From f12a2a513acda68146616368281fb0b63d15bb36 Mon Sep 17 00:00:00 2001
From: Glenn Jocher <glenn.jocher@ultralytics.com>
Date: Sun, 24 Nov 2019 18:29:29 -1000
Subject: [PATCH] updates

---
 test.py              | 2 +-
 train.py             | 2 +-
 utils/torch_utils.py | 8 +++++---
 3 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/test.py b/test.py
index c233adcf..c6656246 100644
--- a/test.py
+++ b/test.py
@@ -20,7 +20,7 @@ def test(cfg,
          model=None):
     # Initialize/load model and set device
     if model is None:
-        device = torch_utils.select_device(opt.device)
+        device = torch_utils.select_device(opt.device, batch_size=batch_size)
         verbose = True
 
         # Initialize model
diff --git a/train.py b/train.py
index 3a4bc34f..ecd8f4fd 100644
--- a/train.py
+++ b/train.py
@@ -423,7 +423,7 @@ if __name__ == '__main__':
     opt = parser.parse_args()
     opt.weights = last if opt.resume else opt.weights
     print(opt)
-    device = torch_utils.select_device(opt.device, apex=mixed_precision)
+    device = torch_utils.select_device(opt.device, apex=mixed_precision, batch_size=opt.batch_size)
     if device.type == 'cpu':
         mixed_precision = False
 
diff --git a/utils/torch_utils.py b/utils/torch_utils.py
index b631262e..466a22ae 100644
--- a/utils/torch_utils.py
+++ b/utils/torch_utils.py
@@ -14,7 +14,7 @@ def init_seeds(seed=0):
         torch.backends.cudnn.benchmark = False
 
 
-def select_device(device='', apex=False):
+def select_device(device='', apex=False, batch_size=None):
     # device = 'cpu' or '0' or '0,1,2,3'
     cpu_request = device.lower() == 'cpu'
     if device and not cpu_request:  # if device requested other than 'cpu'
@@ -25,11 +25,13 @@ def select_device(device='', apex=False):
     if cuda:
         c = 1024 ** 2  # bytes to MB
         ng = torch.cuda.device_count()
+        if ng > 1 and batch_size:  # check that batch_size is compatible with device_count
+            assert batch_size % ng == 0, 'batch-size %g not multiple of GPU count %g' % (batch_size, ng)
         x = [torch.cuda.get_device_properties(i) for i in range(ng)]
-        cuda_str = 'Using CUDA ' + ('Apex ' if apex else '')  # apex for mixed precision https://github.com/NVIDIA/apex
+        s = 'Using CUDA ' + ('Apex ' if apex else '')  # apex for mixed precision https://github.com/NVIDIA/apex
         for i in range(0, ng):
             if i == 1:
-                cuda_str = ' ' * len(cuda_str)
+                s = ' ' * len(s)
             print("%sdevice%g _CudaDeviceProperties(name='%s', total_memory=%dMB)" %
                   (cuda_str, i, x[i].name, x[i].total_memory / c))
     else: