| { |
| "op": "ai.onnx.ReduceL1", |
| "fixtureArrays": { |
| "rank3_axis0_no_keepdims_input_x": [-1, 2, -3, 4, 5, -6, 0, -8, 0.5, -0.25, 1.5, -2.5, 7, -8, 9, -10, -0.75, 1.25, -1.75, 2.25, 3, -4, 5, -6], |
| "rank4_axis0_no_keepdims_input_x": [1, -2, 3, -4, 5, -6, 7, -8, 9, -10, 11, -12, -1.5, 2.5, -3.5, 4.5, -5.5, 6.5, -7.5, 8.5, -9.5, 10.5, -11.5, 12.5] |
| }, |
| "cases": [ |
| { |
| "name": "int32_axis0_splitk_8192x2", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "int32", "shape": [8192, 2], "data": { "kind": "cycle", "values": [1, -1, 2, -2] } } }, |
| "outputs": { "y": { "dtype": "int32", "shape": [2], "tolerance": 0 } } |
| }, |
| { |
| "name": "all_axes_flat_rank1_boundary_8192", |
| "provenance": { |
| "notes": "The parallel full-reduction threshold must supersede the rank1 serial and row-reduction fallbacks." |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [8192], "data": { "kind": "constant", "value": 1.0 } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [], "tolerance": 0 } } |
| }, |
| { |
| "name": "all_axes_flat_fullreduce_32x32x32_keepdims", |
| "attrs": { "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [32, 32, 32], |
| "data": { "kind": "fillFloat32", "sinStep": 0.013, "cosStep": 0.027, "scale": 0.5 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1], "tolerance": 0.05, "relTolerance": 0.0001 } } |
| }, |
| { |
| "name": "dispatch_cliff_noop_abs_16M", |
| "attrs": { "noop_with_empty_axes": 1 }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [16777216], "data": { "kind": "linspace", "start": -1.0, "end": 1.0 } } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [16777216], "tolerance": 0 } } |
| }, |
| { |
| "name": "axis0", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3], |
| "data": { "kind": "values", "values": [-1.0, 2.0, -3.0, 4.0, -5.0, 6.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "axis0_tiled_64x32", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [64, 32], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [32], "tolerance": 0.00002 } } |
| }, |
| { |
| "name": "f32_subnormal_axis0_tilecols_l1_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "Tiled axis-0 companion: L1 over 64 finite subnormal magnitudes should produce a finite subnormal column total." |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [64, 16], "data": { "kind": "constant", "value": -1e-40 } } }, |
| "outputs": { |
| "y": { "dtype": "float32", "shape": [16], "tolerance": 5e-44, "data": { "kind": "constant", "value": 6.4e-39 } } |
| } |
| }, |
| { |
| "name": "axis1", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3], |
| "data": { "kind": "values", "values": [-1.0, 2.0, -3.0, 4.0, -5.0, 6.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "f32_subnormal_axis1_l1_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "L1 reduction must preserve the sum of finite subnormal magnitudes instead of flushing each absolute value to zero." |
| }, |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3], |
| "data": { "kind": "values", "values": [1e-40, -1e-40, 0.0, -1e-40, -1e-40, -1e-40] } |
| } |
| }, |
| "outputs": { |
| "y": { |
| "dtype": "float32", |
| "shape": [2], |
| "tolerance": 2e-45, |
| "data": { "kind": "values", "values": [2e-40, 3e-40] } |
| } |
| } |
| }, |
| { |
| "name": "f32_subnormal_axis0_l1_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "Axis-0 companion: L1 over finite subnormal magnitudes must not take the zero-identity branch." |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2], |
| "data": { "kind": "values", "values": [1e-40, -1e-40, 0.0, -1e-40, -1e-40, 1e-40] } |
| } |
| }, |
| "outputs": { |
| "y": { |
| "dtype": "float32", |
| "shape": [2], |
| "tolerance": 2e-45, |
| "data": { "kind": "values", "values": [2e-40, 3e-40] } |
| } |
| } |
| }, |
| { |
| "name": "f32_subnormal_last_axis_vec4_l1_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "Vec4 last-axis companion: finite subnormal magnitudes should not collapse to a zero L1 total." |
| }, |
| "attrs": { "axes": [-1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 4], |
| "data": { "kind": "values", "values": [1e-40, -1e-40, 0.0, 1e-40, -1e-40, -1e-40, -1e-40, -1e-40] } |
| } |
| }, |
| "outputs": { |
| "y": { |
| "dtype": "float32", |
| "shape": [2], |
| "tolerance": 3e-45, |
| "data": { "kind": "values", "values": [3e-40, 4e-40] } |
| } |
| } |
| }, |
| { |
| "name": "f32_subnormal_last_axis_odd_l1_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "Odd-width last-axis companion: L1 over finite subnormal magnitudes should not collapse to zero in the non-vec4 subgroup reducer." |
