Change dynamic fusion API to return destination tensor info
The new dynamic fusion API is introduced in the following patch:
https://review.mlplatform.org/c/ml/ComputeLibrary/+/8906
For each operator (except Conv2D, which is migrated in the above patch), we
- remove destination tensor from is_supported, validate and create calls
- make create_op return ITensorInfo* to the intermediate destination object
Affected operators:
- DepthwiseConv2D
- Cast
- Elementwise Ops
- Clamp
- Reshape
- Resize
Resolves: COMPMID-5777
Change-Id: Ib60ec8a5f081752808455d7a7d790f2ed0627059
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8991
Reviewed-by: Ramy Elgammal <ramy.elgammal@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Dynamic-Fusion: Ramy Elgammal <ramy.elgammal@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/tests/validation/dynamic_fusion/gpu/cl/Add.cpp b/tests/validation/dynamic_fusion/gpu/cl/Add.cpp
index 1451ab3..0385407 100644
--- a/tests/validation/dynamic_fusion/gpu/cl/Add.cpp
+++ b/tests/validation/dynamic_fusion/gpu/cl/Add.cpp
@@ -48,7 +48,7 @@
// *INDENT-OFF*
// clang-format off
-DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(
framework::dataset::make("Input1Info", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32), // Invalid data type combination
TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16), // S16 is valid data type for Add
@@ -71,41 +71,7 @@
TensorInfo(TensorShape( 3U, 8U, 1U), 1, DataType::S16),
TensorInfo(TensorShape(32U, 13U, 2U, 2), 1, DataType::F32),
})),
- framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S32),
- TensorInfo(TensorShape(48U, 11U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(15U, 23U, 3U), 1, DataType::F32),
- TensorInfo(TensorShape( 3U, 8U, 9U), 1, DataType::S16),
- TensorInfo(TensorShape(32U, 13U, 2U, 2), 1, DataType::F32),
- })),
framework::dataset::make("Expected", { true, false, true, true, false, true, true, false, false, true})),
- input1_info, input2_info, output_info, expected)
-{
- // Create a new workload sketch
- auto cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
- auto gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
- GpuWorkloadSketch sketch{ &gpu_ctx };
-
- // Fuse Elementwise Add
- auto lhs_info = sketch.create_tensor_info(input1_info);
- auto rhs_info = sketch.create_tensor_info(input2_info);
- auto dst_info = sketch.create_tensor_info(output_info);
- bool res = bool(GpuAdd::validate_op(sketch, &lhs_info, &rhs_info, &dst_info));
- ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
-}
-
-DATA_TEST_CASE(ValidateRhsInplace, framework::DatasetMode::ALL, zip(zip(
- framework::dataset::make("Input1Info", { TensorInfo(TensorShape(32U, 1U, 1U), 1, DataType::F32), // Broadcasting allowed for lhs
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- }),
- framework::dataset::make("Input2Info",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(32U, 1U, 1U), 1, DataType::F32), // Broadcasting not allowed for rhs
- })),
- framework::dataset::make("Expected", { true, false})),
input1_info, input2_info, expected)
{
// Create a new workload sketch
@@ -116,29 +82,8 @@
// Fuse Elementwise Add
auto lhs_info = sketch.create_tensor_info(input1_info);
auto rhs_info = sketch.create_tensor_info(input2_info);
- bool res = bool(GpuAdd::validate_op(sketch, &lhs_info, &rhs_info, &rhs_info));
- ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
-}
-DATA_TEST_CASE(ValidateLhsInplace, framework::DatasetMode::ALL, zip(zip(
- framework::dataset::make("Input1Info", { TensorInfo(TensorShape(32U, 1U, 1U), 1, DataType::F32), // Broadcasting not allowed for lhs
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- }),
- framework::dataset::make("Input2Info",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(32U, 1U, 1U), 1, DataType::F32), // Broadcasting allowed for rhs
- })),
- framework::dataset::make("Expected", { false, true})),
- input1_info, input2_info, expected)
-{
- // Create a new workload sketch
- auto cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
- auto gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
- GpuWorkloadSketch sketch{ &gpu_ctx };
-
- // Fuse Elementwise Add
- auto lhs_info = sketch.create_tensor_info(input1_info);
- auto rhs_info = sketch.create_tensor_info(input2_info);
- bool res = bool(GpuAdd::validate_op(sketch, &lhs_info, &rhs_info, &lhs_info));
+ bool res = bool(GpuAdd::validate_op(sketch, &lhs_info, &rhs_info));
ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
}
// clang-format on
diff --git a/tests/validation/dynamic_fusion/gpu/cl/Clamp.cpp b/tests/validation/dynamic_fusion/gpu/cl/Clamp.cpp
