Optimization fixes

Two primary fixes.  First, the save/restore mechanism for FP callee saves
was broken if there were any holes in the save mask (the Arm ld/store
multiple instructions for floating point use a start + count mechanism,
rather than the bit-mask mechanism used for core registers).

The second fix corrects a problem introduced by the recent enhancements
to loading floating point literals.  The load->copy optimization mechanism
for literal loads used the value of the loaded literal to identify
redundant loads.  However, it used only the first 32 bits of the
literal - which worked fine previously because 64-bit literal loads
were treated as a pair of 32-bit loads.  The fix was to use the
label of the literal rather than the value in the aliasInfo - which
works for all sizes.

Change-Id: Ic4779adf73b2c7d80059a988b0ecdef39921a81f
diff --git a/src/compiler/codegen/arm/ArchUtility.cc b/src/compiler/codegen/arm/ArchUtility.cc
index 3ceffae..edce114 100644
--- a/src/compiler/codegen/arm/ArchUtility.cc
+++ b/src/compiler/codegen/arm/ArchUtility.cc
@@ -404,7 +404,8 @@
     LOG(INFO) << "Regs (excluding ins) : " << cUnit->numRegs;
     LOG(INFO) << "Ins                  : " << cUnit->numIns;
     LOG(INFO) << "Outs                 : " << cUnit->numOuts;
-    LOG(INFO) << "Spills               : " << cUnit->numSpills;
+    LOG(INFO) << "CoreSpills           : " << cUnit->numCoreSpills;
+    LOG(INFO) << "FPSpills             : " << cUnit->numFPSpills;
     LOG(INFO) << "Padding              : " << cUnit->numPadding;
     LOG(INFO) << "Frame size           : " << cUnit->frameSize;
     LOG(INFO) << "Start of ins         : " << cUnit->insOffset;
diff --git a/src/compiler/codegen/arm/ArmLIR.h b/src/compiler/codegen/arm/ArmLIR.h
index 3b2e986..99b22e7 100644
--- a/src/compiler/codegen/arm/ArmLIR.h
+++ b/src/compiler/codegen/arm/ArmLIR.h
@@ -99,6 +99,8 @@
 #define FP_REG_OFFSET 32
 /* Offset to distinguish DP FP regs */
 #define FP_DOUBLE 64
+/* First FP callee save */
+#define FP_CALLEE_SAVE_BASE 16
 /* Reg types */
 #define REGTYPE(x) (x & (FP_REG_OFFSET | FP_DOUBLE))
 #define FPREG(x) ((x & FP_REG_OFFSET) == FP_REG_OFFSET)
diff --git a/src/compiler/codegen/arm/ArmRallocUtil.cc b/src/compiler/codegen/arm/ArmRallocUtil.cc
index 84c3792..9a7c642 100644
--- a/src/compiler/codegen/arm/ArmRallocUtil.cc
+++ b/src/compiler/codegen/arm/ArmRallocUtil.cc
@@ -136,7 +136,7 @@
      * machinery is in place, always spill lr.
      */
     cUnit->coreSpillMask |= (1 << rLR);
-    cUnit->numSpills++;
+    cUnit->numCoreSpills++;
     /*
      * Simple hack for testing register allocation.  Just do a static
      * count of the uses of Dalvik registers.  Note that we examine
diff --git a/src/compiler/codegen/arm/MethodCodegenDriver.cc b/src/compiler/codegen/arm/MethodCodegenDriver.cc
index 80118d8..a4e211b 100644
--- a/src/compiler/codegen/arm/MethodCodegenDriver.cc
+++ b/src/compiler/codegen/arm/MethodCodegenDriver.cc
@@ -1853,6 +1853,7 @@
 
     ArmLIR* headLIR = NULL;
 