| }, |
| "attrs": { "axes": [-1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3], |
| "data": { "kind": "values", "values": [1e-40, -1e-40, 0.0, -1e-40, -1e-40, -1e-40] } |
| } |
| }, |
| "outputs": { |
| "y": { |
| "dtype": "float32", |
| "shape": [2], |
| "tolerance": 2e-45, |
| "data": { "kind": "values", "values": [2e-40, 3e-40] } |
| } |
| } |
| }, |
| { |
| "name": "f32_subnormal_rank3_axis1_l1_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "Rank-3 axis-1 companion: L1 over finite subnormal magnitudes should keep finite subnormal totals." |
| }, |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3, 2], |
| "data": { |
| "kind": "values", |
| "values": [1e-40, -1e-40, -1e-40, 1e-40, 1e-40, -1e-40, -1e-40, -1e-40, 1e-40, 1e-40, -1e-40, 1e-40] |
| } |
| } |
| }, |
| "outputs": { |
| "y": { |
| "dtype": "float32", |
| "shape": [2, 2], |
| "tolerance": 2e-45, |
| "data": { "kind": "values", "values": [3e-40, 3e-40, 3e-40, 3e-40] } |
| } |
| } |
| }, |
| { |
| "name": "f32_subnormal_rank3_all_axes_l1_scalar_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "Rank-3 default-axes companion: L1 over finite subnormal magnitudes should preserve a finite scalar total." |
| }, |
| "attrs": { "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [2, 3, 2], "data": { "kind": "constant", "value": -1e-40 } } }, |
| "outputs": { |
| "y": { "dtype": "float32", "shape": [], "tolerance": 1e-44, "data": { "kind": "values", "values": [1.2e-39] } } |
| } |
| }, |
| { |
| "name": "f32_subnormal_rank3_all_axes_keepdims_l1_gpu_gap", |
| "skipGpu": { |
| "category": "permanent", |
| "reason": "Portable WGSL floating-point semantics do not guarantee preservation of the subnormal values required by this fixture. Backend evidence: WebGPU/Metal flushes subnormals to zero in f32; bit-exact subnormal preservation is unattainable on GPU." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1", |
| "notes": "Rank-3 default-axes keepdims companion: L1 over finite subnormal magnitudes should preserve a finite total in shape [1,1,1]." |
| }, |
| "attrs": { "keepdims": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [2, 3, 2], "data": { "kind": "constant", "value": -1e-40 } } }, |
| "outputs": { |
| "y": { |
| "dtype": "float32", |
| "shape": [1, 1, 1], |
| "tolerance": 1e-44, |
| "data": { "kind": "values", "values": [1.2e-39] } |
| } |
| } |
| }, |
| { |
| "name": "axis1_empty_cols_identity_zero", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [2, 0], "data": { "kind": "values", "values": [] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2], "tolerance": 0 } } |
| }, |
| { |
| "name": "axis0_empty_rows_identity_zero", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [0, 3], "data": { "kind": "values", "values": [] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3], "tolerance": 0 } } |
| }, |
| { |
| "name": "axis1_zero_rows_noop", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [0, 3], "data": { "kind": "values", "values": [] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [0], "tolerance": 0 } } |
| }, |
| { |
| "name": "axis1_keepdims", |
| "attrs": { "axes": [1], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3], |
| "data": { "kind": "values", "values": [-1.0, 2.0, -3.0, 4.0, -5.0, 6.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 1], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "rank3_axis0_no_keepdims", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3, 4], |
| "data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/rank3_axis0_no_keepdims_input_x" } } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 4], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "ort_empty_rank3_axis0_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.empty_set_ReduceL1_13", |
| "notes": "Same empty-set identity behavior adapted to axis 0." |
| }, |
| "attrs": { "axes": [0], "keepdims": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [0, 3, 4], "data": { "kind": "values", "values": [] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 3, 4], "tolerance": 0 } } |
| }, |
| { |
| "name": "rank4_axis1_channel_no_keepdims", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 4, 3, 2], |
| "data": { |
| "kind": "values", |
| "values": [1.0, -2.0, 3.0, -4.0, 5.0, -6.0, -1.0, 2.0, -3.0, 4.0, -5.0, 6.0, 0.5, -1.5, 2.5, -3.5, 4.5, -5.5, -0.5, 1.5, -2.5, 3.5, -4.5, 5.5, 6.0, -7.0, 8.0, -9.0, 10.0, -11.0, -6.0, 7.0, -8.0, 9.0, -10.0, 11.0, 1.25, -2.25, 3.25, -4.25, 5.25, -6.25, -1.25, 2.25, -3.25, 4.25, -5.25, 6.25] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 3, 2], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "rank1_axis0_keepdims", |
