OPCODE_LIST(_) \ _(Constant) \ + _(SimdValueX4) \ + _(SimdConstant) \ _(SimdExtractElement) \ + _(SimdBinaryArith) \ _(CloneLiteral) \ _(Parameter) \ _(Callee) \ @@ -103,6 +106,7 @@ namespace jit { _(InitProp) \ _(InitPropGetterSetter) \ _(Start) \ + _(PcOffset) \ _(OsrEntry) \ _(Nop) \ _(LimitedTruncate) \ @@ -157,6 +161,7 @@ namespace jit { _(ArrayPopShift) \ _(ArrayPush) \ _(ArrayConcat) \ + _(ArrayJoin) \ _(LoadTypedArrayElement) \ _(LoadTypedArrayElementHole) \ _(LoadTypedArrayElementStatic) \ diff --git a/js/src/jit/ParallelSafetyAnalysis.cpp b/js/src/jit/ParallelSafetyAnalysis.cpp index 3f0379c6b8b9..e7573d26c52d 100644 --- a/js/src/jit/ParallelSafetyAnalysis.cpp +++ b/js/src/jit/ParallelSafetyAnalysis.cpp @@ -112,7 +112,10 @@ class ParallelSafetyVisitor : public MDefinitionVisitor // obviously safe for now. We can loosen as we need. SAFE_OP(Constant) + SAFE_OP(SimdValueX4) + SAFE_OP(SimdConstant) SAFE_OP(SimdExtractElement) + SAFE_OP(SimdBinaryArith) UNSAFE_OP(CloneLiteral) SAFE_OP(Parameter) SAFE_OP(Callee) @@ -196,6 +199,7 @@ class ParallelSafetyVisitor : public MDefinitionVisitor UNSAFE_OP(InitProp) UNSAFE_OP(InitPropGetterSetter) SAFE_OP(Start) + SAFE_OP(PcOffset) UNSAFE_OP(OsrEntry) SAFE_OP(Nop) SAFE_OP(LimitedTruncate) @@ -287,6 +291,7 @@ class ParallelSafetyVisitor : public MDefinitionVisitor SAFE_OP(NewCallObjectPar) SAFE_OP(LambdaPar) UNSAFE_OP(ArrayConcat) + UNSAFE_OP(ArrayJoin) UNSAFE_OP(GetDOMProperty) UNSAFE_OP(GetDOMMember) UNSAFE_OP(SetDOMProperty) diff --git a/js/src/jit/VMFunctions.cpp b/js/src/jit/VMFunctions.cpp index 18cacf5e9a48..307b8cd598a9 100644 --- a/js/src/jit/VMFunctions.cpp +++ b/js/src/jit/VMFunctions.cpp @@ -457,6 +457,44 @@ ArrayConcatDense(JSContext *cx, HandleObject obj1, HandleObject obj2, HandleObje return &argv[0].toObject(); } +JSString * +ArrayJoin(JSContext *cx, HandleObject array, HandleString sep) +{ + // The annotations in this function follow the first steps of join + // specified in ES5. + + // Step 1 + RootedObject obj(cx, array); + if (!obj) + return nullptr; + + AutoCycleDetector detector(cx, obj); + if (!detector.init()) + return nullptr; + + if (detector.foundCycle()) + return nullptr; + + // Steps 2 and 3 + uint32_t length; + if (!GetLengthProperty(cx, obj, &length)) + return nullptr; + + // Steps 4 and 5 + RootedLinearString sepstr(cx); + if (sep) { + sepstr = sep->ensureLinear(cx); + if (!sepstr) + return nullptr; + } else { + sepstr = cx->names().comma; + } + + // Step 6 to 11 + return js::ArrayJoin(cx, obj, sepstr, length); +} + + bool CharCodeAt(JSContext *cx, HandleString str, int32_t index, uint32_t *code) { diff --git a/js/src/jit/VMFunctions.h b/js/src/jit/VMFunctions.h index f303e9e0a824..527f47ca6a9e 100644 --- a/js/src/jit/VMFunctions.h +++ b/js/src/jit/VMFunctions.h @@ -631,6 +631,7 @@ bool ArrayPopDense(JSContext *cx, HandleObject obj, MutableHandleValue rval); bool ArrayPushDense(JSContext *cx, HandleObject obj, HandleValue v, uint32_t *length); bool ArrayShiftDense(JSContext *cx, HandleObject obj, MutableHandleValue rval); JSObject *ArrayConcatDense(JSContext *cx, HandleObject obj1, HandleObject obj2, HandleObject res); +JSString *ArrayJoin(JSContext *cx, HandleObject array, HandleString sep); bool CharCodeAt(JSContext *cx, HandleString str, int32_t index, uint32_t *code); JSFlatString *StringFromCharCode(JSContext *cx, int32_t code); diff --git a/js/src/jit/arm/Architecture-arm.cpp b/js/src/jit/arm/Architecture-arm.cpp index 6c38a3e134a9..f70eed5bf1ee 100644 --- a/js/src/jit/arm/Architecture-arm.cpp +++ b/js/src/jit/arm/Architecture-arm.cpp @@ -16,28 +16,98 @@ #include "jit/arm/Assembler-arm.h" #include "jit/RegisterSets.h" -#define HWCAP_USE_HARDFP_ABI (1 << 27) - -#if !(defined(ANDROID) || defined(MOZ_B2G)) && !defined(JS_ARM_SIMULATOR) -#define HWCAP_ARMv7 (1 << 28) -#include +#if defined(ANDROID) || defined(MOZ_B2G) || defined(JS_ARM_SIMULATOR) +// The Android NDK does not include the hwcap.h kernel header, and it is not +// defined when building the simulator, so inline the header defines we need. +# define HWCAP_VFP (1 << 6) +# define HWCAP_NEON (1 << 12) +# define HWCAP_VFPv3 (1 << 13) +# define HWCAP_VFPv3D16 (1 << 14) /* also set for VFPv4-D16 */ +# define HWCAP_VFPv4 (1 << 16) +# define HWCAP_IDIVA (1 << 17) +# define HWCAP_IDIVT (1 << 18) +# define HWCAP_VFPD32 (1 << 19) /* set if VFP has 32 regs (not 16) */ +# define AT_HWCAP 16 #else -#define HWCAP_VFP (1<<0) -#define HWCAP_VFPv3 (1<<1) -#define HWCAP_VFPv3D16 (1<<2) -#define HWCAP_VFPv4 (1<<3) -#define HWCAP_IDIVA (1<<4) -#define HWCAP_IDIVT (1<<5) -#define HWCAP_NEON (1<<6) -#define HWCAP_ARMv7 (1<<7) +# include +# if !defined(HWCAP_IDIVA) +# define HWCAP_IDIVA (1 << 17) +# endif +# if !defined(HWCAP_VFPD32) +# define HWCAP_VFPD32 (1 << 19) /* set if VFP has 32 regs (not 16) */ +# endif #endif +// Not part of the HWCAP flag, but we need to know this, and this bit is not +// used so we are using it. +#define HWCAP_ARMv7 (1 << 28) + +// Also take a bit to flag the use of the hardfp ABI. +#define HWCAP_USE_HARDFP_ABI (1 << 27) + namespace js { namespace jit { + +// Parse the Linux kernel cpuinfo features. This is also used to parse the +// override features which has some extensions: 'armv7' and 'hardfp'. +uint32_t +ParseARMCpuFeatures(const char *features, bool override = false) +{ + uint32_t flags = 0; + + for (;;) { + char ch = *features; + if (!ch) { + // End of string. + break; + } + if (ch == ' ' || ch == ',') { + // Skip separator characters. + features++; + continue; + } + // Find the end of the token. + const char *end = features + 1; + for (; ; end++) { + ch = *end; + if (!ch || ch == ' ' || ch == ',') + break; + } + size_t count = end - features; + if (count == 3 && strncmp(features, "vfp", 3) == 0) + flags |= HWCAP_VFP; + else if (count == 4 && strncmp(features, "neon", 4) == 0) + flags |= HWCAP_NEON; + else if (count == 5 && strncmp(features, "vfpv3", 5) == 0) + flags |= HWCAP_VFPv3; + else if (count == 8 && strncmp(features, "vfpv3d16", 8) == 0) + flags |= HWCAP_VFPv3D16; + else if (count == 5 && strncmp(features, "vfpv4", 5) == 0) + flags |= HWCAP_VFPv4; + else if (count == 5 && strncmp(features, "idiva", 5) == 0) + flags |= HWCAP_IDIVA; + else if (count == 5 && strncmp(features, "idivt", 5) == 0) + flags |= HWCAP_IDIVT; + else if (count == 6 && strncmp(features, "vfpd32", 6) == 0) + flags |= HWCAP_VFPD32; + else if (count == 5 && strncmp(features, "armv7", 5) == 0) + flags |= HWCAP_ARMv7; +#if defined(JS_ARM_SIMULATOR) + else if (count == 6 && strncmp(features, "hardfp", 6) == 0) + flags |= HWCAP_USE_HARDFP_ABI; +#endif + else if (override) + fprintf(stderr, "Warning: unexpected ARM feature at: %s\n", features); + features = end; + } + IonSpew(IonSpew_Codegen, "ARM features: '%s'\n flags: 0x%x\n", features, flags); + return flags; +} + // The override flags parsed from the ARMHWCAP environment variable or from the // --arm-hwcap js shell argument. -static uint32_t armHwCapFlags = 0; +volatile static uint32_t armHwCapFlags = 0; bool ParseARMHwCapFlags(const char *armHwCap) @@ -65,6 +135,7 @@ ParseARMHwCapFlags(const char *armHwCap) " vfpv4 \n" " idiva \n" " idivt \n" + " vfpd32 \n" #if defined(JS_ARM_SIMULATOR) " hardfp \n" #endif @@ -74,168 +145,108 @@ ParseARMHwCapFlags(const char *armHwCap) /*NOTREACHED*/ } - // Canonicalize each token to have a leading and trailing space. - const char *start = armHwCap; // Token start. - for (;;) { - char ch = *start; - if (!ch) { - // End of string. - break; - } - if (ch == ' ' || ch == ',') { - // Skip separator characters. - start++; - continue; - } - // Find the end of the token. - const char *end = start + 1; - for (; ; end++) { - ch = *end; - if (!ch || ch == ' ' || ch == ',') - break; - } - size_t count = end - start; - if (count == 3 && strncmp(start, "vfp", 3) == 0) - flags |= HWCAP_VFP; - else if (count == 5 && strncmp(start, "vfpv3", 5) == 0) - flags |= HWCAP_VFPv3; - else if (count == 8 && strncmp(start, "vfpv3d16", 8) == 0) - flags |= HWCAP_VFPv3D16; - else if (count == 5 && strncmp(start, "vfpv4", 5) == 0) - flags |= HWCAP_VFPv4; - else if (count == 5 && strncmp(start, "idiva", 5) == 0) - flags |= HWCAP_IDIVA; - else if (count == 5 && strncmp(start, "idivt", 5) == 0) - flags |= HWCAP_IDIVT; - else if (count == 4 && strncmp(start, "neon", 4) == 0) - flags |= HWCAP_NEON; - else if (count == 5 && strncmp(start, "armv7", 5) == 0) - flags |= HWCAP_ARMv7; -#if defined(JS_ARM_SIMULATOR) - else if (count == 6 && strncmp(start, "hardfp", 6) == 0) - flags |= HWCAP_USE_HARDFP_ABI; -#endif - else - fprintf(stderr, "Warning: unexpected ARMHWCAP flag at: %s\n", start); - start = end; - } -#ifdef DEBUG - IonSpew(IonSpew_Codegen, "ARMHWCAP: '%s'\n flags: 0x%x\n", armHwCap, flags); -#endif - armHwCapFlags = flags; + armHwCapFlags = ParseARMCpuFeatures(armHwCap, /* override = */ true); return true; } uint32_t GetARMFlags() { - static bool isSet = false; - static uint32_t flags = 0; + volatile static bool isSet = false; + volatile static uint32_t flags = 0; if (isSet) return flags; const char *env = getenv("ARMHWCAP"); if (ParseARMHwCapFlags(env) || armHwCapFlags) { - isSet = true; flags = armHwCapFlags; + isSet = true; return flags; } -#ifdef JS_CODEGEN_ARM_HARDFP - flags |= HWCAP_USE_HARDFP_ABI; -#endif - #ifdef JS_ARM_SIMULATOR - isSet = true; flags = HWCAP_ARMv7 | HWCAP_VFP | HWCAP_VFPv3 | HWCAP_VFPv4 | HWCAP_NEON; - return flags; #else -#if WTF_OS_LINUX +#if defined(WTF_OS_LINUX) || defined(WTF_OS_ANDROID) || defined(MOZ_B2G) + bool readAuxv = false; int fd = open("/proc/self/auxv", O_RDONLY); if (fd > 0) { - Elf32_auxv_t aux; - while (read(fd, &aux, sizeof(Elf32_auxv_t))) { + struct { uint32_t a_type; uint32_t a_val; } aux; + while (read(fd, &aux, sizeof(aux))) { if (aux.a_type == AT_HWCAP) { - close(fd); - flags = aux.a_un.a_val; - isSet = true; -#if defined(__ARM_ARCH_7__) || defined (__ARM_ARCH_7A__) - // This should really be detected at runtime, but /proc/*/auxv - // doesn't seem to carry the ISA. We could look in /proc/cpuinfo - // as well, but the chances that it will be different from this - // are low. - flags |= HWCAP_ARMv7; -#endif - return flags; + flags = aux.a_val; + readAuxv = true; + break; } } close(fd); } -#if defined(__ARM_ARCH_7__) || defined (__ARM_ARCH_7A__) - flags = HWCAP_ARMv7; -#endif - isSet = true; - return flags; - -#elif defined(WTF_OS_ANDROID) || defined(MOZ_B2G) - FILE *fp = fopen("/proc/cpuinfo", "r"); - if (!fp) - return false; - - char buf[1024]; - memset(buf, 0, sizeof(buf)); - size_t len = fread(buf, sizeof(char), sizeof(buf) - 2, fp); - fclose(fp); - // Canonicalize each token to have a leading and trailing space. - buf[len] = ' '; - buf[len + 1] = '\0'; - for (size_t i = 0; i < len; i++) { - char ch = buf[i]; - if (!ch) - break; - else if (ch == '\n') - buf[i] = 0x20; - else - buf[i] = ch; + if (!readAuxv) { + // Read the Features if the auxv is not available. + FILE *fp = fopen("/proc/cpuinfo", "r"); + if (fp) { + char buf[1024]; + memset(buf, 0, sizeof(buf)); + size_t len = fread(buf, sizeof(char), sizeof(buf) - 1, fp); + fclose(fp); + buf[len] = '\0'; + char *featureList = strstr(buf, "Features"); + if (featureList) { + if (char *featuresEnd = strstr(featureList, "\n")) + *featuresEnd = '\0'; + flags = ParseARMCpuFeatures(featureList + 8); + } + if (strstr(buf, "ARMv7")) + flags |= HWCAP_ARMv7; + } } +#endif - if (strstr(buf, " vfp ")) - flags |= HWCAP_VFP; + // If compiled to use specialized features then these features can be + // assumed to be present otherwise the compiler would fail to run. - if (strstr(buf, " vfpv3 ")) +#ifdef JS_CODEGEN_ARM_HARDFP + // Compiled to use the hardfp ABI. + flags |= HWCAP_USE_HARDFP_ABI; +#endif + +#if defined(__VFP_FP__) && !defined(__SOFTFP__) + // Compiled to use VFP instructions so assume VFP support. + flags |= HWCAP_VFP; +#endif + +#if defined(__ARM_ARCH_7__) || defined (__ARM_ARCH_7A__) + // Compiled to use ARMv7 instructions so assume the ARMv7 arch. + flags |= HWCAP_ARMv7; +#endif + +#endif // JS_ARM_SIMULATOR + + // Canonicalize the flags. These rules are also applied to the features + // supplied for simulation. + + // The VFPv3 feature is expected when the VFPv3D16 is reported, but add it + // just in case of a kernel difference in feature reporting. + if (flags & HWCAP_VFPv3D16) flags |= HWCAP_VFPv3; - if (strstr(buf, " vfpv3d16 ")) - flags |= HWCAP_VFPv3D16; - - if (strstr(buf, " vfpv4 ")) - flags |= HWCAP_VFPv4; - - if (strstr(buf, " idiva ")) - flags |= HWCAP_IDIVA; - - if (strstr(buf, " idivt ")) - flags |= HWCAP_IDIVT; - - if (strstr(buf, " neon ")) - flags |= HWCAP_NEON; - - // Not part of the HWCAP flag, but we need to know this, and we're not using - // that bit, so... we are using it. - if (strstr(buf, "ARMv7")) + // If VFPv3 or Neon is supported then this must be an ARMv7. + if (flags & (HWCAP_VFPv3 | HWCAP_NEON)) flags |= HWCAP_ARMv7; -#ifdef DEBUG - IonSpew(IonSpew_Codegen, "ARMHWCAP: '%s'\n flags: 0x%x\n", buf, flags); -#endif + // Some old kernels report VFP and not VFPv3, but if ARMv7 then it must be + // VFPv3. + if (flags & HWCAP_VFP && flags & HWCAP_ARMv7) + flags |= HWCAP_VFPv3; + // Older kernels do not implement the HWCAP_VFPD32 flag. + if ((flags & HWCAP_VFPv3) && !