Threaded generation of software paths

All information needed by the thread is captured by the prepare
callback object, the lambda captures a pointer to that, and does the
mask render. Once it's done, it signals the semaphore (also owned by the
callback). The callback defers the semaphore wait even longer (into the
ASAP upload), so the odds of waiting for the thread are REALLY low.

Also did a bunch of cleanup along the way, and put in some trace markers
so we can monitor how well this is working.

Traces of a GM that includes GPU and SW path rendering (path-reverse):

Original:
    https://screenshot.googleplex.com/f5BG3901tQg.png
Threaded, with wait in the callback (notice pre flush callback blocking):
    https://screenshot.googleplex.com/htOSZFE2s04.png
Current version, with wait deferred to ASAP upload function:
    https://screenshot.googleplex.com/GHjD0U3C34q.png

Bug: skia:
Change-Id: I3d5a230bbd68eb35e1f0574b308485c691435790
Reviewed-on: https://skia-review.googlesource.com/36560
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>

4 files changed, 43 insertions, 4 deletions

diff --git a/tools/flags/SkCommonFlags.cpp b/tools/flags/SkCommonFlags.cpp
index facbd3139b..0986d2b57c 100644
--- a/tools/flags/SkCommonFlags.cpp
+++ b/tools/flags/SkCommonFlags.cpp
@@ -6,6 +6,8 @@
  */
 
 #include "SkCommonFlags.h"
+#include "SkExecutor.h"
+#include "SkOnce.h"
 #include "SkOSFile.h"
 #include "SkOSPath.h"
 
@@ -54,6 +56,9 @@ DEFINE_string(svgs, "", "Directory to read SVGs from, or a single SVG file.");
 DEFINE_int32_2(threads, j, -1, "Run threadsafe tests on a threadpool with this many extra threads, "
                                "defaulting to one extra thread per core.");
 
+DEFINE_int32(gpuThreads, 0, "Create this many extra threads to assist with GPU work, "
+                            "including software path rendering.");
+
 DEFINE_bool2(verbose, v, false, "enable verbose output from the test driver.");
 
 DEFINE_bool2(veryVerbose, V, false, "tell individual tests to be verbose.");
@@ -123,3 +128,9 @@ bool CollectImages(SkCommandLineFlags::StringArray images, SkTArray<SkString>* o
     }
     return true;
 }
+
+SkExecutor* GpuExecutorForTools() {
+    static std::unique_ptr<SkExecutor> gGpuExecutor = (0 != FLAGS_gpuThreads)
+        ? SkExecutor::MakeThreadPool(FLAGS_gpuThreads) : nullptr;
+    return gGpuExecutor.get();
+}
diff --git a/tools/flags/SkCommonFlagsConfig.cpp b/tools/flags/SkCommonFlagsConfig.cpp
index 03b3b96822..16fe914f77 100644
--- a/tools/flags/SkCommonFlagsConfig.cpp
+++ b/tools/flags/SkCommonFlagsConfig.cpp
@@ -72,6 +72,7 @@ static const struct {
     { "glesnarrow",            "gpu", "api=gles,color=f16_narrow" },
     { "gldft",                 "gpu", "api=gl,dit=true" },
     { "glesdft",               "gpu", "api=gles,dit=true" },
+    { "gltestthreading",       "gpu", "api=gl,testThreading=true" },
     { "debuggl",               "gpu", "api=debuggl" },
     { "nullgl",                "gpu", "api=nullgl" },
     { "angle_d3d11_es2",       "gpu", "api=angle_d3d11_es2" },
@@ -170,6 +171,8 @@ static const char configExtendedHelp[] =
     "\t    Use multisampling with N samples.\n"
     "\tstencils\ttype: bool\tdefault: true.\n"
     "\t    Allow the use of stencil buffers.\n"
+    "\ttestThreading\ttype: bool\tdefault: false.\n"
+    "\t    Run config with and without worker threads, check that results match.\n"
     "\n"
     "Predefined configs:\n\n"
     // Help text for pre-defined configs is auto-generated from gPredefinedConfigs
@@ -200,7 +203,7 @@ SkCommandLineConfig::~SkCommandLineConfig() {
 SkCommandLineConfigGpu::SkCommandLineConfigGpu(
     const SkString& tag, const SkTArray<SkString>& viaParts, ContextType contextType, bool useNVPR,
     bool useInstanced, bool useDIText, int samples, SkColorType colorType, SkAlphaType alphaType,
-    sk_sp<SkColorSpace> colorSpace, bool useStencilBuffers)
+    sk_sp<SkColorSpace> colorSpace, bool useStencilBuffers, bool testThreading)
         : SkCommandLineConfig(tag, SkString("gpu"), viaParts)
         , fContextType(contextType)
         , fContextOverrides(ContextOverrides::kNone)
@@ -208,7 +211,8 @@ SkCommandLineConfigGpu::SkCommandLineConfigGpu(
         , fSamples(samples)
         , fColorType(colorType)
         , fAlphaType(alphaType)
-        , fColorSpace(std::move(colorSpace)) {
+        , fColorSpace(std::move(colorSpace))
+        , fTestThreading(testThreading) {
     if (useNVPR) {
         fContextOverrides |= ContextOverrides::kRequireNVPRSupport;
     } else if (!useInstanced) {
@@ -419,6 +423,8 @@ SkCommandLineConfigGpu* parse_command_line_config_gpu(const SkString& tag,
     sk_sp<SkColorSpace> colorSpace = nullptr;
     bool seenUseStencils = false;
     bool useStencils = true;
+    bool seenTestThreading = false;
+    bool testThreading = false;
 
