IVGCVSW-7214 Disable BF16-Turbo-Mode and remove conversion layers

 - Remove Bf16ToFp32 Conversion Layer
 - Remove Fp32ToBf16 Conversion Layer
 - Remove B16 Conversion tests
 * Throw exception if m_ReduceFp32ToBf16 optimzer option is set to true
 * Provide comments to enable fast math in order to use bf16
 * Update docs to inform users to enable fast math for bf16

 Execute Network Changes
 * Require bf16_turbo_mode to also have fast_math_enabled set to true
 - Remove setting m_ReduceFp32ToBf16 optimizer option

Signed-off-by: Ryan OShea <ryan.oshea3@arm.com>
Change-Id: Ibaa6da9d29c96a1ce32ff5196b0847fde9f04a1c

7 files changed, 0 insertions, 192 deletions

diff --git a/src/backends/backendsCommon/test/CMakeLists.txt b/src/backends/backendsCommon/test/CMakeLists.txt
index 5e283990f7..232226b5df 100644
--- a/src/backends/backendsCommon/test/CMakeLists.txt
+++ b/src/backends/backendsCommon/test/CMakeLists.txt
@@ -83,12 +83,8 @@ list(APPEND armnnBackendsCommonUnitTests_sources
     layerTests/Conv2dTestImpl.hpp
     layerTests/Conv3dTestImpl.cpp
     layerTests/Conv3dTestImpl.hpp
-    layerTests/ConvertBf16ToFp32TestImpl.cpp
-    layerTests/ConvertBf16ToFp32TestImpl.hpp
     layerTests/ConvertFp16ToFp32TestImpl.cpp
     layerTests/ConvertFp16ToFp32TestImpl.hpp
-    layerTests/ConvertFp32ToBf16TestImpl.cpp
-    layerTests/ConvertFp32ToBf16TestImpl.hpp
     layerTests/ConvertFp32ToFp16TestImpl.cpp
     layerTests/ConvertFp32ToFp16TestImpl.hpp
     layerTests/DebugTestImpl.cpp
diff --git a/src/backends/backendsCommon/test/IsLayerSupportedTestImpl.hpp b/src/backends/backendsCommon/test/IsLayerSupportedTestImpl.hpp
index 5fdcd9c57a..18f11a542e 100644
--- a/src/backends/backendsCommon/test/IsLayerSupportedTestImpl.hpp
+++ b/src/backends/backendsCommon/test/IsLayerSupportedTestImpl.hpp
@@ -630,12 +630,8 @@ DECLARE_LAYER_POLICY_2_PARAM(Concat)
 
 DECLARE_LAYER_POLICY_1_PARAM(Constant)
 
-DECLARE_LAYER_POLICY_1_PARAM(ConvertBf16ToFp32)
-
 DECLARE_LAYER_POLICY_1_PARAM(ConvertFp16ToFp32)
 
-DECLARE_LAYER_POLICY_1_PARAM(ConvertFp32ToBf16)
-
 DECLARE_LAYER_POLICY_1_PARAM(ConvertFp32ToFp16)
 
