开源 企业版 高校版 私有云 模力方舟 AI 队友
代码拉取完成,页面将自动刷新
捐赠
捐赠前请先登录
扫描微信二维码支付
取消
支付完成
支付提示
将跳转至支付宝完成支付
确定
取消
1 Star 0 Fork 9

bugman/ML.NET

加入 Gitee
与超过 1400万 开发者一起发现、参与优秀开源项目,私有仓库也完全免费 :)
免费加入
已有帐号? 立即登录
文件
master
分支 (19)
标签 (13)
master
features/automl
features/IntegrationPackage
release/1.0
eerhardt-patch-1
release/preview
ccbuilddef2
ccbuilddef
montebhoover-docs
newciLeg
danmosemsft-patch-1
tryparse
CESARDELATORRE-patch-1
fixmerge
OSXCIVSTS
RemoveDemandsWindowsCi
add-yaml-ci
samples
release/v0.1
v1.0.0
v1.0.0-preview
v0.11.0
v0.10.0
v0.9.0
v0.8.0
v0.7.0
v0.6.0
v0.5.0
v0.4.0
v0.3.0
v0.2.0
v0.1.0
master
分支 (19)
标签 (13)
master
features/automl
features/IntegrationPackage
release/1.0
eerhardt-patch-1
release/preview
ccbuilddef2
ccbuilddef
montebhoover-docs
newciLeg
danmosemsft-patch-1
tryparse
CESARDELATORRE-patch-1
fixmerge
OSXCIVSTS
RemoveDemandsWindowsCi
add-yaml-ci
samples
release/v0.1
v1.0.0
v1.0.0-preview
v0.11.0
v0.10.0
v0.9.0
v0.8.0
v0.7.0
v0.6.0
v0.5.0
v0.4.0
v0.3.0
v0.2.0
v0.1.0
克隆/下载
克隆/下载
提示
下载代码请复制以下命令到终端执行
为确保你提交的代码身份被 Gitee 正确识别,请执行以下命令完成配置
初次使用 SSH 协议进行代码克隆、推送等操作时,需按下述提示完成 SSH 配置
1 生成 RSA 密钥
2 获取 RSA 公钥内容,并配置到 SSH公钥
在 Gitee 上使用 SVN,请访问 使用指南
使用 HTTPS 协议时,命令行会出现如下账号密码验证步骤。基于安全考虑,Gitee 建议 配置并使用私人令牌 替代登录密码进行克隆、推送等操作
Username for 'https://gitee.com': userName
Password for 'https://userName@gitee.com': # 私人令牌
master
分支 (19)
标签 (13)
master
features/automl
features/IntegrationPackage
release/1.0
eerhardt-patch-1
release/preview
ccbuilddef2
ccbuilddef
montebhoover-docs
newciLeg
danmosemsft-patch-1
tryparse
CESARDELATORRE-patch-1
fixmerge
OSXCIVSTS
RemoveDemandsWindowsCi
add-yaml-ci
samples
release/v0.1
v1.0.0
v1.0.0-preview
v0.11.0
v0.10.0
v0.9.0
v0.8.0
v0.7.0
v0.6.0
v0.5.0
v0.4.0
v0.3.0
v0.2.0
v0.1.0
segment.cpp 13.46 KB
一键复制 编辑 原始数据 按行查看 历史
dotnet-bot 提交于 2018年05月04日 08:22 +08:00 . Initial commit
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
//------------------------------------------------------------------------------
// <copyright company="Microsoft Corporation">
// Copyright (c) Microsoft Corporation. All rights reserved.
// </copyright>
//------------------------------------------------------------------------------
#include "stdafx.h"
#include <iostream> // for qsort
#include <stdio.h>
#include <tmmintrin.h>
#include <smmintrin.h>
#include <emmintrin.h>
EXPORT_API(void) C_SegmentFindOptimalPath15(_In_reads_(valc) unsigned long* valv, _In_ long valc, _Inout_ long long* pBits, _Inout_ long* pTransitions)
{
unsigned long transmap, bitindex = 0, bestindex = 0;
short bestcost;
long long bits = 40;
// Get the preallocated memory.
__m128i masksu[16];
{
__m128i imask = _mm_set1_epi16(~0);
for (int i = 15; i >= 0; --i)
{
imask = _mm_srli_si128(imask, 1);
_mm_storeu_si128(masksu + i, imask);
}
}
__m128i state0f = _mm_setzero_si128();
for (int i = 0; i < valc; ++i)
{
// The bit ops are a little goofy, but it's a lot better than running conditional tests.
_BitScanReverse((unsigned long*)&bitindex, (((unsigned long)valv[i]) << 1) | 1);
__m128i mask = _mm_load_si128(masksu + bitindex);
__m128i staycost = _mm_adds_epu8(state0f, _mm_setr_epi8(
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15));
__m128i unmasked = _mm_min_epu8(staycost, _mm_setr_epi8(
40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55));
// Find the transition map.
transmap = _mm_movemask_epi8(_mm_cmpeq_epi8(unmasked, staycost));
// Calculate the next base state of state0f.
state0f = _mm_or_si128(mask, unmasked);
// Calculate the min cost position in the current iteration.
bestcost = _mm_extract_epi16(_mm_minpos_epu16(_mm_min_epu16(
