From 6e1e9b9782c2e1783cdce32fc65e8e25c9290a64 Mon Sep 17 00:00:00 2001
From: "Christian R. Helmrich" <c.helmrich@ecodis.de>
Date: Tue, 31 Mar 2020 20:00:00 +0200
Subject: [PATCH] finalize release

---
 README.md                    |  4 +++
 include/Release.htm          | 17 +++++++++----
 src/lib/stereoProcessing.cpp | 49 +++++++++++++++++++++++++++---------
 3 files changed, 53 insertions(+), 17 deletions(-)

diff --git a/README.md b/README.md
index 55b6738..6e61997 100644
--- a/README.md
+++ b/README.md
@@ -7,6 +7,10 @@ encode uncompressed WAVE-format audio files into MPEG-4-format files
 complying with the ISO/IEC 23003-3 (MPEG-D) Unified Speech and Audio
 Coding (USAC, also known as Extended High-Efficiency AAC) standard.
 
+In addition, exhale writes program peak-level and loudness data into
+the generated MPEG-4 files according to the ISO/IEC 23003-4, Dynamic
+Range Control (DRC) specification for use by decoders providing DRC.
+
 exhale currently makes use of all frequency-domain (FD) coding tools
 in the scalefactor based MDCT processing path, except for predictive
 joint stereo, which is still being integrated. Its objective is high
diff --git a/include/Release.htm b/include/Release.htm
index e0f4852..9663a96 100644
--- a/include/Release.htm
+++ b/include/Release.htm
@@ -25,9 +25,16 @@
 <td valign="top">
 
 <h1><br><span class="pink">exhale</span> - <span class="pink">e</span>codis e<span class="pink">x</span>tended <span class="pink">h</span>igh-efficiency <span class="pink">a</span>nd <span class="pink">l</span>ow-complexity <span class="pink">e</span>ncoder<br><span class="gray"><sup><br>Software Release Notes, Version History, Known Issues, Upcoming Feature Roadmap</sup></span><br><br></h1>
-<h3>&nbsp; &nbsp;The version of this distribution of the &laquo;exhale&raquo; software release is <b>1.0.1</b> (official pub&shy;lic minor release) from February 2020. Please check <a href="http://www.ecodis.de">www.ecodis.de</a> regularly for new versions of this software. A summary of each version up to this release, a list of known issues with this release, and a roadmap of additional functionality are provided below.</h3>
+<h3>&nbsp; &nbsp;The version of this distribution of the &laquo;exhale&raquo; software release is <b>1.0.2</b> (official pub&shy;lic minor release) from March 2020. Please check <a href="http://www.ecodis.de">www.ecodis.de</a> regularly for new versions of this software. A summary of each version up to this release, a list of known issues with this release, and a roadmap of additional functionality are provided below.</h3>
 <h3><br><b>Chronological Version History</b></h3>
-<h3>&nbsp; &nbsp;Version <b>1.0.1 <span class="gray">&nbsp;Feb. 2020, this release</span></b></h3>
+<h3>&nbsp; &nbsp;Version <b>1.0.2 <span class="gray">&nbsp;Mar. 2020, this release</span></b></h3>
+<ul>
+ <li><h3>added basic low/mid-rate joint-stereo coding functionality, bugfixes, and speedups</h3></li>
+ <li><h3>exhaleApp: support for input sampling rates of up to 48000 Hz with CVBR mode 2</h3></li>
+ <li><h3>exhaleLib: frame adaptive joint-stereo preprocessing and coding (CVBR mode &lt;5)</h3></li>
+ <li><h3>exhaleLib: accelerated R/D opt. coding, stability and quality fixes (issues 2 and 3)</h3></li>
+</ul>
+<h3>&nbsp; &nbsp;Version <b>1.0.1 <span class="gray">&nbsp;Feb. 2020</span></b></h3>
 <ul>
  <li><h3>improved low-bitrate coding efficiency and support for MPEG-D loudness metadata</h3></li>
  <li><h3>exhaleApp: increased MP4 file versatility (issue 1) and calculation of loudness info</h3></li>
@@ -49,20 +56,20 @@
 <h3><br><b>Known Issues with This Release</b></h3>
 <h3>&nbsp; &nbsp;If you notice an issue with this release <b>not</b> mentioned below, please contact ecodis or a contributor with the details (configuration, input file) needed to reproduce the issue.</h3>
 <ul>
- <li><h3>exhaleLib: Coding of stereo or multichannel input yields suboptimal audio quality because the joint-channel coding functionality provided by the ISO/IEC <a href="https://www.iso.org/standard/76385.html">23003-3</a> standard has not been implemented so far. See the functionality roadmap below.</h3></li>
+ <li><h3>exhaleLib: Coding of stereo or multichannel input occasionally leads to suboptimal audio quality because the joint-channel coding functionality provided by ISO/IEC <a href="https://www.iso.org/standard/76385.html">23003-3</a> has not been fully implemented. See the functionality roadmap below.</h3></li>
  <li><h3>exhaleApp: Only basic WAVE input file reading functionality has been implemen&shy;ted. Specifically, 8-bit WAVE input is assumed to contain an even number of audio samples, and the Broadcast and Extensible WAVE file formats are not supported.</h3></li>
 </ul>
 <h3><br><b>Roadmap of Upcoming Features</b></h3>
 <h3>&nbsp; &nbsp;If you are in need of an additional library or application feature <b>not</b> mentioned below, please contact ecodis or a contributor with a request, and we will see what we can do.</h3>
 <ul>
  <li><h3>support for coding with a core coder frame length of 768 samples, no version plan</h3></li>
- <li><h3>exhaleLib: finalized integration of joint-channel coding functionality, version 1.0.2</h3></li>
+ <li><h3>exhaleLib: finalized integration of higher-rate joint-channel coding, version 1.0.3</h3></li>
  <li><h3>exhaleLib: quality tuning and bug fixing for low-rate stereo coding, version 1.0.3</h3></li>
  <li><h3>exhaleLib: finalization of support for 3.0 &#x2013; 5.1 multichannel coding, version 1.0.4</h3></li>
  <li><h3>exhaleLib: speed-ups and further quality tuning for critical signals, version 1.0.5.</h3></li>
 </ul>
 <h3><br></h3>
-<h4><span class="gray">Written by C. R. Helmrich for exhale 1.0.1, Feb. 2020. Available at www.ecodis.de/exhale/release.htm.</span><br><br></h4>
+<h4><span class="gray">Written by C. R. Helmrich for exhale 1.0.2, Mar. 2020. Available at www.ecodis.de/exhale/release.htm.</span><br><br></h4>
 
