From e950c9b10134dba075616f9ef4092b317da91069 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Arttu=20Yl=C3=A4-Outinen?= Date: Thu, 27 Jul 2017 11:16:54 +0300 Subject: [PATCH] Add AVX2 implementation for coefficient sum --- src/strategies/avx2/quant-avx2.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/src/strategies/avx2/quant-avx2.c b/src/strategies/avx2/quant-avx2.c index e9c28ef5..652770ff 100644 --- a/src/strategies/avx2/quant-avx2.c +++ b/src/strategies/avx2/quant-avx2.c @@ -504,6 +504,27 @@ void kvz_dequant_avx2(const encoder_state_t * const state, coeff_t *q_coef, coef #endif //COMPILE_INTEL_AVX2 && defined X86_64 +static uint32_t coeff_abs_sum_avx2(const coeff_t *coeffs, const size_t length) +{ + assert(length % 8 == 0); + + __m256i total = _mm256_abs_epi32(_mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*) coeffs))); + + for (int i = 8; i < length; i += 8) { + __m256i temp = _mm256_abs_epi32(_mm256_cvtepi16_epi32(_mm_loadu_si128((__m128i*) &coeffs[i]))); + total = _mm256_add_epi32(total, temp); + } + + __m128i result128 = _mm_add_epi32( + _mm256_castsi256_si128(total), + _mm256_extractf128_si256(total, 1) + ); + + uint32_t parts[4]; + _mm_storeu_si128((__m128i*) parts, result128); + + return parts[0] + parts[1] + parts[2] + parts[3]; +} int kvz_strategy_register_quant_avx2(void* opaque, uint8_t bitdepth) { @@ -515,6 +536,7 @@ int kvz_strategy_register_quant_avx2(void* opaque, uint8_t bitdepth) success &= kvz_strategyselector_register(opaque, "quantize_residual", "avx2", 40, &kvz_quantize_residual_avx2); success &= kvz_strategyselector_register(opaque, "dequant", "avx2", 40, &kvz_dequant_avx2); } + success &= kvz_strategyselector_register(opaque, "coeff_abs_sum", "avx2", 0, &coeff_abs_sum_avx2); #endif //COMPILE_INTEL_AVX2 && defined X86_64 return success;