From a214bb99775e898fd9233b932f08ff91a57fc053 Mon Sep 17 00:00:00 2001
From: Benjamin Poulain <benjamin.poulain@nokia.com>
Date: Wed, 4 Aug 2010 11:46:40 +0200
Subject: Implement comp_Source with SSE2 when there is a const alpha

On Atom, comp_Source is 280% faster with the SSE2 implementation.

Reviewed-by: Andreas Kling
---
 src/gui/painting/qdrawhelper.cpp               |  2 ++
 src/gui/painting/qdrawhelper_sse2.cpp          | 33 ++++++++++++++++++++++++++
 tests/benchmarks/gui/image/blendbench/main.cpp | 32 +++++++++++++++++++++++++
 3 files changed, 67 insertions(+)

diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index 4ce2bee..1ff3d7b 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -7820,8 +7820,10 @@ void qInitDrawhelperAsm()
                                                   uint const_alpha);
             extern void QT_FASTCALL comp_func_solid_SourceOver_sse2(uint *destPixels, int length, uint color, uint const_alpha);
             extern void QT_FASTCALL comp_func_Plus_sse2(uint *dst, const uint *src, int length, uint const_alpha);
+            extern void QT_FASTCALL comp_func_Source_sse2(uint *dst, const uint *src, int length, uint const_alpha);
 
             functionForModeAsm[0] = comp_func_SourceOver_sse2;
+            functionForModeAsm[QPainter::CompositionMode_Source] = comp_func_Source_sse2;
             functionForModeAsm[QPainter::CompositionMode_Plus] = comp_func_Plus_sse2;
             functionForModeSolidAsm[0] = comp_func_solid_SourceOver_sse2;
 
diff --git a/src/gui/painting/qdrawhelper_sse2.cpp b/src/gui/painting/qdrawhelper_sse2.cpp
index e090ae5..7ab9eda 100644
--- a/src/gui/painting/qdrawhelper_sse2.cpp
+++ b/src/gui/painting/qdrawhelper_sse2.cpp
@@ -229,6 +229,39 @@ void QT_FASTCALL comp_func_Plus_sse2(uint *dst, const uint *src, int length, uin
     }
 }
 
+void QT_FASTCALL comp_func_Source_sse2(uint *dst, const uint *src, int length, uint const_alpha)
+{
+    if (const_alpha == 255) {
+        ::memcpy(dst, src, length * sizeof(uint));
+    } else {
+        const int ialpha = 255 - const_alpha;
+
+        int x = 0;
+
+        // 1) prologue, align on 16 bytes
+        const int offsetToAlignOn16Bytes = (4 - ((reinterpret_cast<quintptr>(dst) >> 2) & 0x3)) & 0x3;
+        const int prologLength = qMin(length, offsetToAlignOn16Bytes);
+        for (; x < prologLength; ++x)
+            dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], ialpha);
+
+        // 2) interpolate pixels with SSE2
+        const __m128i half = _mm_set1_epi16(0x80);
+        const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
+        const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
+        const __m128i oneMinusConstAlpha =  _mm_set1_epi16(ialpha);
+        for (; x < length - 3; x += 4) {
+            const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]);
+            __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
+            INTERPOLATE_PIXEL_255_SSE2(dstVector, srcVector, dstVector, constAlphaVector, oneMinusConstAlpha, colorMask, half)
+            _mm_store_si128((__m128i *)&dst[x], dstVector);
+        }
+
+        // 3) Epilogue
+        for (; x < length; ++x)
+            dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], ialpha);
+    }
+}
+
 void qt_memfill32_sse2(quint32 *dest, quint32 value, int count)
 {
     if (count < 7) {
diff --git a/tests/benchmarks/gui/image/blendbench/main.cpp b/tests/benchmarks/gui/image/blendbench/main.cpp
index 92d1633..f53654b 100644
--- a/tests/benchmarks/gui/image/blendbench/main.cpp
+++ b/tests/benchmarks/gui/image/blendbench/main.cpp
@@ -103,6 +103,9 @@ class BlendBench : public QObject
 private slots:		
     void blendBench_data();
     void blendBench();
+
+    void blendBenchAlpha_data();
+    void blendBenchAlpha();
 };
 
 void BlendBench::blendBench_data()
@@ -147,6 +150,35 @@ void BlendBench::blendBench()
     }
 }
 
+void BlendBench::blendBenchAlpha_data()
+{
+    blendBench_data();
+}
+
+void BlendBench::blendBenchAlpha()
+{
+    QFETCH(int, brushType);
+    QFETCH(int, compositionMode);
+
+    QImage img(512, 512, QImage::Format_ARGB32_Premultiplied);
+    QImage src(512, 512, QImage::Format_ARGB32_Premultiplied);
+    paint(&src);
+    QPainter p(&img);
+    p.setPen(Qt::NoPen);
+
+    p.setCompositionMode(QPainter::CompositionMode(compositionMode));
+    if (brushType == ImageBrush) {
+        p.setBrush(QBrush(src));
+    } else if (brushType == SolidBrush) {
+        p.setBrush(QColor(127, 127, 127, 127));
+    }
+    p.setOpacity(0.7f);
+
+    QBENCHMARK {
+        p.drawRect(0, 0, 512, 512);
+    }
+}
+
 QTEST_MAIN(BlendBench)
 
 #include "main.moc"
-- 
cgit v0.12