 DECLARE_LAYER_POLICY_2_PARAM(Convolution2d)
diff --git a/src/backends/backendsCommon/test/LayerTests.hpp b/src/backends/backendsCommon/test/LayerTests.hpp
index 25435b24ec..00bfea5452 100644
--- a/src/backends/backendsCommon/test/LayerTests.hpp
+++ b/src/backends/backendsCommon/test/LayerTests.hpp
@@ -16,9 +16,7 @@
 #include <backendsCommon/test/layerTests/ChannelShuffleTestImpl.hpp>
 #include <backendsCommon/test/layerTests/ComparisonTestImpl.hpp>
 #include <backendsCommon/test/layerTests/ConcatTestImpl.hpp>
-#include <backendsCommon/test/layerTests/ConvertBf16ToFp32TestImpl.hpp>
 #include <backendsCommon/test/layerTests/ConvertFp16ToFp32TestImpl.hpp>
-#include <backendsCommon/test/layerTests/ConvertFp32ToBf16TestImpl.hpp>
 #include <backendsCommon/test/layerTests/ConvertFp32ToFp16TestImpl.hpp>
 #include <backendsCommon/test/layerTests/Conv2dTestImpl.hpp>
 #include <backendsCommon/test/layerTests/Conv3dTestImpl.hpp>
diff --git a/src/backends/backendsCommon/test/layerTests/ConvertBf16ToFp32TestImpl.cpp b/src/backends/backendsCommon/test/layerTests/ConvertBf16ToFp32TestImpl.cpp
deleted file mode 100644
index 0dd8b598ac..0000000000
--- a/src/backends/backendsCommon/test/layerTests/ConvertBf16ToFp32TestImpl.cpp
+++ /dev/null
@@ -1,62 +0,0 @@
-//
-// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-
-#include "ConvertBf16ToFp32TestImpl.hpp"
-
-#include <armnnTestUtils/TensorCopyUtils.hpp>
-#include <armnnTestUtils/WorkloadTestUtils.hpp>
-
-#include <armnnTestUtils/TensorHelpers.hpp>
-
-LayerTestResult<float, 4> ConvertBf16ToFp32Test(
-    armnn::IWorkloadFactory& workloadFactory,
-    const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
-    const armnn::ITensorHandleFactory& tensorHandleFactory)
-{
-    IgnoreUnused(memoryManager);
-
-    const armnn::TensorInfo inputTensorInfo({1, 3, 2, 3}, armnn::DataType::BFloat16);
-    const armnn::TensorInfo outputTensorInfo({1, 3, 2, 3}, armnn::DataType::Float32);
-
-    std::vector<armnn::BFloat16> inputValues = armnnUtils::QuantizedVector<armnn::BFloat16>(
-        {
-            -37.5f, -15.2f, -8.76f, -2.0f, -1.5f, -1.3f, -0.5f, -0.4f, 0.0f,
-            1.0f, 0.4f, 0.5f, 1.3f, 1.5f, 2.0f, 8.76f, 15.2f, 37.5f
-        },
-        1.0f, 0);
-
-    std::vector<float> actualOutput(outputTensorInfo.GetNumElements());
-    std::vector<float> expectedOutput =
-        {
-            -37.5f, -15.2f, -8.76f, -2.0f, -1.5f, -1.3f, -0.5f, -0.4f, 0.0f,
-            1.0f, 0.4f, 0.5f, 1.3f, 1.5f, 2.0f, 8.76f, 15.2f, 37.5f
-        };
-
-    std::unique_ptr<armnn::ITensorHandle> inputHandle = tensorHandleFactory.CreateTensorHandle(inputTensorInfo);
-    std::unique_ptr<armnn::ITensorHandle> outputHandle = tensorHandleFactory.CreateTensorHandle(outputTensorInfo);
-
-    armnn::ConvertBf16ToFp32QueueDescriptor data;
-    armnn::WorkloadInfo info;
-    AddInputToWorkload(data, info, inputTensorInfo, inputHandle.get());
-    AddOutputToWorkload(data, info, outputTensorInfo, outputHandle.get());
-
-    std::unique_ptr<armnn::IWorkload> workload = workloadFactory.CreateWorkload(armnn::LayerType::ConvertBf16ToFp32,
-                                                                                data,
-                                                                                info);
-
-    inputHandle->Allocate();
-    outputHandle->Allocate();
-
-    CopyDataToITensorHandle(inputHandle.get(), inputValues.data());
-
-    workload->Execute();
-
-    CopyDataFromITensorHandle(actualOutput.data(), outputHandle.get());
-
-    return LayerTestResult<float, 4>(actualOutput,
-                                     expectedOutput,
-                                     outputHandle->GetShape(),
-                                     outputTensorInfo.GetShape());
-}
diff --git a/src/backends/backendsCommon/test/layerTests/ConvertBf16ToFp32TestImpl.hpp b/src/backends/backendsCommon/test/layerTests/ConvertBf16ToFp32TestImpl.hpp
deleted file mode 100644
index bcb0d6f124..0000000000
--- a/src/backends/backendsCommon/test/layerTests/ConvertBf16ToFp32TestImpl.hpp
+++ /dev/null
@@ -1,18 +0,0 @@
-//
-// Copyright © 2020 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-
-#pragma once
-
-#include <armnnTestUtils/LayerTestResult.hpp>
-
-#include <BFloat16.hpp>
-
-#include <armnn/backends/IBackendInternal.hpp>
-#include <armnn/backends/WorkloadFactory.hpp>
-
-LayerTestResult<float, 4> ConvertBf16ToFp32Test(
-    armnn::IWorkloadFactory& workloadFactory,
