[X86][test] Remove useless pattern for VDPBF16PSZmb and add a test for broadcast folding (llvm#80629)

llvm-issue: llvm#68810

Author: KanRobert

llvm/lib/Target/X86/X86InstrAVX512.td

@@ -12716,13 +12716,13 @@ multiclass avx512_dpbf16ps_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
                                (src_v.LdFrag addr:$src3)))>, EVEX, VVVV,
                                Sched<[sched.Folded, sched.ReadAfterFold]>;
 
+  let mayLoad = 1, hasSideEffects = 0 in
   defm mb: AVX512_maskable_3src<opc, MRMSrcMem, _, (outs _.RC:$dst),
                   (ins src_v.RC:$src2, f32mem:$src3),
                   OpcodeStr,
                   !strconcat("${src3}", _.BroadcastStr,", $src2"),
                   !strconcat("$src2, ${src3}", _.BroadcastStr),
-                  (_.VT (OpNode _.RC:$src1, src_v.RC:$src2,
-                  (src_v.VT (src_v.BroadcastLdFrag addr:$src3))))>,
+                  (null_frag)>,
                   EVEX_B, EVEX, VVVV, Sched<[sched.Folded, sched.ReadAfterFold]>;
 
 }

llvm/test/CodeGen/X86/fold-broadcast.ll

@@ -0,0 +1,18 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx512bf16 < %s | FileCheck %s
+
+define <16 x float> @mm512_dpbf16_ps_broadcast_rhs(<16 x float> noundef %acc, <32 x bfloat> noundef %lhs, ptr nocapture noundef readonly %rhs) {
+; CHECK-LABEL: mm512_dpbf16_ps_broadcast_rhs:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vdpbf16ps (%rdi){1to16}, %zmm1, %zmm0
+; CHECK-NEXT:    retq
+entry:
+  %0 = load float, ptr %rhs, align 4
+  %vecinit.i = insertelement <16 x float> poison, float %0, i64 0
+  %vecinit15.i = shufflevector <16 x float> %vecinit.i, <16 x float> poison, <16 x i32> zeroinitializer
+  %1 = bitcast <16 x float> %vecinit15.i to <32 x bfloat>
+  %2 = tail call <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float> %acc, <32 x bfloat> %lhs, <32 x bfloat> %1)
+  ret <16 x float> %2
+}
+
+declare <16 x float> @llvm.x86.avx512bf16.dpbf16ps.512(<16 x float>, <32 x bfloat>, <32 x bfloat>)