 </td>
 <td valign="top" colspan="2">
diff --git a/src/lib/stereoProcessing.cpp b/src/lib/stereoProcessing.cpp
index a4a4fb3..1e6eaf4 100644
--- a/src/lib/stereoProcessing.cpp
+++ b/src/lib/stereoProcessing.cpp
@@ -11,10 +11,6 @@
 #include "exhaleLibPch.h"
 #include "stereoProcessing.h"
 
-// static helper functions
-
-// private helper function
-
 // constructor
 StereoProcessor::StereoProcessor ()
 {
@@ -76,8 +72,8 @@ unsigned StereoProcessor::applyFullFrameMatrix (int32_t* const mdctSpectrum1, in
 
       if (realOnlyCalc) // real data, only MDCTs are available
       {
-        int32_t* sfbNext1 = &sfbMdct1[1];
-        int32_t* sfbNext2 = &sfbMdct2[1];
+        const int32_t* sfbNext1 = &sfbMdct1[1];
+        const int32_t* sfbNext2 = &sfbMdct2[1];
 
         for (uint16_t s = sfbWidth - (sfb + 1 == numSwbFrame ? 1 : 0); s > 0; s--)
         {
@@ -151,7 +147,7 @@ unsigned StereoProcessor::applyFullFrameMatrix (int32_t* const mdctSpectrum1, in
       grpRms1[sfb] = uint32_t ((sumAbsValM + (sfbWidth >> 1)) / sfbWidth);
       grpRms2[sfb] = uint32_t ((sumAbsValS + (sfbWidth >> 1)) / sfbWidth);
 
