//-----------------------------------------------------------------------
// -----------------------------------------------------------------------
// file="MatrixCompute"
// Project: StgSharp
// AuthorGroup: Nitload
// Copyright (c) Nitload. All rights reserved.
//     
// 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.
//     
// -----------------------------------------------------------------------
// -----------------------------------------------------------------------
using StgSharp.HighPerformance;
using StgSharp.Internal;
using System;
using System.Collections.Generic;
using System.Linq;
using System.Numerics;
using System.Runtime.CompilerServices;
using System.Text;
using System.Threading.Tasks;

namespace StgSharp.Mathematics.Numeric
{
    public partial class MatrixCompute
    {

        internal static unsafe void ComputeFmaPanelF32(
                                    MatrixParallelTaskPackage* p
        )
        {
            int f32 = (int)MatrixElementType.F32;
            int size = MatrixPanel.KernelSideCount<float>();
            MatrixKernel* leftBuffer = (MatrixKernel*)p->Left;
            MatrixKernel* rightBuffer = (MatrixKernel*)p->Right;
            MatrixKernel* ansBuffer = (MatrixKernel*)p->Answer;
            ref long offsetRef = ref Unsafe.As<int, long>(ref p->PrimTileOffset);
            ref long lengthRef = ref Unsafe.As<int, long>(ref p->PrimCount);
            int ansWidth = p->AnsPrimOffset;
            int ansHeight = p->AnsSecOffset;
            int commonK = p->LeftPrimOffset;
            int blocksPerRow = ansWidth / size;
            if (NativeIntrinsic.Context.mat[f32].build_panel == null)
            {
                // in this occasion size must be 1
                for (long c = offsetRef; c < offsetRef + lengthRef; c++)
                {
                    int i = (int)(c / blocksPerRow);
                    int j = (int)(c % blocksPerRow);
                    MatrixKernel<float>* ansPanel = GetKernelAddressUnsafe((MatrixKernel<float>*)ansBuffer,
                                                                           ansHeight, i, j);
                    NativeIntrinsic.Context.mat[f32].clear_panel((MatrixPanel*)ansPanel);
                    for (int k = 0; k < commonK; k += size) {
                        NativeIntrinsic.Context
                                       .mat[f32].panel_fma(
                                        (MatrixPanel*)GetKernelAddressUnsafe((MatrixKernel<float>*)leftBuffer,
                                                                             ansHeight, k, j),
                                        (MatrixPanel*)GetKernelAddressUnsafe((MatrixKernel<float>*)rightBuffer,
                                                                             commonK, j, k),
                                        (MatrixPanel*)ansPanel);
                    }
                }
            } else
            {
                MatrixPanel<float>* leftPanel = MatrixPanel.Create<float>();
                MatrixPanel<float>* rightPanel = MatrixPanel.Create<float>();
                MatrixPanel<float>* ansPanel = MatrixPanel.Create<float>();
                for (long c = offsetRef; c < offsetRef + lengthRef; c++)
                {
                    int i = (int)(c / blocksPerRow) * size;
                    int j = (int)(c % blocksPerRow) * size;
                    NativeIntrinsic.Context.mat[f32].clear_panel((MatrixPanel*)ansPanel);
                    for (int k = 0; k < commonK; k += size)
                    {
                        NativeIntrinsic.Context
                                       .mat[f32].build_panel((MatrixPanel*)leftPanel, leftBuffer,
                                                             commonK, ansHeight, k, i);
                        NativeIntrinsic.Context
                                       .mat[f32].build_panel((MatrixPanel*)rightPanel, rightBuffer,
                                                             ansWidth, commonK, j, k);
                        NativeIntrinsic.Context
                                       .mat[f32].panel_fma((MatrixPanel*)leftPanel, (MatrixPanel*)rightPanel, (MatrixPanel*)ansPanel);
                    }
                    NativeIntrinsic.Context
                                   .mat[f32].store_panel((MatrixPanel*)ansPanel, ansBuffer,
                                                         ansWidth, ansHeight, j, i);
                }
                MatrixPanel.Destroy(leftPanel);
                MatrixPanel.Destroy(rightPanel);
                MatrixPanel.Destroy(ansPanel);
            }
        }

