diff options
author | Android Build Coastguard Worker <android-build-coastguard-worker@google.com> | 2024-04-18 22:13:46 +0000 |
---|---|---|
committer | Gerrit Code Review <noreply-gerritcodereview@google.com> | 2024-04-18 22:13:46 +0000 |
commit | 30c00556ad8a5c6b6719589948bfda4e9f324b8d (patch) | |
tree | 7aa355fd0b89ec0b2611e17ee84a14c6fa449e22 /source/compare_msa.cc | |
parent | 8e521e2ea589b4cafb788330f83037a0413f200a (diff) | |
parent | f635a4c50514783b67549770dee0c53c9fa0c773 (diff) | |
download | libyuv-androidx-core-animation-release.tar.gz |
Merge "Snap for 11735041 from 488a2af021e3e7473f083a9435b1472c0d411f3d to androidx-core-animation-release" into androidx-core-animation-releaseandroidx-core-animation-release
Diffstat (limited to 'source/compare_msa.cc')
-rw-r--r-- | source/compare_msa.cc | 97 |
1 files changed, 97 insertions, 0 deletions
diff --git a/source/compare_msa.cc b/source/compare_msa.cc new file mode 100644 index 00000000..0b807d37 --- /dev/null +++ b/source/compare_msa.cc @@ -0,0 +1,97 @@ +/* + * Copyright 2017 The LibYuv 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 "libyuv/basic_types.h" + +#include "libyuv/compare_row.h" +#include "libyuv/row.h" + +// This module is for GCC MSA +#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) +#include "libyuv/macros_msa.h" + +#ifdef __cplusplus +namespace libyuv { +extern "C" { +#endif + +uint32_t HammingDistance_MSA(const uint8_t* src_a, + const uint8_t* src_b, + int count) { + uint32_t diff = 0u; + int i; + v16u8 src0, src1, src2, src3; + v2i64 vec0 = {0}, vec1 = {0}; + + for (i = 0; i < count; i += 32) { + src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0); + src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16); + src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0); + src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16); + src0 ^= src2; + src1 ^= src3; + vec0 += __msa_pcnt_d((v2i64)src0); + vec1 += __msa_pcnt_d((v2i64)src1); + src_a += 32; + src_b += 32; + } + + vec0 += vec1; + diff = (uint32_t)__msa_copy_u_w((v4i32)vec0, 0); + diff += (uint32_t)__msa_copy_u_w((v4i32)vec0, 2); + return diff; +} + +uint32_t SumSquareError_MSA(const uint8_t* src_a, + const uint8_t* src_b, + int count) { + uint32_t sse = 0u; + int i; + v16u8 src0, src1, src2, src3; + v8i16 vec0, vec1, vec2, vec3; + v4i32 reg0 = {0}, reg1 = {0}, reg2 = {0}, reg3 = {0}; + v2i64 tmp0; + + for (i = 0; i < count; i += 32) { + src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0); + src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16); + src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0); + src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16); + vec0 = (v8i16)__msa_ilvr_b((v16i8)src2, (v16i8)src0); + vec1 = (v8i16)__msa_ilvl_b((v16i8)src2, (v16i8)src0); + vec2 = (v8i16)__msa_ilvr_b((v16i8)src3, (v16i8)src1); + vec3 = (v8i16)__msa_ilvl_b((v16i8)src3, (v16i8)src1); + vec0 = __msa_hsub_u_h((v16u8)vec0, (v16u8)vec0); + vec1 = __msa_hsub_u_h((v16u8)vec1, (v16u8)vec1); + vec2 = __msa_hsub_u_h((v16u8)vec2, (v16u8)vec2); + vec3 = __msa_hsub_u_h((v16u8)vec3, (v16u8)vec3); + reg0 = __msa_dpadd_s_w(reg0, vec0, vec0); + reg1 = __msa_dpadd_s_w(reg1, vec1, vec1); + reg2 = __msa_dpadd_s_w(reg2, vec2, vec2); + reg3 = __msa_dpadd_s_w(reg3, vec3, vec3); + src_a += 32; + src_b += 32; + } + + reg0 += reg1; + reg2 += reg3; + reg0 += reg2; + tmp0 = __msa_hadd_s_d(reg0, reg0); + sse = (uint32_t)__msa_copy_u_w((v4i32)tmp0, 0); + sse += (uint32_t)__msa_copy_u_w((v4i32)tmp0, 2); + return sse; +} + +#ifdef __cplusplus +} // extern "C" +} // namespace libyuv +#endif + +#endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) |