remove SkRP-based runtime effects

The SkVM backend can draw all the existing GMs.
Image diffs look good.  Rough local perf changes:

    spiral_rt:            14.3ms ->  5.9ms w/o JIT, or 2.5ms w/JIT
    overdrawcolorfilter:   4.2       4.9*              2.6
    overdraw_canvas:      10.1       8.9               2.7
    runtimecolorfilter:    8.5       8.5               7.5

with one regression marked with a *.

Change-Id: Ib4f11422f3c2e8756960794a965385fa616ad8bf
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308193
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>

src/core/SkRasterPipeline.h

@@ -37,7 +37,7 @@ class SkData;
  */
 
 #define SK_RASTER_PIPELINE_STAGES(M)                               \
-    M(callback) M(interpreter)                                     \
+    M(callback)                                                    \
     M(move_src_dst) M(move_dst_src)                                \
     M(clamp_0) M(clamp_1) M(clamp_a) M(clamp_gamut)                \
     M(unpremul) M(premul) M(premul_dst)                            \
@@ -166,16 +166,6 @@ class ByteCode;
 class ByteCodeFunction;
 }
 
-struct SkRasterPipeline_InterpreterCtx {
-    const SkSL::ByteCode*         byteCode;
-    const SkSL::ByteCodeFunction* fn;
-
-    SkColor4f     paintColor;
-    sk_sp<SkData> inputs;
-    int           ninputs;
-    bool          shaderConvention;  // if false, we're a colorfilter
-};
-
 struct SkRasterPipeline_GradientCtx {
     size_t stopCount;
     float* fs[4];

src/core/SkRuntimeEffect.cpp

@@ -717,25 +717,9 @@ public:
     }
 
     bool onAppendStages(const SkStageRec& rec, bool shaderIsOpaque) const override {
-        auto ctx = rec.fAlloc->make<SkRasterPipeline_InterpreterCtx>();
-        // don't need to set ctx->paintColor
-        ctx->inputs = fInputs;
-        ctx->ninputs = fEffect->uniformSize() / 4;
-        ctx->shaderConvention = false;
-
-        ctx->byteCode = this->byteCode();
-        if (!ctx->byteCode || !ctx->byteCode->canRun()) {
         return false;
     }
 
-        ctx->fn = ctx->byteCode->getFunction("main");
-        if (!ctx->fn) {
-            return false;
-        }
-        rec.fPipeline->append(SkRasterPipeline::interpreter, ctx);
-        return true;
-    }
-
     skvm::Color onProgram(skvm::Builder* p, skvm::Color c,
                           SkColorSpace* /*dstCS*/,
                           skvm::Uniforms* uniforms, SkArenaAlloc*) const override {
@@ -918,35 +902,9 @@ public:
     }
 
     bool onAppendStages(const SkStageRec& rec) const override {
-        SkMatrix inverse;
-        if (!this->computeTotalInverse(rec.fMatrixProvider.localToDevice(), rec.fLocalM,
-                                       &inverse)) {
         return false;
     }
 
-        auto ctx = rec.fAlloc->make<SkRasterPipeline_InterpreterCtx>();
-        ctx->paintColor = rec.fPaint.getColor4f();
-        ctx->inputs = this->getUniforms(rec.fMatrixProvider, rec.fDstCS);
-        if (!ctx->inputs) {
-            return false;
-        }
-        ctx->ninputs = fEffect->uniformSize() / 4;
-        ctx->shaderConvention = true;
-
-        ctx->byteCode = this->byteCode();
-        if (!ctx->byteCode || !ctx->byteCode->canRun()) {
-            return false;
-        }
-        ctx->fn = ctx->byteCode->getFunction("main");
-        if (!ctx->fn) {
-            return false;
-        }
-        rec.fPipeline->append(SkRasterPipeline::seed_shader);
-        rec.fPipeline->append_matrix(rec.fAlloc, inverse);
-        rec.fPipeline->append(SkRasterPipeline::interpreter, ctx);
-        return true;
-    }
-
     skvm::Color onProgram(skvm::Builder* p,
                           skvm::Coord device, skvm::Coord local, skvm::Color paint,
                           const SkMatrixProvider& matrices, const SkMatrix* localM,

src/opts/SkRasterPipeline_opts.h

@@ -11,7 +11,6 @@
 #include "include/core/SkData.h"
 #include "include/core/SkTypes.h"
 #include "src/core/SkUtils.h"  // unaligned_{load,store}
-#include "src/sksl/SkSLByteCode.h"
 
 // Every function in this file should be marked static and inline using SI.
 #if defined(__clang__)
@@ -2662,44 +2661,6 @@ STAGE(callback, SkRasterPipeline_CallbackCtx* c) {
     load4(c->read_from,0, &r,&g,&b,&a);
 }
 
-// shader:      void main(float2 p, inout half4 color)
-// colorfilter: void main(inout half4 color)
-STAGE(interpreter, SkRasterPipeline_InterpreterCtx* c) {
-    // If N is less than the interpreter's VecWidth, then we are doing more work than necessary in
-    // the interpreter. This is a known issue, and will be addressed at some point.
-    float xx[N], yy[N],
-          rr[N], gg[N], bb[N], aa[N];
-
-    float*  args[]  = { xx, yy, rr, gg, bb, aa };
-    float** in_args = args;
-    int     in_count = 6;
-
-    if (c->shaderConvention) {
-        // our caller must have called seed_shader to set these
-        sk_unaligned_store(xx, r);
-        sk_unaligned_store(yy, g);
-        sk_unaligned_store(rr, F(c->paintColor.fR));
-        sk_unaligned_store(gg, F(c->paintColor.fG));
-        sk_unaligned_store(bb, F(c->paintColor.fB));
-        sk_unaligned_store(aa, F(c->paintColor.fA));
-    } else {
-        in_args += 2;   // skip x,y
-        in_count = 4;
-        sk_unaligned_store(rr, r);
-        sk_unaligned_store(gg, g);
-        sk_unaligned_store(bb, b);
-        sk_unaligned_store(aa, a);
-    }
-
-    SkAssertResult(c->byteCode->runStriped(c->fn, tail ? tail : N, in_args, in_count, nullptr, 0,
-                                           (const float*)c->inputs->data(), c->ninputs));
-
-    r = sk_unaligned_load<F>(rr);
-    g = sk_unaligned_load<F>(gg);
-    b = sk_unaligned_load<F>(bb);
-    a = sk_unaligned_load<F>(aa);
-}
-
 STAGE(gauss_a_to_rgba, Ctx::None) {
     // x = 1 - x;
     // exp(-x * x * 4) - 0.018f;