[llvm] [WIP][X86] Replace custom minmax reduction pattern matching with ISD::VEC_REDUCE_*MIN/MAX support (PR #194473)

[via llvm-commits]

================
@@ -38,14 +31,9 @@ define { i16, i16 } @test_reduce_v16i16_with_umin(<16 x i16> %x, <16 x i16> %y)
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; AVX2-NEXT:    vpminuw %xmm2, %xmm0, %xmm2
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[2,3,2,3]
-; AVX2-NEXT:    vpminuw %xmm3, %xmm2, %xmm3
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,1,1]
-; AVX2-NEXT:    vpminuw %xmm4, %xmm3, %xmm3
-; AVX2-NEXT:    vpsrld $16, %xmm3, %xmm4
 ; AVX2-NEXT:    vphminposuw %xmm2, %xmm2
 ; AVX2-NEXT:    vmovd %xmm2, %eax
-; AVX2-NEXT:    vpminuw %xmm4, %xmm3, %xmm2
+; AVX2-NEXT:    vmovd %eax, %xmm2
----------------
ts-numfum wrote:

unnecessary instruction

https://github.com/llvm/llvm-project/pull/194473