Merge branch 'develop' into dot-add

test/gpu/manage_host_buffer.cpp

+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2022 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include <iostream>
+#include <vector>
+#include <hip/hip_runtime_api.h>
+#include <migraphx/gpu/target.hpp>
+#include <migraphx/verify.hpp>
+#include <test.hpp>
+#include <basic_ops.hpp>
+#include <migraphx/gpu/hip.hpp>
+#include <migraphx/make_op.hpp>
+
+#define MIGRAPHX_HIP_ASSERT(x) (EXPECT(x == hipSuccess))
+
+TEST_CASE(host_same_buffer_copy)
+{
+    migraphx::program p;
+    auto* mm = p.get_main_module();
+    migraphx::shape ss{migraphx::shape::float_type, {4, 2}};
+    auto a           = mm->add_parameter("a", ss);
+    auto b           = mm->add_parameter("b", ss);
+    auto aa          = mm->add_instruction(migraphx::make_op("add"), a, a);
+    auto gpu_out     = mm->add_instruction(migraphx::make_op("hip::copy_from_gpu"), aa);
+    auto stream_sync = mm->add_instruction(migraphx::make_op("hip::sync_stream"), gpu_out);
+    auto pass        = mm->add_instruction(unary_pass_op{}, stream_sync);
+    auto alloc       = mm->add_instruction(
+        migraphx::make_op("hip::allocate", {{"shape", migraphx::to_value(ss)}}));
+    auto gpu_in = mm->add_instruction(migraphx::make_op("hip::copy_to_gpu"), pass, alloc);
+    auto aab    = mm->add_instruction(migraphx::make_op("add"), gpu_in, b);
+    mm->add_return({aab});
+    migraphx::parameter_map pp;
+    std::vector<float> a_vec(ss.elements(), -1);
+    std::vector<float> b_vec(ss.elements(), 2);
+    std::vector<float> c_vec(ss.elements(), 0);
+    pp["a"] = migraphx::argument(ss, a_vec.data());
+    pp["b"] = migraphx::argument(ss, b_vec.data());
+    std::vector<float> gpu_result;
+    migraphx::target gpu_t = migraphx::gpu::target{};
+    migraphx::compile_options options;
+    options.offload_copy = true;
+    p.compile(gpu_t, options);
+    auto result = p.eval(pp).back();
+    std::vector<float> results_vector(ss.elements(), -1);
+    result.visit([&](auto output) { results_vector.assign(output.begin(), output.end()); });
+    EXPECT(migraphx::verify_range(c_vec, results_vector));
+}
+
+TEST_CASE(arguments_lifetime)
+{
+    auto use_on_gpu = [](const migraphx::argument& arg, int c) {
+        auto* arg_ptr = arg.data();
+        MIGRAPHX_HIP_ASSERT(hipSetDevice(0));
+        MIGRAPHX_HIP_ASSERT(hipMemset(arg_ptr, c, arg.get_shape().bytes()));
+        MIGRAPHX_HIP_ASSERT(hipDeviceSynchronize());
+        return;
+    };
+
+    auto f = [use_on_gpu](const migraphx::argument& input) {
+        auto a = migraphx::gpu::register_on_gpu(input);
+        auto s = a.get_shape();
+        {
+            auto b = migraphx::gpu::register_on_gpu(input);
+            use_on_gpu(b, 0);
+            std::vector<float> expected_b(s.elements(), 0);
+            auto gold = migraphx::argument(s, expected_b.data());
+        }
+        use_on_gpu(a, 1);
+        return true;
+    };
+
+    migraphx::shape ss{migraphx::shape::float_type, {4, 2}};
+    std::vector<float> x_data(ss.elements(), -1);
+    migraphx::argument x{ss, x_data.data()};
+    EXPECT(f(x));
+}
+
+int main(int argc, const char* argv[]) { test::run(argc, argv); }

