Update mlir to d401987fe349a87c53fe25829215b080b70c0c1a (#3310)

Updates mlir.

Fixes #3298 
Fixes #3297 
Fixes #3296

---------

Co-authored-by: Sasha Lopoukhine <[email protected]>

.github/workflows/ci-mlir.yml

@@ -20,7 +20,7 @@ jobs:
 
     env:
       LLVM_SYMBOLIZER_PATH: /usr/lib/llvm-11/bin/llvm-symbolizer
-      MLIR-Version: 98e674c9f16d677d95c67bc130e267fae331e43c
+      MLIR-Version: d401987fe349a87c53fe25829215b080b70c0c1a
     steps:
     - uses: actions/checkout@v4
       with:

README.md

@@ -44,7 +44,7 @@ pip install xdsl
 
 *Note:* This version of xDSL is validated against a specific MLIR version,
 interoperability with other versions may result in problems. The supported
-MLIR version is commit `98e674c9f16d677d95c67bc130e267fae331e43c`.
+MLIR version is commit `d401987fe349a87c53fe25829215b080b70c0c1a`.
 
 ### Subprojects With Extra Dependencies
 

docs/irdl.ipynb

@@ -1476,11 +1476,11 @@
       "Operation does not verify: result at position 0 does not verify:\n",
       "attribute i32 expected from variable 'T', but got i64\n",
       "\n",
-      "%0 = \"arith.addi\"(%1, %1) : (i32, i32) -> i64\n",
+      "%0 = \"arith.addi\"(%1, %1) <{\"overflowFlags\" = #arith.overflow<none>}> : (i32, i32) -> i64\n",
       "\n",
       "\n",
       "\n",
-      "%0 = \"arith.addi\"(%1, %1) : (i32, i32) -> i64\n",
+      "%0 = \"arith.addi\"(%1, %1) <{\"overflowFlags\" = #arith.overflow<none>}> : (i32, i32) -> i64\n",
       "\n",
       "\n"
      ]

docs/mlir_interoperation.ipynb

@@ -167,7 +167,7 @@
       "\"builtin.module\"() ({\r\n",
       "  %0 = \"arith.constant\"() <{value = 1 : i32}> : () -> i32\r\n",
       "  %1 = \"arith.constant\"() <{value = 2 : i32}> : () -> i32\r\n",
-      "  %2 = \"arith.addi\"(%0, %1) : (i32, i32) -> i32\r\n",
+      "  %2 = \"arith.addi\"(%0, %1) <{overflowFlags = #arith.overflow<none>}> : (i32, i32) -> i32\r\n",
       "}) : () -> ()\r\n",
       "\r\n"
      ]

docs/mlir_interoperation.md

@@ -25,4 +25,4 @@ mlir-opt --convert-scf-to-cf --convert-cf-to-llvm --convert-func-to-llvm \
 The generated `tmp.ll` file contains LLVM IR, so it can be directly passed to
 the clang compiler. Notice that a `main` function is required for clang to
 build. The functionality is tested with the MLIR git commit hash:
-98e674c9f16d677d95c67bc130e267fae331e43c
+d401987fe349a87c53fe25829215b080b70c0c1a

tests/dialects/test_func.py

@@ -202,7 +202,7 @@ def test_call():
 "builtin.module"() ({
   "func.func"() <{"sym_name" = "func0", "function_type" = (i32, i32) -> i32}> ({
   ^0(%0 : i32, %1 : i32):
-    %2 = "arith.addi"(%0, %1) : (i32, i32) -> i32
+    %2 = "arith.addi"(%0, %1) <{"overflowFlags" = #arith.overflow<none>}> : (i32, i32) -> i32
     "func.return"(%2) : (i32) -> ()
   }) : () -> ()
   %0 = "arith.constant"() <{"value" = 1 : i32}> : () -> i32
@@ -248,7 +248,7 @@ def test_call_II():
 "builtin.module"() ({
   "func.func"() <{"sym_name" = "func1", "function_type" = (i32) -> i32}> ({
   ^0(%0 : i32):
-    %1 = "arith.addi"(%0, %0) : (i32, i32) -> i32
+    %1 = "arith.addi"(%0, %0) <{"overflowFlags" = #arith.overflow<none>}> : (i32, i32) -> i32
     "func.return"(%1) : (i32) -> ()
   }) : () -> ()
   %0 = "arith.constant"() <{"value" = 1 : i32}> : () -> i32

tests/dialects/test_gpu.py

@@ -261,13 +261,16 @@ def test_launch():
     assert launch.blockSizeX is ten.result
     assert launch.blockSizeY is ten.result
     assert launch.blockSizeZ is ten.result
+    assert launch.clusterSizeX is None
+    assert launch.clusterSizeY is None
+    assert launch.clusterSizeZ is None
     assert launch.asyncToken is None
     assert launch.asyncDependencies == tuple()
     assert launch.dynamicSharedMemorySize is None
 
     body2 = Region()
 
