Adding an A15 specific optimization pass for interactions between S/D/Q registers...

[oota-llvm.git] / lib / Target / ARM / ARMTargetMachine.cpp

diff --git a/lib/Target/ARM/ARMTargetMachine.cpp b/lib/Target/ARM/ARMTargetMachine.cpp
index 774521852a17c250bd461de05dbe8fe4fdbe4907..3003760df736d853640f3571ec45823128d42d8c 100644 (file)
--- a/lib/Target/ARM/ARMTargetMachine.cpp
+++ b/lib/Target/ARM/ARMTargetMachine.cpp
@@ -28,6 +28,11 @@ EnableGlobalMerge("global-merge", cl::Hidden,
                   cl::desc("Enable global merge pass"),
                   cl::init(true));
 
+static cl::opt<bool>
+DisableA15SDOptimization("disable-a15-sd-optimization", cl::Hidden,
+                   cl::desc("Inhibit optimization of S->D register accesses on A15"),
+                   cl::init(false));
+
 extern "C" void LLVMInitializeARMTarget() {
   // Register the target.
   RegisterTargetMachine<ARMTargetMachine> X(TheARMTarget);
@@ -164,6 +169,12 @@ bool ARMPassConfig::addPreRegAlloc() {
     addPass(createARMLoadStoreOptimizationPass(true));
   if (getOptLevel() != CodeGenOpt::None && getARMSubtarget().isLikeA9())
     addPass(createMLxExpansionPass());
+  // Since the A15SDOptimizer pass can insert VDUP instructions, it can only be
+  // enabled when NEON is available.
+  if (getOptLevel() != CodeGenOpt::None && getARMSubtarget().isCortexA15() &&
+    getARMSubtarget().hasNEON() && !DisableA15SDOptimization) {
+    addPass(createA15SDOptimizerPass());
+  }
   return true;
 }
 
@@ -174,7 +185,8 @@ bool ARMPassConfig::addPreSched2() {
       addPass(createARMLoadStoreOptimizationPass());
       printAndVerify("After ARM load / store optimizer");
     }
-    if (getARMSubtarget().hasNEON())
+    if ((DisableA15SDOptimization || !getARMSubtarget().isCortexA15()) &&
+      getARMSubtarget().hasNEON())
       addPass(createExecutionDependencyFixPass(&ARM::DPRRegClass));
   }