use templates instead of macros to build blitters

BUG=
R=mtklein@google.com

Author: reed@google.com

Review URL: https://codereview.chromium.org/120633004

git-svn-id: http://skia.googlecode.com/svn/trunk@14178 2bbb7eff-a529-9590-31e7-b0007b416f81

3 files changed, 211 insertions, 168 deletions

diff --git a/src/core/SkBitmapProcState_matrix.h b/src/core/SkBitmapProcState_matrix.h
index d796d0b04c..b71c3c3b75 100644
--- a/src/core/SkBitmapProcState_matrix.h
+++ b/src/core/SkBitmapProcState_matrix.h
@@ -9,11 +9,8 @@
 #include "SkMath.h"
 #include "SkMathPriv.h"
 
-#define SCALE_NOFILTER_NAME     MAKENAME(_nofilter_scale)
 #define SCALE_FILTER_NAME       MAKENAME(_filter_scale)
-#define AFFINE_NOFILTER_NAME    MAKENAME(_nofilter_affine)
 #define AFFINE_FILTER_NAME      MAKENAME(_filter_affine)
-#define PERSP_NOFILTER_NAME     MAKENAME(_nofilter_persp)
 #define PERSP_FILTER_NAME       MAKENAME(_filter_persp)
 
 #define PACK_FILTER_X_NAME  MAKENAME(_pack_filter_x)
@@ -28,13 +25,6 @@
 #endif
 
 // declare functions externally to suppress warnings.
-void SCALE_NOFILTER_NAME(const SkBitmapProcState& s,
-                                uint32_t xy[], int count, int x, int y);
-void AFFINE_NOFILTER_NAME(const SkBitmapProcState& s,
-                                 uint32_t xy[], int count, int x, int y);
-void PERSP_NOFILTER_NAME(const SkBitmapProcState& s,
-                                uint32_t* SK_RESTRICT xy,
-                                int count, int x, int y);
 void SCALE_FILTER_NAME(const SkBitmapProcState& s,
                               uint32_t xy[], int count, int x, int y);
 void AFFINE_FILTER_NAME(const SkBitmapProcState& s,
@@ -43,122 +33,6 @@ void PERSP_FILTER_NAME(const SkBitmapProcState& s,
                               uint32_t* SK_RESTRICT xy, int count,
                               int x, int y);
 
