[X86] Add option to specify preferable loop alignment

[skatkov]

Summary

Loop alignment can cause a significant change of
the perfromance for short loops.
To be able to evaluate the impact of loop alignment this change
introduces the new option x86-experimental-pref-loop-alignment.
The alignment will be 2^Value bytes, the default value is 4.

[craig.topper]

lib/Target/X86/X86ISelLowering.cpp

75

prefereable should be spelled preferable.

Probably shouldn't use "ExperimentalPrefLoopAlignment" in the description since that's an variable name and this text is printed to the command line help.

[skatkov]

Craig, Thank you for the review.

I've updated a change.

[skatkov]

Ooops, bad change. Will upload a new one.

[skatkov]

This one should be better. make check passed.

[sanjoy]

lib/Target/X86/X86ISelLowering.cpp

76

I think what @craig.topper wanted to say is that the description should be "the last x86-pref-loop-alignment bits of the ..."

[craig.topper]

LGTM

lib/Target/X86/X86ISelLowering.cpp

 STATISTIC(NumTailCalls, "Number of tail calls");
 
 static cl::opt<bool> ExperimentalVectorWideningLegalization(
     "x86-experimental-vector-widening-legalization", cl::init(false),
     cl::desc("Enable an experimental vector type legalization through widening "
              "rather than promotion."),
     cl::Hidden);
 
+static cl::opt<int> ExperimentalPrefLoopAlignment(
+    "x86-experimental-pref-loop-alignment", cl::init(4),
+    cl::desc("Sets the prefereable loop alignment for experiments "
+             "(the last ExperimentalPrefLoopAlignment bits"
+             " of the loop header PC will be 0)."),
+    cl::Hidden);
+
 X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
                                      const X86Subtarget &STI)
     : TargetLowering(TM), Subtarget(STI) {
   bool UseX87 = !Subtarget.useSoftFloat() && Subtarget.hasX87();
   X86ScalarSSEf64 = Subtarget.hasSSE2();
   X86ScalarSSEf32 = Subtarget.hasSSE1();
   MVT PtrVT = MVT::getIntegerVT(8 * TM.getPointerSize());
 

   computeRegisterProperties(Subtarget.getRegisterInfo());
 
   MaxStoresPerMemset = 16; // For @llvm.memset -> sequence of stores
   MaxStoresPerMemsetOptSize = 8;
   MaxStoresPerMemcpy = 8; // For @llvm.memcpy -> sequence of stores
   MaxStoresPerMemcpyOptSize = 4;
   MaxStoresPerMemmove = 8; // For @llvm.memmove -> sequence of stores
   MaxStoresPerMemmoveOptSize = 4;
-  setPrefLoopAlignment(4); // 2^4 bytes.
+  // Set loop alignment to 2^ExperimentalPrefLoopAlignment bytes (default: 2^4).
+  setPrefLoopAlignment(ExperimentalPrefLoopAlignment);
 
   // An out-of-order CPU can speculatively execute past a predictable branch,
   // but a conditional move could be stalled by an expensive earlier operation.
   PredictableSelectIsExpensive = Subtarget.getSchedModel().isOutOfOrder();
   EnableExtLdPromotion = true;
   setPrefFunctionAlignment(4); // 2^4 bytes.
 
   verifyIntrinsicTables();