Intel ARCHITECTURE IA-32 User Manual Page 300
- Page / 568
- Table of contents
- BOOKMARKS
Rated. / 5. Based on customer reviews
IA-32 Intel® Architecture Optimization
6-10
The Non-temporal Store Instructions
This section describes the behavior of streaming stores and reiterates
some of the information presented in the previous section. In Streaming
SIMD Extensions, the
movntps, movntpd, movntq, movntdq, movnti,
maskmovq
and maskmovdqu instructions are streaming, non-temporal
stores. With regard to memory characteristics and ordering, they are
similar mostly to the Write-Combining (
WC) memory type:
• Write combining – successive writes to the same cache line are
combined
• Write collapsing – successive writes to the same byte(s) result in
only the last write being visible
• Weakly ordered – no ordering is preserved between WC stores, or
between
WC stores and other loads or stores
• Uncacheable and not write-allocating – stored data is written around
the cache and will not generate a read-for-ownership bus request for
the corresponding cache line
Fencing
Because streaming stores are weakly ordered, a fencing operation is
required to ensure that the stored data is flushed from the processor to
memory. Failure to use an appropriate fence may result in data being
“trapped” within the processor and will prevent visibility of this data by
other processors or system agents. WC stores require software to ensure
coherence of data by performing the fencing operation; see “The fence
Instructions” section for more information.
Streaming Non-temporal Stores
Streaming stores can improve performance in the following ways:
• Increase store bandwidth if 64 bytes that fit within a cache line are
written consecutively, since they do not require read-for-ownership
bus requests and 64 bytes are combined into a single bus write
transaction.
- IA-32 Intel® Architecture 1
- Optimization Reference 1
- Contents 3
- Appendix DStack Alignment 14
- Examples 15
- Introduction 23
- Tuning Your Application 24
- About This Manual 24
- Related Documentation 27
- Notational Conventions 28
- IA-32 Intel 29
- Architecture 29
- Processor Family Overview 29
- SIMD Technology 30
- Y4 Y3 Y2 Y1 31
- OP OP OP OP 31
- • inherently parallel 32
- Summary of SIMD Technologies 33
- Streaming SIMD Extensions 2 34
- Streaming SIMD Extensions 3 34
- Intel NetBurst 36
- Microarchitecture 36
- )UHTXHQWO\XVHGSDWKV 38
- /HVVIUHTXHQWO\XVHGSDWKV 38
- The Front End 39
- The Out-of-order Core 40
- Retirement 40
- Front End Pipeline Detail 41
- Execution Trace Cache 42
- Branch Prediction 43
- Execution Core Detail 44
- Data Prefetch 49
- • multiple outstanding misses 52
- • buffering of writes 52
- Pentium 54
- • fetch/decode unit 55
- • instruction cache 55
- Data Prefetching 57
- Out-of-Order Core 58
- Microarchitecture of Intel 59
- Core™ Solo and 59
- Core™ Duo Processors 59
- • Power-optimized bus 60
- • Data Prefetch 60
- • Micro-op fusion 60
- • operational fairness 64
- Shared Resources 65
- Front End Pipeline 66
- Multi-Core Processors 67
- Load and Store Operations 70
- General Optimization 73
- Optimize Memory Access 77
- Enable Vectorization 79
- Performance Tools 81
- VTune™ Performance Analyzer 82
- Processor Perspectives 83
- A and B. If the condition is 88
- Spin-Wait and Idle Loops 90
- Static Prediction 91
- Inlining, Calls and Returns 94
- Branch Type Selection 95
- Loop Unrolling 98
- • inlining where appropriate 100
- Memory Accesses 101
- Line 029e7100h 103
- Line 029e70c0h 103
- Line 029e7140h 103
- Store Forwarding 104
- Alignment 105
- Figure 2-2 106
- Example 2-14 108
- • parameter passing 110
- Data Layout Optimizations 111
- Stack Alignment 114
- Aliasing Cases in the Pentium 117
- 4 and Intel 117
- Processors 117
- Mixing Code and Data 119
- Write Combining 120
- Locality Enhancement 122
