https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70053
Jiu Fu Guo <guojiufu at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
Resolution|--- |FIXED
--- Comment #12 from Jiu Fu Guo <guojiufu at gcc dot gnu.org> ---
The generated code is a lot better on the trunk (both -O2 and -O3):
.cfi_startproc
dcmpuq 0,4,6
beq 0,.L4
blr
.p2align 4,,15
.L4:
fmr 5,7
fmr 4,6
fmr 3,7
fmr 2,6
blr
And the expanded RTL like below, and cse/fwprop/dse optimize it as expected.
2: r119:TD=%2:TD
3: r120:TD=%4:TD
4: r121:TD=%6:TD
5: NOTE_INSN_FUNCTION_BEG
10: r122:CCFP=cmp(r120:TD,r121:TD)
11: pc={(r122:CCFP!=0)?L13:pc}
REG_BR_PROB 708669604
12: NOTE_INSN_BASIC_BLOCK 4
6: r120:TD=r121:TD
7: r119:TD=r121:TD
13: L13:
14: NOTE_INSN_BASIC_BLOCK 5
15: r123:DI=r112:DI+0x10
16: [r123:DI]=r120:TD
17: [r112:DI]=r119:TD
18: r124:TD=[r112:DI]
19: r126:DI=r112:DI+0x10
20: r125:TD=[r126:DI]
21: r117:TD=r124:TD
22: r118:TD=r125:TD
26: %2:TD=r117:TD
27: %4:TD=r118:TD
28: use %2:TD
29: use %4:TD
