[llvm] [AMDGPU][GlobalISel] Improve combining on v_ldexp (PR #190236)

[Petar Avramovic via llvm-commits]

================
@@ -447,3 +402,52 @@ define amdgpu_cs half @fdiv_f16_i16(half inreg %a, i16 inreg %b) {
   %result = fdiv afn half %a, %uint
   ret half %result
 }
+
+; Test that the fmul+select -> fldexp combine works for divergent values.
----------------
petar-avramovic wrote:

Right, should add check lines for gfx1200 to see if we generate s_mul or ldexp

https://github.com/llvm/llvm-project/pull/190236