[Sorry, pressed the wrong key combination again for the other mail... ]
On Friday 14 October 2005 01:41, Evan Cheng wrote:
> gcc produces the following code:
You're not giving enough information for me to reproduce it. I used
"-O2 -march=i686 -dAP" for the follow annotated assembler output:
#(insn:TI 126 125 40 (parallel [
# (set (reg:SI 0 ax [71])
# (minus:SI (reg:SI 0 ax [71])
# (reg:SI 5 di)))
# (clobber (reg:CC 17 flags))
# ]) 242 {*subsi_1} (insn_list:REG_DEP_TRUE 125 (insn_list:REG_DEP_TRUE
37 (nil)))
# (expr_list:REG_DEAD (reg:SI 5 di)
# (expr_list:REG_UNUSED (reg:CC 17 flags)
# (nil))))
subl %edi, %eax # 126 *subsi_1/1 [length = 2]
#(insn:TI 40 126 41 (set (reg:CC 17 flags)
# (compare:CC (reg:SI 0 ax [71])
# (const_int 0 [0x0]))) 5 {*cmpsi_1_insn} (insn_list:REG_DEP_TRUE
126 (nil))
# (expr_list:REG_DEAD (reg:SI 0 ax [71])
# (nil)))
cmpl $0, %eax # 40 *cmpsi_1_insn/1 [length = 3]
> The cmpl is not needed because subl has already set the flags.
And as you already noticed, we don't actually model that in this case.
> My question is: where and how would you suggest we do this
> optimization.
combine.
> With peephole2?
You'd get an explosion of peephole patterns.
Gr.
Steven
