Cleanup redundant code (#139)

* Cleanup redundant code

* Adhering to contributing guideline

* Change import

Co-authored-by: reazul.hoque <reazul.hoque@intel.com>

Author: reazulhoque

numba_dppy/dpctl_functions.py

@@ -2,29 +2,31 @@
 from numba.core.typing import signature
 
 
-class _DPCTL_FUNCTIONS:
-    @classmethod
-    def dpctl_get_current_queue(cls):
-        ret_type = types.voidptr
-        sig = signature(ret_type)
-        return types.ExternalFunction("DPCTLQueueMgr_GetCurrentQueue", sig)
-
-    @classmethod
-    def dpctl_malloc_shared(cls):
-        ret_type = types.voidptr
-        sig = signature(ret_type, types.int64, types.voidptr)
-        return types.ExternalFunction("DPCTLmalloc_shared", sig)
-
-    @classmethod
-    def dpctl_queue_memcpy(cls):
-        ret_type = types.void
-        sig = signature(
-            ret_type, types.voidptr, types.voidptr, types.voidptr, types.int64
-        )
-        return types.ExternalFunction("DPCTLQueue_Memcpy", sig)
-
-    @classmethod
-    def dpctl_free_with_queue(cls):
-        ret_type = types.void
-        sig = signature(ret_type, types.voidptr, types.voidptr)
-        return types.ExternalFunction("DPCTLfree_with_queue", sig)
+def dpctl_get_current_queue():
+    ret_type = types.voidptr
+    sig = signature(ret_type)
+    return types.ExternalFunction("DPCTLQueueMgr_GetCurrentQueue", sig)
+
+
+def dpctl_malloc_shared():
+    ret_type = types.voidptr
+    sig = signature(ret_type, types.int64, types.voidptr)
+    return types.ExternalFunction("DPCTLmalloc_shared", sig)
+
+
+def dpctl_queue_memcpy():
+    ret_type = types.void
+    sig = signature(ret_type, types.voidptr, types.voidptr, types.voidptr, types.int64)
+    return types.ExternalFunction("DPCTLQueue_Memcpy", sig)
+
+
+def dpctl_free_with_queue():
+    ret_type = types.void
+    sig = signature(ret_type, types.voidptr, types.voidptr)
+    return types.ExternalFunction("DPCTLfree_with_queue", sig)
+
+
+get_current_queue = dpctl_get_current_queue()
+malloc_shared = dpctl_malloc_shared()
+queue_memcpy = dpctl_queue_memcpy()
+free_with_queue = dpctl_free_with_queue()

numba_dppy/dpnp_glue/dpnp_linalgimpl.py

@@ -5,14 +5,13 @@
 import numba_dppy.dpnp_glue as dpnp_lowering
 from numba.core.extending import overload, register_jitable
 import numpy as np
-from numba_dppy.dpctl_functions import _DPCTL_FUNCTIONS
+from numba_dppy import dpctl_functions
 
 
 @overload(stubs.dpnp.eig)
 def dpnp_eig_impl(a):
     name = "eig"
     dpnp_lowering.ensure_dpnp(name)
-    dpctl_functions = dpnp_ext._DPCTL_FUNCTIONS()
 
     ret_type = types.void
     """
@@ -26,16 +25,11 @@ def dpnp_eig_impl(a):
     sig = signature(ret_type, types.voidptr, types.voidptr, types.voidptr, types.intp)
     dpnp_eig = dpnp_ext.dpnp_func("dpnp_" + name, [a.dtype.name, "NONE"], sig)
 
-    get_sycl_queue = dpctl_functions.dpctl_get_current_queue()
-    allocate_usm_shared = dpctl_functions.dpctl_malloc_shared()
-    copy_usm = dpctl_functions.dpctl_queue_memcpy()
-    free_usm = dpctl_functions.dpctl_free_with_queue()
-
     res_dtype = np.float64
     if a.dtype == types.float32:
         res_dtype = np.float32
 