-    nd_launch = LaunchOp(body2, gridSize, blockSize, True, [], ten)
+    nd_launch = LaunchOp(body2, gridSize, blockSize, [], True, [], ten)
 
     assert isinstance(launch, LaunchOp)
     assert nd_launch.body is body2
@@ -277,6 +280,9 @@ def test_launch():
     assert nd_launch.blockSizeX is ten.result
     assert nd_launch.blockSizeY is ten.result
     assert nd_launch.blockSizeZ is ten.result
+    assert nd_launch.clusterSizeX is None
+    assert nd_launch.clusterSizeY is None
+    assert nd_launch.clusterSizeZ is None
     assert nd_launch.asyncToken is not None
     assert nd_launch.asyncToken.type == AsyncTokenType()
     assert nd_launch.asyncDependencies == tuple()

tests/dialects/test_scf.py

@@ -215,14 +215,14 @@ def test_parallel_block_arg_indextype():
     si = Constant.from_int_and_width(1, IndexType())
 
     b = Block(arg_types=[IndexType()])
-    b.add_op(Yield())
+    b.add_op(ReduceOp())
 
     body = Region(b)
     p = ParallelOp([lbi], [ubi], [si], body)
     p.verify()
 
     b2 = Block(arg_types=[i32])
-    b2.add_op(Yield())
+    b2.add_op(ReduceOp())
     body2 = Region(b2)
     p2 = ParallelOp([lbi], [ubi], [si], body2)
     with pytest.raises(VerifyException):
@@ -245,8 +245,7 @@ def test_parallel_verify_reduction_and_block_type():
     reduce_block = Block(arg_types=[i32, i32])
     reduce_block.add_ops([reduce_constant, rro])
 
-    b.add_op(ReduceOp(init_val, reduce_block))
-    b.add_op(Yield())
+    b.add_op(ReduceOp((init_val,), (Region(reduce_block),)))
 
     body = Region(b)
     p = ParallelOp([lbi], [ubi], [si], body, initVals)
@@ -270,76 +269,28 @@ def test_parallel_verify_reduction_and_block_type_fails():
     reduce_block = Block(arg_types=[i32, i32])
     reduce_block.add_ops([reduce_constant, rro])
 
-    b.add_op(ReduceOp(init_val, reduce_block))
-    b.add_op(Yield())
+    b.add_op(ReduceOp((init_val,), (Region(reduce_block),)))
 
     body = Region(b)
     p = ParallelOp([lbi], [ubi], [si], body, initVals)
     with pytest.raises(VerifyException):
         p.verify()
 
 
-def test_parallel_verify_yield_zero_ops():
-    lbi = Constant.from_int_and_width(0, IndexType())
-    ubi = Constant.from_int_and_width(10, IndexType())
-    si = Constant.from_int_and_width(1, IndexType())
-
-    val = Constant.from_int_and_width(10, i64)
-
-    b = Block(arg_types=[IndexType()])
-    b.add_op(Yield(val))
-    body = Region(b)
-    p = ParallelOp([lbi], [ubi], [si], body)
-    with pytest.raises(
-        VerifyException,
-        match="Single-block region terminator scf.yield has 1 operands "
-        "but 0 expected inside an scf.parallel",
-    ):
-        p.verify()
-
-    b2 = Block(arg_types=[IndexType()])
-    b2.add_op(Yield())
-    body2 = Region(b2)
-    p2 = ParallelOp([lbi], [ubi], [si], body2)
-    # Should verify as yield is empty
-    p2.verify()
-
-
-def test_parallel_test_count_number_reduction_ops():
-    @Builder.implicit_region
-    def body():
-        for i in range(10):
-            init_val = Constant.from_int_and_width(i, i32)
-            ReduceOp(init_val, Block())
-
-    p = ParallelOp([], [], [], body)
-    assert p.count_number_reduction_ops() == 10
-
-
-def test_parallel_get_arg_type_of_nth_reduction_op():
-    @Builder.implicit_region
-    def body():
-        init_val1 = Constant.from_int_and_width(10, i32)
-        init_val2 = Constant.from_int_and_width(10, i64)
-        for i in range(10):
-            ReduceOp(init_val1 if i % 2 == 0 else init_val2, Block())
-
-    p = ParallelOp([], [], [], body)
-    assert p.count_number_reduction_ops() == 10
-    for i in range(10):
-        assert p.get_arg_type_of_nth_reduction_op(i) == i32 if i % 2 == 0 else i64
-
-
 def test_reduce_op():
     init_val = Constant.from_int_and_width(10, i32)
 
