On Tue, Jul 19, 2011 at 8:20 PM, Ulrich Weigand <uweig...@de.ibm.com> wrote: > Martin Jambor wrote: > >> I had to add a test that the analyzed expression is not an SSA name. >> This has been approved by Rchi on IRC yesterday. Thus, I have >> committed the following as revision 175703 after successful run of c >> and c++ testsuite on sparc64 (and a bootstrap and test with another >> patch on x86_64-linux). >> >> Thanks, >> >> Martin >> >> >> 2011-06-30 Martin Jambor <mjam...@suse.cz> >> >> PR tree-optimization/49094 >> * tree-sra.c (tree_non_mode_aligned_mem_p): New function. >> (build_accesses_from_assign): Use it. > > This causes a regression on spu-elf: > FAIL: gcc.dg/tree-ssa/forwprop-5.c scan-tree-dump-times optimized "disappear" > 0 > > The problem is that in this expression > disappear = VIEW_CONVERT_EXPR<struct VecClass>(x_8); > the rhs is considered unaligned and blocks the SRA transformation. > > The check you added for SSA_NAMEs doesn't hit, because the SSA_NAME is > encapsulated in a VIEW_CONVERT_EXPR. When get_object_alignment is called, > the VIEW_CONVERT_EXPR is stripped off by get_inner_reference and the > SSA_NAME appears, but then get_object_alignment doesn't handle it > and just returns the default alignment of 8 bits. > > Maybe get_object_alignment should itself handle SSA_NAMEs?
But what should it return for a rvalue? There is no "alignment" here. I think SRA should avoid asking for rvalues. Richard. > Bye, > Ulrich > > -- > Dr. Ulrich Weigand > GNU Toolchain for Linux on System z and Cell BE > ulrich.weig...@de.ibm.com >