-    const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
-    const armnn::ITensorHandleFactory& tensorHandleFactory);
diff --git a/src/backends/backendsCommon/test/layerTests/ConvertFp32ToBf16TestImpl.cpp b/src/backends/backendsCommon/test/layerTests/ConvertFp32ToBf16TestImpl.cpp
deleted file mode 100644
index 5ee8f1dd9a..0000000000
--- a/src/backends/backendsCommon/test/layerTests/ConvertFp32ToBf16TestImpl.cpp
+++ /dev/null
@@ -1,84 +0,0 @@
-//
-// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-
-#include "ConvertFp32ToBf16TestImpl.hpp"
-
-#include <armnnTestUtils/TensorCopyUtils.hpp>
-#include <armnnTestUtils/WorkloadTestUtils.hpp>
-
-#include <armnnTestUtils/TensorHelpers.hpp>
-
-LayerTestResult<armnn::BFloat16, 4> ConvertFp32ToBf16Test(
-    armnn::IWorkloadFactory& workloadFactory,
-    const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
-    const armnn::ITensorHandleFactory& tensorHandleFactory)
-{
-    IgnoreUnused(memoryManager);
-
-    const armnn::TensorInfo inputTensorInfo({1, 2, 4, 3}, armnn::DataType::Float32);
-    const armnn::TensorInfo outputTensorInfo({1, 2, 4, 3}, armnn::DataType::BFloat16);
-
-    std::vector<float> input =
-        {
-          -37.5f, -15.2f, -8.76f,
-          -2.0f, -1.5f, -1.3f,
-          -0.5f, -0.4f, 0.0f,
-          1.0f, 0.4f, 0.5f,
-          1.3f, 1.5f, 2.0f,
-          8.76f, 15.2f, 37.5f,
-          3.8f, // 0x40733333 Round down
-          3.1055E+29f, // 0x707ADC3C Round up
-          9.149516E-10f, // 0x307B7FFF Round down
-          -3.8f, // 0xC0733333 Round down
-          -3.1055E+29f, // 0xF07ADC3C Round up
-          -9.149516E-10f // 0xB07B7FFF Round down
-        };
-
-    std::vector<armnn::BFloat16> expectedOutput = armnnUtils::QuantizedVector<armnn::BFloat16>(
-        {
-          -37.5f, -15.2f, -8.76f,
-          -2.0f, -1.5f, -1.3f,
-          -0.5f, -0.4f, 0.0f,
-          1.0f, 0.4f, 0.5f,
-          1.3f, 1.5f, 2.0f,
-          8.76f, 15.2f, 37.5f,
-          3.796875f, // 0x4073
-          3.1072295E29f, // 0x707B
-          9.131327E-10f, // 0x307B
-          -3.796875f, // 0xC073
-          -3.1072295E29f, // 0xF07B
-          -9.131327E-10f // 0xB07B
-        },
-        1.0f, 0);
-
-    std::vector<armnn::BFloat16> actualOutput(outputTensorInfo.GetNumElements());
-
-    std::unique_ptr<armnn::ITensorHandle> inputHandle = tensorHandleFactory.CreateTensorHandle(inputTensorInfo);
-    std::unique_ptr<armnn::ITensorHandle> outputHandle = tensorHandleFactory.CreateTensorHandle(outputTensorInfo);
-
-    armnn::ConvertFp32ToBf16QueueDescriptor data;
-    armnn::WorkloadInfo info;
-    AddInputToWorkload(data, info, inputTensorInfo, inputHandle.get());
-    AddOutputToWorkload(data, info, outputTensorInfo, outputHandle.get());
-
-    std::unique_ptr<armnn::IWorkload> workload = workloadFactory.CreateWorkload(armnn::LayerType::ConvertFp32ToBf16,
-                                                                                data,
-                                                                                info);
-
-    inputHandle->Allocate();
-    outputHandle->Allocate();
-
-    CopyDataToITensorHandle(inputHandle.get(), input.data());
-
-    workload->Execute();
-
-    CopyDataFromITensorHandle(actualOutput.data(), outputHandle.get());
-
-    return LayerTestResult<armnn::BFloat16, 4>(actualOutput,
-                                               expectedOutput,
-                                               outputHandle->GetShape(),
-                                               outputTensorInfo.GetShape());
-
-}
diff --git a/src/backends/backendsCommon/test/layerTests/ConvertFp32ToBf16TestImpl.hpp b/src/backends/backendsCommon/test/layerTests/ConvertFp32ToBf16TestImpl.hpp
deleted file mode 100644
index c2286d9c41..0000000000
--- a/src/backends/backendsCommon/test/layerTests/ConvertFp32ToBf16TestImpl.hpp
+++ /dev/null
@@ -1,18 +0,0 @@
-//
-// Copyright © 2020 Arm Ltd. All rights reserved.
-// SPDX-License-Identifier: MIT
-//
-
-#pragma once
-
-#include <armnnTestUtils/LayerTestResult.hpp>
-
-#include <BFloat16.hpp>
-
-#include <armnn/backends/IBackendInternal.hpp>
-#include <armnn/backends/WorkloadFactory.hpp>
-
-LayerTestResult<armnn::BFloat16, 4> ConvertFp32ToBf16Test(
-    armnn::IWorkloadFactory& workloadFactory,
-    const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
-    const armnn::ITensorHandleFactory& tensorHandleFactory);