(flags & HWCAP_VFPv3D16)) + flags |= HWCAP_VFPD32; + + IonSpew(IonSpew_Codegen, "ARM HWCAP: 0x%x\n", flags); isSet = true; return flags; -#endif - - return 0; -#endif // JS_ARM_SIMULATOR } bool HasMOVWT() @@ -253,20 +264,12 @@ bool HasVFP() bool Has32DP() { - return (GetARMFlags() & HWCAP_VFPv3) && !(GetARMFlags() & HWCAP_VFPv3D16); -} -bool UseConvReg() -{ - return Has32DP(); + return GetARMFlags() & HWCAP_VFPD32; } bool HasIDIV() { -#if defined HWCAP_IDIVA return GetARMFlags() & HWCAP_IDIVA; -#else - return false; -#endif } // This is defined in the header and inlined when not using the simulator. diff --git a/js/src/jit/arm/CodeGenerator-arm.cpp b/js/src/jit/arm/CodeGenerator-arm.cpp index f4b2c01afcf4..4511f52760da 100644 --- a/js/src/jit/arm/CodeGenerator-arm.cpp +++ b/js/src/jit/arm/CodeGenerator-arm.cpp @@ -180,8 +180,12 @@ CodeGeneratorARM::bailoutIf(Assembler::Condition condition, LSnapshot *snapshot) // We could not use a jump table, either because all bailout IDs were // reserved, or a jump table is not optimal for this frame size or // platform. Whatever, we will generate a lazy bailout. + InlineScriptTree *tree = snapshot->mir()->block()->trackedTree(); OutOfLineBailout *ool = new(alloc()) OutOfLineBailout(snapshot, masm.framePushed()); - if (!addOutOfLineCode(ool)) + + // All bailout code is associated with the bytecodeSite of the block we are + // bailing out from. + if (!addOutOfLineCode(ool, BytecodeSite(tree, tree->script()->code()))) return false; masm.ma_b(ool->entry(), condition); @@ -206,10 +210,13 @@ CodeGeneratorARM::bailoutFrom(Label *label, LSnapshot *snapshot) frameClass_.frameSize() == masm.framePushed()); // On ARM we don't use a bailout table. + InlineScriptTree *tree = snapshot->mir()->block()->trackedTree(); OutOfLineBailout *ool = new(alloc()) OutOfLineBailout(snapshot, masm.framePushed()); - if (!addOutOfLineCode(ool)) { + + // All bailout code is associated with the bytecodeSite of the block we are + // bailing out from. + if (!addOutOfLineCode(ool, BytecodeSite(tree, tree->script()->code()))) return false; - } masm.retarget(label, ool->entry()); @@ -1104,7 +1111,7 @@ CodeGeneratorARM::emitTableSwitchDispatch(MTableSwitch *mir, Register index, Reg if (!ool->addCodeLabel(cl)) return false; } - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mir)) return false; return true; @@ -1253,13 +1260,15 @@ CodeGeneratorARM::emitRoundDouble(FloatRegister src, Register dest, Label *fail) bool CodeGeneratorARM::visitTruncateDToInt32(LTruncateDToInt32 *ins) { - return emitTruncateDouble(ToFloatRegister(ins->input()), ToRegister(ins->output())); + return emitTruncateDouble(ToFloatRegister(ins->input()), ToRegister(ins->output()), + ins->mir()); } bool CodeGeneratorARM::visitTruncateFToInt32(LTruncateFToInt32 *ins) { - return emitTruncateFloat32(ToFloatRegister(ins->input()), ToRegister(ins->output())); + return emitTruncateFloat32(ToFloatRegister(ins->input()), ToRegister(ins->output()), + ins->mir()); } static const uint32_t FrameSizes[] = { 128, 256, 512, 1024 }; diff --git a/js/src/jit/arm/CodeGenerator-arm.h b/js/src/jit/arm/CodeGenerator-arm.h index ca2e3e0030a3..040019a8475b 100644 --- a/js/src/jit/arm/CodeGenerator-arm.h +++ b/js/src/jit/arm/CodeGenerator-arm.h @@ -227,8 +227,13 @@ class CodeGeneratorARM : public CodeGeneratorShared public: // Unimplemented SIMD instructions + bool visitSimdValueX4(LSimdValueX4 *lir) { MOZ_ASSUME_UNREACHABLE("NYI"); } + bool visitInt32x4(LInt32x4 *ins) { MOZ_ASSUME_UNREACHABLE("NYI"); } + bool visitFloat32x4(LFloat32x4 *ins) { MOZ_ASSUME_UNREACHABLE("NYI"); } bool visitSimdExtractElementI(LSimdExtractElementI *ins) { MOZ_ASSUME_UNREACHABLE("NYI"); } bool visitSimdExtractElementF(LSimdExtractElementF *ins) { MOZ_ASSUME_UNREACHABLE("NYI"); } + bool visitSimdBinaryArithIx4(LSimdBinaryArithIx4 *lir) { MOZ_ASSUME_UNREACHABLE("NYI"); } + bool visitSimdBinaryArithFx4(LSimdBinaryArithFx4 *lir) { MOZ_ASSUME_UNREACHABLE("NYI"); } }; typedef CodeGeneratorARM CodeGeneratorSpecific; diff --git a/js/src/jit/arm/MacroAssembler-arm.h b/js/src/jit/arm/MacroAssembler-arm.h index 80943cbfdddf..02fad0a5888b 100644 --- a/js/src/jit/arm/MacroAssembler-arm.h +++ b/js/src/jit/arm/MacroAssembler-arm.h @@ -1332,6 +1332,16 @@ class MacroAssemblerARMCompat : public MacroAssemblerARM void loadPrivate(const Address &address, Register dest); + void loadAlignedInt32x4(const Address &addr, FloatRegister dest) { MOZ_ASSUME_UNREACHABLE("NYI"); } + void storeAlignedInt32x4(FloatRegister src, Address addr) { MOZ_ASSUME_UNREACHABLE("NYI"); } + void loadUnalignedInt32x4(const Address &addr, FloatRegister dest) { MOZ_ASSUME_UNREACHABLE("NYI"); } + void storeUnalignedInt32x4(FloatRegister src, Address addr) { MOZ_ASSUME_UNREACHABLE("NYI"); } + + void loadAlignedFloat32x4(const Address &addr, FloatRegister dest) { MOZ_ASSUME_UNREACHABLE("NYI"); } + void storeAlignedFloat32x4(FloatRegister src, Address addr) { MOZ_ASSUME_UNREACHABLE("NYI"); } + void loadUnalignedFloat32x4(const Address &addr, FloatRegister dest) { MOZ_ASSUME_UNREACHABLE("NYI"); } + void storeUnalignedFloat32x4(FloatRegister src, Address addr) { MOZ_ASSUME_UNREACHABLE("NYI"); } + void loadDouble(const Address &addr, FloatRegister dest); void loadDouble(const BaseIndex &src, FloatRegister dest); diff --git a/js/src/jit/mips/CodeGenerator-mips.cpp b/js/src/jit/mips/CodeGenerator-mips.cpp index 4fa1b93ebf30..0a4aa0018dd5 100644 --- a/js/src/jit/mips/CodeGenerator-mips.cpp +++ b/js/src/jit/mips/CodeGenerator-mips.cpp @@ -1365,13 +1365,15 @@ CodeGeneratorMIPS::visitRoundF(LRoundF *lir) bool CodeGeneratorMIPS::visitTruncateDToInt32(LTruncateDToInt32 *ins) { - return emitTruncateDouble(ToFloatRegister(ins->input()), ToRegister(ins->output())); + return emitTruncateDouble(ToFloatRegister(ins->input()), ToRegister(ins->output()), + ins->mir()); } bool CodeGeneratorMIPS::visitTruncateFToInt32(LTruncateFToInt32 *ins) { - return emitTruncateFloat32(ToFloatRegister(ins->input()), ToRegister(ins->output())); + return emitTruncateFloat32(ToFloatRegister(ins->input()), ToRegister(ins->output()), + ins->mir()); } static const uint32_t FrameSizes[] = { 128, 256, 512, 1024 }; diff --git a/js/src/jit/shared/Assembler-x86-shared.h b/js/src/jit/shared/Assembler-x86-shared.h index bce51a09839a..ca92ad3e09a0 100644 --- a/js/src/jit/shared/Assembler-x86-shared.h +++ b/js/src/jit/shared/Assembler-x86-shared.h @@ -340,6 +340,14 @@ class AssemblerX86Shared : public AssemblerShared label->bind(masm.size()); masm.floatConstant(f); } + void writeInt32x4Constant(const SimdConstant &v, Label *label) { + label->bind(masm.size()); + masm.int32x4Constant(v.asInt32x4()); + } + void writeFloat32x4Constant(const SimdConstant &v, Label *label) { + label->bind(masm.size()); + masm.float32x4Constant(v.asFloat32x4()); + } void movl(Imm32 imm32, Register dest) { masm.movl_i32r(imm32.value, dest.code()); } @@ -1463,6 +1471,106 @@ class AssemblerX86Shared : public AssemblerShared JS_ASSERT(HasSSE2()); masm.movd_rr(src.code(), dest.code()); } + void paddd(const Operand &src, FloatRegister dest) { + JS_ASSERT(HasSSE2()); + switch (src.kind()) { + case Operand::FPREG: + masm.paddd_rr(src.fpu(), dest.code()); + break; + case Operand::MEM_REG_DISP: + masm.paddd_mr(src.disp(), src.base(), dest.code()); + break; + case Operand::MEM_ADDRESS32: + masm.paddd_mr(src.address(), dest.code()); + break; + default: + MOZ_ASSUME_UNREACHABLE("unexpected operand kind"); + } + } + void psubd(const Operand &src, FloatRegister dest) { + JS_ASSERT(HasSSE2()); + switch (src.kind()) { + case Operand::FPREG: + masm.psubd_rr(src.fpu(), dest.code()); + break; + case Operand::MEM_REG_DISP: + masm.psubd_mr(src.disp(), src.base(), dest.code()); + break; + case Operand::MEM_ADDRESS32: + masm.psubd_mr(src.address(), dest.code()); + break; + default: + MOZ_ASSUME_UNREACHABLE("unexpected operand kind"); + } + } + void addps(const Operand &src, FloatRegister dest) { + JS_ASSERT(HasSSE2()); + switch (src.kind()) { + case Operand::FPREG: + masm.addps_rr(src.fpu(), dest.code()); + break; + case Operand::MEM_REG_DISP: + masm.addps_mr(src.disp(), src.base(), dest.code()); + break; + case Operand::MEM_ADDRESS32: + masm.addps_mr(src.address(), dest.code()); + break; + default: + MOZ_ASSUME_UNREACHABLE("unexpected operand kind"); + } + } + void subps(const Operand &src, FloatRegister dest) { + JS_ASSERT(HasSSE2()); + switch (src.kind()) { + case Operand::FPREG: + masm.subps_rr(src.fpu(), dest.code()); + break; + case Operand::MEM_REG_DISP: + masm.subps_mr(src.disp(), src.base(), dest.code()); + break; + case Operand::MEM_ADDRESS32: + masm.subps_mr(src.address(), dest.code()); + break; + default: + MOZ_ASSUME_UNREACHABLE("unexpected operand kind"); + } + } + void mulps(const Operand &src, FloatRegister dest) { + JS_ASSERT(HasSSE2()); + switch (src.kind()) { + case Operand::FPREG: + masm.mulps_rr(src.fpu(), dest.code()); + break; + case Operand::MEM_REG_DISP: + masm.mulps_mr(src.disp(), src.base(), dest.code()); + break; + case Operand::MEM_ADDRESS32: + masm.mulps_mr(src.address(), dest.code()); + break; + default: + MOZ_ASSUME_UNREACHABLE("unexpected operand kind"); + } + } + void divps(const Operand &src, FloatRegister dest) { + JS_ASSERT(HasSSE2()); + switch (src.kind()) { + case Operand::FPREG: + masm.divps_rr(src.fpu(), dest.code()); + break; + case Operand::MEM_REG_DISP: + masm.divps_mr(src.disp(), src.base(), dest.code()); + break; + case Operand::MEM_ADDRESS32: + masm.divps_mr(src.address(), dest.code()); + break; + default: + MOZ_ASSUME_UNREACHABLE("unexpected operand kind"); + } + } + void pxor(FloatRegister src, FloatRegister dest) { + JS_ASSERT(HasSSE2()); + masm.pxor_rr(src.code(), dest.code()); + } void pshufd(uint32_t mask, FloatRegister src, FloatRegister dest) { JS_ASSERT(HasSSE2()); masm.pshufd_irr(mask, src.code(), dest.code()); diff --git a/js/src/jit/shared/CodeGenerator-shared.cpp b/js/src/jit/shared/CodeGenerator-shared.cpp index 1e87f97dea3c..c089f7311c23 100644 --- a/js/src/jit/shared/CodeGenerator-shared.cpp +++ b/js/src/jit/shared/CodeGenerator-shared.cpp @@ -8,9 +8,11 @@ #include "mozilla/DebugOnly.h" +#include "jit/CompactBuffer.h" #include "jit/IonCaches.h" #include "jit/IonMacroAssembler.h" #include "jit/IonSpewer.h" +#include "jit/JitcodeMap.h" #include "jit/MIR.h" #include "jit/MIRGenerator.h" #include "jit/ParallelFunctions.h" @@ -49,6 +51,12 @@ CodeGeneratorShared::CodeGeneratorShared(MIRGenerator *gen, LIRGraph *graph, Mac pushedArgs_(0), #endif lastOsiPointOffset_(0), + nativeToBytecodeMap_(nullptr), + nativeToBytecodeMapSize_(0), + nativeToBytecodeTableOffset_(0), + nativeToBytecodeNumRegions_(0), + nativeToBytecodeScriptList_(nullptr), + nativeToBytecodeScriptListLength_(0), sps_(&GetIonContext()->runtime->spsProfiler(), &lastNotInlinedPC_), osrEntryOffset_(0), skipArgCheckEntryOffset_(0), @@ -123,6 +131,13 @@ CodeGeneratorShared::generateOutOfLineCode() { JSScript *topScript = sps_.getPushed(); for (size_t i = 0; i < outOfLineCode_.length(); i++) { + // Add native => bytecode mapping entries for OOL sites. + // Not enabled on asm.js yet since asm doesn't contain bytecode mappings. + if (!gen->compilingAsmJS()) { + if (!addNativeToBytecodeEntry(outOfLineCode_[i]->bytecodeSite())) + return false; + } + if (!gen->alloc().ensureBallast()) return false; @@ -147,20 +162,134 @@ CodeGeneratorShared::generateOutOfLineCode() } bool -CodeGeneratorShared::addOutOfLineCode(OutOfLineCode *code) +CodeGeneratorShared::addOutOfLineCode(OutOfLineCode *code, const MInstruction *mir) +{ + JS_ASSERT(mir); + return addOutOfLineCode(code, mir->trackedSite()); +} + +bool +CodeGeneratorShared::addOutOfLineCode(OutOfLineCode *code, const BytecodeSite &site) { code->setFramePushed(masm.framePushed()); - // If an OOL instruction adds another OOL instruction, then use the original - // instruction's script/pc instead of the basic block's that we're on - // because they're probably not relevant any more. - if (oolIns) - code->setSource(oolIns->script(), oolIns->pc()); - else - code->setSource(current ? current->mir()->info().script() : nullptr, lastPC_); - JS_ASSERT_IF(code->script(), code->script()->containsPC(code->pc())); + code->setBytecodeSite(site); + JS_ASSERT_IF(!gen->compilingAsmJS(), code->script()->containsPC(code->pc())); return outOfLineCode_.append(code); } +bool +CodeGeneratorShared::addNativeToBytecodeEntry(const BytecodeSite &site) +{ + // Skip the table entirely if profiling is not enabled. + if (!isNativeToBytecodeMapEnabled()) + return true; + + JS_ASSERT(site.tree()); + JS_ASSERT(site.pc()); + + InlineScriptTree *tree = site.tree(); + jsbytecode *pc = site.pc(); + uint32_t nativeOffset = masm.currentOffset(); + + JS_ASSERT_IF(nativeToBytecodeList_.empty(), nativeOffset == 0); + + if (!nativeToBytecodeList_.empty()) { + size_t