[TOPI] Specify non-zero absolute tolerance in tests (apache#1925)

Author: sergei-mironov

apps/extension/tests/test_ext.py

@@ -22,7 +22,7 @@ def check_llvm():
         a = tvm.nd.array(np.random.uniform(size=n).astype(A.dtype), ctx)
         b = tvm.nd.array(np.zeros(n, dtype=B.dtype), ctx)
         f(a, b)
-        np.testing.assert_allclose(b.asnumpy(), a.asnumpy() + 1)
+        tvm.testing.assert_allclose(b.asnumpy(), a.asnumpy() + 1)
     check_llvm()
 
 

docs/deploy/aocl_fpga.md

@@ -52,7 +52,7 @@ b = tvm.nd.array(np.random.uniform(size=n).astype("float32"), ctx)
 c = tvm.nd.array(np.zeros(n, dtype="float32"), ctx)
 
 fadd(a, b, c)
-np.testing.assert_allclose(c.asnumpy(), a.asnumpy() + b.asnumpy())
+tvm.testing.assert_allclose(c.asnumpy(), a.asnumpy() + b.asnumpy())
 ```
 
 Setup

docs/deploy/aws_fpga.md

@@ -55,7 +55,7 @@ b = tvm.nd.array(np.random.uniform(size=n).astype("float32"), ctx)
 c = tvm.nd.array(np.zeros(n, dtype="float32"), ctx)
 
 fadd(a, b, c)
-np.testing.assert_allclose(c.asnumpy(), a.asnumpy() + b.asnumpy())
+tvm.testing.assert_allclose(c.asnumpy(), a.asnumpy() + b.asnumpy())
 ```
 
 Setup

nnvm/python/nnvm/testing/check_computation.py

@@ -281,10 +281,10 @@ def check_function(symbol, forward=None, backward=None, grad_input_vars=None,
         Additional parameters for `check_numerical_grads`.
 
     atol : float, optional
-        Absolute tolerance for `np.testing.assert_allclose`. NOT used for numerical gradients.
+        Absolute tolerance for `tvm.testing.assert_allclose`. NOT used for numerical gradients.
 
     rtol : float, optional
-        Relative tolerance for `np.testing.assert_allclose`. NOT used for numerical gradients.
+        Relative tolerance for `tvm.testing.assert_allclose`. NOT used for numerical gradients.
 
     quiet : bool, optional
         Don't dump additional information to stdout on failure.
@@ -466,7 +466,7 @@ def check_function(symbol, forward=None, backward=None, grad_input_vars=None,
                                      .format(len(numpy_res), out_len))
 
                 for i in range(out_len):
-                    np.testing.assert_allclose(nnvm_res[i], numpy_res[i], atol=atol, rtol=rtol)
+                    tvm.testing.assert_allclose(nnvm_res[i], numpy_res[i], atol=atol, rtol=rtol)
 
             if backward is not None:
                 nothing_was_done = False
@@ -495,8 +495,8 @@ def check_function(symbol, forward=None, backward=None, grad_input_vars=None,
                                          .format(set(grad_var_names) - set(numpy_grads)))
 
                 for x_name in numpy_grads:
-                    np.testing.assert_allclose(nnvm_grads[x_name], numpy_grads[x_name],
-                                               atol=atol, rtol=rtol)
+                    tvm.testing.assert_allclose(nnvm_grads[x_name], numpy_grads[x_name],
+                                                atol=atol, rtol=rtol)
 
             if numerical_grads:
                 nothing_was_done = False

nnvm/tests/python/compiler/test_build.py

@@ -27,7 +27,7 @@ def verify(graph, lib):
         # get outputs
         out = tvm.nd.empty(shape, dtype)
         get_output(0, out)
-        np.testing.assert_allclose(
+        tvm.testing.assert_allclose(
             out.asnumpy(), np.exp(na.asnumpy() + nb.asnumpy()))
 
     graph, lib, _ = nnvm.compiler.build(z, "llvm", shape_dict)
@@ -49,7 +49,7 @@ def test_run():
     nx = tvm.nd.array(np.random.uniform(size=shape).astype(dtype))
     ny = tvm.nd.array(np.random.uniform(size=shape).astype(dtype))
     res = _run_graph(z, {"x": nx, "y": ny})
-    np.testing.assert_allclose(
+    tvm.testing.assert_allclose(
         res[0].asnumpy(), np.exp(nx.asnumpy() + ny.asnumpy()))
 
