Chapter 10 x87 Floating-Point Optimizations 247
Software Optimization Guide for AMD64 Processors
25112 Rev. 3.06 September 2005
10.6 Accumulating Precision-Sensitive Quantities in
x87 Registers
Optimization
Accumulate results in the x87 registers rather than the SSE and SSE2 XMM registers, if more than
64 bits of accuracy are required.
Application
This optimization applies to:
• 32-bit software
• 64-bit software
Rationale
More than 64 bits of accuracy may be required, as when accumulating a result (for example, during
the calculation of dot product). The precision of floating-point operations in the x87 registers ST(0)–
ST(7) is 80 bits internally, whereas the precision of operations using SIMD instructions is only
64 bits.