Early CSE and GVN miss an assume optimization when the value isn't used directly

[llvmbot]

Bugzilla Link	28086
Version	trunk
OS	All
Attachments	Minimum reproducing example
Reporter	LLVM Bugzilla Contributor
CC	@hfinkel

Extended Description

See https://groups.google.com/forum/#!topic/llvm-dev/90lQByG6teg for the conversation.

Here's a minimum reproducing example:

define i32 @​foo(i32*) {
entry:
%1 = load i32, i32* %0
%2 = icmp eq i32 %1, 3
call void @​llvm.assume(i1 %2)
%3 = load i32, i32* %0
ret i32 %3
}

Both -early-cse and -gvn independently simplify this IR to:

define i32 @​foo(i32*) {
entry:
%1 = load i32, i32* %0
ret i32 %1
}

Instead of:

define i32 @​foo(i32*) {
entry:
ret i32 3
}

[nikic]

GVN handles this an EarlyCSE at least doesn't lose the assume: https://llvm.godbolt.org/z/T638EM8re So I believe this is resolved.