consider this piece of code:
uint8_t v[8] = { ... };
int ret = 256;
int ret_pos = -1;
for (int i=0; i<8; ++i)
{
if (v[i] < ret)
{
ret = v[i];
ret_pos = i;
}
}
It finds min and position of the min element (ret
and ret_pos
). In arm neon I could use pairwise min to find min element in v, but how do I find position of the min element?
Update: see my own answer, what would you suggest to improve it?
Here's how I've done after spending some time fiddling with bits and math:
#define VMIN8(x, index, value) \
do { \
uint8x8_t m = vpmin_u8(x, x); \
m = vpmin_u8(m, m); \
m = vpmin_u8(m, m); \
uint8x8_t r = vceq_u8(x, m); \
\
uint8x8_t z = vand_u8(vmask, r); \
\
z = vpadd_u8(z, z); \
z = vpadd_u8(z, z); \
z = vpadd_u8(z, z); \
\
unsigned u32 = vget_lane_u32(vreinterpret_u32_u8(z), 0); \
index = __lzcnt(u32); \
value = vget_lane_u8(m, 0); \
} while (0)
uint8_t v[8] = { ... };
static const uint8_t mask[] = { 0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01 };
uint8x8_t vmask = vld1_u8(mask);
uint8x8_t v8 = vld1_u8(v);
int ret;
int ret_pos;
VMIN8(v8, ret_pos, ret);
where __lzcnt is clz (__builtin_clz in gcc).
Here's the how it works. At first using pairwise min set all u8 fields of uint8x8_t to the minimum value:
uint8x8_t m = vpmin_u8(x, x);
m = vpmin_u8(m, m);
m = vpmin_u8(m, m);
then using vector compare set min element to all ones, and all others set to zeros:
uint8x8_t r = vceq_u8(x, m);
Then perform logical AND with the mask that contains values: uint8_t mask[] {1<<7, 1<<6, 1<<5, ... 1<<1, 1<<0 };
:
uint8x8_t z = vand_u8(vmask, r);
and after that using pairwise add add all 8 bytes of
z = vpadd_u8(z, z);
z = vpadd_u8(z, z);
z = vpadd_u8(z, z);
and after that using clz calculate position of the first min element.
unsigned u32 = vget_lane_u32(vreinterpret_u32_u8(z), 0);
index = __lzcnt(u32);
Then, in real code I use VMIN8 multiple times per loop iteration and compiler is able to perfectly interleave multiple VMIN8 calls to avoid data stalls.