[AArch64][GlobalISel] Implement widenScalar for signed overflow

[porglezomp]

Summary

Implement widening for G_SADDO and G_SSUBO. Previously it was only
implemented for G_UADDO and G_USUBO. Also add legalize-add/sub tests for
narrow overflowing add/sub on AArch64.

[paquette]

LGTM

[porglezomp]

The buildbot with the UBSan tests passed for the first commit in this stack, so I think this should be safe to re-land http://lab.llvm.org:8011/#/builders/77/builds/3112

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
   case TargetOpcode::G_EXTRACT:
     return widenScalarExtract(MI, TypeIdx, WideTy);
   case TargetOpcode::G_INSERT:
     return widenScalarInsert(MI, TypeIdx, WideTy);
   case TargetOpcode::G_MERGE_VALUES:
     return widenScalarMergeValues(MI, TypeIdx, WideTy);
   case TargetOpcode::G_UNMERGE_VALUES:
     return widenScalarUnmergeValues(MI, TypeIdx, WideTy);
+  case TargetOpcode::G_SADDO:
+  case TargetOpcode::G_SSUBO: {
+    if (TypeIdx == 1)
+      return UnableToLegalize; // TODO
+    auto LHSExt = MIRBuilder.buildSExt(WideTy, MI.getOperand(2));
+    auto RHSExt = MIRBuilder.buildSExt(WideTy, MI.getOperand(3));
+    unsigned Opcode = MI.getOpcode() == TargetOpcode::G_SADDO
+                          ? TargetOpcode::G_ADD
+                          : TargetOpcode::G_SUB;
+    auto NewOp = MIRBuilder.buildInstr(Opcode, {WideTy}, {LHSExt, RHSExt});
+    LLT OrigTy = MRI.getType(MI.getOperand(0).getReg());
+    auto TruncOp = MIRBuilder.buildTrunc(OrigTy, NewOp);
+    auto ExtOp = MIRBuilder.buildSExt(WideTy, TruncOp);
+    // There is no overflow if the re-extended result is the same as NewOp.
+    MIRBuilder.buildICmp(CmpInst::ICMP_NE, MI.getOperand(1), NewOp, ExtOp);
+    // Now trunc the NewOp to the original result.
+    MIRBuilder.buildTrunc(MI.getOperand(0), NewOp);
+    MI.eraseFromParent();
+    return Legalized;
+  }
   case TargetOpcode::G_UADDO:
   case TargetOpcode::G_USUBO: {
     if (TypeIdx == 1)
       return UnableToLegalize; // TODO
     auto LHSZext = MIRBuilder.buildZExt(WideTy, MI.getOperand(2));
     auto RHSZext = MIRBuilder.buildZExt(WideTy, MI.getOperand(3));
     unsigned Opcode = MI.getOpcode() == TargetOpcode::G_UADDO
                           ? TargetOpcode::G_ADD

llvm/test/CodeGen/AArch64/GlobalISel/legalize-add.mir

bb.0.entry:
     %2:_(s8) = G_TRUNC %0(s64)
     %3:_(s8) = G_TRUNC %1(s64)
     %4:_(s8) = G_ADD %2, %3
     %5:_(s64) = G_ANYEXT %4(s8)
     $x0 = COPY %5(s64)
 
 ...
 ---
+name:            test_scalar_uaddo_small
+body:             |
+  bb.0.entry:
+    ; CHECK-LABEL: name: test_scalar_uaddo_small
+    ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[TRUNC]], [[C]]
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[COPY1]](s64)
+    ; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[TRUNC1]], [[C]]
+    ; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[AND]], [[AND1]]
+    ; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[ADD]], [[C]]
+    ; CHECK: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[ADD]](s32), [[AND2]]
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[ADD]](s32)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s64) = G_ANYEXT [[ICMP]](s32)
+    ; CHECK: $x0 = COPY [[ANYEXT]](s64)
+    ; CHECK: $x1 = COPY [[ANYEXT1]](s64)
+    %0:_(s64) = COPY $x0
+    %1:_(s64) = COPY $x1
+    %2:_(s8) = G_TRUNC %0(s64)
+    %3:_(s8) = G_TRUNC %1(s64)
+    %4:_(s8), %5:_(s1) = G_UADDO %2, %3
+    %6:_(s64) = G_ANYEXT %4(s8)
+    %7:_(s64) = G_ANYEXT %5(s1)
+    $x0 = COPY %6(s64)
+    $x1 = COPY %7(s64)
+
+...
+---
+name:            test_scalar_saddo_small
+body:             |
+  bb.0.entry:
+    ; CHECK-LABEL: name: test_scalar_saddo_small
+    ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
+    ; CHECK: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[TRUNC]], 8
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[COPY1]](s64)
+    ; CHECK: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[TRUNC1]], 8
+    ; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ADD]](s32)
+    ; CHECK: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
+    ; CHECK: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[ADD]](s32), [[SEXT_INREG2]]
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[ADD]](s32)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s64) = G_ANYEXT [[ICMP]](s32)
+    ; CHECK: $x0 = COPY [[ANYEXT]](s64)
+    ; CHECK: $x1 = COPY [[ANYEXT1]](s64)
+    %0:_(s64) = COPY $x0
+    %1:_(s64) = COPY $x1
+    %2:_(s8) = G_TRUNC %0(s64)
+    %3:_(s8) = G_TRUNC %1(s64)
+    %4:_(s8), %5:_(s1) = G_SADDO %2, %3
+    %6:_(s64) = G_ANYEXT %4(s8)
+    %7:_(s64) = G_ANYEXT %5(s1)
+    $x0 = COPY %6(s64)
+    $x1 = COPY %7(s64)
+
+...
+---
 name:            test_scalar_uaddo_32
 body:             |
   bb.0.entry:
     ; CHECK-LABEL: name: test_scalar_uaddo_32
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
     ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $w1
     ; CHECK: [[UADDO:%[0-9]+]]:_(s32), [[UADDO1:%[0-9]+]]:_(s1) = G_UADDO [[COPY]], [[COPY1]]
     ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UADDO1]](s1)

