slice start==stop

Author: cjld

extern/cuda/nccl/src/nccl_warper.cc

@@ -27,9 +27,6 @@ nccl_initer() {
     int device_count = get_device_count();
     if (!device_count) return;
     if (!inside_mpi) return;
-    if (mpi_world_rank == 0)
-        checkCudaErrors(ncclGetUniqueId(&id));
-    MPI_CHECK(MPI_Bcast((void *)&id, sizeof(id), MPI_BYTE, 0, MPI_COMM_WORLD));
     if (mpi_local_rank >= device_count)
         LOGf << "mpi_local_rank(">>mpi_local_rank>>") is larger than device_count("
             >>device_count>>")";
@@ -39,6 +36,9 @@ nccl_initer() {
     event_queue.run_sync([]() {
         checkCudaErrors(cudaSetDevice(nccl_device_id));
     });
+    if (mpi_world_rank == 0)
+        checkCudaErrors(ncclGetUniqueId(&id));
+    MPI_CHECK(MPI_Bcast((void *)&id, sizeof(id), MPI_BYTE, 0, MPI_COMM_WORLD));
     checkCudaErrors(ncclCommInitRank(&comm, mpi_world_size, id, mpi_world_rank));
 }
 

python/jittor/__init__.py

@@ -7,7 +7,7 @@
 # This file is subject to the terms and conditions defined in
 # file 'LICENSE.txt', which is part of this source code package.
 # ***************************************************************
-__version__ = '1.2.0.2'
+__version__ = '1.2.0.3'
 from . import lock
 with lock.lock_scope():
     from . import compiler

python/jittor_utils/__init__.py

@@ -88,10 +88,10 @@ def in_ipynb():
 
 @contextlib.contextmanager
 def simple_timer(name):
-    LOG.i("Timer start", name)
+    print("Timer start", name)
     now = time.time()
     yield
-    LOG.i("Time stop", name, time.time()-now)
+    print("Time stop", name, time.time()-now)
 
 @contextlib.contextmanager
 def import_scope(flags):

src/misc/nano_vector.h

@@ -39,7 +39,7 @@ struct Slice {
         }
         if (start<0) start += size;
         mask = 0;
-        ASSERT(start>=0 && stop>=-1 && start<size && stop<=size)
+        ASSERT(start==step || (start>=0 && stop>=-1 && start<size && stop<=size))
             << "slice overflow:" << start << stop << step;
     }
 };