path: root/src/GPU.cpp

/*
    Copyright 2016-2023 melonDS team

    This file is part of melonDS.

    melonDS is free software: you can redistribute it and/or modify it under
    the terms of the GNU General Public License as published by the Free
    Software Foundation, either version 3 of the License, or (at your option)
    any later version.

    melonDS is distributed in the hope that it will be useful, but WITHOUT ANY
    WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
    FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

    You should have received a copy of the GNU General Public License along
    with melonDS. If not, see http://www.gnu.org/licenses/.
*/

#include <string.h>
#include "NDS.h"
#include "GPU.h"

#include "ARMJIT.h"

#include "GPU2D_Soft.h"
#include "GPU3D_Soft.h"

namespace melonDS
{
using Platform::Log;
using Platform::LogLevel;

#define LINE_CYCLES  (355*6)
#define HBLANK_CYCLES (48+(256*6))
#define FRAME_CYCLES  (LINE_CYCLES * 263)

enum
{
    LCD_StartHBlank = 0,
    LCD_StartScanline,
    LCD_FinishFrame,
};


/*
    VRAM invalidation tracking

    - we want to know when a VRAM region used for graphics changed
    - for some regions unmapping is mandatory to modify them (Texture, TexPal and ExtPal) and
        we don't want to completely invalidate them every time they're unmapped and remapped

    For this reason we don't track the dirtyness per mapping region, but instead per VRAM bank
    with VRAMDirty.

    This is more or less a description of VRAMTrackingSet::DeriveState
        Each time before the memory is read two things could have happened
        to each 16kb piece (16kb is the smallest unit in which mappings can
        be made thus also the size VRAMMap_* use):
            - this piece was remapped compared to last time we checked,
                which means this location in memory is invalid.
            - this piece wasn't remapped, which means we need to check whether
                it was changed. This can be archived by checking VRAMDirty.
                VRAMDirty need to be reset for the respective VRAM bank.
*/

GPU::GPU(melonDS::NDS& nds, std::unique_ptr<Renderer3D>&& renderer3d, std::unique_ptr<GPU2D::Renderer2D>&& renderer2d) noexcept :
    NDS(nds),
    GPU2D_A(0, *this),
    GPU2D_B(1, *this),
    GPU3D(nds, renderer3d ? std::move(renderer3d) : std::make_unique<SoftRenderer>(*this)),
    GPU2D_Renderer(renderer2d ? std::move(renderer2d) : std::make_unique<GPU2D::SoftRenderer>(*this))
{
    NDS.RegisterEventFunc(Event_LCD, LCD_StartHBlank, MemberEventFunc(GPU, StartHBlank));
    NDS.RegisterEventFunc(Event_LCD, LCD_StartScanline, MemberEventFunc(GPU, StartScanline));
    NDS.RegisterEventFunc(Event_LCD, LCD_FinishFrame, MemberEventFunc(GPU, FinishFrame));
    NDS.RegisterEventFunc(Event_DisplayFIFO, 0, MemberEventFunc(GPU, DisplayFIFO));

    FrontBuffer = 0;
}

GPU::~GPU() noexcept
{
    // All unique_ptr fields are automatically cleaned up

    NDS.UnregisterEventFunc(Event_LCD, LCD_StartHBlank);
    NDS.UnregisterEventFunc(Event_LCD, LCD_StartScanline);
    NDS.UnregisterEventFunc(Event_LCD, LCD_FinishFrame);
    NDS.UnregisterEventFunc(Event_DisplayFIFO, 0);
}

void GPU::ResetVRAMCache() noexcept
{
    for (int i = 0; i < 9; i++)
        VRAMDirty[i] = NonStupidBitField<128*1024/VRAMDirtyGranularity>();

