Skip to main content

tensor_fma32

et_kernel::tensor

Function tensor_fma32 

Source 
pub unsafe fn tensor_fma32(xs: u64)
Expand description

Initiate an asynchronous TensorFMA32.

Issues csrrw x0, 0x801, xs where xs is built by fma32_xs. The operation is asynchronous: call tensor_wait(TensorEvent::Fma) before reading the FP register file or issuing a subsequent tensor_store.

ยงSafety

  • The L1 scratchpad must be fully populated (TensorLoad with subsequent tensor_wait(Load0)) before this call when tenb = false, or equivalently tensor_load_b must have been issued before this call for the TenB path.
  • Must be called from the primary hart of the Minion.