llvm/test/CodeGen/AArch64/GlobalISel/legalize-sub.mir

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -march=aarch64 -run-pass=legalizer %s -o - | FileCheck %s
+
 ---
 name:            test_scalar_sub_big
 body:             |
   bb.0.entry:
     ; CHECK-LABEL: name: test_scalar_sub_big
     ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
     ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
     ; CHECK: [[COPY2:%[0-9]+]]:_(s64) = COPY $x2
bb.0.entry:
     %1:_(s32) = COPY $w1
     %2:_(s32), %3:_(s1) = G_USUBO %0, %1
     %4:_(s32) = G_ANYEXT %3
     $w0 = COPY %2(s32)
     $w1 = COPY %4(s32)
 
 ...
 ---
+name:            test_scalar_usubo_small
+body:             |
+  bb.0.entry:
+    ; CHECK-LABEL: name: test_scalar_usubo_small
+    ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
+    ; CHECK: [[AND:%[0-9]+]]:_(s32) = G_AND [[TRUNC]], [[C]]
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[COPY1]](s64)
+    ; CHECK: [[AND1:%[0-9]+]]:_(s32) = G_AND [[TRUNC1]], [[C]]
+    ; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[AND]], [[AND1]]
+    ; CHECK: [[AND2:%[0-9]+]]:_(s32) = G_AND [[ADD]], [[C]]
+    ; CHECK: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[ADD]](s32), [[AND2]]
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[ADD]](s32)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s64) = G_ANYEXT [[ICMP]](s32)
+    ; CHECK: $x0 = COPY [[ANYEXT]](s64)
+    ; CHECK: $x1 = COPY [[ANYEXT1]](s64)
+    %0:_(s64) = COPY $x0
+    %1:_(s64) = COPY $x1
+    %2:_(s8) = G_TRUNC %0(s64)
+    %3:_(s8) = G_TRUNC %1(s64)
+    %4:_(s8), %5:_(s1) = G_UADDO %2, %3
+    %6:_(s64) = G_ANYEXT %4(s8)
+    %7:_(s64) = G_ANYEXT %5(s1)
+    $x0 = COPY %6(s64)
+    $x1 = COPY %7(s64)
+
+...
+---
+name:            test_scalar_ssubo_small
+body:             |
+  bb.0.entry:
+    ; CHECK-LABEL: name: test_scalar_ssubo_small
+    ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
+    ; CHECK: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[TRUNC]], 8
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[COPY1]](s64)
+    ; CHECK: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[TRUNC1]], 8
+    ; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[SEXT_INREG]], [[SEXT_INREG1]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ADD]](s32)
+    ; CHECK: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
+    ; CHECK: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[ADD]](s32), [[SEXT_INREG2]]
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[ADD]](s32)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s64) = G_ANYEXT [[ICMP]](s32)
+    ; CHECK: $x0 = COPY [[ANYEXT]](s64)
+    ; CHECK: $x1 = COPY [[ANYEXT1]](s64)
+    %0:_(s64) = COPY $x0
+    %1:_(s64) = COPY $x1
+    %2:_(s8) = G_TRUNC %0(s64)
+    %3:_(s8) = G_TRUNC %1(s64)
+    %4:_(s8), %5:_(s1) = G_SADDO %2, %3
+    %6:_(s64) = G_ANYEXT %4(s8)
+    %7:_(s64) = G_ANYEXT %5(s1)
+    $x0 = COPY %6(s64)
+    $x1 = COPY %7(s64)
+
+...
+---
 name:            test_scalar_ssubo_32
 body:             |
   bb.0.entry:
     ; CHECK-LABEL: name: test_scalar_ssubo_32
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
     ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $w1
     ; CHECK: [[SSUBO:%[0-9]+]]:_(s32), [[SSUBO1:%[0-9]+]]:_(s1) = G_SSUBO [[COPY]], [[COPY1]]
     ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SSUBO1]](s1)