change shape for some ops to reduce variance

Test Plan: add later

Reviewed By: mingzhe09088

Differential Revision: D17175802

fbshipit-source-id: 41e6ef13e3e920d49bdfa0a7d2bda99e0501a9fc

Author: hl475

benchmarks/operator_benchmark/pt/cat_test.py

@@ -12,15 +12,15 @@
 
 # Configs for PT Cat operator
 cat_short_configs = op_bench.cross_product_configs(
-    M=[64, 128],
-    N=[64],
+    M=[256, 512],
+    N=[512],
     tags=['short']
 )
 
 
 class CatBenchmark(op_bench.TorchBenchmarkBase):
-    def init(self, M, N): 
-        self.input_one = torch.rand(M, N) 
+    def init(self, M, N):
+        self.input_one = torch.rand(M, N)
         self.set_module_name('cat')
 
     def forward(self):

benchmarks/operator_benchmark/pt/chunk_test.py

@@ -12,16 +12,16 @@
 
 # Configs for PT Chunk operator
 chunks_short_configs = op_bench.cross_product_configs(
-    M=[64, 128],
-    N=[64],
-    chunks=[3],
+    M=[256, 512],
+    N=[512],
+    chunks=[2],
     tags=['short']
 )
 
 
 class ChunkBenchmark(op_bench.TorchBenchmarkBase):
-    def init(self, M, N, chunks): 
-        self.input_one = torch.rand(M, N) 
+    def init(self, M, N, chunks):
+        self.input_one = torch.rand(M, N)
         self.chunks = chunks
         self.set_module_name('chunks')
 

benchmarks/operator_benchmark/pt/gather_test.py

@@ -13,8 +13,8 @@
 # An example input from this configuration is M=4, N=4, dim=0.
 configs = op_bench.config_list(
     attrs=[
-        [4, 4, 0],
-        [256, 256, 1],
+        [256, 512, 0],
+        [512, 512, 1],
     ],
     attr_names=["M", "N", "dim"],
     tags=["short"]

benchmarks/operator_benchmark/pt/split_test.py

@@ -12,16 +12,16 @@
 
 # Configs for PT Split operator
 split_short_configs = op_bench.cross_product_configs(
-    M=[64, 128],
-    N=[64],
-    parts=[3],
+    M=[256, 512],
+    N=[512],
+    parts=[2],
     tags=['short']
 )
 
 
 class SplitBenchmark(op_bench.TorchBenchmarkBase):
-    def init(self, M, N, parts): 
-        self.input_one = torch.rand(M, N) 
+    def init(self, M, N, parts):
+        self.input_one = torch.rand(M, N)
         self.split_size = int(M * N / parts)
         self.set_module_name('split')
 

benchmarks/operator_benchmark/pt/unary_test.py

@@ -114,7 +114,6 @@ def forward(self):
         ['digamma_', lambda t: t.digamma_()],
         ['erfinv_', lambda t: t.erfinv_()],
         ['exponential_', lambda t: t.exponential_()],
-        ['lgamma_', lambda t: t.lgamma_()],
         ['normal_', lambda t: t.normal_()],
         ['random_', lambda t: t.random_()],
         ['sign_', lambda t: t.sign_()],