test/gpu/mlir.cpp

@@ -145,7 +145,7 @@ TEST_CASE(conv)
     const std::string mlir_output = R"__migraphx__(
 module {
   func @main(%arg0: tensor<2x8x3x3xf32>, %arg1: tensor<1x8x4x4xf32>) -> tensor<1x2x2x2xf32> attributes {kernel = "mixr"} {
-    %0 = migraphx.convolution(%arg1, %arg0) {dilation = [1, 1], group = 1 : i64, padding = [0, 0, 0, 0], padding_mode = 0 : i64, stride = [1, 1]} : (tensor<1x8x4x4xf32>, tensor<2x8x3x3xf32>) -> tensor<1x2x2x2xf32>
+    %0 = migraphx.convolution(%arg1, %arg0) {dilation = [1, 1], group = 1 : i64, padding = [0, 0, 0, 0], padding_mode = 0 : i64, stride = [1, 1], use_dynamic_same_auto_pad = 0 : i64} : (tensor<1x8x4x4xf32>, tensor<2x8x3x3xf32>) -> tensor<1x2x2x2xf32>
     return %0 : tensor<1x2x2x2xf32>
   }
 }
@@ -168,7 +168,7 @@ TEST_CASE(conv_add_relu)
     const std::string mlir_output = R"__migraphx__(
 module {
   func @main(%arg0: tensor<1x2x2x2xf32>, %arg1: tensor<2x8x3x3xf32>, %arg2: tensor<1x8x4x4xf32>) -> tensor<1x2x2x2xf32> attributes {kernel = "mixr"} {
-    %0 = migraphx.convolution(%arg2, %arg1) {dilation = [1, 1], group = 1 : i64, padding = [0, 0, 0, 0], padding_mode = 0 : i64, stride = [1, 1]} : (tensor<1x8x4x4xf32>, tensor<2x8x3x3xf32>) -> tensor<1x2x2x2xf32>
+    %0 = migraphx.convolution(%arg2, %arg1) {dilation = [1, 1], group = 1 : i64, padding = [0, 0, 0, 0], padding_mode = 0 : i64, stride = [1, 1], use_dynamic_same_auto_pad = 0 : i64} : (tensor<1x8x4x4xf32>, tensor<2x8x3x3xf32>) -> tensor<1x2x2x2xf32>
     %1 = migraphx.add(%0, %arg0) : (tensor<1x2x2x2xf32>, tensor<1x2x2x2xf32>) -> tensor<1x2x2x2xf32>
     %2 = migraphx.relu(%1) : (tensor<1x2x2x2xf32>) -> tensor<1x2x2x2xf32>
     return %2 : tensor<1x2x2x2xf32>

test/onnx/gen_onnx.py

@@ -626,6 +626,46 @@ def constant_scalar_test():
     return ([node], [], [y])
 
 
+@onnx_test
+def constant_empty_scalar_int64_test():
+    x = np.array([]).astype(np.int64)
+    y = helper.make_tensor_value_info('0', TensorProto.INT64, [0])
+
+    node = onnx.helper.make_node(
+        'Constant',
+        inputs=[],
+        outputs=['0'],
+        value=onnx.helper.make_tensor(
+            name='one_element_tensor',
+            data_type=TensorProto.INT64,
+            dims=x.shape,
+            vals=x.flatten().astype(np.int64),
+        ),
+    )
+
+    return ([node], [], [y])
+
+
+@onnx_test
+def constant_one_val_int64_test():
+    x = np.array([1]).astype(np.int64)
+    y = helper.make_tensor_value_info('0', TensorProto.INT64, [0])
+
+    node = onnx.helper.make_node(
+        'Constant',
+        inputs=[],
+        outputs=['0'],
+        value=onnx.helper.make_tensor(
+            name='empty_tensor',
+            data_type=TensorProto.INT64,
+            dims=x.shape,
+            vals=x.flatten().astype(np.int64),
+        ),
+    )
+
+    return ([node], [], [y])
+
+
 @onnx_test
 def const_of_shape_empty_input_test():
     tensor_val = onnx.helper.make_tensor('value', onnx.TensorProto.INT64, [1],
@@ -851,6 +891,96 @@ def conv_bn_relu_maxpool_test():
     return ([node0, node1, node2, node3], [x, y, z, m, n, k, l], [out])
 
