[AArch64][GlobalISel] Scalarize <2 x s64> G_MUL since we don't have native support for it.

[aemerson]

Summary

Adds a predicated variant of scalarize() to do this.

[paquette]

LGTM

llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h

public:
   }
 
   LegalizeRuleSet &scalarize(unsigned TypeIdx) {
     using namespace LegalityPredicates;
     return actionIf(LegalizeAction::FewerElements, isVector(typeIdx(TypeIdx)),
                     LegalizeMutations::scalarize(TypeIdx));
   }
 
+  LegalizeRuleSet &scalarizeIf(LegalityPredicate Predicate, unsigned TypeIdx) {
+    using namespace LegalityPredicates;
+    return actionIf(LegalizeAction::FewerElements,
+                    all(Predicate, isVector(typeIdx(TypeIdx))),
+                    LegalizeMutations::scalarize(TypeIdx));
+  }
+
   /// Ensure the scalar or element is at least as wide as Ty.
   LegalizeRuleSet &minScalarOrElt(unsigned TypeIdx, const LLT Ty) {
     using namespace LegalityPredicates;
     using namespace LegalizeMutations;
     return actionIf(LegalizeAction::WidenScalar,
                     scalarOrEltNarrowerThan(TypeIdx, Ty.getScalarSizeInBits()),
                     changeElementTo(typeIdx(TypeIdx), Ty));
   }

llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp

getActionDefinitionsBuilder(G_PHI)
       .widenScalarToNextPow2(0);
 
   getActionDefinitionsBuilder(G_BSWAP)
       .legalFor({s32, s64, v4s32, v2s32, v2s64})
       .clampScalar(0, s32, s64)
       .widenScalarToNextPow2(0);
 
   getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR})
-      .legalFor({s32, s64, v2s32, v4s32, v2s64, v4s16, v8s16, v16s8})
+      .legalFor({s32, s64, v2s32, v4s32, v4s16, v8s16, v16s8})
+      .scalarizeIf(
+          [=](const LegalityQuery &Query) {
+            return Query.Opcode == G_MUL && Query.Types[0] == v2s64;
+          },
+          0)
+      .legalFor({v2s64})
       .clampScalar(0, s32, s64)
       .widenScalarToNextPow2(0)
       .clampNumElements(0, v2s32, v4s32)
       .clampNumElements(0, v2s64, v2s64)
       .moreElementsToNextPow2(0);
 
   getActionDefinitionsBuilder(G_SHL)
       .customIf([=](const LegalityQuery &Query) {

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

 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -march=aarch64 -run-pass=legalizer %s -o - | FileCheck %s
+# RUN: llc -march=aarch64 -run-pass=legalizer -global-isel-abort=1  %s -o - | FileCheck %s
 ---
 name:            test_scalar_mul_small
 body:             |
   bb.0.entry:
     ; CHECK-LABEL: name: test_scalar_mul_small
     ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
     ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
     ; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)

 body:             |
   bb.0:
     ; CHECK-LABEL: name: test_smul_overflow
     ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
     ; CHECK: [[COPY1:%[0-9]+]]:_(s64) = COPY $x1
     ; CHECK: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[COPY]], [[COPY1]]
     ; CHECK: [[SMULH:%[0-9]+]]:_(s64) = G_SMULH [[COPY]], [[COPY1]]
     ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 63
-    ; CHECK: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[MUL]], [[C]]
+    ; CHECK: [[ASHR:%[0-9]+]]:_(s64) = G_ASHR [[MUL]], [[C]](s64)
     ; CHECK: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ne), [[SMULH]](s64), [[ASHR]]
     ; CHECK: $x0 = COPY [[MUL]](s64)
     ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[ICMP]](s32)
     ; CHECK: $w0 = COPY [[COPY2]](s32)
     %0:_(s64) = COPY $x0
     %1:_(s64) = COPY $x1
     %2:_(s64), %3:_(s1) = G_SMULO %0, %1
     $x0 = COPY %2(s64)
bb.0:
     %0:_(s64) = COPY $x0
     %1:_(s64) = COPY $x1
     %2:_(s64), %3:_(s1) = G_UMULO %0, %1
     $x0 = COPY %2(s64)
     %4:_(s32) = G_ANYEXT %3(s1)
     $w0 = COPY %4(s32)
 
 ...
+---
+name:            vector_mul_scalarize
+liveins:
+  - { reg: '$q0' }
+  - { reg: '$q1' }
+body:             |
+  bb.1:
+    liveins: $q0, $q1
+
+    ; CHECK-LABEL: name: vector_mul_scalarize
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $q0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $q1
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; CHECK: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
+    ; CHECK: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[UV]], [[UV2]]
+    ; CHECK: [[MUL1:%[0-9]+]]:_(s64) = G_MUL [[UV1]], [[UV3]]
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[MUL]](s64), [[MUL1]](s64)
+    ; CHECK: $q0 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    ; CHECK: RET_ReallyLR implicit $q0
+    %0:_(<2 x s64>) = COPY $q0
+    %1:_(<2 x s64>) = COPY $q1
+    %2:_(<2 x s64>) = G_MUL %0, %1
+    $q0 = COPY %2(<2 x s64>)
+    RET_ReallyLR implicit $q0
+...