From 402ee51532762b6612ece4fcfa89e8215fccce00 Mon Sep 17 00:00:00 2001
From: Vikram Hegde <Vikram.Hegde@amd.com>
Date: Sun, 25 Jan 2026 09:45:44 +0530
Subject: [PATCH] [AMDGPU] Fix DomTree preservation in SILowerControlFlow when
 nodes are removed (#176691)

I would think the "applyUpdates" API should also take care of deleting
the nodes from the analyses objects but this does not seem to be the
case. https://godbolt.org/z/38a7rfzjd shows an example where
SILowerControlFlow removes a basic block which is not removed in the
Post dominator tree.
---
 llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp |  5 ++
 ...i-lower-control-flow-preserve-dom-tree.mir | 65 +++++++++++++++++++
 2 files changed, 70 insertions(+)
 create mode 100644 llvm/test/CodeGen/AMDGPU/si-lower-control-flow-preserve-dom-tree.mir

diff --git a/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp b/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
index bb912be85de7..9cc86e84407b 100644
--- a/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
+++ b/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
@@ -744,6 +744,11 @@ bool SILowerControlFlow::removeMBBifRedundant(MachineBasicBlock &MBB) {
   if (PDT)
     PDT->applyUpdates(DTUpdates);
 
+  if (MDT && MDT->getNode(&MBB))
+    MDT->eraseNode(&MBB);
+  if (PDT && PDT->getNode(&MBB))
+    PDT->eraseNode(&MBB);
+
   MBB.clear();
   MBB.eraseFromParent();
   if (FallThrough && !FallThrough->isLayoutSuccessor(Succ)) {
diff --git a/llvm/test/CodeGen/AMDGPU/si-lower-control-flow-preserve-dom-tree.mir b/llvm/test/CodeGen/AMDGPU/si-lower-control-flow-preserve-dom-tree.mir
new file mode 100644
index 000000000000..1ef3d63a0496
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/si-lower-control-flow-preserve-dom-tree.mir
@@ -0,0 +1,65 @@
+# RUN: llc -mcpu=gfx1201 -mtriple=amdgcn-amd-amdhsa -passes="require<machine-post-dom-tree>,require<machine-dom-tree>,si-lower-control-flow,print<machine-post-dom-tree>,print<machine-dom-tree>" -filetype=null %s 2>&1 | FileCheck %s
+
+# CHECK: Inorder PostDominator Tree: 
+# CHECK-NEXT:  [1]  <<exit node>> {4294967295,4294967295} [0]
+# CHECK-NEXT:    [2] %bb.4 {4294967295,4294967295} [1]
+# CHECK-NEXT:      [3] %bb.0 {4294967295,4294967295} [2]
+# CHECK-NEXT:      [3] %bb.1 {4294967295,4294967295} [2]
+# CHECK-NEXT:      [3] %bb.2 {4294967295,4294967295} [2]
+
+# CHECK: Inorder Dominator Tree:
+# CHECK-NEXT:  [1] %bb.0 {4294967295,4294967295} [0]
+# CHECK-NEXT:    [2] %bb.1 {4294967295,4294967295} [1]
+# CHECK-NEXT:      [3] %bb.2 {4294967295,4294967295} [2]
+# CHECK-NEXT:    [2] %bb.4 {4294967295,4294967295} [1]
+
+---
+name:            preserve_dom_tree
+body:             |
+  bb.0.entry:
+    successors: %bb.1(0x40000000), %bb.4(0x40000000)
+    liveins: $vgpr0, $sgpr0_sgpr1, $ttmp9
+  
+    %0:sreg_32 = COPY $ttmp9
+    %1:sgpr_64 = COPY killed $sgpr0_sgpr1
+    %2:vgpr_32 = COPY killed $vgpr0
+    %3:sgpr_128 = S_LOAD_DWORDX4_IMM %1, 0, 0
+    %4:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM killed %1, 28, 0
+    %5:sreg_64_xexec_xnull = REG_SEQUENCE %3.sub2, %subreg.sub0, %3.sub3, %subreg.sub1
+    %6:sreg_32 = S_AND_B32 killed %4, 65535, implicit-def dead $scc
+    %7:vreg_64 = REG_SEQUENCE killed %2(s32), %subreg.sub0, undef %8:vgpr_32, %subreg.sub1
+    %9:vreg_64, $sgpr_null = V_MAD_U64_U32_e64 killed %0, killed %6, killed %7, 0, implicit $exec
+    %10:vgpr_32 = GLOBAL_LOAD_UBYTE_SADDR killed %5, killed %9.sub0, 0, 0, implicit $exec
+    %11:vgpr_32 = V_AND_B32_e64 1, killed %10, implicit $exec
+    %12:sreg_32 = V_CMP_EQ_U32_e64 1, killed %11, implicit $exec
+    %13:sreg_32 = SI_IF killed %12, %bb.4, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+    S_BRANCH %bb.1
+  
+  bb.1:
+    successors: %bb.2(0x40000000), %bb.3(0x40000000)
+  
+    %14:sreg_32 = COPY $exec_lo
+    %15:vgpr_32 = V_MBCNT_LO_U32_B32_e64 %14, 0, implicit $exec
+    %16:sreg_32 = V_CMP_EQ_U32_e64 0, killed %15, implicit $exec
+    %17:sreg_32 = SI_IF killed %16, %bb.3, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+    S_BRANCH %bb.2
+  
+  bb.2:
+    successors: %bb.3(0x80000000)
+  
+    %18:sreg_32 = S_BCNT1_I32_B32 killed %14, implicit-def dead $scc
+    %19:vgpr_32 = V_CVT_F32_UBYTE0_e64 killed %18, 0, 0, implicit $exec
+    %20:vgpr_32 = nofpexcept V_MUL_F32_e64 0, 1092616192, 0, killed %19, 0, 0, implicit $mode, implicit $exec
+    %21:sreg_64_xexec_xnull = REG_SEQUENCE killed %3.sub0, %subreg.sub0, %3.sub1, %subreg.sub1
+    %22:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
+    GLOBAL_ATOMIC_ADD_F32_SADDR killed %22, killed %20, killed %21, 0, 0, implicit $exec
+  
+  bb.3:
+    successors: %bb.4(0x80000000)
+  
+    SI_END_CF killed %17, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+  
+  bb.4:
+    SI_END_CF killed %13, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
+    S_ENDPGM 0
+...