memcpy多字节字节赋值问题

以前没注意过多字节赋值问题，考虑的角度是充分利用 cpu 的 32 位带宽。一次复制 1 个字节和一次复制 4 个字节占用的 cpu 指令周期是一样的，既然我们的 cpu 能同时操作 32 位，为什么我们不能一次复制 4 个字节呢？如此一来，总运行的指令数将降低到原来的 1/4 ！

于是有下面代码（henix转自http://www.embedded.com/columns/technicalinsights/19205567?_requestid=212290）：

代码

 1 void *memcpy(void *dest, void *src, size_t count)
 2 {
 3     size_t c = count >> 2; /* 相当于 count / 4 */
 4     long *pt;
 5     const long *ps;
 6     if (c)
 7     {
 8         /* 首先按 4 个字节复制 */
 9         pt = (long *) dest;
10         ps = (const long *) src;
11         while (c)
12         {
13             *pt = *ps;
14             pt++;
15             ps++;
16             c--;
17         }
18     }
19     c = count & 3; /* 得到 count 模 4 的余数 */
20     if (c)
21     {
22         /* 如果有剩下的，再按 1 个字节复制剩余的 */
23         char *pct = (char *) pt;
24         const char *pcs = (const char *) ps;
25         while (c)
26         {
27             *pct = *pcs;
28             pct++;
29             pcs++;
30             c--;
31         }
32     }
33     return dest;
34 }

通过使用 long 型指针，强行按 4 个字节复制。

还有对 3 种 memcpy 实现的对比，上述算法受到地址对齐的影响较为严重。在地址按4 字节对齐的时候，上述算法的效率比单字节 memcpy 实现高很多，但如果地址没有按4 字节对齐，则其效率并不高，有时甚至还比普通 memcpy 还低。这可能是因为，虽然上述算法减少了 cpu 的指令数，但内存的速度比 cpu 慢得多，速度的瓶颈还是在内存。

刚看了一份memcpy的实现，为了解决字节对齐的问题，从128开始，依次减少，直到1个字节。原文在http://code.google.com/p/dingoo-sdk/source/browse/trunk/dingoo_sdk/src/libc/memcpy.c?spec=svn180&r=180

贴出如下：

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

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

#include "string.h" 
 #include <stdint.h> 
   
   
   
 inline void _memcpy_1x1(void* dst, const void* src) { 
         asm( 
                 "lb $t0, 0(%0)\n\t"
                 "sb $t0, 0(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
 inline void _memcpy_2x1(void* dst, const void* src) { 
         asm( 
                 "lh $t0, 0(%0)\n\t"
                 "sh $t0, 0(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
 inline void _memcpy_4x1(void* dst, const void* src) { 
         asm( 
                 "lw $t0, 0(%0)\n\t"
                 "sw $t0, 0(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
 inline void _memcpy_4x2(void* dst, const void* src) { 
         asm( 
                 "lw $t0, 0(%0)\n\t"
                 "lw $t1, 4(%0)\n\t"
                 "sw $t0, 0(%1)\n\t"
                 "sw $t1, 4(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
 inline void _memcpy_4x4(void* dst, const void* src) { 
         asm( 
                 "lw $t0,  0(%0)\n\t"
                 "lw $t1,  4(%0)\n\t"
                 "lw $t2,  8(%0)\n\t"
                 "lw $t3, 12(%0)\n\t"
                 "sw $t0,  0(%1)\n\t"
                 "sw $t1,  4(%1)\n\t"
                 "sw $t2,  8(%1)\n\t"
                 "sw $t3, 12(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
 inline void _memcpy_4x8(void* dst, const void* src) { 
         asm( 
                 "lw $t0,  0(%0)\n\t"
                 "lw $t1,  4(%0)\n\t"
                 "lw $t2,  8(%0)\n\t"
                 "lw $t3, 12(%0)\n\t"
                 "lw $t4, 16(%0)\n\t"
                 "lw $t5, 20(%0)\n\t"
                 "lw $t6, 24(%0)\n\t"
                 "lw $t7, 28(%0)\n\t"
                 "sw $t0,  0(%1)\n\t"
                 "sw $t1,  4(%1)\n\t"
                 "sw $t2,  8(%1)\n\t"
                 "sw $t3, 12(%1)\n\t"
                 "sw $t4, 16(%1)\n\t"
                 "sw $t5, 20(%1)\n\t"
                 "sw $t6, 24(%1)\n\t"
                 "sw $t7, 28(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
 inline void _memcpy_4x16(void* dst, const void* src) { 
         asm( 
                 "lw $t0,  0(%0)\n\t"
                 "lw $t1,  4(%0)\n\t"
                 "lw $t2,  8(%0)\n\t"
                 "lw $t3, 12(%0)\n\t"
                 "lw $t4, 16(%0)\n\t"
                 "lw $t5, 20(%0)\n\t"
                 "lw $t6, 24(%0)\n\t"
                 "lw $t7, 28(%0)\n\t"
                 "lw $t8, 32(%0)\n\t"
                 "lw $t9, 36(%0)\n\t"
   