 
@@ -73,7 +73,7 @@ def test_precompute_prune():
     m["load_params"](nnvm.compiler.save_param_dict(params))
     m.run()
     out = m.get_output(0, out=res)
-    np.testing.assert_allclose(
+    tvm.testing.assert_allclose(
         res.asnumpy(), nx.asnumpy() + 1 + ny.asnumpy() + na.asnumpy())
 
 
@@ -92,7 +92,7 @@ def test_dtypes():
         m.run(x=data)
         data = (data > 0) * data
         out = m.get_output(0, tvm.nd.empty(oshape, dtype))
-        np.testing.assert_allclose(out.asnumpy(), data, atol=1e-5, rtol=1e-5)
+        tvm.testing.assert_allclose(out.asnumpy(), data, atol=1e-5, rtol=1e-5)
 
 def test_ndarray_output():
     x = sym.Variable("x")
@@ -110,7 +110,7 @@ def test_ndarray_output():
     m.set_input("y", ny)
     m.run()
     out = m.get_output(0)
-    np.testing.assert_allclose(
+    tvm.testing.assert_allclose(
         out.asnumpy(), nx.asnumpy() + ny.asnumpy())
 
 def test_ndarray_input():
@@ -131,12 +131,12 @@ def test_ndarray_input():
     in_y = tvm.nd.empty(shape, dtype)
     m.get_input("x", in_x)
     m.get_input("y", in_y)
-    np.testing.assert_allclose(nx.asnumpy(), in_x.asnumpy())
-    np.testing.assert_allclose(ny.asnumpy(), in_y.asnumpy())
+    tvm.testing.assert_allclose(nx.asnumpy(), in_x.asnumpy())
+    tvm.testing.assert_allclose(ny.asnumpy(), in_y.asnumpy())
     in_nx = m.get_input("x")
     in_ny = m.get_input("y")
-    np.testing.assert_allclose(nx.asnumpy(), in_nx.asnumpy())
-    np.testing.assert_allclose(ny.asnumpy(), in_ny.asnumpy())
+    tvm.testing.assert_allclose(nx.asnumpy(), in_nx.asnumpy())
+    tvm.testing.assert_allclose(ny.asnumpy(), in_ny.asnumpy())
 
 def test_num_outputs():
     x = sym.Variable('x')

nnvm/tests/python/compiler/test_compiler_cache.py

@@ -19,7 +19,7 @@ def verify(graph, lib):
         m.run(x=na, y=nb)
         # get outputs
         out = m.get_output(0, tvm.nd.empty(shape, dtype))
-        np.testing.assert_allclose(
+        tvm.testing.assert_allclose(
             out.asnumpy(), np.exp(na.asnumpy() + nb.asnumpy()))
 
     engine = nnvm.compiler.engine

nnvm/tests/python/compiler/test_fold_axis.py

@@ -1,4 +1,5 @@
 """Unittest cases for fold_axis"""
+import tvm
 import nnvm
 import nnvm.testing.resnet
 import numpy as np
@@ -147,7 +148,7 @@ def run_prune(graph, params, opt_level):
 
     x = run_prune(graph, params, 0)
     y = run_prune(graph, params, 3)
-    np.testing.assert_allclose(y[0].asnumpy(), x[0].asnumpy())
+    tvm.testing.assert_allclose(y[0].asnumpy(), x[0].asnumpy())
 
 
 if __name__ == "__main__":

nnvm/tests/python/compiler/test_nhwc_layout.py

@@ -50,7 +50,7 @@ def test_nhwc():
     oshape_nhwc = (1, 224, 224, out_channel)
     nchw_output = build_and_run(nchw_sym, nchw_params, data, oshape)
     nhwc_output = build_and_run(nhwc_sym, nhwc_params, data.transpose(0, 2, 3, 1), oshape_nhwc)
-    np.testing.assert_allclose(nchw_output, nhwc_output.transpose(0, 3, 1, 2), rtol=1e-5, atol=1e-5)
+    tvm.testing.assert_allclose(nchw_output, nhwc_output.transpose(0, 3, 1, 2), rtol=1e-5, atol=1e-5)
 
