[LoopVectorizer][AArch64] Add a -sve-vscale-for-tuning override option. (llvm#156916)

It can be useful for debugging and tuning to be able to alter the
VScaleForTuning. This adds a quick option to the aarch64 subtarget for
altering it.

Author: davemgreen

llvm/lib/Target/AArch64/AArch64Subtarget.cpp

@@ -91,6 +91,10 @@ static cl::opt<bool> EnableZPRPredicateSpills(
     cl::desc(
         "Enables spilling/reloading SVE predicates as data vectors (ZPRs)"));
 
+static cl::opt<unsigned>
+    VScaleForTuningOpt("sve-vscale-for-tuning", cl::Hidden,
+                       cl::desc("Force a vscale for tuning factor for SVE"));
+
 // Subreg liveness tracking is disabled by default for now until all issues
 // are ironed out. This option allows the feature to be used in tests.
 static cl::opt<bool>
@@ -364,6 +368,8 @@ void AArch64Subtarget::initializeProperties(bool HasMinSize) {
 
   if (AArch64MinimumJumpTableEntries.getNumOccurrences() > 0 || !HasMinSize)
     MinimumJumpTableEntries = AArch64MinimumJumpTableEntries;
+  if (VScaleForTuningOpt.getNumOccurrences() > 0)
+    VScaleForTuning = VScaleForTuningOpt;
 }
 
 AArch64Subtarget::AArch64Subtarget(const Triple &TT, StringRef CPU,

llvm/test/Transforms/LoopVectorize/AArch64/scalable-vectorization-cost-tuning.ll

@@ -7,6 +7,10 @@
 ; RUN:     -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \
 ; RUN:     | FileCheck %s --check-prefixes=VSCALEFORTUNING1
 
+; RUN: opt -mtriple=aarch64 -mattr=+sve -mcpu=generic -sve-vscale-for-tuning=2 \
+; RUN:     -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \
+; RUN:     | FileCheck %s --check-prefixes=VSCALEFORTUNING2
+
 ; RUN: opt -mtriple=aarch64 -mcpu=neoverse-v1 \
 ; RUN:     -force-target-instruction-cost=1 -passes=loop-vectorize -S -debug-only=loop-vectorize --disable-output < %s 2>&1 \
 ; RUN:     | FileCheck %s --check-prefixes=VSCALEFORTUNING2