Manuals / Brands / Computer Equipment / Computer Hardware / AMD / Computer Equipment / Computer Hardware

AMD 250 - page 382

1 384

Download 384 pages, 3.32 Mb

366 SSE and SSE2 Optimizations Appendix E

25112 Rev. 3.06 September 2005

Software Optimization Guide for AMD64 Processors

Contents

Main Page Contents Page Page Page Page Page Page Page Tab le s Page Figures Page Revision History xv Revision History Page Chapter 1 Introduction 1.1 Intended Audience 1.2 Getting Started Quickly 1.3 Using This Guide Special Information Numbering Systems Typographic Notation Providing Feedback 1.4 Important New Terms Primitive Operations Internal Instruction Formats Chapter 1 Introduction 5 Types of Instructions 1.5 Key Optimizations Guideline Key Optimizations by Rank Chapter 2 C and C++ Source-Level Optimizations Page 2.1 Declarations of Floating-Point Values 2.2 Using Arrays and Pointers Chapter 2 C and C++ Source-Level Optimizations 11 Avoid code, such as the following, which uses pointer notation: Additional Considerations 2.3 Unrolling Small Loops 2.4 Expression Order in Compound Branch Conditions Page 2.5 Long Logical Expressions in If Statements 2.6 Arrange Boolean Operands for Quick Expression Evaluation Page 2.7 Dynamic Memory Allocation Consideration 2.8 Unnecessary Store-to-Load Dependencies Page 2.9 Matching Store and Load Size Chapter 2 C and C++ Source-Level Optimizations 23 Listing 6. Preferred 24 C and C++ Source-Level Optimizations Chapter 2 The object code generated for the source code between label1 and label2 typically looks like this: 2.10 SWITCH and Noncontiguous Case Expressions 26 C and C++ Source-Level Optimizations Chapter 2 Page 2.11 Arranging Cases by Probability of Occurrence 2.12 Use of Function Prototypes 2.13 Use of const Type Qualifier 2.14 Generic Loop Hoisting Rationale and Examples Page Page 2.15 Local Static Functions 2.16 Explicit Parallelism in Code Rationale and Examples Page 2.17 Extracting Common Subexpressions Page 2.18 Sorting and Padding C and C++ Structures Page 2.19 Sorting Local Variables Page 2.20 Replacing Integer Division with Multiplication 2.21 Frequently Dereferenced Pointer Arguments Chapter 2 C and C++ Source-Level Optimizations 45 Listing 17. Preferred 2.22 Array Indices 2.23 32-Bit Integral Data Types 2.24 Sign of Integer Operands Page 2.25 Accelerating Floating-Point Division and Square Root Chapter 2 C and C++ Source-Level Optimizations 51 Listing 22. 2.26 Fast Floating-Point-to-Integer Conversion Page 2.27 Speeding Up Branches Based on Comparisons Between Floats Page Page 2.28 Improving Performance in Linux Libraries Page Chapter 3 General 64-Bit Optimizations 3.1 64-Bit Registers and Integer Arithmetic Page 3.2 64-Bit Arithmetic and Large-Integer Multiplication Background Page 64 General 64-Bit Optimizations Chapter 3 Chapter 3 General 64-Bit Optimizations 65 66 General 64-Bit Optimizations Chapter 3 Page 3.4 32-Bit Legacy GPRs and Small Unsigned Integers Caution Page Page Chapter 4 Instruction-Decoding Optimizations 4.1 DirectPath Instructions 4.2 Load-Execute Instructions 4.2.1 Load-Execute Integer Instructions Page Page 4.3 Branch Targets in Program Hot Spots 4.4 32/64-Bit vs. 16-Bit Forms of the LEA Instruction 4.5 Take Advantage of x86 and AMD64 Complex Addressing Modes Page 4.6 Short Instruction Encodings 4.7 Partial-Register Reads and Writes Page 4.8 Using LEAVE for Function Epilogues Background Page 4.9 Alternatives to SHLD Instruction 86 Instruction-Decoding Optimizations