On Mon, Jun 1, 2015 at 12:43 PM, Eric Botcazou <ebotca...@adacore.com> wrote: > Hi, > > this fixes an ICE on a function returning a variable-sized record type but > discovered to be no-return by the optimizer. In this case, the LHS of the > GIMPLE call statement is removed so the RTL expander attempts to allocate a > temporary and fails: > > eric@polaris:~/build/gcc/native> gcc/gnat1 -quiet varsize_return2.ads -O > +===========================GNAT BUG DETECTED==============================+ > | 6.0.0 20150531 (experimental) [trunk revision 223897] (x86_64-suse-linux) > GCC error:| > | in assign_stack_temp_for_type, at function.c:793 > > because it too cannot create temporaries of variable size. > > The attached patch simply preserves the LHS throughout the GIMPLE pipeline. > It also simplifies the relevant test in gimplify_modify_expr_rhs, which was > overly broad. > > Tested on x86_64-suse-linux, OK for the mainline?
Ok. (I wonder if there are any cases where the return value is allocated by the callee?) Thanks, Richard. > > 2015-06-01 Eric Botcazou <ebotca...@adacore.com> > > * gimplify.c (gimplify_modify_expr_rhs): Use simple test on the size. > * cgraph.c (cgraph_redirect_edge_call_stmt_to_callee): Do not remove > the LHS of a no-return call if its type has variable size. > * tree-cfgcleanup.c (fixup_noreturn_call): Likewise. > * tree-cfg.c (verify_gimple_call): Accept these no-return calls. > > > 2015-06-01 Eric Botcazou <ebotca...@adacore.com> > > * gnat.dg/specs/varsize_return2.ads: New test. > * gnat.dg/specs/varsize_return2_pkg.ad[sb]: New helper. > > > -- > Eric Botcazou
