COMPMID-1405: Create our own gemm_native kernel / function.

Change-Id: Ie0a80bd6b4eb5632cac63ccf54bcb07d4309da19
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/140305
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

3 files changed, 89 insertions, 5 deletions

diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 6a3fabca67..fdb1450179 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -104,6 +104,7 @@
 #include "arm_compute/runtime/NEON/functions/NEReshapeLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEScale.h"
 #include "arm_compute/runtime/NEON/functions/NEScharr3x3.h"
+#include "arm_compute/runtime/NEON/functions/NESimpleAssemblyFunction.h"
 #include "arm_compute/runtime/NEON/functions/NESobel3x3.h"
 #include "arm_compute/runtime/NEON/functions/NESobel5x5.h"
 #include "arm_compute/runtime/NEON/functions/NESobel7x7.h"
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h b/arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h
index 4ac6a3cae2..1c9ecb088e 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMAssemblyDispatch.h
@@ -54,6 +54,20 @@ private:
     /** ACL Function */
     std::unique_ptr<IFunction> _function;
 
+    /** If supported create the ACL function corresponding to the GemmMethod provided to process the other passed parameters
+     *
+     * @param[in]  method             GemmMethod to use to perform the matrix multiplication.
+     * @param[in]  a                  Input tensor (Matrix A).
+     * @param[in]  b                  Input tensor (Matrix B).
+     * @param[out] d                  Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0.
+     * @param[in]  alpha              Scalar multiplier to apply to AB matrix product.
+     * @param[in]  beta               Scalar multiplier to apply to input D matrix before adding product.
+     * @param[in]  pretransposed_hint Can the B tensor can be pretransposed (ie shared across invocations)?
+     *
+     * @return True if the method is supported and the function was successfully created, false otherwise.
+     */
+    bool create_function(arm_gemm::GemmMethod method, const ITensor *a, const ITensor *b, ITensor *d, float alpha, float beta, bool pretranspose_hint);
+
     //Fallback: use arm_gemm's AssemblyGemm:
     class Fallback
     {
@@ -63,7 +77,7 @@ private:
          *  The call to set_arrays is needed to deal with the input sizes containing batches (dims > 2)
          */
         void run();
-        void configure(const ITensor *a, const ITensor *b, ITensor *d, float alpha, float beta, bool pretranspose_hint, MemoryGroup &memory_group);
+        void configure(const ITensor *a, const ITensor *b, ITensor *d, arm_gemm::GemmArgs<TypeOutput> &args, MemoryGroup &memory_group);
         void prepare();
         bool is_configured() const;
 #endif /* DOXYGEN_SKIP_THIS */
@@ -102,7 +116,20 @@ private:
     } _arm_gemm;               /**< Fallback in case ACL doesn't have a function */
     MemoryGroup _memory_group; /**< Function memory group */
 public:
+    /** If supported create an ACL function else fallback to the arm_gemm function.
+     *
+     * @param[in]  a                 Input tensor (Matrix A)
+     * @param[in]  b                 Input tensor (Matrix B)
+     * @param[out] d                 Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0.
+     * @param[in]  alpha             Scalar multiplier to apply to AB matrix product.
+     * @param[in]  beta              Scalar multiplier to apply to input D matrix before adding product.
+     * @param[in]  pretranspose_hint Can the B tensor can be pretransposed (ie shared across invocations)?
+     */
     void configure(const ITensor *a, const ITensor *b, ITensor *d, float alpha, float beta, bool pretranspose_hint);
+    /** Was the function successfully configured ?
+     *
+     * @return True if the function is configured and ready to run
+     */
     bool is_configured() const;
     // Inherited methods overridden:
     /** Runs a preparation step, usually for pre-transposing matrix b */
@@ -110,11 +137,11 @@ public:
     void run() override;
 };
 
-/** Float 32 assembly kernel glue */
+/** Float 32 assembly dispatch kernel */
 using NEGEMMAssemblyDispatchF32 = NEGEMMAssemblyDispatch<float, float>;
-/** Uint 8 to Uint 32 kernel glue */
+/** Uint 8 to Uint 32 assembly dispatch kernel */
 using NEGEMMAssemblyDispatchU8U32 = NEGEMMAssemblyDispatch<uint8_t, uint32_t>;
-/** Int 8 to Int 32 kernel glue */
+/** Int 8 to Int 32 assembly dispatch kernel */
 using NEGEMMAssemblyDispatchS8S32 = NEGEMMAssemblyDispatch<int8_t, int32_t>;
-}
+} // namespace arm_compute
 #endif /* __ARM_COMPUTE_NEGEMMASSEMBLYDISPATCH_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NESimpleAssemblyFunction.h b/arm_compute/runtime/NEON/functions/NESimpleAssemblyFunction.h
new file mode 100644
index 0000000000..5a76c15627
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NESimpleAssemblyFunction.h
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 2018 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 __ARM_COMPUTE_NESIMPLEASSEMBLYFUNCTION_H__
+#define __ARM_COMPUTE_NESIMPLEASSEMBLYFUNCTION_H__
+
+#include "arm_compute/core/NEON/kernels/assembly/INEGEMMWrapperKernel.h"
+#include "arm_compute/runtime/IFunction.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+/** Basic interface for functions which have a single NEON GEMM wrapper kernel to run */
+class NESimpleAssemblyFunction : public IFunction
+{
+public:
+    /** Constructor */
+    NESimpleAssemblyFunction();
+
+    /** Configure the function with the kernel to run
+     *
+     * @param[in] kernel GEMM Wrapper kernel configured and ready to run
+     *
+     * @note The kernel is expected to have a 1D window. The function will multi-thread this window across the X dimension.
+     */
+    void configure(std::unique_ptr<INEGEMMWrapperKernel> kernel);
+
+    // Inherited methods overridden:
+    void run() override final;
+
+protected:
+    std::unique_ptr<INEGEMMWrapperKernel> _kernel; /**< Kernel to run */
+};
+} //namespace arm_compute
+#endif /*__ARM_COMPUTE_NESIMPLEASSEMBLYFUNCTION_H__ */