Here's a strange case of poor code generation with -Os. unsigned short foo2 (unsigned char on_off, unsigned short *puls) { return puls[on_off-1]; } With -O2, it's fine: movzbl %dil, %edi movzwl -2(%rsi,%rdi,2), %eax ret With -Os it's really weird: movzbl %dil, %eax movl $1, %edi subq %rax, %rdi imulq $-2, %rdi, %rdi movw (%rdi,%rsi), %ax ret The problem here seems to be that the FE (or the GIMPLE generation) transforms addr + ((on_off-1)*2) into addr + ((1-on_off)*-2) and nothing in -Os has the wit to recover. This badly affects targets with no hardware multiplier, which end up calling libgcc to do the multiply. http://gcc.gnu.org/ml/gcc/2008-12/msg00329.html
Fallout of the change made for PR middle-end/23294 in GCC 4.2.
Subject: Bug 38751 Author: rguenth Date: Wed Jan 7 10:53:30 2009 New Revision: 143152 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=143152 Log: 2009-01-07 Richard Guenther <rguenther@suse.de> PR middle-end/38751 * fold-const.c (extract_muldiv): Remove obsolete comment. (fold_plusminus_mult_expr): Undo MINUS_EXPR to PLUS_EXPR canonicalization for the canonicalization. Modified: trunk/gcc/ChangeLog trunk/gcc/fold-const.c
Fixed on the trunk: -O2: movzbl %dil, %edi movzwl -2(%rsi,%rdi,2), %eax ret -Os: movzbl %dil, %edi movw -2(%rsi,%rdi,2), %ax ret
Closing 4.2 branch.
Fixed.
Subject: Bug 38751 Author: rguenth Date: Thu Jun 25 12:39:01 2009 New Revision: 148943 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=148943 Log: 2009-06-25 Richard Guenther <rguenther@suse.de> Backport from mainline 2009-01-07 Richard Guenther <rguenther@suse.de> PR middle-end/38751 * fold-const.c (extract_muldiv): Remove obsolete comment. (fold_plusminus_mult_expr): Undo MINUS_EXPR to PLUS_EXPR canonicalization for the canonicalization. Modified: branches/gcc-4_3-branch/gcc/ChangeLog branches/gcc-4_3-branch/gcc/fold-const.c