- don't specify lef_inaggregate when getting the llvm type representation

    for function results and parameters (they're not in an aggregate, and this
    caused 80 bits extended floating point types to be translated into arrays
    of 10 bytes, which have different ABI meanings)

git-svn-id: trunk@30714 -

compiler/llvm/llvmdef.pas

@@ -597,7 +597,7 @@ implementation
              encodedstr:=encodedstr+', '
           else
             first:=false;
-          llvmaddencodedtype_intern(usedef,[lef_inaggregate],encodedstr);
+          llvmaddencodedtype_intern(usedef,[],encodedstr);
           { in case signextstr<>'', there should be only one paraloc -> no need
             to clear (reason: it means that the paraloc is larger than the
             original parameter) }
@@ -655,7 +655,7 @@ implementation
             if pddecltype in [lpd_decl] then
               encodedstr:=encodedstr+llvmvalueextension2str[signext];
             encodedstr:=encodedstr+' ';
-            llvmaddencodedtype_intern(usedef,[lef_inaggregate],encodedstr);
+            llvmaddencodedtype_intern(usedef,[],encodedstr);
           end
         else
           begin