Implement qt_memfill32 with Neon.

This patch introduce a implementation of qt_memfill32 with the Neon
instructions set from ARMv7.

The loop is unrolled 1 time to get better performance.

This implementation of memfill is 330% faster on the N900.

Reviewed-by: Samuel Rødal

3 files changed, 40 insertions, 0 deletions

diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index 0b74fc0e3a..be4275c83a 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -7913,6 +7913,7 @@ void qInitDrawhelperAsm()
 
             qMemRotateFunctions[QImage::Format_RGB16][0] = qt_memrotate90_16_neon;
             qMemRotateFunctions[QImage::Format_RGB16][2] = qt_memrotate270_16_neon;
+            qt_memfill32 = qt_memfill32_neon;
         }
 #endif
 
diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp
index c1f815d636..ed15c5cf8f 100644
--- a/src/gui/painting/qdrawhelper_neon.cpp
+++ b/src/gui/painting/qdrawhelper_neon.cpp
@@ -51,6 +51,44 @@
 
 QT_BEGIN_NAMESPACE
 
+void qt_memfill32_neon(quint32 *dest, quint32 value, int count)
+{
+    const int epilogueSize = count % 16;
+    if (count >= 16) {
+        quint32 *const neonEnd = dest + count - epilogueSize;
+        register uint32x4_t valueVector1 asm ("q0") = vdupq_n_u32(value);
+        register uint32x4_t valueVector2 asm ("q1") = valueVector1;
+        while (dest != neonEnd) {
+            asm volatile (
+                "vst2.32     { d0, d1, d2, d3 }, [%[DST]] !\n\t"
+                "vst2.32     { d0, d1, d2, d3 }, [%[DST]] !\n\t"
+                : [DST]"+r" (dest)
+                : [VALUE1]"w"(valueVector1), [VALUE2]"w"(valueVector2)
+                : "memory"
+            );
+        }
+    }
+
+    switch (epilogueSize)
+    {
+    case 15:     *dest++ = value;
+    case 14:     *dest++ = value;
+    case 13:     *dest++ = value;
+    case 12:     *dest++ = value;
+    case 11:     *dest++ = value;
+    case 10:     *dest++ = value;
+    case 9:      *dest++ = value;
+    case 8:      *dest++ = value;
+    case 7:      *dest++ = value;
+    case 6:      *dest++ = value;
+    case 5:      *dest++ = value;
+    case 4:      *dest++ = value;
+    case 3:      *dest++ = value;
+    case 2:      *dest++ = value;
+    case 1:      *dest++ = value;
+    }
+}
+
 static inline uint16x8_t qvdiv_255_u16(uint16x8_t x, uint16x8_t half)
 {
     // result = (x + (x >> 8) + 0x80) >> 8
diff --git a/src/gui/painting/qdrawhelper_neon_p.h b/src/gui/painting/qdrawhelper_neon_p.h
index 182c936627..451edbc36b 100644
--- a/src/gui/painting/qdrawhelper_neon_p.h
+++ b/src/gui/painting/qdrawhelper_neon_p.h
@@ -120,6 +120,7 @@ void qt_transform_image_rgb16_on_rgb16_neon(uchar *destPixels, int dbpl,
                                             const QTransform &targetRectTransform,
                                             int const_alpha);
 
+void qt_memfill32_neon(quint32 *dest, quint32 value, int count);
 void qt_memrotate90_16_neon(const uchar *srcPixels, int w, int h, int sbpl, uchar *destPixels, int dbpl);
 void qt_memrotate270_16_neon(const uchar *srcPixels, int w, int h, int sbpl, uchar *destPixels, int dbpl);