-    def dpnp_eig_impl(a):
+    def dpnp_impl(a):
         n = a.shape[-1]
         if a.shape[-2] != n:
             msg = "Last 2 dimensions of the array must be square."
@@ -49,34 +43,37 @@ def dpnp_eig_impl(a):
         if n == 0:
             return (wr, vr)
 
-        sycl_queue = get_sycl_queue()
-        a_usm = allocate_usm_shared(a.size * a.itemsize, sycl_queue)
-        copy_usm(sycl_queue, a_usm, a.ctypes, a.size * a.itemsize)
+        sycl_queue = dpctl_functions.get_current_queue()
+        a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
+        dpctl_functions.queue_memcpy(sycl_queue, a_usm, a.ctypes, a.size * a.itemsize)
 
-        wr_usm = allocate_usm_shared(wr.size * wr.itemsize, sycl_queue)
-        vr_usm = allocate_usm_shared(vr.size * vr.itemsize, sycl_queue)
+        wr_usm = dpctl_functions.malloc_shared(wr.size * wr.itemsize, sycl_queue)
+        vr_usm = dpctl_functions.malloc_shared(vr.size * vr.itemsize, sycl_queue)
 
         dpnp_eig(a_usm, wr_usm, vr_usm, n)
 
-        copy_usm(sycl_queue, wr.ctypes, wr_usm, wr.size * wr.itemsize)
-        copy_usm(sycl_queue, vr.ctypes, vr_usm, vr.size * vr.itemsize)
+        dpctl_functions.queue_memcpy(
+            sycl_queue, wr.ctypes, wr_usm, wr.size * wr.itemsize
+        )
+        dpctl_functions.queue_memcpy(
+            sycl_queue, vr.ctypes, vr_usm, vr.size * vr.itemsize
+        )
 
-        free_usm(a_usm, sycl_queue)
-        free_usm(wr_usm, sycl_queue)
-        free_usm(vr_usm, sycl_queue)
+        dpctl_functions.free_with_queue(a_usm, sycl_queue)
+        dpctl_functions.free_with_queue(wr_usm, sycl_queue)
+        dpctl_functions.free_with_queue(vr_usm, sycl_queue)
 
         dpnp_ext._dummy_liveness_func([wr.size, vr.size])
 
         return (wr, vr)
 
-    return dpnp_eig_impl
+    return dpnp_impl
 
 
 @overload(stubs.dpnp.matmul)
 @overload(stubs.dpnp.dot)
 def dpnp_dot_impl(a, b):
     dpnp_lowering.ensure_dpnp("dot")
-    dpctl_functions = dpnp_ext._DPCTL_FUNCTIONS()
 
     ret_type = types.void
     """
@@ -100,11 +97,6 @@ def dpnp_dot_impl(a, b):
         types.intp,
     )
 
-    get_sycl_queue = dpctl_functions.dpctl_get_current_queue()
-    allocate_usm_shared = dpctl_functions.dpctl_malloc_shared()
-    copy_usm = dpctl_functions.dpctl_queue_memcpy()
-    free_usm = dpctl_functions.dpctl_free_with_queue()
-
     res_dtype = np.float64
     if a.dtype == types.int32 and b.dtype == types.int32:
         res_dtype = np.int32
@@ -144,30 +136,36 @@ def dpnp_dot_impl(a, b):
         dpnp_func = dpnp_ext.dpnp_func("dpnp_matmul", [a.dtype.name, "NONE"], sig)
 
         def dot_2_mm(a, b):
-            sycl_queue = get_sycl_queue()
+            sycl_queue = dpctl_functions.get_current_queue()
 
             m, k = a.shape
             _k, n = b.shape
 
             if _k != k:
                 raise ValueError("Incompatible array sizes for np.dot(a, b)")
 
