Table of Contents
Advertisement
22007E/0—November 1999
Use Load-Execute Floating-Point Instructions with Floating-Point
Operands
TOP
Avoid Load-Execute Floating-Point Instructions with Integer Operands
TOP
Load-Execute Instruction Usage
When operating on single-precision or double-precision
floating-point data, wherever possible use floating-point
load-execute instructions to increase code density.
Note: This optimization applies only to floating-point instructions
with floating-point operands and not with integer operands,
as described in the next optimization.
This coding style helps in two ways. First, denser code allows
more work to be held in the instruction cache. Second, the
denser code generates fewer internal OPs and, therefore, the
FPU scheduler holds more work, which increases the chances of
extracting parallelism from the code.
Example 1 (Avoid):
FLD
QWORD PTR [TEST1]
FLD
QWORD PTR [TEST2]
FMUL
ST, ST(1)
Example 2 (Preferred):
FLD
QWORD PTR [TEST1]
FMUL
QWORD PTR [TEST2]
Do not use load-execute floating-point instructions with integer
operands: FIADD, FISUB, FISUBR, FIMUL, FIDIV, FIDIVR,
F I C O M , a n d F I C O M P. R e m e m b e r t h a t f l o a t i n g -p o i n t
ins truc tions can have int ege r ope rands while int ege r
instruction cannot have floating-point operands.
Floating-point computations involving integer-memory
operands should use separate FILD and arithmetic instructions.
This optimization has the potential to increase decode
bandwidth and OP density in the FPU scheduler. The floating-
point load-execute instructions with integer operands are
VectorPath and generate two OPs in a cycle, while the discrete
equivalent enables a third DirectPath instruction to be decoded
in the same cycle. In some situations this optimizations can also
reduce execution time if the FILD can be scheduled several
instructions ahead of the arithmetic instruction in order to
cover the FILD latency.
AMD Athlon™ Processor x86 Code Optimization
35
Advertisement
Table of Contents
