Fixed reduction kernel.

Merged with main.

numba_dpex/core/passes/parfor_lowering_pass.py

@@ -154,17 +154,20 @@ def _submit_parfor_kernel(
                     )
                 global_range.append(stop)
         else:
+            stop = None
             if kernel_type == 1:
                 stop = _load_range(lowerer, 1)
             else:
                 stop = reductionHelper.global_size_var
-            stop = _load_range(lowerer, stop)
+                stop = _load_range(lowerer, stop)
+            global_range.append(stop)
 
         local_range = []
         if kernel_type == 2:
             local_range.append(
                 _load_range(lowerer, reductionHelper.work_group_size)
             )
+
         # Submit a synchronous kernel
         ir_builder.submit_sync_ranged_kernel(
             curr_queue,
@@ -311,15 +314,13 @@ def _lower_parfor_as_kernel(self, lowerer, parfor):
                 parfor_reddict,
             )
 
-            print("-------> submit first NDrange kernel")
             self._submit_parfor_kernel(
                 lowerer,
                 psrfor_kernel,
                 loop_ranges,
                 2,
                 reductionHelperList[0],
             )
-            print("-----> first kernel is done ")
 
             psrfor_kernel = create_reduction_remainder_kernel_for_parfor(
                 parfor,
@@ -331,15 +332,13 @@ def _lower_parfor_as_kernel(self, lowerer, parfor):
                 reductionHelperList,
             )
 
-            print("-------> submit second range kernel")
             self._submit_parfor_kernel(
                 lowerer,
                 psrfor_kernel,
                 loop_ranges,
                 1,
                 reductionHelperList[0],
             )
-            print("-----> second kernel is done ")
 
             reductionKernelVar.copy_final_sum_to_host(psrfor_kernel)
             # ---------------------- End of Reduction codegen

numba_dpex/core/pipelines/kernel_compiler.py

@@ -65,10 +65,10 @@ def define_untyped_pipeline(state, name="dpex_kernel_untyped"):
         # Add pass to ensure when users allocate static constant memory the
         # size of the allocation is a constant and not specified by a closure
         # variable.
-        pm.add_pass(
-            ConstantSizeStaticLocalMemoryPass,
-            "dpex constant size for static local memory",
-        )
+        # pm.add_pass(
+        #    ConstantSizeStaticLocalMemoryPass,
+        #    "dpex constant size for static local memory",
+        # )
 
         # --- End of dpex passes added to the untyped pipeline               --#
 

numba_dpex/core/runtime/_dpexrt_python.c

@@ -1287,6 +1287,7 @@ static PyObject *build_c_helpers_dict(void)
     _declpointer("DPEXRTQueue_CreateFromFilterString",
                  &DPEXRTQueue_CreateFromFilterString);
     _declpointer("DpexrtQueue_SubmitRange", &DpexrtQueue_SubmitRange);
+    _declpointer("DpexrtQueue_SubmitNDRange", &DpexrtQueue_SubmitNDRange);
     _declpointer("DPEXRT_MemInfo_alloc", &DPEXRT_MemInfo_alloc);
     _declpointer("DPEXRT_MemInfo_fill", &DPEXRT_MemInfo_fill);
     _declpointer("NRT_ExternalAllocator_new_for_usm",
@@ -1350,6 +1351,8 @@ MOD_INIT(_dpexrt_python)
                        PyLong_FromVoidPtr(&DPEXRTQueue_CreateFromFilterString));
     PyModule_AddObject(m, "DpexrtQueue_SubmitRange",
                        PyLong_FromVoidPtr(&DpexrtQueue_SubmitRange));
+    PyModule_AddObject(m, "DpexrtQueue_SubmitNDRange",
+                       PyLong_FromVoidPtr(&DpexrtQueue_SubmitNDRange));
     PyModule_AddObject(m, "DPEXRT_MemInfo_alloc",
                        PyLong_FromVoidPtr(&DPEXRT_MemInfo_alloc));
     PyModule_AddObject(m, "DPEXRT_MemInfo_fill",

numba_dpex/core/utils/kernel_launcher.py

@@ -414,5 +414,5 @@ def submit_sync_ranged_kernel(
                 ),
                 self.context.get_constant(types.uintp, 0),
             ]
-            print("---------------> this is NDrange kernel")
+
             self.rtctx.submit_ndrange(self.builder, *args)

numba_dpex/core/utils/kernel_templates/partial_reduction_template.py

@@ -262,9 +262,6 @@ def _generate_remainder_kernel_as_string(self):
         for i, redvar in enumerate(self._redvars):
             gufunc_txt += f"        {self._final_sum_var_name[i]}[0] += {self._partial_sum_var_name[i]}[j]\n"
 
