Rework OpenCL Depthwise Convolution

- Remove dedicated kernels for NCHW. Now we only use NHWC with permute
- Remove specialized kernels for 3x3 NHWC
- Simplify CLDepthwiseConvolutionLayer.cpp to call just the native
  implementation for both floating-point and quantized data types
- Develop two parametric opencl kernels for depthwise convolution layer NHWC
  (floating-point and quantized)
- Add support to export the weights to cl_image
- Extend test for depthwise convolution on opencl

Resolves COMPMID-4417

Change-Id: I253dd5d959a70783c82e62b1771a5e9f91621cb0
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5806
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>

1 files changed, 161 insertions, 18 deletions

diff --git a/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp b/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp
index f640ee2b18..f565255719 100644
--- a/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp
+++ b/tests/validation/CL/DepthwiseConvolutionLayerNative.cpp
@@ -62,7 +62,7 @@ RelativeTolerance<half_float::half>  rel_tolerance_f16(half_float::half(0.01f));
 constexpr float                      abs_tolerance_f16(0.03f);
 
 /** Width values to test - Precommit */
-const auto width_values_precommit = framework::dataset::make("width", { 1U, 17U, 32U } );
+const auto width_values_precommit = framework::dataset::make("width", { 1U, 33U } );
 
 /** Width values to test - Nightly */
 const auto width_values_nightly = framework::dataset::make("width", { 53U, 47U } );
@@ -79,6 +79,12 @@ const auto channel_values_precommit = framework::dataset::make("channels", { 15U
 /** Channel values to test - Nightly */
 const auto channel_values_nightly = framework::dataset::make("channels", { 33U, 19U });
 
+/** Channel values to test with cl_image support - Precommit */
+const auto channel_values_export_to_cl_image_precommit = framework::dataset::make("channels", { 16U });
+
+/** Channel values to test with cl_image support - Nightly */
+const auto channel_values_export_to_cl_image_nightly = framework::dataset::make("channels", { 32U });
+
 /** Batch values to test - Precommit */
 const auto batch_values_precommit = framework::dataset::make("batch", { 1U, 2U });
 
@@ -115,11 +121,17 @@ const auto n0_values_precommit = framework::dataset::make("N0", {2, 4});
 /** N0 values to test - Nightly */
 const auto n0_values_nightly = framework::dataset::make("N0", {3, 8});
 
+/** N0 values to test with cl_image support - Precommit */
+const auto n0_values_export_to_cl_image_precommit = framework::dataset::make("N0", {4});
+
+/** N0 values to test with cl_image support - Nightly */
+const auto n0_values_export_to_cl_image_nightly = framework::dataset::make("N0", {8});
+
 /** Activation values to test */
 const auto act_values = framework::dataset::make("Activation",
 {
     ActivationLayerInfo(),
-    ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 8.f, 2.f),
+    ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 6.0f, 0.5f),
 });
 
 } // namespace
@@ -129,7 +141,7 @@ TEST_SUITE(DepthwiseConvolutionLayerNative)
 TEST_SUITE(Float)
 TEST_SUITE(FP32)
 FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_precommit,
                                                                                                 height_values_precommit),
                                                                                                 channel_values_precommit),
@@ -142,14 +154,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<fl
                                                                                                 framework::dataset::make("DataType", DataType::F32)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                n0_values_precommit))
+                                                                                                n0_values_precommit),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
 }
 
 FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::NIGHTLY,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_nightly,
                                                                                                 height_values_nightly),
                                                                                                 channel_values_nightly),
@@ -162,16 +175,79 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<fl
                                                                                                 framework::dataset::make("DataType", DataType::F32)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                n0_values_nightly))
+                                                                                                n0_values_nightly),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
 }
+
+TEST_SUITE(ExportWeightsToCLImage)
+FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::ALL,
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                                                width_values_precommit,
+                                                                                                height_values_precommit),
+                                                                                                channel_values_export_to_cl_image_precommit),
+                                                                                                batch_values_precommit),
+                                                                                                kernel_sz_values_precommit),
+                                                                                                framework::dataset::make("depth_multiplier", 1)),
+                                                                                                dilation_values),
+                                                                                                stride_values),
+                                                                                                padding_valid_values),
+                                                                                                framework::dataset::make("DataType", DataType::F32)),
+                                                                                                data_layout_values),
+                                                                                                act_values),
+                                                                                                n0_values_export_to_cl_image_precommit),
+                                                                                                framework::dataset::make("ExportToCLImage", true)))
+{
+   // Validate output
+    if(_validate_output)
+    {
+        // Validate output
+        validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
+    }
+    else
+    {
+        ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+        framework::ARM_COMPUTE_PRINT_INFO();
+    }
+}
+
+FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::NIGHTLY,
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                                                width_values_nightly,
+                                                                                                height_values_nightly),
+                                                                                                channel_values_export_to_cl_image_nightly),
+                                                                                                batch_values_nightly),
+                                                                                                kernel_sz_values_nightly),
+                                                                                                framework::dataset::make("depth_multiplier", 1)),
+                                                                                                dilation_values),
+                                                                                                stride_values),
+                                                                                                padding_valid_values),
+                                                                                                framework::dataset::make("DataType", DataType::F32)),
+                                                                                                data_layout_values),
+                                                                                                act_values),
+                                                                                                n0_values_export_to_cl_image_nightly),
+                                                                                                framework::dataset::make("ExportToCLImage", true)))
+{
+   // Validate output
+    if(_validate_output)
+    {
+        // Validate output
+        validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
+    }
+    else
+    {
+        ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+        framework::ARM_COMPUTE_PRINT_INFO();
+    }
+}
+TEST_SUITE_END() // ExportWeightsToCLImage
 TEST_SUITE_END() // FP32
 