_mm_cvtepi8_epi16(state0f), _mm_cvtepi8_epi16(_mm_srli_si128(state0f, 8)))), 0);
// Keep the invariant that the min cost position has 0 cost.
state0f = _mm_or_si128(mask, _mm_sub_epi8(state0f, _mm_set1_epi8((char)bestcost)));
// Find the bit index of the best position.
_BitScanReverse((unsigned long*)&bestindex, (long)(_mm_movemask_epi8(_mm_cmpeq_epi8(state0f, _mm_setzero_si128()))));
// Store the vital statistics.
// [31,27]: best bit index
// [26,22]: min bits to encoded current value (m)
// [21, m]: Transition map (implicitly starts at bit 0)
// ( m, 0]: The value
valv[i] |= ((bestindex << 27) | (bitindex << 22) | ((((unsigned int)transmap) >> bitindex) << bitindex));
bits += bestcost;
}
long back = 0, bitness = 0, transitions = 0;
for (int i = valc - 1; i >= 0; --i)
{
bitness = back ? bitness : (valv[i] >> 27);
transitions += 1 - back;
back = (valv[i] >> bitness) & 1;
valv[i] &= (1 << (valv[i] >> 22)) - 1;
valv[i] |= (bitness << 27);
}
*pBits = bits;
*pTransitions = transitions;
}
EXPORT_API(void) C_SegmentFindOptimalPath21(_In_reads_(valc) unsigned long* valv, _In_ long valc, _Inout_ long long* pBits, _Inout_ long* pTransitions)
{
unsigned long transmap, bitindex = 0, bestindex = 0;
short bestcost;
long long bits = 40;
unsigned long* end = valv + valc;
__m128i statelo = _mm_setzero_si128(), statehi = _mm_setzero_si128();
// Get the preallocated memory.
__m128i masksu[32 + 16];
{
__m128i imask = _mm_set1_epi16(~0);
for (int i = 47; i >= 32; --i)
{
_mm_storeu_si128(masksu + i, imask);
}
for (int i = 31; i >= 16; --i)
{
imask = _mm_srli_si128(imask, 1);
_mm_storeu_si128(masksu + i, imask);
}
for (int i = 15; i >= 0; --i)
{
_mm_storeu_si128(masksu + i, _mm_setzero_si128());
}
}
for (int i = 0; i < valc; ++i)
{
// The bit ops are a little goofy, but it's a lot better than running conditional tests.
_BitScanReverse((unsigned long*)&bitindex, (((unsigned long)valv[i]) << 1) | 1);
__m128i masklo = _mm_load_si128(masksu + bitindex + 16);
// Low order bit calculation.
__m128i staycost = _mm_adds_epu8(statelo, _mm_setr_epi8(
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15));
__m128i unmasked = _mm_min_epu8(staycost, _mm_setr_epi8(
40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55));
// Find the transition map.
transmap = _mm_movemask_epi8(_mm_cmpeq_epi8(unmasked, staycost));
// Calculate the next base state.
statelo = _mm_or_si128(masklo, unmasked);
// High order bit calculation.
__m128i maskhi = _mm_load_si128(masksu + bitindex);
// Low order bit calculation.
staycost = _mm_adds_epu8(statehi, _mm_setr_epi8(
16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31));
unmasked = _mm_min_epu8(staycost, _mm_setr_epi8(
56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71));
// Find the transition map.
transmap |= _mm_movemask_epi8(_mm_cmpeq_epi8(unmasked, staycost)) << 16;
// Calculate the next base state.
statehi = _mm_or_si128(maskhi, unmasked);
// Calculate the min cost position in the current iteration.
bestcost = _mm_extract_epi16(_mm_minpos_epu16(_mm_min_epu16(
_mm_min_epu16(_mm_cvtepi8_epi16(statelo), _mm_cvtepi8_epi16(_mm_srli_si128(statelo, 8))),
_mm_min_epu16(_mm_cvtepi8_epi16(statehi), _mm_cvtepi8_epi16(_mm_srli_si128(statehi, 8))))), 0);
// Keep the invariant that the min cost position has 0 cost.
statelo = _mm_or_si128(masklo, _mm_sub_epi8(statelo, _mm_set1_epi8((char)bestcost)));
statehi = _mm_or_si128(maskhi, _mm_sub_epi8(statehi, _mm_set1_epi8((char)bestcost)));
// Find the bit index of the best position.