 
+@onnx_test
+def conv_dynamic_batch_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [None, 3, 5, 5])
+    y = helper.make_tensor_value_info('1', TensorProto.FLOAT, [1, 3, 3, 3])
+    out = helper.make_tensor_value_info('2', TensorProto.FLOAT,
+                                        [None, 1, 3, 3])
+
+    node = onnx.helper.make_node('Conv', inputs=['0', '1'], outputs=['2'])
+    return ([node], [x, y], [out])
+
+
+@onnx_test
+def conv_dynamic_img_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT,
+                                      [1, 3, None, None])
+    y = helper.make_tensor_value_info('1', TensorProto.FLOAT, [1, 3, 3, 3])
+    out = helper.make_tensor_value_info('2', TensorProto.FLOAT,
+                                        [1, 1, None, None])
+
+    node = onnx.helper.make_node('Conv', inputs=['0', '1'], outputs=['2'])
+    return ([node], [x, y], [out])
+
+
+@onnx_test
+def conv_dynamic_weights_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [1, 3, 5, 5])
+    y = helper.make_tensor_value_info('1', TensorProto.FLOAT,
+                                      [1, 3, None, None])
+    out = helper.make_tensor_value_info('2', TensorProto.FLOAT,
+                                        [1, 1, None, None])
+
+    node = onnx.helper.make_node('Conv', inputs=['0', '1'], outputs=['2'])
+    return ([node], [x, y], [out])
+
+
+@onnx_test
+def conv_dynamic_img_and_weights_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT,
+                                      [1, 3, None, None])
+    y = helper.make_tensor_value_info('1', TensorProto.FLOAT,
+                                      [1, 3, None, None])
+    out = helper.make_tensor_value_info('2', TensorProto.FLOAT,
+                                        [1, 1, None, None])
+
+    node = onnx.helper.make_node('Conv', inputs=['0', '1'], outputs=['2'])
+    return ([node], [x, y], [out])
+
+
+@onnx_test
+def conv_dynamic_batch_same_upper_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [None, 3, 5, 5])
+    y = helper.make_tensor_value_info('1', TensorProto.FLOAT, [1, 3, 3, 3])
+    out = helper.make_tensor_value_info('2', TensorProto.FLOAT, [1, 1, 5, 5])
+
+    node = onnx.helper.make_node('Conv',
+                                 inputs=['0', '1'],
+                                 outputs=['2'],
+                                 auto_pad='SAME_UPPER')
+    return ([node], [x, y], [out])
+
+
+@onnx_test
+def conv_dynamic_img_same_upper_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT,
+                                      [1, 3, None, None])
+    y = helper.make_tensor_value_info('1', TensorProto.FLOAT, [1, 3, 3, 3])
+    out = helper.make_tensor_value_info('2', TensorProto.FLOAT,
+                                        [1, 1, None, None])
+
+    node = onnx.helper.make_node('Conv',
+                                 inputs=['0', '1'],
+                                 outputs=['2'],
+                                 auto_pad='SAME_UPPER')
+    return ([node], [x, y], [out])
+
+
+@onnx_test
+def conv_dynamic_kernel_same_lower_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [1, 3, 5, 5])
+    y = helper.make_tensor_value_info('1', TensorProto.FLOAT,
+                                      [1, 3, None, None])
+    out = helper.make_tensor_value_info('2', TensorProto.FLOAT, [1, 1, 5, 5])
+
+    node = onnx.helper.make_node('Conv',
+                                 inputs=['0', '1'],
+                                 outputs=['2'],
+                                 auto_pad='SAME_LOWER')
+    return ([node], [x, y], [out])
+
+
 @onnx_test
 def conv_relu_maxpool_test():
     x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [1, 3, 32, 32])
@@ -2496,6 +2626,62 @@ def instance_norm_test():
     return ([node], [x, scale, bias], [y])
 
