blob: 133ffea10f8347a184a0606365d23156b2a204da [file] [log] [blame]
/*
* Copyright 2016 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "GrAuditTrail.h"
#include "ops/GrOp.h"
const int GrAuditTrail::kGrAuditTrailInvalidID = -1;
void GrAuditTrail::addOp(const GrOp* op, GrGpuResource::UniqueID renderTargetID) {
SkASSERT(fEnabled);
Op* auditOp = new Op;
fOpPool.emplace_back(auditOp);
auditOp->fName = op->name();
auditOp->fBounds = op->bounds();
auditOp->fClientID = kGrAuditTrailInvalidID;
auditOp->fOpListID = kGrAuditTrailInvalidID;
auditOp->fChildID = kGrAuditTrailInvalidID;
// consume the current stack trace if any
auditOp->fStackTrace = fCurrentStackTrace;
fCurrentStackTrace.reset();
if (fClientID != kGrAuditTrailInvalidID) {
auditOp->fClientID = fClientID;
Ops** opsLookup = fClientIDLookup.find(fClientID);
Ops* ops = nullptr;
if (!opsLookup) {
ops = new Ops;
fClientIDLookup.set(fClientID, ops);
} else {
ops = *opsLookup;
}
ops->push_back(auditOp);
}
// Our algorithm doesn't bother to reorder inside of an OpNode so the ChildID will start at 0
auditOp->fOpListID = fOpList.count();
auditOp->fChildID = 0;
// We use the op pointer as a key to find the OpNode we are 'glomming' ops onto
fIDLookup.set(op->uniqueID(), auditOp->fOpListID);
OpNode* opNode = new OpNode(renderTargetID);
opNode->fBounds = op->bounds();
opNode->fChildren.push_back(auditOp);
fOpList.emplace_back(opNode);
}
void GrAuditTrail::opsCombined(const GrOp* consumer, const GrOp* consumed) {
// Look up the op we are going to glom onto
int* indexPtr = fIDLookup.find(consumer->uniqueID());
SkASSERT(indexPtr);
int index = *indexPtr;
SkASSERT(index < fOpList.count() && fOpList[index]);
OpNode& consumerOp = *fOpList[index];
// Look up the op which will be glommed
int* consumedPtr = fIDLookup.find(consumed->uniqueID());
SkASSERT(consumedPtr);
int consumedIndex = *consumedPtr;
SkASSERT(consumedIndex < fOpList.count() && fOpList[consumedIndex]);
OpNode& consumedOp = *fOpList[consumedIndex];
// steal all of consumed's ops
for (int i = 0; i < consumedOp.fChildren.count(); i++) {
Op* childOp = consumedOp.fChildren[i];
// set the ids for the child op
childOp->fOpListID = index;
childOp->fChildID = consumerOp.fChildren.count();
consumerOp.fChildren.push_back(childOp);
}
// Update the bounds for the combineWith node
consumerOp.fBounds = consumer->bounds();
// remove the old node from our opList and clear the combinee's lookup
// NOTE: because we can't change the shape of the oplist, we use a sentinel
fOpList[consumedIndex].reset(nullptr);
fIDLookup.remove(consumed->uniqueID());
}
void GrAuditTrail::copyOutFromOpList(OpInfo* outOpInfo, int opListID) {
SkASSERT(opListID < fOpList.count());
const OpNode* bn = fOpList[opListID].get();
SkASSERT(bn);
outOpInfo->fBounds = bn->fBounds;
outOpInfo->fRenderTargetUniqueID = bn->fRenderTargetUniqueID;
for (int j = 0; j < bn->fChildren.count(); j++) {
OpInfo::Op& outOp = outOpInfo->fOps.push_back();
const Op* currentOp = bn->fChildren[j];
outOp.fBounds = currentOp->fBounds;
outOp.fClientID = currentOp->fClientID;
}
}
void GrAuditTrail::getBoundsByClientID(SkTArray<OpInfo>* outInfo, int clientID) {
Ops** opsLookup = fClientIDLookup.find(clientID);
if (opsLookup) {
// We track which oplistID we're currently looking at. If it changes, then we need to push
// back a new op info struct. We happen to know that ops are in sequential order in the
// oplist, otherwise we'd have to do more bookkeeping
int currentOpListID = kGrAuditTrailInvalidID;
for (int i = 0; i < (*opsLookup)->count(); i++) {
const Op* op = (**opsLookup)[i];
// Because we will copy out all of the ops associated with a given op list id everytime
// the id changes, we only have to update our struct when the id changes.