-void SCALE_NOFILTER_NAME(const SkBitmapProcState& s,
-                                uint32_t xy[], int count, int x, int y) {
-    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
-                             SkMatrix::kScale_Mask)) == 0);
-
-    PREAMBLE(s);
-    // we store y, x, x, x, x, x
-
-    const unsigned maxX = s.fBitmap->width() - 1;
-    SkFractionalInt fx;
-    {
-        SkPoint pt;
-        s.fInvProc(s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
-                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
-        fx = SkScalarToFractionalInt(pt.fY);
-        const unsigned maxY = s.fBitmap->height() - 1;
-        *xy++ = TILEY_PROCF(SkFractionalIntToFixed(fx), maxY);
-        fx = SkScalarToFractionalInt(pt.fX);
-    }
-
-    if (0 == maxX) {
-        // all of the following X values must be 0
-        memset(xy, 0, count * sizeof(uint16_t));
-        return;
-    }
-
-    const SkFractionalInt dx = s.fInvSxFractionalInt;
-
-#ifdef CHECK_FOR_DECAL
-    if (can_truncate_to_fixed_for_decal(fx, dx, count, maxX)) {
-        decal_nofilter_scale(xy, SkFractionalIntToFixed(fx),
-                             SkFractionalIntToFixed(dx), count);
-    } else
-#endif
-    {
-        int i;
-        for (i = (count >> 2); i > 0; --i) {
-            unsigned a, b;
-            a = TILEX_PROCF(SkFractionalIntToFixed(fx), maxX); fx += dx;
-            b = TILEX_PROCF(SkFractionalIntToFixed(fx), maxX); fx += dx;
-#ifdef SK_CPU_BENDIAN
-            *xy++ = (a << 16) | b;
-#else
-            *xy++ = (b << 16) | a;
-#endif
-            a = TILEX_PROCF(SkFractionalIntToFixed(fx), maxX); fx += dx;
-            b = TILEX_PROCF(SkFractionalIntToFixed(fx), maxX); fx += dx;
-#ifdef SK_CPU_BENDIAN
-            *xy++ = (a << 16) | b;
-#else
-            *xy++ = (b << 16) | a;
-#endif
-        }
-        uint16_t* xx = (uint16_t*)xy;
-        for (i = (count & 3); i > 0; --i) {
-            *xx++ = TILEX_PROCF(SkFractionalIntToFixed(fx), maxX); fx += dx;
-        }
-    }
-}
-
-// note: we could special-case on a matrix which is skewed in X but not Y.
-// this would require a more general setup thatn SCALE does, but could use
-// SCALE's inner loop that only looks at dx
-
-void AFFINE_NOFILTER_NAME(const SkBitmapProcState& s,
-                                 uint32_t xy[], int count, int x, int y) {
-    SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
-    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
-                             SkMatrix::kScale_Mask |
-                             SkMatrix::kAffine_Mask)) == 0);
-
-    PREAMBLE(s);
-    SkPoint srcPt;
-    s.fInvProc(s.fInvMatrix,
-               SkIntToScalar(x) + SK_ScalarHalf,
-               SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
-
-    SkFractionalInt fx = SkScalarToFractionalInt(srcPt.fX);
-    SkFractionalInt fy = SkScalarToFractionalInt(srcPt.fY);
-    SkFractionalInt dx = s.fInvSxFractionalInt;
-    SkFractionalInt dy = s.fInvKyFractionalInt;
-    int maxX = s.fBitmap->width() - 1;
-    int maxY = s.fBitmap->height() - 1;
-
-    for (int i = count; i > 0; --i) {
-        *xy++ = (TILEY_PROCF(SkFractionalIntToFixed(fy), maxY) << 16) |
-                 TILEX_PROCF(SkFractionalIntToFixed(fx), maxX);
-        fx += dx; fy += dy;
-    }
-}
-
-void PERSP_NOFILTER_NAME(const SkBitmapProcState& s,
-                                uint32_t* SK_RESTRICT xy,
-                                int count, int x, int y) {
-    SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
-
-    PREAMBLE(s);
-    int maxX = s.fBitmap->width() - 1;
-    int maxY = s.fBitmap->height() - 1;
-
-    SkPerspIter   iter(s.fInvMatrix,
-                       SkIntToScalar(x) + SK_ScalarHalf,
-                       SkIntToScalar(y) + SK_ScalarHalf, count);
-
-    while ((count = iter.next()) != 0) {
-        const SkFixed* SK_RESTRICT srcXY = iter.getXY();
-        while (--count >= 0) {
-            *xy++ = (TILEY_PROCF(srcXY[1], maxY) << 16) |
-                     TILEX_PROCF(srcXY[0], maxX);
-            srcXY += 2;
-        }
-    }
-}
-
-//////////////////////////////////////////////////////////////////////////////
-
 static inline uint32_t PACK_FILTER_Y_NAME(SkFixed f, unsigned max,
                                           SkFixed one PREAMBLE_PARAM_Y) {
     unsigned i = TILEY_PROCF(f, max);
@@ -270,15 +144,6 @@ void PERSP_FILTER_NAME(const SkBitmapProcState& s,
     }
 }
 
-static SkBitmapProcState::MatrixProc MAKENAME(_Procs)[] = {
-    SCALE_NOFILTER_NAME,
-    SCALE_FILTER_NAME,
-    AFFINE_NOFILTER_NAME,
-    AFFINE_FILTER_NAME,
-    PERSP_NOFILTER_NAME,
-    PERSP_FILTER_NAME
-};
-
 #undef MAKENAME
 #undef TILEX_PROCF
 #undef TILEY_PROCF
@@ -286,11 +151,8 @@ static SkBitmapProcState::MatrixProc MAKENAME(_Procs)[] = {
     #undef CHECK_FOR_DECAL
 #endif
 
