[PATCH] D104815: [MCA] [TimelineView] Fixing a bug (that I caused) in TimelineView.cpp that was causing instructions that execute after timeline-max-cycles to still be displayed.

[Patrick Holland via Phabricator via llvm-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rG70040de32d73: [MCA][TimelineView] Fixed a bug that was causing instructions outside of the… (authored by holland11).

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D104815/new/

https://reviews.llvm.org/D104815

Files:
  llvm/tools/llvm-mca/Views/TimelineView.cpp


Index: llvm/tools/llvm-mca/Views/TimelineView.cpp
===================================================================
--- llvm/tools/llvm-mca/Views/TimelineView.cpp
+++ llvm/tools/llvm-mca/Views/TimelineView.cpp
@@ -288,6 +288,15 @@
   for (unsigned Iteration = 0; Iteration < Iterations; ++Iteration) {
     for (const MCInst &Inst : Source) {
       const TimelineViewEntry &Entry = Timeline[IID];
+      // When an instruction is retired after timeline-max-cycles,
+      // its CycleRetired is left at 0. However, it's possible for
+      // a 0 latency instruction to be retired during cycle 0 and we
+      // don't want to early exit in that case. The CycleExecuted
+      // attribute is set correctly whether or not it is greater
+      // than timeline-max-cycles so we can use that to ensure
+      // we don't early exit because of a 0 latency instruction.
+      if (Entry.CycleRetired == 0 && Entry.CycleExecuted != 0)
+        return;
 
       unsigned SourceIndex = IID % Source.size();
       printTimelineViewEntry(FOS, Entry, Iteration, SourceIndex);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D104815.354085.patch
Type: text/x-patch
Size: 1070 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210623/06f2fcca/attachment.bin>