index 947201f..177c02c 100644
--- a/tests/validation/dynamic_fusion/gpu/cl/Clamp.cpp
+++ b/tests/validation/dynamic_fusion/gpu/cl/Clamp.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -51,33 +51,21 @@
TEST_SUITE(CLAMP)
// *INDENT-OFF*
// clang-format off
-DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
framework::dataset::make("InputInfo", { TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32),
TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
- TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32), // Mismatching data types
- TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32), // Mismatching shapes
TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32), // Minimum value larger than maximum value
}),
- framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
- TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F16),
- TensorInfo(TensorShape(30U, 11U, 2U), 1, DataType::F32),
- TensorInfo(TensorShape(30U, 11U, 2U), 1, DataType::F32),
- })),
framework::dataset::make("MinVal", { 0.2f,
1.5f,
- 0.1f,
- 3.0f,
9.0f,
})),
framework::dataset::make("MaxVal", { 0.5f,
2.0f,
1.0f,
- 4.0f,
- 1.0f,
})),
- framework::dataset::make("Expected", { true, true, false, false, false })),
- input_info, output_info, min_val, max_val, expected)
+ framework::dataset::make("Expected", { true, true, false })),
+ input_info, min_val, max_val, expected)
{
// Create a new workload sketch
CLCompileContext cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
@@ -86,13 +74,12 @@
// Fuse Clamp
const TensorInfo src_info = sketch.create_tensor_info(input_info);
- const TensorInfo dst_info = sketch.create_tensor_info(output_info);
ClampAttributes attributes {};
attributes.min_val(min_val)
.max_val(max_val);
- const bool res = static_cast<bool>(GpuClamp::validate_op(sketch, &src_info, &dst_info, attributes));
+ const bool res = static_cast<bool>(GpuClamp::validate_op(sketch, &src_info, attributes));
ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
}
// clang-format on
diff --git a/tests/validation/dynamic_fusion/gpu/cl/DepthwiseConv2d.cpp b/tests/validation/dynamic_fusion/gpu/cl/DepthwiseConv2d.cpp
index f08cc60..b6331d7 100644
--- a/tests/validation/dynamic_fusion/gpu/cl/DepthwiseConv2d.cpp
+++ b/tests/validation/dynamic_fusion/gpu/cl/DepthwiseConv2d.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -53,31 +53,29 @@
// *INDENT-OFF*
// clang-format off
-DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zip(zip(
- framework::dataset::make("InputInfo", { TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Mismatching data type input/weights
- TensorInfo(TensorShape(3U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Mismatching input feature maps
- TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Mismatching depth multiplier
- TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Invalid biases size
- TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Invalid biases dimensions
- TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Invalid output size
- TensorInfo(TensorShape(8U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // patch size bigger than input width
- TensorInfo(TensorShape(8U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // dilation < 1
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(zip(zip( // Explanations of failing tests
+ framework::dataset::make("InputInfo", { TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Mismatching data type input/weights
+ TensorInfo(TensorShape(3U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Mismatching input feature maps
+ TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Mismatching depth multiplier
+ TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Invalid biases size
+ TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // Invalid biases dimensions
+ TensorInfo(TensorShape(8U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC), // dilation < 1
TensorInfo(TensorShape(8U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QASYMM8, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QASYMM8_SIGNED, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QSYMM16, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QSYMM8, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QSYMM8_PER_CHANNEL, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QASYMM16, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::U8, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::S8, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::U16, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::S16, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::U32, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::S32, DataLayout::NHWC), // Unsupported data type
