Currently, these are the matchers that exist in mlir query:
"hasArgument", hasArgument(op, index) // Matches operation at the given argument index
"definedBy", definedBy(op) // Matches operation defined by op
"getDefinitions", getDefinitions(op, hops) // Matches definitions with hops=hops
"getAllDefinitions", getAllDefinitions(op, hops) // Matches all definitions with hops<=hops
"usedBy", usedBy(op) // Matches all operations used by op
"getUses", getUses(op, hops) // Matches uses with hops=hops
"getAllUses", getAllUses(op, hops) // Matches all uses with hops<=hops
"isConstantOp", isConstantOp() // Matches all contant operations
"hasOpAttr", hasOpAttr(attrName) // Matches all operations that has the given attrName
"hasOpName", hasOpName(opName) // Matches all operations with the given opName
To extract all the definitions of the operation test.use_coo upto 2 hops away (into a function), we can do:
$ mlir-query test1.mlir
mlir-query> m getAllDefinitions(hasOpName("test.use_coo"), 2).extract()which will return
func.func @extracted(%arg0: i32, %arg1: i32, %arg2: i32, %arg3: i32, %arg4: i32) {
%c1_i32 = arith.constant 1 : i32
%0 = "test.one_result"(%arg0, %arg1) : (i32, i32) -> i32
%1 = "test.boo"(%arg2, %arg3) : (i32, i32) -> i32
%2 = "test.coo"(%1, %0, %c1_i32) : (i32, i32, i32) -> i32
return
}