        internal static unsafe void ComputeFmaPanelF64(
                                    MatrixParallelTaskPackage* p
        )
        {
            int f64 = (int)MatrixElementType.F64;
            int size = MatrixPanel.KernelSideCount<double>();
            MatrixKernel* leftBuffer = (MatrixKernel*)p->Left;
            MatrixKernel* rightBuffer = (MatrixKernel*)p->Right;
            MatrixKernel* ansBuffer = (MatrixKernel*)p->Answer;
            ref long offsetRef = ref Unsafe.As<int, long>(ref p->PrimTileOffset);
            ref long lengthRef = ref Unsafe.As<int, long>(ref p->PrimCount);
            int ansWidth = p->AnsPrimOffset;
            int ansHeight = p->AnsSecOffset;
            int commonK = p->LeftPrimOffset;
            int blocksPerRow = ansWidth / size;
            if (NativeIntrinsic.Context.mat[f64].build_panel == null)
            {
                // in this occasion size must be 1
                for (long c = offsetRef; c < offsetRef + lengthRef; c++)
                {
                    int i = (int)(c / blocksPerRow);
                    int j = (int)(c % blocksPerRow);
                    MatrixKernel<double>* ansPanel = GetKernelAddressUnsafe((MatrixKernel<double>*)ansBuffer,
                                                                            ansHeight, i, j);
                    NativeIntrinsic.Context.mat[f64].clear_panel((MatrixPanel*)ansPanel);
                    for (int k = 0; k < commonK; k += size) {
                        NativeIntrinsic.Context
                                       .mat[f64].panel_fma(
                                        (MatrixPanel*)GetKernelAddressUnsafe((MatrixKernel<double>*)leftBuffer,
                                                                             ansHeight, k, j),
                                        (MatrixPanel*)GetKernelAddressUnsafe((MatrixKernel<double>*)rightBuffer,
                                                                             commonK, j, k),
                                        (MatrixPanel*)ansPanel);
                    }
                }
            } else
            {
                MatrixPanel<double>* leftPanel = MatrixPanel.Create<double>();
                MatrixPanel<double>* rightPanel = MatrixPanel.Create<double>();
                MatrixPanel<double>* ansPanel = MatrixPanel.Create<double>();
                for (long c = offsetRef; c < offsetRef + lengthRef; c++)
                {
                    int i = (int)(c / blocksPerRow) * size;
                    int j = (int)(c % blocksPerRow) * size;
                    NativeIntrinsic.Context.mat[f64].clear_panel((MatrixPanel*)ansPanel);
                    for (int k = 0; k < commonK; k += size)
                    {
                        NativeIntrinsic.Context
                                       .mat[f64].build_panel((MatrixPanel*)leftPanel, leftBuffer,
                                                             commonK, ansHeight, k, i);
                        NativeIntrinsic.Context
                                       .mat[f64].build_panel((MatrixPanel*)rightPanel, rightBuffer,
                                                             ansWidth, commonK, j, k);
                        NativeIntrinsic.Context
                                       .mat[f64].panel_fma((MatrixPanel*)leftPanel, (MatrixPanel*)rightPanel, (MatrixPanel*)ansPanel);
                    }
                    NativeIntrinsic.Context
                                   .mat[f64].store_panel((MatrixPanel*)ansPanel, ansBuffer,
                                                         ansWidth, ansHeight, j, i);
                }
                MatrixPanel.Destroy(leftPanel);
                MatrixPanel.Destroy(rightPanel);
                MatrixPanel.Destroy(ansPanel);
            }
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        internal static unsafe MatrixKernel<T>* GetKernelAddressUnsafe<T>(
                                                MatrixKernel<T>* buffer,
                                                int colLength,
                                                int x,
                                                int y
        ) where T : unmanaged, INumber<T>
        {
            return buffer + (colLength * x + y);
        }

    }
}