// Copyright 2023 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.

$assert NR >= 8
$assert NR % 8 == 0
$assert KBLOCK >= 4
$assert KBLOCK % 4 == 0
$ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"

#include <assert.h>
#include <stddef.h>
#include <stdint.h>

#include <arm_neon.h>

#include "xnnpack/packw.h"
$if PREFETCH:
  #include "xnnpack/prefetch.h"


$IFWHILE = "if" if KBLOCK==8 else "while"
void xnn_x16_packw_gemm_goi_ukernel_x${NR}__neon_ld4lane_u${KBLOCK}${"_prfm" if PREFETCH else ""}(
  size_t g,
  size_t nc,
  size_t kc,
  size_t nr,
  size_t kr,
  size_t sr,
  const uint16_t* weights,
  const uint16_t* bias,
  const void* scale,
  uint16_t* packed_weights,
  size_t extra_bytes,
  const void* params)
{
  assert(g != 0);
  assert(nc != 0);
  assert(kc != 0);
  assert(nr == ${NR});
  assert(kr == 1);
  assert(sr == 1);
  assert(weights != NULL);
  assert(packed_weights != NULL);

  $for N in range(0, NR, 8):
    $for K in range(0, KBLOCK, 4):
      uint16x8x4_t vtmp${ABC[K:K+4]}x${ABC[N:N+8]};
      $for L in range(4):
        vtmp${ABC[K:K+4]}x${ABC[N:N+8]}.val[${L}] = vdupq_n_u16(0);

  do {
    // NC main loop multiple of ${NR}
    const uint16_t* w0 = weights;
    size_t n = nc;

    for (; n >= ${NR}; n -= ${NR}) {
      if XNN_LIKELY(bias != NULL) {
        $for N in range(0, NR, 8):
          uint16x8_t vb${N} = vld1q_u16(bias); bias += 8;
        $for N in range(0, NR, 8):
          vst1q_u16(packed_weights, vb${N}); packed_weights += 8;
      } else {
        const uint16x8_t vzero = vmovq_n_u16(0);
        $for N in range(0, NR, 8):
          vst1q_u16(packed_weights, vzero); packed_weights += 8;
      }

      $for N in range(1, NR):
        const uint16_t* w${N} = w${N-1} + kc;
      $if PREFETCH:
        $for N in range(0, NR):
          xnn_prefetch_to_l1((const int8_t*) w${N});
          xnn_prefetch_to_l1((const int8_t*) w${N} + 64);

      // KC main loop multiple of ${KBLOCK}
      size_t k = kc;
      for (; k >= ${KBLOCK}; k -= ${KBLOCK}) {
        $for K in range(0, KBLOCK, 4):
          $for N in range(0, NR, 8):
            $for L in range(8):
              vtmp${ABC[K:K+4]}x${ABC[N:N+8]} = vld4q_lane_u16(w${N+L}, vtmp${ABC[K:K+4]}x${ABC[N:N+8]}, ${L}); w${N+L} += 4;
        $if PREFETCH:
          $for N in range(0, NR):
            xnn_prefetch_to_l1((const int8_t*) w${N} + 128);
        $for K in range(0, KBLOCK, 4):
          $for L in range(4):
            $for N in range(0, NR, 8):
              vst1q_u16(packed_weights, vtmp${ABC[K:K+4]}x${ABC[N:N+8]}.val[${L}]); packed_weights += 8;
      }
      $if KBLOCK > 4:
        // KC remainder multiple of 4
        ${IFWHILE} (k >= 4) {
          $for N in range(0, NR, 8):
            $for L in range(8):
              vtmp${ABC[0:4]}x${ABC[N:N+8]} = vld4q_lane_u16(w${N+L}, vtmp${ABC[0:4]}x${ABC[N:N+8]}, ${L}); w${N+L} += 4;
          $for L in range(4):
            $for N in range(0, NR, 8):
              vst1q_u16(packed_weights, vtmp${ABC[0:4]}x${ABC[N:N+8]}.val[${L}]); packed_weights += 8;
          k -= 4;
        }

      // KC remainder of 1..3
      // Same as main loop but ld1, ld2 or ld3
      if XNN_UNLIKELY(k != 0) {
        assert(k >= 1);
        assert(k <= 3);
        switch (k) {
          $for K in range(1, 4):
            // KC remainder of ${NR}x${K}
            case ${K}:
            {
              $if K == 1:
                $for N in range(0, NR, 8):
                  uint16x8_t vtmp${ABC[0:K]}x${ABC[N:N+8]} = vdupq_n_u16(0);
              $else:
                $for N in range(0, NR, 8):
                  uint16x8x${K}_t vtmp${ABC[0:K]}x${ABC[N:N+8]};
                  $for L in range(0, K):
                    vtmp${ABC[0:K]}x${ABC[N:N+8]}.val[${L}] = vdupq_n_u16(0);

              $for N in range(0, NR, 8):
                $for L in range(8):
                  vtmp${ABC[0:K]}x${ABC[N:N+8]} = vld${K}q_lane_u16(w${N+L}, vtmp${ABC[0:K]}x${ABC[N:N+8]}, ${L}); w${N+L} += ${K};