| "attrs": { "axes": [-1], "keepdims": 1 }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [5], "data": { "kind": "values", "values": [-1.25, 2.5, -3.75, 0.0, 4.5] } } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "ort_axis2_rank3_no_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_do_not_keep_dims" |
| }, |
| "attrs": { "axes": [2], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "ort_axis2_rank3_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_keepdims" |
| }, |
| "attrs": { "axes": [2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2, 1], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "ort_axis0_rank1_scalar", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_do_not_keep_dims_2" |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [3], "data": { "kind": "values", "values": [1.0, 2.0, 3.0] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "ort_axis0_all_negative", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_float_multi_element_all_negative" |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2], |
| "data": { "kind": "values", "values": [-1.0, -2.0, -3.0, -4.0, -5.0, -6.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "ort_axis0_singleton_negative_no_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_float_singleton_axis_negative_input" |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [1, 1], "data": { "kind": "values", "values": [-4.0] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_axis0_singleton_negative_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_int32_keepdims_singleton_axis_negative_input", |
| "notes": "Float32 projection of ORT's singleton keepdims absolute-value edge." |
| }, |
| "attrs": { "axes": [0], "keepdims": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [1, 2], "data": { "kind": "values", "values": [-3.0, -7.0] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 2], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_rank0_negative_scalar", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_0DTensor_negative_input" |
| }, |
| "inputs": { "x": { "dtype": "float32", "shape": [], "data": { "kind": "values", "values": [-3.0] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [], "tolerance": 0 } } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_do_not_keepdims_example", |
| "attrs": { "keepdims": 0, "axes": [2] }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2] } }, |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_example", |
| "notes": "The ONNX int64 axes input is materialized as this compile-time axes list." |
| } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_do_not_keepdims_random", |
| "attrs": { "keepdims": 0, "axes": [2] }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { |
| "kind": "values", |
| "values": [0.9762700796127319, 4.3037872314453125, 2.055267572402954, 0.8976636528968811, -1.5269039869308472, 2.917882204055786, -1.248255729675293, 7.835460186004639, 9.273255348205566, -2.331169605255127, 5.834500789642334, 0.577898383140564] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2] } }, |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_do_not_keepdims_random", |
| "notes": "The ONNX int64 axes input is materialized as this compile-time axes list." |
| } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_empty_set", |
| "attrs": { "keepdims": 1, "axes": [1] }, |
| "inputs": { "x": { "dtype": "float32", "shape": [2, 0, 4], "data": { "kind": "values", "values": [] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 1, 4] } }, |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_empty_set", |
| "notes": "The ONNX int64 axes input is materialized as this compile-time axes list." |
| } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_keep_dims_example", |
| "attrs": { "keepdims": 1, "axes": [2] }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2, 1] } }, |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_keep_dims_example", |
| "notes": "The ONNX int64 axes input is materialized as this compile-time axes list." |
| } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_keep_dims_random", |
| "attrs": { "keepdims": 1, "axes": [2] }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { |
| "kind": "values", |
| "values": [0.9762700796127319, 4.3037872314453125, 2.055267572402954, 0.8976636528968811, -1.5269039869308472, 2.917882204055786, -1.248255729675293, 7.835460186004639, 9.273255348205566, -2.331169605255127, 5.834500789642334, 0.577898383140564] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2, 1] } }, |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_keep_dims_random", |
