[llvm-branch-commits] [libc] [llvm] [libc][math] Implement double precision cosh correctly rounded to all rounding modes. (PR #235181)

[via llvm-branch-commits]

zimmermann6 wrote:

all CORE-MATH tests pass on my side, and the efficiency is quite good. Reciprocal throughput:
```
GNU libc version: 2.43
GNU libc release: stable
[####################] 100 %
Ntrial = 20 ; Min = 33.270 + 0.209 clc/call; Median-Min = 0.210 clc/call; Max = 33.637 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 18.203 + 0.313 clc/call; Median-Min = 0.323 clc/call; Max = 19.476 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 23.759 + 0.388 clc/call; Median-Min = 0.273 clc/call; Max = 24.782 clc/call;
```
Latency:
```
GNU libc version: 2.43
GNU libc release: stable
[####################] 100 %
Ntrial = 20 ; Min = 47.520 + 0.465 clc/call; Median-Min = 0.301 clc/call; Max = 48.965 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 42.192 + 0.208 clc/call; Median-Min = 0.184 clc/call; Max = 42.638 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 46.438 + 0.569 clc/call; Median-Min = 0.272 clc/call; Max = 48.486 clc/call;
```

https://github.com/llvm/llvm-project/pull/235181