using System;
using System.Collections.Generic;
using System.Linq;
using System.Runtime.CompilerServices;
using System.Text;
using System.Threading.Tasks;
using StgSharp.HighPerformance;
using StgSharp.Internal;

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

        internal static unsafe void ComputeFmaPanelF32(
                                    MatrixParallelTaskPackage* p
        )
        {
            int f32 = (int)MatrixElementType.F32;
            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;

            M128* enumeration = stackalloc M128[1];
            enumeration->Member<int>(2) = commonK;
            enumeration->Member<int>(3) = ansHeight;

            // in this occasion size must be 1
            for (long c = offsetRef; c < offsetRef + lengthRef; c++)
            {
                int i = (int)(c / ansHeight);
                int j = (int)(c % ansHeight);
                enumeration->Member<int>(0) = i;
                enumeration->Member<int>(1) = j;
                NativeIntrinsic.Context.mat[f32].kernel_tile_fma(leftBuffer, rightBuffer, ansBuffer,
                                                                 enumeration);
            }
        }

        internal static unsafe void ComputeFmaPanelF64(
                                    MatrixParallelTaskPackage* p
        )
        {
            int f64 = (int)MatrixElementType.F64;

            // int size = 1;
            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;

            M128* enumeration = stackalloc M128[1];
            enumeration->Member<int>(2) = commonK;
            enumeration->Member<int>(3) = ansHeight;

            // in this occasion size must be 1
            for (long c = offsetRef; c < offsetRef + lengthRef; c++)
            {
                int i = (int)(c / ansHeight);
                int j = (int)(c % ansHeight);
                enumeration->Member<int>(0) = i;
                enumeration->Member<int>(1) = j;
                NativeIntrinsic.Context.mat[f64].kernel_tile_fma(leftBuffer, rightBuffer, ansBuffer,
                                                                 enumeration);
            }
        }
    }
}