 
+@onnx_test
+def instance_norm_half_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT16, [1, 2, 3, 3])
+    scale = helper.make_tensor_value_info('1', TensorProto.FLOAT16, [2])
+    bias = helper.make_tensor_value_info('2', TensorProto.FLOAT16, [2])
+    y = helper.make_tensor_value_info('3', TensorProto.FLOAT16, [1, 2, 3, 3])
+
+    node = onnx.helper.make_node('InstanceNormalization',
+                                 inputs=['0', '1', '2'],
+                                 outputs=['3'])
+
+    return ([node], [x, scale, bias], [y])
+
+
+@onnx_test
+def instance_norm_type_mismatch_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [1, 2, 3, 3])
+    scale = helper.make_tensor_value_info('1', TensorProto.FLOAT16, [2])
+    bias = helper.make_tensor_value_info('2', TensorProto.FLOAT16, [2])
+    y = helper.make_tensor_value_info('3', TensorProto.FLOAT, [1, 2, 3, 3])
+
+    node = onnx.helper.make_node('InstanceNormalization',
+                                 inputs=['0', '1', '2'],
+                                 outputs=['3'])
+
+    return ([node], [x, scale, bias], [y])
+
+
+@onnx_test
+def instance_norm_invalid_type_test():
+    x = helper.make_tensor_value_info('0', TensorProto.INT32, [1, 2, 3, 3])
+    scale = helper.make_tensor_value_info('1', TensorProto.FLOAT, [2])
+    bias = helper.make_tensor_value_info('2', TensorProto.FLOAT, [2])
+    y = helper.make_tensor_value_info('3', TensorProto.FLOAT, [1, 2, 3, 3])
+
+    node = onnx.helper.make_node('InstanceNormalization',
+                                 inputs=['0', '1', '2'],
+                                 outputs=['3'])
+
+    return ([node], [x, scale, bias], [y])
+
+
+@onnx_test
+def instance_norm_nonbroadcastable_test():
+    x = helper.make_tensor_value_info('0', TensorProto.FLOAT, [1, 2, 3, 3])
+    scale = helper.make_tensor_value_info('1', TensorProto.FLOAT, [4])
+    bias = helper.make_tensor_value_info('2', TensorProto.FLOAT, [4])
+    y = helper.make_tensor_value_info('3', TensorProto.FLOAT, [1, 2, 3, 3])
+
+    node = onnx.helper.make_node('InstanceNormalization',
+                                 inputs=['0', '1', '2'],
+                                 outputs=['3'])
+
+    return ([node], [x, scale, bias], [y])
+
+
 @onnx_test
 def instance_norm_val_test():
     x = np.array([[[[0, 1, 2], [3, 4, 5], [6, 7, 8]],
@@ -3086,7 +3272,7 @@ def max_test():
     a = helper.make_tensor_value_info('0', TensorProto.FLOAT, [3])
     b = helper.make_tensor_value_info('1', TensorProto.FLOAT, [3])
     c = helper.make_tensor_value_info('2', TensorProto.FLOAT, [3])
-    y = helper.make_tensor_value_info('2', TensorProto.FLOAT, [3])
+    y = helper.make_tensor_value_info('3', TensorProto.FLOAT, [3])
 
     node = onnx.helper.make_node(
         'Max',
@@ -3220,7 +3406,7 @@ def min_test():
     a = helper.make_tensor_value_info('0', TensorProto.FLOAT, [3])
     b = helper.make_tensor_value_info('1', TensorProto.FLOAT, [3])
     c = helper.make_tensor_value_info('2', TensorProto.FLOAT, [3])
-    y = helper.make_tensor_value_info('2', TensorProto.FLOAT, [3])
+    y = helper.make_tensor_value_info('3', TensorProto.FLOAT, [3])
 
     node = onnx.helper.make_node(
         'Min',
@@ -3231,6 +3417,89 @@ def min_test():
     return ([node], [a, b, c], [y])
 
 
+@onnx_test
+def mod_test():
+    a = helper.make_tensor_value_info('0', TensorProto.INT32, [3, 3, 3])
+    b = helper.make_tensor_value_info('1', TensorProto.INT32, [3, 3, 3])
+    y = helper.make_tensor_value_info('2', TensorProto.INT32, [3, 3, 3])
+
+    node = onnx.helper.make_node('Mod', inputs=['0', '1'], outputs=['2'])
+
+    return ([node], [a, b], [y])
+
+
+@onnx_test
+def mod_test_half():
+    a = helper.make_tensor_value_info('0', TensorProto.FLOAT16, [3, 3, 3])
+    b = helper.make_tensor_value_info('1', TensorProto.FLOAT16, [3, 3, 3])
+    y = helper.make_tensor_value_info('2', TensorProto.FLOAT16, [3, 3, 3])
+
+    node = onnx.helper.make_node('Mod', inputs=['0', '1'], outputs=['2'])
+
+    return ([node], [a, b], [y])
+
+
+@onnx_test
+def mod_test_different_dtypes():
+    a = helper.make_tensor_value_info('0', TensorProto.INT16, [3, 3, 3])
+    b = helper.make_tensor_value_info('1', TensorProto.INT32, [3, 3, 3])
+    y = helper.make_tensor_value_info('2', TensorProto.INT32, [3, 3, 3])
+
+    node = onnx.helper.make_node(
+        'Mod',
+        inputs=['0', '1'],
+        outputs=['2'],
+    )
+
+    return ([node], [a, b], [y])
+
+
+@onnx_test
+def mod_test_fmod():
+    a = helper.make_tensor_value_info('0', TensorProto.FLOAT, [3, 3, 3])
+    b = helper.make_tensor_value_info('1', TensorProto.FLOAT, [3, 3, 3])
+    y = helper.make_tensor_value_info('2', TensorProto.FLOAT, [3, 3, 3])
+
+    node = onnx.helper.make_node(
+        'Mod',
+        inputs=['0', '1'],
+        outputs=['2'],
+        fmod=1  #fmod flag = 1
+    )
+
+    return ([node], [a, b], [y])
+
+
+@onnx_test
+def mod_test_fmod_half():
+    a = helper.make_tensor_value_info('0', TensorProto.FLOAT16, [3, 3, 3])
+    b = helper.make_tensor_value_info('1', TensorProto.FLOAT16, [3, 3, 3])
+    y = helper.make_tensor_value_info('2', TensorProto.FLOAT16, [3, 3, 3])
+
+    node = onnx.helper.make_node('Mod',
+                                 inputs=['0', '1'],
+                                 outputs=['2'],
+                                 fmod=1)
+
+    return ([node], [a, b], [y])
+
+
+@onnx_test
+def mod_test_fmod_different_dtypes():
+    a = helper.make_tensor_value_info('0', TensorProto.FLOAT, [3, 3, 3])
+    b = helper.make_tensor_value_info('1', TensorProto.INT32, [3, 3, 3])
+    y = helper.make_tensor_value_info('2', TensorProto.FLOAT, [3, 3, 3])
+
+    node = onnx.helper.make_node(
+        'Mod',
+        inputs=['0', '1'],
+        outputs=['2'],
+        fmod=1  #fmod flag = 1
+    )
+
+    return ([node], [a, b], [y])
+
+
 @onnx_test
 def multinomial_test():
     sample_size = 10

