[pypy-commit] stmgc default: Another llvm "fix", in quotes, because it might be bogus.

[arigo]

Author: Armin Rigo <arigo at tunes.org>
Branch: 
Changeset: r998:321767c428c6
Date: 2014-03-13 15:38 +0100
http://bitbucket.org/pypy/stmgc/changeset/321767c428c6/

Log:	Another llvm "fix", in quotes, because it might be bogus.

	If 'a' is a global with an address_space(256), then this fix allows
	us to use "(void *)(long)&a" in prebuilt structs.

	It looks bogus because "(void *)&a" still doesn't work, with clang
	giving the error "error: initializer element is not a compile-time
	constant". Too bad, we can always generate the first variant in
	pypy and be happy.

diff --git a/c7/llvmfix/addrspacecast-in-constant.diff b/c7/llvmfix/addrspacecast-in-constant.diff
new file mode 100644
--- /dev/null
+++ b/c7/llvmfix/addrspacecast-in-constant.diff
@@ -0,0 +1,13 @@
+Index: lib/CodeGen/AsmPrinter/AsmPrinter.cpp
+===================================================================
+--- lib/CodeGen/AsmPrinter/AsmPrinter.cpp       (revision 203791)
++++ lib/CodeGen/AsmPrinter/AsmPrinter.cpp       (working copy)
+@@ -1529,6 +1529,8 @@
+                                    Ctx);
+   }
+ 
++  case Instruction::AddrSpaceCast:
++    // XXX???  I *think* it is correct at least in PyPy's case
+   case Instruction::Trunc:
+     // We emit the value and depend on the assembler to truncate the generated
+     // expression properly.  This is important for differences between