-            a_usm = allocate_usm_shared(a.size * a.itemsize, sycl_queue)
-            copy_usm(sycl_queue, a_usm, a.ctypes, a.size * a.itemsize)
+            a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, a_usm, a.ctypes, a.size * a.itemsize
+            )
 
-            b_usm = allocate_usm_shared(b.size * b.itemsize, sycl_queue)
-            copy_usm(sycl_queue, b_usm, b.ctypes, b.size * b.itemsize)
+            b_usm = dpctl_functions.malloc_shared(b.size * b.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, b_usm, b.ctypes, b.size * b.itemsize
+            )
 
             out = np.empty((m, n), dtype=res_dtype)
-            out_usm = allocate_usm_shared(out.size * out.itemsize, sycl_queue)
+            out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize, sycl_queue)
 
             dpnp_func(a_usm, b_usm, out_usm, m, n, k)
 
-            copy_usm(sycl_queue, out.ctypes, out_usm, out.size * out.itemsize)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, out.ctypes, out_usm, out.size * out.itemsize
+            )
 
-            free_usm(a_usm, sycl_queue)
-            free_usm(b_usm, sycl_queue)
-            free_usm(out_usm, sycl_queue)
+            dpctl_functions.free_with_queue(a_usm, sycl_queue)
+            dpctl_functions.free_with_queue(b_usm, sycl_queue)
+            dpctl_functions.free_with_queue(out_usm, sycl_queue)
 
             dpnp_ext._dummy_liveness_func([a.size, b.size, out.size])
 
@@ -178,7 +176,7 @@ def dot_2_mm(a, b):
         dpnp_func = dpnp_ext.dpnp_func("dpnp_matmul", [a.dtype.name, "NONE"], sig)
 
         def dot_2_mv(a, b):
-            sycl_queue = get_sycl_queue()
+            sycl_queue = dpctl_functions.get_current_queue()
 
             m, k = a.shape
             (_n,) = b.shape
@@ -187,22 +185,28 @@ def dot_2_mv(a, b):
             if _n != k:
                 raise ValueError("Incompatible array sizes for np.dot(a, b)")
 
-            a_usm = allocate_usm_shared(a.size * a.itemsize, sycl_queue)
-            copy_usm(sycl_queue, a_usm, a.ctypes, a.size * a.itemsize)
+            a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, a_usm, a.ctypes, a.size * a.itemsize
+            )
 
-            b_usm = allocate_usm_shared(b.size * b.itemsize, sycl_queue)
-            copy_usm(sycl_queue, b_usm, b.ctypes, b.size * b.itemsize)
+            b_usm = dpctl_functions.malloc_shared(b.size * b.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, b_usm, b.ctypes, b.size * b.itemsize
+            )
 
             out = np.empty((m,), dtype=res_dtype)
-            out_usm = allocate_usm_shared(out.size * out.itemsize, sycl_queue)
+            out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize, sycl_queue)
 
             dpnp_func(a_usm, b_usm, out_usm, m, n, k)
 
-            copy_usm(sycl_queue, out.ctypes, out_usm, out.size * out.itemsize)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, out.ctypes, out_usm, out.size * out.itemsize
+            )
 
-            free_usm(a_usm, sycl_queue)
-            free_usm(b_usm, sycl_queue)
-            free_usm(out_usm, sycl_queue)
+            dpctl_functions.free_with_queue(a_usm, sycl_queue)
+            dpctl_functions.free_with_queue(b_usm, sycl_queue)
+            dpctl_functions.free_with_queue(out_usm, sycl_queue)
 
             dpnp_ext._dummy_liveness_func([a.size, b.size, out.size])
 
@@ -213,30 +217,36 @@ def dot_2_mv(a, b):
         dpnp_func = dpnp_ext.dpnp_func("dpnp_matmul", [a.dtype.name, "NONE"], sig)
 
         def dot_2_vm(a, b):
-            sycl_queue = get_sycl_queue()
+            sycl_queue = dpctl_functions.get_current_queue()
 