| "notes": "The ONNX int64 axes input is materialized as this compile-time axes list." |
| } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_negative_axes_keep_dims_example", |
| "attrs": { "keepdims": 1, "axes": [-1] }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2, 1] } }, |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_example", |
| "notes": "The ONNX int64 axes input is materialized as this compile-time axes list." |
| } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_negative_axes_keep_dims_random", |
| "attrs": { "keepdims": 1, "axes": [-1] }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { |
| "kind": "values", |
| "values": [0.9762700796127319, 4.3037872314453125, 2.055267572402954, 0.8976636528968811, -1.5269039869308472, 2.917882204055786, -1.248255729675293, 7.835460186004639, 9.273255348205566, -2.331169605255127, 5.834500789642334, 0.577898383140564] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 2, 1] } }, |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_negative_axes_keep_dims_random", |
| "notes": "The ONNX int64 axes input is materialized as this compile-time axes list." |
| } |
| }, |
| { |
| "name": "default_axes_rank3_no_keepdims_scalar", |
| "attrs": { "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [-1.0, 2.0, -3.0, 4.0, -5.0, 6.0, -7.0, 8.0, -9.0, 10.0, -11.0, 12.0] } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "float32", "shape": [], "tolerance": 0, "data": { "kind": "values", "values": [78.0] } } |
| } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_default_axes_keepdims_example", |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_example" |
| }, |
| "attrs": { "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1] } } |
| }, |
| { |
| "name": "onnx_backend_reduce_l1_default_axes_keepdims_random", |
| "provenance": { |
| "source": "cmake/external/onnx/onnx/backend/test/data/node/test_reduce_l1_default_axes_keepdims_random" |
| }, |
| "attrs": { "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { |
| "kind": "values", |
| "values": [0.9762700796127319, 4.3037872314453125, 2.055267572402954, 0.8976636528968811, -1.5269039869308472, 2.917882204055786, -1.248255729675293, 7.835460186004639, 9.273255348205566, -2.331169605255127, 5.834500789642334, 0.577898383140564] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 1, 1] } } |
| }, |
| { |
| "name": "subgroup_vec4_last_axis_2x256", |
| "attrs": { "axes": [-1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 256], |
| "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.29 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2], "tolerance": 0.0002, "relTolerance": 0.0001 } } |
| }, |
| { |
| "name": "subgroup_scalar_last_axis_2x65", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 65], |
| "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.11 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2], "tolerance": 0.0002, "relTolerance": 0.0001 } } |
| }, |
| { |
| "name": "ort_noop_empty_axes_2d_elementwise_abs", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_NoopWithEmptyAxes_2D_ElementwiseAbs" |
| }, |
| "attrs": { "noop_with_empty_axes": 1 }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [2, 2], "data": { "kind": "values", "values": [-2.0, 0.0, 3.5, -4.0] } } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 2], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_noop_empty_axes_scalar_abs", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_NoopWithEmptyAxes_Scalar" |
| }, |
| "attrs": { "noop_with_empty_axes": 1 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [], "data": { "kind": "values", "values": [-3.0] } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_noop_empty_axes_3d_elementwise_abs", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_NoopWithEmptyAxes_3D_ElementwiseAbs" |
| }, |
| "attrs": { "noop_with_empty_axes": 1 }, |
| "inputs": { |
| "x": { "dtype": "float32", "shape": [1, 2, 2], "data": { "kind": "values", "values": [-2.0, 0.0, 3.5, -4.0] } } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 2, 2], "tolerance": 0 } } |
| }, |
| { |
| "name": "ort_int32_singleton_axis_negative", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_int32_singleton_axis_negative_input" |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "int32", "shape": [1, 1], "data": { "kind": "values", "values": [-4] } } }, |
| "outputs": { |
| "y": { "dtype": "int32", "shape": [1], "data": { "kind": "values", "values": [4] }, "tolerance": 0 } |
| } |