test/onnx/instance_norm_half_test.onnx

+instance_norm_half_test:

+#
+
0
+
1
+
2
3"InstanceNormalizationinstance_norm_half_testZ
+
0
+
+
+

+
+
+Z
+
1
+
+
+
+Z
+
2
+
+
+
+b
+
3
+
+
+

+
+
+B
\ No newline at end of file

test/onnx/instance_norm_invalid_type_test.onnx

+instance_norm_invalid_type_test:

+#
+
0
+
1
+
2
3"InstanceNormalizationinstance_norm_invalid_type_testZ
+
0
+
+

+
+
+Z
+
1
+
+

+Z
+
2
+
+

+b
+
3
+

+

+
+
+B
\ No newline at end of file

test/onnx/instance_norm_nonbroadcastable_test.onnx

+#instance_norm_nonbroadcastable_test:

+#
+
0
+
1
+
2
3"InstanceNormalization#instance_norm_nonbroadcastable_testZ
+
0
+

+

+
+
+Z
+
1
+
+

+Z
+
2
+
+

+b
+
3
+

+

+
+
+B
\ No newline at end of file

test/onnx/instance_norm_type_mismatch_test.onnx

+ instance_norm_type_mismatch_test:

+#
+
0
+
1
+
2
3"InstanceNormalization instance_norm_type_mismatch_testZ
+
0
+

+

+
+
+Z
+
1
+
+
+
+Z
+
2
+
+
+
+b
+
3
+

+

+
+
+B
\ No newline at end of file

test/onnx/max_test.onnx

-max-example:e
+max_test:a
 
 
0
 
1
-
2
3"Maxtest-dropoutZ
+
2
3"Maxmax_testZ
 
0
 
 

@@ -15,7 +15,7 @@
 
 

 b
-
2
+
3
 
 

-B
\ No newline at end of file
+B
\ No newline at end of file

test/onnx/min_test.onnx

-min-example:e
+min_test:a
 
 
0
 
1
-
2
3"Mintest-dropoutZ
+
2
3"Minmin_testZ
 
0
 
 

@@ -15,7 +15,7 @@
 
 

 b
-
2
+
3
 
 

-B
\ No newline at end of file
+B
\ No newline at end of file

test/onnx/mod_test.onnx

+mod_test:e
+
+
0
+
1
2"Modmod_testZ
+
0
+
+
+
+Z
+
1
+
+
+
+b
+
2
+
+
+
+B
\ No newline at end of file

test/onnx/mod_test_different_dtypes.onnx

+mod_test_different_dtypes:v
+
+
0
+
1
2"Modmod_test_different_dtypesZ
+
0
+
+
+
+Z
+
1
+
+
+
+b
+
2
+
+
+
+B
\ No newline at end of file