lastIdx = nativeToBytecodeList_.length() - 1; + NativeToBytecode &lastEntry = nativeToBytecodeList_[lastIdx]; + + JS_ASSERT(nativeOffset >= lastEntry.nativeOffset.offset()); + + // If the new entry is for the same inlineScriptTree and same + // bytecodeOffset, but the nativeOffset has changed, do nothing. + // The same site just generated some more code. + if (lastEntry.tree == tree && lastEntry.pc == pc) { + IonSpew(IonSpew_Profiling, " => In-place update [%u-%u]", + lastEntry.nativeOffset.offset(), nativeOffset); + return true; + } + + // If the new entry is for the same native offset, then update the + // previous entry with the new bytecode site, since the previous + // bytecode site did not generate any native code. + if (lastEntry.nativeOffset.offset() == nativeOffset) { + lastEntry.tree = tree; + lastEntry.pc = pc; + IonSpew(IonSpew_Profiling, " => Overwriting zero-length native region."); + + // This overwrite might have made the entry merge-able with a + // previous one. If so, merge it. + if (lastIdx > 0) { + NativeToBytecode &nextToLastEntry = nativeToBytecodeList_[lastIdx - 1]; + if (nextToLastEntry.tree == lastEntry.tree && nextToLastEntry.pc == lastEntry.pc) { + IonSpew(IonSpew_Profiling, " => Merging with previous region"); + nativeToBytecodeList_.erase(&lastEntry); + } + } + + dumpNativeToBytecodeEntry(nativeToBytecodeList_.length() - 1); + return true; + } + } + + // Otherwise, some native code was generated for the previous bytecode site. + // Add a new entry for code that is about to be generated. + NativeToBytecode entry; + entry.nativeOffset = CodeOffsetLabel(nativeOffset); + entry.tree = tree; + entry.pc = pc; + if (!nativeToBytecodeList_.append(entry)) + return false; + + IonSpew(IonSpew_Profiling, " => Push new entry."); + dumpNativeToBytecodeEntry(nativeToBytecodeList_.length() - 1); + return true; +} + +void +CodeGeneratorShared::dumpNativeToBytecodeEntries() +{ +#ifdef DEBUG + InlineScriptTree *topTree = gen->info().inlineScriptTree(); + IonSpewStart(IonSpew_Profiling, "Native To Bytecode Entries for %s:%d\n", + topTree->script()->filename(), topTree->script()->lineno()); + for (unsigned i = 0; i < nativeToBytecodeList_.length(); i++) + dumpNativeToBytecodeEntry(i); +#endif +} + +void +CodeGeneratorShared::dumpNativeToBytecodeEntry(uint32_t idx) +{ +#ifdef DEBUG + NativeToBytecode &ref = nativeToBytecodeList_[idx]; + InlineScriptTree *tree = ref.tree; + JSScript *script = tree->script(); + uint32_t nativeOffset = ref.nativeOffset.offset(); + unsigned nativeDelta = 0; + unsigned pcDelta = 0; + if (idx + 1 < nativeToBytecodeList_.length()) { + NativeToBytecode *nextRef = &ref + 1; + nativeDelta = nextRef->nativeOffset.offset() - nativeOffset; + if (nextRef->tree == ref.tree) + pcDelta = nextRef->pc - ref.pc; + } + IonSpewStart(IonSpew_Profiling, " %08x [+%-6d] => %-6d [%-4d] {%-10s} (%s:%d", + ref.nativeOffset.offset(), + nativeDelta, + ref.pc - script->code(), + pcDelta, + js_CodeName[JSOp(*ref.pc)], + script->filename(), script->lineno()); + + for (tree = tree->caller(); tree; tree = tree->caller()) { + IonSpewCont(IonSpew_Profiling, " <= %s:%d", tree->script()->filename(), + tree->script()->lineno()); + } + IonSpewCont(IonSpew_Profiling, ")"); + IonSpewFin(IonSpew_Profiling); +#endif +} + // see OffsetOfFrameSlot static inline int32_t ToStackIndex(LAllocation *a) @@ -415,6 +544,206 @@ CodeGeneratorShared::encodeSafepoints() } } +bool +CodeGeneratorShared::createNativeToBytecodeScriptList(JSContext *cx) +{ + js::Vector scriptList; + InlineScriptTree *tree = gen->info().inlineScriptTree(); + for (;;) { + // Add script from current tree. + bool found = false; + for (uint32_t i = 0; i < scriptList.length(); i++) { + if (scriptList[i] == tree->script()) { + found = true; + break; + } + } + if (!found) { + if (!scriptList.append(tree->script())) + return false; + } + + // Process rest of tree + + // If children exist, emit children. + if (tree->hasChildren()) { + tree = tree->firstChild(); + continue; + } + + // Otherwise, find the first tree up the chain (including this one) + // that contains a next sibling. + while (!tree->hasNextCallee() && tree->hasCaller()) + tree = tree->caller(); + + // If we found a sibling, use it. + if (tree->hasNextCallee()) { + tree = tree->nextCallee(); + continue; + } + + // Otherwise, we must have reached the top without finding any siblings. + JS_ASSERT(tree->isOutermostCaller()); + break; + } + + // Allocate array for list. + JSScript **data = (JSScript **) cx->malloc_(scriptList.length() * sizeof(JSScript **)); + if (!data) + return false; + + for (uint32_t i = 0; i < scriptList.length(); i++) + data[i] = scriptList[i]; + + // Success. + nativeToBytecodeScriptListLength_ = scriptList.length(); + nativeToBytecodeScriptList_ = data; + return true; +} + +bool +CodeGeneratorShared::generateCompactNativeToBytecodeMap(JSContext *cx, JitCode *code) +{ + JS_ASSERT(nativeToBytecodeScriptListLength_ == 0); + JS_ASSERT(nativeToBytecodeScriptList_ == nullptr); + JS_ASSERT(nativeToBytecodeMap_ == nullptr); + JS_ASSERT(nativeToBytecodeMapSize_ == 0); + JS_ASSERT(nativeToBytecodeTableOffset_ == 0); + JS_ASSERT(nativeToBytecodeNumRegions_ == 0); + + // Iterate through all nativeToBytecode entries, fix up their masm offsets. + for (unsigned i = 0; i < nativeToBytecodeList_.length(); i++) { + NativeToBytecode &entry = nativeToBytecodeList_[i]; + + // Fixup code offsets. + entry.nativeOffset = CodeOffsetLabel(masm.actualOffset(entry.nativeOffset.offset())); + } + + if (!createNativeToBytecodeScriptList(cx)) + return false; + + JS_ASSERT(nativeToBytecodeScriptListLength_ > 0); + JS_ASSERT(nativeToBytecodeScriptList_ != nullptr); + + CompactBufferWriter writer; + uint32_t tableOffset = 0; + uint32_t numRegions = 0; + + if (!JitcodeIonTable::WriteIonTable( + writer, nativeToBytecodeScriptList_, nativeToBytecodeScriptListLength_, + &nativeToBytecodeList_[0], + &nativeToBytecodeList_[0] + nativeToBytecodeList_.length(), + &tableOffset, &numRegions)) + { + return false; + } + + JS_ASSERT(tableOffset > 0); + JS_ASSERT(numRegions > 0); + + // Writer is done, copy it to sized buffer. + uint8_t *data = (uint8_t *) cx->malloc_(writer.length()); + if (!data) + return false; + + memcpy(data, writer.buffer(), writer.length()); + nativeToBytecodeMap_ = data; + nativeToBytecodeMapSize_ = writer.length(); + nativeToBytecodeTableOffset_ = tableOffset; + nativeToBytecodeNumRegions_ = numRegions; + + verifyCompactNativeToBytecodeMap(code); + + IonSpew(IonSpew_Profiling, "Compact Native To Bytecode Map [%p-%p]", + data, data + nativeToBytecodeMapSize_); + + return true; +} + +void +CodeGeneratorShared::verifyCompactNativeToBytecodeMap(JitCode *code) +{ +#ifdef DEBUG + JS_ASSERT(nativeToBytecodeScriptListLength_ > 0); + JS_ASSERT(nativeToBytecodeScriptList_ != nullptr); + JS_ASSERT(nativeToBytecodeMap_ != nullptr); + JS_ASSERT(nativeToBytecodeMapSize_ > 0); + JS_ASSERT(nativeToBytecodeTableOffset_ > 0); + JS_ASSERT(nativeToBytecodeNumRegions_ > 0); + + // The pointer to the table must be 4-byte aligned + const uint8_t *tablePtr = nativeToBytecodeMap_ + nativeToBytecodeTableOffset_; + JS_ASSERT(uintptr_t(tablePtr) % sizeof(uint32_t) == 0); + + // Verify that numRegions was encoded correctly. + const JitcodeIonTable *ionTable = reinterpret_cast(tablePtr); + JS_ASSERT(ionTable->numRegions() == nativeToBytecodeNumRegions_); + + // Region offset for first region should be at the start of the payload region. + // Since the offsets are backward from the start of the table, the first entry + // backoffset should be equal to the forward table offset from the start of the + // allocated data. + JS_ASSERT(ionTable->regionOffset(0) == nativeToBytecodeTableOffset_); + + // Verify each region. + for (uint32_t i = 0; i < ionTable->numRegions(); i++) { + // Back-offset must point into the payload region preceding the table, not before it. + JS_ASSERT(ionTable->regionOffset(i) <= nativeToBytecodeTableOffset_); + + // Back-offset must point to a later area in the payload region than previous + // back-offset. This means that back-offsets decrease monotonically. + JS_ASSERT_IF(i > 0, ionTable->regionOffset(i) < ionTable->regionOffset(i - 1)); + + JitcodeRegionEntry entry = ionTable->regionEntry(i); + + // Ensure native code offset for region falls within jitcode. + JS_ASSERT(entry.nativeOffset() <= code->instructionsSize()); + + // Read out script/pc stack and verify. + JitcodeRegionEntry::ScriptPcIterator scriptPcIter = entry.scriptPcIterator(); + while (scriptPcIter.hasMore()) { + uint32_t scriptIdx = 0, pcOffset = 0; + scriptPcIter.readNext(&scriptIdx, &pcOffset); + + // Ensure scriptIdx refers to a valid script in the list. + JS_ASSERT(scriptIdx < nativeToBytecodeScriptListLength_); + JSScript *script = nativeToBytecodeScriptList_[scriptIdx]; + + // Ensure pcOffset falls within the script. + JS_ASSERT(pcOffset < script->length()); + } + + // Obtain the original nativeOffset and pcOffset and script. + uint32_t curNativeOffset = entry.nativeOffset(); + JSScript *script = nullptr; + uint32_t curPcOffset = 0; + { + uint32_t scriptIdx = 0; + scriptPcIter.reset(); + scriptPcIter.readNext(&scriptIdx, &curPcOffset); + script = nativeToBytecodeScriptList_[scriptIdx]; + } + + // Read out nativeDeltas and pcDeltas and verify. + JitcodeRegionEntry::DeltaIterator deltaIter = entry.deltaIterator(); + while (deltaIter.hasMore()) { + uint32_t nativeDelta = 0; + int32_t pcDelta = 0; + deltaIter.readNext(&nativeDelta, &pcDelta); + + curNativeOffset += nativeDelta; + curPcOffset = uint32_t(int32_t(curPcOffset) + pcDelta); + + // Ensure that nativeOffset still falls within jitcode after delta. + JS_ASSERT(curNativeOffset <= code->instructionsSize()); + + // Ensure that pcOffset still falls within bytecode after delta. + JS_ASSERT(curPcOffset < script->length()); + } + } +#endif // DEBUG +} + bool CodeGeneratorShared::markSafepoint(LInstruction *ins) { @@ -766,18 +1095,18 @@ class OutOfLineTruncateSlow : public OutOfLineCodeBase }; OutOfLineCode * -CodeGeneratorShared::oolTruncateDouble(FloatRegister src, Register dest) +CodeGeneratorShared::oolTruncateDouble(FloatRegister src, Register dest, MInstruction *mir) { OutOfLineTruncateSlow *ool = new(alloc()) OutOfLineTruncateSlow(src, dest); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mir)) return nullptr; return ool; } bool -CodeGeneratorShared::emitTruncateDouble(FloatRegister src, Register dest) +CodeGeneratorShared::emitTruncateDouble(FloatRegister src, Register dest, MInstruction *mir) { - OutOfLineCode *ool = oolTruncateDouble(src, dest); + OutOfLineCode *ool = oolTruncateDouble(src, dest, mir); if (!ool) return false; @@ -787,10 +1116,10 @@ CodeGeneratorShared::emitTruncateDouble(FloatRegister src, Register dest) } bool -CodeGeneratorShared::emitTruncateFloat32(FloatRegister src, Register dest) +CodeGeneratorShared::emitTruncateFloat32(FloatRegister src, Register dest, MInstruction *mir) { OutOfLineTruncateSlow *ool = new(alloc()) OutOfLineTruncateSlow(src, dest, true); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mir)) return false; masm.branchTruncateFloat32(src, dest, ool->entry()); diff --git a/js/src/jit/shared/CodeGenerator-shared.h b/js/src/jit/shared/CodeGenerator-shared.h index 76843002274f..530c4e8974db 100644 --- a/js/src/jit/shared/CodeGenerator-shared.h +++ b/js/src/jit/shared/CodeGenerator-shared.h @@ -97,11 +97,32 @@ class CodeGeneratorShared : public LInstructionVisitor js::Vector patchableTLScripts_; #endif + public: + struct NativeToBytecode { + CodeOffsetLabel nativeOffset; + InlineScriptTree *tree; + jsbytecode *pc; + }; + + protected: + js::Vector nativeToBytecodeList_; + uint8_t *nativeToBytecodeMap_; + uint32_t nativeToBytecodeMapSize_; + uint32_t nativeToBytecodeTableOffset_; + uint32_t nativeToBytecodeNumRegions_; + + JSScript **nativeToBytecodeScriptList_; + uint32_t nativeToBytecodeScriptListLength_; + // When profiling is enabled, this is the instrumentation manager which // maintains state of what script is currently being generated (for inline // scripts) and when instrumentation needs to be emitted or skipped. IonInstrumentation sps_; + bool isNativeToBytecodeMapEnabled() { + return gen->isNativeToBytecodeMapEnabled(); + } + protected: // The offset of the first instruction of the OSR entry block from the // beginning of the code buffer. @@ -223,6 +244,10 @@ class CodeGeneratorShared : public LInstructionVisitor void verifyOsiPointRegs(LSafepoint *safepoint); #endif + bool addNativeToBytecodeEntry(const BytecodeSite &site); + void dumpNativeToBytecodeEntries(); + void dumpNativeToBytecodeEntry(uint32_t idx); + public: MIRGenerator &mirGen() const { return *gen; @@ -289,6 +314,11 @@ class CodeGeneratorShared : public LInstructionVisitor // safepoint offsets. void encodeSafepoints(); + // Fixup offsets of native-to-bytecode map. + bool createNativeToBytecodeScriptList(JSContext *cx); + bool generateCompactNativeToBytecodeMap(JSContext *cx, JitCode *code); + void