| { |
| "op": "ai.onnx.IsNaN", |
| "cases": [ |
| { |
| "name": "f32_nan_inf_finite", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 4], |
| "data": { "kind": "values", "values": [1.0, "NaN", "Infinity", "-Infinity", 0.0, "NaN", -5.5, 3.25] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [2, 4] } } |
| }, |
| { |
| "name": "f16_nan_values", |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [2, 3], |
| "data": { "kind": "values", "values": ["NaN", 0.0, 1.0, -2.0, "NaN", "Infinity"] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [2, 3] } } |
| }, |
| { |
| "name": "scalar_false", |
| "inputs": { "x": { "dtype": "float32", "shape": [], "data": { "kind": "values", "values": [42.0] } } }, |
| "outputs": { "y": { "dtype": "bool", "shape": [] } } |
| }, |
| { |
| "name": "scalar_true", |
| "inputs": { "x": { "dtype": "float32", "shape": [], "data": { "kind": "values", "values": ["NaN"] } } }, |
| "outputs": { "y": { "dtype": "bool", "shape": [], "tolerance": 0 } } |
| }, |
| { |
| "name": "zero_size_noop", |
| "inputs": { "x": { "dtype": "float16", "shape": [0, 3], "data": { "kind": "values", "values": [] } } }, |
| "outputs": { "y": { "dtype": "bool", "shape": [0, 3], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_float_opset20_2x2", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/tensor/isnan_test.cc", |
| "test": "IsNaNOpTest.IsNaNFloat20" |
| }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [2, 2], "data": { "kind": "values", "values": [1.0, "NaN", 2.0, "NaN"] } } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [2, 2], "tolerance": 0 } } |
| }, |
| { |
| "name": "onnx_backend_float_specials", |
| "provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_isnan", "test": "test_isnan" }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [6], |
| "data": { "kind": "values", "values": [-1.2, "NaN", "Infinity", 2.8, "-Infinity", "Infinity"] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [6], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_float16_opset20_2x2", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/tensor/isnan_test.cc", |
| "test": "IsNaNOpTest.IsNaNFloat16_20" |
| }, |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [2, 2], "data": { "kind": "values", "values": [1.0, "NaN", 2.0, "NaN"] } } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [2, 2], "tolerance": 0 } } |
| }, |
| { |
| "name": "onnx_backend_float16_specials", |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_isnan_float16", |
| "test": "test_isnan_float16" |
| }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [6], |
| "data": { "kind": "values", "values": [-1.2, "NaN", "Infinity", 2.8, "-Infinity", "Infinity"] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [6], "tolerance": 0 } } |
| }, |
| { |
| "name": "onnx_backend_isnan", |
| "provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_isnan" }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [6], |
| "data": { |
| "kind": "values", |
| "values": [-1.2000000476837158, "NaN", "Infinity", 2.799999952316284, "-Infinity", "Infinity"] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [6], "tolerance": 0 } } |
| }, |
| { |
| "name": "onnx_backend_isnan_float16", |
| "provenance": { "source": "cmake/external/onnx/onnx/backend/test/data/node/test_isnan_float16" }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [6], |
| "data": { |
| "kind": "values", |
| "values": [-1.2001953125, "NaN", "Infinity", 2.80078125, "-Infinity", "Infinity"] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [6], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_bool_exact_empty_float16_zero_dim", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/tensor/isnan_test.cc", |
| "test": "IsNaNOpTest.IsNaNFloat16_20", |
| "notes": "Zero-dimension projection of ORT's float16 IsNaN coverage." |
| }, |
| "inputs": { "x": { "dtype": "float16", "shape": [0, 2], "data": { "kind": "values", "values": [] } } }, |
| "outputs": { "y": { "dtype": "bool", "shape": [0, 2], "tolerance": 0 } } |
| }, |
| { |
| "name": "f32_vec4_mixed_nan_16", |
| "provenance": { |
| "notes": "Compact aligned f32 vec4 sibling for the 1M IsNaN benchmark; keeps mixed NaN/Inf/finite lanes on a same-layout vectorized path." |
| }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [16], |
| "data": { |
| "kind": "values", |
| "values": ["NaN", 0.0, "Infinity", "-Infinity", 1.5, "NaN", -2.0, 3.0, 4.0, 5.0, "NaN", 6.0, "-Infinity", 7.0, 8.0, "NaN"] |
| } |
| } |
| }, |
| "outputs": { |
| "y": { |
| "dtype": "bool", |
| "shape": [16], |
| "tolerance": 0, |
| "data": { "kind": "values", "values": [1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1] } |
| } |
| } |
| }, |
| { |
| "name": "f16_all_nan_vec4", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [4], "data": { "kind": "values", "values": ["NaN", "NaN", "NaN", "NaN"] } } |
| }, |
| "outputs": { |
| "y": { "dtype": "bool", "shape": [4], "tolerance": 0, "data": { "kind": "values", "values": [1, 1, 1, 1] } } |
| } |
| }, |
| { |
| "name": "f16_mixed_nan_non_nan_vec4", |
| "inputs": { |
| "x": { "dtype": "float16", "shape": [4], "data": { "kind": "values", "values": ["NaN", 1.0, -1.0, "NaN"] } } |
| }, |
| "outputs": { |
| "y": { "dtype": "bool", "shape": [4], "tolerance": 0, "data": { "kind": "values", "values": [1, 0, 0, 1] } } |
| } |
| }, |
| { |
| "name": "rank7_vec4", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [1, 1, 1, 1, 1, 1, 8], |
| "data": { "kind": "values", "values": ["NaN", 1.0, "Infinity", "-Infinity", 0.0, "NaN", 2.0, "NaN"] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [1, 1, 1, 1, 1, 1, 8], "tolerance": 0 } }, |
| "provenance": { |
| "notes": "Rank-7 flat input on the shared vec4 unary kernel; rank is unused (sibling ops impose no rank cap). Covers the regime the removed 'ranks.x <= 6' guard rejected." |
| } |
| }, |
| { |
| "name": "rank7_scalar", |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [1, 1, 1, 1, 1, 1, 3], |
| "data": { "kind": "values", "values": ["NaN", 1.0, "NaN"] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "bool", "shape": [1, 1, 1, 1, 1, 1, 3], "tolerance": 0 } }, |
| "provenance": { "notes": "Rank-7 input, numel 3 routes to the scalar elementwise unary variant." } |
| } |
| ] |
| } |
|
|
