https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117890
Bug ID: 117890
Summary: Wrong code with -fvect-cost-model=unlimited
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: lin1.hu at intel dot com
Target Milestone: ---
Created attachment 59774
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=59774&action=edit
A short case
Compile the test file using the option -O2 -march=x86-64-v3
-fvect-cost-model=unlimited will generate wrong code.
Without -fvect-cost-model=unlimited, the output should be 4, but the output is
currently 0.
According to my investigation, in case.c.183t.cunroll.
<bb 23> [local count: 761014530]:
_63 = _29 >> 2;
_64 = ~_63;
_65 = (_Bool) _64;
if (_65 != 0)
goto <bb 18>; [5.50%]
else
goto <bb 3>; [94.50%]
<bb 18> [local count: 223857727]:
# vect_vec_iv_.29_45 = PHI <{ 5, 6 }(3), { 1, 2 }(7), { 3, 4 }(23)>
# vect_vec_iv_.31_50 = PHI <{ 3, 2 }(3), { 7, 6 }(7), { 5, 4 }(23)>
_51 = BIT_FIELD_REF <vect_vec_iv_.31_50, 32, 0>;
_46 = BIT_FIELD_REF <vect_vec_iv_.29_45, 32, 0>;
goto <bb 17>; [100.00%]
when the program jump from bb 23. _46 will be 3.
<bb 17> [local count: 563714454]:
# i_38 = PHI <7(3), _46(18)>
# ivtmp_40 = PHI <1(3), _51(18)>
<bb 14> [local count: 1127428908]:
# i_9 = PHI <i_35(15), i_38(17)>
# ivtmp_17 = PHI <ivtmp_36(15), ivtmp_40(17)>
_32 = _29 >> i_9;
_33 = ~_32;
_34 = (_Bool) _33;
if (_34 != 0)
goto <bb 4>; [5.50%]
else
goto <bb 16>; [94.50%]
<bb 4> [local count: 51673825]:
# i_4 = PHI <i_1(11)>
_2 = 1 << i_4;
_19 = (unsigned char) _2;
goto <bb 6>; [100.00%]
Hence, i_9 will be 3. It's strange. The program jump because the 2nd bit
(counting from 0) is 0, i_9 should be 2, _2 will be 4. But i_9 starts at 3,
making the program semantics change.
