[llvm-branch-commits] [libc] [llvm] [libc][math] Implement double precision tanh correctly rounded to all rounding modes. (PR #235618)

[via llvm-branch-commits]

zimmermann6 wrote:

great work! I get the following figures (reciprocal throughput then latency):
```
GNU libc version: 2.43
GNU libc release: stable
[####################] 100 %
Ntrial = 20 ; Min = 37.187 + 1.662 clc/call; Median-Min = 1.477 clc/call; Max = 40.182 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 46.730 + 0.230 clc/call; Median-Min = 0.116 clc/call; Max = 47.596 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 29.025 + 0.795 clc/call; Median-Min = 0.214 clc/call; Max = 31.137 clc/call;
GNU libc version: 2.43
GNU libc release: stable
[####################] 100 %
Ntrial = 20 ; Min = 85.692 + 0.130 clc/call; Median-Min = 0.040 clc/call; Max = 87.070 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 102.373 + 0.098 clc/call; Median-Min = 0.072 clc/call; Max = 103.126 clc/call;
[####################] 100 %
Ntrial = 20 ; Min = 67.133 + 0.205 clc/call; Median-Min = 0.147 clc/call; Max = 68.380 clc/call;
```


https://github.com/llvm/llvm-project/pull/235618