                 "sw $t0,  0(%1)\n\t"
                 "lw $t0, 40(%0)\n\t"
                 "sw $t1,  4(%1)\n\t"
                 "lw $t1, 44(%0)\n\t"
                 "sw $t2,  8(%1)\n\t"
                 "lw $t2, 48(%0)\n\t"
                 "sw $t3, 12(%1)\n\t"
                 "lw $t3, 52(%0)\n\t"
                 "sw $t4, 16(%1)\n\t"
                 "lw $t4, 56(%0)\n\t"
                 "sw $t5, 20(%1)\n\t"
                 "lw $t5, 60(%0)\n\t"
   
                 "sw $t6, 24(%1)\n\t"
                 "sw $t7, 28(%1)\n\t"
                 "sw $t8, 32(%1)\n\t"
                 "sw $t9, 36(%1)\n\t"
   
                 "sw $t0, 40(%1)\n\t"
                 "sw $t1, 44(%1)\n\t"
                 "sw $t2, 48(%1)\n\t"
                 "sw $t3, 52(%1)\n\t"
                 "sw $t4, 56(%1)\n\t"
                 "sw $t5, 60(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
 inline void _memcpy_4x32(void* dst, const void* src) { 
         asm( 
                 "lw $t0,   0(%0)\n\t"
                 "lw $t1,   4(%0)\n\t"
                 "lw $t2,   8(%0)\n\t"
                 "lw $t3,  12(%0)\n\t"
                 "lw $t4,  16(%0)\n\t"
                 "lw $t5,  20(%0)\n\t"
                 "lw $t6,  24(%0)\n\t"
                 "lw $t7,  28(%0)\n\t"
                 "lw $t8,  32(%0)\n\t"
                 "lw $t9,  36(%0)\n\t"
   
                 "sw $t0,   0(%1)\n\t"
                 "lw $t0,  40(%0)\n\t"
                 "sw $t1,   4(%1)\n\t"
                 "lw $t1,  44(%0)\n\t"
                 "sw $t2,   8(%1)\n\t"
                 "lw $t2,  48(%0)\n\t"
                 "sw $t3,  12(%1)\n\t"
                 "lw $t3,  52(%0)\n\t"
                 "sw $t4,  16(%1)\n\t"
                 "lw $t4,  56(%0)\n\t"
                 "sw $t5,  20(%1)\n\t"
                 "lw $t5,  60(%0)\n\t"
                 "sw $t6,  24(%1)\n\t"
                 "lw $t6,  64(%0)\n\t"
                 "sw $t7,  28(%1)\n\t"
                 "lw $t7,  68(%0)\n\t"
                 "sw $t8,  32(%1)\n\t"
                 "lw $t8,  72(%0)\n\t"
                 "sw $t9,  36(%1)\n\t"
                 "lw $t9,  76(%0)\n\t"
   
                 "sw $t0,  40(%1)\n\t"
                 "lw $t0,  80(%0)\n\t"
                 "sw $t1,  44(%1)\n\t"
                 "lw $t1,  84(%0)\n\t"
                 "sw $t2,  48(%1)\n\t"
                 "lw $t2,  88(%0)\n\t"
                 "sw $t3,  52(%1)\n\t"
                 "lw $t3,  92(%0)\n\t"
                 "sw $t4,  56(%1)\n\t"
                 "lw $t4,  96(%0)\n\t"
                 "sw $t5,  60(%1)\n\t"
                 "lw $t5, 100(%0)\n\t"
                 "sw $t6,  64(%1)\n\t"
                 "lw $t6, 104(%0)\n\t"
                 "sw $t7,  68(%1)\n\t"
                 "lw $t7, 108(%0)\n\t"
                 "sw $t8,  72(%1)\n\t"
                 "lw $t8, 112(%0)\n\t"
                 "sw $t9,  76(%1)\n\t"
                 "lw $t9, 116(%0)\n\t"
   
                 "sw $t0,  80(%1)\n\t"
                 "lw $t0, 120(%0)\n\t"
                 "sw $t1,  84(%1)\n\t"
                 "lw $t1, 124(%0)\n\t"
   
                 "sw $t2,  88(%1)\n\t"
                 "sw $t3,  92(%1)\n\t"
                 "sw $t4,  96(%1)\n\t"
                 "sw $t5, 100(%1)\n\t"
                 "sw $t6, 104(%1)\n\t"
                 "sw $t7, 108(%1)\n\t"
                 "sw $t8, 112(%1)\n\t"
                 "sw $t9, 116(%1)\n\t"
   
                 "sw $t0, 120(%1)\n\t"
                 "sw $t1, 124(%1)\n\t"
                 : : "r"(dst), "r"(src) 
         ); 
 } 
   
