/ prepack A according to gemm kernel A block size: (<4x2>x1) x2, with unroll=2 can be described as below: origin A data: A_origin(no trans, m x k): r0

/ prepack A according to gemm kernel A block size: <4x2>x2, unroll x4, can be described as below: origin A data: A_origin(no trans, k x m): r0: ==>

! \brief Appends an element after the last valid element. * Calling this on a vector that has reached capacity will throw an * exception if

Reallocate `new_size` bytes of memory, replacing the old region of size `old_size` at `p`. In contrast to a normal realloc, this grows downwards, and

Run the PaddlePredictor. @return true if run successfully.

Test a single case.

Set the tensor float data. @param buf the float array buffer which will be copied into tensor. @return true if set data successfully.

Set model from file. @return

! * Specify a vector of SVM pointers that the kernel may access in * addition to its arguments. */

DEPRECATED**, use set_xpu_l3_cache_method() in the future

__aarch64__ * \brief gemm with ablock = 6, bblock = 8, output 6x8 * @param A * @param B * @param C * @param M * @param N * @param K * @param th

evaluation of 4 sines & cosines at once. The code is the exact rewriting of the cephes sinf function. Precision is excellent as long as x < 8192 (I d

* \brief Semi-structured Sparse calculation implementation of 1x1 convolution, * both input and output are int8. * Semi-structured Sparse matrix mul