-      sfbStereoData[sfb + numSwbFrame * gr] = 16; // alpha = 0
+      if (applyPredSte) sfbStereoData[sfb + numSwbFrame * gr] = 16; // initialize to alpha=0
 
       if ((sfbIsOdd) || (sfb + 1 == maxSfbSte)) // finish pair
       {
@@ -175,7 +171,36 @@ unsigned StereoProcessor::applyFullFrameMatrix (int32_t* const mdctSpectrum1, in
           }
           if (realOnlyCalc) // real data, only MDCTs available
           {
-            // TODO
+            const int32_t* nextA = (alterPredDir ? &mdctSpectrum2[offEv + 1] : &mdctSpectrum1[offEv + 1]);
+            const int32_t* nextB = (alterPredDir ? &mdctSpectrum1[offEv + 1] : &mdctSpectrum2[offEv + 1]);
+
+            if (sfbEv == 0) // exclude unavailable DC estimate
+            {
+              mdctA -= width; nextA++;
+              mdctB -= width; nextB++;
+            }
+            else
+            {
+              mdctA -= (width + 1); // point to pre-
+              mdctB -= (width + 1); // vious samples
+            }
+            for (uint16_t s = width - (sfbEv == 0 ? 2 : 1); s > 0; s--) // exclude unavailable final estimate
+            {
+              const int64_t mdstA = ((int64_t) *(nextA++) - (int64_t) *(mdctA++)) >> 1; // estimate, see also
+              const int64_t mdstB = ((int64_t) *(nextB++) - (int64_t) *(mdctB++)) >> 1; // getMeanAbsValues()
+
+              sumPrdRefRes += (mdstA * mdstB + SA_BW) >> (SA_BW_SHIFT + 1);
+              sumPrdRefRef += (mdstA * mdstA + SA_BW) >> (SA_BW_SHIFT + 1);
+            }
+            if (sfb + 1 < numSwbFrame) // process final sample
+            {
+              const int64_t msSgn = (alterPredDir ? -1 : 1);
+              const int64_t mdstA = ((((int64_t) *nextB + *nextA * msSgn + 1) >> 1) - (int64_t) *mdctA) >> 1;
+              const int64_t mdstB = ((((int64_t) *nextA - *nextB * msSgn + 1) >> 1) - (int64_t) *mdctB) >> 1;
+
+              sumPrdRefRes += (mdstA * mdstB + SA_BW) >> (SA_BW_SHIFT + 1);
+              sumPrdRefRef += (mdstA * mdstA + SA_BW) >> (SA_BW_SHIFT + 1);
+            }
           }
           else // complex data, both MDCTs and MDSTs available
           {
@@ -277,7 +302,7 @@ unsigned StereoProcessor::applyFullFrameMatrix (int32_t* const mdctSpectrum1, in
 
   if (numSfbPredSte <= 1) // discard prediction coefficients and stay with legacy M/S stereo
   {
-    memset (sfbStereoData, 16, numSwbFrame * grp.numWindowGroups * sizeof (uint8_t));
+    if (applyPredSte) memset (sfbStereoData, 16, numSwbFrame * grp.numWindowGroups * sizeof (uint8_t));
 
     numSfbPredSte = 0;
   }
@@ -290,7 +315,7 @@ unsigned StereoProcessor::applyFullFrameMatrix (int32_t* const mdctSpectrum1, in
       const uint16_t*  grpOff = &grp.sfbOffsets[numSwbFrame * gr];
       uint32_t* const grpRms1 = &groupingData1.sfbRmsValues[numSwbFrame * gr];
       uint32_t* const grpRms2 = &groupingData2.sfbRmsValues[numSwbFrame * gr];
-      int32_t b = 0, prevResi = 0;
+      int32_t prevResi = 0;
 
       if (realOnlyCalc) // preparation of res. magnitude value
       {
@@ -316,8 +341,8 @@ unsigned StereoProcessor::applyFullFrameMatrix (int32_t* const mdctSpectrum1, in
 
         if (realOnlyCalc) // real data, only MDCT is available
         {
-          int32_t* sfbNextD = &sfbMdctD[1];
-          int32_t* sfbNextR = &sfbMdctR[1];
+          const int32_t* sfbNextD = &sfbMdctD[1];
+          const int32_t* sfbNextR = &sfbMdctR[1];
 
           for (uint16_t s = sfbWidth - (sfb + 1 == numSwbFrame ? 1 : 0); s > 0; s--)
           {