From 221f28123ab0adb2df3007a64836a06e73b40a0b Mon Sep 17 00:00:00 2001
From: Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com>
Date: Mon, 8 Sep 2025 13:40:09 -0700
Subject: [PATCH] [AMDGPU] Update hard-clauses-load-monitor.mir. NFC

---
 .../AMDGPU/hard-clauses-load-monitor.mir      | 40 ++++++++++++++++++-
 1 file changed, 38 insertions(+), 2 deletions(-)

diff --git a/llvm/test/CodeGen/AMDGPU/hard-clauses-load-monitor.mir b/llvm/test/CodeGen/AMDGPU/hard-clauses-load-monitor.mir
index db4b946be25bb..2dccac136d24b 100644
--- a/llvm/test/CodeGen/AMDGPU/hard-clauses-load-monitor.mir
+++ b/llvm/test/CodeGen/AMDGPU/hard-clauses-load-monitor.mir
@@ -1,6 +1,42 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -run-pass si-insert-hard-clauses %s -o - | FileCheck %s -check-prefixes=GCN
-# RUN: llc -mtriple=amdgcn -mcpu=gfx1250 -run-pass si-insert-hard-clauses %s -o - | FileCheck %s -check-prefixes=GCN
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -verify-machineinstrs -run-pass si-insert-hard-clauses %s -o - | FileCheck %s -check-prefixes=GCN,GFX1200
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1250 -run-pass si-insert-hard-clauses %s -o - | FileCheck %s -check-prefixes=GCN,GFX1250
+
+---
+name: non_flat_vmem_loads
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2
+    ; GFX1200-LABEL: name: non_flat_vmem_loads
+    ; GFX1200: liveins: $vgpr0, $vgpr1, $vgpr2
+    ; GFX1200-NEXT: {{  $}}
+    ; GFX1200-NEXT: BUNDLE implicit-def $vgpr6_vgpr7, implicit-def $vgpr8_vgpr9, implicit-def $vgpr10_vgpr11, implicit $vgpr0_vgpr1, implicit $exec, implicit $m0 {
+    ; GFX1200-NEXT:   S_CLAUSE 2
+    ; GFX1200-NEXT:   $vgpr6_vgpr7 = GLOBAL_LOAD_TR_B64_w32 $vgpr0_vgpr1, 8, 0, implicit $exec
+    ; GFX1200-NEXT:   $vgpr8_vgpr9 = CLUSTER_LOAD_B64 $vgpr0_vgpr1, 16, 0, implicit $m0, implicit $exec
+    ; GFX1200-NEXT:   $vgpr10_vgpr11 = GLOBAL_LOAD_MONITOR_B64 $vgpr0_vgpr1, 24, 0, implicit $exec
+    ; GFX1200-NEXT: }
+    ; GFX1200-NEXT: GLOBAL_LOAD_ASYNC_TO_LDS_B32 $vgpr2, $vgpr0_vgpr1, 0, 0, implicit-def $asynccnt, implicit $exec, implicit $asynccnt
+    ; GFX1200-NEXT: $vgpr4 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 32, 0, implicit $exec
+    ;
+    ; GFX1250-LABEL: name: non_flat_vmem_loads
+    ; GFX1250: liveins: $vgpr0, $vgpr1, $vgpr2
+    ; GFX1250-NEXT: {{  $}}
+    ; GFX1250-NEXT: BUNDLE implicit-def $vgpr6_vgpr7, implicit-def $vgpr8_vgpr9, implicit-def $vgpr10_vgpr11, implicit-def $asynccnt, implicit-def $vgpr4, implicit $vgpr0_vgpr1, implicit $exec, implicit $m0, implicit $vgpr2, implicit $asynccnt {
+    ; GFX1250-NEXT:   S_CLAUSE 4
+    ; GFX1250-NEXT:   $vgpr6_vgpr7 = GLOBAL_LOAD_TR_B64_w32 $vgpr0_vgpr1, 8, 0, implicit $exec
+    ; GFX1250-NEXT:   $vgpr8_vgpr9 = CLUSTER_LOAD_B64 $vgpr0_vgpr1, 16, 0, implicit $m0, implicit $exec
+    ; GFX1250-NEXT:   $vgpr10_vgpr11 = GLOBAL_LOAD_MONITOR_B64 $vgpr0_vgpr1, 24, 0, implicit $exec
+    ; GFX1250-NEXT:   GLOBAL_LOAD_ASYNC_TO_LDS_B32 $vgpr2, $vgpr0_vgpr1, 0, 0, implicit-def $asynccnt, implicit $exec, implicit $asynccnt
+    ; GFX1250-NEXT:   $vgpr4 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 32, 0, implicit $exec
+    ; GFX1250-NEXT: }
+    $vgpr6_vgpr7 = GLOBAL_LOAD_TR_B64_w32 $vgpr0_vgpr1, 8, 0, implicit $exec
+    $vgpr8_vgpr9 = CLUSTER_LOAD_B64 $vgpr0_vgpr1, 16, 0, implicit $m0, implicit $exec
+    $vgpr10_vgpr11 = GLOBAL_LOAD_MONITOR_B64 $vgpr0_vgpr1, 24, 0, implicit $exec
+    GLOBAL_LOAD_ASYNC_TO_LDS_B32 $vgpr2, $vgpr0_vgpr1, 0, 0, implicit-def $asynccnt, implicit $exec, implicit $asynccnt
+    $vgpr4 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 32, 0, implicit $exec
+...
 
 ---
 name: async_load_flat_monitor_load