Graduate matrix map-point procs out of SkOpts.

src/core/SkMatrix.cpp

 /*
  * Copyright 2006 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
+#include "SkFloatBits.h"
 #include "SkMatrix.h"
-#include "SkFloatBits.h"
+#include "SkNx.h"
+#include "SkPaint.h"
 #include "SkRSXform.h"
 #include "SkString.h"
-#include "SkNx.h"
-#include "SkOpts.h"
-
 #include <stddef.h>
 
 static void normalize_perspective(SkScalar mat[9]) {
     // If it was interesting to never store the last element, we could divide all 8 other
     // elements here by the 9th, making it 1.0...
     //
     // When SkScalar was SkFixed, we would sometimes rescale the entire matrix to keep its
     // component values from getting too large. This is not a concern when using floats/doubles,
     // so we do nothing now.
 
@@ skipping 885 matching lines @@
 
 void SkMatrix::Identity_pts(const SkMatrix& m, SkPoint dst[], const SkPoint src[], int count) {
     SkASSERT(m.getType() == 0);
 
     if (dst != src && count > 0) {
         memcpy(dst, src, count * sizeof(SkPoint));
     }
 }
 
 void SkMatrix::Trans_pts(const SkMatrix& m, SkPoint dst[], const SkPoint src[], int count) {
-    return SkOpts::matrix_translate(m,dst,src,count);
+    SkASSERT(m.getType() <= SkMatrix::kTranslate_Mask);
+    if (count > 0) {
+        SkScalar tx = m.getTranslateX();
+        SkScalar ty = m.getTranslateY();
+        if (count & 1) {
+            dst->fX = src->fX + tx;
+            dst->fY = src->fY + ty;
+            src += 1;
+            dst += 1;
+        }
+        Sk4s trans4(tx, ty, tx, ty);
+        count >>= 1;
+        if (count & 1) {
+            (Sk4s::Load(src) + trans4).store(dst);
+            src += 2;
+            dst += 2;
+        }
+        count >>= 1;
+        for (int i = 0; i < count; ++i) {
+            (Sk4s::Load(src+0) + trans4).store(dst+0);
+            (Sk4s::Load(src+2) + trans4).store(dst+2);
+            src += 4;
+            dst += 4;
+        }
+    }
 }
 
 void SkMatrix::Scale_pts(const SkMatrix& m, SkPoint dst[], const SkPoint src[], int count) {
-    return SkOpts::matrix_scale_translate(m,dst,src,count);
+    SkASSERT(m.getType() <= (SkMatrix::kScale_Mask | SkMatrix::kTranslate_Mask));
+    if (count > 0) {
+        SkScalar tx = m.getTranslateX();
+        SkScalar ty = m.getTranslateY();
+        SkScalar sx = m.getScaleX();
+        SkScalar sy = m.getScaleY();
+        if (count & 1) {
+            dst->fX = src->fX * sx + tx;
+            dst->fY = src->fY * sy + ty;
+            src += 1;
+            dst += 1;
+        }
+        Sk4s trans4(tx, ty, tx, ty);
+        Sk4s scale4(sx, sy, sx, sy);
+        count >>= 1;
+        if (count & 1) {
+            (Sk4s::Load(src) * scale4 + trans4).store(dst);
+            src += 2;
+            dst += 2;
+        }
+        count >>= 1;
+        for (int i = 0; i < count; ++i) {
+            (Sk4s::Load(src+0) * scale4 + trans4).store(dst+0);
+            (Sk4s::Load(src+2) * scale4 + trans4).store(dst+2);
+            src += 4;
+            dst += 4;
+        }
+    }
 }
 
 void SkMatrix::Persp_pts(const SkMatrix& m, SkPoint dst[],
                          const SkPoint src[], int count) {
     SkASSERT(m.hasPerspective());
 
     if (count > 0) {
         do {
             SkScalar sy = src->fY;
             SkScalar sx = src->fX;
@@ skipping 11 matching lines @@
             }
 
             dst->fY = y * z;
             dst->fX = x * z;
             dst += 1;
         } while (--count);
     }
 }
 
 void SkMatrix::Affine_vpts(const SkMatrix& m, SkPoint dst[], const SkPoint src[], int count) {
-    return SkOpts::matrix_affine(m,dst,src,count);
+    SkASSERT(m.getType() != SkMatrix::kPerspective_Mask);
+    if (count > 0) {
+        SkScalar tx = m.getTranslateX();
+        SkScalar ty = m.getTranslateY();
+        SkScalar sx = m.getScaleX();
+        SkScalar sy = m.getScaleY();
+        SkScalar kx = m.getSkewX();
+        SkScalar ky = m.getSkewY();
+        if (count & 1) {
+            dst->set(src->fX * sx + src->fY * kx + tx,
+                     src->fX * ky + src->fY * sy + ty);
+            src += 1;
+            dst += 1;
+        }
+        Sk4s trans4(tx, ty, tx, ty);
+        Sk4s scale4(sx, sy, sx, sy);
+        Sk4s  skew4(kx, ky, kx, ky);    // applied to swizzle of src4
+        count >>= 1;
+        for (int i = 0; i < count; ++i) {
+            Sk4s src4 = Sk4s::Load(src);
+            Sk4s swz4 = SkNx_shuffle<1,0,3,2>(src4);  // y0 x0, y1 x1
+            (src4 * scale4 + swz4 * skew4 + trans4).store(dst);
+            src += 2;
+            dst += 2;
+        }
+    }
 }
 
 const SkMatrix::MapPtsProc SkMatrix::gMapPtsProcs[] = {
     SkMatrix::Identity_pts, SkMatrix::Trans_pts,
     SkMatrix::Scale_pts,   SkMatrix::Scale_pts,
     SkMatrix::Affine_vpts,  SkMatrix::Affine_vpts,
     SkMatrix::Affine_vpts,  SkMatrix::Affine_vpts,
     // repeat the persp proc 8 times
     SkMatrix::Persp_pts,    SkMatrix::Persp_pts,
     SkMatrix::Persp_pts,    SkMatrix::Persp_pts,
@@ skipping 815 matching lines @@
     const SkScalar m10 = -m01;
     const SkScalar m11 = m00;
     const SkScalar m12 = fTy;
 
     quad[0].set(m02, m12);
     quad[1].set(m00 * width + m02, m10 * width + m12);
     quad[2].set(m00 * width + m01 * height + m02, m10 * width + m11 * height + m12);
     quad[3].set(m01 * height + m02, m11 * height + m12);
 #endif
 }