1 files changed, 53 insertions, 2 deletions
diff --git a/src/gui/painting/qdrawhelper_sse4.cpp b/src/gui/painting/qdrawhelper_sse4.cpp
index 257bad9eca..14bfaabf09 100644
--- a/src/gui/painting/qdrawhelper_sse4.cpp
+++ b/src/gui/painting/qdrawhelper_sse4.cpp
@@ -44,16 +44,67 @@
 
 QT_BEGIN_NAMESPACE
 
+template<bool RGBA>
+static inline void convertARGBToARGB32PM_sse4(uint *buffer, const uint *src, int count)
+{
+    int i = 0;
+    const __m128i alphaMask = _mm_set1_epi32(0xff000000);
+    const __m128i rgbaMask = _mm_setr_epi8(2, 1, 0, 3, 6, 5, 4, 7, 10, 9, 8, 11, 14, 13, 12, 15);
+    const __m128i shuffleMask = _mm_setr_epi8(6, 7, 6, 7, 6, 7, 6, 7, 14, 15, 14, 15, 14, 15, 14, 15);
+    const __m128i half = _mm_set1_epi16(0x0080);
+    const __m128i zero = _mm_setzero_si128();
+
+    for (; i < count - 3; i += 4) {
+        __m128i srcVector = _mm_loadu_si128((const __m128i *)&src[i]);
+        if (!_mm_testz_si128(srcVector, alphaMask)) {
+            if (!_mm_testc_si128(srcVector, alphaMask)) {
+                if (RGBA)
+                    srcVector = _mm_shuffle_epi8(srcVector, rgbaMask);
+                __m128i src1 = _mm_unpacklo_epi8(srcVector, zero);
+                __m128i src2 = _mm_unpackhi_epi8(srcVector, zero);
+                __m128i alpha1 = _mm_shuffle_epi8(src1, shuffleMask);
+                __m128i alpha2 = _mm_shuffle_epi8(src2, shuffleMask);
+                src1 = _mm_mullo_epi16(src1, alpha1);
+                src2 = _mm_mullo_epi16(src2, alpha2);
+                src1 = _mm_add_epi16(src1, _mm_srli_epi16(src1, 8));
+                src2 = _mm_add_epi16(src2, _mm_srli_epi16(src2, 8));
+                src1 = _mm_add_epi16(src1, half);
+                src2 = _mm_add_epi16(src2, half);
+                src1 = _mm_srli_epi16(src1, 8);
+                src2 = _mm_srli_epi16(src2, 8);
+                src1 = _mm_blend_epi16(src1, alpha1, 0x88);
+                src2 = _mm_blend_epi16(src2, alpha2, 0x88);
+                srcVector = _mm_packus_epi16(src1, src2);
+                _mm_storeu_si128((__m128i *)&buffer[i], srcVector);
+            } else {
+                if (RGBA)
+                    _mm_storeu_si128((__m128i *)&buffer[i], _mm_shuffle_epi8(srcVector, rgbaMask));
+                else if (buffer != src)
+                    _mm_storeu_si128((__m128i *)&buffer[i], srcVector);
+            }
+        } else {
+            _mm_storeu_si128((__m128i *)&buffer[i], _mm_setzero_si128());
+        }
+    }
+
+    SIMD_EPILOGUE(i, count, 3) {
+        uint v = qPremultiply(src[i]);
+        buffer[i] = RGBA ? RGBA2ARGB(v) : v;
+    }
+}
+
 const uint *QT_FASTCALL convertARGB32ToARGB32PM_sse4(uint *buffer, const uint *src, int count,
                                                      const QVector<QRgb> *, QDitherInfo *)
 {
-    return qt_convertARGB32ToARGB32PM(buffer, src, count);
+    convertARGBToARGB32PM_sse4<false>(buffer, src, count);
+    return buffer;
 }
 
 const uint *QT_FASTCALL convertRGBA8888ToARGB32PM_sse4(uint *buffer, const uint *src, int count,
                                                        const QVector<QRgb> *, QDitherInfo *)
 {
-    return qt_convertRGBA8888ToARGB32PM(buffer, src, count);
+    convertARGBToARGB32PM_sse4<true>(buffer, src, count);
+    return buffer;
 }
 
 const uint *QT_FASTCALL convertARGB32FromARGB32PM_sse4(uint *buffer, const uint *src, int count,