_BitScanReverse((unsigned long*)&bestindex, (long)(_mm_movemask_epi8(_mm_cmpeq_epi8(statelo, _mm_setzero_si128())) | (_mm_movemask_epi8(_mm_cmpeq_epi8(statehi, _mm_setzero_si128())) << 16)));
// Store the vital statistics.
// [31,27]: best bit index
// [26,22]: min bits to encoded current value (m)
// [21, m]: Transition map (implicitly starts at bit 0)
// ( m, 0]: The value
valv[i] |= ((bestindex << 27) | (bitindex << 22) | ((((unsigned int)transmap & 0x003fffff) >> bitindex) << bitindex));
bits += bestcost;
}
long back = 0, bitness = 0, transitions = 0;
for (int i = valc - 1; i >= 0; --i)
{
bitness = back ? bitness : (valv[i] >> 27);
transitions += 1 - back;
back = (valv[i] >> bitness) & 1;
valv[i] &= (1 << (valv[i] >> 22)) - 1;
valv[i] |= (bitness << 27);
}
*pBits = bits;
*pTransitions = transitions;
}
EXPORT_API(void) C_SegmentFindOptimalPath7(_In_reads_(valc) unsigned long* valv, _In_ long valc, _Inout_ long long* pBits, _Inout_ long* pTransitions)
{
// TODO: Speed this up.
// In principle it seems like, with fewer and simpler operations,
// and fewer conversions from 8-bit to 16-bit, that this should be
// faster than the 15-bit version of this function, but this does
// not seem to actually be the case? This should be improved.
unsigned long transmap, bitindex = 0, bestindex = 0;
short bestcost;
long long bits = 40;
unsigned long* end = valv + valc;
__m128i state0f, stay, transition;
state0f = _mm_setzero_si128();
stay = _mm_setr_epi16(0, 1, 2, 3, 4, 5, 6, 7);
transition = _mm_add_epi16(_mm_set1_epi16(40), stay);
// Get the preallocated memory.
__m128i masksu[8];
{
__m128i imask = _mm_set1_epi8(~0);
for (int i = 7; i >= 0; --i)
{
imask = _mm_srli_si128(imask, 2);
_mm_storeu_si128(masksu + i, imask);
}
}
for (int i = 0; i < valc; ++i)
{
// The bit ops are a little goofy, but it's a lot better than running conditional tests.
_BitScanReverse((unsigned long*)&bitindex, (((unsigned long)valv[i]) << 1) | 1);
__m128i mask = _mm_load_si128(masksu + bitindex);
__m128i staycost = _mm_adds_epu16(state0f, stay);
__m128i unmasked = _mm_min_epu16(staycost, transition);
// Find the transition map.
transmap = _mm_movemask_epi8(_mm_packus_epi16(_mm_cmpeq_epi16(unmasked, staycost), _mm_setzero_si128()));
// Calculate the next base state of state0f.
state0f = _mm_or_si128(mask, unmasked);
// Calculate the min cost position in the current iteration.
__m128i min = _mm_minpos_epu16(state0f);
bestcost = _mm_extract_epi16(min, 0);
bestindex = _mm_extract_epi16(min, 1);
// Keep the invariant that the min cost position has 0 cost.
state0f = _mm_or_si128(mask, _mm_sub_epi8(state0f, _mm_set1_epi16(bestcost)));
// Store the vital statistics.
// [31,27]: best bit index
// [26,22]: min bits to encoded current value (m)
// [21, m]: Transition map (implicitly starts at bit 0)
// ( m, 0]: The value
valv[i] |= ((bestindex << 27) | (bitindex << 22) | ((((unsigned int)transmap) >> bitindex) << bitindex));
bits += bestcost;
}
long back = 0, bitness = 0, transitions = 0;
for (int i = valc - 1; i >= 0; --i)
{
bitness = back ? bitness : (valv[i] >> 27);
transitions += 1 - back;
back = (valv[i] >> bitness) & 1;
valv[i] &= (1 << (valv[i] >> 22)) - 1;
valv[i] |= (bitness << 27);
}
*pBits = bits;
*pTransitions = transitions;
}
EXPORT_API(void) C_SegmentFindOptimalCost15(_In_reads_(valc) unsigned int* valv, _In_ const int valc, _Inout_ long* pBits)
{
unsigned int val, numbits, transitions = 0;
__m128i state0f, stay, transition;
long bits = 40;
unsigned long bitindex = 0;
// Get the preallocated masks.
__m128i masksu[16];
stay = _mm_set1_epi16(~0);
for (int i = 15; i >= 0; --i)
{
