[llvm] [WIP][SPARC] Mark branches as being expensive in early Niagara CPUs (PR #166489)

[via llvm-commits]

koachan wrote:

So I tried some pgbench on a 48-thread SPARC T2, and the patch does seem to increase performance (with only one exception on 48-thread SELECT workload):

| threads | SELECT   |          | INS/UPD |         |
| ------- | -------- | -------- | ------- | ------- |
|         | main     | +patch   | main    | +patch  |
| 1       | 2318.04  | 2413.27  | 351.07  | 372.09  |
| 8       | 17273.82 | 17568.66 | 2513.91 | 2619.37 |
| 16      | 30384.08 | 31282.25 | 4322.87 | 4611.78 |
| 24      | 41794.27 | 43145.46 | 5719.44 | 5987.82 |
| 32      | 48902.99 | 50403.88 | 6498.09 | 6579.41 |
| 40      | 58342.80 | 60795.85 | 6935.46 | 7009.69 |
| 48      | 61119.86 | 59821.99 | 7363.66 | 7424.10 |

The speedup is quite modest (at around 2-3%), but given that it's only from setting two codegen tunables I'd say that this is a good result.

https://github.com/llvm/llvm-project/pull/166489