[ARM] Enable useAA() for the in-order Cortex-R52

[dmgreen]

Summary

This option allows codegen (such as DAGCombine or MI scheduling) to use alias analysis information, which can help with the codegen on in-order cpu's. Here I have done things the same way as AArch64, adding a subtarget feature to enable this for specific cores.

I was going to enable this for A53 too, but seeing as we happen to not have a AArch32 A53 schedule, the usefulness is not as high as R52.

[dmgreen]

Requires D48029 to survives a bootstrap, but that looks like a more generic error than having to use this option. Otherwise I believe this is safe.

[javed.absar]

LGTM but will wait for others to comment as well before accepting

[dmgreen]

Thanks

[efriedma]

I'm generally not a fan of having features like this, which have widespread implications, turned on only for certain target CPUs; it tends to make it much harder to find bugs, since the code gets little testing. But I guess this is okay for now.

[dmgreen]

Yes I can see that. I would have liked to turn this on for more in-order cores, but without scheduling enough to at least say that a load takes multiple cycles, I didn't feel I had a great justification. For the record, these were the changes I saw on a A53 with useAA returning true (units are time, so lower is better. these are more than 2%):

SingleSource/Benchmarks/BenchmarkGame/n-body -14.38%
SingleSource/Benchmarks/Shootout/Shootout-lists -6.40%
SingleSource/Benchmarks/Misc-C++/Large/ray -6.20%
MultiSource/Applications/ALAC/encode/alacconvert-encode -5.44%
MultiSource/Benchmarks/McCat/17-bintr/bintr -3.27%
SingleSource/Benchmarks/CoyoteBench/huffbench -3.15%
MultiSource/Benchmarks/SciMark2-C/scimark2 -2.97%
MultiSource/Benchmarks/Bullet/bullet -2.50%
MultiSource/Benchmarks/TSVC/LoopRerolling-flt/LoopRerolling-flt -2.33%
SingleSource/Benchmarks/Misc/richards_benchmark -2.20%
MultiSource/Benchmarks/Ptrdist/yacr2/yacr2 +4.60%
MultiSource/Benchmarks/Trimaran/enc-pc1/enc-pc1 +9.66%

They don't look too bad, but there are some decreases. enc-pc1 is genuinely worse, yacr2 might be noise. And without instruction scheduling, they may be getting lucky. Compile time increase was roughly 0.25% on CT-mark (may not be statistically significant, but it was enough alternating runs to make me think it's probably close).

I tried it on the A72 too, on both T32 and A64, with more varied results, both showing several large increases in places (including a memcpy benchmark). This option, as far as I can tell, should give more freedom to the DAG, but that may not be used in the best way all the time.

lib/Target/ARM/ARM.td

 // Use the MachineScheduler for instruction scheduling for the subtarget.
 def FeatureUseMISched: SubtargetFeature<"use-misched", "UseMISched", "true",
                                         "Use the MachineScheduler">;
 
 def FeatureNoPostRASched : SubtargetFeature<"disable-postra-scheduler",
     "DisablePostRAScheduler", "true",
     "Don't schedule again after register allocation">;
 
+// Enable use of alias analysis during code generation
+def FeatureUseAA : SubtargetFeature<"use-aa", "UseAA", "true",
+                                    "Use alias analysis during codegen">;
+
 //===----------------------------------------------------------------------===//
 // ARM architecture class
 //
 
 // A-series ISA
 def FeatureAClass : SubtargetFeature<"aclass", "ARMProcClass", "AClass",
                                      "Is application profile ('A' series)">;
 

 def : ProcNoItin<"kryo",                                [ARMv8a, ProcKryo,
                                                          FeatureHWDivThumb,
                                                          FeatureHWDivARM,
                                                          FeatureCrypto,
                                                          FeatureCRC]>;
 
 def : ProcessorModel<"cortex-r52", CortexR52Model,      [ARMv8r, ProcR52,
                                                          FeatureUseMISched,
-                                                         FeatureFPAO]>;
+                                                         FeatureFPAO,
+                                                         FeatureUseAA]>;
 
 //===----------------------------------------------------------------------===//
 // Register File Description
 //===----------------------------------------------------------------------===//
 
 include "ARMRegisterInfo.td"
 include "ARMRegisterBanks.td"
 include "ARMCallingConv.td"

lib/Target/ARM/ARMSubtarget.h

protected:
 
   /// UseMISched - True if MachineScheduler should be used for this subtarget.
   bool UseMISched = false;
 
   /// DisablePostRAScheduler - False if scheduling should happen again after
   /// register allocation.
   bool DisablePostRAScheduler = false;
 
+  /// UseAA - True if using AA during codegen (DAGCombine, MISched, etc)
+  bool UseAA = false;
+
   /// HasThumb2 - True if Thumb2 instructions are supported.
   bool HasThumb2 = false;
 
   /// NoARM - True if subtarget does not support ARM mode execution.
   bool NoARM = false;
 
   /// ReserveR9 - True if R9 is not available as a general purpose register.
   bool ReserveR9 = false;
public:
   unsigned getMispredictionPenalty() const;
 
   /// Returns true if machine scheduler should be enabled.
   bool enableMachineScheduler() const override;
 
   /// True for some subtargets at > -O0.
   bool enablePostRAScheduler() const override;
 
+  /// Enable use of alias analysis during code generation (during MI
+  /// scheduling, DAGCombine, etc.).
+  bool useAA() const override { return UseAA; }
+
   // enableAtomicExpand- True if we need to expand our atomics.
   bool enableAtomicExpand() const override;
 
   /// getInstrItins - Return the instruction itineraries based on subtarget
   /// selection.
   const InstrItineraryData *getInstrItineraryData() const override {
     return &InstrItins;
   }

test/CodeGen/ARM/useaa.ll

+; RUN: llc < %s -mtriple=armv8r-eabi -mcpu=cortex-r52 | FileCheck %s --check-prefix=CHECK --check-prefix=USEAA
+; RUN: llc < %s -mtriple=armv8r-eabi -mcpu=generic | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
+
+; Check we use AA during codegen, so can interleave these loads/stores.
+
+; CHECK-LABEL: test
+; GENERIC: ldr
+; GENERIC: str
+; GENERIC: ldr
+; GENERIC: str
+; USEAA: ldr
+; USEAA: ldr
+; USEAA: str
+; USEAA: str
+
+define void @test(i32* nocapture %a, i32* noalias nocapture %b) {
+entry:
+  %0 = load i32, i32* %a, align 4
+  %add = add nsw i32 %0, 10
+  store i32 %add, i32* %a, align 4
+  %1 = load i32, i32* %b, align 4
+  %add2 = add nsw i32 %1, 20
+  store i32 %add2, i32* %b, align 4
+  ret void
+}
+