diff options
author | James Zern <jzern@google.com> | 2015-07-29 00:47:09 +0000 |
---|---|---|
committer | Gerrit Code Review <noreply-gerritcodereview@google.com> | 2015-07-29 00:47:09 +0000 |
commit | f42012e526f218f4607e5d678c5bd44bb5978bff (patch) | |
tree | 88e1a2cd2c282d043a19a523ba5b17799f927a26 /vp9/encoder/arm | |
parent | 4cbf36b10569f3973f653f74b0c1467d07633d02 (diff) | |
parent | ea990af7f5e780bd90d67249a2db631404a42dc1 (diff) | |
download | libvpx-f42012e526f218f4607e5d678c5bd44bb5978bff.tar libvpx-f42012e526f218f4607e5d678c5bd44bb5978bff.tar.gz libvpx-f42012e526f218f4607e5d678c5bd44bb5978bff.tar.bz2 libvpx-f42012e526f218f4607e5d678c5bd44bb5978bff.zip |
Merge "add vp9_block_error_fp_neon"
Diffstat (limited to 'vp9/encoder/arm')
-rw-r--r-- | vp9/encoder/arm/neon/vp9_error_neon.c | 41 |
1 files changed, 41 insertions, 0 deletions
diff --git a/vp9/encoder/arm/neon/vp9_error_neon.c b/vp9/encoder/arm/neon/vp9_error_neon.c new file mode 100644 index 000000000..1c7503139 --- /dev/null +++ b/vp9/encoder/arm/neon/vp9_error_neon.c @@ -0,0 +1,41 @@ +/* + * Copyright (c) 2015 The WebM 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 in the root of the source + * tree. An additional intellectual property rights grant can be found + * in the file PATENTS. All contributing project authors may + * be found in the AUTHORS file in the root of the source tree. + */ + +#include <arm_neon.h> +#include <assert.h> + +#include "./vp9_rtcd.h" + +int64_t vp9_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); +} |