rsqrt_tile

void ckernel::rsqrt_tile_init()

Please refer to documentation for any_init.

template<RsqrtMode mode = RsqrtMode::Default, bool is_fp32_dest_acc_en = DST_ACCUM_MODE>
void ckernel::rsqrt_tile(uint32_t idst)

Performs element-wise computation of reciprocal sqrt on each element of a tile in DST register at index idst. The DST register buffer must be in acquired state via acquire_dst call. This call is blocking and is only available on the compute engine. On Wormhole and Blackhole, RsqrtMode::Default returns NaN for negative inputs. RsqrtMode::Fast skips that negative-input check, leaving results for negative inputs unspecified. It does not change approximation precision for nonnegative inputs: both modes use the kernel’s APPROX setting to select the approximation. RsqrtMode::Fast is not supported on Quasar and is rejected at compile time. Legacy Boolean template arguments are not accepted.

Return value: None

Argument Description

idst The index of the tile in DST register buffer to perform the computation on