[TableGen][NFCI] Speed up generating *GenRegisterInfo.inc files on builds with expensive checks.

[kosarev]

This is mostly AMDGPU-specific. When the expensive checks are enabled, generating of AMDGPUGenRegisterInfo.inc currently takes about 20 minutes on my machine for release+asserts builds, which effectively prevents such testing from regular use. This patch fixes this by reducing the time to about 2 minutes.

Generation times for AMDGPUGenRegisterInfo.inc without expensive checks and other *GenRegisterInfo.inc files with and without the expensive checks remain approximately the same.

The patch doesn't cause any changes in the contents of the generated files.

The root cause of the current poor performance is that where glibcxx is used, enabling the expensive checks defines _GLIBCXX_DEBUG, which enables various consistency checks in the library. One such check is in std::binary_search() to make sure the range is ordered. As CodeGenRegisterClass::contains() relies on std::binary_search() and it is called very a large number of times from within CodeGenRegBank::inferMatchingSuperRegClass(), the libcxx checks heavily affect the runtimes.

[kosarev]

This helped working on #67245.

Note that the Debian expensive-checks bot, https://lab.llvm.org/buildbot/#/builders/16, explicitly undefines _GLIBCXX_DEBUG, which is a bit unfortunate because when the checks are not disabled, they do catch some problems on the current ToT:

Failed Tests (16):
  Clang :: Misc/warning-flags-tree.c
  Clang :: Misc/warning-wall.c
  LLVM :: CodeGen/PowerPC/fp-strict.ll
  LLVM :: CodeGen/PowerPC/pr55463.ll
  LLVM :: CodeGen/PowerPC/register-pressure.ll
  LLVM :: CodeGen/PowerPC/spe.ll
  LLVM :: CodeGen/X86/code_placement_ext_tsp.ll
  LLVM :: CodeGen/X86/code_placement_ext_tsp_large.ll
  LLVM :: Transforms/Inline/nested-inline.ll
  LLVM :: Transforms/SampleProfile/csspgo-import-list.ll
  LLVM-Unit :: ADT/./ADTTests/14/47
  LLVM-Unit :: ADT/./ADTTests/22/47
  LLVM-Unit :: ADT/./ADTTests/30/47
  LLVM-Unit :: ADT/./ADTTests/38/47
  LLVM-Unit :: ADT/./ADTTests/46/47
  LLVM-Unit :: ADT/./ADTTests/7/47

#67337 should fix the ADT failures.

As a side note, I don't know why AMDGPU is that much different in regards to the number of (generated) register classes; it might be worth looking into that at some point.

Tagging Galina @gkistanova for awarness.

[dwblaikie]

Sounds good to me (if this restores sufficient performance, perhaps you can re-enable the glibcxx debug checks on the buildbot?)