Port MatMul to Dynamic Fusion + CKW boilerplate code

- Port Matmaul to to Dynamic Fusion.
- Prepare a CKW boilerplate code.
- Implement the following classes:
 - MatMulAttributes
 - GPUMatMulSettings
 - GpuMatMul
 - ClComponentMatMul
 - GpuCkwMatMul

Signed-off-by: Adnan AlSinan <adnan.alsinan@arm.com>
Change-Id: I5a7c183b293973e8a4233b554b2affe0bb28f44d
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10453
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 274 insertions, 0 deletions

diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentMatMul.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentMatMul.cpp
new file mode 100644
index 0000000000..eada61e1b3
--- /dev/null
+++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentMatMul.cpp
@@ -0,0 +1,151 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifdef ACL_INTERNAL_TEST_CKW_IN_DF
+
+#include "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentMatMul.h"
+
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/dynamic_fusion/sketch/attributes/MatMulAttributes.h"
+
+#include "src/core/CL/CLValidate.h"
+#include "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwMatMul.h"
+#include "src/gpu/cl/kernels/helpers/MatMulKernelHelpers.h"
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+namespace
+{
+using Attributes = MatMulAttributes;
+using Settings   = GpuMatMulSettings;
+
+Status validate_matmul_kernel_info(Attributes attributes, Settings settings)
+{
+    const bool adj_lhs = attributes.adj_lhs();
+    const bool adj_rhs = attributes.adj_rhs();
+    const int  m0      = settings.m0();
+    const int  n0      = settings.n0();
+    const int  k0      = settings.k0();
+
+    // Validate M0
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(m0 < 1, "Only positive integers are supported for M0");
+
+    if (adj_lhs)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(((m0 & (m0 - 1)) && (m0 != 3)) || (m0 > 16),
+                                        "Only 1,2,3,4,8,16 are supported for M0 for Lhs transposed");
+    }
+
+    // Validate N0
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(n0 < 1, "Only positive integers are supported for N0");
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(((n0 & (n0 - 1)) && (n0 != 3)) || (n0 > 16),
+                                    "Only 1,2,3,4,8,16 are supported for N0");
+
+    // Validate K0
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(k0 < 1, "Only positive integers are supported for K0");
+    if (!adj_lhs || adj_rhs)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(((k0 & (k0 - 1)) && (k0 != 3)) || (k0 > 16),
+                                        "Only 1,2,3,4,8,16 are supported for K0");
+    }
+
+    return Status{};
+}
+
+} // namespace
+
+Status ClComponentMatMul::validate(const Properties                &properties,
+                                   const ArgumentPack<ITensorInfo> &tensors,
+                                   const Attributes                &attributes,
+                                   const Settings                  &settings)
+{
+    ARM_COMPUTE_UNUSED(properties);
+    ARM_COMPUTE_UNUSED(attributes);
+
+    const auto lhs = tensors.get_const_tensor(TensorType::ACL_SRC_0);
+    const auto rhs = tensors.get_const_tensor(TensorType::ACL_SRC_1);
+    const auto dst = tensors.get_const_tensor(TensorType::ACL_DST_0);
+
+    // Check if Matching data type
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(lhs, rhs);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(lhs, dst);
+
+    // Data type
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::F16, DataType::F32);
+    // Data layout
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(lhs, DataLayout::NHWC);
+
+    // All tensor infos are initialized
+    ARM_COMPUTE_RETURN_ERROR_ON(lhs->tensor_shape().total_size() == 0);
+    ARM_COMPUTE_RETURN_ERROR_ON(rhs->tensor_shape().total_size() == 0);
+    ARM_COMPUTE_RETURN_ERROR_ON(dst->tensor_shape().total_size() == 0);
+
+    // Device requirements are met
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(lhs);
+
+    // Check if dst shape is correct
+    MatMulKernelInfo matmul_kernel_info =
+        MatMulKernelInfo(attributes.adj_lhs(), attributes.adj_rhs(), settings.m0(), settings.n0(), settings.k0());
+    const auto expected_dst_shape =
+        misc::shape_calculator::compute_matmul_shape(lhs->tensor_shape(), rhs->tensor_shape(), matmul_kernel_info);
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), expected_dst_shape);
+
+    // Check if block sizes are supported
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_matmul_kernel_info(attributes, settings));
+
+    ARM_COMPUTE_RETURN_ON_ERROR(
+        opencl::kernels::validate_matmul_input_shapes(lhs->tensor_shape(), rhs->tensor_shape(), matmul_kernel_info));
