[AMDGPU] Fix more MSVC signed/unsigned and shift warnings (#241461)
Fix three MSVC warnings in the AMDGPU backend:
- AMDGPUISelLowering.cpp: C4018 (signed/unsigned mismatch) comparing the
unsigned DivBits against the int literals 23 and 22.
- AMDGPUDisassembler.cpp: C4334 (32-bit shift implicitly converted to 64
bits) in an assert comparing a uint64_t against 1U << EncSize. Use a
64-bit shift.
- AMDGPUInstCombineIntrinsic.cpp: C4334 EnBits is a uint64_t but the
masks were built with 32-bit shifts. Use 64-bit shifts.