verifyCompactNativeToBytecodeMap(JitCode *code); + // Mark the safepoint on |ins| as corresponding to the current assembler location. // The location should be just after a call. bool markSafepoint(LInstruction *ins); @@ -307,9 +337,9 @@ class CodeGeneratorShared : public LInstructionVisitor // an invalidation marker. void ensureOsiSpace(); - OutOfLineCode *oolTruncateDouble(FloatRegister src, Register dest); - bool emitTruncateDouble(FloatRegister src, Register dest); - bool emitTruncateFloat32(FloatRegister src, Register dest); + OutOfLineCode *oolTruncateDouble(FloatRegister src, Register dest, MInstruction *mir); + bool emitTruncateDouble(FloatRegister src, Register dest, MInstruction *mir); + bool emitTruncateFloat32(FloatRegister src, Register dest, MInstruction *mir); void emitPreBarrier(Register base, const LAllocation *index, MIRType type); void emitPreBarrier(Address address, MIRType type); @@ -438,7 +468,8 @@ class CodeGeneratorShared : public LInstructionVisitor ReciprocalMulConstants computeDivisionConstants(int d); protected: - bool addOutOfLineCode(OutOfLineCode *code); + bool addOutOfLineCode(OutOfLineCode *code, const MInstruction *mir); + bool addOutOfLineCode(OutOfLineCode *code, const BytecodeSite &site); bool hasOutOfLineCode() { return !outOfLineCode_.empty(); } bool generateOutOfLineCode(); @@ -498,14 +529,12 @@ class OutOfLineCode : public TempObject Label entry_; Label rejoin_; uint32_t framePushed_; - jsbytecode *pc_; - JSScript *script_; + BytecodeSite site_; public: OutOfLineCode() : framePushed_(0), - pc_(nullptr), - script_(nullptr) + site_() { } virtual bool generate(CodeGeneratorShared *codegen) = 0; @@ -525,15 +554,17 @@ class OutOfLineCode : public TempObject uint32_t framePushed() const { return framePushed_; } - void setSource(JSScript *script, jsbytecode *pc) { - script_ = script; - pc_ = pc; + void setBytecodeSite(const BytecodeSite &site) { + site_ = site; } - jsbytecode *pc() { - return pc_; + const BytecodeSite &bytecodeSite() const { + return site_; } - JSScript *script() { - return script_; + jsbytecode *pc() const { + return site_.pc(); + } + JSScript *script() const { + return site_.script(); } }; @@ -729,8 +760,11 @@ inline OutOfLineCode * CodeGeneratorShared::oolCallVM(const VMFunction &fun, LInstruction *lir, const ArgSeq &args, const StoreOutputTo &out) { + JS_ASSERT(lir->mirRaw()); + JS_ASSERT(lir->mirRaw()->isInstruction()); + OutOfLineCode *ool = new(alloc()) OutOfLineCallVM(lir, fun, args, out); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, lir->mirRaw()->toInstruction())) return nullptr; return ool; } diff --git a/js/src/jit/shared/CodeGenerator-x86-shared.cpp b/js/src/jit/shared/CodeGenerator-x86-shared.cpp index 077ca7c61d56..aaa228242701 100644 --- a/js/src/jit/shared/CodeGenerator-x86-shared.cpp +++ b/js/src/jit/shared/CodeGenerator-x86-shared.cpp @@ -420,8 +420,12 @@ CodeGeneratorX86Shared::bailout(const T &binder, LSnapshot *snapshot) // We could not use a jump table, either because all bailout IDs were // reserved, or a jump table is not optimal for this frame size or // platform. Whatever, we will generate a lazy bailout. + // + // All bailout code is associated with the bytecodeSite of the block we are + // bailing out from. + InlineScriptTree *tree = snapshot->mir()->block()->trackedTree(); OutOfLineBailout *ool = new(alloc()) OutOfLineBailout(snapshot); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, BytecodeSite(tree, tree->script()->code()))) return false; binder(masm, ool->entry()); @@ -623,7 +627,7 @@ CodeGeneratorX86Shared::visitAddI(LAddI *ins) if (ins->snapshot()) { if (ins->recoversInput()) { OutOfLineUndoALUOperation *ool = new(alloc()) OutOfLineUndoALUOperation(ins); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; masm.j(Assembler::Overflow, ool->entry()); } else { @@ -645,7 +649,7 @@ CodeGeneratorX86Shared::visitSubI(LSubI *ins) if (ins->snapshot()) { if (ins->recoversInput()) { OutOfLineUndoALUOperation *ool = new(alloc()) OutOfLineUndoALUOperation(ins); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; masm.j(Assembler::Overflow, ool->entry()); } else { @@ -763,7 +767,7 @@ CodeGeneratorX86Shared::visitMulI(LMulI *ins) if (mul->canBeNegativeZero()) { // Jump to an OOL path if the result is 0. MulNegativeZeroCheck *ool = new(alloc()) MulNegativeZeroCheck(ins); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mul)) return false; masm.testl(ToRegister(lhs), ToRegister(lhs)); @@ -843,7 +847,7 @@ CodeGeneratorX86Shared::visitUDivOrMod(LUDivOrMod *ins) } if (ool) { - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; masm.bind(ool->rejoin()); } @@ -1090,7 +1094,7 @@ CodeGeneratorX86Shared::visitDivI(LDivI *ins) masm.bind(&done); if (ool) { - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mir)) return false; masm.bind(ool->rejoin()); } @@ -1276,13 +1280,13 @@ CodeGeneratorX86Shared::visitModI(LModI *ins) masm.bind(&done); if (overflow) { - if (!addOutOfLineCode(overflow)) + if (!addOutOfLineCode(overflow, ins->mir())) return false; masm.bind(overflow->done()); } if (ool) { - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; masm.bind(ool->rejoin()); } @@ -1488,7 +1492,7 @@ CodeGeneratorX86Shared::emitTableSwitchDispatch(MTableSwitch *mir, Register inde // generate the case entries (we don't yet know their offsets in the // instruction stream). OutOfLineTableSwitch *ool = new(alloc()) OutOfLineTableSwitch(mir); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mir)) return false; // Compute the position where a pointer to the right case stands. @@ -2053,6 +2057,57 @@ CodeGeneratorX86Shared::visitNegF(LNegF *ins) return true; } +bool +CodeGeneratorX86Shared::visitInt32x4(LInt32x4 *ins) +{ + const LDefinition *out = ins->getDef(0); + masm.loadConstantInt32x4(ins->getValue(), ToFloatRegister(out)); + return true; +} + +bool +CodeGeneratorX86Shared::visitFloat32x4(LFloat32x4 *ins) +{ + const LDefinition *out = ins->getDef(0); + masm.loadConstantFloat32x4(ins->getValue(), ToFloatRegister(out)); + return true; +} + +bool +CodeGeneratorX86Shared::visitSimdValueX4(LSimdValueX4 *ins) +{ + FloatRegister output = ToFloatRegister(ins->output()); + + MSimdValueX4 *mir = ins->mir(); + JS_ASSERT(IsSimdType(mir->type())); + JS_STATIC_ASSERT(sizeof(float) == sizeof(int32_t)); + + masm.reserveStack(Simd128DataSize); + // TODO see bug 1051860 for possible optimizations. + switch (mir->type()) { + case MIRType_Int32x4: { + for (size_t i = 0; i < 4; ++i) { + Register r = ToRegister(ins->getOperand(i)); + masm.store32(r, Address(StackPointer, i * sizeof(int32_t))); + } + masm.loadAlignedInt32x4(Address(StackPointer, 0), output); + break; + } + case MIRType_Float32x4: { + for (size_t i = 0; i < 4; ++i) { + FloatRegister r = ToFloatRegister(ins->getOperand(i)); + masm.storeFloat32(r, Address(StackPointer, i * sizeof(float))); + } + masm.loadAlignedFloat32x4(Address(StackPointer, 0), output); + break; + } + default: MOZ_ASSUME_UNREACHABLE("Unknown SIMD kind"); + } + + masm.freeStack(Simd128DataSize); + return true; +} + bool CodeGeneratorX86Shared::visitSimdExtractElementI(LSimdExtractElementI *ins) { @@ -2092,6 +2147,56 @@ CodeGeneratorX86Shared::visitSimdExtractElementF(LSimdExtractElementF *ins) return true; } +bool +CodeGeneratorX86Shared::visitSimdBinaryArithIx4(LSimdBinaryArithIx4 *ins) +{ + FloatRegister lhs = ToFloatRegister(ins->lhs()); + Operand rhs = ToOperand(ins->rhs()); + JS_ASSERT(ToFloatRegister(ins->output()) == lhs); + + MSimdBinaryArith::Operation op = ins->operation(); + switch (op) { + case MSimdBinaryArith::Add: + masm.packedAddInt32(rhs, lhs); + return true; + case MSimdBinaryArith::Sub: + masm.packedSubInt32(rhs, lhs); + return true; + case MSimdBinaryArith::Mul: + // we can do mul with a single instruction only if we have SSE4.1 + // using the PMULLD instruction. + case MSimdBinaryArith::Div: + // x86 doesn't have SIMD i32 div. + break; + } + MOZ_ASSUME_UNREACHABLE("unexpected SIMD op"); +} + +bool +CodeGeneratorX86Shared::visitSimdBinaryArithFx4(LSimdBinaryArithFx4 *ins) +{ + FloatRegister lhs = ToFloatRegister(ins->lhs()); + Operand rhs = ToOperand(ins->rhs()); + JS_ASSERT(ToFloatRegister(ins->output()) == lhs); + + MSimdBinaryArith::Operation op = ins->operation(); + switch (op) { + case MSimdBinaryArith::Add: + masm.packedAddFloat32(rhs, lhs); + return true; + case MSimdBinaryArith::Sub: + masm.packedSubFloat32(rhs, lhs); + return true; + case MSimdBinaryArith::Mul: + masm.packedMulFloat32(rhs, lhs); + return true; + case MSimdBinaryArith::Div: + masm.packedDivFloat32(rhs, lhs); + return true; + } + MOZ_ASSUME_UNREACHABLE("unexpected SIMD op"); +} + bool CodeGeneratorX86Shared::visitForkJoinGetSlice(LForkJoinGetSlice *ins) { diff --git a/js/src/jit/shared/CodeGenerator-x86-shared.h b/js/src/jit/shared/CodeGenerator-x86-shared.h index 6d33809f36bb..9b2bc6a74caf 100644 --- a/js/src/jit/shared/CodeGenerator-x86-shared.h +++ b/js/src/jit/shared/CodeGenerator-x86-shared.h @@ -205,8 +205,13 @@ class CodeGeneratorX86Shared : public CodeGeneratorShared bool visitNegF(LNegF *lir); // SIMD operators + bool visitSimdValueX4(LSimdValueX4 *lir); + bool visitInt32x4(LInt32x4 *ins); + bool visitFloat32x4(LFloat32x4 *ins); bool visitSimdExtractElementI(LSimdExtractElementI *lir); bool visitSimdExtractElementF(LSimdExtractElementF *lir); + bool visitSimdBinaryArithIx4(LSimdBinaryArithIx4 *lir); + bool visitSimdBinaryArithFx4(LSimdBinaryArithFx4 *lir); // Out of line visitors. bool visitOutOfLineBailout(OutOfLineBailout *ool); diff --git a/js/src/jit/shared/MacroAssembler-x86-shared.h b/js/src/jit/shared/MacroAssembler-x86-shared.h index 6c49f9900162..f540f823ff80 100644 --- a/js/src/jit/shared/MacroAssembler-x86-shared.h +++ b/js/src/jit/shared/MacroAssembler-x86-shared.h @@ -482,6 +482,12 @@ class MacroAssemblerX86Shared : public Assembler void storeUnalignedInt32x4(FloatRegister src, const Address &dest) { movdqu(src, Operand(dest)); } + void packedAddInt32(const Operand &src, FloatRegister dest) { + paddd(src, dest); + } + void packedSubInt32(const Operand &src, FloatRegister dest) { + psubd(src, dest); + } void loadAlignedFloat32x4(const Address &src, FloatRegister dest) { movaps(Operand(src), dest); @@ -498,6 +504,18 @@ class MacroAssemblerX86Shared : public Assembler void storeUnalignedFloat32x4(FloatRegister src, const Address &dest) { movups(src, Operand(dest)); } + void packedAddFloat32(const Operand &src, FloatRegister dest) { + addps(src, dest); + } + void packedSubFloat32(const Operand &src, FloatRegister dest) { + subps(src, dest); + } + void packedMulFloat32(const Operand &src, FloatRegister dest) { + mulps(src, dest); + } + void packedDivFloat32(const Operand &src, FloatRegister dest) { + divps(src, dest); + } static uint32_t ComputeShuffleMask(SimdLane x, SimdLane y = LaneX, SimdLane z = LaneX, SimdLane w = LaneX) @@ -665,6 +683,30 @@ class MacroAssemblerX86Shared : public Assembler return false; } + bool maybeInlineInt32x4(const SimdConstant &v, const FloatRegister &dest) { + static const SimdConstant zero = SimdConstant::CreateX4(0, 0, 0, 0); + static const SimdConstant minusOne = SimdConstant::CreateX4(-1, -1, -1, -1); + if (v == zero) { + pxor(dest, dest); + return true; + } + if (v == minusOne) { + pcmpeqw(dest, dest); + return true; + } + return false; + } + bool maybeInlineFloat32x4(const SimdConstant &v, const FloatRegister &dest) { + static const SimdConstant zero = SimdConstant::CreateX4(0.f, 0.f, 0.f, 0.f); + if (v == zero) { + // This won't get inlined if the SimdConstant v contains -0 in any + // lane, as operator== here does a memcmp. + xorps(dest, dest); + return true; + } + return false; + } + void convertBoolToInt32(Register source, Register dest) { // Note that C++ bool is only 1 byte, so zero extend it to clear the // higher-order bits. diff --git a/js/src/jit/x64/CodeGenerator-x64.cpp b/js/src/jit/x64/CodeGenerator-x64.cpp index b9b38684229e..8ebabe805b1e 100644 --- a/js/src/jit/x64/CodeGenerator-x64.cpp +++ b/js/src/jit/x64/CodeGenerator-x64.cpp @@ -262,7 +262,7 @@ CodeGeneratorX64::visitAsmJSLoadHeap(LAsmJSLoadHeap *ins) if (!mir->skipBoundsCheck()) { bool isFloat32Load = vt == Scalar::Float32; ool = new(alloc()) OutOfLineLoadTypedArrayOutOfBounds(ToAnyRegister(out), isFloat32Load); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; CodeOffsetLabel cmp = masm.cmplWithPatch(ToRegister(ptr), Imm32(0)); @@ -416,7 +416,7 @@ CodeGeneratorX64::visitTruncateDToInt32(LTruncateDToInt32 *ins) // On x64, branchTruncateDouble uses cvttsd2sq. Unlike the x86 // implementation, this should handle most doubles and we can just // call a stub if it fails. - return emitTruncateDouble(input, output); + return emitTruncateDouble(input, output, ins->mir()); } bool @@ -428,5 +428,5 @@ CodeGeneratorX64::visitTruncateFToInt32(LTruncateFToInt32 *ins) // On x64, branchTruncateFloat32 uses cvttss2sq. Unlike the x86 // implementation, this should handle most floats and we can just // call a stub if it fails. - return emitTruncateFloat32(input, output); + return