PWR028: Remove pointer increment preventing performance optimization
Issue
A loop contains a pointer increment statement enabling access to the elements of an array, but the pointer increment introduces dependencies between the iterations of the loop that defeat performance optimization.
Actions
Remove the pointer increment instruction from the loop body, updating the dereferences of the pointer in the loop body and recalculating the offsets correspondingly.
Relevance
Performance optimization best practices give recommendations regarding the efficient memory access pattern, but do not give recommendations regarding a coding style that favors performance optimizations for the compiler.
Programmers may write programs where the array references are driven by the pointer variable being incremented in the loop body. From a memory access pattern perspective, both ways of coding lead to accessing the same data in the same order at run-time. However, there is a fundamental difference from the point of view of the instructions executed by the processor: using another pointer variable incremented in the loop body introduces dependencies between the iterations of the loop.
Programmers may write different programs, and tools are expected to detect these challenging situations, help programmers to understand the implications from the performance viewpoint, and suggest how to overcome these issues through alternative ways of coding that are more hardware-friendly.
Code example
The following loop uses pointer increment notation to access the elements of
array bTemp1.
void example(float *a, float *b, float *c, unsigned size, unsigned inc) {
float *bTemp1 = b;
for (unsigned i = 0; i < size; i++) {
c[0] += (a[i] * bTemp1[0]);
bTemp1 -= inc;
}
}
We can get rid of pointer based notation and replace it with array based notation like this:
void example(float *a, float *b, float *c, unsigned size, unsigned inc) {
for (unsigned i = 0; i < size; i++) {
c[0] += (a[i] * bTemp1[-i * inc]);
}
}