[X86] Search through associative operators for BMI patterns (BLSI, BLSR, BLSMSK)

[goldstein.w.n]

Summary

(a & (-b)) & b is often lowered as:

%sub  = sub i32     0, %b
%and0 = and i32  %sub, %a
%and1 = and i32 %and0, %b

Which won't get detected by the BLSI pattern as b & -b are never in
the same SDNode.

This patch will do a small search through associative operators and try
and place BMI patterns in the same node so they will hit the pattern.

[craig.topper]

Isn’t the property “associative” rather than “commutative”?

[craig.topper]

llvm/lib/Target/X86/X86ISelLowering.cpp

48801

Do not use all caps for OPC

48808

If you only look through binops without going through bitcasts, extends, or truncates, you can assume the types will always match.

48828

You can use isNullConstant(Op.getOperand(0) here.

48837

isOneConstant(Op.getOperand(1))

48846

isAllOnesConstant(Op.getOperand(OpIdx)

48854

Lower case start of function name

48858

Just check VT != MVT::i32 && VT != MVT::i64 unless you're trying to support i33-i63?

48865

You can drop all the curly braces here.

49088

Space after if

49089

Over indented

52005

Space after if

52006

Over indented

[goldstein.w.n]

Fix format issues.
Remove unneeded code.
Cleanup code.

[goldstein.w.n]

Change commutable -> assosative in commit / code.

[goldstein.w.n]

Remove unneeded braces in a few more places

[goldstein.w.n]

ping.

[goldstein.w.n]

ping.

[goldstein.w.n]

ping2.

[pengfei]

llvm/lib/Target/X86/X86ISelLowering.cpp

48807

Add period in the end. The same for others below.

48811

SDLoc DL(Op);

48841

Can we assume it is canonical and just check operand 1?

[goldstein.w.n]

Fix nits + rebase

[pengfei]

LGTM.

[goldstein.w.n]

@pengfei I was thinking patch, is it liable to cause an infinite loop? Its just reordering and another place might undo its changes. Maybe it makes more sense in X86::ISelDAGToDag?