    VRAMDirty_ABG.Reset();
    VRAMDirty_BBG.Reset();
    VRAMDirty_AOBJ.Reset();
    VRAMDirty_BOBJ.Reset();
    VRAMDirty_ABGExtPal.Reset();
    VRAMDirty_BBGExtPal.Reset();
    VRAMDirty_AOBJExtPal.Reset();
    VRAMDirty_BOBJExtPal.Reset();
    VRAMDirty_Texture.Reset();
    VRAMDirty_TexPal.Reset();

    memset(VRAMFlat_ABG, 0, sizeof(VRAMFlat_ABG));
    memset(VRAMFlat_BBG, 0, sizeof(VRAMFlat_BBG));
    memset(VRAMFlat_AOBJ, 0, sizeof(VRAMFlat_AOBJ));
    memset(VRAMFlat_BOBJ, 0, sizeof(VRAMFlat_BOBJ));
    memset(VRAMFlat_ABGExtPal, 0, sizeof(VRAMFlat_ABGExtPal));
    memset(VRAMFlat_BBGExtPal, 0, sizeof(VRAMFlat_BBGExtPal));
    memset(VRAMFlat_AOBJExtPal, 0, sizeof(VRAMFlat_AOBJExtPal));
    memset(VRAMFlat_BOBJExtPal, 0, sizeof(VRAMFlat_BOBJExtPal));
    memset(VRAMFlat_Texture, 0, sizeof(VRAMFlat_Texture));
    memset(VRAMFlat_TexPal, 0, sizeof(VRAMFlat_TexPal));
}

void GPU::Reset() noexcept
{
    VCount = 0;
    NextVCount = -1;
    TotalScanlines = 0;

    DispStat[0] = 0;
    DispStat[1] = 0;
    VMatch[0] = 0;
    VMatch[1] = 0;

    memset(Palette, 0, 2*1024);
    memset(OAM, 0, 2*1024);

    memset(VRAM_A, 0, 128*1024);
    memset(VRAM_B, 0, 128*1024);
    memset(VRAM_C, 0, 128*1024);
    memset(VRAM_D, 0, 128*1024);
    memset(VRAM_E, 0,  64*1024);
    memset(VRAM_F, 0,  16*1024);
    memset(VRAM_G, 0,  16*1024);
    memset(VRAM_H, 0,  32*1024);
    memset(VRAM_I, 0,  16*1024);

    memset(VRAMCNT, 0, 9);
    VRAMSTAT = 0;

    VRAMMap_LCDC = 0;

    memset(VRAMMap_ABG, 0, sizeof(VRAMMap_ABG));
    memset(VRAMMap_AOBJ, 0, sizeof(VRAMMap_AOBJ));
    memset(VRAMMap_BBG, 0, sizeof(VRAMMap_BBG));
    memset(VRAMMap_BOBJ, 0, sizeof(VRAMMap_BOBJ));

    memset(VRAMMap_ABGExtPal, 0, sizeof(VRAMMap_ABGExtPal));
    VRAMMap_AOBJExtPal = 0;
    memset(VRAMMap_BBGExtPal, 0, sizeof(VRAMMap_BBGExtPal));
    VRAMMap_BOBJExtPal = 0;

    memset(VRAMMap_Texture, 0, sizeof(VRAMMap_Texture));
    memset(VRAMMap_TexPal, 0, sizeof(VRAMMap_TexPal));

    VRAMMap_ARM7[0] = 0;
    VRAMMap_ARM7[1] = 0;

    memset(VRAMPtr_ABG, 0, sizeof(VRAMPtr_ABG));
    memset(VRAMPtr_AOBJ, 0, sizeof(VRAMPtr_AOBJ));
    memset(VRAMPtr_BBG, 0, sizeof(VRAMPtr_BBG));
    memset(VRAMPtr_BOBJ, 0, sizeof(VRAMPtr_BOBJ));

    size_t fbsize;
    if (GPU3D.IsRendererAccelerated())
        fbsize = (256*3 + 1) * 192;
    else
        fbsize = 256 * 192;

    for (size_t i = 0; i < fbsize; i++)
    {
        Framebuffer[0][0][i] = 0xFFFFFFFF;
        Framebuffer[1][0][i] = 0xFFFFFFFF;
    }
    for (size_t i = 0; i < fbsize; i++)
    {
        Framebuffer[0][1][i] = 0xFFFFFFFF;
        Framebuffer[1][1][i] = 0xFFFFFFFF;
    }