- TensorInfo(TensorShape(32U, 13U, 8U), 1, DataType::F32, DataLayout::NCHW), // Unsupported data layout
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QASYMM8, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QASYMM8_SIGNED, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QSYMM16, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QSYMM8, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QSYMM8_PER_CHANNEL, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::QASYMM16, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::U8, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::S8, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::U16, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::S16, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::U32, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(8U, 32U, 13U), 1, DataType::S32, DataLayout::NHWC), // Unsupported data type
+ TensorInfo(TensorShape(32U, 13U, 8U), 1, DataType::F32, DataLayout::NCHW), // Unsupported data layout
TensorInfo(TensorShape(8U, 32U, 13U, 4U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(8U, 32U, 13U, 4U), 1, DataType::F32, DataLayout::NHWC), // weight dimension > 3
+ TensorInfo(TensorShape(8U, 32U, 13U, 4U), 1, DataType::F32, DataLayout::NHWC), // weight dimension > 3
TensorInfo(TensorShape(8U, 32U, 13U, 4U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(8U, 32U, 13U, 4U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(8U, 32U, 13U, 4U), 1, DataType::F32, DataLayout::NHWC),
@@ -87,8 +85,6 @@
TensorInfo(TensorShape(2U, 3U, 3U, 2U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(2U, 3U, 3U, 2U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(2U, 3U, 3U, 2U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(2U, 3U, 3U, 2U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(16U, 3U, 3U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(16U, 3U, 3U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(16U, 3U, 3U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(24U, 3U, 3U), 1, DataType::QASYMM8, DataLayout::NHWC),
@@ -115,8 +111,6 @@
TensorInfo(TensorShape(2U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(4U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(2U, 2U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(2U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(16U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(16U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(16U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(24U), 1, DataType::S32, DataLayout::NHWC),
@@ -138,34 +132,6 @@
TensorInfo(TensorShape(24U), 1, DataType::F32, DataLayout::NHWC),
TensorInfo(TensorShape(24U), 1, DataType::F32, DataLayout::NHWC),
})),
- framework::dataset::make("OutputInfo", { TensorInfo(TensorShape(2U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(2U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(2U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(2U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(2U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(2U, 27U, 13U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(16U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(16U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(16U, 25U, 11U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::QASYMM8, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::QASYMM8_SIGNED, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::QSYMM16, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::QSYMM8, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::QSYMM8_PER_CHANNEL, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::QASYMM16, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::U8, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::S8, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::U16, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::S16, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::U32, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U), 1, DataType::S32, DataLayout::NHWC),
- TensorInfo(TensorShape(32U, 11U, 24U), 1, DataType::F32, DataLayout::NCHW),
- TensorInfo(TensorShape(24U, 32U, 11U, 4U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 32U, 11U, 4U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 33U, 14U, 4U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 17U, 5U, 4U), 1, DataType::F32, DataLayout::NHWC),
