[AArch64] Enable PostRAScheduler for AArch64 generic build

[MinSeongKIM]

Summary

This patch enables PostRAScheduler specifically for AArch64 generic build,
which is beneficial from the performance perspective.
Speedups up to 2 to 7% for some benchmarks on A57 and A53 are observed.
Also benchmarks from LLVM test-suite did not regress.

[mcrosier]

Seems like a reasonable change to me. Did you happen to measure compile-time as well? IIRC, the post-RA scheduler can be expensive in terms of compile-time.

[aadg]

Beside the inherent compile time increase due to scheduling, the algorithm currently implemented exhibits exponential behavior which can be triggered by some real life programs; see https://llvm.org/bugs/show_bug.cgi?id=25794 for an example. There is some work being done in parallel by Jonas Paulsson to improve the scheduler: see http://reviews.llvm.org/D8705.

[mcrosier]

Arnaud,
I'm a proponent of this change going in. James provided a LGTM. IYO, should be wait for the ongoing improvements to the post-RA scheduler to be complete prior to this being commit or is this ready to land in your opinion? We might also solicit feedback from @t.p.northover.

Chda

[aadg]

I'm also in favor of this patch going in.

This is orthogonal to http://reviews.llvm.org/D8705. My point is really just a heads-up as it may create compile time regressions for people using generic cores, and which were not impacted by those regressions; this may shakes some bug out a bit more in the scheduler y broadening the user base (which is good) and give us a good incentive to have Jonas patch go in quicker as well.

So all in all, this patch is a good move forward.

[mcrosier]

LGT James, Arnaud and I.

[MinSeongKIM]

In D15557#311964, @mcrosier wrote:

Seems like a reasonable change to me. Did you happen to measure compile-time as well? IIRC, the post-RA scheduler can be expensive in terms of compile-time.

Many thanks for reviewing this patch, Chad. Based on the results from LLVM test-suite, the compile time (CC_Time) is increased by 11.6% on average when post-RA scheduler is on.

[MinSeongKIM]

In D15557#312006, @aadg wrote:

I'm also in favor of this patch going in.

This is orthogonal to http://reviews.llvm.org/D8705. My point is really just a heads-up as it may create compile time regressions for people using generic cores, and which were not impacted by those regressions; this may shakes some bug out a bit more in the scheduler y broadening the user base (which is good) and give us a good incentive to have Jonas patch go in quicker as well.

So all in all, this patch is a good move forward.

I appreciate your concern, Arnaud. I will keep this issue in mind.

[MinSeongKIM]

I am very thankful to you guys all for making this patch ready to land. I will commit this patch after getting the permission from Chris.

llvm/trunk/lib/Target/AArch64/AArch64Subtarget.h

public:
   }
   const AArch64InstrInfo *getInstrInfo() const override { return &InstrInfo; }
   const AArch64RegisterInfo *getRegisterInfo() const override {
     return &getInstrInfo()->getRegisterInfo();
   }
   const Triple &getTargetTriple() const { return TargetTriple; }
   bool enableMachineScheduler() const override { return true; }
   bool enablePostRAScheduler() const override {
-    return isCortexA53() || isCortexA57();
+    return isGeneric() || isCortexA53() || isCortexA57();
   }
 
   bool hasV8_1aOps() const { return HasV8_1aOps; }
   bool hasV8_2aOps() const { return HasV8_2aOps; }
 
   bool hasZeroCycleRegMove() const { return HasZeroCycleRegMove; }
 
   bool hasZeroCycleZeroing() const { return HasZeroCycleZeroing; }
public:
   bool isTargetLinux() const { return TargetTriple.isOSLinux(); }
   bool isTargetWindows() const { return TargetTriple.isOSWindows(); }
   bool isTargetAndroid() const { return TargetTriple.isAndroid(); }
 
   bool isTargetCOFF() const { return TargetTriple.isOSBinFormatCOFF(); }
   bool isTargetELF() const { return TargetTriple.isOSBinFormatELF(); }
   bool isTargetMachO() const { return TargetTriple.isOSBinFormatMachO(); }
 
+  bool isGeneric() const { return CPUString == "generic"; }
   bool isCyclone() const { return CPUString == "cyclone"; }
   bool isCortexA57() const { return CPUString == "cortex-a57"; }
   bool isCortexA53() const { return CPUString == "cortex-a53"; }
 
   bool useAA() const override { return isCortexA53(); }
 
   /// getMaxInlineSizeThreshold - Returns the maximum memset / memcpy size
   /// that still makes it profitable to inline the call.