Unbalanced x87 FPU stack push/pop operations

[andreas-abel]

There are several basic blocks with an unbalanced number of push/pop operations to the x87 FPU stack. An example is d9ee4885ff (fldz; test rdi,rdi). These basic blocks have a low throughput due to the overhead of handling FPU stack overflows/underflows.

I don't think these benchmarks are very meaningful, as reasonable programs would not execute such basic blocks repeatedly without additional instructions that keep the stack balanced.