/* * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under the BSD-style license found in the * LICENSE file in the root directory of this source tree. */ #include // Declares the operator #include #include #include #include #include #include #include using namespace ::testing; using exec_aten::ArrayRef; using exec_aten::ScalarType; using exec_aten::Tensor; using torch::executor::testing::TensorFactory; class OpBmmOutTest : public OperatorTest { protected: Tensor& op_bmm_out(const Tensor& self, const Tensor& mat2, Tensor& out) { return torch::executor::aten::bmm_outf(context_, self, mat2, out); } template void test_dtype() { TensorFactory tf; // Gives 4 * 2 * 3 = 24, shape (10, 3, 5) Tensor x = tf.full({10, 3, 4}, 2); Tensor y = tf.full({10, 4, 5}, 3); Tensor out = tf.zeros({10, 3, 5}); op_bmm_out(x, y, out); Tensor expected = tf.full({10, 3, 5}, 24); EXPECT_TENSOR_EQ(out, expected); } }; TEST_F(OpBmmOutTest, OutputDim) { TensorFactory tf; // Two tensors with compatible dimensions: (10, 3, 4) and (10, 4, 5). Tensor x = tf.ones({10, 3, 4}); Tensor y = tf.ones({10, 4, 5}); // Output shape should be (10, 3, 5) Tensor out = tf.zeros({10, 3, 5}); Tensor ret = op_bmm_out(x, y, out); // Should always return the provided out Tensor. EXPECT_TENSOR_EQ(ret, out); // Expected tensor, filled with 4. Tensor expected = tf.full({10, 3, 5}, 4); EXPECT_TENSOR_EQ(out, expected); } TEST_F(OpBmmOutTest, OutputDimFloat) { TensorFactory tf; // clang-format off Tensor x = tf.make( {2, 4, 5}, { 4., 3., 1., 1., 1., 3., 1., 4., 4., 2., 1., 1., 1., 3., 3., 4., 2., 2., 2., 3., 1., 3., 1., 4., 4., 1., 1., 2., 4., 3., 4., 3., 4., 1., 2., 1., 4., 4., 4., 4., }); // clang-format on // clang-format off Tensor y = tf.make( {2, 5, 3}, { 4., 4., 4., 2., 3., 1., 1., 4., 4., 3., 1., 2., 1., 4., 3., 1., 4., 4., 4., 4., 4., 2., 1., 4., 1., 4., 3., 1., 4., 4., }); // clang-format on // Output shape should be (10, 3, 5) Tensor out = tf.zeros({2, 4, 3}); Tensor ret = op_bmm_out(x, y, out); // Should always return the provided out Tensor. EXPECT_TENSOR_EQ(ret, out); // clang-format off Tensor expected = tf.make( {2, 4, 3}, { 27., 34., 28., 32., 43., 43., 19., 26., 24., 31., 44., 39., 23., 49., 48., 16., 38., 40., 27., 44., 55., 33., 56., 64., }); // clang-format on EXPECT_TENSOR_EQ(out, expected); } /// A generic smoke test that works for any dtype that supports ones() and /// zeros(). TEST_F(OpBmmOutTest, AllDtypesSupported) { #define TEST_ENTRY(ctype, dtype) test_dtype(); ET_FORALL_REAL_TYPES(TEST_ENTRY); #undef TEST_ENTRY // TODO: Also add tests for half, complex, quantized, and other types. Easiest // way to do that would be to make TensorFactory support zeros() and ones() // for those types. } TEST_F(OpBmmOutTest, EmptyInputWithEmptyOutTensorPasses) { TensorFactory tf; Tensor x = tf.full({2, 2, 2}, 3); Tensor y = tf.make({2, 2, 0}, {}); // Make an empty out tensor and demonstrate that it's empty. Tensor out = tf.make({2, 2, 0}, {}); EXPECT_EQ(out.numel(), 0); op_bmm_out(x, y, out); EXPECT_EQ(out.numel(), 