A method for transforming C code with floating-point values into C code with integer variables is developed. The objective is to avoid any operations with floating-point data types, thereby increasing the execution speed of the program on a microprocessor without a math coprocessor. The original C code must be a dot product with floating-point literals and integer variables with known interval bounds. The transformation algorithm remodels the dot product form into a tree structure, to maximize the accuracy, but, on the other side, keeps the number of shift operations reduced. The integer code that is generated is ANSI C compliant. It is tested on 8-bit and 32-bit microprocessors using different compilers. The results show that the integer code is several times faster than the floating-point code, the only loss being a very low accuracy drop.