-#undef SCALE_NOFILTER_NAME
 #undef SCALE_FILTER_NAME
-#undef AFFINE_NOFILTER_NAME
 #undef AFFINE_FILTER_NAME
-#undef PERSP_NOFILTER_NAME
 #undef PERSP_FILTER_NAME
 
 #undef PREAMBLE
diff --git a/src/core/SkBitmapProcState_matrixProcs.cpp b/src/core/SkBitmapProcState_matrixProcs.cpp
index 57376d5699..904e9488e7 100644
--- a/src/core/SkBitmapProcState_matrixProcs.cpp
+++ b/src/core/SkBitmapProcState_matrixProcs.cpp
@@ -33,6 +33,10 @@ static inline int sk_int_mod(int x, int n) {
 void decal_nofilter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count);
 void decal_filter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count);
 
+#include "SkBitmapProcState_matrix_template.h"
+
+///////////////////////////////////////////////////////////////////////////////
+
 // Compile neon code paths if needed
 #if !SK_ARM_NEON_IS_NONE
 
@@ -52,6 +56,35 @@ extern const SkBitmapProcState::MatrixProc RepeatX_RepeatY_Procs_neon[];
 #define CHECK_FOR_DECAL
 #include "SkBitmapProcState_matrix.h"
 
+struct ClampTileProcs {
+    static unsigned X(const SkBitmapProcState&, SkFixed fx, int max) {
+        return SkClampMax(fx >> 16, max);
+    }
+    static unsigned Y(const SkBitmapProcState&, SkFixed fy, int max) {
+        return SkClampMax(fy >> 16, max);
+    }
+};
+
+// Referenced in opts_check_SSE2.cpp
+void ClampX_ClampY_nofilter_scale(const SkBitmapProcState& s, uint32_t xy[],
+                                  int count, int x, int y) {
+    return NoFilterProc_Scale<ClampTileProcs, true>(s, xy, count, x, y);
+}
+void ClampX_ClampY_nofilter_affine(const SkBitmapProcState& s, uint32_t xy[],
+                                  int count, int x, int y) {
+    return NoFilterProc_Affine<ClampTileProcs>(s, xy, count, x, y);
+}
+
+static SkBitmapProcState::MatrixProc ClampX_ClampY_Procs[] = {
+    // only clamp lives in the right coord space to check for decal
+    ClampX_ClampY_nofilter_scale,
+    ClampX_ClampY_filter_scale,
+    ClampX_ClampY_nofilter_affine,
+    ClampX_ClampY_filter_affine,
+    NoFilterProc_Persp<ClampTileProcs>,
+    ClampX_ClampY_filter_persp
+};
+
 #define MAKENAME(suffix)        RepeatX_RepeatY ## suffix
 #define TILEX_PROCF(fx, max)    SK_USHIFT16(((fx) & 0xFFFF) * ((max) + 1))
 #define TILEY_PROCF(fy, max)    SK_USHIFT16(((fy) & 0xFFFF) * ((max) + 1))
@@ -60,6 +93,24 @@ extern const SkBitmapProcState::MatrixProc RepeatX_RepeatY_Procs_neon[];
 #include "SkBitmapProcState_matrix.h"
 #endif
 
