[llvm] [X86] Narrow vXi32/vXi64 usubsat to vpsubusb/vpsubusw when LHS is known to fit in fewer bits (PR #206592)

[via llvm-commits]

sipher-01 wrote:

I've updated all the affected test files to reflect the new codegen produced by the generic DAGCombiner narrowing:

- `usub_sat_vec.ll`
- `combine-icmp.ll`
- `combine-sub-usat.ll`
- `var-permute-128.ll`
- `vec_setcc-2.ll`
- `vector-extract-last-active.ll`
- `intrinsic-cttz-elts.ll`

Both `check-llvm-codegen-x86` and `check-llvm-codegen-aarch64` pass cleanly locally:

**X86:**

~/De/p/co/o/llvm-project fix/avx512-sub-sat !5 ❯ ninja -C build check-llvm-codegen-x86 2>&1 | tail -10                               base 02:27:25 AM
[0/1] Running lit suite /Users/sipher/Desktop/pc/coding/openSource/llvm-project/llvm/test/CodeGen/X86
-- Testing: 5523 tests, 8 workers --
Testing:  0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90..

Testing Time: 171.08s

Total Discovered Tests: 5523
  Unsupported      :    7 (0.13%)
  Passed           : 5501 (99.60%)
  Expectedly Failed:   15 (0.27%)
  
  **AArch64:**
  
  ~/De/p/co/o/llvm-project fix/avx512-sub-sat !5 ❯ ninja -C build check-llvm-codegen-aarch64 2>&1 | tail -10                   2m 54s  base 02:30:26 AM
ninja: Entering directory `build'
[0/1] Running lit suite /Users/sipher/Desktop/pc/coding/openSource/llvm-project/llvm/test/CodeGen/AArch64
-- Testing: 4089 tests, 8 workers --
Testing:  0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90.. 

Testing Time: 107.59s

Total Discovered Tests: 4089
  Passed           : 4084 (99.88%)
  Expectedly Failed:    5 (0.12%)

https://github.com/llvm/llvm-project/pull/206592