Chapter 4 Example 3 Replace this instruction: with this code sequence: 4.10 8-Bit Sign-Extended Immediate Values 4.11 8-Bit Sign-Extended Displacements 4.12 Code Padding with Operand-Size Override and NOP Page Chapter 5 Cache and Memory Optimizations 5.1 Memory-Size Mismatches ExamplesStore-to-Load-Forwarding Stalls ExamplesLarge-to-small Mismatches Page 5.2 Natural Alignment of Data Objects 5.3 Cache-Coherent Nonuniform Memory Access (ccNUMA) Page OS Implications 5.4 Multiprocessor Considerations 5.5 Store-to-Load Forwarding Restrictions Store-to-Load Forwarding PitfallsTrue Dependencies Narrow-to-Wide Store-Buffer Data-Forwarding Restriction Wide-to-Narrow Store-Buffer Data-Forwarding Restriction Misaligned Store-Buffer Data-Forwarding Restriction High-Byte Store-Buffer Data-Forwarding Restriction One Supported Store-to-Load Forwarding Case Store-to-Load ForwardingFalse Dependencies Summary of Store-to-Load-Forwarding Pitfalls to Avoid 5.6 Prefetch Instructions Page Page Page Page Chapter 5 Cache and Memory Optimizations 109 25112 Rev. 3.06 September 2005 Prefetchnta [ esi + 64 * 2] C1 C5C2 C3 C4 ] C1 C5C2 C3 C4 Prefetchnta [ esi + 64 * ] Prefetchnta [ esi + 64 * Figure 5. Processor-Limited Code Page Page 5.7 Streaming-Store/Non-Temporal Instructions 5.8 Write-combining 5.9 L1 Data Cache Bank Conflicts Fields Used to Address the Multibank L1 Data Cache How to Know If a Bank Conflict Exists Page 5.10 Placing Code and Data in the Same 64-Byte Cache Line 5.11 Sorting and Padding C and C++ Structures Sorting and Padding C and C++ Structures Page 5.12 Sorting Local Variables 5.13 Memory Copy Chapter 5 Cache and Memory Optimizations 121 5.14 Stack Considerations 5.15 Cache Issues when Writing Instruction Bytes to 5.16 Interleave Loads and Stores Chapter 6 Branch Optimizations In This Chapter 6.1 Density of Branches Page 6.2 Two-Byte Near-Return RET Instruction Page 6.3 Branches That Depend on Random Data Chapter 6 Branch Optimizations 131 Conditional Write 6.4 Pairing CALL and RETURN 6.5 Recursive Functions Page 6.6 Nonzero Code-Segment Base Values 6.7 Replacing Branches with Computation Muxing Constructs Chapter 6 Branch Optimizations 137 SSE Solution (Preferred) MMX Solution (Avoid) Sample Code Translated into AMD64 Code Chapter 6 Branch Optimizations 139 Example 3: C Code Example 3: 3DNow! Code Example 4: C Code Example 4: 3DNow! Code 140 Branch Optimizations Chapter 6 Example 5: C Code Example 5: 3DNow! Code 6.8 The LOOP Instruction 6.9 Far Control-Transfer Instructions Chapter 7 Scheduling Optimizations 7.1 Instruction Scheduling by Latency 7.2 Loop Unrolling Loop Unrolling Complete Loop Unrolling Example: Complete Loop Unrolling Partial Loop Unrolling Example: Partial Loop Unrolling Page Deriving the Loop Control for Partially Unrolled Loops 7.3 Inline Functions Additional Recommendations 7.4 Address-Generation Interlocks Page 7.5 MOVZX and MOVSX 7.6 Pointer Arithmetic in Loops Page Page 7.7 Pushing Memory Data Directly onto the Stack Page Chapter 8 Integer Optimizations 8.1 Replacing Division with Multiplication Multiplication by Reciprocal (Division) Utility Unsigned Division by Multiplication of Constant Signed Division by Multiplication of Constant Chapter 8 Integer Optimizations 163 Signed Division by 2 Signed Division by 2 Signed Division by (2n) Remainder of Signed Division by 2 or 2 8.2 Alternative Code for Multiplying by a Constant Chapter 8 Integer Optimizations 165 166 Integer Optimizations Chapter 8 8.3 Repeated