[llvm] [TableGen] Add `CompressedTupleNameStride` to compress tuple names. (PR #169196)

[Jason Eckhardt via llvm-commits]

nvjle wrote:

FWIW:  I was curious about the impact on the tuple-heavy AMDGPU target. I applied the feature with this quick experiment:
```
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.td
@@ -71,7 +71,10 @@ class SIRegisterTuples<list<SubRegIndex> Indices, RegisterClass RC,
                        int last_reg, int stride, int size, string prefix> :
   RegisterTuples<Indices,
                  RegSeqDags<RC, last_reg, stride, size>.ret,
-                 RegSeqNames<last_reg, stride, size, prefix>.ret>;
+                 RegSeqNames<last_reg, stride, size, prefix>.ret> {
+  let CompressedTupleNameStride =
+      !if(!and(!gt(size, 4), !ne(prefix, "ttmp")), 1, 0);
+}
``` 
That is, compress the names of all tuple classes greater than 4-wide  (note: I excluded `TTMP` because there are actually three of those classes-- one defined by `RegisterTuples` and two others that are hand-rolled).

The sizes in bytes are as follows:
```
                                        before       after
AMDGPUGenRegisterInfoEnums.inc            1.6M      758K  
AMDGPUGenRegisterInfoMCDesc.inc           26M       17M 
```

https://github.com/llvm/llvm-project/pull/169196