Foundation for Quick LLVM compiler
Ready for review - probably better get this cleaned up and
checked in even though much work remains.
Basic conversion from MIR to GreenlandIR and from GreenlandIR
back to LIR. Support sufficient to run Fibonacci test.
Note some structural changes in MIR to support this work:
o retaining incoming label for phi nodes
o constant propagation
o include object reference detection in type inference pass
Change-Id: I8ba63c73e76d071aa40cae0f744e598b96f68699
diff --git a/src/compiler/Frontend.cc b/src/compiler/Frontend.cc
index a764583..70de4fc 100644
--- a/src/compiler/Frontend.cc
+++ b/src/compiler/Frontend.cc
@@ -712,7 +712,7 @@
* THROW_VERIFICATION_ERROR is also an unconditional
* branch, but we shouldn't treat it as such until we have
* a dead code elimination pass (which won't be important
- * until inlining w/ constant propogation is implemented.
+ * until inlining w/ constant propagation is implemented.
*/
if (insn->dalvikInsn.opcode != Instruction::THROW) {
curBlock->fallThrough = fallthroughBlock;
@@ -764,6 +764,9 @@
cUnit->numIns = code_item->ins_size_;
cUnit->numRegs = code_item->registers_size_ - cUnit->numIns;
cUnit->numOuts = code_item->outs_size_;
+#if defined(ART_USE_QUICK_COMPILER)
+ cUnit->genBitcode = PrettyMethod(method_idx, dex_file).find("Fibonacci.fibonacci") != std::string::npos;
+#endif
/* Adjust this value accordingly once inlining is performed */
cUnit->numDalvikRegisters = code_item->registers_size_;
// TODO: set this from command line
@@ -778,6 +781,9 @@
cUnit->printMe = VLOG_IS_ON(compiler) ||
(cUnit->enableDebug & (1 << kDebugVerbose));
}
+#if defined(ART_USE_QUICK_COMPILER)
+ if (cUnit->genBitcode) cUnit->printMe = true;
+#endif
if (cUnit->instructionSet == kX86) {
// Disable optimizations on X86 for now
cUnit->disableOpt = -1;
@@ -985,6 +991,13 @@
}
}
+#if defined(ART_USE_QUICK_COMPILER)
+ if (cUnit->genBitcode) {
+ // Bitcode generation requires full dataflow analysis, no qdMode
+ cUnit->qdMode = false;
+ }
+#endif
+
if (cUnit->qdMode) {
cUnit->disableDataflow = true;
// Disable optimization which require dataflow/ssa
@@ -1012,6 +1025,18 @@
/* Perform SSA transformation for the whole method */
oatMethodSSATransformation(cUnit.get());
+ /* Do constant propagation */
+ // TODO: Probably need to make these expandable to support new ssa names
+ // introducted during MIR optimization passes
+ cUnit->isConstantV = oatAllocBitVector(cUnit.get(), cUnit->numSSARegs,
+ false /* not expandable */);
+ cUnit->constantValues =
+ (int*)oatNew(cUnit.get(), sizeof(int) * cUnit->numSSARegs, true,
+ kAllocDFInfo);
+ oatDataFlowAnalysisDispatcher(cUnit.get(), oatDoConstantPropagation,
+ kAllNodes,
+ false /* isIterative */);
+
/* Detect loops */
oatMethodLoopDetection(cUnit.get());
@@ -1029,19 +1054,31 @@
/* Allocate Registers using simple local allocation scheme */
oatSimpleRegAlloc(cUnit.get());
- if (specialCase != kNoHandler) {
- /*
- * Custom codegen for special cases. If for any reason the
- * special codegen doesn't success, cUnit->firstLIRInsn will
- * set to NULL;
- */
- oatSpecialMIR2LIR(cUnit.get(), specialCase);
- }
+#if defined(ART_USE_QUICK_COMPILER)
+ /* Go the LLVM path? */
+ if (cUnit->genBitcode) {
+ // MIR->Bitcode
+ oatMethodMIR2Bitcode(cUnit.get());
+ // Bitcode->LIR
+ oatMethodBitcode2LIR(cUnit.get());
+ } else {
+#endif
+ if (specialCase != kNoHandler) {
+ /*
+ * Custom codegen for special cases. If for any reason the
+ * special codegen doesn't succeed, cUnit->firstLIRInsn will
+ * set to NULL;
+ */
+ oatSpecialMIR2LIR(cUnit.get(), specialCase);
+ }
- /* Convert MIR to LIR, etc. */
- if (cUnit->firstLIRInsn == NULL) {
- oatMethodMIR2LIR(cUnit.get());
+ /* Convert MIR to LIR, etc. */
+ if (cUnit->firstLIRInsn == NULL) {
+ oatMethodMIR2LIR(cUnit.get());
+ }
+#if defined(ART_USE_QUICK_COMPILER)
}
+#endif
// Debugging only
if (cUnit->enableDebug & (1 << kDebugDumpCFG)) {