+struct RepeatTileProcs {
+    static unsigned X(const SkBitmapProcState&, SkFixed fx, int max) {
+        return SK_USHIFT16(((fx) & 0xFFFF) * ((max) + 1));
+    }
+    static unsigned Y(const SkBitmapProcState&, SkFixed fy, int max) {
+        return SK_USHIFT16(((fy) & 0xFFFF) * ((max) + 1));
+    }
+};
+
+static SkBitmapProcState::MatrixProc RepeatX_RepeatY_Procs[] = {
+    NoFilterProc_Scale<RepeatTileProcs, false>,
+    RepeatX_RepeatY_filter_scale,
+    NoFilterProc_Affine<RepeatTileProcs>,
+    RepeatX_RepeatY_filter_affine,
+    NoFilterProc_Persp<RepeatTileProcs>,
+    RepeatX_RepeatY_filter_persp
+};
+
 #define MAKENAME(suffix)        GeneralXY ## suffix
 #define PREAMBLE(state)         SkBitmapProcState::FixedTileProc tileProcX = (state).fTileProcX; (void) tileProcX; \
                                 SkBitmapProcState::FixedTileProc tileProcY = (state).fTileProcY; (void) tileProcY; \
@@ -75,8 +126,27 @@ extern const SkBitmapProcState::MatrixProc RepeatX_RepeatY_Procs_neon[];
 #define TILEY_LOW_BITS(fy, max) tileLowBitsProcY(fy, (max) + 1)
 #include "SkBitmapProcState_matrix.h"
 
-static inline U16CPU fixed_clamp(SkFixed x)
-{
+struct GeneralTileProcs {
+    static unsigned X(const SkBitmapProcState& s, SkFixed fx, int max) {
+        return SK_USHIFT16(s.fTileProcX(fx) * ((max) + 1));
+    }
+    static unsigned Y(const SkBitmapProcState& s, SkFixed fy, int max) {
+        return SK_USHIFT16(s.fTileProcY(fy) * ((max) + 1));
+    }
+};
+
+static SkBitmapProcState::MatrixProc GeneralXY_Procs[] = {
+    NoFilterProc_Scale<GeneralTileProcs, false>,
+    GeneralXY_filter_scale,
+    NoFilterProc_Affine<GeneralTileProcs>,
+    GeneralXY_filter_affine,
+    NoFilterProc_Persp<GeneralTileProcs>,
+    GeneralXY_filter_persp
+};
+
+///////////////////////////////////////////////////////////////////////////////
+
+static inline U16CPU fixed_clamp(SkFixed x) {
     if (x < 0) {
         x = 0;
     }
@@ -86,8 +156,7 @@ static inline U16CPU fixed_clamp(SkFixed x)
     return x;
 }
 
-static inline U16CPU fixed_repeat(SkFixed x)
-{
+static inline U16CPU fixed_repeat(SkFixed x) {
     return x & 0xFFFF;
 }
 
@@ -97,8 +166,7 @@ static inline U16CPU fixed_repeat(SkFixed x)
 #pragma optimize("", off)
 #endif
 
-static inline U16CPU fixed_mirror(SkFixed x)
-{
+static inline U16CPU fixed_mirror(SkFixed x) {
     SkFixed s = x << 15 >> 31;
     // s is FFFFFFFF if we're on an odd interval, or 0 if an even interval
     return (x ^ s) & 0xFFFF;
@@ -108,12 +176,13 @@ static inline U16CPU fixed_mirror(SkFixed x)
 #pragma optimize("", on)
 #endif
 
-static SkBitmapProcState::FixedTileProc choose_tile_proc(unsigned m)
-{
-    if (SkShader::kClamp_TileMode == m)
+static SkBitmapProcState::FixedTileProc choose_tile_proc(unsigned m) {
+    if (SkShader::kClamp_TileMode == m) {
         return fixed_clamp;
-    if (SkShader::kRepeat_TileMode == m)
+    }
+    if (SkShader::kRepeat_TileMode == m) {
         return fixed_repeat;
+    }
     SkASSERT(SkShader::kMirror_TileMode == m);
     return fixed_mirror;
 }
@@ -178,12 +247,10 @@ static SkBitmapProcState::IntTileProc choose_int_tile_proc(unsigned tm) {
 
 //////////////////////////////////////////////////////////////////////////////
 
-void decal_nofilter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count)
-{
+void decal_nofilter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count) {
     int i;
 
-    for (i = (count >> 2); i > 0; --i)
-    {
+    for (i = (count >> 2); i > 0; --i) {
         *dst++ = pack_two_shorts(fx >> 16, (fx + dx) >> 16);
         fx += dx+dx;
         *dst++ = pack_two_shorts(fx >> 16, (fx + dx) >> 16);
@@ -197,18 +264,13 @@ void decal_nofilter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count)
     }
 }
 
