SkRasterPipeline shader adapter

Reland of https://skia-review.googlesource.com/c/7615.

(lifted from https://skia-review.googlesource.com/c/7088/)

CQ_INCLUDE_TRYBOTS=skia.primary:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD

Change-Id: I797a2f0ae80209c8637875418e08d2fa03249672
Reviewed-on: https://skia-review.googlesource.com/7731
Commit-Queue: Florin Malita <fmalita@google.com>
Reviewed-by: Mike Reed <reed@google.com>

5 files changed, 59 insertions, 4 deletions

diff --git a/include/core/SkShader.h b/include/core/SkShader.h
index a885f0b982..30a10dcc9f 100644
--- a/include/core/SkShader.h
+++ b/include/core/SkShader.h
@@ -503,9 +503,7 @@ protected:
 
     virtual bool onAppendStages(SkRasterPipeline*, SkColorSpace*, SkArenaAlloc*,
                                 const SkMatrix&, const SkPaint&,
-                                const SkMatrix* /*local matrix*/) const {
-        return false;
-    }
+                                const SkMatrix* /*local matrix*/) const;
 
 private:
     // This is essentially const, but not officially so it can be modified in
diff --git a/src/core/SkRasterPipeline.h b/src/core/SkRasterPipeline.h
index 5143d523e4..2dd2c32a99 100644
--- a/src/core/SkRasterPipeline.h
+++ b/src/core/SkRasterPipeline.h
@@ -95,7 +95,8 @@
     M(bicubic_n3y) M(bicubic_n1y) M(bicubic_p1y) M(bicubic_p3y)  \
     M(save_xy) M(accumulate)                                     \
     M(linear_gradient_2stops)                                    \
-    M(byte_tables)
+    M(byte_tables)                                               \
+    M(shader_adapter)
 
 class SkRasterPipeline {
 public:
diff --git a/src/core/SkShader.cpp b/src/core/SkShader.cpp
index 0e03fb7112..9e562c9276 100644
--- a/src/core/SkShader.cpp
+++ b/src/core/SkShader.cpp
@@ -5,6 +5,7 @@
  * found in the LICENSE file.
  */
 
+#include "SkArenaAlloc.h"
 #include "SkAtomics.h"
 #include "SkBitmapProcShader.h"
 #include "SkColorShader.h"
@@ -13,9 +14,12 @@
 #include "SkPaint.h"
 #include "SkPicture.h"
 #include "SkPictureShader.h"
+#include "SkPM4fPriv.h"
+#include "SkRasterPipeline.h"
 #include "SkReadBuffer.h"
 #include "SkScalar.h"
 #include "SkShader.h"
+#include "SkTLazy.h"
 #include "SkWriteBuffer.h"
 
 #if SK_SUPPORT_GPU
@@ -262,6 +266,40 @@ bool SkShader::appendStages(SkRasterPipeline* pipeline,
     return this->onAppendStages(pipeline, dst, scratch, ctm, paint, nullptr);
 }
 
+bool SkShader::onAppendStages(SkRasterPipeline* p,
+                              SkColorSpace* cs,
+                              SkArenaAlloc* alloc,
+                              const SkMatrix& ctm,
+                              const SkPaint& paint,
+                              const SkMatrix* localM) const {
+    // Legacy shaders handle the paint opacity internally,
+    // but RP applies it as a separate stage.
+    SkTCopyOnFirstWrite<SkPaint> opaquePaint(paint);
+    if (paint.getAlpha() != SK_AlphaOPAQUE) {
+        opaquePaint.writable()->setAlpha(SK_AlphaOPAQUE);
+    }
+
+    ContextRec rec(*opaquePaint, ctm, localM, ContextRec::kPM4f_DstType, cs);
+    if (auto* ctx = this->createContext(rec,
+                                        alloc->makeArrayDefault<char>(this->contextSize(rec)))) {
+        struct ContextDestroyer {
+            ContextDestroyer(Context* ctx) : fContext(ctx) {}
+            ~ContextDestroyer() { fContext->~Context(); }
+
+            Context* fContext;
+        };
+
+        alloc->make<ContextDestroyer>(ctx);
+        p->append(SkRasterPipeline::shader_adapter, ctx);
+
+        // Legacy shaders aren't aware of color spaces. We can pretty
+        // safely assume they're in sRGB gamut.
+        return append_gamut_transform(p, alloc,
+                                      SkColorSpace::MakeNamed(SkColorSpace::kSRGB_Named).get(), cs);
+    }
+    return false;
+}
+
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 
 sk_sp<SkFlattenable> SkEmptyShader::CreateProc(SkReadBuffer&) {
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index fe74a77545..e81872291e 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -623,6 +623,16 @@ public:
             _mm256_storeu_ps((float*)ptr + 2*8, _45);
             _mm256_storeu_ps((float*)ptr + 3*8, _67);
         }
+        AI static void Load4(const void* ptr, SkNx* r, SkNx* g, SkNx* b, SkNx* a) {
+            Sk4f rl, gl, bl, al,
+                 rh, gh, bh, ah;
+            Sk4f::Load4((const float*)ptr +  0, &rl, &gl, &bl, &al);
+            Sk4f::Load4((const float*)ptr + 16, &rh, &gh, &bh, &ah);
+            *r = _mm256_setr_m128(rl.fVec, rh.fVec);
+            *g = _mm256_setr_m128(gl.fVec, gh.fVec);
+            *b = _mm256_setr_m128(bl.fVec, bh.fVec);
+            *a = _mm256_setr_m128(al.fVec, ah.fVec);
+        }
 
         AI SkNx operator+(const SkNx& o) const { return _mm256_add_ps(fVec, o.fVec); }
         AI SkNx operator-(const SkNx& o) const { return _mm256_sub_ps(fVec, o.fVec); }
diff --git a/src/opts/SkRasterPipeline_opts.h b/src/opts/SkRasterPipeline_opts.h
index 1b27fc25d4..dec81d990a 100644
--- a/src/opts/SkRasterPipeline_opts.h
+++ b/src/opts/SkRasterPipeline_opts.h
@@ -18,6 +18,7 @@
 #include "SkPM4f.h"
 #include "SkPM4fPriv.h"
 #include "SkRasterPipeline.h"
+#include "SkShader.h"
 #include "SkSRGB.h"
 
 namespace {
@@ -1085,6 +1086,13 @@ STAGE_CTX(byte_tables, const void*) {
     a = SkNf_from_byte(gather(tail, tables->a, SkNf_round(255.0f, a)));
 }
 
+STAGE_CTX(shader_adapter, SkShader::Context*) {
+    SkPM4f buf[N];
+    static_assert(sizeof(buf) == sizeof(r) + sizeof(g) + sizeof(b) + sizeof(a), "");
+    ctx->shadeSpan4f(x, (int)g[0], buf, N);
+    SkNf::Load4(buf, &r, &g, &b, &a);
+}
+
 SI Fn enum_to_Fn(SkRasterPipeline::StockStage st) {
     switch (st) {
     #define M(stage) case SkRasterPipeline::stage: return stage;