4f linear gradient shader blitters

src/effects/gradients/Sk4fLinearGradient.cpp

 /*
  * Copyright 2016 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "Sk4fGradientPriv.h"
 #include "Sk4fLinearGradient.h"
+#include "SkXfermode.h"
 
 namespace {
 
-template<typename DstType, bool do_premul>
+template<typename DstType, SkColorProfileType, bool do_premul>
 void fill(const Sk4f& c, DstType* dst, int n);
 
 template<>
-void fill<SkPM4f, false>(const Sk4f& c, SkPM4f* dst, int n) {
+void fill<SkPM4f, kLinear_SkColorProfileType, false>(const Sk4f& c, SkPM4f* dst, int n) {
     while (n > 0) {
         c.store(dst++);
         n--;
     }
 }
 
 template<>
-void fill<SkPM4f, true>(const Sk4f& c, SkPM4f* dst, int n) {
-    fill<SkPM4f, false>(premul_4f(c), dst, n);
+void fill<SkPM4f, kLinear_SkColorProfileType, true>(const Sk4f& c, SkPM4f* dst, int n) {
+    fill<SkPM4f, kLinear_SkColorProfileType, false>(premul_4f(c), dst, n);
 }
 
 template<>
-void fill<SkPMColor, false>(const Sk4f& c, SkPMColor* dst, int n) {
+void fill<SkPMColor, kLinear_SkColorProfileType, false>(const Sk4f& c, SkPMColor* dst, int n) {
     sk_memset32(dst, trunc_from_255<false>(c), n);
 }
 
 template<>
-void fill<SkPMColor, true>(const Sk4f& c, SkPMColor* dst, int n) {
+void fill<SkPMColor, kLinear_SkColorProfileType, true>(const Sk4f& c, SkPMColor* dst, int n) {
     sk_memset32(dst, trunc_from_255<true>(c), n);
 }
 
-template<typename DstType, bool do_premul>
+template<>
+void fill<SkPMColor, kSRGB_SkColorProfileType, false>(const Sk4f& c, SkPMColor* dst, int n) {
+    // FIXME: this assumes opaque colors.  Handle unpremultiplication.
+    const SkPMColor pmc = Sk4f_toS32(c);
+    sk_memset32(dst, pmc, n);
+}
+
+template<>
+void fill<SkPMColor, kSRGB_SkColorProfileType, true>(const Sk4f& c, SkPMColor* dst, int n) {
+    const SkPMColor s32 = Sk4f_toS32(c);
+    const SkPMColor pmc = SkPreMultiplyARGB(SkGetPackedA32(s32), SkGetPackedR32(s32),
+                                            SkGetPackedG32(s32), SkGetPackedB32(s32));
+    sk_memset32(dst, pmc, n);
+}
+
+template<>
+void fill<uint64_t, kLinear_SkColorProfileType, false>(const Sk4f& c, uint64_t* dst, int n) {
+    const uint64_t c64 = SkFloatToHalf_01(c);
+    while (n > 0) {

[reed1, 2016/03/18 14:28:59]

sk_memset64() ?

[f(malita), 2016/03/18 16:54:11]

added/done.

+        *dst++ = c64;
+        n--;
+    }
+}
+
+template<>
+void fill<uint64_t, kLinear_SkColorProfileType, true>(const Sk4f& c, uint64_t* dst, int n) {
+    fill<uint64_t, kLinear_SkColorProfileType, false>(premul_4f(c), dst, n);
+}
+
+template<typename DstType, SkColorProfileType profile, bool do_premul>
 void ramp(const Sk4f& c, const Sk4f& dc, DstType* dst, int n) {
     SkASSERT(n > 0);
 
     const Sk4f dc2 = dc + dc;
     const Sk4f dc4 = dc2 + dc2;
 
