Changeset 1356:59c2aa9def23

Timestamp:

05/15/09 11:05:35 (3 years ago)

Author:

Tomas Lindquist Olsen <tomas.l.olsen gmail com>

branch:

default

Message:

Enabled fix for ticket #294 . It will try it's best to let LLVM handle the alignment, and fall back to manual padding when that's not possible. If this causes problems again, we'll have to consistently pad manually using packed structs.

Files:

• gen/llvmhelpers.cpp (modified) (1 diff)

gen/llvmhelpers.cpp

@@ -1452 +1452 @@
-        return alignedoffset;
-
-disabled since this can fail for opaques. if the check above is not in plac
-it does so for gcx.d!!!
-// this needs to be investigated, but I don't have time right now!
-
-// if not, we have to make sure it agrees with what llvm thinks is the alignment
-// sometimes this is different from what we really need (in case of unions, see #294)
-// so if there we get different results we don't realign the offset at all and instead
-
-IF_LOG Logger::cout() << "getting alignment for type " << type->toChars()
-    << " with llvm type " << *DtoType(type) << std::endl
-
-
+we cannot get the llvm alignment if the type is still opaque, this can happen in som
+forward reference situations, so when this happens we fall back to manual padding.
+const llvm::Type* T = DtoType(type);
+
+{
+    return offset;
+}
+
+// then we check against the llvm alignment
+size_t alignsize2 = gTargetData->getABITypeAlignment(T)
+
+
-    if (alignsize != alignsize2)
-    {
-        assert(alignsize > alignsize2 && "this is not good, the D and LLVM "
-            "type alignments differ, but LLVM's is bigger! This will break "
-the type mapping algorithms
+aggregate type mapping
-        // don't try and align the offset, and let the mappers pad 100% manually
-        return offset;
-    }
-
-
+
+
-    return alignedoffset;
-}