    GPU2D_A.Reset();
    GPU2D_B.Reset();
    GPU3D.Reset();

    int backbuf = FrontBuffer ? 0 : 1;
    GPU2D_Renderer->SetFramebuffer(Framebuffer[backbuf][1].get(), Framebuffer[backbuf][0].get());

    ResetVRAMCache();

    OAMDirty = 0x3;
    PaletteDirty = 0xF;
}

void GPU::Stop() noexcept
{
    int fbsize;
    if (GPU3D.IsRendererAccelerated())
        fbsize = (256*3 + 1) * 192;
    else
        fbsize = 256 * 192;

    memset(Framebuffer[0][0].get(), 0, fbsize*4);
    memset(Framebuffer[0][1].get(), 0, fbsize*4);
    memset(Framebuffer[1][0].get(), 0, fbsize*4);
    memset(Framebuffer[1][1].get(), 0, fbsize*4);

    GPU3D.Stop();
}

void GPU::DoSavestate(Savestate* file) noexcept
{
    file->Section("GPUG");

    file->Var16(&VCount);
    file->Var32(&NextVCount);
    file->Var16(&TotalScanlines);

    file->Var16(&DispStat[0]);
    file->Var16(&DispStat[1]);
    file->Var16(&VMatch[0]);
    file->Var16(&VMatch[1]);

    file->VarArray(Palette, 2*1024);
    file->VarArray(OAM, 2*1024);

    file->VarArray(VRAM_A, 128*1024);
    file->VarArray(VRAM_B, 128*1024);
    file->VarArray(VRAM_C, 128*1024);
    file->VarArray(VRAM_D, 128*1024);
    file->VarArray(VRAM_E,  64*1024);
    file->VarArray(VRAM_F,  16*1024);
    file->VarArray(VRAM_G,  16*1024);
    file->VarArray(VRAM_H,  32*1024);
    file->VarArray(VRAM_I,  16*1024);

    file->VarArray(VRAMCNT, 9);
    file->Var8(&VRAMSTAT);

    file->Var32(&VRAMMap_LCDC);

    file->VarArray(VRAMMap_ABG, sizeof(VRAMMap_ABG));
    file->VarArray(VRAMMap_AOBJ, sizeof(VRAMMap_AOBJ));
    file->VarArray(VRAMMap_BBG, sizeof(VRAMMap_BBG));
    file->VarArray(VRAMMap_BOBJ, sizeof(VRAMMap_BOBJ));

    file->VarArray(VRAMMap_ABGExtPal, sizeof(VRAMMap_ABGExtPal));
    file->Var32(&VRAMMap_AOBJExtPal);
    file->VarArray(VRAMMap_BBGExtPal, sizeof(VRAMMap_BBGExtPal));
    file->Var32(&VRAMMap_BOBJExtPal);

    file->VarArray(VRAMMap_Texture, sizeof(VRAMMap_Texture));
    file->VarArray(VRAMMap_TexPal, sizeof(VRAMMap_TexPal));

    file->Var32(&VRAMMap_ARM7[0]);
    file->Var32(&VRAMMap_ARM7[1]);

    if (!file->Saving)
    {
        for (int i = 0; i < 0x20; i++)
            VRAMPtr_ABG[i] = GetUniqueBankPtr(VRAMMap_ABG[i], i << 14);
        for (int i = 0; i < 0x10; i++)
            VRAMPtr_AOBJ[i] = GetUniqueBankPtr(VRAMMap_AOBJ[i], i << 14);
        for (int i = 0; i < 0x8; i++)
            VRAMPtr_BBG[i] = GetUniqueBankPtr(VRAMMap_BBG[i], i << 14);
        for (int i = 0; i < 0x8; i++)
            VRAMPtr_BOBJ[i] = GetUniqueBankPtr(VRAMMap_BOBJ[i], i << 14);
    }

