Idiom Recognizer

Idiom recognition is an optimization technique used in compilers to identify and replace common programming patterns with more efficient or specialized implementations. This transformation recognizes known mathematical or logical expressions and substitutes them with optimized equivalents, often leveraging platform-specific or hardware-accelerated functions. By standardizing these patterns, idiom recognition improves performance, enhances portability, and enables further optimizations such as vectorization or automatic replacement of low-level operations. This technique is particularly useful in optimizing code for specific hardware architectures while maintaining the original program's behavior.

Idiom Recognizer Transformation in emmtrix Studio

emmtrix Studio can implement idiom recognizer using #pragma directives or via the GUI. Idiom Recognizer is a transformation that identifies known math.h functions by the implementation and replaces them with internal EMX functions. The EMX functions are defined in the emx_operators.h header file.

Typical Usage and Benefits

The transformation is used to introduce flexibility with regard to function implementation. This flexibility increases potential for target platform optimizations. The optimizations can be performed when for instance the hardware manufacturer has specially efficient implementations of respective functions.

Common application areas are vectorization and automatic replacement of assembly intrinsics.

Example

/* The following code tests idiom recognizer transformation applied to main function.
 */   

#pragma EMX_TRANSFORMATION IdiomRecognizer
int main(void) {
    int i = 3;
    i = 2 < i ? 2 : i;
    return 0;
}

/* In the given example (on the left side), i = 2 < i ? 2 : i; is recognized as the math min function and it is replaced
   with an internal EMX implementation.
 */

#include ”emx_operators.h”

int main(void) {
    int i = 3;
    i = __emx_min(2, i);
    return 0;
}

Note

This transformation currently supports limited number of functions and it is hence work in progress. More functions will be supported in the future.

Idiom Recognizer

Contents