             (m,) = a.shape
             k, n = b.shape
 
             if m != k:
                 raise ValueError("Incompatible array sizes for np.dot(a, b)")
 
-            a_usm = allocate_usm_shared(a.size * a.itemsize, sycl_queue)
-            copy_usm(sycl_queue, a_usm, a.ctypes, a.size * a.itemsize)
+            a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, a_usm, a.ctypes, a.size * a.itemsize
+            )
 
-            b_usm = allocate_usm_shared(b.size * b.itemsize, sycl_queue)
-            copy_usm(sycl_queue, b_usm, b.ctypes, b.size * b.itemsize)
+            b_usm = dpctl_functions.malloc_shared(b.size * b.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, b_usm, b.ctypes, b.size * b.itemsize
+            )
 
             out = np.empty((n,), dtype=res_dtype)
-            out_usm = allocate_usm_shared(out.size * out.itemsize, sycl_queue)
+            out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize, sycl_queue)
 
             dpnp_func(a_usm, b_usm, out_usm, m, n, k)
 
-            copy_usm(sycl_queue, out.ctypes, out_usm, out.size * out.itemsize)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, out.ctypes, out_usm, out.size * out.itemsize
+            )
 
-            free_usm(a_usm, sycl_queue)
-            free_usm(b_usm, sycl_queue)
-            free_usm(out_usm, sycl_queue)
+            dpctl_functions.free_with_queue(a_usm, sycl_queue)
+            dpctl_functions.free_with_queue(b_usm, sycl_queue)
+            dpctl_functions.free_with_queue(out_usm, sycl_queue)
 
             dpnp_ext._dummy_liveness_func([a.size, b.size, out.size])
 
@@ -250,30 +260,36 @@ def dot_2_vm(a, b):
         dpnp_func = dpnp_ext.dpnp_func("dpnp_dot", [a.dtype.name, "NONE"], sig)
 
         def dot_2_vv(a, b):
-            sycl_queue = get_sycl_queue()
+            sycl_queue = dpctl_functions.get_current_queue()
 
             (m,) = a.shape
             (n,) = b.shape
 
             if m != n:
                 raise ValueError("Incompatible array sizes for np.dot(a, b)")
 
-            a_usm = allocate_usm_shared(a.size * a.itemsize, sycl_queue)
-            copy_usm(sycl_queue, a_usm, a.ctypes, a.size * a.itemsize)
+            a_usm = dpctl_functions.malloc_shared(a.size * a.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, a_usm, a.ctypes, a.size * a.itemsize
+            )
 
-            b_usm = allocate_usm_shared(b.size * b.itemsize, sycl_queue)
-            copy_usm(sycl_queue, b_usm, b.ctypes, b.size * b.itemsize)
+            b_usm = dpctl_functions.malloc_shared(b.size * b.itemsize, sycl_queue)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, b_usm, b.ctypes, b.size * b.itemsize
+            )
 
             out = np.empty(1, dtype=res_dtype)
-            out_usm = allocate_usm_shared(out.size * out.itemsize, sycl_queue)
+            out_usm = dpctl_functions.malloc_shared(out.size * out.itemsize, sycl_queue)
 
             dpnp_func(a_usm, b_usm, out_usm, m)
 
-            copy_usm(sycl_queue, out.ctypes, out_usm, out.size * out.itemsize)
+            dpctl_functions.queue_memcpy(
+                sycl_queue, out.ctypes, out_usm, out.size * out.itemsize
+            )
 
-            free_usm(a_usm, sycl_queue)
-            free_usm(b_usm, sycl_queue)
-            free_usm(out_usm, sycl_queue)
+            dpctl_functions.free_with_queue(a_usm, sycl_queue)
+            dpctl_functions.free_with_queue(b_usm, sycl_queue)
+            dpctl_functions.free_with_queue(out_usm, sycl_queue)
 
             dpnp_ext._dummy_liveness_func([a.size, b.size, out.size])