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_precommit,
                                                                                                 height_values_precommit),
                                                                                                 channel_values_precommit),
@@ -184,14 +260,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<ha
                                                                                                 framework::dataset::make("DataType", DataType::F16)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                n0_values_precommit))
+                                                                                                n0_values_precommit),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
         validate(CLAccessor(_target), _reference, rel_tolerance_f16);
 }
 
 FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::NIGHTLY,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_nightly,
                                                                                                 height_values_nightly),
                                                                                                 channel_values_nightly),
@@ -204,18 +281,80 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<ha
                                                                                                 framework::dataset::make("DataType", DataType::F16)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                n0_values_nightly))
+                                                                                                n0_values_nightly),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);
 }
+TEST_SUITE(ExportWeightsToCLImage)
+FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::ALL,
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                                                width_values_precommit,
+                                                                                                height_values_precommit),
+                                                                                                channel_values_export_to_cl_image_precommit),
+                                                                                                batch_values_precommit),
+                                                                                                kernel_sz_values_precommit),
+                                                                                                framework::dataset::make("depth_multiplier", 1)),
+                                                                                                dilation_values),
+                                                                                                stride_values),
+                                                                                                padding_valid_values),
+                                                                                                framework::dataset::make("DataType", DataType::F16)),
+                                                                                                data_layout_values),
+                                                                                                act_values),
+                                                                                                n0_values_export_to_cl_image_precommit),
+                                                                                                framework::dataset::make("ExportToCLImage", true)))
+{
+   // Validate output
+    if(_validate_output)
+    {
+        // Validate output
+        validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);
+    }
+    else
+    {
+        ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+        framework::ARM_COMPUTE_PRINT_INFO();
+    }
+}
+
+FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::NIGHTLY,
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                                                                                                width_values_nightly,
+                                                                                                height_values_nightly),
+                                                                                                channel_values_export_to_cl_image_nightly),
+                                                                                                batch_values_nightly),
+                                                                                                kernel_sz_values_nightly),
+                                                                                                framework::dataset::make("depth_multiplier", 1)),
+                                                                                                dilation_values),
+                                                                                                stride_values),
+                                                                                                padding_valid_values),
+                                                                                                framework::dataset::make("DataType", DataType::F16)),
+                                                                                                data_layout_values),
+                                                                                                act_values),
+                                                                                                n0_values_export_to_cl_image_nightly),
+                                                                                                framework::dataset::make("ExportToCLImage", true)))
+{
+   // Validate output
+    if(_validate_output)
+    {
+        // Validate output
+        validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);
+    }
+    else
+    {
+        ARM_COMPUTE_TEST_INFO("cl_khr_image2d_from_buffer not supported. TEST skipped");
+        framework::ARM_COMPUTE_PRINT_INFO();
+    }
+}
+TEST_SUITE_END() // ExportWeightsToCLImage
 TEST_SUITE_END() // FP16
 TEST_SUITE_END() // Float
 TEST_SUITE(DepthMultiplier)
 TEST_SUITE(Float)
 TEST_SUITE(FP32)
 FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_precommit,
                                                                                                 height_values_precommit),
                                                                                                 channel_values_precommit),
@@ -228,14 +367,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<fl
                                                                                                 framework::dataset::make("DataType", DataType::F32)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                framework::dataset::make("N0", 1)))
+                                                                                                framework::dataset::make("N0", 1)),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
 }
 
 FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<float>, framework::DatasetMode::NIGHTLY,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_nightly,
                                                                                                 height_values_nightly),
                                                                                                 channel_values_nightly),
@@ -248,7 +388,8 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<fl
                                                                                                 framework::dataset::make("DataType", DataType::F32)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                framework::dataset::make("N0", 1)))
+                                                                                                framework::dataset::make("N0", 1)),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, rel_tolerance_f32, 0.f, abs_tolerance_f32);
@@ -257,7 +398,7 @@ TEST_SUITE_END() // FP32
 
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::ALL,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_precommit,
                                                                                                 height_values_precommit),
                                                                                                 channel_values_precommit),
@@ -270,14 +411,15 @@ FIXTURE_DATA_TEST_CASE_NEW(RunSmall, CLDepthwiseConvolutionLayerNativeFixture<ha
                                                                                                 framework::dataset::make("DataType", DataType::F16)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                framework::dataset::make("N0", 1)))
+                                                                                                framework::dataset::make("N0", 1)),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
         validate(CLAccessor(_target), _reference, rel_tolerance_f16);
 }
 
 FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<half>, framework::DatasetMode::NIGHTLY,
-                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
+                combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(combine(
                                                                                                 width_values_nightly,
                                                                                                 height_values_nightly),
                                                                                                 channel_values_nightly),
@@ -290,7 +432,8 @@ FIXTURE_DATA_TEST_CASE_NEW(RunLarge, CLDepthwiseConvolutionLayerNativeFixture<ha
                                                                                                 framework::dataset::make("DataType", DataType::F16)),
                                                                                                 data_layout_values),
                                                                                                 act_values),
-                                                                                                framework::dataset::make("N0", 1)))
+                                                                                                framework::dataset::make("N0", 1)),
+                                                                                                framework::dataset::make("ExportToCLImage", false)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, rel_tolerance_f16, 0.f, abs_tolerance_f16);