     Sk4f c0 = c ;
     Sk4f c1 = c + dc;
     Sk4f c2 = c0 + dc2;
     Sk4f c3 = c1 + dc2;
 
     while (n >= 4) {
-        store4x<DstType, do_premul>(c0, c1, c2, c3, dst);
+        store4x<DstType, profile, do_premul>(c0, c1, c2, c3, dst);
         dst += 4;
 
         c0 = c0 + dc4;
         c1 = c1 + dc4;
         c2 = c2 + dc4;
         c3 = c3 + dc4;
         n -= 4;
     }
     if (n & 2) {
-        store<DstType, do_premul>(c0, dst++);
-        store<DstType, do_premul>(c1, dst++);
+        store<DstType, profile, do_premul>(c0, dst++);
+        store<DstType, profile, do_premul>(c1, dst++);
         c0 = c0 + dc2;
     }
     if (n & 1) {
-        store<DstType, do_premul>(c0, dst);
+        store<DstType, profile, do_premul>(c0, dst);
     }
 }
 
 template<SkShader::TileMode>
 SkScalar pinFx(SkScalar);
 
 template<>
 SkScalar pinFx<SkShader::kClamp_TileMode>(SkScalar fx) {
     return fx;
 }
@@ skipping 78 matching lines @@
 void SkLinearGradient::
 LinearGradient4fContext::shadeSpan(int x, int y, SkPMColor dst[], int count) {
     if (!this->isFast()) {
         this->INHERITED::shadeSpan(x, y, dst, count);
         return;
     }
 
     // TODO: plumb dithering
     SkASSERT(count > 0);
     if (fColorsArePremul) {
-        this->shadePremulSpan<SkPMColor, false>(x, y, dst, count);
+        this->shadePremulSpan<SkPMColor, kLinear_SkColorProfileType, false>(x, y, dst, count);
     } else {
-        this->shadePremulSpan<SkPMColor, true>(x, y, dst, count);
+        this->shadePremulSpan<SkPMColor, kLinear_SkColorProfileType, true>(x, y, dst, count);
     }
 }
 
 void SkLinearGradient::
 LinearGradient4fContext::shadeSpan4f(int x, int y, SkPM4f dst[], int count) {
     if (!this->isFast()) {
         this->INHERITED::shadeSpan4f(x, y, dst, count);
         return;
     }
 
     // TONOTDO: plumb dithering
     SkASSERT(count > 0);
     if (fColorsArePremul) {
-        this->shadePremulSpan<SkPM4f, false>(x, y, dst, count);
+        this->shadePremulSpan<SkPM4f, kLinear_SkColorProfileType, false>(x, y, dst, count);
     } else {
-        this->shadePremulSpan<SkPM4f, true>(x, y, dst, count);
+        this->shadePremulSpan<SkPM4f, kLinear_SkColorProfileType, true>(x, y, dst, count);
     }
 }
 
-template<typename DstType, bool do_premul>
+template<typename DstType, SkColorProfileType profile, bool do_premul>
 void SkLinearGradient::
 LinearGradient4fContext::shadePremulSpan(int x, int y,
                                          DstType dst[],
                                          int count) const {
     const SkLinearGradient& shader =
         static_cast<const SkLinearGradient&>(fShader);
     switch (shader.fTileMode) {
     case kClamp_TileMode:
         this->shadeSpanInternal<DstType,
+                                profile,
                                 do_premul,
                                 kClamp_TileMode>(x, y, dst, count);
         break;
     case kRepeat_TileMode:
         this->shadeSpanInternal<DstType,
+                                profile,
                                 do_premul,
                                 kRepeat_TileMode>(x, y, dst, count);
         break;
     case kMirror_TileMode:
         this->shadeSpanInternal<DstType,
+                                profile,
                                 do_premul,
                                 kMirror_TileMode>(x, y, dst, count);
         break;
     }
 }
 
