Fix ir generation for ttg.async_copy_global_to_local without mask (#7444)

We are calling `ir.value()` in the `async_copy_global_to_shared`,
however the default constructor for `value` was never defined, leading
to a crash.

Author: pawelszczerbuk

python/src/ir.cc

@@ -349,6 +349,7 @@ void init_triton_ir(py::module &&m) {
       });
 
   py::class_<Value>(m, "value", py::module_local())
+      .def(py::init<>())
       .def("set_attr",
            [](Value &self, std::string &name, Attribute &attr) -> void {
              if (Operation *definingOp = self.getDefiningOp())

python/test/gluon/test_frontend.py

@@ -1094,7 +1094,7 @@ def async_copy_kernel(inp, xnumel, XBLOCK: ttgl.constexpr):
     xindex = ttgl.arange(0, XBLOCK, block_layout)
     mask = tl.max_constancy(xindex < xnumel, 2)
 
-    async_copy.async_copy_global_to_shared(smem, inp + xindex, mask)
+    async_copy.async_copy_global_to_shared(smem, inp + xindex)
     async_copy.async_copy_global_to_shared(smem, inp + xindex, mask, cache_modifier=".ca", eviction_policy="evict_last",
                                            volatile=True)
 
@@ -1124,7 +1124,7 @@ def test_async_copy(fresh_knobs):
     %3 = arith.cmpi slt, %1, %2 {tt.constancy = dense<2> : tensor<1xi32>} : tensor<128xi32, #blocked> loc(#loc)
     %4 = tt.splat %arg0 : !tt.ptr<f16> -> tensor<128x!tt.ptr<f16>, #blocked> loc(#loc)
     %5 = tt.addptr %4, %1 : tensor<128x!tt.ptr<f16>, #blocked>, tensor<128xi32, #blocked> loc(#loc)
-    %6 = ttg.async_copy_global_to_local %5, %0 mask %3 : tensor<128x!tt.ptr<f16>, #blocked> -> <128xf16, #shared, #smem, mutable> loc(#loc)
+    %6 = ttg.async_copy_global_to_local %5, %0 : tensor<128x!tt.ptr<f16>, #blocked> -> <128xf16, #shared, #smem, mutable> loc(#loc)
     %7 = ttg.async_copy_global_to_local %5, %0 mask %3 cacheModifier = ca evictionPolicy = evict_last {isVolatile = true} : tensor<128x!tt.ptr<f16>, #blocked> -> <128xf16, #shared, #smem, mutable> loc(#loc)
     %8 = ttg.local_alloc : () -> !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc)
     ttng.async_copy_mbarrier_arrive %8 : !ttg.memdesc<1xi64, #shared, #smem, mutable> loc(#loc)