String Instructions Rational Page 8.4 Using XOR to Clear Integer Registers 8.5 Efficient 64-Bit Integer Arithmetic in 32-Bit Mode Chapter 8 Integer Optimizations 171 64-Bit Right Shift 64-Bit Multiplication 64-Bit Unsigned Division 172 Integer Optimizations Chapter 8 Chapter 8 Integer Optimizations 173 64-Bit Signed Division 174 Integer Optimizations Chapter 8 64-Bit Unsigned Remainder Computation Chapter 8 Integer Optimizations 175 176 Integer Optimizations Chapter 8 64-Bit Signed Remainder Computation Chapter 8 Integer Optimizations 177 178 Integer Optimizations Chapter 8 8.6 Efficient Implementation of Population-Count Function in 32-Bit Mode 180 Integer Optimizations Chapter 8 Integer Version 8.7 Efficient Binary-to-ASCII Decimal Conversion Binary-to-ASCII Decimal Conversion Retaining Leading Zeros 182 Integer Optimizations Chapter 8 Chapter 8 Integer Optimizations 183 Binary-to-ASCII Decimal Conversion Suppressing Leading Zeros 184 Integer Optimizations Chapter 8 Chapter 8 Integer Optimizations 185 8.8 Derivation of Algorithm, Multiplier, and Shift Factor for Integer Division by Constants Unsigned Integer Division Chapter 8 Integer Optimizations 187 188 Integer Optimizations Chapter 8 Chapter 8 Integer Optimizations 189 Signed Integer Division 190 Integer Optimizations Chapter 8 Chapter 8 Integer Optimizations 191 8.9 Optimizing Integer Division Chapter 9 Optimizing with SIMD Instructions Page 9.1 Ensure All Packed Floating-Point Data are Aligned RationaleSingle Precision RationalDouble Precision 9.3 Use MOVLPx/MOVHPx Instructions for Unaligned Data Access 9.4 Use MOVAPD and MOVAPS Instead of MOVUPD and MOVUPS 9.5 Structuring Code with Prefetch Instructions to Hide Memory Latency Page Page Chapter 9 Optimizing with SIMD Instructions 203 204 Optimizing with SIMD Instructions Chapter 9 Chapter 9 Optimizing with SIMD Instructions 205 9.6 Avoid Moving Data Directly Between General-Purpose and MMX Registers 9.7 Use MMX Instructions to Construct Fast Block- Copy Routines in 32-Bit Mode 9.8 Passing Data between MMX and 3DNow! 9.9 Storing Floating-Point Data in MMX Registers 9.10 EMMS and FEMMS Usage 9.11 Using SIMD Instructions for Fast Square Roots and Fast Reciprocal Square Roots 212 Optimizing with SIMD Instructions Chapter 9 Chapter 9 Optimizing with SIMD Instructions 213 The preceding code illustrates the use of separate loops for optimal performance. The loop titled keep the processor busy by masking the latencies of the reciprocal and square-root instructions. In Page 9.12 Use XOR Operations to Negate Operands of SSE, SSE2, and 3DNow! Instructions 9.13 Clearing MMX and XMM Registers with XOR 9.14 Finding the Floating-Point Absolute Value of Operands of SSE, SSE2, and 3DNow! 9.15 Accumulating Single-Precision Floating-Point Numbers Using SSE, SSE2, and 3DNow! Chapter 9 Optimizing with SIMD Instructions 219 220 Optimizing with SIMD Instructions Chapter 9 9.16 Complex-Number Arithmetic Using SSE, SSE2, and 3DNow! Instructions 222 Optimizing with SIMD Instructions Chapter 9 Listing 25. Complex Multiplication of Streams of Complex Numbers (SSE) Chapter 9 Optimizing with SIMD Instructions 223 224 Optimizing with SIMD Instructions Chapter 9 Listing 26. Complex Multiplication of Streams of Complex Numbers (3DNow! Technology) Chapter 9 Optimizing with SIMD Instructions 225 226 Optimizing with SIMD Instructions Chapter 9 Chapter 9 Optimizing with SIMD Instructions 227 The PFPNACC instruction is specifically designed for use in complex arithmetic operations. Page Page 9.17 Optimized 4 4 Matrix Multiplication on 4 1 Column Vector Routines Chapter 9 Optimizing with SIMD Instructions 231 Listing 27. 