Don't use the Sk4f gradient impl without SIMD

Also remove the SK_SUPPORT_LEGACY_LINEAR_GRADIENT_TABLE guard since it is no
longer used in Chromium.

BUG=chromium:563492
R=reed@google.com,mtklein@google.com
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

Review URL: https://codereview.chromium.org/1489233005

src/effects/gradients/SkLinearGradient.cpp

@@ -282,7 +282,8 @@ void SkLinearGradient::LinearGradientContext::shadeSpan(int x, int y, SkPMColor*
     SkASSERT(count > 0);
     const SkLinearGradient& linearGradient = static_cast<const SkLinearGradient&>(fShader);
 
-#ifndef SK_SUPPORT_LEGACY_LINEAR_GRADIENT_TABLE
+// Only use the Sk4f impl when known to be fast.
+#if defined(SKNX_IS_FAST)
     if (SkShader::kClamp_TileMode == linearGradient.fTileMode &&
         kLinear_MatrixClass == fDstToIndexClass)
     {

src/opts/SkNx_avx.h

@@ -13,6 +13,8 @@
 // All the SSE specializations are still good ideas.  We'll just add Sk8f.
 #include "SkNx_sse.h"
 
+// SkNx_sse.h defines SKNX_IS_FAST.
+
 namespace {  // See SkNx.h
 
 template <>

src/opts/SkNx_neon.h

@@ -8,6 +8,8 @@
 #ifndef SkNx_neon_DEFINED
 #define SkNx_neon_DEFINED
 
+#define SKNX_IS_FAST
+
 namespace {  // See SkNx.h
 
 // Well, this is absurd.  The shifts require compile-time constant arguments.

src/opts/SkNx_sse.h

@@ -10,6 +10,8 @@
 
 // This file may assume <= SSE2, but must check SK_CPU_SSE_LEVEL for anything more recent.
 
+#define SKNX_IS_FAST
+
 namespace {  // See SkNx.h