[llvm] [SDAG] Fix fmaximum legalization errors (PR #142170)

[Matt Arsenault via llvm-commits]

================
@@ -2649,3 +2649,180 @@ define <4 x bfloat> @test_fmaximum_v4bf16(<4 x bfloat> %x, <4 x bfloat> %y) {
   %r = call <4 x bfloat> @llvm.maximum.v4bf16(<4 x bfloat> %x, <4 x bfloat> %y)
   ret <4 x bfloat> %r
 }
+
----------------
arsenm wrote:

Does this also fix the legalization on x86_32? The current run lines add AVX, need one without sse or avx 

https://github.com/llvm/llvm-project/pull/142170