stay = _mm_srli_si128(stay, 1);
_mm_storeu_si128(masksu + i, stay);
}
state0f = _mm_setzero_si128();
stay = _mm_setr_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15);
transition = _mm_add_epi8(_mm_set1_epi8(40), stay);
for (int i = 0; i < valc; ++i)
{
val = valv[i];
bitindex = 0;
_BitScanReverse((unsigned long*)&bitindex, (((unsigned long)valv[i]) << 1) | 1);
numbits = bitindex;
__m128i mask = _mm_load_si128(masksu + numbits);
// Calculate the next base state of state0f.
state0f = _mm_or_si128(mask, _mm_min_epu8(_mm_adds_epu8(state0f, stay), transition));
// Calculate the min cost position in the current iteration.
short bestcost = _mm_extract_epi16(_mm_minpos_epu16(_mm_min_epu16(
_mm_cvtepi8_epi16(state0f), _mm_cvtepi8_epi16(_mm_srli_si128(state0f, 8)))), 0);
// Keep the invariant that the min cost position has 0 cost.
state0f = _mm_or_si128(mask, _mm_sub_epi8(state0f, _mm_set1_epi8((char)bestcost)));
// Find the position of the best position.
bits += bestcost;
}
*pBits = bits;
}
EXPORT_API(void) C_SegmentFindOptimalCost31(_In_reads_(valc) unsigned long* valv, _In_ long valc, _Inout_ long long* pBits)
{
unsigned long bitindex = 0, bestindex = 0;
short bestcost;
long long bits = 40;
unsigned long* end = valv + valc;
__m128i statelo = _mm_setzero_si128(), statehi = _mm_setzero_si128();
// Get the preallocated memory.
__m128i masksu[32 + 16];
{
__m128i imask = _mm_set1_epi16(~0);
for (int i = 47; i >= 32; --i)
{
_mm_storeu_si128(masksu + i, imask);
}
for (int i = 31; i >= 16; --i)
{
imask = _mm_srli_si128(imask, 1);
_mm_storeu_si128(masksu + i, imask);
}
for (int i = 15; i >= 0; --i)
{
_mm_storeu_si128(masksu + i, _mm_setzero_si128());
}
}
for (int i = 0; i < valc; ++i)
{
// The bit ops are a little goofy, but it's a lot better than running conditional tests.
_BitScanReverse((unsigned long*)&bitindex, (((unsigned long)valv[i]) << 1) | 1);
__m128i masklo = _mm_load_si128(masksu + bitindex + 16);
statelo = _mm_or_si128(masklo, _mm_min_epu8(
_mm_adds_epu8(statelo, _mm_setr_epi8(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)),
_mm_setr_epi8(40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55)));
// High order bit calculation.
__m128i maskhi = _mm_load_si128(masksu + bitindex);
statehi = _mm_or_si128(maskhi, _mm_min_epu8(
_mm_adds_epu8(statehi, _mm_setr_epi8(16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31)),
_mm_setr_epi8(56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71)));
// Calculate the min cost position in the current iteration.
bestcost = _mm_extract_epi16(_mm_minpos_epu16(_mm_min_epu16(
_mm_min_epu16(_mm_cvtepi8_epi16(statelo), _mm_cvtepi8_epi16(_mm_srli_si128(statelo, 8))),
_mm_min_epu16(_mm_cvtepi8_epi16(statehi), _mm_cvtepi8_epi16(_mm_srli_si128(statehi, 8))))), 0);
// Keep the invariant that the min cost position has 0 cost.
statelo = _mm_or_si128(masklo, _mm_sub_epi8(statelo, _mm_set1_epi8((char)bestcost)));
statehi = _mm_or_si128(maskhi, _mm_sub_epi8(statehi, _mm_set1_epi8((char)bestcost)));
bits += bestcost;
}
*pBits = bits;
}
Loading...
举报
举报成功
我们将于2个工作日内通过站内信反馈结果给你!
请认真填写举报原因,尽可能描述详细。
请选择举报类型
取消
发送
误判申诉

此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。

如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。

取消
提交

简介

ML.NET 是一个跨平台的开源机器学习框架,旨在让 .NET 开发者更快上手机器学习
取消

发行版

暂无发行版

贡献者

全部

近期动态

不能加载更多了
编辑仓库简介
简介内容
主页
马建仓 AI 助手
尝试更多
代码解读
代码找茬
代码优化
1
https://gitee.com/skygit/ML.NET.git
git@gitee.com:skygit/ML.NET.git
skygit
ML.NET
ML.NET
master
点此查找更多帮助

搜索帮助

评论
仓库举报
回到顶部
登录提示
该操作需登录 Gitee 帐号,请先登录后再操作。
立即登录
没有帐号,去注册

AltStyle によって変換されたページ (->オリジナル) /