-    reduce_op = ReduceOp(init_val, Block(arg_types=[i32, i32]))
+    reduce_op = ReduceOp((init_val,), (Region(Block(arg_types=[i32, i32])),))
 
-    assert reduce_op.argument is init_val.results[0]
-    assert reduce_op.argument.type is i32
-    assert len(reduce_op.body.blocks) == 1
-    assert len(reduce_op.body.block.args) == 2
-    assert reduce_op.body.block.args[0].type == reduce_op.body.block.args[0].type == i32
+    assert reduce_op.args[0] is init_val.results[0]
+    assert reduce_op.args[0].type is i32
+    assert len(reduce_op.reductions[0].blocks) == 1
+    assert len(reduce_op.reductions[0].block.args) == 2
+    assert (
+        reduce_op.reductions[0].block.args[0].type
+        == reduce_op.reductions[0].block.args[0].type
+        == i32
+    )
 
 
 def test_reduce_op_num_block_args():
@@ -351,21 +302,23 @@ def test_reduce_op_num_block_args():
         match="scf.reduce block must have exactly two arguments, but ",
     ):
         rro = ReduceReturnOp(reduce_constant)
-        ReduceOp(init_val, Block([rro], arg_types=[i32, i32, i32])).verify()
+        ReduceOp(
+            (init_val,), (Region(Block([rro], arg_types=[i32, i32, i32])),)
+        ).verify()
 
     with pytest.raises(
         VerifyException,
         match="scf.reduce block must have exactly two arguments, but ",
     ):
         rro = ReduceReturnOp(reduce_constant)
-        ReduceOp(init_val, Block([rro], arg_types=[i32])).verify()
+        ReduceOp((init_val,), (Region(Block([rro], arg_types=[i32])),)).verify()
 
     with pytest.raises(
         VerifyException,
         match="scf.reduce block must have exactly two arguments, but ",
     ):
         rro = ReduceReturnOp(reduce_constant)
-        ReduceOp(init_val, Block([rro], arg_types=[])).verify()
+        ReduceOp((init_val,), (Region(Block([rro], arg_types=[])),)).verify()
 
 
 def test_reduce_op_num_block_arg_types():
@@ -377,21 +330,21 @@ def test_reduce_op_num_block_arg_types():
         match="scf.reduce block argument types must be the same but have",
     ):
         rro = ReduceReturnOp(reduce_constant)
-        ReduceOp(init_val, Block([rro], arg_types=[i32, i64])).verify()
+        ReduceOp((init_val,), (Region(Block([rro], arg_types=[i32, i64])),)).verify()
 
     with pytest.raises(
         VerifyException,
         match="scf.reduce block argument types must be the same but have",
     ):
         rro = ReduceReturnOp(reduce_constant)
-        ReduceOp(init_val, Block([rro], arg_types=[i64, i32])).verify()
+        ReduceOp((init_val,), (Region(Block([rro], arg_types=[i64, i32])),)).verify()
 
 
 def test_reduce_op_num_block_arg_types_match_operand_type():
     init_val = Constant.from_int_and_width(10, i32)
 
     with pytest.raises(VerifyException):
-        ReduceOp(init_val, Block(arg_types=[i64, i64])).verify()
+        ReduceOp((init_val,), (Region(Block(arg_types=[i64, i64])),)).verify()
 
 
 def test_reduce_return_op_at_end():
@@ -401,13 +354,15 @@ def test_reduce_return_op_at_end():
     reduce_block.add_ops([reduce_constant, rro])
 
     init_val = Constant.from_int_and_width(10, i32)
-    ReduceOp(init_val, reduce_block).verify()
+    ReduceOp((init_val,), (Region(reduce_block),)).verify()
 
     with pytest.raises(
         VerifyException,
-        match="Block inside scf.reduce must terminate with an scf.reduce.return",
+        match="'scf.reduce' terminates with operation test.termop instead of scf.reduce.return",
     ):
-        ReduceOp(init_val, Block([TestTermOp.create()], arg_types=[i32, i32])).verify()
+        ReduceOp(
+            (init_val,), (Region(Block([TestTermOp.create()], arg_types=[i32, i32])),)
+        ).verify()
 
 
 def test_reduce_return_type_is_arg_type():
@@ -418,7 +373,7 @@ def test_reduce_return_type_is_arg_type():
 
     init_val = Constant.from_int_and_width(10, i64)
     with pytest.raises(VerifyException):
-        ReduceOp(init_val, reduce_block).verify()
+        ReduceOp((init_val,), (Region(reduce_block),)).verify()
 