-template<typename DstType, bool do_premul, SkShader::TileMode tileMode>
+template<typename DstType, SkColorProfileType profile, bool do_premul,
+         SkShader::TileMode tileMode>
 void SkLinearGradient::
 LinearGradient4fContext::shadeSpanInternal(int x, int y,
                                            DstType dst[],
                                            int count) const {
     SkPoint pt;
     fDstToPosProc(fDstToPos,
                   x + SK_ScalarHalf,
                   y + SK_ScalarHalf,
                   &pt);
     const SkScalar fx = pinFx<tileMode>(pt.x());
     const SkScalar dx = fDstToPos.getScaleX();
-    LinearIntervalProcessor<DstType, tileMode> proc(fIntervals.begin(),
-                                                    fIntervals.end() - 1,
-                                                    this->findInterval(fx),
-                                                    fx,
-                                                    dx,
-                                                    SkScalarNearlyZero(dx * count));
+    LinearIntervalProcessor<DstType, profile, tileMode> proc(fIntervals.begin(),
+                                                             fIntervals.end() - 1,
+                                                             this->findInterval(fx),
+                                                             fx,
+                                                             dx,
+                                                             SkScalarNearlyZero(dx * count));
     while (count > 0) {
         // What we really want here is SkTPin(advance, 1, count)
         // but that's a significant perf hit for >> stops; investigate.
         const int n = SkScalarTruncToInt(
             SkTMin<SkScalar>(proc.currentAdvance() + 1, SkIntToScalar(count)));
 
         // The current interval advance can be +inf (e.g. when reaching
         // the clamp mode end intervals) - when that happens, we expect to
         //   a) consume all remaining count in one swoop
         //   b) return a zero color gradient
         SkASSERT(SkScalarIsFinite(proc.currentAdvance())
             || (n == count && proc.currentRampIsZero()));
 
         if (proc.currentRampIsZero()) {
-            fill<DstType, do_premul>(proc.currentColor(),
-                                     dst, n);
+            fill<DstType, profile, do_premul>(proc.currentColor(),
+                                              dst, n);
         } else {
-            ramp<DstType, do_premul>(proc.currentColor(),
-                                     proc.currentColorGrad(),
-                                     dst, n);
+            ramp<DstType, profile, do_premul>(proc.currentColor(),
+                                              proc.currentColorGrad(),
+                                              dst, n);
         }
 
         proc.advance(SkIntToScalar(n));
         count -= n;
         dst   += n;
     }
 }
 
-template<typename DstType, SkShader::TileMode tileMode>
+template<typename DstType, SkColorProfileType profile, SkShader::TileMode tileMode>
 class SkLinearGradient::
 LinearGradient4fContext::LinearIntervalProcessor {
 public:
     LinearIntervalProcessor(const Interval* firstInterval,
                             const Interval* lastInterval,
                             const Interval* i,
                             SkScalar fx,
                             SkScalar dx,
                             bool is_vertical)
-        : fDstComponentScale(dst_component_scale<DstType>())
-        , fAdvX((i->fP1 - fx) / dx)
+        : fAdvX((i->fP1 - fx) / dx)
         , fFirstInterval(firstInterval)
         , fLastInterval(lastInterval)
         , fInterval(i)
         , fDx(dx)
         , fIsVertical(is_vertical)
     {
         SkASSERT(firstInterval <= lastInterval);
         SkASSERT(in_range(fx, i->fP0, i->fP1));
         this->compute_interval_props(fx - i->fP0);
     }
@@ skipping 19 matching lines @@
 
         fCc = fCc + fDcDx * Sk4f(advX);
         fAdvX -= advX;
     }
 
 private:
     void compute_interval_props(SkScalar t) {
         fDc   = dst_swizzle<DstType>(fInterval->fDc);
         fCc   = dst_swizzle<DstType>(fInterval->fC0);
         fCc   = fCc + fDc * Sk4f(t);
-        fCc   = fCc * fDstComponentScale;
-        fDcDx = fDc * fDstComponentScale * Sk4f(fDx);
+        fCc   = scale_for_dest<DstType, profile>(fCc);
+        fDcDx = scale_for_dest<DstType, profile>(fDc * Sk4f(fDx));
         fZeroRamp = fIsVertical || fInterval->isZeroRamp();
     }
 
     const Interval* next_interval(const Interval* i) const {
         SkASSERT(i >= fFirstInterval);
         SkASSERT(i <= fLastInterval);
         i++;
 
         if (tileMode == kClamp_TileMode) {
             SkASSERT(i <= fLastInterval);
@@ skipping 12 matching lines @@
             fAdvX = (fInterval->fP1 - fInterval->fP0) / fDx;
             SkASSERT(fAdvX > 0);
         } while (advX >= fAdvX);
 
         compute_interval_props(0);
 