    GPU2D_A.DoSavestate(file);
    GPU2D_B.DoSavestate(file);
    GPU3D.DoSavestate(file);

    ResetVRAMCache();
}

void GPU::AssignFramebuffers() noexcept
{
    int backbuf = FrontBuffer ? 0 : 1;
    if (NDS.PowerControl9 & (1<<15))
    {
        GPU2D_Renderer->SetFramebuffer(Framebuffer[backbuf][0].get(), Framebuffer[backbuf][1].get());
    }
    else
    {
        GPU2D_Renderer->SetFramebuffer(Framebuffer[backbuf][1].get(), Framebuffer[backbuf][0].get());
    }
}

void GPU::SetRenderer3D(std::unique_ptr<Renderer3D>&& renderer) noexcept
{
    if (renderer == nullptr)
        GPU3D.SetCurrentRenderer(std::make_unique<SoftRenderer>(*this));
    else
        GPU3D.SetCurrentRenderer(std::move(renderer));

    int fbsize;
    if (GPU3D.IsRendererAccelerated())
        fbsize = (256*3 + 1) * 192;
    else
        fbsize = 256 * 192;

    Framebuffer[0][0] = std::make_unique<u32[]>(fbsize);
    Framebuffer[1][0] = std::make_unique<u32[]>(fbsize);
    Framebuffer[0][1] = std::make_unique<u32[]>(fbsize);
    Framebuffer[1][1] = std::make_unique<u32[]>(fbsize);

    memset(Framebuffer[0][0].get(), 0, fbsize*4);
    memset(Framebuffer[1][0].get(), 0, fbsize*4);
    memset(Framebuffer[0][1].get(), 0, fbsize*4);
    memset(Framebuffer[1][1].get(), 0, fbsize*4);

    AssignFramebuffers();
}


// VRAM mapping notes
//
// mirroring:
// unmapped range reads zero
// LCD is mirrored every 0x100000 bytes, the gap between each mirror reads zero
// ABG:
//   bank A,B,C,D,E mirror every 0x80000 bytes
//   bank F,G mirror at base+0x8000, mirror every 0x80000 bytes
// AOBJ:
//   bank A,B,E mirror every 0x40000 bytes
//   bank F,G mirror at base+0x8000, mirror every 0x40000 bytes
// BBG:
//   bank C mirrors every 0x20000 bytes
//   bank H mirrors every 0x10000 bytes
//   bank I mirrors at base+0x4000, mirrors every 0x10000 bytes
// BOBJ:
//   bank D mirrors every 0x20000 bytes
//   bank I mirrors every 0x4000 bytes
//
// untested:
//   ARM7 (TODO)
//   extended palette (mirroring doesn't apply)
//   texture/texpal (does mirroring apply?)
//   -> trying to use extpal/texture/texpal with no VRAM mapped.
//      would likely read all black, but has to be tested.
//
// overlap:
// when reading: values are read from each bank and ORed together
// when writing: value is written to each bank

u8* GPU::GetUniqueBankPtr(u32 mask, u32 offset) noexcept
{
    if (!mask || (mask & (mask - 1)) != 0) return NULL;
    int num = __builtin_ctz(mask);
    return &VRAM[num][offset & VRAMMask[num]];
}

const u8* GPU::GetUniqueBankPtr(u32 mask, u32 offset) const noexcept
{
    if (!mask || (mask & (mask - 1)) != 0) return NULL;
    int num = __builtin_ctz(mask);
    return &VRAM[num][offset & VRAMMask[num]];
}

#define MAP_RANGE(map, base, n)    for (int i = 0; i < n; i++) VRAMMap_##map[(base)+i] |= bankmask;
#define UNMAP_RANGE(map, base, n)  for (int i = 0; i < n; i++) VRAMMap_##map[(base)+i] &= ~bankmask;

