[llvm] [AMDGPU][SIInsertWaitcnts] Fix iota_range assertion when OtherMarks is empty in mergeAsyncMarks() (PR #193499)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,147 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1250 -run-pass=si-insert-waitcnts -o - %s | FileCheck %s
+
+# Regression test for mergeAsyncMarks() asserting when OtherMarks is empty.
+#
+# At a CFG join point where one predecessor has an ASYNCMARK (non-empty
+# AsyncMarks) and the other does not (empty OtherMarks), MergeCount becomes
+# min(0, N) = 0. Before the fix, seq_inclusive<unsigned>(1, 0) asserted
+# Begin <= End. After the fix the function returns early when either side
+# is empty.
+#
+# GLOBAL_LOAD_ASYNC_TO_LDS_B32 is a GFX1250 async LDS DMA instruction tracked
+# via ASYNC_CNT.  isAsync() returns true for it, so the score is recorded into
+# AsyncScore before ASYNCMARK pushes it onto AsyncMarks.
+#
+# The join block contains WAIT_ASYNCMARK 0 to consume the pending mark.
+# Before the fix, mergeAsyncMarks() asserted before reaching the wait.
+# After the fix the pass completes without asserting.
+#
+# Two patterns are tested:
+#   asyncmark_in_then - ASYNCMARK in the then-successor, else-successor is sync
+#   asyncmark_in_else - ASYNCMARK in the else-successor, then-successor is sync
+
+---
+# Pattern 1: ASYNCMARK in then-successor, else-successor is sync.
+name:            asyncmark_in_then
+tracksRegLiveness: true
+machineFunctionInfo:
+  occupancy:       8
+body:             |
+  ; CHECK-LABEL: name: asyncmark_in_then
+  ; CHECK: bb.0:
+  ; CHECK-NEXT:   successors: %bb.1(0x40000000), %bb.2(0x40000000)
+  ; CHECK-NEXT:   liveins: $sgpr0, $sgpr1, $vgpr0_vgpr1, $vgpr2
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   S_WAIT_LOADCNT_DSCNT 0
+  ; CHECK-NEXT:   S_WAIT_KMCNT 0
+  ; CHECK-NEXT:   S_CMP_LG_U32 $sgpr0, $sgpr1, implicit-def $scc
+  ; CHECK-NEXT:   S_CBRANCH_SCC1 %bb.2, implicit killed $scc
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
+  ; CHECK-NEXT:   liveins: $vgpr0_vgpr1, $vgpr2
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   GLOBAL_LOAD_ASYNC_TO_LDS_B32 $vgpr2, $vgpr0_vgpr1, 0, 0, implicit-def $asynccnt, implicit $exec, implicit $asynccnt :: (load (s32) from `ptr addrspace(1) poison`, addrspace 1), (store (s32) into `ptr addrspace(3) poison`, addrspace 3)
+  ; CHECK-NEXT:   ASYNCMARK
+  ; CHECK-NEXT:   S_BRANCH %bb.3
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   successors: %bb.3(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   S_BRANCH %bb.3
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.3:
+  ; CHECK-NEXT:   WAIT_ASYNCMARK 0
+  ; CHECK-NEXT:   S_ENDPGM 0
+  bb.0:
+    successors: %bb.1(0x40000000), %bb.2(0x40000000)
----------------
arsenm wrote:

```suggestion
```

https://github.com/llvm/llvm-project/pull/193499