- Minimizing Bus Latency 124
- • software prefetch for data 127
- Cacheability Instructions 128
- Applications 129
- • arithmetic overflow 132
- • arithmetic underflow 132
- • denormalized operand 132
- Floating-point Modes 134
- Core Duo Processors 142
- Memory Operands 143
- Floating-Point Stalls 144
- Instruction Selection 145
- Complex Instructions 146
- Use of the lea Instruction 146
- Flag Register Accesses 147
- Integer Divide 148
- Alternate Sequence without 150
- Partial Register Stall 150
- • Operand size prefix (0x66) 152
- • Address size prefix (0x67) 152
- REP Prefix and Data Movement 153
- • Throughput per iteration: 154
- • Address alignment: 154
- • Cache eviction: 155
- Destination 157
- • immediate constant 158
- • base register 158
- • scaled index register 158
- Clearing Registers 159
- Compares 159
- Floating Point/SIMD Operands 160
- Prolog Sequences 162
- Instruction Scheduling 163
- Spill Scheduling 164
- Vectorization 165
- • avoid global pointers 166
- • avoid global variables 166
- Miscellaneous 167
- User/Source Coding Rules 169
- PUSH, CALL, RET). 2-84 179
- Tuning Suggestions 180
- Coding for SIMD 181
- Architectures 181
- Technologies 182
- bool OSSupportCheck() { 184
- Programming 188
- Identifying Hot Spots 190
- Coding Techniques 192
- Coding Methodologies 193
- Assembly 195
- Intrinsics 195
- +”, “>>”) 197
- Automatic Vectorization 198
- Stack and Data Alignment 200
- __m128* datatypes 202
- __m128* 203
- Compiler-Supported Alignment 204
- Improving Memory Utilization 207
- SoA Data Structure 208
- Strip Mining 212
- Example 3-19 Strip Mined Code 213
- Loop Blocking 214
- Example 3-20 Loop Blocking 215
- A. Original Loop 215
- Blocking 216
- Tuning the Final Application 219
- Optimizing for SIMD Integer 221
- Using the EMMS Instruction 223
- Data Alignment 226
- Signed Unpack 227
- MM/M64 mm 229
- Non-Interleaved Unpack 231
- Extract Word 233
- Insert Word 234
- Figure 4-6 pinsrw Instruction 235
- Move Byte Mask to Integer 236
- 55 47 39 23 15 7 237
- X4 X3 X2 X1 238
- X1 X2 X3 X4 238
- Generating Constants 241
- Building Blocks 243
- Absolute Value 245
- 0x8000800080008000 246
- Highly Efficient Clipping 247
- Signed Word 249
- Packed Multiply High Unsigned 250
- Packed Average (Byte/Word) 251
- Packed 32*32 Multiply 253
- Packed 64-bit Add/Subtract 253
- 128-bit Shifts 253
- Memory Optimizations 254
- Partial Memory Accesses 255
- Instruction 259
- Optimizing for SIMD 263
- Floating-point Applications 263
- Planning Considerations 264
- Scalar Floating-point Code 265
- Data Swizzling 271
- Example 5-3 Swizzling Data 272
- Data Deswizzling 276
- Instructions 277
- Instructions (continued) 278
- Functions 279
- Horizontal ADD Using SSE 280
- C1 C2 C3 C4 D1 D2 D3 D4 281
- C1 C2 D1 D2 C3 C4 D3 D4 281
- MXCSR register should be 283
- SSE3 and Complex Arithmetics 285
- Optimizing Cache Usage 291
- Optimizing Cache Usage 6 293
- Hardware Prefetching of Data 294
- Prefetch 296
- Implementation 298
- Cacheability Control 299
- Fencing 300
- Streaming Non-temporal Stores 300
- WB) or Write-Through (WT) 301
- WC semantics) 301
- Write-Combining 302
- Streaming Store Usage Models 303
- • hand-crafted code 305
- The lfence Instruction 306
- The mfence Instruction 306
- The clflush Instruction 307
- Software-controlled Prefetch 308
- Hardware Prefetch 309
- Constant Stride 311
- Non-Adjacent Passes Loops 326
- 60 invis 332
- • write-once (non-temporal) 333
- Cache Management 334
- Video Encoder 335
- Video Decoder 335
- • alignment of data 337
- • cache size 337
- Bit Location Name Meaning 344
- • Determine prefetch stride 345
- Parameters 346
- Multi-Core and 347
- Hyper-Threading Technology 347
- Performance and Usage Models 348
- Single Thread 349
- Multi-Thread on MP 349
- Multitasking Environment 350
- • workload 352
