Function tfe_einsum_rrule_f64 

#[unsafe(no_mangle)]
pub unsafe extern "C" fn tfe_einsum_rrule_f64(
    _subscripts: *const c_char,
    _operands: *const *const TfeTensorF64,
    _num_operands: usize,
    _cotangent: *const TfeTensorF64,
    _grads_out: *mut *mut TfeTensorF64,
    _status: *mut tfe_status_t,
)

Reverse-mode rule (VJP) for einsum.

Computes one gradient tensor per input operand given the output cotangent. The caller must provide grads_out as a pre-allocated array of num_operands pointers. Each returned tensor must be released by the caller.

Safety

• subscripts must be a valid null-terminated C string.
• operands must point to an array of num_operands valid tensor pointers.
• cotangent must be a valid, non-null tensor pointer.
• grads_out must point to a caller-allocated array of num_operands mutable *mut TfeTensorF64 pointers.
• status must be a valid, non-null pointer.

Examples (C)

// After computing c = einsum("ij,jk->ik", [a, b]):
tfe_tensor_f64 *grads[2];
tfe_status_t status;
const tfe_tensor_f64 *ops[] = {a, b};
tfe_einsum_rrule_f64("ij,jk->ik", ops, 2, grad_c, grads, &status);
// grads[0] = gradient w.r.t. a
// grads[1] = gradient w.r.t. b
tfe_tensor_f64_release(grads[0]);
tfe_tensor_f64_release(grads[1]);