Fix overeager alignment upgrading when vectorizer tries to upgrade allocas

Author: dakersnar

llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp

@@ -934,7 +934,8 @@ std::vector<Chain> Vectorizer::splitChainByAlignment(Chain &C) {
                             isa<AllocaInst>(PtrOperand->stripPointerCasts());
       Align Alignment = getLoadStoreAlignment(C[CBegin].Inst);
       Align PrefAlign = Align(StackAdjustedAlignment);
-      if (IsAllocaAccess && Alignment.value() % SizeBytes != 0) {
+      if (IsAllocaAccess && Alignment.value() % SizeBytes != 0 &&
+          accessIsAllowedAndFast(SizeBytes, AS, PrefAlign, VecElemBits)) {
         Align NewAlign = getOrEnforceKnownAlignment(
             PtrOperand, PrefAlign, DL, C[CBegin].Inst, nullptr, &DT);
         if (NewAlign >= Alignment) {
@@ -943,6 +944,7 @@ std::vector<Chain> Vectorizer::splitChainByAlignment(Chain &C) {
                      << Alignment.value() << " to " << NewAlign.value()
                      << "\n");
           Alignment = NewAlign;
+          setLoadStoreAlignment(C[CBegin].Inst, Alignment);
         }
       }
 
@@ -1117,14 +1119,6 @@ bool Vectorizer::vectorizeChain(Chain &C) {
   Type *VecTy = FixedVectorType::get(VecElemTy, NumElem);
 
   Align Alignment = getLoadStoreAlignment(C[0].Inst);
-  // If this is a load/store of an alloca, we might have upgraded the alloca's
-  // alignment earlier.  Get the new alignment.
-  if (AS == DL.getAllocaAddrSpace()) {
-    Alignment = std::max(
-        Alignment,
-        getOrEnforceKnownAlignment(getLoadStorePointerOperand(C[0].Inst),
-                                   MaybeAlign(), DL, C[0].Inst, nullptr, &DT));
-  }
 
   // All elements of the chain must have the same scalar-type size.
 #ifndef NDEBUG

llvm/test/CodeGen/NVPTX/variadics-backend.ll

@@ -110,7 +110,7 @@ define dso_local i32 @foo() {
 ; CHECK-PTX-NEXT:  // %bb.0: // %entry
 ; CHECK-PTX-NEXT:    mov.b64 %SPL, __local_depot1;
 ; CHECK-PTX-NEXT:    cvta.local.u64 %SP, %SPL;
-; CHECK-PTX-NEXT:    st.v2.b32 [%SP], {1, 1};
+; CHECK-PTX-NEXT:    st.b64 [%SP], 4294967297;
 ; CHECK-PTX-NEXT:    st.b32 [%SP+8], 1;
 ; CHECK-PTX-NEXT:    st.b64 [%SP+16], 1;
 ; CHECK-PTX-NEXT:    st.b64 [%SP+24], 4607182418800017408;

llvm/test/Transforms/LoadStoreVectorizer/NVPTX/vectorize_i8.ll

@@ -58,7 +58,7 @@ define void @int8x3a4(ptr nocapture align 4 %ptr) {
   %l1 = load i8, ptr %ptr1, align 1
   %l2 = load i8, ptr %ptr2, align 2
 
-  store i8 %l2, ptr %ptr0, align 2
+  store i8 %l2, ptr %ptr0, align 4
   store i8 %l1, ptr %ptr1, align 1
   store i8 %l0, ptr %ptr2, align 4
 

llvm/test/Transforms/LoadStoreVectorizer/X86/massive_indirection.ll

@@ -155,7 +155,7 @@ define void @variadics1(ptr %vlist) {
 ; CHECK-NEXT:    [[ARGP_NEXT12:%.*]] = getelementptr i8, ptr [[ARGP_CUR11_ALIGNED]], i64 8
 ; CHECK-NEXT:    [[X2:%.*]] = getelementptr i8, ptr [[ARGP_NEXT12]], i32 7
 ; CHECK-NEXT:    [[ARGP_CUR16_ALIGNED:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[X2]], i64 0)
-; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[ARGP_CUR16_ALIGNED]], align 4294967296
+; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr [[ARGP_CUR16_ALIGNED]], align 8
 ; CHECK-NEXT:    [[X31:%.*]] = extractelement <2 x double> [[TMP1]], i32 0
 ; CHECK-NEXT:    [[X42:%.*]] = extractelement <2 x double> [[TMP1]], i32 1
 ; CHECK-NEXT:    [[X5:%.*]] = fadd double [[X42]], [[X31]]