| }, |
| { |
| "name": "ort_int32_keepdims_singleton_axis_negative", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_int32_keepdims_singleton_axis_negative_input" |
| }, |
| "attrs": { "axes": [0], "keepdims": 1 }, |
| "inputs": { "x": { "dtype": "int32", "shape": [1, 2], "data": { "kind": "values", "values": [-3, -7] } } }, |
| "outputs": { |
| "y": { "dtype": "int32", "shape": [1, 2], "data": { "kind": "values", "values": [3, 7] }, "tolerance": 0 } |
| } |
| }, |
| { |
| "name": "ort_int32_multi_axis_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_int32" |
| }, |
| "attrs": { "axes": [0, 2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "int32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12] } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "int32", "shape": [1, 2, 1], "data": { "kind": "values", "values": [33, 45] }, "tolerance": 0 } |
| } |
| }, |
| { |
| "name": "ort_int32_abs_int_min_saturates_gpu_gap", |
| "skipGpu": { |
| "category": "todo", |
| "reason": "The current integer reduction route uses an i32 accumulator, so it cannot reproduce the fixture's widened intermediate arithmetic and final int32 saturation. A portable multiword accumulator can implement this behavior." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_int32_INT_MIN" |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "int32", "shape": [1], "data": { "kind": "values", "values": [-2147483648] } } }, |
| "outputs": { |
| "y": { "dtype": "int32", "shape": [], "data": { "kind": "values", "values": [2147483647] }, "tolerance": 0 } |
| } |
| }, |
| { |
| "name": "ort_int32_summation_overflow_saturates_gpu_gap", |
| "skipGpu": { |
| "category": "todo", |
| "reason": "The current integer reduction route uses an i32 accumulator, so it cannot reproduce the fixture's widened intermediate arithmetic and final int32 saturation. A portable multiword accumulator can implement this behavior." |
| }, |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_int32_summation_overflow" |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "int32", |
| "shape": [3], |
| "data": { "kind": "values", "values": [1000000000, 1000000000, 1000000000] } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "int32", "shape": [], "data": { "kind": "values", "values": [2147483647] }, "tolerance": 0 } |
| } |
| }, |
| { |
| "name": "ort_float_multi_axis_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1" |
| }, |
| "attrs": { "axes": [0, 2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 2, 2], |
| "data": { "kind": "values", "values": [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 2, 1], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "rank3_lastaxis_cols1024_tree_nosubgroup", |
| "attrs": { "axes": [2], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 2, 1024], |
| "data": { |
| "kind": "cycle", |
| "values": [1.0, -2.0, 0.5, 3.25, -1.5, 2.0, -0.75, 4.0, -3.5, 1.25, 0.0, -2.25, 5.0, -4.0, 2.75, -1.0] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 2], "tolerance": 0.00001 } } |
| }, |
| { |
| "name": "axis0_splitk_8192x32", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [8192, 32], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [32], "tolerance": 0.001 } } |
| }, |
| { |
| "name": "axis0_splitk_8192x48_keepdims", |
| "attrs": { "axes": [0], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [8192, 48], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 48], "tolerance": 0.001 } } |
| }, |
| { |
| "name": "rank4_axis0_no_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1 (rank4 single-axis projection)", |
| "notes": "Single-axis ReduceL1 over axis 0 of a rank-4 tensor. Valid ONNX float32 reduction; ORT computes the elementwise L1 over the leading axis." |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 2, 3, 2], |
| "data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/rank4_axis0_no_keepdims_input_x" } } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 3, 2], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "rank4_axis2_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_keepdims (rank4 middle-axis projection)", |
| "notes": "Single-axis ReduceL1 over middle axis 2 of a rank-4 NCHW-like tensor with keepdims. Valid ONNX float32; ORT computes per-element L1 over axis 2." |
| }, |
| "attrs": { "axes": [2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3, 2, 2], |
| "data": { |
| "kind": "values", |
