Add experimental PostOp interface to ClGemmMatrixMultiplyReshapedKernel Part 1

This interface supports the fusion of multiple elementwise operations

Partially resolves: COMPMID-4435

Change-Id: If68dd7dd98dcf239fde7cb1f0a4a6d4d1e899a6f
Signed-off-by: SiCongLi <sicong.li@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6483
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 85 insertions, 1 deletions

diff --git a/src/core/CL/CLUtils.h b/src/core/CL/CLUtils.h
index b65d547756..d133e4fe6f 100644
--- a/src/core/CL/CLUtils.h
+++ b/src/core/CL/CLUtils.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,10 +26,13 @@
 #define ARM_COMPUTE_CL_CLUTILS_H
 
 #include "arm_compute/core/CL/OpenCL.h"
+#include "arm_compute/core/experimental/IPostOp.h"
 
 namespace arm_compute
 {
 class TensorShape;
+class CLBuildOptions;
+class ITensorInfo;
 
 /** Create a cl::Image2D object from an OpenCL buffer
  *
@@ -51,6 +54,87 @@ class TensorShape;
  */
 cl::Image2D create_image2d_from_buffer(const cl::Context &ctx, const cl::Buffer &buffer, const TensorShape &shape2d, DataType data_type, size_t image_row_pitch);
 
+namespace experimental
+{
+/** @name (EXPERIMENTAL_POST_OPS)
+ * @{
+ */
+
+/** Manage validation, building and configurations of PostOp CL kernels */
+class PostOpCLKernelUtils final
+{
+public:
+    /** CL kernel name postfix for post ops */
+    using NamePostfix = std::string;
+    /** CL kernels that supports post ops assign each post op to a 'slot', in accordance with the postfix
+     * For example, for a kernel with postfix '_act_prelu_eltwiseadd', there are 3 slots
+     * slot 1: (unary) activation, slot 2: pRelu, slot 3: elementwise addition
+     *
+     * Some kernels may allow some slots to be optional, to support multiple combinations of post op sequences.
+     * In such cases, we need to explicitly set up a mapping between each post op and the slots for that kernel.
+     * For example, suppose we have 2 kernels with postfixes: _eltwiseadd_prelu, _act_eltwiseadd_act_prelu, where the activations in the
+     * second kernel are optional. Say we want to support an eltwise addition, followed by a prelu (sequence { eltwiseadd, prelu }).
+     * Now we can choose which one of the 2 kernels to use, since they both support this post op sequence.
+     * We can either:
+     *  1. assign the elementwise to slot 1 and prelu to slot 2 of kernel 1
+     *  { { Eltwise_Add, PRelu } -> {"_eltwise_act", {1, 2} } } or
+     *  2. assign the elementwise to slot 2 and prelu to slot 4 of kernel 1
+     *  { { Eltwise_Add, PRelu } -> {"_act_eltwiseadd_act_prelu", {2, 4} } }
+     */
+    using Slots  = std::vector<unsigned int>;
+    using Config = std::map<PostOpTypeSequence, std::tuple<NamePostfix, Slots>>;
+
+public:
+    explicit PostOpCLKernelUtils(const Config &config);
+
+    /** Check if post op argument tensor shapes are compliant
+     * All post ops must not alter the shape of the original dst tensor (even after broadcasting)
+     *
+     * @param[in] dst      Dst tensor to apply the post ops to
+     * @param[in] post_ops Post ops
+     *
+     * @return true if shapes are compliant and false otherwise
+     */
+    static bool are_post_op_shapes_compliant(const ITensorInfo *dst, const experimental::PostOpList<ITensorInfo *> &post_ops);
+    /** Check if the post op sequence is supported in the current configuration
+     *
+     * @param[in] post_ops Post ops
+     *
+     * @return true if the post op sequence is supported and false otherwise
+     */
+    bool is_post_op_sequence_supported(const PostOpList<ITensorInfo *> &post_ops) const;
+    /** Helper function to set PostOp related build options
+     * @note Convention
+     *      1. Each post op "slot" is prefixed with "P<slot number>", followed by the usual parameters for that post op.
+     *      E.g. If the first slot is an activation, we need to pass 3 definitions in this way:
+     *          -P1_ACTIVATION_TYPE=...  -P1_ACTIVATION_A_VAL=...   -P1_ACTIVATION_B_VAL=...
+     *
+     *      2. For multi-ary post ops, to pass the position of the previous op's dest tensor,
+     *         we append "_X_POS_<pos>" to the post op type.
+     *      E.g. for a single post op add(dst, x), where dst is the result of the main op.
+     *         In this case, the position of the previous op's dest is 0, so we pass
+     *         -P1_ELTWISE_OP=ADD_X_POS_0
+     *
+     * @param[out] built_opts OpenCL kernel build options
+     * @param[in]  post_ops   Post ops
+     *
+     */
+    void set_post_ops_cl_build_options(CLBuildOptions &built_opts, const PostOpList<ITensorInfo *> &post_ops) const;
+    /** Helper function to set PostOp kernel name
+     *
+     * @param[out] kernel_name OpenCL kernel name
+     * @param[in]  post_ops    Post ops
+     *
+     */
+    void set_post_ops_cl_kernel_name(std::string &kernel_name, const PostOpList<ITensorInfo *> &post_ops) const;
+
+private:
+    Config _supported_config{};
+};
+/** @} */ // end of group (EXPERIMENTAL_POST_OPS)
+
+} // namespace experimental
+
 } // arm_compute
 
 #endif /* ARM_COMPUTE_CL_CLUTILS_H */