 
 if __name__ == "__main__":

nnvm/tests/python/compiler/test_op_fusion.py

@@ -22,7 +22,7 @@ def test_ewise_injective():
         x_np = np.random.uniform(size=dshape).astype(dtype)
         m.run(x=x_np)
         out = m.get_output(0, tvm.nd.empty((10, 6)))
-        np.testing.assert_allclose(
+        tvm.testing.assert_allclose(
             out.asnumpy(),  x_np.reshape(out.shape) * 2 + 1,
             atol=1e-5, rtol=1e-5)
 
@@ -54,7 +54,7 @@ def test_conv_ewise_injective():
             data.asnumpy(), kernel.asnumpy(), (1,1), 'SAME')
         c_np = c_np + bias.asnumpy().reshape(kshape[0], 1, 1) + 1
         c_np = c_np.reshape(c_np.shape[0], np.prod(c_np.shape[1:])) + 1
-        np.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
+        tvm.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
 
 
 def test_injective_reduce_injective():
@@ -74,7 +74,7 @@ def test_injective_reduce_injective():
         c_np = np.sum(data.reshape(32, 18 * 18) + 1, axis=1)
         # get output
         out = m.get_output(0, tvm.nd.empty(c_np.shape, dtype))
-        np.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
+        tvm.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
 
 
 def test_injective_conv2d():
@@ -107,7 +107,7 @@ def test_injective_conv2d():
             data.asnumpy(), kernel.asnumpy(), (1,1), 'SAME')
         weight = np.mean(data.asnumpy(), axis=(2, 3))
         c_np = weight[:, :, np.newaxis, np.newaxis] * data.asnumpy() + residual
-        np.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
+        tvm.testing.assert_allclose(out.asnumpy(), c_np, rtol=1e-5)
 
 
 def test_concatenate_conv2d():
@@ -140,7 +140,7 @@ def test_concatenate_conv2d():
         conv = topi.testing.conv2d_nchw_python(
             concat, kernel.asnumpy(), (1,1), 'SAME')
         ref = concat + conv
-        np.testing.assert_allclose(out.asnumpy(), ref, rtol=1e-5)
+        tvm.testing.assert_allclose(out.asnumpy(), ref, rtol=1e-5)
 
 
 def test_residual_block_layout_transform():
@@ -178,7 +178,7 @@ def test_residual_block_layout_transform():
     conv2 = topi.testing.conv2d_nchw_python(
         conv1, kernel2.asnumpy(), (1,1), 'SAME')
     ref = np.maximum(conv1 + conv2, 0)
-    np.testing.assert_allclose(out.asnumpy(), ref, rtol=1e-5)
+    tvm.testing.assert_allclose(out.asnumpy(), ref, rtol=1e-5)
 
 
 def build_and_run(sym, params, data, out_shape, target, ctx, opt_level=2):
@@ -218,7 +218,7 @@ def get_sym(out_channel):
         _, params2 = utils.create_workload(sym2, 1, dshape[1:], seed=0)
         output1, g1 = build_and_run(sym1, params1, data, oshape, target, ctx, opt_level=2)
         output2, g2 = build_and_run(sym2, params2, data, oshape, target, ctx, opt_level=0)
-        np.testing.assert_allclose(output1, output2, rtol=1e-5, atol=1e-5)
+        tvm.testing.assert_allclose(output1, output2, rtol=1e-5, atol=1e-5)
         # data, conv weight, bias, batch norm gamma, batch norm beta, conv op
         assert g1.index.num_nodes == 6
 

nnvm/tests/python/compiler/test_optimizer.py

@@ -27,7 +27,7 @@ def helper(symbol, inputs, params, update_func, run_times, target, ctx, dtype="f
         m.run()
     y_np = update_func(**np_inputs)
     out = m.get_output(0, tvm.nd.empty(y_np.shape, dtype))
-    np.testing.assert_allclose(out.asnumpy(), y_np, atol=1e-5, rtol=1e-5)
+    tvm.testing.assert_allclose(out.asnumpy(), y_np, atol=1e-5, rtol=1e-5)
 
 
 def test_sgd():