[llvm] [AMDGPU] Use correct number of bits needed for div/rem shrinking (PR #80622)

[Matt Arsenault via llvm-commits]

================
@@ -258,7 +258,7 @@ define amdgpu_kernel void @no_srem25_i24_i25_i32(ptr addrspace(1) %out, ptr addr
 }
 
 ; FUNC-LABEL: {{^}}no_srem25_i25_i24_i32:
-; SI-NOT: v_cvt_f32_i32
----------------
arsenm wrote:

changing -NOT checks makes me nervous. make this looser to check both types of conversions?

https://github.com/llvm/llvm-project/pull/80622