diff options
Diffstat (limited to 'third_party/aom/av1/encoder/arm/neon/error_neon.c')
-rw-r--r-- | third_party/aom/av1/encoder/arm/neon/error_neon.c | 42 |
1 files changed, 42 insertions, 0 deletions
diff --git a/third_party/aom/av1/encoder/arm/neon/error_neon.c b/third_party/aom/av1/encoder/arm/neon/error_neon.c new file mode 100644 index 000000000..fe5233f89 --- /dev/null +++ b/third_party/aom/av1/encoder/arm/neon/error_neon.c @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include <arm_neon.h> +#include <assert.h> + +#include "./av1_rtcd.h" + +int64_t av1_block_error_fp_neon(const int16_t *coeff, const int16_t *dqcoeff, + int block_size) { + int64x2_t error = vdupq_n_s64(0); + + assert(block_size >= 8); + assert((block_size % 8) == 0); + + do { + const int16x8_t c = vld1q_s16(coeff); + const int16x8_t d = vld1q_s16(dqcoeff); + const int16x8_t diff = vsubq_s16(c, d); + const int16x4_t diff_lo = vget_low_s16(diff); + const int16x4_t diff_hi = vget_high_s16(diff); + // diff is 15-bits, the squares 30, so we can store 2 in 31-bits before + // accumulating them in 64-bits. + const int32x4_t err0 = vmull_s16(diff_lo, diff_lo); + const int32x4_t err1 = vmlal_s16(err0, diff_hi, diff_hi); + const int64x2_t err2 = vaddl_s32(vget_low_s32(err1), vget_high_s32(err1)); + error = vaddq_s64(error, err2); + coeff += 8; + dqcoeff += 8; + block_size -= 8; + } while (block_size != 0); + + return vgetq_lane_s64(error, 0) + vgetq_lane_s64(error, 1); +} |