[llvm] [FuncSpec] Enable SpecializeLiteralConstant by default (PR #113442)

[Hari Limaye via llvm-commits]

hazzlim wrote:

Change in compile-times for CTMark (measured as instruction-count) from applying this patch, measured on an AArch64 (Graviton 3) machine:

O3 (Non-LTO):

```
Benchmark        |  Change  |
-----------------+----------+
ClamAV           |   0.00%  |
7zip             |  +0.01%  |
kimwitu++        |  -0.01%  |
tramp3d-v4       |   0.00%  |
sqlite3          |  +0.02%  |
mafft            |   0.00%  |
SPASS            |   0.00%  |
lencod           |   0.00%  |
consumer-typeset |  +0.03%  |
Bullet           |   0.00%  |
-----------------+----------+
GEOMEAN          |   0.00%  |
```

ReleaseLTO-g:

```
Benchmark        |  Change  |
-----------------+----------+
ClamAV           |  +0.01%  |
7zip             |  -0.01%  |
kimwitu++        |   0.00%  |
tramp3d-v4       |  +0.01%  |
sqlite3          |  +0.01%  |
mafft            |   0.00%  |
SPASS            |  -0.01%  |
lencod           |  +0.02%  |
consumer-typeset |  +0.05%  |
Bullet           |  -0.01%  |
-----------------+----------+
GEOMEAN          |  +0.01%  |
```


https://github.com/llvm/llvm-project/pull/113442