4 4 Matrix Multiplication (SSE) 232 Optimizing with SIMD Instructions Chapter 9 Chapter 9 Optimizing with SIMD Instructions 233 Listing 28. 4 4 Matrix Multiplication (3DNow! Technology) 234 Optimizing with SIMD Instructions Chapter 9 Chapter 9 Optimizing with SIMD Instructions 235 Page Chapter 10 x87 Floating-Point Optimizations 10.1 Using Multiplication Rather Than Division 10.2 Achieving Two Floating-Point Operations per Clock Cycle Page Page Page Page 10.3 Floating-Point Compare Instructions 10.4 Using the FXCH Instruction Rather Than FST/FLD Pairs 10.5 Floating-Point Subexpression Elimination 10.6 Accumulating Precision-Sensitive Quantities in x87 Registers 10.7 Avoiding Extended-Precision Data Appendix A Microarchitecture for AMD Athlon 64 and AMD Opteron Processors A.1 Key Microarchitecture Features Page A.5 L1 Instruction Cache A.6 Branch-Prediction Table A.7 Fetch-Decode Unit A.8 Instruction Control Unit A.9 Translation-Lookaside Buffer L1 Instruction TLB Specifications L1 Data TLB Specifications A.10 L1 Data Cache A.11 Integer Scheduler A.12 Integer Execution Unit A.13 Floating-Point Scheduler A.14 Floating-Point Execution Unit A.15 Load-Store Unit A.16 L2 Cache A.17 Write-combining A.18 Buses for AMD Athlon 64 and AMD Opteron Processor A.19 Integrated Memory Controller A.20 HyperTransport Technology Interface HyperTransport Technology Page Appendix B Implementation of Write-Combining B.1 Write-Combining Definitions and Abbreviations B.2 Programming Details B.3 Write-combining Operations Page B.4 Sending Write-Buffer Data to the System B.5 Write-Combining Optimization on Revision D and E AMD Athlon 64 and AMD Opteron Processors Optimizations Page Appendix C Instruction Latencies C.1 Understanding Instruction Entries Example: Instruction Entry Parts of the Instruction Entry Appendix C Instruction Latencies 271 Interpreting Placeholders Interpreting Latencies Appendix C Instruction Latencies 273 C.2 Integer Instructions Table 13. Integer Instructions 274 Instruction Latencies Appendix C Appendix C Instruction Latencies 275 276 Instruction Latencies Appendix C Appendix C Instruction Latencies 277 278 Instruction Latencies Appendix C Appendix C Instruction Latencies 279 280 Instruction Latencies Appendix C Appendix C Instruction Latencies 281 282 Instruction Latencies Appendix C Appendix C Instruction Latencies 283 284 Instruction Latencies Appendix C Appendix C Instruction Latencies 285 286 Instruction Latencies Appendix C Appendix C Instruction Latencies 287 288 Instruction Latencies Appendix C Appendix C Instruction Latencies 289 290 Instruction Latencies Appendix C Appendix C Instruction Latencies 291 292 Instruction Latencies Appendix C Appendix C Instruction Latencies 293 294 Instruction Latencies Appendix C Appendix C Instruction Latencies 295 296 Instruction Latencies Appendix C Appendix C Instruction Latencies 297 298 Instruction Latencies Appendix C Appendix C Instruction Latencies 299 300 Instruction Latencies Appendix C Appendix C Instruction Latencies 301 302 Instruction Latencies Appendix C Appendix C Instruction Latencies 303 C.3 MMX Technology Instructions Table 14. MMX Technology Instructions 304 Instruction Latencies Appendix C Appendix C Instruction Latencies 305 306 Instruction Latencies Appendix C Appendix C Instruction Latencies 307 C.4 x87 Floating-Point Instructions Table 15. x87 Floating-Point Instructions 308 Instruction Latencies Appendix C Appendix C Instruction Latencies 309 310 Instruction Latencies Appendix C Appendix C Instruction Latencies 311 312 Instruction Latencies Appendix C Appendix C Instruction Latencies 313 314 Instruction Latencies Appendix C C.5 3DNow! Technology Instructions Table 16. 