   
   
 inline void _memcpy_down_1(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 1) { 
                 _memcpy_1x1(dst, src); 
                 dst += 1; 
                 src += 1; 
         } 
 } 
   
 inline void _memcpy_down_2(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 2) { 
                 _memcpy_2x1(dst, src); 
                 dst += 2; 
                 src += 2; 
         } 
         _memcpy_down_1(dst, src, dst_end); 
 } 
   
 inline void _memcpy_down_4(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 4) { 
                 _memcpy_4x1(dst, src); 
                 dst += 4; 
                 src += 4; 
         } 
         _memcpy_down_2(dst, src, dst_end); 
 } 
   
 inline void _memcpy_down_8(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 8) { 
                 _memcpy_4x2(dst, src); 
                 dst += 8; 
                 src += 8; 
         } 
         _memcpy_down_4(dst, src, dst_end); 
 } 
   
 inline void _memcpy_down_16(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 16) { 
                 _memcpy_4x4(dst, src); 
                 dst += 16; 
                 src += 16; 
         } 
         _memcpy_down_8(dst, src, dst_end); 
 } 
   
 inline void _memcpy_down_32(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 32) { 
                 _memcpy_4x8(dst, src); 
                 dst += 32; 
                 src += 32; 
         } 
         _memcpy_down_16(dst, src, dst_end); 
 } 
   
 inline void _memcpy_down_64(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 64) { 
                 _memcpy_4x16(dst, src); 
                 dst += 64; 
                 src += 64; 
         } 
         _memcpy_down_32(dst, src, dst_end); 
 } 
   
 inline void _memcpy_down_128(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         if(((uintptr_t)dst_end - (uintptr_t)dst) >= 128) { 
                 _memcpy_4x32(dst, src); 
                 dst += 128; 
                 src += 128; 
         } 
         _memcpy_down_64(dst, src, dst_end); 
 } 
   
   
   
 inline void _memcpy_1(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 1); dst += 1, src += 1) 
                 _memcpy_1x1(dst, src); 
 } 
   
 inline void _memcpy_2(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 2); dst += 2, src += 2) 
                 _memcpy_2x1(dst, src); 
         _memcpy_down_1(dst, src, dst_end); 
 } 
   
 inline void _memcpy_4(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 4); dst += 4, src += 4) 
                 _memcpy_4x1(dst, src); 
         _memcpy_down_2(dst, src, dst_end); 
 } 
   
 inline void _memcpy_8(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 8); dst += 8, src += 8) 
                 _memcpy_4x2(dst, src); 
         _memcpy_down_4(dst, src, dst_end); 
 } 
   
 inline void _memcpy_16(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 16); dst += 16, src += 16) 
                 _memcpy_4x4(dst, src); 
         _memcpy_down_8(dst, src, dst_end); 
 } 
   
 inline void _memcpy_32(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 32); dst += 32, src += 32) 
                 _memcpy_4x8(dst, src); 
         _memcpy_down_16(dst, src, dst_end); 
 } 
   
 inline void _memcpy_64(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 64); dst += 64, src += 64) 
                 _memcpy_4x16(dst, src); 
         _memcpy_down_32(dst, src, dst_end); 
 } 
   
 inline void _memcpy_128(uint8_t* dst, const uint8_t* src, uint8_t* dst_end) { 
         for(; dst <= (dst_end - 128); dst += 128, src += 128) 
                 _memcpy_4x32(dst, src); 
         _memcpy_down_64(dst, src, dst_end); 
 } 
   
   
   
 void* _memcpy_fast(void* dst, const void* src, uintptr_t size) { 
         if((dst == NULL) || (src == NULL)) 
                 return NULL; 
   
         uint8_t*       _dst = (uint8_t*)dst; 
         uint8_t*       _dst_end = &_dst[size]; 
         const uint8_t* _src = (const uint8_t*)src; 
   
         if((((uintptr_t)_dst & 1) != 0) && (((uintptr_t)_dst & 1) == ((uintptr_t)_src & 1)) && ((_dst + 1) <= _dst_end)) { 
                 _memcpy_1x1(_dst, _src); 
                 _dst += 1; _src += 1; 
         } 
         if(((uintptr_t)_dst & 1) == 0) { 
                 if((((uintptr_t)_dst & 3) != 0) && (((uintptr_t)_dst & 3) == ((uintptr_t)_src & 3)) && ((_dst + 2) <= _dst_end)) { 
                         _memcpy_2x1(_dst, _src); 
                         _dst += 2; _src += 2; 
                 } 
                 if(((uintptr_t)_dst & 3) == 0) { 
                         if((((uintptr_t)_dst & 7) != 0) && (((uintptr_t)_dst & 7) == ((uintptr_t)_src & 7)) && ((_dst + 4) <= _dst_end)) { 
                                 _memcpy_4x1(_dst, _src); 
                                 _dst += 4; _src += 4; 
                         } 
                         if(((uintptr_t)_dst & 7) == 0) { 
                                 _memcpy_128(_dst, _src, _dst_end); 
                         } else _memcpy_4(_dst, _src, _dst_end); 
                 } else _memcpy_2(_dst, _src, _dst_end); 
         } else _memcpy_1(_dst, _src, _dst_end); 
   
         return dst; 
 } 