#define MAP_RANGE_PTR(map, base, n) \
    for (int i = 0; i < n; i++) { VRAMMap_##map[(base)+i] |= bankmask; VRAMPtr_##map[(base)+i] = GetUniqueBankPtr(VRAMMap_##map[(base)+i], ((base)+i)<<14); }
#define UNMAP_RANGE_PTR(map, base, n) \
    for (int i = 0; i < n; i++) { VRAMMap_##map[(base)+i] &= ~bankmask; VRAMPtr_##map[(base)+i] = GetUniqueBankPtr(VRAMMap_##map[(base)+i], ((base)+i)<<14); }

void GPU::MapVRAM_AB(u32 bank, u8 cnt) noexcept
{
    cnt &= 0x9B;

    u8 oldcnt = VRAMCNT[bank];
    VRAMCNT[bank] = cnt;

    if (oldcnt == cnt) return;

    u8 oldofs = (oldcnt >> 3) & 0x3;
    u8 ofs = (cnt >> 3) & 0x3;
    u32 bankmask = 1 << bank;

    if (oldcnt & (1<<7))
    {
        switch (oldcnt & 0x3)
        {
        case 0: // LCDC
            VRAMMap_LCDC &= ~bankmask;
            break;

        case 1: // ABG
            UNMAP_RANGE_PTR(ABG, oldofs<<3, 8);
            break;

        case 2: // AOBJ
            oldofs &= 0x1;
            UNMAP_RANGE_PTR(AOBJ, oldofs<<3, 8);
            break;

        case 3: // texture
            VRAMMap_Texture[oldofs] &= ~bankmask;
            break;
        }
    }

    if (cnt & (1<<7))
    {
        switch (cnt & 0x3)
        {
        case 0: // LCDC
            VRAMMap_LCDC |= bankmask;
            break;

        case 1: // ABG
            MAP_RANGE_PTR(ABG, ofs<<3, 8);
            break;

        case 2: // AOBJ
            ofs &= 0x1;
            MAP_RANGE_PTR(AOBJ, ofs<<3, 8);
            break;

        case 3: // texture
            VRAMMap_Texture[ofs] |= bankmask;
            break;
        }
    }
}

void GPU::MapVRAM_CD(u32 bank, u8 cnt) noexcept
{
    cnt &= 0x9F;

    u8 oldcnt = VRAMCNT[bank];
    VRAMCNT[bank] = cnt;

    VRAMSTAT &= ~(1 << (bank-2));

    if (oldcnt == cnt) return;

    u8 oldofs = (oldcnt >> 3) & 0x7;
    u8 ofs = (cnt >> 3) & 0x7;
    u32 bankmask = 1 << bank;

    if (oldcnt & (1<<7))
    {
        switch (oldcnt & 0x7)
        {
        case 0: // LCDC
            VRAMMap_LCDC &= ~bankmask;
            break;

        case 1: // ABG
            UNMAP_RANGE_PTR(ABG, oldofs<<3, 8);
            break;

        case 2: // ARM7 VRAM
            oldofs &= 0x1;
            VRAMMap_ARM7[oldofs] &= ~bankmask;
            break;

        case 3: // texture
            VRAMMap_Texture[oldofs] &= ~bankmask;
            break;

        case 4: // BBG/BOBJ
            if (bank == 2)
            {
                UNMAP_RANGE_PTR(BBG, 0, 8);
            }
            else
            {
                UNMAP_RANGE_PTR(BOBJ, 0, 8);
            }
            break;
        }
    }

    if (cnt & (1<<7))
    {
        switch (cnt & 0x7)
        {
        case 0: // LCDC
            VRAMMap_LCDC |= bankmask;
            break;

        case 1: // ABG
            MAP_RANGE_PTR(ABG, ofs<<3, 8);
            break;