 
 def test_reduce_return_op():
@@ -441,7 +396,7 @@ def test_reduce_return_type_is_operand_type():
         VerifyException,
         match="scf.reduce.return result type at end of scf.reduce block must",
     ):
-        ReduceOp(init_val, reduce_block).verify()
+        ReduceOp((init_val,), (Region(reduce_block),)).verify()
 
 
 def test_empty_else():

tests/dialects/test_transform.py

@@ -124,8 +124,8 @@ def test_get_result():
     target = test.TestOp(result_types=[transform.AnyOpType()]).results[0]
     result_number = 0
     assert_print_op(
-        transform.GetResultOp(target=target, result_number=result_number),
-        """%0 = "transform.get_result"(%1) <{"result_number" = 0 : i64}> : (!transform.any_op) -> !transform.any_value""",
+        transform.GetResultOp(target=target, raw_position_list=[result_number]),
+        """%0 = "transform.get_result"(%1) <{"raw_position_list" = array<i64: 0>}> : (!transform.any_op) -> !transform.any_value""",
         None,
     )
 

tests/filecheck/dialects/arith/arith_ops.mlir

@@ -171,6 +171,10 @@
 
   // CHECK-NEXT: %cmpf = arith.cmpf ogt, %lhsf32, %rhsf32 : f32
 
+  %cmpf_fm = "arith.cmpf"(%lhsf32, %rhsf32) {"predicate" = 2 : i64, "fastmath" = #arith.fastmath<fast>} : (f32, f32) -> i1
+
+  // CHECK-NEXT: %cmpf_fm = arith.cmpf ogt, %lhsf32, %rhsf32 fastmath<fast> : f32
+
   %selecti = "arith.select"(%lhsi1, %lhsi32, %rhsi32) : (i1, i32, i32) -> i32
   %selectf = "arith.select"(%lhsi1, %lhsf32, %rhsf32) : (i1, f32, f32) -> f32
 

tests/filecheck/dialects/csl/ops.mlir

@@ -658,7 +658,7 @@ csl.func @builtins() {
 // CHECK-GENERIC-NEXT:       %variable_with_default = "csl.variable"() <{"default" = 42 : i32}> : () -> !csl.var<i32>
 // CHECK-GENERIC-NEXT:       %variable = "csl.variable"() : () -> !csl.var<i32>
 // CHECK-GENERIC-NEXT:       %value = "csl.load_var"(%variable_with_default) : (!csl.var<i32>) -> i32
-// CHECK-GENERIC-NEXT:       %new_value = "arith.addi"(%value, %one) : (i32, i32) -> i32
+// CHECK-GENERIC-NEXT:       %new_value = "arith.addi"(%value, %one) <{"overflowFlags" = #arith.overflow<none>}> : (i32, i32) -> i32
 // CHECK-GENERIC-NEXT:       "csl.store_var"(%variable_with_default, %new_value) : (!csl.var<i32>, i32) -> ()
 // CHECK-GENERIC-NEXT:       "csl.store_var"(%variable, %new_value) : (!csl.var<i32>, i32) -> ()
 // CHECK-GENERIC-NEXT:       "csl.return"() : () -> ()

tests/filecheck/dialects/gpu/invalid.mlir

@@ -93,7 +93,7 @@
             }) {"op" = #gpu<all_reduce_op add>} : (index) -> index
             %final = "arith.muli"(%sum, %one) : (index, index) -> index
             "gpu.terminator"() : () -> ()
-        }) {"operandSegmentSizes" = array<i32: 0, 1, 1, 1, 1, 1, 1, 0>} : (index, index, index, index, index, index) -> ()
+        }) {"operandSegmentSizes" = array<i32: 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0>} : (index, index, index, index, index, index) -> ()
         "gpu.module_end"() : () -> ()
     }) {"sym_name" = "gpu"} : () -> ()
 }) {} : () -> ()
@@ -108,7 +108,7 @@
         %one = "arith.constant"() {"value" = 1 : index} : () -> index
 
         "gpu.launch"(%one, %one, %n, %one, %one, %one) ({})
-        {"operandSegmentSizes" = array<i32: 0, 1, 1, 1, 1, 1, 1, 0>} : (index, index, index, index, index, index) -> ()
+        {"operandSegmentSizes" = array<i32: 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0>} : (index, index, index, index, index, index) -> ()
         "gpu.module_end"() : () -> ()
     }) {"sym_name" = "gpu"} : () -> ()
 }) {} : () -> ()
@@ -136,7 +136,7 @@
     }) {} : () -> ()
 }) {} : () -> ()
 
-// CHECK: attribute sym_name expected
+// CHECK: property sym_name expected
 
 // -----