Scalable Numerical Algorithms and Methods on the ASCI Machines

1
CS61V
64-Bit Computing
2
What is 64-bit Computing ?

• A 64-bit computing architecture is one that
  contains general purpose registers (GPR) that can
  hold 64-bit data.
• Similarly, a 64-bit instruction is one that can
  operate on 64-bit operands.

3
64-bit Computing
Code Stream
Data Stream
64-bit data
Mixture of 32 and 64-bit data
64-bit registers
4
Programming Models
Doubled register size
Doubled PC size
5
What is 64-bit Computing ?

• In 64-bit processors
• The data stream doubles in size and hence the
  GPRs are doubled in capacity
• The instruction stream remains the same size and
  therefore the Instruction Register (IR) remains
  fixed.

6
The Data Stream

• In modern processors the data stream is composed
  of
• Integer and address data which are operated on in
  the ALU with associated registers.
• Floating Point data which are operated on in the
  FPU with associated registers.
• Vector (SIMD) data which are operated on in the
  Vector Unit with associated registers.

7
64-bit Architecture

• The only difference in moving to 64-bits is in
• the integer and address hardware.

Intel ISA
8
64-bit Integer Data

• Wider data registers increases the dynamic range
  of integer representation
• A 32-bit processor can manipulate
• 232 4.3e9 integers
• A 64-bit processor can manipulate
• 264 1.8e19 integers
• By doubling the bits we have increased the
  dynamic range by a factor of 4.3 Billion.

9
64-bit Address Data

• Wider data registers also increases the dynamic
  range of addressable memory
• A 32-bit processor can manipulate
• 232 4.3e9 addresses 4GB
• A 64-bit processor can manipulate
• 264 1.8e19 addresses 180 PetaBytes.
• Quote 640K ought to be enough for anybody.
• Bill Gates 1981

10
Increased Memory Addressing
11
What can you do with over 4GB of Memory?

• Applications that use very large objects,
• i.e., gt 2 gigabyte objects
• Video editing
• Simulation
• Visualization
• Applications that use files that are larger than
  232 bytes.
• Can memory-map very LARGE files!
• e.g. Back-end Servers and Databases.

12
What can you do with over 4GB of Memory?

• Large scientific applications
• They work with numbers outside the dynamic range
  of 32-bit integers.
• This causes overflows or underflows and hence
  wrong answers in the output registers
• ( you can check the PSW to determine if such a
  situation occurs )
• Cryptography
• Modern cryptographic algorithms rely on the
  factoring and multiplication of very large
  numbers
• The larger the number the more secure the
  encryption.

13
64-bit Drawbacks?

• Memory address values ( called pointers ) are now
  twice as large and take up twice the space.
• Pointers normally take up a fraction of the space
  in cache
• Now they are doubled in size and can squeeze out
  other useful data from the cache and reduce
  performance.
• ( slight improvement is to tag 64-bit integers
  during programming e.g. REX mnemonic prefix,
  increase size lt 10 for current applications)

14
Conclusions

• 64-bit code can be compiled on 32-bit
  architectures.
• 64-bit data can be processed as two 32-bit
  calculations although a performance penalty is
  incurred.
• Only applications designed to process 64-bit data
  and are implemented on 64-bit hardware will
  achieve any speedup.

15
Conclusions

• On a daily basis we're running into the Windows
  2GB barrier with our next-generation content
  development and preprocessing tools.
• If cost-effective, backwards-compatible 64-bit
  CPU's were available today, we'd buy them today.
  We need them today. It looks like we'll get them
  in April.
• Regarding this "far off" application
  compatibility, we've been running the 64-bit SuSE
  Linux distribution on Hammer for over 3 months.
  We're going to ship the 64-bit version of UT2003
  at or before the consumer Athlon64 launch. And
  our next-generation engine won't just support
  64-bit, but will basically REQUIRE it on the
  content-authoring side.
• -Tim Sweeney, Unreal Engine Guru, Epic Games