0); } TEST_F(OpBmmOutTest, MismatchedDimensionsDies) { TensorFactory tf; Tensor x = tf.ones({2, 10, 3}); // wrong_y has incompatible shape Tensor wrong_y = tf.ones({3, 7, 4}); Tensor right_y = tf.ones({2, 3, 4}); Tensor out = tf.ones({2, 10, 4}); ET_EXPECT_KERNEL_FAILURE(context_, op_bmm_out(x, wrong_y, out)); EXPECT_TENSOR_EQ(op_bmm_out(x, right_y, out), tf.full({2, 10, 4}, 3)); } TEST_F(OpBmmOutTest, MismatchedDimensionSizeDies) { if (torch::executor::testing::SupportedFeatures::get()->is_aten) { GTEST_SKIP() << "ATen kernel can handle mismatched dimension size"; } TensorFactory tf; Tensor x = tf.ones({2, 10, 3}); Tensor y = tf.ones({2, 3, 4}); // wrong_y has incompatible dim Tensor wrong_y = tf.ones({7, 4}); Tensor right_y = tf.ones({2, 3, 4}); // wrong_out has incompatible dim Tensor right_out = tf.ones({2, 10, 4}); Tensor wrong_out = tf.ones({7, 5}); ET_EXPECT_KERNEL_FAILURE(context_, op_bmm_out(x, right_y, wrong_out)); ET_EXPECT_KERNEL_FAILURE(context_, op_bmm_out(x, wrong_y, right_out)); } TEST_F(OpBmmOutTest, WrongOutShapeDies) { if (torch::executor::testing::SupportedFeatures::get()->is_aten) { GTEST_SKIP() << "ATen kernel can handle wrong out shape"; } TensorFactory tf; Tensor x = tf.ones({2, 10, 3}); Tensor y = tf.ones({2, 3, 4}); // wrong_out has incompatible shape Tensor right_out = tf.ones({2, 10, 4}); Tensor wrong_out = tf.ones({3, 7, 5}); ET_EXPECT_KERNEL_FAILURE(context_, op_bmm_out(x, y, wrong_out)); EXPECT_TENSOR_EQ(op_bmm_out(x, y, right_out), tf.full({2, 10, 4}, 3)); } TEST_F(OpBmmOutTest, DynamicShapeUpperBoundSameAsExpected) { TensorFactory tf; auto x = tf.make( {3, 3, 6}, {0.7231091856956482, 0.7423362731933594, 0.5262957811355591, 0.24365824460983276, 0.584592342376709, 0.033152639865875244, 0.13871687650680542, 0.242235004901886, 0.815468966960907, 0.793160617351532, 0.2782524824142456, 0.48195880651474, 0.8197803497314453, 0.9970665574073792, 0.6984410881996155, 0.5675464272499084, 0.8352431654930115, 0.2055988311767578, 0.593172013759613, 0.11234724521636963, 0.1534569263458252, 0.24170821905136108, 0.7262365221977234, 0.7010802030563354, 0.2038237452507019, 0.6510535478591919, 0.7744860053062439, 0.4368913173675537, 0.5190907716751099, 0.6158523559570312, 0.8101882934570312, 0.9800970554351807, 0.1146882176399231, 0.3167651295661926, 0.6965049505233765, 0.9142746925354004, 0.9351036548614502, 0.9411783814430237, 0.5995072722434998, 0.06520867347717285, 0.5459962487220764, 0.18719732761383057, 0.03402292728424072, 0.944246232509613, 0.8801798820495605, 0.0012360215187072754, 0.5935860276222229, 0.4157699942588806, 0.41771942377090454, 0.2711215615272522, 0.6922780871391296, 0.2038482427597046, 0.6832956671714783, 0.75285404920578}); auto y = tf.make( {3, 6, 2}, {0.8579357862472534, 0.6869555711746216, 0.0051323771476745605, 0.17565155029296875, 0.7496575117111206, 0.6046506762504578, 0.1099579930305481, 0.21209025382995605, 0.9703746438026428, 0.8369089365005493, 0.28198742866516113, 0.3741576075553894, 0.023700952529907227, 0.49101293087005615, 0.12347054481506348, 0.11432164907455444, 