3DNow! Technology Instructions Appendix C Instruction Latencies 315 Table 16. 3DNow! Technology Instructions (Continued) 316 Instruction Latencies Appendix C C.6 3DNow! Technology Extensions Table 17. 3DNow! Technology Extensions Appendix C Instruction Latencies 317 C.7 SSE Instructions Table 18. SSE Instructions 318 Instruction Latencies Appendix C Appendix C Instruction Latencies 319 320 Instruction Latencies Appendix C Appendix C Instruction Latencies 321 322 Instruction Latencies Appendix C Appendix C Instruction Latencies 323 324 Instruction Latencies Appendix C Appendix C Instruction Latencies 325 326 Instruction Latencies Appendix C C.8 SSE2 Instructions Table 19. SSE2 Instructions Appendix C Instruction Latencies 327 328 Instruction Latencies Appendix C Appendix C Instruction Latencies 329 330 Instruction Latencies Appendix C Appendix C Instruction Latencies 331 332 Instruction Latencies Appendix C Appendix C Instruction Latencies 333 334 Instruction Latencies Appendix C Appendix C Instruction Latencies 335 336 Instruction Latencies Appendix C Appendix C Instruction Latencies 337 338 Instruction Latencies Appendix C Appendix C Instruction Latencies 339 340 Instruction Latencies Appendix C Appendix C Instruction Latencies 341 342 Instruction Latencies Appendix C C.9 SSE3 Instructions Table 20. SSE3 Instructions Appendix C Instruction Latencies 343 Table 20. SSE3 Instructions (Continued) Page Appendix D AGP Considerations D.1 Fast-Write Optimizations D.2 Fast-Write Optimizations for Graphics-Engine Programming Page Page D.3 Fast-Write Optimizations for Video-Memory Copies 350 AGP Considerations Appendix D D.4 Memory Optimizations 352 AGP Considerations Appendix D 25112 Rev. 3.06 September 2005 Figure 12. Northbridge Command Flow D.5 Memory Optimizations for Graphics-Engine Programming Using the DMA Model D.6 Optimizations for Texture-Map Copies to AGP D.7 Optimizations for Vertex-Geometry Copies to AGP Page Appendix E SSE and SSE2 Optimizations Types of XMM-Register Data Types of SSE and SSE2 Instructions E.1 Half-Register Operations E.2 Zeroing Out an XMM Register Page E.3 Reuse of Dead Registers E.4 Moving Data Between XMM Registers and GPRs E.5 Saving and Restoring Registers of Unknown Format E.6 SSE and SSE2 Copy Loops E.7 Explicit Load Instructions E.8 Data Conversion Appendix E SSE and SSE2 Optimizations 365 Table 23. Converting Vector Values Table 24. Converting Directly from Memory Table 22. Converting Scalar Values (Continued) Page Index 367 Index Numerics A B C M N O P R

Related pages Troubleshooting section at end of manual for Harbor Freight Tools 2433 Engineering Data & Specifications for SPX Cooling Technologies 1000 Where playback starts for Sanyo VA-SW8000LITE Networking FAQ for Lexmark 20E INSTALLATION for Panasonic WV-CP484 Panel Parts List No for Whirlpool DU3003XL0 Language code list for Technics SC-DV290 …I get the error message “Recycle Bin Maximum Size Reached.” for HP Palmtop 660LX ADVANCED MENU - SETTING DEVICE FEATURE CONTROLS for Fujitsu C2330 KITCHENAID OUTDOOR PRODUCT WARRANTY for KitchenAid KBAU272V SAFETY INFORMATION for Desa VS18PRA