emitTruncateFloat32(input, output, ins->mir()); } diff --git a/js/src/jit/x64/MacroAssembler-x64.cpp b/js/src/jit/x64/MacroAssembler-x64.cpp index f159169e87d7..65a3038df749 100644 --- a/js/src/jit/x64/MacroAssembler-x64.cpp +++ b/js/src/jit/x64/MacroAssembler-x64.cpp @@ -79,6 +79,66 @@ MacroAssemblerX64::loadConstantFloat32(float f, FloatRegister dest) masm.setNextJump(j, prev); } +MacroAssemblerX64::SimdData * +MacroAssemblerX64::getSimdData(const SimdConstant &v) +{ + if (!simdMap_.initialized()) { + enoughMemory_ &= simdMap_.init(); + if (!enoughMemory_) + return nullptr; + } + + size_t index; + if (SimdMap::AddPtr p = simdMap_.lookupForAdd(v)) { + index = p->value(); + } else { + index = simds_.length(); + enoughMemory_ &= simds_.append(SimdData(v)); + enoughMemory_ &= simdMap_.add(p, v, index); + if (!enoughMemory_) + return nullptr; + } + return &simds_[index]; +} + +void +MacroAssemblerX64::loadConstantInt32x4(const SimdConstant &v, FloatRegister dest) +{ + JS_ASSERT(v.type() == SimdConstant::Int32x4); + if (maybeInlineInt32x4(v, dest)) + return; + + SimdData *val = getSimdData(v); + if (!val) + return; + + JS_ASSERT(!val->uses.bound()); + JS_ASSERT(val->type() == SimdConstant::Int32x4); + + JmpSrc j = masm.movdqa_ripr(dest.code()); + JmpSrc prev = JmpSrc(val->uses.use(j.offset())); + masm.setNextJump(j, prev); +} + +void +MacroAssemblerX64::loadConstantFloat32x4(const SimdConstant&v, FloatRegister dest) +{ + JS_ASSERT(v.type() == SimdConstant::Float32x4); + if (maybeInlineFloat32x4(v, dest)) + return; + + SimdData *val = getSimdData(v); + if (!val) + return; + + JS_ASSERT(!val->uses.bound()); + JS_ASSERT(val->type() == SimdConstant::Float32x4); + + JmpSrc j = masm.movaps_ripr(dest.code()); + JmpSrc prev = JmpSrc(val->uses.use(j.offset())); + masm.setNextJump(j, prev); +} + void MacroAssemblerX64::finish() { @@ -98,6 +158,19 @@ MacroAssemblerX64::finish() masm.floatConstant(flt.value); } + // SIMD memory values must be suitably aligned. + if (!simds_.empty()) + masm.align(SimdStackAlignment); + for (size_t i = 0; i < simds_.length(); i++) { + SimdData &v = simds_[i]; + bind(&v.uses); + switch(v.type()) { + case SimdConstant::Int32x4: masm.int32x4Constant(v.value.asInt32x4()); break; + case SimdConstant::Float32x4: masm.float32x4Constant(v.value.asFloat32x4()); break; + default: MOZ_ASSUME_UNREACHABLE("unexpected SimdConstant type"); + } + } + MacroAssemblerX86Shared::finish(); } diff --git a/js/src/jit/x64/MacroAssembler-x64.h b/js/src/jit/x64/MacroAssembler-x64.h index bb9b73c2c9e2..79bb3007d3b9 100644 --- a/js/src/jit/x64/MacroAssembler-x64.h +++ b/js/src/jit/x64/MacroAssembler-x64.h @@ -66,6 +66,17 @@ class MacroAssemblerX64 : public MacroAssemblerX86Shared typedef HashMap, SystemAllocPolicy> FloatMap; FloatMap floatMap_; + struct SimdData { + SimdConstant value; + NonAssertingLabel uses; + + SimdData(const SimdConstant &v) : value(v) {} + SimdConstant::Type type() { return value.type(); } + }; + Vector simds_; + typedef HashMap SimdMap; + SimdMap simdMap_; + void setupABICall(uint32_t arg); protected: @@ -1199,6 +1210,11 @@ class MacroAssemblerX64 : public MacroAssemblerX86Shared void loadConstantDouble(double d, FloatRegister dest); void loadConstantFloat32(float f, FloatRegister dest); + private: + SimdData *getSimdData(const SimdConstant &v); + public: + void loadConstantInt32x4(const SimdConstant &v, FloatRegister dest); + void loadConstantFloat32x4(const SimdConstant &v, FloatRegister dest); void branchTruncateDouble(FloatRegister src, Register dest, Label *fail) { cvttsd2sq(src, dest); diff --git a/js/src/jit/x86/CodeGenerator-x86.cpp b/js/src/jit/x86/CodeGenerator-x86.cpp index 35339a75a0d7..adfd2ae2370b 100644 --- a/js/src/jit/x86/CodeGenerator-x86.cpp +++ b/js/src/jit/x86/CodeGenerator-x86.cpp @@ -310,7 +310,7 @@ CodeGeneratorX86::visitLoadTypedArrayElementStatic(LLoadTypedArrayElementStatic bool isFloat32Load = (vt == Scalar::Float32); if (!mir->fallible()) { ool = new(alloc()) OutOfLineLoadTypedArrayOutOfBounds(ToAnyRegister(out), isFloat32Load); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; } @@ -356,7 +356,7 @@ CodeGeneratorX86::visitAsmJSLoadHeap(LAsmJSLoadHeap *ins) bool isFloat32Load = vt == Scalar::Float32; OutOfLineLoadTypedArrayOutOfBounds *ool = new(alloc()) OutOfLineLoadTypedArrayOutOfBounds(ToAnyRegister(out), isFloat32Load); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, mir)) return false; CodeOffsetLabel cmp = masm.cmplWithPatch(ptrReg, Imm32(0)); @@ -641,7 +641,7 @@ CodeGeneratorX86::visitTruncateDToInt32(LTruncateDToInt32 *ins) Register output = ToRegister(ins->output()); OutOfLineTruncate *ool = new(alloc()) OutOfLineTruncate(ins); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; masm.branchTruncateDouble(input, output, ool->entry()); @@ -656,7 +656,7 @@ CodeGeneratorX86::visitTruncateFToInt32(LTruncateFToInt32 *ins) Register output = ToRegister(ins->output()); OutOfLineTruncateFloat32 *ool = new(alloc()) OutOfLineTruncateFloat32(ins); - if (!addOutOfLineCode(ool)) + if (!addOutOfLineCode(ool, ins->mir())) return false; masm.branchTruncateFloat32(input, output, ool->entry()); diff --git a/js/src/jit/x86/MacroAssembler-x86.cpp b/js/src/jit/x86/MacroAssembler-x86.cpp index 0c46c7e2f3a6..05fb7a890d32 100644 --- a/js/src/jit/x86/MacroAssembler-x86.cpp +++ b/js/src/jit/x86/MacroAssembler-x86.cpp @@ -110,6 +110,58 @@ MacroAssemblerX86::addConstantFloat32(float f, FloatRegister dest) flt->uses.setPrev(masm.size()); } +MacroAssemblerX86::SimdData * +MacroAssemblerX86::getSimdData(const SimdConstant &v) +{ + if (!simdMap_.initialized()) { + enoughMemory_ &= simdMap_.init(); + if (!enoughMemory_) + return nullptr; + } + size_t index; + SimdMap::AddPtr p = simdMap_.lookupForAdd(v); + if (p) { + index = p->value(); + } else { + index = simds_.length(); + enoughMemory_ &= simds_.append(SimdData(v)); + enoughMemory_ &= simdMap_.add(p, v, index); + if (!enoughMemory_) + return nullptr; + } + SimdData &simd = simds_[index]; + JS_ASSERT(!simd.uses.bound()); + return &simd; +} + +void +MacroAssemblerX86::loadConstantInt32x4(const SimdConstant &v, FloatRegister dest) +{ + JS_ASSERT(v.type() == SimdConstant::Int32x4); + if (maybeInlineInt32x4(v, dest)) + return; + SimdData *i4 = getSimdData(v); + if (!i4) + return; + JS_ASSERT(i4->type() == SimdConstant::Int32x4); + masm.movdqa_mr(reinterpret_cast(i4->uses.prev()), dest.code()); + i4->uses.setPrev(masm.size()); +} + +void +MacroAssemblerX86::loadConstantFloat32x4(const SimdConstant &v, FloatRegister dest) +{ + JS_ASSERT(v.type() == SimdConstant::Float32x4); + if (maybeInlineFloat32x4(v, dest)) + return; + SimdData *f4 = getSimdData(v); + if (!f4) + return; + JS_ASSERT(f4->type() == SimdConstant::Float32x4); + masm.movaps_mr(reinterpret_cast(f4->uses.prev()), dest.code()); + f4->uses.setPrev(masm.size()); +} + void MacroAssemblerX86::finish() { @@ -132,6 +184,22 @@ MacroAssemblerX86::finish() if (!enoughMemory_) return; } + + // SIMD memory values must be suitably aligned. + if (!simds_.empty()) + masm.align(SimdStackAlignment); + for (size_t i = 0; i < simds_.length(); i++) { + CodeLabel cl(simds_[i].uses); + SimdData &v = simds_[i]; + switch (v.type()) { + case SimdConstant::Int32x4: writeInt32x4Constant(v.value, cl.src()); break; + case SimdConstant::Float32x4: writeFloat32x4Constant(v.value, cl.src()); break; + default: MOZ_ASSUME_UNREACHABLE("unexpected SimdConstant type"); + } + enoughMemory_ &= addCodeLabel(cl); + if (!enoughMemory_) + return; + } } void diff --git a/js/src/jit/x86/MacroAssembler-x86.h b/js/src/jit/x86/MacroAssembler-x86.h index a3bd5181c6da..aaa7cdd3efc5 100644 --- a/js/src/jit/x86/MacroAssembler-x86.h +++ b/js/src/jit/x86/MacroAssembler-x86.h @@ -38,14 +38,24 @@ class MacroAssemblerX86 : public MacroAssemblerX86Shared Float(float value) : value(value) {} }; Vector floats_; + struct SimdData { + SimdConstant value; + AbsoluteLabel uses; + SimdData(const SimdConstant &v) : value(v) {} + SimdConstant::Type type() { return value.type(); } + }; + Vector simds_; typedef HashMap, SystemAllocPolicy> DoubleMap; DoubleMap doubleMap_; typedef HashMap, SystemAllocPolicy> FloatMap; FloatMap floatMap_; + typedef HashMap SimdMap; + SimdMap simdMap_; Double *getDouble(double d); Float *getFloat(float f); + SimdData *getSimdData(const SimdConstant &v); protected: MoveResolver moveResolver_; @@ -929,6 +939,8 @@ class MacroAssemblerX86 : public MacroAssemblerX86Shared void addConstantDouble(double d, FloatRegister dest); void loadConstantFloat32(float f, FloatRegister dest); void addConstantFloat32(float f, FloatRegister dest); + void loadConstantInt32x4(const SimdConstant &v, FloatRegister dest); + void loadConstantFloat32x4(const SimdConstant &v, FloatRegister dest); void branchTruncateDouble(FloatRegister src, Register dest, Label *fail) { cvttsd2si(src, dest); diff --git a/js/src/jsapi.cpp b/js/src/jsapi.cpp index 4361ee3d86da..c2d15765045b 100644 --- a/js/src/jsapi.cpp +++ b/js/src/jsapi.cpp @@ -1879,7 +1879,9 @@ JS_GC(JSRuntime *rt) JS_PUBLIC_API(void) JS_MaybeGC(JSContext *cx) { - MaybeGC(cx); + GCRuntime &gc = cx->runtime()->gc; + if (!gc.maybeGC(cx->zone())) + gc.maybePeriodicFullGC(); } JS_PUBLIC_API(void) diff --git a/js/src/jsarray.cpp b/js/src/jsarray.cpp index c01b8b4feb96..e7438265fa12 100644 --- a/js/src/jsarray.cpp +++ b/js/src/jsarray.cpp @@ -1039,13 +1039,74 @@ ArrayJoinKernel(JSContext *cx, SeparatorOp sepOp, HandleObject obj, uint32_t len } template -static bool -ArrayJoin(JSContext *cx, CallArgs &args) +JSString * +js::ArrayJoin(JSContext *cx, HandleObject obj, HandleLinearString sepstr, uint32_t length) { // This method is shared by Array.prototype.join and // Array.prototype.toLocaleString. The steps in ES5 are nearly the same, so // the annotations in this function apply to both toLocaleString and join. + // Steps 1 to 6, should be done by the caller. + + JS::Anchor anchor(sepstr); + + // Step 6 is implicit in the loops below. + + // An optimized version of a special case of steps 7-11: when length==1 and + // the 0th element is a string, ToString() of that element is a no-op and + // so it can be immediately returned as the result. + if (length == 1 && !Locale && obj->is() && + obj->getDenseInitializedLength() == 1) + { + const Value &elem0 = obj->getDenseElement(0); + if (elem0.isString()) { + return elem0.toString(); + } + } + + StringBuffer sb(cx); + if (sepstr->hasTwoByteChars() && !sb.ensureTwoByteChars()) + return nullptr; + + // The separator will be added |length - 1| times, reserve space for that + // so that we don't have to unnecessarily grow the buffer. + size_t seplen = sepstr->length(); + if (length > 0 && !sb.reserve(seplen * (length - 1))) + return nullptr; + + // Various optimized versions of steps 7-10. + if (seplen == 0) { + EmptySeparatorOp op; + if (!ArrayJoinKernel(cx, op, obj, length, sb)) + return nullptr; + } else if (seplen == 1) { + jschar c = sepstr->latin1OrTwoByteChar(0); + if (c <= JSString::MAX_LATIN1_CHAR) { + CharSeparatorOp op(c); + if (!ArrayJoinKernel(cx, op, obj, length, sb)) + return nullptr; + } else { + CharSeparatorOp op(c); + if (!ArrayJoinKernel(cx, op, obj, length, sb)) + return nullptr; + } + } else { + StringSeparatorOp op(sepstr); + if (!ArrayJoinKernel(cx, op, obj, length, sb)) + return nullptr; + } + + // Step 11 + JSString *str = sb.finishString(); + if (!str) + return nullptr; + return str; +} + +template +bool +ArrayJoin(JSContext *cx, CallArgs &args) +{ // Step 1 RootedObject obj(cx, ToObject(cx, args.thisv())); if (!obj) @@ -1078,60 +1139,12 @@ ArrayJoin(JSContext *cx, CallArgs &args) sepstr = cx->names().comma; } - JS::Anchor anchor(sepstr); - - // Step 6 is implicit in the loops below. - - // An optimized version of a special case of steps 7-11: when length==1 and - // the 0th element is a string, ToString() of that element is a no-op and - // so it can be immediately returned as the result. - if (length == 1 && !Locale && obj->is() && - obj->getDenseInitializedLength() == 1) - { - const Value &elem0 = obj->getDenseElement(0); - if (elem0.isString()) { - args.rval().setString(elem0.toString()); - return true; - } - } - - StringBuffer sb(cx); - if (sepstr->hasTwoByteChars() && !sb.ensureTwoByteChars()) + // Step 6 to 11 + JSString *res = js::ArrayJoin(cx, obj, sepstr, length); + if (!res) return false; - // The separator will be added |length - 1| times, reserve space for that - // so that we don't have to unnecessarily grow the buffer. - size_t seplen = sepstr->length(); - if (length > 0 && !sb.reserve(seplen * (length - 1))) - return false; - - // Various optimized versions of steps 7-10. - if (seplen == 0) { - EmptySeparatorOp op; - if (!ArrayJoinKernel(cx, op, obj, length, sb)) - return false; - } else if (seplen == 1) { - jschar c = sepstr->latin1OrTwoByteChar(0); - if (c <= JSString::MAX_LATIN1_CHAR) { - CharSeparatorOp op(c); - if (!ArrayJoinKernel(cx, op, obj, length, sb)) - return false; - } else { - CharSeparatorOp op(c); - if (!ArrayJoinKernel(cx, op, obj, length, sb)) - return