[llvm] [AMDGPU] Skip fold_pow constant-exponent shortcuts for powr on possibly-negative base (PR #200579)

[Matt Arsenault via llvm-commits]

================
@@ -6109,6 +6110,47 @@ define <2 x float> @test_pow_v2f32_known_integral_constant_vector_poison_elt(<2
   ret <2 x float> %pow
 }
 
+define float @test_powr_f32_unknown_base_two(float %x) {
----------------
arsenm wrote:

Don't think these new tests are mirrored in the powr-fast test?

https://github.com/llvm/llvm-project/pull/200579