[SROA] Skip uses of allocas where the type is scalable

[Allen]

Summary

When visiting load and store instructions in SROA skip scalable vectors.
This is relevant in the implementation of the 'arm_sve_vector_bits'
attribute that is used to define VLS types, similar to D85725.

Fix https://gcc.godbolt.org/z/o561P9zj4

[nikic]

Looks fine, but please test the store case as well.

[Allen]

Add store case according comment

[nikic]

LGTM

llvm/lib/Transforms/Scalar/SROA.cpp

Instruction *hasUnsafePHIOrSelectUse(Instruction *Root, uint64_t &Size) {
     // If there are no loads or stores, the access is dead. We mark that as
     // a size zero access.
     Size = 0;
     do {
       Instruction *I, *UsedI;
       std::tie(UsedI, I) = Uses.pop_back_val();
 
       if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
-        Size =
-            std::max(Size, DL.getTypeStoreSize(LI->getType()).getFixedValue());
+        TypeSize LoadSize = DL.getTypeStoreSize(LI->getType());
+        if (LoadSize.isScalable()) {
+          PI.setAborted(LI);
+          return nullptr;
+        }
+        Size = std::max(Size, LoadSize.getFixedValue());
         continue;
       }
       if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
         Value *Op = SI->getOperand(0);
         if (Op == UsedI)
           return SI;
-        Size =
-            std::max(Size, DL.getTypeStoreSize(Op->getType()).getFixedValue());
+        TypeSize StoreSize = DL.getTypeStoreSize(Op->getType());
+        if (StoreSize.isScalable()) {
+          PI.setAborted(SI);
+          return nullptr;
+        }
+        Size = std::max(Size, StoreSize.getFixedValue());
         continue;
       }
 
       if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
         if (!GEP->hasAllZeroIndices())
           return GEP;
       } else if (!isa<BitCastInst>(I) && !isa<PHINode>(I) &&
                  !isa<SelectInst>(I) && !isa<AddrSpaceCastInst>(I)) {

llvm/test/Transforms/SROA/scalable-vectors.ll

 ;
   %retval = alloca <16 x i32>
   %retval.coerce = alloca <vscale x 4 x i32>
   call void @llvm.memcpy.p0.p0.i64(ptr align 16 %retval.coerce, ptr align 16 %retval, i64 64, i1 false)
   %1 = load <vscale x 4 x i32>, ptr %retval.coerce
   ret <vscale x 4 x i32> %1
 }
 
+; Test we bail out when using an alloca of a fixed-length vector (VLS) that was
+; bitcasted to a scalable vector.
+define void @select_load_alloca_to_svdouble_t() {
+; CHECK-LABEL: @select_load_alloca_to_svdouble_t(
+; CHECK-NEXT:    [[Z:%.*]] = alloca <16 x half>, align 32
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 0, 0
+; CHECK-NEXT:    [[COND:%.*]] = select i1 [[CMP]], ptr [[Z]], ptr null
+; CHECK-NEXT:    [[VAL:%.*]] = load <vscale x 2 x double>, ptr [[COND]], align 16
+; CHECK-NEXT:    ret void
+;
+  %z = alloca <16 x half>
+  %cmp = icmp eq i32 0, 0
+  %cond = select i1 %cmp, ptr %z, ptr null
+  %val = load <vscale x 2 x double>, ptr %cond, align 16
+  ret void
+}
+
+define void @select_store_alloca_to_svdouble_t(<vscale x 2 x double> %val) {
+; CHECK-LABEL: @select_store_alloca_to_svdouble_t(
+; CHECK-NEXT:    [[Z:%.*]] = alloca <16 x half>, align 32
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 0, 0
+; CHECK-NEXT:    [[COND:%.*]] = select i1 [[CMP]], ptr [[Z]], ptr null
+; CHECK-NEXT:    store <vscale x 2 x double> [[VAL:%.*]], ptr [[COND]], align 16
+; CHECK-NEXT:    ret void
+;
+  %z = alloca <16 x half>
+  %cmp = icmp eq i32 0, 0
+  %cond = select i1 %cmp, ptr %z, ptr null
+  store <vscale x 2 x double> %val, ptr %cond, align 16
+  ret void
+}
+
 declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1) nounwind
 ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
 ; CHECK-MODIFY-CFG: {{.*}}
 ; CHECK-PRESERVE-CFG: {{.*}}