        case 2: // ARM7 VRAM
            ofs &= 0x1;
            VRAMMap_ARM7[ofs] |= bankmask;
            memset(VRAMDirty[bank].Data, 0xFF, sizeof(VRAMDirty[bank].Data));
            VRAMSTAT |= (1 << (bank-2));
            NDS.JIT.CheckAndInvalidateWVRAM(ofs);
            break;

        case 3: // texture
            VRAMMap_Texture[ofs] |= bankmask;
            break;

        case 4: // BBG/BOBJ
            if (bank == 2)
            {
                MAP_RANGE_PTR(BBG, 0, 8);
            }
            else
            {
                MAP_RANGE_PTR(BOBJ, 0, 8);
            }
            break;
        }
    }
}

void GPU::MapVRAM_E(u32 bank, u8 cnt) noexcept
{
    cnt &= 0x87;

    u8 oldcnt = VRAMCNT[bank];
    VRAMCNT[bank] = cnt;

    if (oldcnt == cnt) return;

    u32 bankmask = 1 << bank;

    if (oldcnt & (1<<7))
    {
        switch (oldcnt & 0x7)
        {
        case 0: // LCDC
            VRAMMap_LCDC &= ~bankmask;
            break;

        case 1: // ABG
            UNMAP_RANGE_PTR(ABG, 0, 4);
            break;

        case 2: // AOBJ
            UNMAP_RANGE_PTR(AOBJ, 0, 4);
            break;

        case 3: // texture palette
            UNMAP_RANGE(TexPal, 0, 4);
            break;

        case 4: // ABG ext palette
            UNMAP_RANGE(ABGExtPal, 0, 4);
            break;
        }
    }

    if (cnt & (1<<7))
    {
        switch (cnt & 0x7)
        {
        case 0: // LCDC
            VRAMMap_LCDC |= bankmask;
            break;

        case 1: // ABG
            MAP_RANGE_PTR(ABG, 0, 4);
            break;

        case 2: // AOBJ
            MAP_RANGE_PTR(AOBJ, 0, 4);
            break;

        case 3: // texture palette
            MAP_RANGE(TexPal, 0, 4);
            break;

        case 4: // ABG ext palette
            MAP_RANGE(ABGExtPal, 0, 4);
            break;
        }
    }
}

void GPU::MapVRAM_FG(u32 bank, u8 cnt) noexcept
{
    cnt &= 0x9F;

    u8 oldcnt = VRAMCNT[bank];
    VRAMCNT[bank] = cnt;

    if (oldcnt == cnt) return;

    u8 oldofs = (oldcnt >> 3) & 0x7;
    u8 ofs = (cnt >> 3) & 0x7;
    u32 bankmask = 1 << bank;

    if (oldcnt & (1<<7))
    {
        switch (oldcnt & 0x7)
        {
        case 0: // LCDC
            VRAMMap_LCDC &= ~bankmask;
            break;

        case 1: // ABG
            {
                u32 base = (oldofs & 0x1) + ((oldofs & 0x2) << 1);
                VRAMMap_ABG[base] &= ~bankmask;
                VRAMMap_ABG[base + 2] &= ~bankmask;
                VRAMPtr_ABG[base] = GetUniqueBankPtr(VRAMMap_ABG[base], base << 14);
                VRAMPtr_ABG[base + 2] = GetUniqueBankPtr(VRAMMap_ABG[base + 2], (base + 2) << 14);
            }
            break;

        case 2: // AOBJ
            {
                u32 base = (oldofs & 0x1) + ((oldofs & 0x2) << 1);
                VRAMMap_AOBJ[base] &= ~bankmask;
                VRAMMap_AOBJ[base + 2] &= ~bankmask;
                VRAMPtr_AOBJ[base] = GetUniqueBankPtr(VRAMMap_AOBJ[base], base << 14);
                VRAMPtr_AOBJ[base + 2] = GetUniqueBankPtr(VRAMMap_AOBJ[base + 2], (base + 2) << 14);
            }
            break;

