Native vector size queries for OpenCL 1.1 devices.

4 files changed, 112 insertions, 2 deletions

diff --git a/doc/src/opencl11.qdoc b/doc/src/opencl11.qdoc
index 000d98e..3bf1862 100644
--- a/doc/src/opencl11.qdoc
+++ b/doc/src/opencl11.qdoc
@@ -57,7 +57,10 @@
     \o New QCLDevice queries:
        \l{QCLDevice::languageVersion()}{languageVersion()},
        \l{QCLDevice::preferredHalfFloatVectorSize()}{preferredHalfFloatVectorSize()},
-       and \l{QCLDevice::hasUnifiedMemory()}{hasUnifiedMemory()}.
+       \l{QCLDevice::hasUnifiedMemory()}{hasUnifiedMemory()},
+       \l{QCLDevice::nativeCharVectorSize()}{nativeCharVectorSize()},
+       \l{QCLDevice::nativeShortVectorSize()}{nativeShortVectorSize()},
+       etc.
     \o The \c{setOutOfOrder()} and \c{setProfilingEnabled()}
        functions have been removed from QCLCommandQueue because
        OpenCL 1.1 has deprecated the functionality, and we want
diff --git a/src/opencl/qcldevice.cpp b/src/opencl/qcldevice.cpp
index 986c920..30b510a 100644
--- a/src/opencl/qcldevice.cpp
+++ b/src/opencl/qcldevice.cpp
@@ -563,6 +563,89 @@ int QCLDevice::preferredHalfFloatVectorSize() const
 }
 
 /*!
+    Returns the native size for vectors of type \c{char}
+    in the device.  For example, 4 indicates that 4 \c{char}
+    values can be packed into a vector and operated on as a
+    unit for optimal performance.  Returns zero on OpenCL 1.0.
+*/
+int QCLDevice::nativeCharVectorSize() const
+{
+    return qt_cl_paramInt(m_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_CHAR);
+}
+
+/*!
+    Returns the native size for vectors of type \c{short}
+    in the device.  For example, 4 indicates that 4 \c{short}
+    values can be packed into a vector and operated on as a
+    unit for optimal performance.  Returns zero on OpenCL 1.0.
+*/
+int QCLDevice::nativeShortVectorSize() const
+{
+    return qt_cl_paramInt(m_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_SHORT);
+}
+
+/*!
+    Returns the native size for vectors of type \c{int}
+    in the device.  For example, 4 indicates that 4 \c{int}
+    values can be packed into a vector and operated on as a
+    unit for optimal performance.  Returns zero on OpenCL 1.0.
+*/
+int QCLDevice::nativeIntVectorSize() const
+{
+    return qt_cl_paramInt(m_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_INT);
+}
+
+/*!
+    Returns the native size for vectors of type \c{long}
+    in the device.  For example, 2 indicates that 2 \c{long}
+    values can be packed into a vector and operated on as a
+    unit for optimal performance.  Returns zero on OpenCL 1.0.
+*/
+int QCLDevice::nativeLongVectorSize() const
+{
+    return qt_cl_paramInt(m_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_LONG);
+}
+
+/*!
+    Returns the native size for vectors of type \c{float}
+    in the device.  For example, 4 indicates that 4 \c{float}
+    values can be packed into a vector and operated on as a
+    unit for optimal performance.  Returns zero on OpenCL 1.0.
+*/
+int QCLDevice::nativeFloatVectorSize() const
+{
+    return qt_cl_paramInt(m_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_FLOAT);
+}
+
+/*!
+    Returns the native size for vectors of type \c{double}
+    in the device.  For example, 2 indicates that 2 \c{double}
+    values can be packed into a vector and operated on as a
+    unit for optimal performance.  Returns zero on OpenCL 1.0,
+    or if the device does not support \c{double}.
+
+    \sa hasDouble()
+*/
+int QCLDevice::nativeDoubleVectorSize() const
+{
+    return qt_cl_paramInt(m_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_DOUBLE);
+}
+
+/*!
+    Returns the native size for vectors of type \c{half}
+    in the device.  For example, 2 indicates that 2 \c{half}
+    values can be packed into a vector and operated on as a
+    unit for optimal performance.  Returns zero on OpenCL 1.0,
+    or if the device does not support \c{half}.
+
+    \sa hasHalfFloat()
+*/
+int QCLDevice::nativeHalfFloatVectorSize() const
+{
+    return qt_cl_paramInt(m_id, CL_DEVICE_NATIVE_VECTOR_WIDTH_HALF);
+}
+
+/*!
     \enum QCLDevice::FloatCapability
     This enum defines the floating-point capabilities of the
     \c{float} or \c{double} type on an OpenCL device.
diff --git a/src/opencl/qcldevice.h b/src/opencl/qcldevice.h
index a874da2..b73d0cd 100644
--- a/src/opencl/qcldevice.h
+++ b/src/opencl/qcldevice.h
@@ -107,6 +107,14 @@ public:
     int preferredDoubleVectorSize() const;
     int preferredHalfFloatVectorSize() const;
 
+    int nativeCharVectorSize() const;
+    int nativeShortVectorSize() const;
+    int nativeIntVectorSize() const;
+    int nativeLongVectorSize() const;
+    int nativeFloatVectorSize() const;
+    int nativeDoubleVectorSize() const;
+    int nativeHalfFloatVectorSize() const;
+
     enum FloatCapability
     {
         NotSupported        = 0x0000,
diff --git a/util/clinfo/clinfo.cpp b/util/clinfo/clinfo.cpp
index b13ffb8..276d9f7 100644
--- a/util/clinfo/clinfo.cpp
+++ b/util/clinfo/clinfo.cpp
@@ -183,8 +183,24 @@ int main(int argc, char *argv[])
             printf(", double%d", dsize);
         int hsize = dev.preferredHalfFloatVectorSize();
         if (hsize)
-            printf(", half%d\n", hsize);
+            printf(", half%d", hsize);
         printf("\n");
+        if (dev.nativeCharVectorSize()) {
+            printf("    Native Vector Sizes:\n");
+            printf("        char%d, short%d, int%d, long%d, float%d",
+                   dev.nativeCharVectorSize(),
+                   dev.nativeShortVectorSize(),
+                   dev.nativeIntVectorSize(),
+                   dev.nativeLongVectorSize(),
+                   dev.nativeFloatVectorSize());
+            dsize = dev.nativeDoubleVectorSize();
+            if (dsize)
+                printf(", double%d", dsize);
+            hsize = dev.nativeHalfFloatVectorSize();
+            if (hsize)
+                printf(", half%d", hsize);
+            printf("\n");
+        }
         printf("    Extensions        :\n");
         QStringList extns = dev.extensions();
         foreach (QString ext, extns)