- • thread interaction 352
- • hardware utilization 352
- • domain decomposition 353
- • functional decomposition 353
- Functional Decomposition 354
- P(1)P(1) C(1)C(1)P(1) 355
- P: producer 356
- C: consumer 356
- Thread 0 358
- Thread 1 358
- Optimization Guidelines 362
- Thread Synchronization 365
- Optimization with Spin-Locks 371
- PAUSE instruction in the 372
- Example 7-5 373
- System Bus Optimization 379
- Conserve Bus Bandwidth 380
- Memory Optimization 384
- Shared-Memory Optimization 385
- 4 KB in each thread 388
- Per-thread Stack Offset 390
- Per-instance Stack Offset 392
- Front-end Optimization 394
- Resources 395
- Processor 397
- Processor (Contd.) 398
- Sharing the Same Cache 401
- 64-bit Mode Coding 409
- Guidelines 409
- Only When Necessary 410
- Assembly/Compiler Coding rule 411
- 64-Bit Arithmetic 412
- Assembly/Compiler Coding Rule 413
- Possible 414
- Using Software Prefetch 414
- Power Optimization for 415
- Mobile Usages 415
- Mobile Usage Scenarios 416
- ACPI C-States 418
- Reducing Amount of Work 423
- • Switch off unused devices 424
- Technology 426
- Enabling Intel 428
- Enhanced Deeper Sleep 428
- Multi-Core Considerations 429
- (C1-C4) 433
- Application Performance 435
- Compilers 436
- Code Optimization Options 437
- Vectorizer Switch Options 439
- Multithreading with OpenMP* 440
- VTune™ Performance Analyzer 442
- Sampling 443
- Event-based Sampling 444
- Workload Characterization 445
- Call Graph 447
- Performance Libraries 448
- Benefits Summary 449
- Optimizations with the Intel 450
- Enhanced Debugger (EDB) 451
- Threading Tools 451
- Thread Profiler 453
- Software College 454
- Using Performance Monitoring 455
- Bogus, Non-bogus, Retire 456
- Bus Ratio 456
- Counting Clocks 458
- Non-Halted Clockticks 459
- Non-Sleep Clockticks 460
- Time Stamp Counter 461
- Microarchitecture Notes 462
- Side Bus 464
- Reads due to program loads 465
- Writebacks (dirty evictions) 466
- Usage Notes on Bus Activities 469
- Tags for replay_event 500
- Tags for front_end_event 502
- Tags for execution_event 502
- Technology 504
- Parallel Counting 505
- Parallel Counting (continued) 506
- Intel Core Duo processors 510
- Ratio Interpretation 511
- Notes on Selected Events 512
- Throughput 515
- Overview 516
- PADDQ and PMULUDQ, each have 517
- Definitions 518
- Latency and Throughput 518
- See “Table Footnotes” 520
- Instructions (continued) 524
- Table Footnotes 533
- Stack Alignment D 539
- & 0x0f) == 0x08 542
- Stack Frame Optimizations 545
- Inlined Assembly and ebx 546
- Mathematics of Prefetch 547
- Scheduling Distance 547
- Mathematical Model for PSD 548
- L2 lookup miss latency 550
- • Optimize T 551
- No Preloading or Prefetch 552
- Front-Side Bus 553
- Execution pipeline 553
- Execution cycles 553
- Compute Bound (Case: T 554
- >= T 556
- INTEL SALES OFFICES 567
Related products and manuals for Computer Accessories Intel ARCHITECTURE IA-32
Computer Accessories Intel 1520 User Manual
(176 pages)
(176 pages)
Computer Accessories Intel 82600 User Manual
(40 pages)
(40 pages)
Computer Accessories Intel 220T User Manual
(24 pages)
(24 pages)
Computer Accessories Intel 520T User Manual
(31 pages)
(31 pages)
Computer Accessories Intel 410T User Manual
(40 pages)
(40 pages)
(144 pages)
Computer Accessories Intel 130T User Manual
(18 pages)
(18 pages)
Computer Accessories Intel 4 User Manual
(10 pages)
(10 pages)
Computer Accessories Intel IA-32 User Manual
(636 pages)
(636 pages)
Computer Accessories Intel AXXSW1GB User Manual
(220 pages)
(220 pages)
Computer Accessories Intel ZT8101 User Manual
(124 pages)
(124 pages)
© 2020, manymanuals.com. All rights reserved. | 1.200 s |
Manymanuals.com
Manymanuals.de
Manymanuals.fr
Manymanuals.it
Manymanuals.pl
Manymanuals.cz
Manymanuals.es
Manymanuals-pt.com
Comments to this Manuals