[Bug rtl-optimization/41455] New: memcpy not tail called if it's a struct assignment

[TabonyEE at austin dot rr dot com]

For the following code,

typedef struct {
  char c[11111];
} s;

void f(s *a, s *b){
  *a = *b;
}

compiling for ARM, with -Os, mainline GCC generates the following:

        str     lr, [sp, #-4]!
        ldr     r2, .L2
        bl      memcpy
        ldr     pc, [sp], #4
.L3:
        .align  2
.L2:
        .word   11112

GCC 3.4.6 generates a tail call:

        ldr     r2, .L2
        @ lr needed for prologue
        b       memcpy
.L3:
        .align  2
.L2:
        .word   11112

The first release of GCC to regress in this way was 4.0.0.

Here is a backtrace from emit_block_move_hints.

#0  emit_block_move_hints (x=0x2ba56160cae0, y=0x2ba56160cb20,
size=0x2ba5615cedc0, method=BLOCK_OP_NORMAL, expected_align=0,
expected_size=-1)
    at ../../gcc/expr.c:1163
#1  0x0000000000656f33 in emit_block_move (x=0x2ba56160cae0, y=0x2ba56160cb20,
size=0x2ba5615cedc0, method=BLOCK_OP_NORMAL) at ../../gcc/expr.c:1233
#2  0x000000000066076d in store_expr (exp=0x2ba5610fa0c0,
target=0x2ba56160cae0, call_param_p=0, nontemporal=0 '\0') at
../../gcc/expr.c:4779
#3  0x000000000065ee57 in expand_assignment (to=0x2ba5610fa080,
from=0x2ba5610fa0c0, nontemporal=0 '\0') at ../../gcc/expr.c:4428

Is there any way store_expr() could pass BLOCK_OP_TAILCALL to
emit_block_move()?  If not, then can this be transformed into a tail call
later?


-- 
           Summary: memcpy not tail called if it's a struct assignment
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: TabonyEE at austin dot rr dot com
 GCC build triplet: x86_64-unknown-linux-gnu
  GCC host triplet: x86_64-unknown-linux-gnu
GCC target triplet: arm-elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41455