0.4724501967430115, 0.5750725269317627, 0.2952348589897156, 0.7966887950897217, 0.19573044776916504, 0.9536850452423096, 0.8426499366760254, 0.07835853099822998, 0.3755578398704529, 0.5225613117218018, 0.572950541973114, 0.6185871362686157, 0.6962141394615173, 0.5299500823020935, 0.25603562593460083, 0.7365944981575012, 0.020375549793243408, 0.20364665985107422, 0.3748350739479065, 0.2564433217048645}); Tensor expected_result = tf.make( {3, 3, 2}, {1.6221470832824707, 1.498693823814392, 1.224705696105957, 1.2123372554779053, 2.1629090309143066, 2.05692195892334, 0.9047035574913025, 1.3324503898620605, 1.2006582021713257, 1.5112680196762085, 1.1946606636047363, 1.5640640258789062, 1.405808448791504, 1.5957869291305542, 1.3348338603973389, 1.2967426776885986, 1.1425018310546875, 1.2352378368377686}); Tensor out = tf.zeros({3, 3, 2}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND); Tensor ret = op_bmm_out(x, y, out); EXPECT_TENSOR_CLOSE(out, expected_result); } TEST_F(OpBmmOutTest, DynamicShapeUpperBoundLargerThanExpected) { TensorFactory tf; auto x = tf.make( {3, 3, 6}, {0.7231091856956482, 0.7423362731933594, 0.5262957811355591, 0.24365824460983276, 0.584592342376709, 0.033152639865875244, 0.13871687650680542, 0.242235004901886, 0.815468966960907, 0.793160617351532, 0.2782524824142456, 0.48195880651474, 0.8197803497314453, 0.9970665574073792, 0.6984410881996155, 0.5675464272499084, 0.8352431654930115, 0.2055988311767578, 0.593172013759613, 0.11234724521636963, 0.1534569263458252, 0.24170821905136108, 0.7262365221977234, 0.7010802030563354, 0.2038237452507019, 0.6510535478591919, 0.7744860053062439, 0.4368913173675537, 0.5190907716751099, 0.6158523559570312, 0.8101882934570312, 0.9800970554351807, 0.1146882176399231, 0.3167651295661926, 0.6965049505233765, 0.9142746925354004, 0.9351036548614502, 0.9411783814430237, 0.5995072722434998, 0.06520867347717285, 0.5459962487220764, 0.18719732761383057, 0.03402292728424072, 0.944246232509613, 0.8801798820495605, 0.0012360215187072754, 0.5935860276222229, 0.4157699942588806, 0.41771942377090454, 0.2711215615272522, 0.6922780871391296, 0.2038482427597046, 0.6832956671714783, 0.75285404920578}); auto y = tf.make( {3, 6, 2}, {0.8579357862472534, 0.6869555711746216, 0.0051323771476745605, 0.17565155029296875, 0.7496575117111206, 0.6046506762504578, 0.1099579930305481, 0.21209025382995605, 0.9703746438026428, 0.8369089365005493, 0.28198742866516113, 0.3741576075553894, 0.023700952529907227, 0.49101293087005615, 0.12347054481506348, 0.11432164907455444, 0.4724501967430115, 0.5750725269317627, 0.2952348589897156, 0.7966887950897217, 0.19573044776916504, 0.9536850452423096, 0.8426499366760254, 0.07835853099822998, 0.3755578398704529, 0.5225613117218018, 0.572950541973114, 0.6185871362686157, 0.6962141394615173, 0.5299500823020935, 0.25603562593460083, 0.7365944981575012, 0.020375549793243408, 0.20364665985107422, 0.3748350739479065, 0.2564433217048645}); Tensor expected_result = tf.make( {3, 3, 2}, {1.6221470832824707, 1.498693823814392, 1.224705696105957, 