diff --git a/autotests/CMakeLists.txt b/autotests/CMakeLists.txt
index 78772c1..adaab43 100644
--- a/autotests/CMakeLists.txt
+++ b/autotests/CMakeLists.txt
@@ -161,9 +161,7 @@ if (LibJXR_FOUND)
     )
 endif()
 
-# Allow some fuzziness when reading this formats, to allow for
-# rounding errors (eg: in alpha blending).
-kimageformats_read_tests(FUZZ 1
+kimageformats_read_tests(
     xcf
 )
 
diff --git a/autotests/read/dds/rgba16dx10.dds.json b/autotests/read/dds/rgba16dx10.dds.json
new file mode 100644
index 0000000..afc2536
--- /dev/null
+++ b/autotests/read/dds/rgba16dx10.dds.json
@@ -0,0 +1,7 @@
+[
+    {
+        "fileName" : "rgba16dx10.png",
+        "fuzziness" : 1,
+        "description" : "Minimum fuzziness value to pass the test on all architectures."
+    }
+]
diff --git a/autotests/read/dds/rgba_f16.dds.json b/autotests/read/dds/rgba_f16.dds.json
new file mode 100644
index 0000000..9f974b6
--- /dev/null
+++ b/autotests/read/dds/rgba_f16.dds.json
@@ -0,0 +1,7 @@
+[
+    {
+        "fileName" : "rgba_f16.png",
+        "fuzziness" : 1,
+        "description" : "Minimum fuzziness value to pass the test on all architectures."
+    }
+]
diff --git a/autotests/readtest.cpp b/autotests/readtest.cpp
index bdef725..020eb7e 100644
--- a/autotests/readtest.cpp
+++ b/autotests/readtest.cpp
@@ -377,7 +377,12 @@ int main(int argc, char **argv)
                                         << " to " << formatToString(cmpFormat) << '\n';
                     expImage = expImage.convertToFormat(cmpFormat);
                 }
-                if (fuzzyeq(inputImage, expImage, fuzziness)) {
+                auto tmpFuzziness = fuzziness;
+                if (tmpFuzziness == 0) {
+                    // If the fuzziness value is not explicitly set I use the one set for the current image.
+                    tmpFuzziness = timg.fuzziness();
+                }
+                if (fuzzyeq(inputImage, expImage, tmpFuzziness)) {
                     QTextStream(stdout) << "PASS : " << fi.fileName() << "\n";
                     ++passed;
                 } else {
diff --git a/autotests/templateimage.cpp b/autotests/templateimage.cpp
index 4e44097..afd0b2c 100644
--- a/autotests/templateimage.cpp
+++ b/autotests/templateimage.cpp
@@ -126,6 +126,15 @@ bool TemplateImage::checkOptionaInfo(const QImage& image, QString& error) const
     return true;
 }
 
+quint8 TemplateImage::fuzziness() const
+{
+    auto obj = searchObject(m_fi);
+    if (obj.isEmpty()) {
+        return quint8(0);
+    }
+    return quint8(obj.value("fuzziness").toInt());
+}
+
 QStringList TemplateImage::suffixes()
 {
     return QStringList({"png", "tif", "tiff", "json"});
diff --git a/autotests/templateimage.h b/autotests/templateimage.h
index c613a59..0e515c1 100644
--- a/autotests/templateimage.h
+++ b/autotests/templateimage.h
@@ -63,13 +63,25 @@ public:
 
     /*!
      * \brief checkOptionaInfo
-     * Verify the optional information (resolution, metadata, etc.) of the image with that in the template if present.
+     * Verify the optional information (resolution, metadata, etc.) of the
+     * image with that in the template if present.
      * \param image The image to check optional information on.
      * \param error The error message when returns false.
      * \return True on success, otherwise false.
      */
     bool checkOptionaInfo(const QImage& image, QString& error) const;
 
+    /*!
+     * \brief fuzziness
+     * The fuzziness value that ensures the test works correctly. Normally
+     * set for lossy codecs and images that require floating point
+     * conversions.
+     * Floating point conversions may give slightly different results from
+     * one architecture to another (Intel, PowerPC, Arm, etc...).
+     * \return The default fuzziness value for the image. Zero means no fuzziness.
+     */
+    quint8 fuzziness() const;
+
     /*!
      * \brief suffixes
      * \return The list of suffixes considered templates.
diff --git a/src/imageformats/dds.cpp b/src/imageformats/dds.cpp
index eb2b592..9755709 100644
--- a/src/imageformats/dds.cpp
+++ b/src/imageformats/dds.cpp
@@ -1031,11 +1031,7 @@ static QImage readUnsignedImage(QDataStream &s, const DDSHeader &dds, quint32 wi
 static qfloat16 readFloat16(QDataStream &s)
 {
     qfloat16 f16;
-
-    quint16 rawData;
-    s >> rawData;
-    memcpy(&f16, &rawData, sizeof(rawData));
-
+    s >> f16;
     return f16;
 }