| "values": [1.0, -2.0, 3.0, -4.0, 5.0, -6.0, 7.0, -8.0, 9.0, -10.0, 11.0, -12.0, -1.25, 2.25, -3.25, 4.25, -5.25, 6.25, -7.25, 8.25, -9.25, 10.25, -11.25, 12.25] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 3, 1, 2], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "rank4_multi_axes_0_2_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1 (rank4 multi-axis projection)", |
| "notes": "Multi-axis ReduceL1 over axes [0,2] of a rank-4 tensor with keepdims. Valid ONNX float32 multi-axis reduction; ORT computes L1 over the union of axes." |
| }, |
| "attrs": { "axes": [0, 2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 2, 2, 2], |
| "data": { |
| "kind": "values", |
| "values": [1.0, -2.0, 3.0, -4.0, 5.0, -6.0, 7.0, -8.0, 9.0, -10.0, 11.0, -12.0, -1.5, 2.5, -3.5, 4.5] |
| } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 2, 1, 2], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "rank3_multi_axes_1_2_keepdims", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1 (rank3 multi-axis [1,2] subset)", |
| "notes": "Exercises the hasAxis(1)+hasAxis(2) branch of multi_axis_rank3 (only axes=[0,2] is currently covered). Valid ONNX float32; WORKING path." |
| }, |
| "attrs": { "axes": [1, 2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 3, 4], |
| "data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/rank3_axis0_no_keepdims_input_x" } } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 1, 1], "tolerance": 0.000001 } } |
| }, |
| { |
| "name": "rank3_axis1_middle_4x128x256", |
| "provenance": { |
| "notes": "Compact sibling for the rank3 middle-axis benchmark; preserves a wide trailing dimension and axis=1 reduction so baseline tiers exercise the generic rank3_axis1 path." |
| }, |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [4, 128, 256], |
| "data": { "kind": "cycle", "values": [1.0, -2.0, 0.5, -0.25, 3.0, -4.0, 1.5, -0.75] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [4, 256], "tolerance": 0.0002, "relTolerance": 0.0002 } } |
| }, |
| { |
| "name": "rank4_lastaxis_scalar_2x2x2x3", |
| "provenance": { |
| "source": "onnxruntime/test/providers/cpu/reduction/reduction_ops_test.cc", |
| "test": "ReductionOpTest.ReduceL1_do_not_keep_dims (rank4 last-axis odd-width)", |
| "notes": "Rank4 last-axis (axis 3) reduction with cols=3 (%4!=0) selects the subgroup_last_axis scalar reducer. Valid ONNX float32; WORKING path, previously unbenched/untested at rank4." |
| }, |
| "attrs": { "axes": [-1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 2, 2, 3], |
| "data": { "kind": "values", "values": { "$ref": "#/fixtureArrays/rank4_axis0_no_keepdims_input_x" } } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 2, 2], "tolerance": 0.00001 } } |
| }, |
| { |
| "name": "lastaxis_unaligned_256x513_scalar", |
| "provenance": { |
| "notes": "Compact sibling for the unaligned last-axis ReduceL1 benchmark; cols=513 keeps the scalar last-axis reducer while the row count is large enough to exercise fallback scheduling." |
| }, |
| "attrs": { "axes": [-1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [256, 513], |
| "data": { "kind": "cycle", "values": [1.0, -2.0, 0.5, -0.25, 3.0, -4.0, 1.5, -0.75] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [256], "tolerance": 0.0002, "relTolerance": 0.0002 } } |
| }, |
| { |
| "name": "reduce_size1_axis_returns_abs_value", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 1, 4], |
| "data": { "kind": "values", "values": [-1.0, 2.0, -3.0, 4.0, -5.0, 6.0, -7.0, 8.0, 9.0, -10.0, 11.0, -12.0] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 4] } } |
| }, |
| { |
| "name": "int32_large_values_l1_no_overflow", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "int32", |
| "shape": [3, 1], |
| "data": { "kind": "values", "values": [-710000000, 710000000, -710000000] } |
| } |
| }, |
| "outputs": { |
| "y": { "dtype": "int32", "shape": [1], "data": { "kind": "values", "values": [2130000000] }, "tolerance": 0 } |
| } |
| }, |
| { |
| "name": "negative_axis_rank3_minus2", |
| "attrs": { "axes": [-2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 4, 3], |
| "data": { "kind": "fillFloat32", "sinStep": 0.19, "cosStep": 0.11, "scale": 1.5 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 1, 3] } } |
| }, |
| { |
| "name": "multi_axis_rank3_axes_0_1_keepdims", |
| "attrs": { "axes": [0, 1], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 4, 5], |
| "data": { "kind": "fillFloat32", "sinStep": 0.07, "cosStep": 0.13, "scale": 2.0 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [1, 1, 5] } } |
| }, |
| { |
| "name": "axis0_narrow_f32_8192x3_splitk_guard_lock", |
| "provenance": { |
| "notes": "Compact lock below the historical 16-column split-K guard. Negative ones validate absolute-value accumulation and an exact result." |