         SkASSERT(advX >= 0);
         return advX;
     }
 
-    const Sk4f      fDstComponentScale; // cached dst scale (PMC: 255, PM4f: 1)
-
     // Current interval properties.
     Sk4f            fDc;        // local color gradient (dc/dt)
     Sk4f            fDcDx;      // dst color gradient (dc/dx)
     Sk4f            fCc;        // current color, interpolated in dst
     SkScalar        fAdvX;      // remaining interval advance in dst
     bool            fZeroRamp;  // current interval color grad is 0
 
     const Interval* fFirstInterval;
     const Interval* fLastInterval;
     const Interval* fInterval;  // current interval
@@ skipping 38 matching lines @@
             *ts++ = t4[0];
         }
     } else {
         for (int i = 0; i < count; ++i) {
             fDstToPosProc(fDstToPos, sx, sy, &pt);
             ts[i] = pt.x();
             sx += SK_Scalar1;
         }
     }
 }
+
+SkShader::Context::BlitProc SkLinearGradient::
+LinearGradient4fContext::onChooseBlitProc(const SkImageInfo& info, BlitState* state) {
+    SkXfermode::Mode mode;
+    if (!SkXfermode::AsMode(state->fXfer, &mode)) {
+        return nullptr;
+    }
+
+    const SkGradientShaderBase& shader = static_cast<const SkGradientShaderBase&>(fShader);
+    if (mode != SkXfermode::kSrc_Mode &&
+        !(mode == SkXfermode::kSrcOver_Mode && shader.colorsAreOpaque())) {
+        return nullptr;
+    }
+
+    switch (info.colorType()) {
+        case kN32_SkColorType:
+            return D32_BlitProc;
+        case kRGBA_F16_SkColorType:
+            return D64_BlitProc;
+        default:
+            return nullptr;
+    }
+}
+
+void SkLinearGradient::
+LinearGradient4fContext::D32_BlitProc(BlitState* state, int x, int y, const SkPixmap& dst,
+                                      int count, const SkAlpha aa[]) {
+    // FIXME: ignoring coverage for now
+    const LinearGradient4fContext* ctx =
+        static_cast<const LinearGradient4fContext*>(state->fCtx);
+
+    if (dst.info().isLinear()) {
+        if (ctx->fColorsArePremul) {
+            ctx->shadePremulSpan<SkPMColor, kLinear_SkColorProfileType, false>(
+                x, y, dst.writable_addr32(x, y), count);
+        } else {
+            ctx->shadePremulSpan<SkPMColor, kLinear_SkColorProfileType, true>(
+                x, y, dst.writable_addr32(x, y), count);
+        }
+    } else {
+        if (ctx->fColorsArePremul) {
+            ctx->shadePremulSpan<SkPMColor, kSRGB_SkColorProfileType, false>(
+                x, y, dst.writable_addr32(x, y), count);
+        } else {
+            ctx->shadePremulSpan<SkPMColor, kSRGB_SkColorProfileType, true>(
+                x, y, dst.writable_addr32(x, y), count);
+        }
+    }
+}
+
+void SkLinearGradient::
+LinearGradient4fContext::D64_BlitProc(BlitState* state, int x, int y, const SkPixmap& dst,
+                                      int count, const SkAlpha aa[]) {
+    // FIXME: ignoring coverage for now
+    const LinearGradient4fContext* ctx =
+        static_cast<const LinearGradient4fContext*>(state->fCtx);
+
+    if (ctx->fColorsArePremul) {
+        ctx->shadePremulSpan<uint64_t, kLinear_SkColorProfileType, false>(
+            x, y, dst.writable_addr64(x, y), count);
+    } else {
+        ctx->shadePremulSpan<uint64_t, kLinear_SkColorProfileType, true>(
+            x, y, dst.writable_addr64(x, y), count);
+    }
+}