summaryrefslogtreecommitdiff
path: root/Documentation
diff options
context:
space:
mode:
authorAndy Ayers <andya@microsoft.com>2019-03-04 23:30:48 -0800
committerGitHub <noreply@github.com>2019-03-04 23:30:48 -0800
commit04fed62162092da2a824e425aa65b8fcfc70ce14 (patch)
tree3b439f5336944c2f163f1eb2946ff53e22878de3 /Documentation
parent7d54e6130f0db16b2eaa80272076ae3bd725cf69 (diff)
downloadcoreclr-04fed62162092da2a824e425aa65b8fcfc70ce14.tar.gz
coreclr-04fed62162092da2a824e425aa65b8fcfc70ce14.tar.bz2
coreclr-04fed62162092da2a824e425aa65b8fcfc70ce14.zip
JIT: remove unneeded ref count updating traversal from optimizer (#22954)
The ref count update traversal in the optimizer is not doing anything, so remove it. This was overlooked when we changed away from incremental updates in #19345. Also: fix up comments and documentation to reflect the current approach to local var ref counts.
Diffstat (limited to 'Documentation')
-rw-r--r--Documentation/botr/ryujit-overview.md15
1 files changed, 13 insertions, 2 deletions
diff --git a/Documentation/botr/ryujit-overview.md b/Documentation/botr/ryujit-overview.md
index bf8b2d4890..fad15f47c8 100644
--- a/Documentation/botr/ryujit-overview.md
+++ b/Documentation/botr/ryujit-overview.md
@@ -195,7 +195,9 @@ At this point, a number of analyses and transformations are done on the flowgrap
At this point, a number of properties are computed on the IR, and must remain valid for the remaining phases. We will call this “normalization”
-* `lvaMarkLocalVars` – set the reference counts (raw and weighted) for lclVars, sort them, and determine which will be tracked (currently up to 128). Note that after this point any transformation that adds or removes lclVar references must update the reference counts.
+* `lvaMarkLocalVars` – if this jit is optimizing, set the reference counts (raw and weighted) for lclVars, sort them, and determine which
+will be tracked (currently up to 512). If not optimizing, all locals are given an implicit reference count of one. Reference counts are not incrementally
+maintained. They can be recomputed if accurate counts are needed.
* `optOptimizeBools` – this optimizes Boolean expressions, and may change the flowgraph (why is it not done prior to reachability and dominators?)
* Link the trees in evaluation order (setting `gtNext` and `gtPrev` fields): and `fgFindOperOrder()` and `fgSetBlockOrder()`.
@@ -492,7 +494,16 @@ TreeNodeInfo:
## LclVar phase-dependent properties
-Prior to normalization, the reference counts (`lvRefCnt` and `lvRefCntWtd`) are not valid. After normalization they must be updated when lclVar references are added or removed.
+LclVar ref counts track the number of uses and weighted used of a local in the jit IR. There are two sequences of phases over which ref counts are valid,
+tracked via `lvaRefCountState`: an early sequence (state `RCS_EARLY`) and the normal sequence (state `RCS_NORMAL`). Requests for ref counts via `lvRefCnt`
+and `lvRefCntWtd` must be aware of the ref count state.
+
+Before struct promotion the ref counts are invalid. Struct promotion enables `RCS_EARLY` and it and subsequent phases through morph compute and uses ref counts
+on some locals to guide some struct optimizations. After morph the counts go back to longer being valid.
+
+The `RCS_NORMAL` sequence begins at normalization. Ref counts are computed and generally available via for the rest of the compilation phases.
+The counts are not incrementally maintained and may go stale as the IR is optimized or transformed, or maybe very approximate if the jit is not optimizing.
+They can be recomputed via `lvaComputeRefCounts` at points where accurate counts are valuable. Currently this happens before and after lower.
# Supporting technologies and components