https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127824
Bug ID: 127824
Summary: fails to vectorize bounded double comparison with and
reduction
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: manu at gcc dot gnu.org
Target Milestone: ---
https://godbolt.org/z/Pbdc986Wa
unsigned
weakly_dominates(const double * restrict a, const double * restrict b, const
unsigned dim)
{
enum { DIM_BLOCK=4 };
int d = 0;
for (; d+DIM_BLOCK <= dim; d += DIM_BLOCK) {
unsigned a_leq_b = 1;
for (unsigned i = 0; i < DIM_BLOCK; i++)
a_leq_b &= (a[d+i] <= b[d+i]);
if (!a_leq_b)
return 0;
}
for (; d < dim; d++)
if (a[d] > b[d])
return 0;
return 1;
}
unsigned
weakly_dominates_2(const double * restrict a, const double * restrict b, const
unsigned dim)
{
unsigned a_leq_b = 1;
for (unsigned d = 0; d < dim; d++)
a_leq_b &= (a[d] <= b[d]);
return a_leq_b;
}
unsigned
weakly_dominates_3(const double * restrict a, const double * restrict b, const
unsigned dim)
{
for (unsigned d = 0; d < dim; d++)
if (!(a[d] <= b[d]))
return 0;
return 1;
}
With gcc -O3 -march=x86-64-v3 , the inner for-loop in weakly_dominates() is not
vectorized no matter the value of DIM_BLOCK, but my understanding is that it
should get vectorized exactly like weakly_dominates_3 is.
Clang does seem to be able to vectorize it as:
weakly_dominates:
mov ecx, edx
cmp edx, 4
jae .LBB0_3
xor r8d, r8d
jmp .LBB0_2
.LBB0_3:
xor eax, eax
xor r9d, r9d
.LBB0_5:
vmovupd ymm0, ymmword ptr [rdi + 8*r9]
vcmpnlepd ymm0, ymm0, ymmword ptr [rsi + 8*r9]
vtestpd ymm0, ymm0
jne .LBB0_9
lea r8, [r9 + 4]
add r9, 8
cmp r9, rcx
mov r9, r8
jbe .LBB0_5
.LBB0_2:
mov eax, 1
cmp edx, r8d
jbe .LBB0_9
.LBB0_7:
vmovsd xmm0, qword ptr [rdi + 8*r8]
vucomisd xmm0, qword ptr [rsi + 8*r8]
ja .LBB0_8
inc r8
cmp rcx, r8
jne .LBB0_7
.LBB0_9:
vzeroupper
ret
.LBB0_8:
xor eax, eax
vzeroupper
ret
