Make sure we don't promise alignments that are larger than the heap alignment to LLVM (#56938)

Fixes #56937

---------

Co-authored-by: Oscar Smith <oscardssmith@gmail.com>
(cherry picked from commit 1e2758e)

Author: 2 people

src/codegen.cpp

@@ -8508,6 +8508,8 @@ static jl_llvm_functions_t
             Type *RT = Arg->getParamStructRetType();
             TypeSize sz = DL.getTypeAllocSize(RT);
             Align al = DL.getPrefTypeAlign(RT);
+            if (al > MAX_ALIGN)
+                al = Align(MAX_ALIGN);
             param.addAttribute(Attribute::NonNull);
             // The `dereferenceable` below does not imply `nonnull` for non addrspace(0) pointers.
             param.addDereferenceableAttr(sz);

src/datatype.c

@@ -754,6 +754,8 @@ void jl_compute_field_offsets(jl_datatype_t *st)
             if (al > alignm)
                 alignm = al;
         }
+        if (alignm > MAX_ALIGN)
+            alignm = MAX_ALIGN; // We cannot guarantee alignments over 16 bytes because that's what our heap is aligned as
         if (LLT_ALIGN(sz, alignm) > sz) {
             haspadding = 1;
             sz = LLT_ALIGN(sz, alignm);

test/compiler/codegen.jl

@@ -916,3 +916,9 @@ end
 
 # Core.getptls() special handling
 @test !occursin("call ptr @jlplt", get_llvm(Core.getptls, Tuple{})) #It should lower to a direct load of the ptls and not a ccall
+
+
+struct Vec56937 x::NTuple{8, VecElement{Int}} end
+
+x56937 = Ref(Vec56937(ntuple(_->VecElement(1),8)))
+@test x56937[].x[1] == VecElement{Int}(1) # shouldn't crash