Review issues & conda recipe

AlexanderKalistratov · Jun 1, 2020 · May 18, 2020 · May 29, 2020 · May 29, 2020 · May 29, 2020
commit 7317e55241ec62015ef5db3dc260a68b5c5ac34f
diff --git a/conda-recipe/meta.yaml b/conda-recipe/meta.yaml
@@ -29,6 +29,8 @@ requirements:
     - pandas  {{ PANDAS_VERSION }}
     - pyarrow {{ PYARROW_VERSION }}
     - wheel
+    - tbb
+    - tbb-devel
 
   run:
     - python
@@ -37,6 +39,7 @@ requirements:
     - pyarrow {{ PYARROW_VERSION }}
     - numba   {{ NUMBA_VERSION }}
     - setuptools
+    - tbb
 
 test:
   imports:
@@ -58,6 +61,8 @@ outputs:
         - numpy
         - pandas  {{ PANDAS_VERSION }}
         - pyarrow {{ PYARROW_VERSION }}
+        - tbb
+        - tbb-devel
 
 about:
   home: https://github.com/IntelPython/sdc

diff --git a/requirements.txt b/requirements.txt
@@ -2,3 +2,5 @@ numpy>=1.16
 pandas==0.25.3
 pyarrow==0.17.0
 numba==0.49.1
+tbb
+tbb-devel
diff --git a/sdc/functions/sort.py b/sdc/functions/sort.py
@@ -7,15 +7,17 @@
 
 from sdc import concurrent_sort
 
+
 def bind(sym, sig):
     # Returns ctypes binding to symbol sym with signature sig
     addr = getattr(concurrent_sort, sym)
     return ct.cast(addr, sig)
 
+
 parallel_sort_arithm_sig = ct.CFUNCTYPE(None, ct.c_void_p, ct.c_uint64)
 
 parallel_sort_sig = ct.CFUNCTYPE(None, ct.c_void_p, ct.c_uint64,
-                               ct.c_uint64, ct.c_void_p,)
+                                 ct.c_uint64, ct.c_void_p,)
 
 parallel_sort_sym = bind('parallel_sort',
                          parallel_sort_sig)
@@ -30,7 +32,7 @@ def less(left, right):
     pass
 
 
-@overload(less, jit_options={'locals':{'result':types.int8}})
+@overload(less, jit_options={'locals': {'result': types.int8}})
 def less_overload(left, right):
     def less_impl(left, right):
         result = left < right
@@ -74,12 +76,15 @@ def deref(_ityctx, _x):
             # to deref the void * passed. TODO: nrt awareness
             catchthing = thing
             sig = catchthing(_x)
+
             def codegen(cgctx, builder, sig, args):
                 toty = cgctx.get_value_type(sig.return_type).as_pointer()
                 addressable = builder.bitcast(args[0], toty)
                 zero_intpt = cgctx.get_constant(types.intp, 0)
                 vref = builder.gep(addressable, [zero_intpt], inbounds=True)
+
                 return builder.load(vref)
+
             return sig, codegen
 
         @njit
@@ -103,11 +108,14 @@ def wrapper(a, b):
 @intrinsic
 def asvoidp(tyctx, thing):
     sig = types.voidptr(thing)
+
     def codegen(cgctx, builder, sig, args):
         dm_thing = cgctx.data_model_manager[sig.args[0]]
         data_thing = dm_thing.as_data(builder, args[0])
         ptr_thing = cgutils.alloca_once_value(builder, data_thing)
+
         return builder.bitcast(ptr_thing, cgutils.voidptr_t)
+
     return sig, codegen
 
 
@@ -116,22 +124,22 @@ def sizeof(context, t):
     sig = types.uint64(t)
 
     def codegen(cgctx, builder, sig, args):
-        size =  cgctx.get_abi_sizeof(t)
+        size = cgctx.get_abi_sizeof(t)
         return cgctx.get_constant(types.uint64, size)
 
     return sig, codegen
 
 
-types_to_postfix = {types.int8    : 'i8',
-                    types.uint8   : 'u8',
-                    types.int16   : 'i16',
-                    types.uint16  : 'u16',
-                    types.int32   : 'i32',
-                    types.uint32  : 'u32',
-                    types.int64   : 'i64',
-                    types.uint64  : 'u64',
-                    types.float32 : 'f32',
-                    types.float64 : 'f64' }
+types_to_postfix = {types.int8: 'i8',
+                    types.uint8: 'u8',
+                    types.int16: 'i16',
+                    types.uint16: 'u16',
+                    types.int32: 'i32',
+                    types.uint32: 'u32',
+                    types.int64: 'i64',
+                    types.uint64: 'u64',
+                    types.float32: 'f32',
+                    types.float64: 'f64'}
 
 
 def load_symbols(name, sig, types):
@@ -148,6 +156,7 @@ def load_symbols(name, sig, types):
 sort_map = load_symbols('parallel_sort', parallel_sort_arithm_sig, types_to_postfix)
 stable_sort_map = load_symbols('parallel_stable_sort', parallel_sort_arithm_sig, types_to_postfix)
 