-        gufunc_txt += f"        print({self._final_sum_var_name[0]}[0])\n"
-        gufunc_txt += f"        print({self._final_sum_var_name[1]}[0])\n"
-
         gufunc_txt += (
             f"    for j in range ({self._global_size_mod_var_name[0]}) :\n"
         )
@@ -280,9 +277,6 @@ def _generate_remainder_kernel_as_string(self):
             + " = "
             + f"{self._global_size_var_name[0]} + j\n"
         )
-        gufunc_txt += f"        print({self._legal_loop_indices[0]})\n"
-        gufunc_txt += f"        print(a[{self._legal_loop_indices[0]}])\n"
-        gufunc_txt += f"        print(b[{self._legal_loop_indices[0]}])\n"
 
         for redvar in self._redvars:
             rtyp = str(self._typemap[redvar])
@@ -306,7 +300,6 @@ def _generate_remainder_kernel_as_string(self):
             redop = self._parfor_reddict[redvar].redop
             if redop == operator.iadd:
                 gufunc_txt += f"        {self._final_sum_var_name[i]}[0] +=  local_sums_{legal_redvar}[0]\n"
-                gufunc_txt += f"        print(local_sums_{legal_redvar}[0])\n"
             elif redop == operator.imul:
                 gufunc_txt += f"        {self._final_sum_var_name[i]}[0] *=  local_sums_{legal_redvar}[0]\n"
             else:

numba_dpex/core/utils/reduction_helper.py

@@ -122,7 +122,7 @@ def _allocate_partail_reduction_arrays(
                 types.literal(inputArrayType.usm_type),
             ),
         )
-        # shape, dtype=None, order="C", usm_type=None, device=None, sycl_queue=None
+        # shape, dtype=None, order="C", device=None, usm_type=None, sycl_queue=None
         sizeVar = pfbdr.make_tuple_variable(
             [self.partial_sum_size_var], name="tuple_sizeVar"
         )

numba_dpex/core/utils/reduction_kernel_builder.py

@@ -9,6 +9,8 @@
     get_unused_var_name,
     legalize_names,
     mk_unique_var,
+    remove_dels,
+    rename_labels,
     replace_var_names,
 )
 from numba.core.typing import signature
@@ -143,8 +145,8 @@ def create_reduction_main_kernel_for_parfor(
 
     # FIXME: Why rename and remove dels causes the partial_sum array update
     # instructions to be removed.
-    # kernel_ir.blocks = rename_labels(kernel_ir.blocks)
-    # remove_dels(kernel_ir.blocks)
+    kernel_ir.blocks = rename_labels(kernel_ir.blocks)
+    remove_dels(kernel_ir.blocks)
 
     old_alias = flags.noalias
 
@@ -154,7 +156,6 @@ def create_reduction_main_kernel_for_parfor(
     kernel_sig = signature(types.none, *kernel_param_types)
     exec_queue = typemap[parfor_args[0]].queue
 
-    breakpoint()
     sycl_kernel = _compile_kernel_parfor(
         exec_queue,
         kernel_name,

numba_dpex/dpnp_iface/_intrinsic.py

@@ -185,7 +185,7 @@ def impl_dpnp_zeros(
     ty_shape,
     ty_dtype,
     ty_order,
-    ty_like,
+    #    ty_like,
     ty_device,
     ty_usm_type,
     ty_sycl_queue,
@@ -223,7 +223,7 @@ def impl_dpnp_zeros(
         ty_shape,
         ty_dtype,
         ty_order,
-        ty_like,
+        #        ty_like,
         ty_device,
         ty_usm_type,
         ty_sycl_queue,
@@ -245,7 +245,7 @@ def impl_dpnp_ones(
     ty_shape,
     ty_dtype,
     ty_order,
-    ty_like,
+    # ty_like,
     ty_device,
     ty_usm_type,
     ty_sycl_queue,
@@ -283,7 +283,7 @@ def impl_dpnp_ones(
         ty_shape,
         ty_dtype,
         ty_order,
-        ty_like,
+        # ty_like,
         ty_device,
         ty_usm_type,
         ty_sycl_queue,

numba_dpex/dpnp_iface/arrayobj.py

@@ -309,7 +309,7 @@ def ol_dpnp_zeros(
     shape,
     dtype=None,
     order="C",
-    like=None,
+    # like=None,
     device=None,
     usm_type="device",
     sycl_queue=None,
@@ -355,7 +355,7 @@ def ol_dpnp_zeros(
     """
 
     _ndim = _ty_parse_shape(shape)
-    _layout = _parse_layout(order)
+    # _layout = _parse_layout(order)
     _dtype = _parse_dtype(dtype)
     _layout = _parse_layout(order)
     _usm_type = _parse_usm_type(usm_type) if usm_type is not None else "device"
@@ -377,7 +377,7 @@ def impl(
                 shape,
                 dtype=None,
                 order="C",
-                like=None,
+                # like=None,
                 device=None,
                 usm_type="device",
                 sycl_queue=None,
@@ -386,7 +386,7 @@ def impl(
                     shape,
                     _dtype,
                     order,
-                    like,
+                    # like,
                     _device,
                     _usm_type,
                     sycl_queue,
@@ -408,7 +408,7 @@ def ol_dpnp_ones(
     shape,
     dtype=None,
     order="C",
-    like=None,
+    # like=None,
     device=None,
     usm_type="device",
     sycl_queue=None,
@@ -475,7 +475,7 @@ def impl(
                 shape,
                 dtype=None,
                 order="C",
-                like=None,
+                # like=None,
                 device=None,
                 usm_type="device",
                 sycl_queue=None,
@@ -484,7 +484,7 @@ def impl(
                     shape,
                     _dtype,
                     order,
-                    like,
+                    # like,
                     _device,
                     _usm_type,
                     sycl_queue,