From 2b9fa593a0a172bf36a02b5cdb840c6b9b361d7c Mon Sep 17 00:00:00 2001
From: Gunes Bayir <gunes.bayir@arm.com>
Date: Wed, 17 Jan 2024 16:07:03 +0000
Subject: Use the stable CKW API in the GPU dynamic fusion backend

- Refactor all kernels to work with the CKW stable API
- Add support for sub-tile in the op_load/op_store CKW operator
- Fix mismatch in resize
- Add comments in all kernels written with CKW to help developers
understand the structure of the code
- Add texture image support in depthwise convolution written with CKW
- Add support for different block sizes in depthwise convolution
- Remove the use of the dynamic fusion helper functions.
- Add support for floor in the op_unary() of CKW

Resolves: COMPMID-6708, COMPMID-6743, COMPMID-6530

Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Signed-off-by: Jakub Sujak <jakub.sujak@arm.com>

Change-Id: I8104ce4d04a3138a1aeb0b84940e1f1c89e76069
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10914
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
---
 .../dynamic_fusion/gpu/cl/MatMulKernelFixture.h    |  2 +-
 .../fixtures/dynamic_fusion/gpu/cl/Pool2dFixture.h | 41 +++++-----------------
 2 files changed, 9 insertions(+), 34 deletions(-)

(limited to 'tests/validation/fixtures/dynamic_fusion/gpu')

diff --git a/tests/validation/fixtures/dynamic_fusion/gpu/cl/MatMulKernelFixture.h b/tests/validation/fixtures/dynamic_fusion/gpu/cl/MatMulKernelFixture.h
index 65a3363e24..4c1cc94d3d 100644
--- a/tests/validation/fixtures/dynamic_fusion/gpu/cl/MatMulKernelFixture.h
+++ b/tests/validation/fixtures/dynamic_fusion/gpu/cl/MatMulKernelFixture.h
@@ -203,7 +203,7 @@ protected:
                                       bool               pretranspose_b,
                                       DataType           data_type)
     {
-        // We collapse dimensions > 3 onto dimension 3, i.e. 5D+ tensors will look like 4D
+        // We collapse dimensions > 3 onto dimension 3, i.e. 5D+ tensors will look like 3D
         // This is necessary unless we choose to extend gemm reference for 5D+ tensors
         TensorShape output_shape_collapsed = output_shape.collapsed_from(Window::DimZ);
         TensorShape shape_a_collapsed      = shape_a.collapsed_from(Window::DimZ);
diff --git a/tests/validation/fixtures/dynamic_fusion/gpu/cl/Pool2dFixture.h b/tests/validation/fixtures/dynamic_fusion/gpu/cl/Pool2dFixture.h
index dd3519b549..b0c7143d91 100644
--- a/tests/validation/fixtures/dynamic_fusion/gpu/cl/Pool2dFixture.h
+++ b/tests/validation/fixtures/dynamic_fusion/gpu/cl/Pool2dFixture.h
@@ -51,11 +51,11 @@ template <typename TensorType, typename AccessorType, typename FunctionType, typ
 class DynamicFusionGpuPool2dValidationGenericFixture : public framework::Fixture
 {
 public:
-    void setup(TensorShape input_shape, const Pool2dAttributes &pool_attr, DataType data_type, bool mixed_precision)
+    void setup(TensorShape input_shape, const Pool2dAttributes &pool_attr, DataType data_type)
     {
-        _target = compute_target(input_shape, pool_attr, data_type, mixed_precision);
-        _reference =
-            compute_reference(input_shape, convert_pool_attr_to_pool_info(pool_attr, mixed_precision), data_type);
+        _target    = compute_target(input_shape, pool_attr, data_type);
+        _reference = compute_reference(
+            input_shape, convert_pool_attr_to_pool_info(pool_attr, true /* mixed_precision */), data_type);
     }
 
 protected:
@@ -82,10 +82,7 @@ protected:
     }
 
     // Given input is in nchw format
-    TensorType compute_target(TensorShape             input_shape,
-                              const Pool2dAttributes &pool_attr,
-                              const DataType          data_type,
-                              bool                    mixed_precision)
+    TensorType compute_target(TensorShape input_shape, const Pool2dAttributes &pool_attr, const DataType data_type)
     {
         CLScheduler::get().default_reinit();
 
@@ -102,7 +99,7 @@ protected:
         auto dst_info   = context.create_tensor_info();
 
         // Create Pool2dSettings
-        GpuPool2dSettings pool_settings = GpuPool2dSettings().mixed_precision(mixed_precision);
+        GpuPool2dSettings pool_settings = GpuPool2dSettings();
 
         ITensorInfo *ans_info = FunctionType::create_op(sketch, input_info, pool_attr, pool_settings);
         GpuOutput::create_op(sketch, ans_info, dst_info);
@@ -168,29 +165,7 @@ public:
             input_shape,
             Pool2dAttributes().pool_type(pool_type).pool_size(pool_size).pad(pad).stride(stride).exclude_padding(
                 exclude_padding),
-            data_type, false);
-    }
-};
-
-template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
-class DynamicFusionGpuPool2dMixedPrecisionValidationFixture
-    : public DynamicFusionGpuPool2dValidationGenericFixture<TensorType, AccessorType, FunctionType, T>
-{
-public:
-    void setup(TensorShape input_shape,
-               PoolingType pool_type,
-               Size2D      pool_size,
-               Padding2D   pad,
-               Size2D      stride,
-               bool        exclude_padding,
-               DataType    data_type,
-               bool        mixed_precision)
-    {
-        DynamicFusionGpuPool2dValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(
-            input_shape,
-            Pool2dAttributes().pool_type(pool_type).pool_size(pool_size).pad(pad).stride(stride).exclude_padding(
-                exclude_padding),
-            data_type, mixed_precision);
+            data_type);
     }
 };
 
@@ -202,7 +177,7 @@ public:
     void setup(TensorShape input_shape, Pool2dAttributes pool_attr, DataType data_type)
     {
         DynamicFusionGpuPool2dValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(
-            input_shape, pool_attr, data_type, false);
+            input_shape, pool_attr, data_type);
     }
 };
 
-- 
cgit v1.2.1