search for: float_size

On Wed, Oct 27, 2010 at 1:29 PM, Brian West <bnwest at rice.edu> wrote: > Here is the patch for the new Operator Strength Reduction optimization > pass that I have written.  The bulk of the code is in > > lib/Transforms/Scalar/OperatorStrengthReduce.cpp > > The algorithm finds reduction opportunities in both array accesses and > explicit multiplications within loops.

Here is the patch for the new Operator Strength Reduction optimization pass that I have written. The bulk of the code is in lib/Transforms/Scalar/OperatorStrengthReduce.cpp The optimization is based on the algorithm described within the paper that can be found here: http://portal.acm.org/citation.cfm?id=504709.504710 Keith D. Cooper , L. Taylor Simpson , Christopher A. Vick, Operator strength

...op-reduce (LSR) only looks at implicit multiplication in array accesses. The following code has an explicit multiplication within a loop that LSR will not reduce: extern float loop4(float* a, int start, int stop) { int i; float sum; float *pEntry; char *p; unsigned long offset; const unsigned int float_size = sizeof(float); sum = 0.0; p = (char*)a; for (i=start-1; i<stop; ++i) { offset = i * float_size; pEntry = (float*) (p + offset); sum += *pEntry; } return sum; } OSR also finds reductions in some esoteric array accesses that LSR misses. I found these LSR misses while writing less-than-real-w...