     SkTArray<SkString> optionParts;
     SkStrSplit(options.c_str(), ",", kStrict_SkStrSplitMode, &optionParts);
@@ -452,6 +458,9 @@ SkCommandLineConfigGpu* parse_command_line_config_gpu(const SkString& tag,
         } else if (key.equals("stencils") && !seenUseStencils) {
             valueOk = parse_option_bool(value, &useStencils);
             seenUseStencils = true;
+        } else if (key.equals("testThreading") && !seenTestThreading) {
+            valueOk = parse_option_bool(value, &testThreading);
+            seenTestThreading = true;
         }
         if (!valueOk) {
             return nullptr;
@@ -461,7 +470,8 @@ SkCommandLineConfigGpu* parse_command_line_config_gpu(const SkString& tag,
         return nullptr;
     }
     return new SkCommandLineConfigGpu(tag, vias, contextType, useNVPR, useInstanced, useDIText,
-                                      samples, colorType, alphaType, colorSpace, useStencils);
+                                      samples, colorType, alphaType, colorSpace, useStencils,
+                                      testThreading);
 }
 #endif
 
diff --git a/tools/flags/SkCommonFlagsConfig.h b/tools/flags/SkCommonFlagsConfig.h
index 77f31c32c2..18b138c468 100644
--- a/tools/flags/SkCommonFlagsConfig.h
+++ b/tools/flags/SkCommonFlagsConfig.h
@@ -56,7 +56,8 @@ class SkCommandLineConfigGpu : public SkCommandLineConfig {
     SkCommandLineConfigGpu(const SkString& tag, const SkTArray<SkString>& viaParts,
                            ContextType contextType, bool useNVPR, bool useInstanced, bool useDIText,
                            int samples, SkColorType colorType, SkAlphaType alphaType,
-                           sk_sp<SkColorSpace> colorSpace, bool useStencilBuffers);
+                           sk_sp<SkColorSpace> colorSpace, bool useStencilBuffers,
+                           bool testThreading);
     const SkCommandLineConfigGpu* asConfigGpu() const override { return this; }
     ContextType getContextType() const { return fContextType; }
     ContextOverrides getContextOverrides() const { return fContextOverrides; }
@@ -71,6 +72,7 @@ class SkCommandLineConfigGpu : public SkCommandLineConfig {
     SkColorType getColorType() const { return fColorType; }
     SkAlphaType getAlphaType() const { return fAlphaType; }
     SkColorSpace* getColorSpace() const { return fColorSpace.get(); }
+    bool getTestThreading() const { return fTestThreading; }
 
   private:
     ContextType fContextType;
@@ -80,6 +82,7 @@ class SkCommandLineConfigGpu : public SkCommandLineConfig {
     SkColorType fColorType;
     SkAlphaType fAlphaType;
     sk_sp<SkColorSpace> fColorSpace;
+    bool fTestThreading;
 };
 #endif
 
diff --git a/tools/flags/SkCommonFlagsGpuThreads.h b/tools/flags/SkCommonFlagsGpuThreads.h
new file mode 100644
index 0000000000..a6042fa821
--- /dev/null
+++ b/tools/flags/SkCommonFlagsGpuThreads.h
@@ -0,0 +1,15 @@
+/*
+ * Copyright 2017 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#ifndef SK_COMMON_FLAGS_GPU_THREADS
+#define SK_COMMON_FLAGS_GPU_THREADS
+
+class SkExecutor;
+
+SkExecutor* GpuExecutorForTools();
+
+#endif