-void decal_filter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count)
-{
-
-
-    if (count & 1)
-    {
+void decal_filter_scale(uint32_t dst[], SkFixed fx, SkFixed dx, int count) {
+    if (count & 1) {
         SkASSERT((fx >> (16 + 14)) == 0);
         *dst++ = (fx >> 12 << 14) | ((fx >> 16) + 1);
         fx += dx;
     }
-    while ((count -= 2) >= 0)
-    {
+    while ((count -= 2) >= 0) {
         SkASSERT((fx >> (16 + 14)) == 0);
         *dst++ = (fx >> 12 << 14) | ((fx >> 16) + 1);
         fx += dx;
@@ -412,8 +474,7 @@ static void mirrorx_nofilter_trans(const SkBitmapProcState& s,
 
 ///////////////////////////////////////////////////////////////////////////////
 
-SkBitmapProcState::MatrixProc
-SkBitmapProcState::chooseMatrixProc(bool trivial_matrix) {
+SkBitmapProcState::MatrixProc SkBitmapProcState::chooseMatrixProc(bool trivial_matrix) {
 //    test_int_tileprocs();
     // check for our special case when there is no scale/affine/perspective
     if (trivial_matrix) {
@@ -439,9 +500,7 @@ SkBitmapProcState::chooseMatrixProc(bool trivial_matrix) {
         index += 2;
     }
 
-    if (SkShader::kClamp_TileMode == fTileModeX &&
-        SkShader::kClamp_TileMode == fTileModeY)
-    {
+    if (SkShader::kClamp_TileMode == fTileModeX && SkShader::kClamp_TileMode == fTileModeY) {
         // clamp gets special version of filterOne
         fFilterOneX = SK_Fixed1;
         fFilterOneY = SK_Fixed1;
@@ -452,9 +511,7 @@ SkBitmapProcState::chooseMatrixProc(bool trivial_matrix) {
     fFilterOneX = SK_Fixed1 / fBitmap->width();
     fFilterOneY = SK_Fixed1 / fBitmap->height();
 
-    if (SkShader::kRepeat_TileMode == fTileModeX &&
-        SkShader::kRepeat_TileMode == fTileModeY)
-    {
+    if (SkShader::kRepeat_TileMode == fTileModeX && SkShader::kRepeat_TileMode == fTileModeY) {
         return SK_ARM_NEON_WRAP(RepeatX_RepeatY_Procs)[index];
     }
 
