r37894 causes a 10% regression on richards. Caused by: - #define CTI_SAMPLER + #define CTI_SAMPLER 0 Measured on my Mac Pro.
I am an idiot.
lol
Gavin mentioned in person that he saw this effect disappear in later revisions.
I tried testing against TOT, but changing to "#define CTI_SAMPLER" causes the infamous GCC 4.2 hang.
It might be nice to do a little more investigation when GCC 4.2 is in a more cooperative mood. For now, though, I think we can close this bug based on Gavin's existing data.