- TensorInfo(TensorShape(24U, 15U, 4U, 4U), 1, DataType::F32, DataLayout::NHWC),
- })),
framework::dataset::make("Padding", { Padding2D(0, 0, 0, 0),
Padding2D(0, 0, 0, 0),
Padding2D(0, 0, 0, 0),
@@ -173,8 +139,6 @@
Padding2D(0, 0, 0, 0),
Padding2D(0, 0, 0, 0),
Padding2D(0, 0, 0, 0),
- Padding2D(0, 0, 0, 0),
- Padding2D(0, 0, 0, 0),
Padding2D(1, 1, 0, 0),
Padding2D(1, 1, 0, 0),
Padding2D(1, 1, 0, 0),
@@ -217,8 +181,6 @@
Size2D(1, 1),
Size2D(1, 1),
Size2D(1, 1),
- Size2D(1, 1),
- Size2D(1, 1),
Size2D(2, 3),
Size2D(2, 3),
})),
@@ -227,8 +189,6 @@
3,
1,
1,
- 1,
- 2,
2,
2,
3,
@@ -255,8 +215,6 @@
Size2D(1U, 1U),
Size2D(1U, 1U),
Size2D(1U, 1U),
- Size2D(1U, 1U),
- Size2D(20U, 1U),
Size2D(0U, 1U),
Size2D(1U, 1U),
Size2D(1U, 1U),
@@ -278,10 +236,10 @@
Size2D(1U, 1U),
Size2D(2U, 3U),
})),
- framework::dataset::make("Expected", { false, false, false, false, false, false, false, false, true, false,
+ framework::dataset::make("Expected", { false, false, false, false, false, false, true, false,
false, false, false, false, false, false, false, false, false, false,
false, false, true, false, true, true, true })),
- input_info, weights_info, biases_info, output_info, padding, stride, depth_multiplier, dilation, expected)
+ input_info, weights_info, biases_info, padding, stride, depth_multiplier, dilation, expected)
{
CLCompileContext cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
GpuWorkloadContext gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
@@ -290,7 +248,6 @@
const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
const TensorInfo sketch_weights_info = sketch.create_tensor_info(weights_info);
const TensorInfo sketch_biases_info = sketch.create_tensor_info(biases_info);
- const TensorInfo sketch_output_info = sketch.create_tensor_info(output_info);
DepthwiseConv2dAttributes attributes {};
attributes.pad(padding)
@@ -298,7 +255,7 @@
.dilation(dilation)
.depth_multiplier(depth_multiplier);
- const Status status = GpuDepthwiseConv2d::validate_op(sketch, &sketch_input_info, &sketch_weights_info, &sketch_biases_info, &sketch_output_info, attributes);
+ const Status status = GpuDepthwiseConv2d::validate_op(sketch, &sketch_input_info, &sketch_weights_info, &sketch_biases_info, attributes);
const bool res = bool(status);
ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
}
diff --git a/tests/validation/dynamic_fusion/gpu/cl/Reshape.cpp b/tests/validation/dynamic_fusion/gpu/cl/Reshape.cpp
index b7a71db..51822b0 100644
--- a/tests/validation/dynamic_fusion/gpu/cl/Reshape.cpp
+++ b/tests/validation/dynamic_fusion/gpu/cl/Reshape.cpp
@@ -44,17 +44,15 @@
TensorInfo(TensorShape(9U, 5U, 7U, 3U), 1, DataType::F32),
TensorInfo(TensorShape(8U, 4U, 6U, 4U), 1, DataType::F32),
TensorInfo(TensorShape(8U, 4U, 6U, 4U), 1, DataType::F32), // mismatching dimensions
- TensorInfo(TensorShape(9U, 5U, 7U, 3U), 1, DataType::F16), // mismatching types
}),
-framework::dataset::make("OutputInfo",
+framework::dataset::make("OutputShape",
{
- TensorInfo(TensorShape(9U, 5U, 21U), 1, DataType::F32),
- TensorInfo(TensorShape(8U, 24U, 4U), 1, DataType::F32),
- TensorInfo(TensorShape(192U, 192U), 1, DataType::F32),
- TensorInfo(TensorShape(9U, 5U, 21U), 1, DataType::F32),
+ TensorShape(9U, 5U, 21U),
+ TensorShape(8U, 24U, 4U),
+ TensorShape(192U, 192U),
})),
-framework::dataset::make("Expected", { true, true, false, false })),
-input_info, output_info, expected)
+framework::dataset::make("Expected", { true, true, false })),
+input_info, output_shape, expected)
{
// Create a new workload sketch
auto cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
@@ -62,13 +60,12 @@
GpuWorkloadSketch sketch{ &gpu_ctx };
// Create sketch tensors
- auto input_shape = input_info.tensor_shape();
- auto output_shape = output_info.tensor_shape();
- auto src_info = sketch.create_tensor_info(input_info);
- auto dst_info = sketch.create_tensor_info(output_info);
+ TensorShape input_shape = input_info.tensor_shape();
+ TensorInfo src_info = sketch.create_tensor_info(input_info);
+
ReshapeAttributes attributes;
attributes.shape(output_shape);
- Status status = GpuReshape::validate_op(sketch, &src_info, &dst_info, attributes);
+ Status status = GpuReshape::validate_op(sketch, &src_info, attributes);
ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
}
diff --git a/tests/validation/dynamic_fusion/gpu/cl/Resize.cpp b/tests/validation/dynamic_fusion/gpu/cl/Resize.cpp
index 3791aef..696be54 100644
--- a/tests/validation/dynamic_fusion/gpu/cl/Resize.cpp
+++ b/tests/validation/dynamic_fusion/gpu/cl/Resize.cpp
@@ -1,5 +1,5 @@
/*
-* Copyright (c) 2022 Arm Limited.
