add srcover_bgra_8888

Chrome generally uses BGRA buffers, so srcover_rgba_8888 isn't really
doing them any good.  Probably a good idea to cover both kN32 options
any time we specialize like this?

There's one small diff, so I've lazily guarded this by
SK_LEGACY_LOWP_STAGES, which I want to rebaseline today anyway.

Change-Id: Ice672aa01a3fc83be0798580d6730a54df075478
Reviewed-on: https://skia-review.googlesource.com/63301
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Mike Reed <reed@google.com>

src/core/SkRasterPipeline.h

@@ -64,7 +64,7 @@ struct SkJumper_Engine;
     M(colorburn) M(colordodge) M(darken) M(difference)           \
     M(exclusion) M(hardlight) M(lighten) M(overlay) M(softlight) \
     M(hue) M(saturation) M(color) M(luminosity)                  \
-    M(srcover_rgba_8888)                                         \
+    M(srcover_rgba_8888) M(srcover_bgra_8888)                    \
     M(luminance_to_alpha)                                        \
     M(matrix_translate) M(matrix_scale_translate)                \
     M(matrix_2x3) M(matrix_3x4) M(matrix_4x5) M(matrix_4x3)      \

src/core/SkRasterPipelineBlitter.cpp

@@ -305,6 +305,7 @@ void SkRasterPipelineBlitter::blitRect(int x, int y, int w, int h) {
     if (!fBlitRect) {
         SkRasterPipeline p(fAlloc);
         p.extend(fColorPipeline);
+    #if defined(SK_LEGACY_LOWP_STAGES)
         if (fBlend == SkBlendMode::kSrcOver
                 && fDst.info().colorType() == kRGBA_8888_SkColorType
                 && !fDst.colorSpace()
@@ -312,6 +313,19 @@ void SkRasterPipelineBlitter::blitRect(int x, int y, int w, int h) {
                 && fDitherRate == 0.0f) {
             p.clamp_if_unclamped(kPremul_SkAlphaType);
             p.append(SkRasterPipeline::srcover_rgba_8888, &fDstPtr);
+    #else
+        if (fBlend == SkBlendMode::kSrcOver
+                && (fDst.info().colorType() == kRGBA_8888_SkColorType ||
+                    fDst.info().colorType() == kBGRA_8888_SkColorType)
+                && !fDst.colorSpace()
+                && fDst.info().alphaType() != kUnpremul_SkAlphaType
+                && fDitherRate == 0.0f) {
+            p.clamp_if_unclamped(kPremul_SkAlphaType);
+            auto stage = fDst.info().colorType() == kRGBA_8888_SkColorType
+                       ? SkRasterPipeline::srcover_rgba_8888
+                       : SkRasterPipeline::srcover_bgra_8888;
+            p.append(stage, &fDstPtr);
+    #endif
         } else {
             if (fBlend != SkBlendMode::kSrc) {
                 this->append_load_dst(&p);

src/jumper/SkJumper.cpp

@@ -199,7 +199,7 @@ extern "C" {
     LOWP(load_g8)   LOWP(load_g8_dst)
     LOWP(load_565)  LOWP(load_565_dst)  LOWP(store_565)
     LOWP(swap_rb)
-    LOWP(srcover_rgba_8888)
+    LOWP(srcover_rgba_8888) LOWP(srcover_bgra_8888)
     LOWP(lerp_1_float)
     LOWP(lerp_u8)
     LOWP(lerp_565)

src/jumper/SkJumper_stages.cpp

@@ -502,6 +502,30 @@ STAGE(srcover_rgba_8888, const SkJumper_MemoryCtx* ctx) {
     store(ptr, dst, tail);
 }
 
+STAGE(srcover_bgra_8888, const SkJumper_MemoryCtx* ctx) {
+    auto ptr = ptr_at_xy<uint32_t>(ctx, dx,dy);
+
+    U32 dst = load<U32>(ptr, tail);
+    db = cast((dst      ) & 0xff);
+    dg = cast((dst >>  8) & 0xff);
+    dr = cast((dst >> 16) & 0xff);
+    da = cast((dst >> 24)       );
+    // {dr,dg,db,da} are in [0,255]
+    // { r, g, b, a} are in [0,  1]
+
+    r = mad(dr, inv(a), r*255.0f);
+    g = mad(dg, inv(a), g*255.0f);
+    b = mad(db, inv(a), b*255.0f);
+    a = mad(da, inv(a), a*255.0f);
+    // { r, g, b, a} are now in [0,255]
+
+    dst = round(b, 1.0f)
+        | round(g, 1.0f) <<  8
+        | round(r, 1.0f) << 16
+        | round(a, 1.0f) << 24;
+    store(ptr, dst, tail);
+}
+
 STAGE(clamp_0, Ctx::None) {
     r = max(r, 0);
     g = max(g, 0);

src/jumper/SkJumper_stages_lowp.cpp

@@ -696,5 +696,15 @@ STAGE_PP(srcover_rgba_8888, const SkJumper_MemoryCtx* ctx) {
     a = a + div255( da*inv(a) );
     store_8888(ptr, tail, r,g,b,a);
 }
+STAGE_PP(srcover_bgra_8888, const SkJumper_MemoryCtx* ctx) {
+    auto ptr = ptr_at_xy<uint32_t>(ctx, dx,dy);
+
+    load_8888(ptr, tail, &db,&dg,&dr,&da);
+    r = r + div255( dr*inv(a) );
+    g = g + div255( dg*inv(a) );
+    b = b + div255( db*inv(a) );
+    a = a + div255( da*inv(a) );
+    store_8888(ptr, tail, b,g,r,a);
+}
 
 #endif//defined(__clang__)