if (kGrAuditTrailInvalidID == currentOpListID || op->fOpListID != currentOpListID) {
OpInfo& outOpInfo = outInfo->push_back();
// copy out all of the ops so the client can display them even if they have a
// different clientID
this->copyOutFromOpList(&outOpInfo, op->fOpListID);
}
}
}
}
void GrAuditTrail::getBoundsByOpListID(OpInfo* outInfo, int opListID) {
this->copyOutFromOpList(outInfo, opListID);
}
void GrAuditTrail::fullReset() {
SkASSERT(fEnabled);
fOpList.reset();
fIDLookup.reset();
// free all client ops
fClientIDLookup.foreach ([](const int&, Ops** ops) { delete *ops; });
fClientIDLookup.reset();
fOpPool.reset(); // must be last, frees all of the memory
}
template <typename T>
void GrAuditTrail::JsonifyTArray(SkString* json, const char* name, const T& array,
bool addComma) {
if (array.count()) {
if (addComma) {
json->appendf(",");
}
json->appendf("\"%s\": [", name);
const char* separator = "";
for (int i = 0; i < array.count(); i++) {
// Handle sentinel nullptrs
if (array[i]) {
json->appendf("%s", separator);
json->append(array[i]->toJson());
separator = ",";
}
}
json->append("]");
}
}
// This will pretty print a very small subset of json
// The parsing rules are straightforward, aside from the fact that we do not want an extra newline
// before ',' and after '}', so we have a comma exception rule.
class PrettyPrintJson {
public:
SkString prettify(const SkString& json) {
fPrettyJson.reset();
fTabCount = 0;
fFreshLine = false;
fCommaException = false;
for (size_t i = 0; i < json.size(); i++) {
if ('[' == json[i] || '{' == json[i]) {
this->newline();
this->appendChar(json[i]);
fTabCount++;
this->newline();
} else if (']' == json[i] || '}' == json[i]) {
fTabCount--;
this->newline();
this->appendChar(json[i]);
fCommaException = true;
} else if (',' == json[i]) {
this->appendChar(json[i]);
this->newline();
} else {
this->appendChar(json[i]);
}
}
return fPrettyJson;
}
private:
void appendChar(char appendee) {
if (fCommaException && ',' != appendee) {
this->newline();
}
this->tab();
fPrettyJson += appendee;
fFreshLine = false;
fCommaException = false;
}
void tab() {
if (fFreshLine) {
for (int i = 0; i < fTabCount; i++) {
fPrettyJson += '\t';
}
}
}
void newline() {
if (!fFreshLine) {
fFreshLine = true;
fPrettyJson += '\n';
}
}
SkString fPrettyJson;
int fTabCount;
bool fFreshLine;
bool fCommaException;
};
static SkString pretty_print_json(SkString json) {
class PrettyPrintJson prettyPrintJson;
return prettyPrintJson.prettify(json);
}
SkString GrAuditTrail::toJson(bool prettyPrint) const {
SkString json;
json.append("{");
JsonifyTArray(&json, "Ops", fOpList, false);
json.append("}");
if (prettyPrint) {
return pretty_print_json(json);
} else {
return json;
}
}
SkString GrAuditTrail::toJson(int clientID, bool prettyPrint) const {
SkString json;
json.append("{");
Ops** ops = fClientIDLookup.find(clientID);
if (ops) {
JsonifyTArray(&json, "Ops", **ops, false);
}
json.appendf("}");
if (prettyPrint) {
return pretty_print_json(json);
} else {
return json;
}
}
static void skrect_to_json(SkString* json, const char* name, const SkRect& rect) {
json->appendf("\"%s\": {", name);
json->appendf("\"Left\": %f,", rect.fLeft);
json->appendf("\"Right\": %f,", rect.fRight);
json->appendf("\"Top\": %f,", rect.fTop);
json->appendf("\"Bottom\": %f", rect.fBottom);
json->append("}");
}
SkString GrAuditTrail::Op::toJson() const {
SkString json;
json.append("{");
json.appendf("\"Name\": \"%s\",", fName.c_str());
json.appendf("\"ClientID\": \"%d\",", fClientID);
json.appendf("\"OpListID\": \"%d\",", fOpListID);
json.appendf("\"ChildID\": \"%d\",", fChildID);
skrect_to_json(&json, "Bounds", fBounds);
if (fStackTrace.count()) {
json.append(",\"Stack\": [");
for (int i = 0; i < fStackTrace.count(); i++) {
json.appendf("\"%s\"", fStackTrace[i].c_str());
if (i < fStackTrace.count() - 1) {
json.append(",");
}
}
json.append("]");
}
json.append("}");
return json;
}
SkString GrAuditTrail::OpNode::toJson() const {
SkString json;
json.append("{");
json.appendf("\"RenderTarget\": \"%u\",", fRenderTargetUniqueID.asUInt());
skrect_to_json(&json, "Bounds", fBounds);
JsonifyTArray(&json, "Ops", fChildren, true);
json.append("}");
return json;
}