AMD 250 manual Optimizing Integer Division, 192

Software Optimization Guide for AMD64 Processors 25112 Rev. 3.06 September 2005

8.9Optimizing Integer Division

Optimization

When possible, use smaller data types for integer division.

Application

This optimization applies to:

•32-bit software

•64-bit software

Rationale

Division by a 16-bit value is significantly faster than division by a 32-bit value—about a 26 clock latency versus 42. Likewise, division by a 32-bit value is faster than division by a 64-bit value—about 42 clocks versus 74. Refer to IDIV in table 15. In algorithms in which integer division contributes a substantial component to performance, it may be beneficial to check whether using a smaller divide type is possible. Study the assembly language output generated by high-level language compilers to verify that the desired code is generated. Compilers often generate code that converts 16-bit types into 32-bit values that are then used to perform 32-bit division, thus eliminating the advantage of using 16- bit integer types. If the compiler cannot be coerced into producing the desired code, then compiler intrinsics or assembly language are required.

192 Integer Optimizations Chapter 8