On 10/8/25 6:32 PM, Vineet Gupta wrote:
On 10/8/25 11:54, Vineet Gupta wrote:
We are seeing about 100 billion dynamic instructions saved which is about 5%
on cactuBSSN and a 2% improvement in performance on the BPI.
For Cactu, QEMU dynamic icounts went down from 2,010,916,571,027 to
1,894,197,931,855
Right in line with expectations. Note that in an OOO core performance
may not improve -- the code in question is dominated by fp loads and fp
ops. So there's a ton of excess scalar integer execution resources
available to hide the inefficiencies in the old code.
This is just awesome. Thx Shreya for doing this.
Agreed. Interestingly enough Philip R. is looking at this exact same
issue right now in the LLVM side.
jeff
