/* * By downloading, copying, installing or using the software you agree to this license. * If you do not agree to this license, do not download, install, * copy or use the software. * * * License Agreement * For Open Source Computer Vision Library * (3-clause BSD License) * * Copyright (C) 2012-2015, NVIDIA Corporation, all rights reserved. * Third party copyrights are property of their respective owners. * * Redistribution and use in source and binary forms, with or without modification, * are permitted provided that the following conditions are met: * * * Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * * Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * * * Neither the names of the copyright holders nor the names of the contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * This software is provided by the copyright holders and contributors "as is" and * any express or implied warranties, including, but not limited to, the implied * warranties of merchantability and fitness for a particular purpose are disclaimed. * In no event shall copyright holders or contributors be liable for any direct, * indirect, incidental, special, exemplary, or consequential damages * (including, but not limited to, procurement of substitute goods or services; * loss of use, data, or profits; or business interruption) however caused * and on any theory of liability, whether in contract, strict liability, * or tort (including negligence or otherwise) arising in any way out of * the use of this software, even if advised of the possibility of such damage. */ #include "common.hpp" namespace CAROTENE_NS { f64 dotProduct(const Size2D &_size, const u8 * src0Base, ptrdiff_t src0Stride, const u8 * src1Base, ptrdiff_t src1Stride) { internal::assertSupportedConfiguration(); #ifdef CAROTENE_NEON Size2D size(_size); if (src0Stride == src1Stride && src0Stride == (ptrdiff_t)(size.width)) { size.width *= size.height; size.height = 1; } // It is possible to accumulate up to 66051 uchar multiplication results in uint32 without overflow // We process 16 elements and accumulate two new elements per step. So we could handle 66051/2*16 elements #define DOT_UINT_BLOCKSIZE 66050*8 f64 result = 0.0; for (size_t row = 0; row < size.height; ++row) { const u8 * src0 = internal::getRowPtr(src0Base, src0Stride, row); const u8 * src1 = internal::getRowPtr(src1Base, src1Stride, row); size_t i = 0; uint64x2_t ws = vmovq_n_u64(0); while(i + 16 <= size.width) { size_t lim = std::min(i + DOT_UINT_BLOCKSIZE, size.width) - 16; uint32x4_t s1 = vmovq_n_u32(0); uint32x4_t s2 = vmovq_n_u32(0); for (; i <= lim; i += 16) { internal::prefetch(src0 + i); internal::prefetch(src1 + i); uint8x16_t vs1 = vld1q_u8(src0 + i); uint8x16_t vs2 = vld1q_u8(src1 + i); uint16x8_t vdot1 = vmull_u8(vget_low_u8(vs1), vget_low_u8(vs2)); uint16x8_t vdot2 = vmull_u8(vget_high_u8(vs1), vget_high_u8(vs2)); s1 = vpadalq_u16(s1, vdot1); s2 = vpadalq_u16(s2, vdot2); } ws = vpadalq_u32(ws, s1); ws = vpadalq_u32(ws, s2); } if(i + 8 <= size.width) { uint8x8_t vs1 = vld1_u8(src0 + i); uint8x8_t vs2 = vld1_u8(src1 + i); ws = vpadalq_u32(ws, vpaddlq_u16(vmull_u8(vs1, vs2))); i += 8; } result += (double)vget_lane_u64(vadd_u64(vget_low_u64(ws), vget_high_u64(ws)), 0); for (; i < size.width; ++i) result += s32(src0[i]) * s32(src1[i]); } return