+    int spillCount = cUnit->numCoreSpills + cUnit->numFPSpills;
     if (bb->blockType == kEntryBlock) {
         /*
          * On entry, r0, r1, r2 & r3 are live.  Let the register allocation
@@ -1882,17 +1883,22 @@
         newLIR1(cUnit, kThumb2Push, cUnit->coreSpillMask);
         /* Need to spill any FP regs? */
         if (cUnit->numFPSpills) {
+            /*
+             * NOTE: fp spills are a little different from core spills in that
+             * they are pushed as a contiguous block.  When promoting from
+             * the fp set, we must allocate all singles from s16..highest-promoted
+             */
             newLIR1(cUnit, kThumb2VPushCS, cUnit->numFPSpills);
         }
         if (!skipOverflowCheck) {
             opRegRegImm(cUnit, kOpSub, rLR, rSP,
-                        cUnit->frameSize - (cUnit->numSpills * 4));
+                        cUnit->frameSize - (spillCount * 4));
             genRegRegCheck(cUnit, kArmCondCc, rLR, r12, NULL,
                            kArmThrowStackOverflow);
             genRegCopy(cUnit, rSP, rLR);         // Establish stack
         } else {
             opRegImm(cUnit, kOpSub, rSP,
-                     cUnit->frameSize - (cUnit->numSpills * 4));
+                     cUnit->frameSize - (spillCount * 4));
         }
         storeBaseDisp(cUnit, rSP, 0, r0, kWord);
         flushIns(cUnit);
@@ -1902,7 +1908,7 @@
         oatFreeTemp(cUnit, r3);
     } else if (bb->blockType == kExitBlock) {
         newLIR0(cUnit, kArmPseudoMethodExit);
-        opRegImm(cUnit, kOpAdd, rSP, cUnit->frameSize - (cUnit->numSpills * 4));
+        opRegImm(cUnit, kOpAdd, rSP, cUnit->frameSize - (spillCount * 4));
         /* Need to restore any FP callee saves? */
         if (cUnit->numFPSpills) {
             newLIR1(cUnit, kThumb2VPopCS, cUnit->numFPSpills);
@@ -2121,7 +2127,8 @@
                 funcOffset =
                     OFFSETOF_MEMBER(Thread, pThrowStackOverflowFromCode);
                 // Restore stack alignment
-                opRegImm(cUnit, kOpAdd, rSP, cUnit->numSpills * 4);
+                opRegImm(cUnit, kOpAdd, rSP,
+                         (cUnit->numCoreSpills + cUnit->numFPSpills) * 4);
                 break;
             default:
                 LOG(FATAL) << "Unexpected throw kind: " << lab->operands[0];
diff --git a/src/compiler/codegen/arm/Thumb2/Factory.cc b/src/compiler/codegen/arm/Thumb2/Factory.cc
index 3ee23ea..45c7377 100644
--- a/src/compiler/codegen/arm/Thumb2/Factory.cc
+++ b/src/compiler/codegen/arm/Thumb2/Factory.cc
@@ -75,7 +75,7 @@
     loadPcRel->operands[1] = r15pc;
     setupResourceMasks(loadPcRel);
     setMemRefType(loadPcRel, true, kLiteral);
-    loadPcRel->aliasInfo = dataTarget->operands[0];
+    loadPcRel->aliasInfo = (intptr_t)dataTarget;
     oatAppendLIR(cUnit, (LIR* ) loadPcRel);
     return loadPcRel;
 }
@@ -183,7 +183,7 @@
     loadPcRel->operands[0] = rDest;
     setupResourceMasks(loadPcRel);
     setMemRefType(loadPcRel, true, kLiteral);
-    loadPcRel->aliasInfo = dataTarget->operands[0];
+    loadPcRel->aliasInfo = (intptr_t)dataTarget;
     res = loadPcRel;
     oatAppendLIR(cUnit, (LIR* ) loadPcRel);
 
@@ -661,7 +661,7 @@
             loadPcRel->operands[1] = r15pc;
             setupResourceMasks(loadPcRel);
             setMemRefType(loadPcRel, true, kLiteral);
-            loadPcRel->aliasInfo = dataTarget->operands[0];
+            loadPcRel->aliasInfo = (intptr_t)dataTarget;
             oatAppendLIR(cUnit, (LIR* ) loadPcRel);
             res = loadPcRel;
         }