[PATCH] D37648: [SLPVectorizer] Fix PR21780 Expansion of 256 bit vector loads fails to fold into shuffles

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added inline comments.


================
Comment at: docs/LangRef.rst:4893
+
+``speculation.marker`` metadata must be attached to a load. It consist of
+set of ``i64`` type offsets indicating that the memory from load pointer address
----------------
"consists"


================
Comment at: docs/LangRef.rst:4894
+``speculation.marker`` metadata must be attached to a load. It consist of
+set of ``i64`` type offsets indicating that the memory from load pointer address
+is accessable for read operation with provided offsets. The intent of this
----------------
Is it a set or a min/max range? The examples only ever have 2 entries. If a set do they have to be sorted?


================
Comment at: docs/LangRef.rst:4896
+is accessable for read operation with provided offsets. The intent of this
+metadata is keep notion of certain memory as dereferanceable after load
+operations to that memory were deleted, but it might be beneficially to keep
----------------
"is to keep track of dereferanceable memory locations after load operations to that memory were deleted, as it might be beneficial for future optimizations."


================
Comment at: include/llvm/IR/LLVMContext.h:104
     MD_callees = 23,                  // "callees"
+    MD_speculation_marker = 24,        // "speculation.marker"
   };
----------------
// alignment


================
Comment at: include/llvm/Transforms/Vectorize/SLPVectorizer.h:154
   WeakTrackingVHListMap GEPs;
+
 };
----------------
Unnecessary whitespace change


https://reviews.llvm.org/D37648