[llvm-dev] Should MemorySSA be updated when load/store address changes ?

[Momchil Velikov via llvm-dev]

This is a test case from `Transforms/GVN/PRE/pre-after-rle.ll`.

Before GVN, the MemorySSA beginning looks like:

    define void @fn1(i32** noalias %start, i32* %width, i32 %h) {
    entry:
    ; 1 = MemoryDef(liveOnEntry)
      %call = tail call noalias i8* @malloc(i64 1024)
      %call.cast = bitcast i8* %call to i32*
    ; 2 = MemoryDef(1)
      store i32* %call.cast, i32** %start, align 8
      br label %preheader

    preheader:                                        ; preds = %body,
%entry
    ; 5 = MemoryPhi({entry,2},{body,3})
      %cmp = icmp slt i32 1, %h
      br i1 %cmp, label %body, label %exit

    body:                                             ; preds = %body,
%preheader
    ; 4 = MemoryPhi({preheader,5},{body,3})
      %j = phi i32 [ 0, %preheader ], [ %j.next, %body ]
    ; MemoryUse(2) MayAlias
      %s = load i32*, i32** %start, align 8
      %idx = getelementptr inbounds i32, i32* %s, i64 0
    ; 3 = MemoryDef(4)
      store i32 0, i32* %idx, align 4
      %j.next = add nuw nsw i32 %j, 1
    ; MemoryUse(3) MayAlias
      %w = load i32, i32* %width, align 8
      %cmp3 = icmp slt i32 %j.next, %w
      br i1 %cmp3, label %body, label %preheader

    exit:                                             ; preds = %preheader
      ret void
    }


Note that `%w = load ...` is a `MemoryUse` of `store i32 0, i32* %idx`.

After GVN has removed the fully redundant load `%s = ...`, the
MemorySSA looks like:

    define void @fn1(i32** noalias %start, i32* %width, i32 %h) {
    entry:
    ; 1 = MemoryDef(liveOnEntry)
      %call = tail call noalias i8* @malloc(i64 1024)
      %call.cast = bitcast i8* %call to i32*
    ; 2 = MemoryDef(1)
      store i32* %call.cast, i32** %start, align 8
      br label %preheader

    preheader:                                        ; preds = %body,
%entry
    ; 5 = MemoryPhi({entry,2},{body,3})
      %cmp = icmp slt i32 1, %h
      br i1 %cmp, label %body, label %exit

    body:                                             ; preds = %body,
%preheader
    ; 4 = MemoryPhi({preheader,5},{body,3})
      %j = phi i32 [ 0, %preheader ], [ %j.next, %body ]
    ; 3 = MemoryDef(4)
      store i32 0, i32* %call.cast, align 4
      %j.next = add nuw nsw i32 %j, 1
    ; MemoryUse(3) MayAlias
      %w = load i32, i32* %width, align 8
      %cmp3 = icmp slt i32 %j.next, %w
      br i1 %cmp3, label %body, label %preheader

    exit:                                             ; preds = %preheader
      ret void
    }

The `%w = load ...` is still a use of the `store i32 0, i32*...`.

However, the locations now can be determined to not alias, and the
MemorySSA walker in fact skips `3 = MemoryDef(4)` and returns
`liveOnEntry` when queried for the clobber of `%w = load ...`.

So far so good, but then the `%w = load ... ` is not among the uses of
`liveOnEntry`.

This seems to contradict this part from the MemorySSA docs:

    "Unlike other partitioned forms, LLVM’s MemorySSA does make one
    useful guarantee - all loads are optimized to point at the thing
    that actually clobbers them. This gives some nice properties. For
    example, for a given store, you can find all loads actually
    clobbered by that store by walking the immediate uses of the
    store."

Is this the intended behaviour? Should MemorySSAUpdater have an update
API for when the address operands of loads/stores change?

~chill

-- 
Compiler scrub, Arm
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20220105/67d2e1aa/attachment.html>