AMD K5 User Manual Page 19

  • Download
  • Add to my manuals
  • Print
  • Page
    / 100
  • Table of contents
  • BOOKMARKS
  • Rated. / 5. Based on customer reviews
Page view 18
9
18522F/0Jan1997 AMD-K5 Processor Data Sheet
PRELIMINARY INFORMATION
a given memory location returns valid data. Each cache line is
assigned one of the four protocol states to identify the status of
the information stored in the cache. The writeback cache
design updates memory only when necessary. This keeps the
system bus free for use by other devices and improves the
overall system performance.
4.7 Branch Prediction
A branch occurs on average once every seven x86 instructions.
When a branch is encountered, the processor predicts which
direction the instruction flow will follow. The AMD-K5 proces-
sor adds branch prediction information to each instruction
cache line in the form of a predicted address tag that indicates
the target address of the first branch that is predicted to be
taken in the cache line. The processor’s dynamic branch pre-
diction mechanism allows for 1024 branch targets and a 75%
branch prediction accuracy. Combined with a minimal 3-cycle
mispredict penalty, the branch prediction mechanism opti-
mizes the processor’s speculative execution of x86 software,
such as the Microsoft Windows operating system and associ-
ated applications.
The dynamic branch prediction of the processor enables
instructions to be fetched and fed into the processor’s execu-
tion core, eliminating many pipeline bubbles and contributing
to the superior performance of the AMD-K5 processor.
4.8 Unique x86 Instruction Conversion and Decoding
The logical instruction flow within the AMD-K5 processor con-
tinues as up to 32-bytes of predecoded x86 instructions are
fetched from the byte queue of the instruction cache and for-
warded in order to the decoder.
The processor's decoder converts complex x86 instructions into
relatively simple, fast-executing ROPs that are of fixed length
and easy to process. Simultaneously, the operands needed to
perform the ROPs’ operations are fetched from the register
file or from the reorder buffer.
Page view 18
1 2 ... 14 15 16 17 18 19 20 21 22 23 24 ... 99 100

Comments to this Manuals

No comments