false; - } - } else { - StringSeparatorOp op(sepstr); - if (!ArrayJoinKernel(cx, op, obj, length, sb)) - return false; - } - - // Step 11 - JSString *str = sb.finishString(); - if (!str) - return false; - args.rval().setString(str); + args.rval().setString(res); return true; } @@ -1183,8 +1196,8 @@ array_toLocaleString(JSContext *cx, unsigned argc, Value *vp) } /* ES5 15.4.4.5 */ -static bool -array_join(JSContext *cx, unsigned argc, Value *vp) +bool +js::array_join(JSContext *cx, unsigned argc, Value *vp) { JS_CHECK_RECURSION(cx, return false); @@ -2955,7 +2968,7 @@ static const JSFunctionSpec array_methods[] = { JS_FN(js_toLocaleString_str,array_toLocaleString,0,0), /* Perl-ish methods. */ - JS_FN("join", array_join, 1,JSFUN_GENERIC_NATIVE), + JS_FN("join", js::array_join, 1,JSFUN_GENERIC_NATIVE), JS_FN("reverse", array_reverse, 0,JSFUN_GENERIC_NATIVE), JS_FN("sort", array_sort, 1,JSFUN_GENERIC_NATIVE), JS_FN("push", array_push, 1,JSFUN_GENERIC_NATIVE), diff --git a/js/src/jsarray.h b/js/src/jsarray.h index 7ed72df7bb69..0f6e5b38bc70 100644 --- a/js/src/jsarray.h +++ b/js/src/jsarray.h @@ -140,10 +140,20 @@ array_splice_impl(JSContext *cx, unsigned argc, js::Value *vp, bool pop); extern bool array_concat(JSContext *cx, unsigned argc, js::Value *vp); +template +JSString * +ArrayJoin(JSContext *cx, HandleObject obj, HandleLinearString sepstr, uint32_t length); + extern bool array_concat_dense(JSContext *cx, Handle arr1, Handle arr2, Handle result); +bool +array_join(JSContext *cx, unsigned argc, js::Value *vp); + +extern JSString * +array_join_impl(JSContext *cx, HandleValue array, HandleString sep); + extern void ArrayShiftMoveElements(JSObject *obj); diff --git a/js/src/jsgc.cpp b/js/src/jsgc.cpp index f85a1ee7d9c2..de8708c8e7b9 100644 --- a/js/src/jsgc.cpp +++ b/js/src/jsgc.cpp @@ -1112,12 +1112,13 @@ GCRuntime::GCRuntime(JSRuntime *rt) : chunkAllocationSinceLastGC(false), nextFullGCTime(0), lastGCTime(0), - jitReleaseTime(0), mode(JSGC_MODE_INCREMENTAL), decommitThreshold(32 * 1024 * 1024), cleanUpEverything(false), grayBitsValid(false), isNeeded(0), + majorGCNumber(0), + jitReleaseNumber(0), number(0), startNumber(0), isFull(false), @@ -1248,8 +1249,11 @@ GCRuntime::initZeal() #endif -/* Lifetime for type sets attached to scripts containing observed types. */ -static const int64_t JIT_SCRIPT_RELEASE_TYPES_INTERVAL = 60 * 1000 * 1000; +/* + * Lifetime in number of major GCs for type sets attached to scripts containing + * observed types. + */ +static const uint64_t JIT_SCRIPT_RELEASE_TYPES_PERIOD = 20; bool GCRuntime::init(uint32_t maxbytes, uint32_t maxNurseryBytes) @@ -1276,9 +1280,7 @@ GCRuntime::init(uint32_t maxbytes, uint32_t maxNurseryBytes) tunables.setParameter(JSGC_MAX_BYTES, maxbytes); setMaxMallocBytes(maxbytes); -#ifndef JS_MORE_DETERMINISTIC - jitReleaseTime = PRMJ_Now() + JIT_SCRIPT_RELEASE_TYPES_INTERVAL; -#endif + jitReleaseNumber = majorGCNumber + JIT_SCRIPT_RELEASE_TYPES_PERIOD; #ifdef JSGC_GENERATIONAL if (!nursery.init(maxNurseryBytes)) @@ -2415,13 +2417,7 @@ GCRuntime::triggerZoneGC(Zone *zone, JS::gcreason::Reason reason) return true; } -void -js::MaybeGC(JSContext *cx) -{ - cx->runtime()->gc.maybeGC(cx->zone()); -} - -void +bool GCRuntime::maybeGC(Zone *zone) { JS_ASSERT(CurrentThreadCanAccessRuntime(rt)); @@ -2430,13 +2426,13 @@ GCRuntime::maybeGC(Zone *zone) if (zealMode == ZealAllocValue || zealMode == ZealPokeValue) { JS::PrepareForFullGC(rt); GC(rt, GC_NORMAL, JS::gcreason::MAYBEGC); - return; + return true; } #endif if (isNeeded) { GCSlice(rt, GC_NORMAL, JS::gcreason::MAYBEGC); - return; + return true; } double factor = schedulingState.inHighFrequencyGCMode() ? 0.85 : 0.9; @@ -2447,15 +2443,25 @@ GCRuntime::maybeGC(Zone *zone) { PrepareZoneForGC(zone); GCSlice(rt, GC_NORMAL, JS::gcreason::MAYBEGC); - return; + return true; } -#ifndef JS_MORE_DETERMINISTIC + return false; +} + +void +GCRuntime::maybePeriodicFullGC() +{ /* + * Trigger a periodic full GC. + * + * This is a source of non-determinism, but is not called from the shell. + * * Access to the counters and, on 32 bit, setting gcNextFullGCTime below * is not atomic and a race condition could trigger or suppress the GC. We * tolerate this. */ +#ifndef JS_MORE_DETERMINISTIC int64_t now = PRMJ_Now(); if (nextFullGCTime && nextFullGCTime <= now) { if (chunkAllocationSinceLastGC || @@ -2881,7 +2887,7 @@ GCHelperState::onBackgroundThread() } bool -GCRuntime::releaseObservedTypes() +GCRuntime::shouldReleaseObservedTypes() { bool releaseTypes = false; @@ -2890,13 +2896,12 @@ GCRuntime::releaseObservedTypes() releaseTypes = true; #endif -#ifndef JS_MORE_DETERMINISTIC - int64_t now = PRMJ_Now(); - if (now >= jitReleaseTime) + /* We may miss the exact target GC due to resets. */ + if (majorGCNumber >= jitReleaseNumber) releaseTypes = true; + if (releaseTypes) - jitReleaseTime = now + JIT_SCRIPT_RELEASE_TYPES_INTERVAL; -#endif + jitReleaseNumber = majorGCNumber + JIT_SCRIPT_RELEASE_TYPES_PERIOD; return releaseTypes; } @@ -4157,10 +4162,9 @@ GCRuntime::beginSweepingZoneGroup() zone->discardJitCode(&fop); } - bool releaseTypes = releaseObservedTypes(); for (GCCompartmentGroupIter c(rt); !c.done(); c.next()) { gcstats::AutoSCC scc(stats, zoneGroupIndex); - c->sweep(&fop, releaseTypes && !c->zone()->isPreservingCode()); + c->sweep(&fop, releaseObservedTypes && !c->zone()->isPreservingCode()); } for (GCZoneGroupIter zone(rt); !zone.done(); zone.next()) { @@ -4174,7 +4178,7 @@ GCRuntime::beginSweepingZoneGroup() // code and new script information in the zone, the only things // whose correctness depends on the type constraints. bool oom = false; - zone->sweep(&fop, releaseTypes && !zone->isPreservingCode(), &oom); + zone->sweep(&fop, releaseObservedTypes && !zone->isPreservingCode(), &oom); if (oom) { zone->setPreservingCode(false); @@ -4264,6 +4268,8 @@ GCRuntime::beginSweepPhase(bool lastGC) sweepOnBackgroundThread = !lastGC && !TraceEnabled() && CanUseExtraThreads(); + releaseObservedTypes = shouldReleaseObservedTypes(); + #ifdef DEBUG for (CompartmentsIter c(rt, SkipAtoms); !c.done(); c.next()) { JS_ASSERT(!c->gcIncomingGrayPointers); @@ -4986,6 +4992,8 @@ GCRuntime::gcCycle(bool incremental, int64_t budget, JSGCInvocationKind gckind, interFrameGC = true; number++; + if (incrementalState == NO_INCREMENTAL) + majorGCNumber++; // It's ok if threads other than the main thread have suppressGC set, as // they are operating on zones which will not be collected from here. diff --git a/js/src/jsgc.h b/js/src/jsgc.h index 3f3f4265b88b..71ae41121c7e 100644 --- a/js/src/jsgc.h +++ b/js/src/jsgc.h @@ -971,9 +971,6 @@ TriggerGC(JSRuntime *rt, JS::gcreason::Reason reason); extern bool TriggerZoneGC(Zone *zone, JS::gcreason::Reason reason); -extern void -MaybeGC(JSContext *cx); - extern void ReleaseAllJITCode(FreeOp *op); diff --git a/js/src/jsgcinlines.h b/js/src/jsgcinlines.h index 2ae37ea6aba7..8e764d08d0d6 100644 --- a/js/src/jsgcinlines.h +++ b/js/src/jsgcinlines.h @@ -691,7 +691,7 @@ AllocateObjectForCacheHit(JSContext *cx, AllocKind kind, InitialHeap heap) JSObject *obj = AllocateObject(cx, kind, 0, heap); if (!obj && allowGC) { - MaybeGC(cx); + cx->runtime()->gc.maybeGC(cx->zone()); return nullptr; } diff --git a/js/src/moz.build b/js/src/moz.build index 779df8c9163b..7a5c0f04ad14 100644 --- a/js/src/moz.build +++ b/js/src/moz.build @@ -167,6 +167,7 @@ UNIFIED_SOURCES += [ 'jit/IonMacroAssembler.cpp', 'jit/IonOptimizationLevels.cpp', 'jit/IonSpewer.cpp', + 'jit/JitcodeMap.cpp', 'jit/JitOptions.cpp', 'jit/JSONSpewer.cpp', 'jit/LICM.cpp', @@ -378,10 +379,6 @@ if CONFIG['ENABLE_ION']: SOURCES += [ 'assembler/assembler/MacroAssemblerX86Common.cpp', ] - elif CONFIG['JS_CODEGEN_ARM']: - SOURCES += [ - 'assembler/assembler/MacroAssemblerARM.cpp', - ] if CONFIG['JS_HAS_CTYPES']: SOURCES += [ diff --git a/js/src/vm/Debugger.cpp b/js/src/vm/Debugger.cpp index 065d8517acce..b9503edcb426 100644 --- a/js/src/vm/Debugger.cpp +++ b/js/src/vm/Debugger.cpp @@ -22,6 +22,7 @@ #include "js/Vector.h" #include "vm/ArgumentsObject.h" #include "vm/DebuggerMemory.h" +#include "vm/SPSProfiler.h" #include "vm/WrapperObject.h" #include "jsgcinlines.h" @@ -2300,6 +2301,7 @@ Debugger::construct(JSContext *cx, unsigned argc, Value *vp) bool Debugger::addDebuggeeGlobal(JSContext *cx, Handle global) { + AutoSuppressProfilerSampling suppressProfilerSampling(cx); AutoDebugModeInvalidation invalidate(global->compartment()); return addDebuggeeGlobal(cx, global, invalidate); } diff --git a/js/src/vm/Runtime.cpp b/js/src/vm/Runtime.cpp index 4a75154790cf..9c1729d7c655 100644 --- a/js/src/vm/Runtime.cpp +++ b/js/src/vm/Runtime.cpp @@ -178,6 +178,7 @@ JSRuntime::JSRuntime(JSRuntime *parentRuntime) debugMode(false), spsProfiler(thisFromCtor()), profilingScripts(false), + suppressProfilerSampling(false), hadOutOfMemory(false), haveCreatedContext(false), data(nullptr), @@ -235,12 +236,12 @@ JSRuntime::JSRuntime(JSRuntime *parentRuntime) static bool JitSupportsFloatingPoint() { - if (!JSC::MacroAssembler::supportsFloatingPoint()) - return false; - -#if WTF_ARM_ARCH_VERSION == 6 +#if defined(JS_CODEGEN_ARM) if (!js::jit::HasVFP()) return false; +#else + if (!JSC::MacroAssembler::supportsFloatingPoint()) + return false; #endif return true; diff --git a/js/src/vm/Runtime.h b/js/src/vm/Runtime.h index 11064cca35c5..9cb754774fac 100644 --- a/js/src/vm/Runtime.h +++ b/js/src/vm/Runtime.h @@ -1004,6 +1004,21 @@ struct JSRuntime : public JS::shadow::Runtime, /* If true, new scripts must be created with PC counter information. */ bool profilingScripts; + /* Whether sampling should be enabled or not. */ + private: + bool suppressProfilerSampling; + + public: + bool isProfilerSamplingEnabled() const { + return !suppressProfilerSampling; + } + void disableProfilerSampling() { + suppressProfilerSampling = true; + } + void enableProfilerSampling() { + suppressProfilerSampling = false; + } + /* Had an out-of-memory error which did not populate an exception. */ bool hadOutOfMemory; diff --git a/js/src/vm/SPSProfiler.cpp b/js/src/vm/SPSProfiler.cpp index 63482457b2d0..fa2d69b09e5d 100644 --- a/js/src/vm/SPSProfiler.cpp +++ b/js/src/vm/SPSProfiler.cpp @@ -374,3 +374,31 @@ js::ProfilingGetPC(JSRuntime *rt, JSScript *script, void *ip) { return rt->spsProfiler.ipToPC(script, size_t(ip)); } + + + +AutoSuppressProfilerSampling::AutoSuppressProfilerSampling(JSContext *cx + MOZ_GUARD_OBJECT_NOTIFIER_PARAM_IN_IMPL) + : rt_(cx->runtime()), + previouslyEnabled_(rt_->isProfilerSamplingEnabled()) +{ + MOZ_GUARD_OBJECT_NOTIFIER_INIT; + if (previouslyEnabled_) + rt_->disableProfilerSampling(); +} + +AutoSuppressProfilerSampling::AutoSuppressProfilerSampling(JSRuntime *rt + MOZ_GUARD_OBJECT_NOTIFIER_PARAM_IN_IMPL) + : rt_(rt), + previouslyEnabled_(rt_->isProfilerSamplingEnabled()) +{ + MOZ_GUARD_OBJECT_NOTIFIER_INIT; + if (previouslyEnabled_) + rt_->disableProfilerSampling(); +} + +AutoSuppressProfilerSampling::~AutoSuppressProfilerSampling() +{ + if (previouslyEnabled_) + rt_->enableProfilerSampling(); +} diff --git a/js/src/vm/SPSProfiler.h b/js/src/vm/SPSProfiler.h index abb8e1f09b7c..68955174b008 100644 --- a/js/src/vm/SPSProfiler.h +++ b/js/src/vm/SPSProfiler.h @@ -220,6 +220,24 @@ class AutoSPSLock PRLock *lock_; }; +/* + * This class is used to suppress profiler sampling during + * critical sections where stack state is not valid. + */ +class AutoSuppressProfilerSampling +{ + public: + explicit AutoSuppressProfilerSampling(JSContext *cx MOZ_GUARD_OBJECT_NOTIFIER_PARAM); + explicit AutoSuppressProfilerSampling(JSRuntime *rt MOZ_GUARD_OBJECT_NOTIFIER_PARAM); + + ~AutoSuppressProfilerSampling(); + + private: + JSRuntime *rt_; + bool previouslyEnabled_; + MOZ_DECL_USE_GUARD_OBJECT_NOTIFIER +}; + inline size_t SPSProfiler::stringsCount() { diff --git a/layout/svg/nsSVGClipPathFrame.cpp b/layout/svg/nsSVGClipPathFrame.cpp index 0e0f02b8f992..76df0c59af99 100644 --- a/layout/svg/nsSVGClipPathFrame.cpp +++ b/layout/svg/nsSVGClipPathFrame.cpp @@ -28,9 +28,9 @@ NS_NewSVGClipPathFrame(nsIPresShell* aPresShell, nsStyleContext* aContext) NS_IMPL_FRAMEARENA_HELPERS(nsSVGClipPathFrame) nsresult -nsSVGClipPathFrame::ClipPaint(nsRenderingContext* aContext, - nsIFrame* aParent, - const gfxMatrix &aMatrix) +nsSVGClipPathFrame::ApplyClipOrPaintClipMask(nsRenderingContext* aContext, + nsIFrame* aClippedFrame, + const gfxMatrix& aMatrix) { // If the flag is set when we get here, it means this clipPath frame // has already been used painting the current clip, and the document @@ -41,12 +41,7 @@ nsSVGClipPathFrame::ClipPaint(nsRenderingContext* aContext, } AutoClipPathReferencer clipRef(this); - mClipParent = aParent; - if (mClipParentMatrix) { - *mClipParentMatrix = aMatrix; - } else { - mClipParentMatrix = new gfxMatrix(aMatrix); - } + mMatrixForChildren = GetClipPathTransform(aClippedFrame) * aMatrix; gfxContext *gfx = aContext->ThebesContext(); @@ -83,7 +78,7 @@ nsSVGClipPathFrame::ClipPaint(nsRenderingContext* aContext, referencedClipIsTrivial = clipPathFrame->IsTrivial(); gfx->Save(); if (referencedClipIsTrivial) { - clipPathFrame->ClipPaint(aContext, aParent, aMatrix); + clipPathFrame->ApplyClipOrPaintClipMask(aContext, aClippedFrame, aMatrix); } else { gfx->PushGroup(gfxContentType::ALPHA); } @@ -109,7 +104,7 @@ nsSVGClipPathFrame::ClipPaint(nsRenderingContext* aContext, isTrivial = clipPathFrame->IsTrivial(); gfx->Save(); if (isTrivial) { - clipPathFrame->ClipPaint(aContext, aParent, aMatrix); + clipPathFrame->ApplyClipOrPaintClipMask(aContext, aClippedFrame, aMatrix); } else { gfx->PushGroup(gfxContentType::ALPHA); } @@ -124,7 +119,7 @@ nsSVGClipPathFrame::ClipPaint(nsRenderingContext* aContext, nsRefPtr clipMaskSurface; gfx->PushGroup(gfxContentType::ALPHA); - clipPathFrame->ClipPaint(aContext, aParent, aMatrix); + clipPathFrame->ApplyClipOrPaintClipMask(aContext, aClippedFrame, aMatrix); clipMaskSurface = gfx->PopGroup(); if (clipMaskSurface) { @@ -143,7 +138,7 @@ nsSVGClipPathFrame::ClipPaint(nsRenderingContext* aContext, nsRefPtr clipMaskSurface; gfx->PushGroup(gfxContentType::ALPHA); - clipPathFrame->ClipPaint(aContext, aParent, aMatrix); + clipPathFrame->ApplyClipOrPaintClipMask(aContext, aClippedFrame, aMatrix); clipMaskSurface = gfx->PopGroup(); if (clipMaskSurface) { @@ -325,15 +320,7 @@ nsSVGClipPathFrame::GetType() const gfxMatrix nsSVGClipPathFrame::GetCanvasTM(uint32_t aFor, nsIFrame* aTransformRoot) { - SVGClipPathElement *content = static_cast(mContent); - - gfxMatrix tm = - content->PrependLocalTransformsTo(mClipParentMatrix ? - *mClipParentMatrix : gfxMatrix()); - - return nsSVGUtils::AdjustMatrixForUnits(tm, - &content->mEnumAttributes[SVGClipPathElement::CLIPPATHUNITS], - mClipParent); + return mMatrixForChildren; } gfxMatrix diff --git a/layout/svg/nsSVGClipPathFrame.h b/layout/svg/nsSVGClipPathFrame.h index 602bcbc48c18..20f717a6fedb 100644 --- a/layout/svg/nsSVGClipPathFrame.h +++ b/layout/svg/nsSVGClipPathFrame.h @@ -37,9 +37,21 @@ public: const nsDisplayListSet& aLists) MOZ_OVERRIDE {} // nsSVGClipPathFrame methods: - nsresult ClipPaint(nsRenderingContext* aContext, - nsIFrame* aParent, - const gfxMatrix &aMatrix); + + /** + * If the SVG clipPath is simple (as determined by the IsTrivial() method), + * calling this method simply pushes a clip path onto the DrawTarget. If the + * SVG clipPath is not simple then calling this method will paint the + * clipPath's contents (geometry being filled only, with opaque black) to the + * DrawTarget. In this latter case callers are expected to first push a + * group before calling this method, then pop the group after calling and use + * it as a mask to mask the clipped frame. + * + * XXXjwatt Maybe split this into two methods. + */ + nsresult ApplyClipOrPaintClipMask(nsRenderingContext* aContext, + nsIFrame* aClippedFrame, + const gfxMatrix &aMatrix); /** * aPoint is expected to be in aClippedFrame's SVG user space. @@ -110,8 +122,7 @@ public: MOZ_DECL_USE_GUARD_OBJECT_NOTIFIER }; - nsIFrame *mClipParent; - nsAutoPtr mClipParentMatrix; + gfxMatrix mMatrixForChildren; // recursion prevention flag bool mInUse; diff --git a/layout/svg/nsSVGFilterFrame.cpp b/layout/svg/nsSVGFilterFrame.cpp index 13aa7de85229..d0c8bdce2111 100644 --- a/layout/svg/nsSVGFilterFrame.cpp +++ b/layout/svg/nsSVGFilterFrame.cpp @@ -14,7 +14,6 @@ #include "nsSVGElement.h" #include "mozilla/dom/SVGFilterElement.h" #include "nsSVGFilterInstance.h" -#include "nsSVGFilterPaintCallback.h" #include "nsSVGIntegrationUtils.h" #include "nsSVGUtils.h" #include "nsContentUtils.h" diff --git a/layout/svg/nsSVGFilterFrame.h b/layout/svg/nsSVGFilterFrame.h index f1c252d73c5d..e60845614a9d 100644 --- a/layout/svg/nsSVGFilterFrame.h +++ b/layout/svg/nsSVGFilterFrame.h @@ -18,7 +18,6 @@ class nsIFrame; class nsIPresShell; class nsRenderingContext; class nsStyleContext; -class nsSVGFilterPaintCallback; class nsSVGIntegerPair; class nsSVGLength2; diff --git a/layout/svg/nsSVGFilterInstance.cpp b/layout/svg/nsSVGFilterInstance.cpp index c21dea6a0089..eebddd79c203 100644 --- a/layout/svg/nsSVGFilterInstance.cpp +++ b/layout/svg/nsSVGFilterInstance.cpp @@ -14,7 +14,6 @@ #include "mozilla/dom/SVGFilterElement.h" #include "nsReferencedElement.h" #include "nsSVGFilterFrame.h" -#include "nsSVGFilterPaintCallback.h" #include "nsSVGUtils.h" #include "SVGContentUtils.h" #include "FilterSupport.h" diff --git a/layout/svg/nsSVGFilterInstance.h b/layout/svg/nsSVGFilterInstance.h index 5912ea6dc0a0..24ce14544b7e 100644 --- a/layout/svg/nsSVGFilterInstance.h +++ b/layout/svg/nsSVGFilterInstance.h @@ -16,7 +16,6 @@ class nsIFrame; class nsSVGFilterFrame; -class nsSVGFilterPaintCallback; namespace mozilla { namespace dom { diff --git a/layout/svg/nsSVGForeignObjectFrame.cpp b/layout/svg/nsSVGForeignObjectFrame.cpp index ccbc88fe4438..2d55d5fe295d 100644 --- a/layout/svg/nsSVGForeignObjectFrame.cpp +++ b/layout/svg/nsSVGForeignObjectFrame.cpp @@ -570,11 +570,14 @@ nsSVGForeignObjectFrame::DoReflow() nsRect nsSVGForeignObjectFrame::GetInvalidRegion() { + MOZ_ASSERT(!NS_SVGDisplayListPaintingEnabled(), + "Only called by nsDisplayOuterSVG code"); + nsIFrame* kid = GetFirstPrincipalChild(); if (kid->HasInvalidFrameInSubtree()) { gfxRect r(mRect.x, mRect.y, mRect.width, mRect.height); r.Scale(1.0 / nsPresContext::AppUnitsPerCSSPixel()); - nsRect rect = nsSVGUtils::ToCanvasBounds(r, GetCanvasTM(FOR_PAINTING), PresContext()); + nsRect rect = nsSVGUtils::ToCanvasBounds(r, GetCanvasTM(FOR_OUTERSVG_TM), PresContext()); rect = nsSVGUtils::GetPostFilterVisualOverflowRect(this, rect); return rect; } diff --git a/layout/svg/nsSVGIntegrationUtils.cpp b/layout/svg/nsSVGIntegrationUtils.cpp index e5bd8df6bc3b..ae708b1c41b0 100644 --- a/layout/svg/nsSVGIntegrationUtils.cpp +++ b/layout/svg/nsSVGIntegrationUtils.cpp @@ -515,7 +515,7 @@ nsSVGIntegrationUtils::PaintFramesWithEffects(nsRenderingContext* aCtx, */ if (clipPathFrame && isTrivialClip) { gfx->Save(); - clipPathFrame->ClipPaint(aCtx, aFrame, cssPxToDevPxMatrix); + clipPathFrame->ApplyClipOrPaintClipMask(aCtx, aFrame, cssPxToDevPxMatrix); } /* Paint the child */ @@ -552,7 +552,7 @@ nsSVGIntegrationUtils::PaintFramesWithEffects(nsRenderingContext* aCtx, if (clipPathFrame && !isTrivialClip) { gfx->PushGroup(gfxContentType::COLOR_ALPHA); - nsresult rv = clipPathFrame->ClipPaint(aCtx, aFrame, cssPxToDevPxMatrix); + nsresult rv = clipPathFrame->ApplyClipOrPaintClipMask(aCtx, aFrame, cssPxToDevPxMatrix); clipMaskSurface = gfx->PopGroup(); if (NS_SUCCEEDED(rv) && clipMaskSurface) { diff --git a/layout/svg/nsSVGUtils.cpp b/layout/svg/nsSVGUtils.cpp index 89232c05a654..4c5f17c92162 100644 --- a/layout/svg/nsSVGUtils.cpp +++ b/layout/svg/nsSVGUtils.cpp @@ -609,7 +609,7 @@ nsSVGUtils::PaintFrameWithEffects(nsRenderingContext *aContext, */ if (clipPathFrame && isTrivialClip) { gfx->Save(); - clipPathFrame->ClipPaint(aContext, aFrame, matrix); + clipPathFrame->ApplyClipOrPaintClipMask(aContext, aFrame, matrix); } /* Paint the child */ @@ -661,7 +661,7 @@ nsSVGUtils::PaintFrameWithEffects(nsRenderingContext *aContext, if (clipPathFrame && !isTrivialClip) { gfx->PushGroup(gfxContentType::COLOR_ALPHA); - nsresult rv = clipPathFrame->ClipPaint(aContext, aFrame, matrix); + nsresult rv = clipPathFrame->ApplyClipOrPaintClipMask(aContext, aFrame, matrix); clipMaskSurface = gfx->PopGroup(); if (NS_SUCCEEDED(rv) && clipMaskSurface) { diff --git a/media/webrtc/signaling/src/peerconnection/PeerConnectionCtx.cpp b/media/webrtc/signaling/src/peerconnection/PeerConnectionCtx.cpp index 84f283367cfb..a6a486973d5a 100644 --- a/media/webrtc/signaling/src/peerconnection/PeerConnectionCtx.cpp +++ b/media/webrtc/signaling/src/peerconnection/PeerConnectionCtx.cpp @@ -401,7 +401,7 @@ nsresult PeerConnectionCtx::Initialize() { codecMask |= VCM_CODEC_RESOURCE_G711; codecMask |= VCM_CODEC_RESOURCE_OPUS; //codecMask |= VCM_CODEC_RESOURCE_LINEAR; - //codecMask |= VCM_CODEC_RESOURCE_G722; + codecMask |= VCM_CODEC_RESOURCE_G722; //codecMask |= VCM_CODEC_RESOURCE_iLBC; //codecMask |= VCM_CODEC_RESOURCE_iSAC; mCCM->setAudioCodecs(codecMask); diff --git a/media/webrtc/signaling/src/sipcc/core/common/prot_configmgr.c b/media/webrtc/signaling/src/sipcc/core/common/prot_configmgr.c index 32ff6005c38e..b8e6eedbbd7c 100755 --- a/media/webrtc/signaling/src/sipcc/core/common/prot_configmgr.c +++ b/media/webrtc/signaling/src/sipcc/core/common/prot_configmgr.c @@ -735,6 +735,11 @@ config_set_current_codec_table (int codec_mask, rtp_ptype *codecs) idx++; } + if (codec_mask & VCM_CODEC_RESOURCE_G722) { + codecs[idx] = RTP_G722; + idx++; + } + if (codec_mask & VCM_CODEC_RESOURCE_G711) { codecs[idx] = RTP_PCMU; idx++; @@ -752,11 +757,6 @@ config_set_current_codec_table (int codec_mask, rtp_ptype *codecs) idx++; } - if (codec_mask & VCM_CODEC_RESOURCE_G722) { - codecs[idx] = RTP_G722; - idx++; - } - if (codec_mask & VCM_CODEC_RESOURCE_iLBC) { codecs[idx] = RTP_ILBC; idx++; diff --git a/media/webrtc/signaling/src/sipcc/core/gsm/gsm_sdp.c b/media/webrtc/signaling/src/sipcc/core/gsm/gsm_sdp.c index 77a01c4b353a..62c2533eaed6 100644 --- a/media/webrtc/signaling/src/sipcc/core/gsm/gsm_sdp.c +++ b/media/webrtc/signaling/src/sipcc/core/gsm/gsm_sdp.c @@ -3485,6 +3485,27 @@ gsmsdp_negotiate_codec (fsmdef_dcb_t *dcb_p, cc_sdp_t *sdp_p, payload_info->audio.bitrate = 32000; break; + case RTP_G722: + /* RFC 3551 + + G722 is specified in ITU-T Recommendation G.722, "7 kHz audio-coding + within 64 kbit/s". The G.722 encoder produces a stream of octets, + each of which SHALL be octet-aligned in an RTP packet. The first bit + transmitted in the G.722 octet, which is the most significant bit of + the higher sub-band sample, SHALL correspond to the most significant + bit of the octet in the RTP packet. + + Even though the actual sampling rate for G.722 audio is 16,000 Hz, + the RTP clock rate for the G722 payload format is 8,000 Hz because + that value was erroneously assigned in RFC 1890 and must remain + unchanged for backward compatibility. The octet rate or sample-pair + rate is 8,000 Hz. + */ + payload_info->audio.frequency = 16000; + payload_info->audio.packet_size = 320; + payload_info->audio.bitrate = 64000; + break; + case RTP_ILBC: payload_info->ilbc.mode = (uint16_t)sdp_attr_get_fmtp_mode_for_payload_type( @@ -3511,6 +3532,7 @@ gsmsdp_negotiate_codec (fsmdef_dcb_t *dcb_p, cc_sdp_t *sdp_p, dcb_p->call_id, fname), codec); payload_info->audio.packet_size = -1; payload_info->audio.bitrate = -1; + MOZ_ASSERT(0); } /* end switch */ diff --git a/media/webrtc/signaling/test/signaling_unittests.cpp b/media/webrtc/signaling/test/signaling_unittests.cpp index aba9e0b5b9ba..4b9c52243eaf 100644 --- a/media/webrtc/signaling/test/signaling_unittests.cpp +++ b/media/webrtc/signaling/test/signaling_unittests.cpp @@ -1445,6 +1445,7 @@ private: case SHOULD_CHECK_AUDIO: ASSERT_NE(sdp.find("a=rtpmap:109 opus/48000"), std::string::npos); if (offer) { + ASSERT_NE(sdp.find("a=rtpmap:9 G722/8000"), std::string::npos); ASSERT_NE(sdp.find("a=rtpmap:0 PCMU/8000"), std::string::npos); } break; @@ -1452,6 +1453,7 @@ private: ASSERT_NE(sdp.find("a=rtpmap:109 opus/48000"), std::string::npos); ASSERT_NE(sdp.find(" 0-15\r\na=sendonly"), std::string::npos); if (offer) { + ASSERT_NE(sdp.find("a=rtpmap:9 G722/8000"), std::string::npos); ASSERT_NE(sdp.find("a=rtpmap:0 PCMU/8000"), std::string::npos); } break; @@ -1459,6 +1461,7 @@ private: ASSERT_NE(sdp.find("a=rtpmap:109 opus/48000"), std::string::npos); ASSERT_NE(sdp.find(" 0-15\r\na=recvonly"), std::string::npos); if (offer) { + ASSERT_NE(sdp.find("a=rtpmap:9 G722/8000"), std::string::npos); ASSERT_NE(sdp.find("a=rtpmap:0 PCMU/8000"), std::string::npos); } break; @@ -1466,6 +1469,7 @@ private: ASSERT_NE(sdp.find("a=rtpmap:109 opus/48000"), std::string::npos); ASSERT_NE(sdp.find(" 0-15\r\na=sendrecv"), std::string::npos); if (offer) { + ASSERT_NE(sdp.find("a=rtpmap:9 G722/8000"), std::string::npos); ASSERT_NE(sdp.find("a=rtpmap:0 PCMU/8000"), std::string::npos); } break; @@ -2053,6 +2057,17 @@ TEST_F(SignalingTest, OfferAnswerNothingDisabled) SHOULD_SENDRECV_AV, SHOULD_SENDRECV_AV); } +TEST_F(SignalingTest, OfferAnswerNothingDisabledFullCycle) +{ + sipcc::OfferOptions options; + OfferAnswer(options, OFFER_AV | ANSWER_AV, true, + SHOULD_SENDRECV_AV, SHOULD_SENDRECV_AV); + // verify the default codec priorities + ASSERT_NE(a1_->getLocalDescription().find("RTP/SAVPF 109 9 0 8 101\r"), std::string::npos); + // verify that opus got selected + ASSERT_NE(a2_->getLocalDescription().find("RTP/SAVPF 109 101\r"), std::string::npos); +} + // XXX reject streams has changed. Re-enable when we can stop() received stream TEST_F(SignalingTest, DISABLED_OfferAnswerDontReceiveAudioOnOffer) { @@ -3201,6 +3216,97 @@ TEST_F(SignalingTest, AudioOnlyCalleeNoRtcpMux) a2_->CheckMediaPipeline(0, 1, 0); } + + +TEST_F(SignalingTest, AudioOnlyG722Only) +{ + EnsureInit(); + + sipcc::OfferOptions options; + + a1_->CreateOffer(options, OFFER_AUDIO, SHOULD_SENDRECV_AUDIO); + a1_->SetLocal(TestObserver::OFFER, a1_->offer(), false); + ParsedSDP