+
+    return Status{};
+}
+
+ClComponentMatMul::ClComponentMatMul(ComponentId                      id,
+                                     const Properties                &properties,
+                                     const ArgumentPack<ITensorInfo> &tensors,
+                                     const Attributes                &attributes,
+                                     const Settings                  &settings)
+    : IGpuKernelComponent{id, properties, tensors},
+      _component_writer{std::make_unique<GpuCkwMatMul>(id, tensors, attributes, settings)}
+{
+}
+
+ClComponentMatMul::~ClComponentMatMul()
+{
+}
+
+const IGpuCkwComponentDriver *ClComponentMatMul::ckw_component_driver() const
+{
+    return _component_writer.get();
+}
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+
+#endif // ACL_INTERNAL_TEST_CKW_IN_DF
diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentMatMul.h b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentMatMul.h
new file mode 100644
index 0000000000..41833e4adb
--- /dev/null
+++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentMatMul.h
@@ -0,0 +1,123 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTMATMUL_H
+#define ACL_SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTMATMUL_H
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/KernelDescriptors.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuMatMul.h"
+
+#include "src/dynamic_fusion/sketch/gpu/components/IGpuKernelComponent.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+/** Forward declaration */
+class ITensorInfo;
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** Forward declaration */
+template <typename T>
+class ArgumentPack;
+class MatMulAttributes;
+class GpuCkwMatMul;
+
+class ClComponentMatMul final : public IGpuKernelComponent
+{
+public:
+    /** Attributes are a set of backend-agnostic parameters that define what a component does */
+    using Attributes = MatMulAttributes;
+    /** Settings are a set of backend-specific parameters that influence the implementation of a component */
+    using Settings = GpuMatMulSettings;
+
+    /** Validate the component
+     *
+     * @param[in]     properties Component properties
+     * @param[in,out] tensors    Tensor arguments to the component
+     * @param[in]     attributes Component attributes
+     * @param[in]     settings   Component settings
+     *
+     * @return Status       Validation results
+     *
+     * Tensor argument names:
+     * - ACL_SRC_0: LHS
+     * - ACL_SRC_1: RHS
+     * - ACL_DST_0: Output
+     *
+     * Tensor argument constness:
+     * - ACL_SRC_0: Const
+     * - ACL_SRC_1: Const
+     * - ACL_DST_0: Const
+     *
+     * Valid data layouts:
+     * - NHWC
+     *
+     * Valid data type configurations:
+     * |ACL_SRC_0      |ACL_SRC_1      |ACL_DST_0      |
+     * |:--------------|:--------------|:--------------|
+     * |F16            |F16            |F16            |
+     * |F32            |F32            |F32            |
+     */
+    static Status validate(const Properties                &properties,
+                           const ArgumentPack<ITensorInfo> &tensors,
+                           const Attributes                &attributes,
+                           const Settings                  &settings);
+
+    /** Constructor
+     *
+     * Similar to @ref ClComponentMatMul::validate()
+     */
+    ClComponentMatMul(ComponentId                      id,
+                      const Properties                &properties,
+                      const ArgumentPack<ITensorInfo> &tensors,
+                      const Attributes                &attributes,
+                      const Settings                  &settings);
+    /** Destructor */
+    ~ClComponentMatMul() override;
+    /** Prevent instances of this class from being copy constructed */
+    ClComponentMatMul(const ClComponentMatMul &component) = delete;
+    /** Prevent instances of this class from being copied */
+    ClComponentMatMul &operator=(const ClComponentMatMul &component) = delete;
+    /** Allow instances of this class to be move constructed */
+    ClComponentMatMul(ClComponentMatMul &&component) = default;
+    /** Allow instances of this class to be moved */
+    ClComponentMatMul &operator=(ClComponentMatMul &&component) = default;
+    /** Get writer for the component */
+    const IGpuCkwComponentDriver *ckw_component_driver() const override;
+    /** Get component type */
+    GpuComponentType type() const override
+    {
+        return GpuComponentType::Complex;
+    }
+
+private:
+    std::unique_ptr<GpuCkwMatMul> _component_writer;
+};
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif // ACL_SRC_DYNAMIC_FUSION_SKETCH_GPU_COMPONENTS_CL_CLCOMPONENTMATMUL_H