| }, |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float32", "shape": [8192, 3], "data": { "kind": "constant", "value": -1.0 } } }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3], "tolerance": 0 } } |
| }, |
| { |
| "name": "contiguous_suffix_axes12_parallel", |
| "provenance": { |
| "notes": "Contiguous axes {1,2} exercise the shared cooperative suffix reduction instead of one serial lane per output." |
| }, |
| "attrs": { "axes": [1, 2], "keepdims": 1 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [3, 16, 16], |
| "data": { "kind": "fillFloat32", "sinStep": 0.13, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [3, 1, 1], "tolerance": 0.00001 } } |
| }, |
| { |
| "name": "axis_split_rank3_axis1_2x8192x4", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 8192, 4], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 4], "tolerance": 0.001 } } |
| }, |
| { |
| "name": "f16_axis_split_tiled_narrow_2x8192x4", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [2, 8192, 4], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [2, 4], "tolerance": 0.05, "relTolerance": 0.01 } } |
| }, |
| { |
| "name": "axis_split_rank3_axis1_wide_2x8192x32", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float32", |
| "shape": [2, 8192, 32], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float32", "shape": [2, 32], "tolerance": 0.001 } } |
| }, |
| { |
| "name": "f16_axis_split_wide_2x8192x32", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [2, 8192, 32], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [2, 32], "tolerance": 0.05, "relTolerance": 0.01 } } |
| }, |
| { |
| "name": "f16_rank3_axis1_serial", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [4, 128, 256], |
| "data": { "kind": "cycle", "values": [1.0, -2.0, 0.5, -0.25, 3.0, -4.0, 1.5, -0.75] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [4, 256], "tolerance": 0.05, "relTolerance": 0.0002 } } |
| }, |
| { |
| "name": "f16_last_axis_serial_fallback", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [2, 65], |
| "data": { "kind": "fillFloat32", "sinStep": 0.23, "cosStep": 0.11 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [2], "tolerance": 0.05, "relTolerance": 0.0001 } } |
| }, |
| { |
| "name": "f16_all_axes", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { "x": { "dtype": "float16", "shape": [8192], "data": { "kind": "constant", "value": 1.0 } } }, |
| "outputs": { "y": { "dtype": "float16", "shape": [], "tolerance": 0.05 } } |
| }, |
| { |
| "name": "f16_axis0_splitk_8192x8", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [8192, 8], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [8], "tolerance": 0.05, "relTolerance": 0.002 } } |
| }, |
| { |
| "name": "f16_last_axis_vec4_8x1024", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [8, 1024], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [8], "tolerance": 0.05, "relTolerance": 0.002 } } |
| }, |
| { |
| "name": "f16_last_axis_scalar_8x1023", |
| "attrs": { "axes": [1], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [8, 1023], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [8], "tolerance": 0.05, "relTolerance": 0.002 } } |
| }, |
| { |
| "name": "f16_all_axes_flat_65543", |
| "attrs": { "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [65543], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [], "tolerance": 0.05, "relTolerance": 0.002 } } |
| }, |
| { |
| "name": "f16_suffix_vec4_4x8x128", |
| "attrs": { "axes": [1, 2], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [4, 8, 128], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [4], "tolerance": 0.05, "relTolerance": 0.002 } } |
| }, |
| { |
| "name": "f16_suffix_scalar_4x7x37", |
| "attrs": { "axes": [1, 2], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [4, 7, 37], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [4], "tolerance": 0.05, "relTolerance": 0.002 } } |
| }, |
| { |
| "name": "f16_axis0_tilecols_4096x64", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "float16", |
| "shape": [4096, 64], |
| "data": { "kind": "fillFloat32", "sinStep": 0.17, "cosStep": 0.07, "scale": 0.2 } |
| } |
| }, |
| "outputs": { "y": { "dtype": "float16", "shape": [64], "tolerance": 0.05, "relTolerance": 0.002 } } |
| }, |
| { |
| "name": "int32_axis0_tiled_64x32", |
| "attrs": { "axes": [0], "keepdims": 0 }, |
| "inputs": { |
| "x": { |
| "dtype": "int32", |
| "shape": [64, 32], |
| "data": { "kind": "cycle", "values": [-16777217, 3, 5, -16777219, 7, 11, -2] } |
| } |
| }, |
| "outputs": { "y": { "dtype": "int32", "shape": [32], "tolerance": 0 } } |
| } |
| ] |
| } |
|
|
