From 1ce552f3ef8d6455c10a9886191c1898594975e0 Mon Sep 17 00:00:00 2001
From: Changpeng Fang <changpeng.fang@gmail.com>
Date: Fri, 25 Oct 2019 13:08:04 -0700
Subject: [PATCH] AMDGPU: Fix the broken dominator tree when creating waterfall
 loop for resource descriptor

Summary:
  In loadSRsrcFromVGPR, if MBB is the same as Succ, Remiander is not the immediate dominator of Succ.

Reviewer:
  arsenm

Differential Revision:
  https://reviews.llvm.org/D69358
---
 llvm/lib/Target/AMDGPU/SIInstrInfo.cpp        |  4 +-
 ...r-descriptor-waterfall-loop-idom-update.ll | 48 +++++++++++++++++++
 2 files changed, 50 insertions(+), 2 deletions(-)
 create mode 100644 llvm/test/CodeGen/AMDGPU/vgpr-descriptor-waterfall-loop-idom-update.ll

diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index d97e6a62971b..6e9143305426 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -4451,12 +4451,12 @@ static void loadSRsrcFromVGPR(const SIInstrInfo &TII, MachineInstr &MI,
   // Update dominators. We know that MBB immediately dominates LoopBB, that
   // LoopBB immediately dominates RemainderBB, and that RemainderBB immediately
   // dominates all of the successors transferred to it from MBB that MBB used
-  // to dominate.
+  // to properly dominate.
   if (MDT) {
     MDT->addNewBlock(LoopBB, &MBB);
     MDT->addNewBlock(RemainderBB, LoopBB);
     for (auto &Succ : RemainderBB->successors()) {
-      if (MDT->dominates(&MBB, Succ)) {
+      if (MDT->properlyDominates(&MBB, Succ)) {
         MDT->changeImmediateDominator(Succ, RemainderBB);
       }
     }
diff --git a/llvm/test/CodeGen/AMDGPU/vgpr-descriptor-waterfall-loop-idom-update.ll b/llvm/test/CodeGen/AMDGPU/vgpr-descriptor-waterfall-loop-idom-update.ll
new file mode 100644
index 000000000000..3d00d845202b
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/vgpr-descriptor-waterfall-loop-idom-update.ll
@@ -0,0 +1,48 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -march=amdgcn -mcpu=gfx1010 | FileCheck %s --check-prefix=GCN
+
+define void @vgpr_descriptor_waterfall_loop_idom_update(<4 x i32>* %arg) {
+; GCN-LABEL: vgpr_descriptor_waterfall_loop_idom_update:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-NEXT:    s_waitcnt_vscnt null, 0x0
+; GCN-NEXT:    ; implicit-def: $vcc_hi
+; GCN-NEXT:  BB0_1: ; %bb0
+; GCN-NEXT:    ; =>This Loop Header: Depth=1
+; GCN-NEXT:    ; Child Loop BB0_2 Depth 2
+; GCN-NEXT:    v_add_co_u32_e64 v4, vcc_lo, v0, 8
+; GCN-NEXT:    s_mov_b32 s5, exec_lo
+; GCN-NEXT:    v_add_co_ci_u32_e32 v5, vcc_lo, 0, v1, vcc_lo
+; GCN-NEXT:    flat_load_dwordx2 v[2:3], v[0:1]
+; GCN-NEXT:    flat_load_dwordx2 v[4:5], v[4:5]
+; GCN-NEXT:  BB0_2: ; Parent Loop BB0_1 Depth=1
+; GCN-NEXT:    ; => This Inner Loop Header: Depth=2
+; GCN-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
+; GCN-NEXT:    v_readfirstlane_b32 s8, v2
+; GCN-NEXT:    v_readfirstlane_b32 s9, v3
+; GCN-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GCN-NEXT:    v_readfirstlane_b32 s10, v4
+; GCN-NEXT:    v_readfirstlane_b32 s11, v5
+; GCN-NEXT:    v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[2:3]
+; GCN-NEXT:    v_cmp_eq_u64_e64 s4, s[10:11], v[4:5]
+; GCN-NEXT:    s_and_b32 s4, vcc_lo, s4
+; GCN-NEXT:    s_and_saveexec_b32 s4, s4
+; GCN-NEXT:    s_nop 0
+; GCN-NEXT:    buffer_store_dword v0, v0, s[8:11], 0 offen
+; GCN-NEXT:    v_nop
+; GCN-NEXT:    s_xor_b32 exec_lo, exec_lo, s4
+; GCN-NEXT:    s_cbranch_execnz BB0_2
+; GCN-NEXT:  ; %bb.3: ; in Loop: Header=BB0_1 Depth=1
+; GCN-NEXT:    s_mov_b32 exec_lo, s5
+; GCN-NEXT:    s_branch BB0_1
+entry:
+  br label %bb0
+
+bb0:
+  %desc = load <4 x i32>, <4 x i32>* %arg, align 8
+  tail call void @llvm.amdgcn.buffer.store.f32(float undef, <4 x i32> %desc, i32 0, i32 undef, i1 zeroext false, i1 zeroext false)
+  br label %bb0
+}
+
+; Function Attrs: nounwind writeonly
+declare void @llvm.amdgcn.buffer.store.f32(float, <4 x i32>, i32, i32, i1 immarg, i1 immarg)