result; #else (void)_size; (void)src0Base; (void)src0Stride; (void)src1Base; (void)src1Stride; return 0; #endif } f64 dotProduct(const Size2D &_size, const s8 * src0Base, ptrdiff_t src0Stride, const s8 * src1Base, ptrdiff_t src1Stride) { internal::assertSupportedConfiguration(); #ifdef CAROTENE_NEON Size2D size(_size); if (src0Stride == src1Stride && src0Stride == (ptrdiff_t)(size.width)) { size.width *= size.height; size.height = 1; } // It is possible to accumulate up to 131071 schar multiplication results in sint32 without overflow // We process 16 elements and accumulate two new elements per step. So we could handle 131071/2*16 elements #define DOT_INT_BLOCKSIZE 131070*8 f64 result = 0.0; for (size_t row = 0; row < size.height; ++row) { const s8 * src0 = internal::getRowPtr(src0Base, src0Stride, row); const s8 * src1 = internal::getRowPtr(src1Base, src1Stride, row); size_t i = 0; int64x2_t ws = vmovq_n_s64(0); while(i + 16 <= size.width) { size_t lim = std::min(i + DOT_UINT_BLOCKSIZE, size.width) - 16; int32x4_t s1 = vmovq_n_s32(0); int32x4_t s2 = vmovq_n_s32(0); for (; i <= lim; i += 16) { internal::prefetch(src0 + i); internal::prefetch(src1 + i); int8x16_t vs1 = vld1q_s8(src0 + i); int8x16_t vs2 = vld1q_s8(src1 + i); int16x8_t vdot1 = vmull_s8(vget_low_s8(vs1), vget_low_s8(vs2)); int16x8_t vdot2 = vmull_s8(vget_high_s8(vs1), vget_high_s8(vs2)); s1 = vpadalq_s16(s1, vdot1); s2 = vpadalq_s16(s2, vdot2); } ws = vpadalq_s32(ws, s1); ws = vpadalq_s32(ws, s2); } if(i + 8 <= size.width) { int8x8_t vs1 = vld1_s8(src0 + i); int8x8_t vs2 = vld1_s8(src1 + i); ws = vpadalq_s32(ws, vpaddlq_s16(vmull_s8(vs1, vs2))); i += 8; } result += (double)vget_lane_s64(vadd_s64(vget_low_s64(ws), vget_high_s64(ws)), 0); for (; i < size.width; ++i) result += s32(src0[i]) * s32(src1[i]); } return result; #else (void)_size; (void)src0Base; (void)src0Stride; (void)src1Base; (void)src1Stride; return 0; #endif } f64 dotProduct(const Size2D &_size, const f32 * src0Base, ptrdiff_t src0Stride, const f32 * src1Base, ptrdiff_t src1Stride) { internal::assertSupportedConfiguration(); #ifdef CAROTENE_NEON Size2D size(_size); if (src0Stride == src1Stride && src0Stride == (ptrdiff_t)(size.width * sizeof(f32))) { size.width *= size.height; size.height = 1; } #define DOT_FLOAT_BLOCKSIZE (1 << 13) f64 result = 0.0; for (size_t row = 0; row < size.height; ++row) { const f32 * src0 = internal::getRowPtr(src0Base, src0Stride, row); const f32 * src1 = internal::getRowPtr(src1Base, src1Stride, row); size_t i = 0; while(i + 4 <= size.width) { size_t lim = std::min(i + DOT_FLOAT_BLOCKSIZE, size.width) - 4; float32x4_t v_sum = vdupq_n_f32(0.0f); for( ; i <= lim; i += 4 ) { internal::prefetch(src0 + i); internal::prefetch(src1 + i); v_sum = vmlaq_f32(v_sum, vld1q_f32(src0 + i), vld1q_f32(src1 + i)); } float32x2_t vres = vpadd_f32(vget_low_f32(v_sum),vget_high_f32(v_sum)); result += vget_lane_f32(vres, 0) + vget_lane_f32(vres, 1); } if(i + 2 <= size.width) { float32x2_t vres = vmul_f32(vld1_f32(src0 + i), vld1_f32(src1 + i)); result += vget_lane_f32(vres, 0) + vget_lane_f32(vres, 1); i += 2; } for (; i < size.width; ++i) result += src0[i] * src1[i]; } return result; #else (void)_size; (void)src0Base; (void)src0Stride; (void)src1Base; (void)src1Stride; return 0; #endif } } // namespace CAROTENE_NS