Use first/second instead of min/max in 2pt conical gradient

Here's the tiny performance gain:

$python tools/calmbench/calmbench.py firstsecond --extraarg "-m conic"

firstsecond (compared to master) is likely
    4.23% faster in gradient_conicalOut_clamp_3color
    4.23% faster in gradient_conicalOutZero_clamp_3color
    4.79% faster in gradient_conical_clamp_shallow_dither
    6.04% faster in gradient_conical_clamp_3color
    6.04% faster in gradient_conicalZero_clamp_3color
    6.42% faster in gradient_conicalOut_clamp
    6.43% faster in gradient_conicalOutZero_clamp
    6.74% faster in gradient_conical_clamp
    6.98% faster in gradient_conical_clamp_shallow
    6.98% faster in gradient_conicalZero_clamp

Bug: skia:
Change-Id: Id74866908b99753ed8b16a657d3f67c9255d0043
Reviewed-on: https://skia-review.googlesource.com/76561
Commit-Queue: Yuqian Li <liyuqian@google.com>
Reviewed-by: Mike Klein <mtklein@chromium.org>
Reviewed-by: Florin Malita <fmalita@chromium.org>

src/core/SkRasterPipeline.h

@@ -85,8 +85,8 @@ struct SkJumper_Engine;
     M(evenly_spaced_2_stop_gradient)                               \
     M(xy_to_unit_angle)                                            \
     M(xy_to_radius)                                                \
-    M(xy_to_2pt_conical_quadratic_min)                             \
-    M(xy_to_2pt_conical_quadratic_max)                             \
+    M(xy_to_2pt_conical_quadratic_first)                           \
+    M(xy_to_2pt_conical_quadratic_second)                          \
     M(xy_to_2pt_conical_linear)                                    \
     M(mask_2pt_conical_degenerates) M(apply_vector_mask)           \
     M(byte_tables) M(byte_tables_rgb)                              \

src/jumper/SkJumper.cpp

@@ -259,8 +259,8 @@ extern "C" {
     LOWP(evenly_spaced_2_stop_gradient)
     LOWP(xy_to_unit_angle)
     LOWP(xy_to_radius)
-    TODO(xy_to_2pt_conical_quadratic_min)
-    TODO(xy_to_2pt_conical_quadratic_max)
+    TODO(xy_to_2pt_conical_quadratic_first)
+    TODO(xy_to_2pt_conical_quadratic_second)
     TODO(xy_to_2pt_conical_linear)
     TODO(mask_2pt_conical_degenerates) TODO(apply_vector_mask)
     TODO(byte_tables) TODO(byte_tables_rgb)

src/jumper/SkJumper_stages.cpp

@@ -304,6 +304,9 @@ STAGE(store_rgba, float* ptr) {
 
 SI F inv(F x) { return 1.0f - x; }
 SI F two(F x) { return x + x; }
+SI F first (F a, F b) { return a; }
+SI F second(F a, F b) { return b; }
+
 
 BLEND_MODE(clear)    { return 0; }
 BLEND_MODE(srcatop)  { return s*da + d*inv(sa); }
@@ -1287,12 +1290,12 @@ SI F solve_2pt_conical_quadratic(const SkJumper_2PtConicalCtx* c, F x, F y, F (*
                   (-coeffB - sqrt_disc) * (invCoeffA * 0.5f));
 }
 
-STAGE(xy_to_2pt_conical_quadratic_max, const SkJumper_2PtConicalCtx* ctx) {
-    r = solve_2pt_conical_quadratic(ctx, r, g, max);
+STAGE(xy_to_2pt_conical_quadratic_first, const SkJumper_2PtConicalCtx* ctx) {
+    r = solve_2pt_conical_quadratic(ctx, r, g, first);
 }
 
-STAGE(xy_to_2pt_conical_quadratic_min, const SkJumper_2PtConicalCtx* ctx) {
-    r = solve_2pt_conical_quadratic(ctx, r, g, min);
+STAGE(xy_to_2pt_conical_quadratic_second, const SkJumper_2PtConicalCtx* ctx) {
+    r = solve_2pt_conical_quadratic(ctx, r, g, second);
 }
 
 STAGE(xy_to_2pt_conical_linear, const SkJumper_2PtConicalCtx* c) {

src/shaders/gradients/SkTwoPointConicalGradient.cpp

@@ -210,6 +210,7 @@ void SkTwoPointConicalGradient::appendGradientStages(SkArenaAlloc* alloc, SkRast
     } else {
         isWellBehaved = SkScalarAbs(dRadius) >= dCenter;
         bool isFlipped = isWellBehaved && dRadius < 0;
+        bool isFirst = (coeffA > 0) != isFlipped;
 
         // We want the larger root, per spec:
         //   "For all values of ω where r(ω) > 0, starting with the value of ω nearest
@@ -221,8 +222,11 @@ void SkTwoPointConicalGradient::appendGradientStages(SkArenaAlloc* alloc, SkRast
         // (https://html.spec.whatwg.org/multipage/canvas.html#dom-context-2d-createradialgradient)
         //
         // ... except when the gradient is flipped.
-        p->append(isFlipped ? SkRasterPipeline::xy_to_2pt_conical_quadratic_min
-                            : SkRasterPipeline::xy_to_2pt_conical_quadratic_max, ctx);
+        //
+        // The first root (-b + sqrt(...)) / 2a is larger if a (i.e. coeffA) > 0 and the second
+        // root (-b - sqrt(...)) / 2a is larger if a < 0.
+        p->append(isFirst ? SkRasterPipeline::xy_to_2pt_conical_quadratic_first
+                          : SkRasterPipeline::xy_to_2pt_conical_quadratic_second, ctx);
     }
 
     if (!isWellBehaved) {