apeforest commented on a change in pull request #17456: Implement remaining 
nn_basic ops in opperf
URL: https://github.com/apache/incubator-mxnet/pull/17456#discussion_r373260284
 
 

 ##########
 File path: benchmark/opperf/utils/benchmark_utils.py
 ##########
 @@ -143,19 +145,35 @@ def run_performance_test(ops, inputs, run_backward=True,
 
 
 def run_op_benchmarks(ops, dtype, ctx, profiler, warmup, runs):
+    # Running SoftmaxOutput backwards on GPU results in errors
+    # track issue here: https://github.com/apache/incubator-mxnet/issues/880
+    gpu_backwards_disabled_ops = ['SoftmaxOutput']
+
+    # Running im2col either forwards or backwards on GPU results in errors
+    # track issue here: https://github.com/apache/incubator-mxnet/issues/17493
+    gpu_disabled_ops = ['im2col']
+
     # For each operator, run benchmarks
     mx_op_benchmark_results = []
     for op, op_params in ops.items():
-        # Prepare inputs for the operator
-        inputs = prepare_op_inputs(op, op_params)
-        # Run benchmarks
-        cur_op_res = run_performance_test(op_params["nd_op_handle"],
-                                          
run_backward=op_params["has_backward"],
-                                          dtype=dtype, ctx=ctx,
-                                          profiler=profiler,
-                                          inputs=inputs,
-                                          warmup=warmup, runs=runs)
-        mx_op_benchmark_results += cur_op_res
+        if not (ctx == mx.gpu() and op in gpu_disabled_ops):
 
 Review comment:
   Can we change the logic here to `ctx == mx.cpu() or op not in 
gpu_disabled_ops` for readability?

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to