remove lowp bilerp_clamp_8888 stage

I encountered some trouble getting this turned on in Chromium,
so we've got some undesirable skew between what's tested and shipping.

No one has complained about software rasterization perf to me lately,
so let's just drop this, keeping the already-launched highp
bilerp_clamp_8888 stage.

Change-Id: I9a6557d9b468765989f97369a976c5028f81eab3
Reviewed-on: https://skia-review.googlesource.com/135563
Commit-Queue: Mike Klein <mtklein@chromium.org>
Commit-Queue: Florin Malita <fmalita@chromium.org>
Auto-Submit: Mike Klein <mtklein@chromium.org>
Reviewed-by: Florin Malita <fmalita@chromium.org>

1 files changed, 1 insertions, 64 deletions

diff --git a/src/opts/SkRasterPipeline_opts.h b/src/opts/SkRasterPipeline_opts.h
index f7c757de46..fac5d1b93b 100644
--- a/src/opts/SkRasterPipeline_opts.h
+++ b/src/opts/SkRasterPipeline_opts.h
@@ -3180,70 +3180,6 @@ STAGE_PP(srcover_bgra_8888, const SkJumper_MemoryCtx* ctx) {
     store_8888_(ptr, tail, b,g,r,a);
 }
 
-#if defined(SK_DISABLE_LOWP_BILERP_CLAMP_CLAMP_STAGE)
-    static void(*bilerp_clamp_8888)(void) = nullptr;
-#else
-STAGE_GP(bilerp_clamp_8888, const SkJumper_GatherCtx* ctx) {
-    // (cx,cy) are the center of our sample.
-    F cx = x,
-      cy = y;
-
-    // All sample points are at the same fractional offset (fx,fy).
-    // They're the 4 corners of a logical 1x1 pixel surrounding (x,y) at (0.5,0.5) offsets.
-    F fx = fract(cx + 0.5f),
-      fy = fract(cy + 0.5f);
-
-    // We'll accumulate the color of all four samples into {r,g,b,a} directly.
-    r = g = b = a = 0;
-
-    // The first three sample points will calculate their area using math
-    // just like in the float code above, but the fourth will take up all the rest.
-    //
-    // Logically this is the same as doing the math for the fourth pixel too,
-    // but rounding error makes this a better strategy, keeping opaque opaque, etc.
-    //
-    // We can keep up to 8 bits of fractional precision without overflowing 16-bit,
-    // so our "1.0" area is 256.
-    const uint16_t bias = 256;
-    U16 remaining = bias;
-
-    for (float dy = -0.5f; dy <= +0.5f; dy += 1.0f)
-    for (float dx = -0.5f; dx <= +0.5f; dx += 1.0f) {
-        // (x,y) are the coordinates of this sample point.
-        F x = cx + dx,
-          y = cy + dy;
-
-        // ix_and_ptr() will clamp to the image's bounds for us.
-        const uint32_t* ptr;
-        U32 ix = ix_and_ptr(&ptr, ctx, x,y);
-
-        U16 sr,sg,sb,sa;
-        from_8888(gather<U32>(ptr, ix), &sr,&sg,&sb,&sa);
-
-        // In bilinear interpolation, the 4 pixels at +/- 0.5 offsets from the sample pixel center
-        // are combined in direct proportion to their area overlapping that logical query pixel.
-        // At positive offsets, the x-axis contribution to that rectangle is fx,
-        // or (1-fx) at negative x.  Same deal for y.
-        F sx = (dx > 0) ? fx : 1.0f - fx,
-          sy = (dy > 0) ? fy : 1.0f - fy;
-
-        U16 area = (dy == 0.5f && dx == 0.5f) ? remaining
-                                              : cast<U16>(sx * sy * bias);
-        remaining -= area;
-
-        r += sr * area;
-        g += sg * area;
-        b += sb * area;
-        a += sa * area;
-    }
-
-    r /= bias;
-    g /= bias;
-    b /= bias;
-    a /= bias;
-}
-#endif
-
 // Now we'll add null stand-ins for stages we haven't implemented in lowp.
 // If a pipeline uses these stages, it'll boot it out of lowp into highp.
 
@@ -3268,6 +3204,7 @@ static NotImplemented
         mirror_x, repeat_x,
         mirror_y, repeat_y,
         negate_x,
+        bilerp_clamp_8888,
         bilinear_nx, bilinear_ny, bilinear_px, bilinear_py,
         bicubic_n3x, bicubic_n1x, bicubic_p1x, bicubic_p3x,
         bicubic_n3y, bicubic_n1y, bicubic_p1y, bicubic_p3y,