// Copyright (c) the JPEG XL Project Authors. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

/* This file is automatically generated. Do not modify it directly. */
#if HWY_TARGET != HWY_NEON
#error "only include this file from fast_dct-inl.h"
#endif

constexpr size_t FastIDCTIntegerBits(FastDCTTag<8>) { return 1; }

void FastIDCT(FastDCTTag<8>, const int16_t* in, size_t in_stride, int16_t* out,
              size_t out_stride, size_t count) {
  JXL_ASSERT(count % 8 == 0);
  for (size_t i = 0; i < count; i += 8) {
    int16x8_t v0 = vld1q_s16(in + in_stride * 0 + i);
    int16x8_t v1 = vld1q_s16(in + in_stride * 4 + i);
    int16x8_t v2 = vaddq_s16(v0, v1);
    int16x8_t v3 = vld1q_s16(in + in_stride * 2 + i);
    int16x8_t v4_tmp = vqrdmulhq_n_s16(v3, 13573);
    int16x8_t v4 = vaddq_s16(v4_tmp, v3);
    int16x8_t v5 = vld1q_s16(in + in_stride * 6 + i);
    int16x8_t v6 = vaddq_s16(v5, v3);
    int16x8_t v7 = vaddq_s16(v4, v6);
    int16x8_t v8 = vqrdmulhq_n_s16(v7, 17734);
    int16x8_t v9 = vaddq_s16(v2, v8);
    int16x8_t v10 = vld1q_s16(in + in_stride * 1 + i);
    int16x8_t v11_tmp = vqrdmulhq_n_s16(v10, 13573);
    int16x8_t v11 = vaddq_s16(v11_tmp, v10);
    int16x8_t v12 = vld1q_s16(in + in_stride * 5 + i);
    int16x8_t v13 = vld1q_s16(in + in_stride * 3 + i);
    int16x8_t v14 = vaddq_s16(v12, v13);
    int16x8_t v15 = vaddq_s16(v11, v14);
    int16x8_t v16 = vaddq_s16(v13, v10);
    int16x8_t v17 = vqrdmulhq_n_s16(v16, 25080);
    int16x8_t v18 = vld1q_s16(in + in_stride * 7 + i);
    int16x8_t v19 = vaddq_s16(v18, v12);
    int16x8_t v20 = vaddq_s16(v16, v19);
    int16x8_t v21 = vqrdmulhq_n_s16(v20, 17734);
    int16x8_t v22 = vaddq_s16(v17, v21);
    int16x8_t v23 = vaddq_s16(v15, v22);
    int16x8_t v24 = vqrdmulhq_n_s16(v23, 16705);
    int16x8_t v25 = vaddq_s16(v9, v24);
    int16x8_t v26 = vsubq_s16(v0, v1);
    int16x8_t v27 = vsubq_s16(v4, v6);
    int16x8_t v28_tmp = vqrdmulhq_n_s16(v27, 10045);
    int16x8_t v28 = vaddq_s16(v28_tmp, v27);
    int16x8_t v29 = vaddq_s16(v26, v28);
    int16x8_t v30 = vsubq_s16(v11, v14);
    int16x8_t v31 = vqrdmulhq_n_s16(v16, 17734);
    int16x8_t v32_tmp = vqrdmulhq_n_s16(v19, 10045);
    int16x8_t v32 = vaddq_s16(v32_tmp, v19);
    int16x8_t v33 = vsubq_s16(v31, v32);
    int16x8_t v34 = vaddq_s16(v30, v33);
    int16x8_t v35 = vqrdmulhq_n_s16(v34, 19705);
    int16x8_t v36 = vaddq_s16(v29, v35);
    int16x8_t v37 = vsubq_s16(v26, v28);
    int16x8_t v38 = vsubq_s16(v30, v33);
    int16x8_t v39 = vqrdmulhq_n_s16(v38, 29490);
    int16x8_t v40 = vaddq_s16(v37, v39);
    int16x8_t v41 = vsubq_s16(v2, v8);
    int16x8_t v42 = vsubq_s16(v15, v22);
    int16x8_t v43_tmp = vqrdmulhq_n_s16(v42, 18446);
    int16x8_t v43 = vmlaq_n_s16(v43_tmp, v42, 2);
    int16x8_t v44 = vaddq_s16(v41, v43);
    int16x8_t v45 = vsubq_s16(v41, v43);
    int16x8_t v46 = vsubq_s16(v37, v39);
    int16x8_t v47 = vsubq_s16(v29, v35);
    int16x8_t v48 = vsubq_s16(v9, v24);
    vst1q_s16(out + out_stride * 0 + i, v25);
    vst1q_s16(out + out_stride * 1 + i, v36);
    vst1q_s16(out + out_stride * 2 + i, v40);
    vst1q_s16(out + out_stride * 3 + i, v44);
    vst1q_s16(out + out_stride * 4 + i, v45);
    vst1q_s16(out + out_stride * 5 + i, v46);
    vst1q_s16(out + out_stride * 6 + i, v47);
    vst1q_s16(out + out_stride * 7 + i, v48);
  }
}