Dataflow analysis rework
Rework key portions of the dataflow analysis code to avoid the
massive slowdowns we're seeing with methods containing more than
20,000 basic blocks.
Add a somewhat hacky lookup cache in front of findBlock. Later
we'll want to rework the "GrowableList" mechanism to incorporate
this properly.
Add memory to the bit vector utilities to speed up the iterators
for large sparse vectors. (Similarly, in the future we'll want
to include a rewrite of the home-grown bit-vector manipluation
utilites to be better with large ones).
With this CL, compilation speed on the degenerate cases is > 10x
faster. Compilation of typically-sized methods will see a smallish
improvement.
Change-Id: I7f086c1229dd4fe62f0a5fc361234bf204ebc2b1
diff --git a/src/compiler/Frontend.cc b/src/compiler/Frontend.cc
index 8070e6f..1b1215a 100644
--- a/src/compiler/Frontend.cc
+++ b/src/compiler/Frontend.cc
@@ -44,6 +44,7 @@
//(1 << kDebugSlowestFieldPath) |
//(1 << kDebugSlowestStringPath) |
//(1 << kDebugExerciseResolveMethod) |
+ //(1 << kDebugVerifyDataflow) |
0;
std::string compilerMethodMatch; // Method name match to apply above flags
@@ -136,6 +137,10 @@
bottomBlock->firstMIRInsn = insn;
bottomBlock->lastMIRInsn = origBlock->lastMIRInsn;
+ /* Add it to the quick lookup cache */
+ cUnit->blockMap.insert(std::make_pair(bottomBlock->startOffset,
+ bottomBlock));
+
/* Handle the taken path */
bottomBlock->taken = origBlock->taken;
if (bottomBlock->taken) {
@@ -196,6 +201,7 @@
* is not non-null and is the block being split, update *immedPredBlockP to
* point to the bottom block so that outgoing edges can be set up properly
* (by the caller)
+ * Utilizes a map for fast lookup of the typical cases.
*/
STATIC BasicBlock *findBlock(CompilationUnit* cUnit,
unsigned int codeOffset,
@@ -205,28 +211,36 @@
GrowableList* blockList = &cUnit->blockList;
BasicBlock* bb;
unsigned int i;
+ std::map<unsigned int, BasicBlock*>::iterator it;
- for (i = 0; i < blockList->numUsed; i++) {
- bb = (BasicBlock *) blockList->elemList[i];
- if (bb->blockType != kDalvikByteCode) continue;
- if (bb->startOffset == codeOffset) return bb;
- /* Check if a branch jumps into the middle of an existing block */
- if ((split == true) && (codeOffset > bb->startOffset) &&
- (bb->lastMIRInsn != NULL) &&
- (codeOffset <= bb->lastMIRInsn->offset)) {
- BasicBlock *newBB = splitBlock(cUnit, codeOffset, bb,
- bb == *immedPredBlockP ?
- immedPredBlockP : NULL);
- return newBB;
+ it = cUnit->blockMap.find(codeOffset);
+ if (it != cUnit->blockMap.end()) {
+ return it->second;
+ } else if (!create) {
+ return NULL;
+ }
+
+ if (split) {
+ for (i = 0; i < blockList->numUsed; i++) {
+ bb = (BasicBlock *) blockList->elemList[i];
+ if (bb->blockType != kDalvikByteCode) continue;
+ /* Check if a branch jumps into the middle of an existing block */
+ if ((codeOffset > bb->startOffset) && (bb->lastMIRInsn != NULL) &&
+ (codeOffset <= bb->lastMIRInsn->offset)) {
+ BasicBlock *newBB = splitBlock(cUnit, codeOffset, bb,
+ bb == *immedPredBlockP ?
+ immedPredBlockP : NULL);
+ return newBB;
+ }
}
}
- if (create) {
- bb = oatNewBB(kDalvikByteCode, cUnit->numBlocks++);
- oatInsertGrowableList(&cUnit->blockList, (intptr_t) bb);
- bb->startOffset = codeOffset;
- return bb;
- }
- return NULL;
+
+ /* Create a new one */
+ bb = oatNewBB(kDalvikByteCode, cUnit->numBlocks++);
+ oatInsertGrowableList(&cUnit->blockList, (intptr_t) bb);
+ bb->startOffset = codeOffset;
+ cUnit->blockMap.insert(std::make_pair(bb->startOffset, bb));
+ return bb;
}
/* Dump the CFG into a DOT graph */
@@ -748,6 +762,8 @@
cUnit->numOuts = code_item->outs_size_;
/* Adjust this value accordingly once inlining is performed */
cUnit->numDalvikRegisters = code_item->registers_size_;
+ cUnit->blockMap = std::map<unsigned int, BasicBlock*>();
+ cUnit->blockMap.clear();
bool useMatch = compilerMethodMatch.length() != 0;
bool match = useMatch && (compilerFlipMatch ^
(PrettyMethod(method_idx, dex_file).find(compilerMethodMatch) != std::string::npos));
@@ -794,6 +810,8 @@
BasicBlock *curBlock = oatNewBB(kDalvikByteCode, numBlocks++);
curBlock->startOffset = 0;
oatInsertGrowableList(&cUnit->blockList, (intptr_t) curBlock);
+ /* Add first block to the fast lookup cache */
+ cUnit->blockMap.insert(std::make_pair(curBlock->startOffset, curBlock));
entryBlock->fallThrough = curBlock;
oatSetBit(curBlock->predecessors, entryBlock->id);
@@ -885,8 +903,10 @@
oatDumpCompilationUnit(cUnit.get());
}
- /* Verify if all blocks are connected as claimed */
- oatDataFlowAnalysisDispatcher(cUnit.get(), verifyPredInfo, kAllNodes, false /* isIterative */);
+ if (cUnit->enableDebug & (1 << kDebugVerifyDataflow)) {
+ /* Verify if all blocks are connected as claimed */
+ oatDataFlowAnalysisDispatcher(cUnit.get(), verifyPredInfo, kAllNodes, false /* isIterative */);
+ }
/* Perform SSA transformation for the whole method */
oatMethodSSATransformation(cUnit.get());