[AArch64] Make machine combiner patterns preserve MIFlags

[john.brawn]

Summary

This is mainly done so that we don't lose the nofpexcept flag once we start emitting it.

[dmgreen]

Thanks for looking into this. Is the same or something similar needed in TargetInstrInfo::genAlternativeCodeSequence, for the commutative fp instructions?

[john.brawn]

Also preserve flags in TargetInstrInfo::reassociateOps (which is called from TargetInstrInfo::genAlternativeCodeSequence).

[dmgreen]

LGTM Thanks

llvm/lib/Target/AArch64/AArch64InstrInfo.cpp

genIndexedMultiply(Root, InsInstrs, IdxDupOp, AArch64::FMULv8i16_indexed,
                        &AArch64::FPR128_loRegClass, MRI);
     break;
   }
   } // end switch (Pattern)
   // Record MUL and ADD/SUB for deletion
   if (MUL)
     DelInstrs.push_back(MUL);
   DelInstrs.push_back(&Root);
+
+  // Set the flags on the inserted instructions to be the merged flags of the
+  // instructions that we have combined.
+  uint16_t Flags = Root.getFlags();
+  if (MUL)
+    Flags = Root.mergeFlagsWith(*MUL);
+  for (auto *MI : InsInstrs)
+    MI->setFlags(Flags);
 }
 
 /// Replace csincr-branch sequence by simple conditional branch
 ///
 /// Examples:
 /// 1. \code
 ///   csinc  w9, wzr, wzr, <condition code>
 ///   tbnz   w9, #0, 0x44

llvm/test/CodeGen/AArch64/machine-combiner-instr-fmf.mir

 # RUN: llc -run-pass=machine-combiner -o - -simplify-mir -mtriple=arm64-apple-iphoneos %s | FileCheck %s
 
 # Can create FMADD, because both the fmul and fadd have all fast-math flags.
 #
 # CHECK-LABEL: name: scalar_fmadd_fast
 # CHECK:        [[C:%.*]]:fpr32 = COPY $s2
 # CHECK-NEXT:   [[B:%.*]]:fpr32 = COPY $s1
 # CHECK-NEXT:   [[A:%.*]]:fpr32 = COPY $s0
-# CHECK-NEXT:   :fpr32 = FMADDSrrr [[B]], [[A]], [[C]]
+# CHECK-NEXT:   :fpr32 = nnan ninf nsz arcp contract afn reassoc FMADDSrrr [[B]], [[A]], [[C]]
 ---
 name:            scalar_fmadd_fast
 alignment:       4
 tracksRegLiveness: true
 registers:
   - { id: 0, class: fpr32 }
   - { id: 1, class: fpr32 }
   - { id: 2, class: fpr32 }

 ...
 
 # Can create FMADD, because both the fmul and fadd have the contract fast-math flag.
 #
 # CHECK-LABEL: name: scalar_fmadd_contract
 # CHECK:        [[C:%.*]]:fpr32 = COPY $s2
 # CHECK-NEXT:   [[B:%.*]]:fpr32 = COPY $s1
 # CHECK-NEXT:   [[A:%.*]]:fpr32 = COPY $s0
-# CHECK-NEXT:   :fpr32 = FMADDSrrr [[B]], [[A]], [[C]]
+# CHECK-NEXT:   :fpr32 = contract FMADDSrrr [[B]], [[A]], [[C]]
 
 ---
 name:            scalar_fmadd_contract
 alignment:       4
 tracksRegLiveness: true
 registers:
   - { id: 0, class: fpr32 }
   - { id: 1, class: fpr32 }

 ...
 
 # Do create FMADD, because we have the contract flag on the FADD.
 #
 # CHECK-LABEL: name: scalar_fmadd_contract_op1
 # CHECK:        [[C:%.*]]:fpr32 = COPY $s2
 # CHECK-NEXT:   [[B:%.*]]:fpr32 = COPY $s1
 # CHECK-NEXT:   [[A:%.*]]:fpr32 = COPY $s0
-# CHECK-NEXT:   :fpr32 = FMADDSrrr [[B]], [[A]], [[C]]
+# CHECK-NEXT:   :fpr32 = contract FMADDSrrr [[B]], [[A]], [[C]]
 
 ---
 name:            scalar_fmadd_contract_op1
 alignment:       4
 tracksRegLiveness: true
 registers:
   - { id: 0, class: fpr32 }
   - { id: 1, class: fpr32 }

 ...
 
 # Can create FMLA, because both the fmul and fadd have all fast-math flags.
 #
 # CHECK-LABEL: name: vector_fmadd_fast
 # CHECK:       [[C:%.*]]:fpr128 = COPY $q2
 # CHECK-NEXT:  [[B:%.*]]:fpr128 = COPY $q1
 # CHECK-NEXT:  [[A:%.*]]:fpr128 = COPY $q0
-# CHECK-NEXT:  fpr128 = FMLAv2f64 [[C]], [[B]], [[A]]
+# CHECK-NEXT:  fpr128 = nnan ninf nsz arcp contract afn reassoc FMLAv2f64 [[C]], [[B]], [[A]]
 ---
 name:            vector_fmadd_fast
 alignment:       4
 tracksRegLiveness: true
 registers:
   - { id: 0, class: fpr128 }
   - { id: 1, class: fpr128 }
   - { id: 2, class: fpr128 }

 ...
 
 # Can create FMLA, because both the fmul and fadd have the contract fast-math flag.
 #
 # CHECK-LABEL: name: vector_fmadd_contract
 # CHECK:       [[C:%.*]]:fpr128 = COPY $q2
 # CHECK-NEXT:  [[B:%.*]]:fpr128 = COPY $q1
 # CHECK-NEXT:  [[A:%.*]]:fpr128 = COPY $q0
-# CHECK-NEXT:  fpr128 = FMLAv2f64 [[C]], [[B]], [[A]]
+# CHECK-NEXT:  fpr128 = contract FMLAv2f64 [[C]], [[B]], [[A]]
 ---
 name:            vector_fmadd_contract
 alignment:       4
 tracksRegLiveness: true
 registers:
   - { id: 0, class: fpr128 }
   - { id: 1, class: fpr128 }
   - { id: 2, class: fpr128 }

 ...
 
 # Do create FMLA, because we have the contract flag on the FADD.
 #
 # CHECK-LABEL: name: vector_fmadd_contract_op1
 # CHECK:       [[C:%.*]]:fpr128 = COPY $q2
 # CHECK-NEXT:  [[B:%.*]]:fpr128 = COPY $q1
 # CHECK-NEXT:  [[A:%.*]]:fpr128 = COPY $q0
-# CHECK-NEXT:  fpr128 = FMLAv2f64 [[C]], [[B]], [[A]]
+# CHECK-NEXT:  fpr128 = contract FMLAv2f64 [[C]], [[B]], [[A]]
 
 ---
 name:            vector_fmadd_contract_op1
 alignment:       4
 tracksRegLiveness: true
 registers:
   - { id: 0, class: fpr128 }
   - { id: 1, class: fpr128 }