+
 @intrinsic
 def list_itemsize(tyctx, list_ty):
     sig = types.uint64(list_ty)
@@ -209,6 +218,7 @@ def parallel_sort_impl(arr):
 
     return parallel_sort_impl
 
+
 def parallel_stable_sort(arr):
     pass
 

diff --git a/sdc/native/module.cpp b/sdc/native/module.cpp
@@ -77,4 +77,4 @@ PyMODINIT_FUNC PyInit_concurrent_sort()
     REGISTER(parallel_stable_sort_f64)
 #undef REGISTER
     return m;
-}
+}
diff --git a/sdc/native/sort.cpp b/sdc/native/sort.cpp
@@ -33,9 +33,6 @@ declare_int_sort(64)
 declare_sort(f32, float)
 declare_sort(f64, double)
 
-#undef declare_int_sort
-#undef declare_sort
-
 void parallel_sort(void* begin, uint64_t len, uint64_t size, void* compare)
 {
     auto compare_f = reinterpret_cast<compare_func>(compare);
@@ -64,15 +61,15 @@ void parallel_sort(void* begin, uint64_t len, uint64_t size, void* compare)
         break;
     default:
         // fallback to c qsort?
-        if      (size < 4)    run_sort(_void_range<4>)
-        else if (size < 8)    run_sort(_void_range<8>)
-        else if (size < 16)   run_sort(_void_range<16>)
-        else if (size < 32)   run_sort(_void_range<32>)
-        else if (size < 64)   run_sort(_void_range<64>)
-        else if (size < 128)  run_sort(_void_range<128>)
-        else if (size < 256)  run_sort(_void_range<256>)
-        else if (size < 512)  run_sort(_void_range<512>)
-        else if (size < 1024) run_sort(_void_range<1024>)
+        if      (size <= 4)    run_sort(_void_range<4>)
+        else if (size <= 8)    run_sort(_void_range<8>)
+        else if (size <= 16)   run_sort(_void_range<16>)
+        else if (size <= 32)   run_sort(_void_range<32>)
+        else if (size <= 64)   run_sort(_void_range<64>)
+        else if (size <= 128)  run_sort(_void_range<128>)
+        else if (size <= 256)  run_sort(_void_range<256>)
+        else if (size <= 512)  run_sort(_void_range<512>)
+        else if (size <= 1024) run_sort(_void_range<1024>)
         else
         {
             std::cout << "Unsupported item size " << size << std::endl;
@@ -84,4 +81,7 @@ void parallel_sort(void* begin, uint64_t len, uint64_t size, void* compare)
 #undef run_sort
 }
 
-}
+}
+
+#undef declare_int_sort
+#undef declare_sort
diff --git a/sdc/native/stable_sort.cpp b/sdc/native/stable_sort.cpp
@@ -227,15 +227,15 @@ void stable_sort_inner_sort<void>(void* data, int begin, int end, int item_size,
         break;
     default:
         // fallback to own implementation?
-        if      (item_size < 4)    run_sort(_void_range<4>)
-        else if (item_size < 8)    run_sort(_void_range<8>)
-        else if (item_size < 16)   run_sort(_void_range<16>)
-        else if (item_size < 32)   run_sort(_void_range<32>)
-        else if (item_size < 64)   run_sort(_void_range<64>)
-        else if (item_size < 128)  run_sort(_void_range<128>)
-        else if (item_size < 256)  run_sort(_void_range<256>)
-        else if (item_size < 512)  run_sort(_void_range<512>)
-        else if (item_size < 1024) run_sort(_void_range<1024>)
+        if      (item_size <= 4)    run_sort(_void_range<4>)
+        else if (item_size <= 8)    run_sort(_void_range<8>)
+        else if (item_size <= 16)   run_sort(_void_range<16>)
+        else if (item_size <= 32)   run_sort(_void_range<32>)
+        else if (item_size <= 64)   run_sort(_void_range<64>)
+        else if (item_size <= 128)  run_sort(_void_range<128>)
+        else if (item_size <= 256)  run_sort(_void_range<256>)
+        else if (item_size <= 512)  run_sort(_void_range<512>)
+        else if (item_size <= 1024) run_sort(_void_range<1024>)
         else
         {
             std::cout << "Unsupported item size " << item_size << std::endl;
@@ -322,12 +322,12 @@ declare_int_sort(64)
 declare_sort(f32, float)
 declare_sort(f64, double)
 
-#undef declare_int_sort
-#undef declare_sort
-
 void parallel_stable_sort(void* begin, uint64_t len, uint64_t size, void* compare)
 {
     parallel_stable_sort_<void>(begin, len, size, compare);
 }
 
-}
+}
+
+#undef declare_int_sort
+#undef declare_sort
diff --git a/setup.py b/setup.py
@@ -106,7 +106,8 @@ def check_file_at_path(path2file):
 if not tbb_root:
     tbb_root = check_file_at_path(['include', 'tbb', 'tbb.h'])
 
-print("tbb_root", tbb_root)
+if tbb_root is None:
+    raise RuntimeError("tbb headers were not found. Try to install tbb-devel package")
 
 ind = [PREFIX_DIR + '/include', ]
 lid = [PREFIX_DIR + '/lib', ]
@@ -375,7 +376,8 @@ def run(self):
           'numpy>=1.16',
           'pandas==0.25.3',
           'pyarrow==0.17.0',
-          'numba>=0.49.1,<0.50.0'
+          'numba>=0.49.1,<0.50.0',
+          'tbb'
           ],
       cmdclass=sdc_build_commands,
       ext_modules=_ext_mods,