sdpWrapper(a1_->offer()); + sdpWrapper.ReplaceLine("m=audio", + "m=audio 65375 RTP/SAVPF 9\r\n"); + std::cout << "Modified SDP " << std::endl + << indent(sdpWrapper.getSdp()) << std::endl; + a2_->SetRemote(TestObserver::OFFER, sdpWrapper.getSdp(), false); + a2_->CreateAnswer(sdpWrapper.getSdp(), + OFFER_AUDIO | ANSWER_AUDIO); + a2_->SetLocal(TestObserver::ANSWER, a2_->answer(), false); + a1_->SetRemote(TestObserver::ANSWER, a2_->answer(), false); + ASSERT_NE(a2_->getLocalDescription().find("RTP/SAVPF 9\r"), std::string::npos); + ASSERT_NE(a2_->getLocalDescription().find("a=rtpmap:9 G722/8000"), std::string::npos); + + ASSERT_TRUE_WAIT(a1_->IceCompleted() == true, kDefaultTimeout); + ASSERT_TRUE_WAIT(a2_->IceCompleted() == true, kDefaultTimeout); + + // Wait for some data to get written + ASSERT_TRUE_WAIT(a1_->GetPacketsSent(0) >= 40 && + a2_->GetPacketsReceived(0) >= 40, kDefaultTimeout * 2); + + a1_->CloseSendStreams(); + a2_->CloseReceiveStreams(); + + ASSERT_GE(a1_->GetPacketsSent(0), 40); + ASSERT_GE(a2_->GetPacketsReceived(0), 40); +} + +TEST_F(SignalingTest, AudioOnlyG722MostPreferred) +{ + EnsureInit(); + + sipcc::OfferOptions options; + + a1_->CreateOffer(options, OFFER_AUDIO, SHOULD_SENDRECV_AUDIO); + a1_->SetLocal(TestObserver::OFFER, a1_->offer(), false); + ParsedSDP sdpWrapper(a1_->offer()); + sdpWrapper.ReplaceLine("m=audio", + "m=audio 65375 RTP/SAVPF 9 0 8 109 101\r\n"); + std::cout << "Modified SDP " << std::endl + << indent(sdpWrapper.getSdp()) << std::endl; + a2_->SetRemote(TestObserver::OFFER, sdpWrapper.getSdp(), false); + a2_->CreateAnswer(sdpWrapper.getSdp(), + OFFER_AUDIO | ANSWER_AUDIO); + a2_->SetLocal(TestObserver::ANSWER, a2_->answer(), false); + a1_->SetRemote(TestObserver::ANSWER, a2_->answer(), false); + ASSERT_NE(a2_->getLocalDescription().find("RTP/SAVPF 9 101\r"), std::string::npos); + ASSERT_NE(a2_->getLocalDescription().find("a=rtpmap:9 G722/8000"), std::string::npos); + + a1_->CloseSendStreams(); + a2_->CloseReceiveStreams(); +} + +TEST_F(SignalingTest, AudioOnlyG722Rejected) +{ + EnsureInit(); + + sipcc::OfferOptions options; + + a1_->CreateOffer(options, OFFER_AUDIO, SHOULD_SENDRECV_AUDIO); + // creating different SDPs as a workaround for rejecting codecs + // this way the answerer should pick a codec with lower priority + a1_->SetLocal(TestObserver::OFFER, a1_->offer(), false); + ParsedSDP sdpWrapper(a1_->offer()); + sdpWrapper.ReplaceLine("m=audio", + "m=audio 65375 RTP/SAVPF 0 8 101\r\n"); + std::cout << "Modified SDP offer " << std::endl + << indent(sdpWrapper.getSdp()) << std::endl; + a2_->SetRemote(TestObserver::OFFER, sdpWrapper.getSdp(), false); + a2_->CreateAnswer(sdpWrapper.getSdp(), + OFFER_AUDIO | ANSWER_AUDIO); + a2_->SetLocal(TestObserver::ANSWER, a2_->answer(), false); + a1_->SetRemote(TestObserver::ANSWER, a2_->answer(), false); + ASSERT_NE(a2_->getLocalDescription().find("RTP/SAVPF 0 101\r"), std::string::npos); + ASSERT_NE(a2_->getLocalDescription().find("a=rtpmap:0 PCMU/8000"), std::string::npos); + ASSERT_EQ(a2_->getLocalDescription().find("a=rtpmap:109 opus/48000/2"), std::string::npos); + ASSERT_EQ(a2_->getLocalDescription().find("a=rtpmap:9 G722/8000"), std::string::npos); + + a1_->CloseSendStreams(); + a2_->CloseReceiveStreams(); +} + TEST_F(SignalingTest, FullCallAudioNoMuxVideoMux) { EnsureInit(); diff --git a/media/webrtc/trunk/webrtc/modules/audio_coding/main/acm2/audio_coding_module.gypi b/media/webrtc/trunk/webrtc/modules/audio_coding/main/acm2/audio_coding_module.gypi index 35b72f97b1fd..e347dc9823dc 100644 --- a/media/webrtc/trunk/webrtc/modules/audio_coding/main/acm2/audio_coding_module.gypi +++ b/media/webrtc/trunk/webrtc/modules/audio_coding/main/acm2/audio_coding_module.gypi @@ -46,6 +46,8 @@ 'acm_common_defs.h', 'acm_dtmf_playout.cc', 'acm_dtmf_playout.h', + 'acm_g722.cc', + 'acm_g722.h', 'acm_g729.cc', 'acm_g729.h', 'acm_g7291.cc', diff --git a/mobile/android/chrome/content/browser.js b/mobile/android/chrome/content/browser.js index bd13ff76a0dc..b1712880f286 100644 --- a/mobile/android/chrome/content/browser.js +++ b/mobile/android/chrome/content/browser.js @@ -6202,7 +6202,7 @@ var ViewportHandler = { let document = target.ownerDocument; let browser = BrowserApp.getBrowserForDocument(document); let tab = BrowserApp.getTabForBrowser(browser); - if (tab) + if (tab && tab.contentDocumentIsDisplayed) this.updateMetadata(tab, false); break; } diff --git a/modules/libpref/init/all.js b/modules/libpref/init/all.js index fd45f7e94c52..ba6c82c5ea7f 100644 --- a/modules/libpref/init/all.js +++ b/modules/libpref/init/all.js @@ -34,6 +34,8 @@ pref("general.warnOnAboutConfig", true); // maximum number of dated backups to keep at any time pref("browser.bookmarks.max_backups", 5); +// Delete HTTP cache v1 data +pref("browser.cache.auto_delete_cache_version", 0); // Preference for switching the cache backend, can be changed freely at runtime // 0 - use the old (Darin's) cache // 1 - use the new cache back-end (cache v2) diff --git a/netwerk/cache2/CacheObserver.cpp b/netwerk/cache2/CacheObserver.cpp index 8c85e1470430..7a21777101a2 100644 --- a/netwerk/cache2/CacheObserver.cpp +++ b/netwerk/cache2/CacheObserver.cpp @@ -108,7 +108,7 @@ CacheObserver::Init() obs->AddObserver(sSelf, "prefservice:after-app-defaults", true); obs->AddObserver(sSelf, "profile-do-change", true); - obs->AddObserver(sSelf, "sessionstore-windows-restored", true); + obs->AddObserver(sSelf, "browser-delayed-startup-finished", true); obs->AddObserver(sSelf, "profile-before-change", true); obs->AddObserver(sSelf, "xpcom-shutdown", true); obs->AddObserver(sSelf, "last-pb-context-exited", true); @@ -450,7 +450,7 @@ CacheObserver::Observe(nsISupports* aSubject, return NS_OK; } - if (!strcmp(aTopic, "sessionstore-windows-restored")) { + if (!strcmp(aTopic, "browser-delayed-startup-finished")) { uint32_t activeVersion = UseNewCache() ? 1 : 0; CacheStorageService::CleaupCacheDirectories(sAutoDeleteCacheVersion, activeVersion); return NS_OK; diff --git a/netwerk/cache2/CacheStorageService.cpp b/netwerk/cache2/CacheStorageService.cpp index ddd45b1004a7..d070feae35ca 100644 --- a/netwerk/cache2/CacheStorageService.cpp +++ b/netwerk/cache2/CacheStorageService.cpp @@ -594,6 +594,7 @@ NS_IMETHODIMP CleaupCacheDirectoriesRunnable::Run() } #if defined(MOZ_WIDGET_ANDROID) if (mCache2Profileless) { + nsDeleteDir::RemoveOldTrashes(mCache2Profileless); // Always delete the profileless cache on Android nsDeleteDir::DeleteDir(mCache2Profileless, true, 30000); } diff --git a/netwerk/dns/effective_tld_names.dat b/netwerk/dns/effective_tld_names.dat index 3336a443f1e0..108d88a92f8d 100644 --- a/netwerk/dns/effective_tld_names.dat +++ b/netwerk/dns/effective_tld_names.dat @@ -404,7 +404,7 @@ mil.bo tv.bo // br : http://registro.br/dominio/categoria.html -// Submitted by registry 2014-03-04 +// Submitted by registry 2014-08-11 br adm.br adv.br @@ -452,7 +452,7 @@ mil.br mp.br mus.br net.br -nom.br +*.nom.br not.br ntr.br odo.br @@ -6012,11 +6012,35 @@ mil.to // Submitted by Ryan Sleevi 2014-01-03 tp -// tr : http://en.wikipedia.org/wiki/.tr -*.tr -!nic.tr -// Used by government in the TRNC -// http://en.wikipedia.org/wiki/.nc.tr +// subTLDs: https://www.nic.tr/forms/eng/policies.pdf +// and: https://www.nic.tr/forms/politikalar.pdf +// Submitted by 2014-07-19 +tr +com.tr +info.tr +biz.tr +net.tr +org.tr +web.tr +gen.tr +tv.tr +av.tr +dr.tr +bbs.tr +name.tr +tel.tr +gov.tr +bel.tr +pol.tr +mil.tr +k12.tr +edu.tr +kep.tr + +// Used by Northern Cyprus +nc.tr + +// Used by government agencies of Northern Cyprus gov.nc.tr // travel : http://en.wikipedia.org/wiki/.travel @@ -6291,7 +6315,7 @@ k12.mo.us k12.ms.us k12.mt.us k12.nc.us -k12.nd.us +// k12.nd.us Bug 1028347 - Removed at request of Travis Rosso k12.ne.us k12.nh.us k12.nj.us @@ -6316,7 +6340,6 @@ k12.wa.us k12.wi.us // k12.wv.us Bug 947705 - Removed at request of Verne Britton k12.wy.us - cc.ak.us cc.al.us cc.ar.us @@ -6372,7 +6395,6 @@ cc.wa.us cc.wi.us cc.wv.us cc.wy.us - lib.ak.us lib.al.us lib.ar.us @@ -6428,7 +6450,6 @@ lib.wa.us lib.wi.us // lib.wv.us Bug 941670 - Removed at request of Larry W Arnold lib.wy.us - // k12.ma.us contains school districts in Massachusetts. The 4LDs are // managed indepedently except for private (PVT), charter (CHTR) and // parochial (PAROCH) schools. Those are delegated dorectly to the @@ -7999,13 +8020,13 @@ cern healthcare // xn--30rr7y : 2014-06-13 Excellent First Limited -xn--30rr7y +慈善 // band : 2014-06-13 Auburn Hollow, LLC band // xn--9et52u : 2014-06-13 RISE VICTORY LIMITED -xn--9et52u +时尚 // world : 2014-06-13 Bitter Fields, LLC world @@ -8032,10 +8053,10 @@ pharmacy dvag // xn--vermgensberater-ctb : 2014-06-23 Deutsche Vermögensberatung Aktiengesellschaft DVAG -xn--vermgensberater-ctb +vermögensberater // xn--vermgensberatung-pwb : 2014-06-23 Deutsche Vermögensberatung Aktiengesellschaft DVAG -xn--vermgensberatung-pwb +vermögensberatung // montblanc : 2014-06-23 Richemont DNS Inc. montblanc @@ -8586,6 +8607,10 @@ global.prod.fastly.net // Submitted by Chris Raynor 2014-01-21 firebaseapp.com +// Flynn : https://flynn.io +// Submitted by Jonathan Rudenberg 2014-07-12 +flynnhub.com + // GitHub, Inc. // Submitted by Ben Toews 2014-02-06 github.io @@ -8693,6 +8718,10 @@ rhcloud.com // Submitted by registry 2008-06-09 priv.at +// Yola : https://www.yola.com/ +// Submitted by Stefano Rivera 2014-07-09 +yolasite.com + // ZaNiC : http://www.za.net/ // Submitted by registry 2009-10-03 za.net diff --git a/testing/mochitest/runtests.py b/testing/mochitest/runtests.py index 9374a5284ec0..908370bf1e83 100644 --- a/testing/mochitest/runtests.py +++ b/testing/mochitest/runtests.py @@ -1332,7 +1332,7 @@ class Mochitest(MochitestUtilsMixin): symbolsPath=None, timeout=-1, onLaunch=None, - webapprtChrome=False, + detectShutdownLeaks=False, screenshotOnFail=False, testPath=None, bisectChunk=None, @@ -1383,7 +1383,7 @@ class Mochitest(MochitestUtilsMixin): testUrl = testUrl.replace("&", "\\&") args.append(testUrl) - if mozinfo.info["debug"] and not webapprtChrome: + if detectShutdownLeaks: shutdownLeaks = ShutdownLeaks(log.info) else: shutdownLeaks = None @@ -1743,6 +1743,9 @@ class Mochitest(MochitestUtilsMixin): if options.vmwareRecording: self.startVMwareRecording(options); + # detect shutdown leaks for m-bc runs + detectShutdownLeaks = mozinfo.info["debug"] and options.browserChrome and not options.webapprtChrome + log.info("runtests.py | Running tests: start.\n") try: status = self.runApp(testURL, @@ -1755,7 +1758,7 @@ class Mochitest(MochitestUtilsMixin): symbolsPath=options.symbolsPath, timeout=timeout, onLaunch=onLaunch, - webapprtChrome=options.webapprtChrome, + detectShutdownLeaks=detectShutdownLeaks, screenshotOnFail=options.screenshotOnFail, testPath=options.testPath, bisectChunk=options.bisectChunk, diff --git a/testing/mochitest/runtestsb2g.py b/testing/mochitest/runtestsb2g.py index c84e847d17c9..259c5a470e30 100644 --- a/testing/mochitest/runtestsb2g.py +++ b/testing/mochitest/runtestsb2g.py @@ -167,9 +167,9 @@ class B2GMochitest(MochitestUtilsMixin): # empty env. self.browserEnv = self.buildBrowserEnv(options, env={}) - # XXXkhuey MOZ_DISABLE_NONLOCAL_CONNECTIONS is busted on b2g, so make - # sure we don't pass it through (bug 1039019). - if 'MOZ_DISABLE_NONLOCAL_CONNECTIONS' in self.browserEnv: + # B2G emulator debug tests still make external connections, so don't + # pass MOZ_DISABLE_NONLOCAL_CONNECTIONS to them for now (bug 1039019). + if mozinfo.info['debug'] and 'MOZ_DISABLE_NONLOCAL_CONNECTIONS' in self.browserEnv: del self.browserEnv['MOZ_DISABLE_NONLOCAL_CONNECTIONS'] self.runner.env.update(self.browserEnv) diff --git a/testing/profiles/prefs_general.js b/testing/profiles/prefs_general.js index 9e92579d070e..b78a5e6235cb 100644 --- a/testing/profiles/prefs_general.js +++ b/testing/profiles/prefs_general.js @@ -104,6 +104,12 @@ user_pref("extensions.getAddons.search.browseURL", "http://%(server)s/extensions user_pref("extensions.getAddons.search.url", "http://%(server)s/extensions-dummy/repositorySearchURL"); // Make sure that opening the plugins check page won't hit the network user_pref("plugins.update.url", "http://%(server)s/plugins-dummy/updateCheckURL"); +// Make sure SNTP requests don't hit the network +user_pref("network.sntp.pools", "%(server)s"); +// We know the SNTP request will fail, since localhost isn't listening on +// port 135. The default number of retries (10) is excessive, but retrying +// at least once will mean that codepath is still tested in automation. +user_pref("network.sntp.maxRetryCount", 1); // Existing tests don't wait for the notification button security delay user_pref("security.notification_enable_delay", 0); diff --git a/toolkit/content/tests/chrome/test_bug437844.xul b/toolkit/content/tests/chrome/test_bug437844.xul index 8a651a6bc2b3..d58dfc68bec8 100644 --- a/toolkit/content/tests/chrome/test_bug437844.xul +++ b/toolkit/content/tests/chrome/test_bug437844.xul @@ -37,7 +37,7 @@ https://bugzilla.mozilla.org/show_bug.cgi?id=348233