+* Copyright (c) 2022-2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -86,8 +86,8 @@
const auto default_input_shape = TensorShape{ 2, 3, 3, 2 };
const auto default_output_shape = TensorShape{ 4, 6, 3, 2 };
-constexpr auto default_data_type = DataType::U8;
-constexpr auto default_data_layout = DataLayout::NHWC;
+constexpr auto default_data_type = DataType::U8;
+constexpr auto default_data_layout = DataLayout::NHWC;
TEST_CASE(NullPtr, framework::DatasetMode::ALL)
{
@@ -98,15 +98,10 @@
GpuWorkloadContext gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
GpuWorkloadSketch sketch{ &gpu_ctx };
- const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- const TensorInfo sketch_output_info = sketch.create_tensor_info(output_info);
+ const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
// nullptr is given as input
- Status status = GpuResize::validate_op(sketch, nullptr, &sketch_output_info, ResizeAttributes());
- ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
-
- // nullptr is given as output
- status = GpuResize::validate_op(sketch, &sketch_input_info, nullptr, ResizeAttributes());
+ Status status = GpuResize::validate_op(sketch, nullptr, ResizeAttributes());
ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
}
@@ -137,18 +132,19 @@
for(auto &kv : supported_data_types)
{
- const TensorInfo input_info = TensorInfo{ default_input_shape, 1, kv.first, default_data_layout };
- const TensorInfo output_info = TensorInfo{ default_output_shape, 1, kv.first, default_data_layout };
+ const TensorInfo input_info = TensorInfo{ default_input_shape, 1, kv.first, default_data_layout };
CLCompileContext cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
GpuWorkloadContext gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
GpuWorkloadSketch sketch{ &gpu_ctx };
- const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- const TensorInfo sketch_output_info = sketch.create_tensor_info(output_info);
+ const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- // nullptr is given as input
- Status status = GpuResize::validate_op(sketch, &sketch_input_info, &sketch_output_info, ResizeAttributes());
+ ResizeAttributes attributes;
+ attributes.output_width(default_output_shape[0]); // shape is not important unless it's empty
+ attributes.output_height(default_output_shape[1]);
+
+ Status status = GpuResize::validate_op(sketch, &sketch_input_info, attributes);
ARM_COMPUTE_EXPECT(bool(status) == kv.second, framework::LogLevel::ERRORS);
}
}
@@ -164,10 +160,9 @@
GpuWorkloadContext gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
GpuWorkloadSketch sketch{ &gpu_ctx };
- const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- const TensorInfo sketch_output_info = sketch.create_tensor_info(output_info);
+ const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- Status status = GpuResize::validate_op(sketch, &sketch_input_info, &sketch_output_info, ResizeAttributes());
+ Status status = GpuResize::validate_op(sketch, &sketch_input_info, ResizeAttributes());
ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
}
@@ -185,15 +180,14 @@
GpuWorkloadContext gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
GpuWorkloadSketch sketch{ &gpu_ctx };
- const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- const TensorInfo sketch_output_info = sketch.create_tensor_info(output_info);
+ const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
ResizeAttributes attributes{};
attributes.interpolation_policy(interpolation_policy)
.sampling_policy(sampling_policy)
.align_corners(align_corners);
- Status status = GpuResize::validate_op(sketch, &sketch_input_info, &sketch_output_info, attributes);
+ Status status = GpuResize::validate_op(sketch, &sketch_input_info, attributes);
ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
}
@@ -207,13 +201,12 @@
GpuWorkloadContext gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
GpuWorkloadSketch sketch{ &gpu_ctx };
- const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- const TensorInfo sketch_output_info = sketch.create_tensor_info(output_info);
+ const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
ResizeAttributes attributes{};
attributes.interpolation_policy(interpolation_policy);
- Status status = GpuResize::validate_op(sketch, &sketch_input_info, &sketch_output_info, attributes);
+ Status status = GpuResize::validate_op(sketch, &sketch_input_info, attributes);
ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
}
@@ -227,13 +220,12 @@
GpuWorkloadContext gpu_ctx = GpuWorkloadContext{ &cl_compile_ctx };
GpuWorkloadSketch sketch{ &gpu_ctx };
- const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
- const TensorInfo sketch_output_info = sketch.create_tensor_info(output_info);
+ const TensorInfo sketch_input_info = sketch.create_tensor_info(input_info);
ResizeAttributes attributes{};
attributes.interpolation_policy(interpolation_policy);
- Status status = GpuResize::validate_op(sketch, &sketch_input_info, &sketch_output_info, attributes);
+ Status status = GpuResize::validate_op(sketch, &sketch_input_info, attributes);
ARM_COMPUTE_EXPECT(bool(status) == false, framework::LogLevel::ERRORS);
}