skia2/tools/flags/CommonFlagsGpu.cpp
Brian Osman f0095cf86a Add option to use a program binary cache in DM, etc
Change-Id: I2117e41388962682a40f9db9ffc62150b30c7847
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/221779
Reviewed-by: Brian Salomon <bsalomon@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
2019-06-19 00:00:11 +00:00

109 lines
4.1 KiB
C++

/*
* Copyright 2014 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "include/core/SkExecutor.h"
#include "include/gpu/GrContextOptions.h"
#include "tools/flags/CommonFlags.h"
#include "tools/gpu/MemoryCache.h"
DEFINE_int(gpuThreads,
2,
"Create this many extra threads to assist with GPU work, "
"including software path rendering. Defaults to two.");
static DEFINE_bool(cachePathMasks, true,
"Allows path mask textures to be cached in GPU configs.");
static DEFINE_bool(noGS, false, "Disables support for geometry shaders.");
static DEFINE_bool(cc, false, "Allow coverage counting shortcuts to render paths?");
static DEFINE_string(pr, "",
"Set of enabled gpu path renderers. Defined as a list of: "
"[~]none [~]dashline [~]nvpr [~]ccpr [~]aahairline [~]aaconvex [~]aalinearizing "
"[~]small [~]tess] [~]all");
static DEFINE_bool(disableDriverCorrectnessWorkarounds, false,
"Disables all GPU driver correctness workarounds");
static DEFINE_bool(reduceOpListSplitting, false, "Improve opList sorting");
static DEFINE_bool(dontReduceOpListSplitting, false, "Allow more opList splitting");
static DEFINE_bool(programBinaryCache, false, "Use in-memory program binary cache");
static GpuPathRenderers get_named_pathrenderers_flags(const char* name) {
if (!strcmp(name, "none")) {
return GpuPathRenderers::kNone;
} else if (!strcmp(name, "dashline")) {
return GpuPathRenderers::kDashLine;
} else if (!strcmp(name, "nvpr")) {
return GpuPathRenderers::kStencilAndCover;
} else if (!strcmp(name, "ccpr")) {
return GpuPathRenderers::kCoverageCounting;
} else if (!strcmp(name, "aahairline")) {
return GpuPathRenderers::kAAHairline;
} else if (!strcmp(name, "aaconvex")) {
return GpuPathRenderers::kAAConvex;
} else if (!strcmp(name, "aalinearizing")) {
return GpuPathRenderers::kAALinearizing;
} else if (!strcmp(name, "small")) {
return GpuPathRenderers::kSmall;
} else if (!strcmp(name, "tess")) {
return GpuPathRenderers::kTessellating;
} else if (!strcmp(name, "all")) {
return GpuPathRenderers::kAll;
}
SK_ABORT(SkStringPrintf("error: unknown named path renderer \"%s\"\n", name).c_str());
return GpuPathRenderers::kNone;
}
static GpuPathRenderers collect_gpu_path_renderers_from_flags() {
if (FLAGS_pr.isEmpty()) {
return GpuPathRenderers::kAll;
}
GpuPathRenderers gpuPathRenderers = ('~' == FLAGS_pr[0][0])
? GpuPathRenderers::kAll
: GpuPathRenderers::kNone;
for (int i = 0; i < FLAGS_pr.count(); ++i) {
const char* name = FLAGS_pr[i];
if (name[0] == '~') {
gpuPathRenderers &= ~get_named_pathrenderers_flags(&name[1]);
} else {
gpuPathRenderers |= get_named_pathrenderers_flags(name);
}
}
return gpuPathRenderers;
}
void SetCtxOptionsFromCommonFlags(GrContextOptions* ctxOptions) {
static std::unique_ptr<SkExecutor> gGpuExecutor = (0 != FLAGS_gpuThreads)
? SkExecutor::MakeFIFOThreadPool(FLAGS_gpuThreads)
: nullptr;
ctxOptions->fExecutor = gGpuExecutor.get();
ctxOptions->fDisableCoverageCountingPaths = !FLAGS_cc;
ctxOptions->fAllowPathMaskCaching = FLAGS_cachePathMasks;
ctxOptions->fSuppressGeometryShaders = FLAGS_noGS;
ctxOptions->fGpuPathRenderers = collect_gpu_path_renderers_from_flags();
ctxOptions->fDisableDriverCorrectnessWorkarounds = FLAGS_disableDriverCorrectnessWorkarounds;
if (FLAGS_programBinaryCache) {
static sk_gpu_test::MemoryCache gMemoryCache;
ctxOptions->fPersistentCache = &gMemoryCache;
}
if (FLAGS_reduceOpListSplitting) {
SkASSERT(!FLAGS_dontReduceOpListSplitting);
ctxOptions->fReduceOpListSplitting = GrContextOptions::Enable::kYes;
} else if (FLAGS_dontReduceOpListSplitting) {
ctxOptions->fReduceOpListSplitting = GrContextOptions::Enable::kNo;
}
}