Optimize unpremultiply on SSE4.1

Adds an SSE4.1 optimized version of qUnpremultiply and uses it in the
most drawing conversions methods. This gives a speed-up of little over
2x.

Change-Id: Ieb858a94ada1eb86d7af715ac1a100f1587f360d
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>

2 files changed, 33 insertions, 1 deletions

diff --git a/tests/auto/gui/painting/qcolor/qcolor.pro b/tests/auto/gui/painting/qcolor/qcolor.pro
index f7439c243c..44d65bb50b 100644
--- a/tests/auto/gui/painting/qcolor/qcolor.pro
+++ b/tests/auto/gui/painting/qcolor/qcolor.pro
@@ -2,4 +2,4 @@ CONFIG += testcase
 CONFIG += parallel_test
 TARGET = tst_qcolor
 SOURCES  += tst_qcolor.cpp
-QT += testlib
+QT += testlib gui-private core-private
diff --git a/tests/auto/gui/painting/qcolor/tst_qcolor.cpp b/tests/auto/gui/painting/qcolor/tst_qcolor.cpp
index 3b3334ba1f..95f1da1354 100644
--- a/tests/auto/gui/painting/qcolor/tst_qcolor.cpp
+++ b/tests/auto/gui/painting/qcolor/tst_qcolor.cpp
@@ -38,6 +38,7 @@
 
 #include <qcolor.h>
 #include <qdebug.h>
+#include <private/qdrawingprimitive_sse2_p.h>
 
 class tst_QColor : public QObject
 {
@@ -102,6 +103,9 @@ private slots:
 
     void achromaticHslHue();
 
+    void premultiply();
+    void unpremultiply_sse4();
+
 #ifdef Q_DEAD_CODE_FROM_QT4_X11
     void setallowX11ColorNames();
 #endif
@@ -1432,5 +1436,33 @@ void tst_QColor::setallowX11ColorNames()
 }
 #endif
 
+void tst_QColor::premultiply()
+{
+    // Tests that qPremultiply(qUnpremultiply(x)) returns x.
+    for (uint a = 0; a < 256; a++) {
+        for (uint c = 0; c <= a; c++) {
+            QRgb p = qRgba(c, a-c, c, a);
+            QCOMPARE(p, qPremultiply(qUnpremultiply(p)));
+        }
+    }
+}
+
+void tst_QColor::unpremultiply_sse4()
+{
+    // Tests that qUnpremultiply_sse4 returns the same as qUnpremultiply.
+#if QT_COMPILER_SUPPORTS_HERE(SSE4_1)
+    if (qCpuHasFeature(SSE4_1)) {
+        for (uint a = 0; a < 256; a++) {
+            for (uint c = 0; c <= a; c++) {
+                QRgb p = qRgba(c, a-c, c, a);
+                QCOMPARE(qUnpremultiply(p), qUnpremultiply_sse4(p));
+            }
+        }
+        return;
+    }
+#endif
+    QSKIP("SSE4 not supported on this CPU.");
+}
+
 QTEST_MAIN(tst_QColor)
 #include "tst_qcolor.moc"