https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102991
--- Comment #4 from luoxhu at gcc dot gnu.org ---
vect-simd-17.p10.c.335r.final:
3379: %v1:V16QI=unspec[%v1:V16QI,%v1:V16QI,%v9:V16QI] 254
3372: {%v11:V4SI=~%v0:V4SI&%v13:V4SI|%v11:V4SI;clobber %r10:V4SI;} // wrong
code.
REG_DEAD %v0:V4SI
REG_UNUSED %r10:V4SI
3373: [%r1:DI+0x80]=%v11:V4SI
ASM:
.L149:
lxvx %vs32,%r9,%r8
vadduwm %v12,%v12,%v13
mfvsrd %r5,%vs42
mfvsrld %r4,%vs42
vadduwm %v10,%v10,%v8
stxv %vs44,112(%r1)
xxperm %vs32,%vs32,%vs41
vadduwm %v1,%v13,%v0
vcmpequw %v0,%v0,%v13
xxperm %vs33,%vs33,%vs41
vandc %r10,%v13,%v0 // wrong code
vor %v11,%r10,%v11 // wrong code
stxv %vs43,128(%r1)
stxvx %vs33,%r9,%r8
addi %r8,%r8,-16
bdnz .L149
But the binary is (/opt/binutils-power10/bin/objdump -d vect-simd-17.p10 |
less):
10002ea0: 19 42 09 7c lxvx vs32,r9,r8
10002ea4: 80 68 8c 11 vadduwm v12,v12,v13
10002ea8: 67 00 45 7d mfvrd r5,v10
10002eac: 67 02 44 7d mfvsrld r4,vs42
10002eb0: 80 40 4a 11 vadduwm v10,v10,v8
10002eb4: 7d 00 81 f5 stxv vs44,112(r1)
10002eb8: d7 48 00 f0 xxperm vs32,vs32,vs41
10002ebc: 80 00 2d 10 vadduwm v1,v13,v0
10002ec0: 86 68 00 10 vcmpequw v0,v0,v13
10002ec4: d7 48 21 f0 xxperm vs33,vs33,vs41
10002ec8: 44 04 4d 11 vandc v10,v13,v0 // wrong code
10002ecc: 84 5c 6a 11 vor v11,v10,v11 // wrong code
10002ed0: 8d 00 61 f5 stxv vs43,128(r1)
10002ed4: 19 43 29 7c stxvx vs33,r9,r8
10002ed8: f0 ff 08 39 addi r8,r8,-16
10002edc: c4 ff 00 42 bdnz 10002ea0 <doit+0x26f0>
%vs42 is a global constant data loaded from memory, it was modified at address
0x10002ec8, there r10 is changed to v10 from ASM to binary, which was supposed
to be never change in the loop.
(gdb)
0x0000000010002eb4 <doit+9988>: 7d 00 81 f5 stxv vs44,112(r1)
0x0000000010002eb8 <doit+9992>: d7 48 00 f0 xxperm vs32,vs32,vs41
0x0000000010002ebc <doit+9996>: 80 00 2d 10 vadduwm v1,v13,v0
0x0000000010002ec0 <doit+10000>: 86 68 00 10 vcmpequw v0,v0,v13
0x0000000010002ec4 <doit+10004>: d7 48 21 f0 xxperm vs33,vs33,vs41
=> 0x0000000010002ec8 <doit+10008>: 44 04 4d 11 vandc v10,v13,v0
0x0000000010002ecc <doit+10012>: 84 5c 6a 11 vor v11,v10,v11
0x0000000010002ed0 <doit+10016>: 8d 00 61 f5 stxv vs43,128(r1)
7: $vs42.v4_int32 = {-30, -29, -28, -27}
(gdb) si
0x0000000010002eb4 <doit+9988>: 7d 00 81 f5 stxv vs44,112(r1)
0x0000000010002eb8 <doit+9992>: d7 48 00 f0 xxperm vs32,vs32,vs41
0x0000000010002ebc <doit+9996>: 80 00 2d 10 vadduwm v1,v13,v0
0x0000000010002ec0 <doit+10000>: 86 68 00 10 vcmpequw v0,v0,v13
0x0000000010002ec4 <doit+10004>: d7 48 21 f0 xxperm vs33,vs33,vs41
0x0000000010002ec8 <doit+10008>: 44 04 4d 11 vandc v10,v13,v0
=> 0x0000000010002ecc <doit+10012>: 84 5c 6a 11 vor v11,v10,v11
0x0000000010002ed0 <doit+10016>: 8d 00 61 f5 stxv vs43,128(r1)
7: $vs42.v4_int32 = {0, 0, 0, 0}