1.2123372554779053, 2.1629090309143066, 2.05692195892334, 0.9047035574913025, 1.3324503898620605, 1.2006582021713257, 1.5112680196762085, 1.1946606636047363, 1.5640640258789062, 1.405808448791504, 1.5957869291305542, 1.3348338603973389, 1.2967426776885986, 1.1425018310546875, 1.2352378368377686}); Tensor out = tf.zeros({6, 6, 4}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND); Tensor ret = op_bmm_out(x, y, out); EXPECT_TENSOR_CLOSE(out, expected_result); } TEST_F(OpBmmOutTest, DynamicShapeUnbound) { GTEST_SKIP() << "Dynamic shape unbound not supported"; TensorFactory tf; auto x = tf.make( {3, 3, 6}, {0.7231091856956482, 0.7423362731933594, 0.5262957811355591, 0.24365824460983276, 0.584592342376709, 0.033152639865875244, 0.13871687650680542, 0.242235004901886, 0.815468966960907, 0.793160617351532, 0.2782524824142456, 0.48195880651474, 0.8197803497314453, 0.9970665574073792, 0.6984410881996155, 0.5675464272499084, 0.8352431654930115, 0.2055988311767578, 0.593172013759613, 0.11234724521636963, 0.1534569263458252, 0.24170821905136108, 0.7262365221977234, 0.7010802030563354, 0.2038237452507019, 0.6510535478591919, 0.7744860053062439, 0.4368913173675537, 0.5190907716751099, 0.6158523559570312, 0.8101882934570312, 0.9800970554351807, 0.1146882176399231, 0.3167651295661926, 0.6965049505233765, 0.9142746925354004, 0.9351036548614502, 0.9411783814430237, 0.5995072722434998, 0.06520867347717285, 0.5459962487220764, 0.18719732761383057, 0.03402292728424072, 0.944246232509613, 0.8801798820495605, 0.0012360215187072754, 0.5935860276222229, 0.4157699942588806, 0.41771942377090454, 0.2711215615272522, 0.6922780871391296, 0.2038482427597046, 0.6832956671714783, 0.75285404920578}); auto y = tf.make( {3, 6, 2}, {0.8579357862472534, 0.6869555711746216, 0.0051323771476745605, 0.17565155029296875, 0.7496575117111206, 0.6046506762504578, 0.1099579930305481, 0.21209025382995605, 0.9703746438026428, 0.8369089365005493, 0.28198742866516113, 0.3741576075553894, 0.023700952529907227, 0.49101293087005615, 0.12347054481506348, 0.11432164907455444, 0.4724501967430115, 0.5750725269317627, 0.2952348589897156, 0.7966887950897217, 0.19573044776916504, 0.9536850452423096, 0.8426499366760254, 0.07835853099822998, 0.3755578398704529, 0.5225613117218018, 0.572950541973114, 0.6185871362686157, 0.6962141394615173, 0.5299500823020935, 0.25603562593460083, 0.7365944981575012, 0.020375549793243408, 0.20364665985107422, 0.3748350739479065, 0.2564433217048645}); Tensor expected_result = tf.make( {3, 3, 2}, {1.6221470832824707, 1.498693823814392, 1.224705696105957, 1.2123372554779053, 2.1629090309143066, 2.05692195892334, 0.9047035574913025, 1.3324503898620605, 1.2006582021713257, 1.5112680196762085, 1.1946606636047363, 1.5640640258789062, 1.405808448791504, 1.5957869291305542, 1.3348338603973389, 1.2967426776885986, 1.1425018310546875, 1.2352378368377686}); Tensor out = tf.zeros( {1, 1, 1}, torch::executor::TensorShapeDynamism::DYNAMIC_UNBOUND); Tensor ret = op_bmm_out(x, y, out); EXPECT_TENSOR_CLOSE(out, expected_result); }