hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-23 18:14:06 +00:00
Code Issues Actions Projects Releases Wiki Activity

[avx2] check_rd_costs_avx2 done

Browse source
This commit is contained in:
Joose Sainio 2023-04-09 14:05:50 +03:00
parent c6e6f5da33
commit 8b19c468cf
2 changed files with 2 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
CMakeLists.txt
View file

@ -144,6 +144,7 @@ target_include_directories(uvg266 PUBLIC src/extras)
target_include_directories(uvg266 PUBLIC src/strategies)
file(GLOB LIB_SOURCES_STRATEGIES_AVX2 RELATIVE ${PROJECT_SOURCE_DIR} "src/strategies/avx2/*.c")
file(GLOB LIB_SOURCES_STRATEGIES_AVX2 RELATIVE ${PROJECT_SOURCE_DIR} "src/dep_quant.c")
file(GLOB LIB_SOURCES_STRATEGIES_SSE41 RELATIVE ${PROJECT_SOURCE_DIR} "src/strategies/sse41/*.c")
file(GLOB LIB_SOURCES_STRATEGIES_SSE42 RELATIVE ${PROJECT_SOURCE_DIR} "src/strategies/sse42/*.c")

2
src/dep_quant.c
View file

@ -601,7 +601,7 @@ static void check_rd_costs_avx2(const all_depquant_states* const state, const en
__m128i max_rice = _mm_set1_epi32(31);
value = _mm_min_epi32(value, max_rice);
__m128i go_rice_tab = _mm_cvtepi8_epi32(_mm_loadu_si32(&state->m_goRicePar[start]));
go_rice_tab = _mm_slli_epi32(value, 5);
go_rice_tab = _mm_slli_epi32(go_rice_tab, 5);
value = _mm_add_epi32(value, go_rice_tab);
__m128i temp = _mm_add_epi32(coeff_frac_bits, _mm_i32gather_epi32(&g_goRiceBits[0][0], value, 4));