              $for L in range(K):
                $for N in range(0, NR, 8):
                  $if K == 1:
                    vst1q_u16(packed_weights, vtmp${ABC[0:K]}x${ABC[N:N+8]}); packed_weights += 8;
                  $else:
                    vst1q_u16(packed_weights, vtmp${ABC[0:K]}x${ABC[N:N+8]}.val[${L}]); packed_weights += 8;
              break;
            }
          default:
            XNN_UNREACHABLE;
        }
      }
      packed_weights = (uint16_t*) ((uintptr_t) packed_weights + extra_bytes);
      w0 = w${NR-1};
    }

    // NC remainder (1..${NR-1})
    if XNN_UNLIKELY(n != 0) {
      assert(n >= 1);
      assert(n <= ${NR-1});
      if XNN_LIKELY(bias != NULL) {
        size_t nb = n;
        do {
          *packed_weights++  = *bias++;
        } while (--nb != 0);
        packed_weights += (${NR} - n);
      } else {
        const uint16x8_t vzero = vmovq_n_u16(0);
        $for N in range(0, NR, 8):
          vst1q_u16(packed_weights, vzero); packed_weights += 8;
      }

      // NR remainder has less than ${NR} rows so last row is not loaded
      $for N in range(1, NR-1):
        const uint16_t* w${N} = w${N-1} + kc;
        $if N % 2 == 0:
          if XNN_UNPREDICTABLE(n <= ${N}) {
            w${N} = w${N-1};
          }
        $else:
          if XNN_UNPREDICTABLE(n < ${N+1}) {
            w${N} = w${N-1};
          }

      // KC main loop multiple of ${KBLOCK}
      size_t k = kc;
      for (; k >= ${KBLOCK}; k -= ${KBLOCK}) {
        $for K in range(0, KBLOCK, 4):
          $for N in range(0, NR, 8):
            $for L in range(8):
              $if N+L != NR-1:
                vtmp${ABC[K:K+4]}x${ABC[N:N+8]} = vld4q_lane_u16(w${N+L}, vtmp${ABC[K:K+4]}x${ABC[N:N+8]}, ${L}); w${N+L} += 4;
        $if PREFETCH:
          $for N in range(0, NR-1):
            xnn_prefetch_to_l1((const int8_t*) w${N} + 128);
        $for K in range(0, KBLOCK, 4):
          $for L in range(4):
            $for N in range(0, NR, 8):
              vst1q_u16(packed_weights, vtmp${ABC[K:K+4]}x${ABC[N:N+8]}.val[${L}]); packed_weights += 8;
      }

      $if KBLOCK > 4:
        // KC remainder multiple of 4
        ${IFWHILE} (k >= 4) {
          $for N in range(0, NR, 8):
            $for L in range(8):
              $if N+L != NR-1:
                vtmp${ABC[0:4]}x${ABC[N:N+8]} = vld4q_lane_u16(w${N+L}, vtmp${ABC[0:4]}x${ABC[N:N+8]}, ${L}); w${N+L} += 4;
          $for L in range(4):
            $for N in range(0, NR, 8):
              vst1q_u16(packed_weights, vtmp${ABC[0:4]}x${ABC[N:N+8]}.val[${L}]); packed_weights += 8;
          k -= 4;
        }

      // KC remainder of 1..3
      // Same as main loop but ld1, ld2 or ld3
      if XNN_UNLIKELY(k != 0) {
        assert(k >= 1);
        assert(k <= 3);
        switch (k) {
          $for K in range(1, 4):
            // KC remainder of ${NR}x${K}
            case ${K}:
            {
              $if K == 1:
                $for N in range(0, NR, 8):
                  uint16x8_t vtmp${ABC[0:K]}x${ABC[N:N+8]} = vdupq_n_u16(0);
              $else:
                $for N in range(0, NR, 8):
                  uint16x8x${K}_t vtmp${ABC[0:K]}x${ABC[N:N+8]};
                  $for L in range(0, K):
                    vtmp${ABC[0:K]}x${ABC[N:N+8]}.val[${L}] = vdupq_n_u16(0);

              $for N in range(0, NR, 8):
                $for L in range(8):
                  $if N+L != NR-1:
                    vtmp${ABC[0:K]}x${ABC[N:N+8]} = vld${K}q_lane_u16(w${N+L}, vtmp${ABC[0:K]}x${ABC[N:N+8]}, ${L}); w${N+L} += ${K};

              $for L in range(K):
                $for N in range(0, NR, 8):
                  $if K == 1:
                    vst1q_u16(packed_weights, vtmp${ABC[0:K]}x${ABC[N:N+8]}); packed_weights += 8;
                  $else:
                    vst1q_u16(packed_weights, vtmp${ABC[0:K]}x${ABC[N:N+8]}.val[${L}]); packed_weights += 8;
              break;
            }
          default:
            XNN_UNREACHABLE;
        }
      }
      packed_weights = (uint16_t*) ((uintptr_t) packed_weights + extra_bytes);
    }

    weights += nc * kc;
  } while (--g != 0);
}