diff --git a/src/core/SkBitmapProcState_matrix_template.h b/src/core/SkBitmapProcState_matrix_template.h
new file mode 100644
index 0000000000..4b0044b46a
--- /dev/null
+++ b/src/core/SkBitmapProcState_matrix_template.h
@@ -0,0 +1,124 @@
+/*
+ * Copyright 2013 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#ifndef SkBitmapProcState_MatrixTemplates_DEFINED
+#define SkBitmapProcState_MatrixTemplates_DEFINED
+
+#include "SkMath.h"
+#include "SkMathPriv.h"
+
+template <typename TileProc, bool tryDecal>
+void NoFilterProc_Scale(const SkBitmapProcState& s, uint32_t xy[],
+                        int count, int x, int y) {
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask)) == 0);
+
+    // we store y, x, x, x, x, x
+
+    const unsigned maxX = s.fBitmap->width() - 1;
+    SkFractionalInt fx;
+    {
+        SkPoint pt;
+        s.fInvProc(s.fInvMatrix, SkIntToScalar(x) + SK_ScalarHalf,
+                                  SkIntToScalar(y) + SK_ScalarHalf, &pt);
+        fx = SkScalarToFractionalInt(pt.fY);
+        const unsigned maxY = s.fBitmap->height() - 1;
+        *xy++ = TileProc::Y(s, SkFractionalIntToFixed(fx), maxY);
+        fx = SkScalarToFractionalInt(pt.fX);
+    }
+
+    if (0 == maxX) {
+        // all of the following X values must be 0
+        memset(xy, 0, count * sizeof(uint16_t));
+        return;
+    }
+
+    const SkFractionalInt dx = s.fInvSxFractionalInt;
+
+    if (tryDecal && can_truncate_to_fixed_for_decal(fx, dx, count, maxX)) {
+        decal_nofilter_scale(xy, SkFractionalIntToFixed(fx),
+                             SkFractionalIntToFixed(dx), count);
+    } else {
+        int i;
+        for (i = (count >> 2); i > 0; --i) {
+            unsigned a, b;
+            a = TileProc::X(s, SkFractionalIntToFixed(fx), maxX); fx += dx;
+            b = TileProc::X(s, SkFractionalIntToFixed(fx), maxX); fx += dx;
+#ifdef SK_CPU_BENDIAN
+            *xy++ = (a << 16) | b;
+#else
+            *xy++ = (b << 16) | a;
+#endif
+            a = TileProc::X(s, SkFractionalIntToFixed(fx), maxX); fx += dx;
+            b = TileProc::X(s, SkFractionalIntToFixed(fx), maxX); fx += dx;
+#ifdef SK_CPU_BENDIAN
+            *xy++ = (a << 16) | b;
+#else
+            *xy++ = (b << 16) | a;
+#endif
+        }
+        uint16_t* xx = (uint16_t*)xy;
+        for (i = (count & 3); i > 0; --i) {
+            *xx++ = TileProc::X(s, SkFractionalIntToFixed(fx), maxX); fx += dx;
+        }
+    }
+}
+
+// note: we could special-case on a matrix which is skewed in X but not Y.
+// this would require a more general setup thatn SCALE does, but could use
+// SCALE's inner loop that only looks at dx
+
+template <typename TileProc>
+void NoFilterProc_Affine(const SkBitmapProcState& s, uint32_t xy[],
+                         int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kAffine_Mask);
+    SkASSERT((s.fInvType & ~(SkMatrix::kTranslate_Mask |
+                             SkMatrix::kScale_Mask |
+                             SkMatrix::kAffine_Mask)) == 0);
+
+    SkPoint srcPt;
+    s.fInvProc(s.fInvMatrix,
+               SkIntToScalar(x) + SK_ScalarHalf,
+               SkIntToScalar(y) + SK_ScalarHalf, &srcPt);
+
+    SkFractionalInt fx = SkScalarToFractionalInt(srcPt.fX);
+    SkFractionalInt fy = SkScalarToFractionalInt(srcPt.fY);
+    SkFractionalInt dx = s.fInvSxFractionalInt;
+    SkFractionalInt dy = s.fInvKyFractionalInt;
+    int maxX = s.fBitmap->width() - 1;
+    int maxY = s.fBitmap->height() - 1;
+
+    for (int i = count; i > 0; --i) {
+        *xy++ = (TileProc::Y(s, SkFractionalIntToFixed(fy), maxY) << 16) |
+                 TileProc::X(s, SkFractionalIntToFixed(fx), maxX);
+        fx += dx; fy += dy;
+    }
+}
+
+template <typename TileProc>
+void NoFilterProc_Persp(const SkBitmapProcState& s, uint32_t* SK_RESTRICT xy,
+                        int count, int x, int y) {
+    SkASSERT(s.fInvType & SkMatrix::kPerspective_Mask);
+
+    int maxX = s.fBitmap->width() - 1;
+    int maxY = s.fBitmap->height() - 1;
+
+    SkPerspIter   iter(s.fInvMatrix,
+                       SkIntToScalar(x) + SK_ScalarHalf,
+                       SkIntToScalar(y) + SK_ScalarHalf, count);
+
+    while ((count = iter.next()) != 0) {
+        const SkFixed* SK_RESTRICT srcXY = iter.getXY();
+        while (--count >= 0) {
+            *xy++ = (TileProc::Y(s, srcXY[1], maxY) << 16) |
+                     TileProc::X(s, srcXY[0], maxX);
+            srcXY += 2;
+        }
+    }
+}
+
+#endif