        case 3: // texture palette
            VRAMMap_TexPal[(oldofs & 0x1) + ((oldofs & 0x2) << 1)] &= ~bankmask;
            break;

        case 4: // ABG ext palette
            VRAMMap_ABGExtPal[((oldofs & 0x1) << 1)] &= ~bankmask;
            VRAMMap_ABGExtPal[((oldofs & 0x1) << 1) + 1] &= ~bankmask;
            break;

        case 5: // AOBJ ext palette
            VRAMMap_AOBJExtPal &= ~bankmask;
            break;
        }
    }

    if (cnt & (1<<7))
    {
        switch (cnt & 0x7)
        {
        case 0: // LCDC
            VRAMMap_LCDC |= bankmask;
            break;

        case 1: // ABG
            {
                u32 base = (ofs & 0x1) + ((ofs & 0x2) << 1);
                VRAMMap_ABG[base] |= bankmask;
                VRAMMap_ABG[base + 2] |= bankmask;
                VRAMPtr_ABG[base] = GetUniqueBankPtr(VRAMMap_ABG[base], base << 14);
                VRAMPtr_ABG[base + 2] = GetUniqueBankPtr(VRAMMap_ABG[base + 2], (base + 2) << 14);
            }
            break;

        case 2: // AOBJ
            {
                u32 base = (ofs & 0x1) + ((ofs & 0x2) << 1);
                VRAMMap_AOBJ[base] |= bankmask;
                VRAMMap_AOBJ[base + 2] |= bankmask;
                VRAMPtr_AOBJ[base] = GetUniqueBankPtr(VRAMMap_AOBJ[base], base << 14);
                VRAMPtr_AOBJ[base + 2] = GetUniqueBankPtr(VRAMMap_AOBJ[base + 2], (base + 2) << 14);
            }
            break;

        case 3: // texture palette
            VRAMMap_TexPal[(ofs & 0x1) + ((ofs & 0x2) << 1)] |= bankmask;
            break;

        case 4: // ABG ext palette
            VRAMMap_ABGExtPal[((ofs & 0x1) << 1)] |= bankmask;
            VRAMMap_ABGExtPal[((ofs & 0x1) << 1) + 1] |= bankmask;
            break;

        case 5: // AOBJ ext palette
            VRAMMap_AOBJExtPal |= bankmask;
            break;
        }
    }
}

void GPU::MapVRAM_H(u32 bank, u8 cnt) noexcept
{
    cnt &= 0x83;

    u8 oldcnt = VRAMCNT[bank];
    VRAMCNT[bank] = cnt;

    if (oldcnt == cnt) return;

    u32 bankmask = 1 << bank;

    if (oldcnt & (1<<7))
    {
        switch (oldcnt & 0x3)
        {
        case 0: // LCDC
            VRAMMap_LCDC &= ~bankmask;
            break;

        case 1: // BBG
            VRAMMap_BBG[0] &= ~bankmask;
            VRAMMap_BBG[1] &= ~bankmask;
            VRAMMap_BBG[4] &= ~bankmask;
            VRAMMap_BBG[5] &= ~bankmask;
            VRAMPtr_BBG[0] = GetUniqueBankPtr(VRAMMap_BBG[0], 0 << 14);
            VRAMPtr_BBG[1] = GetUniqueBankPtr(VRAMMap_BBG[1], 1 << 14);
            VRAMPtr_BBG[4] = GetUniqueBankPtr(VRAMMap_BBG[4], 4 << 14);
            VRAMPtr_BBG[5] = GetUniqueBankPtr(VRAMMap_BBG[5], 5 << 14);
            break;

        case 2: // BBG ext palette
            UNMAP_RANGE(BBGExtPal, 0, 4);
            break;
        }
    }

    if (cnt & (1<<7))
    {
        switch (cnt & 0x3)
        {
        case 0: // LCDC
            VRAMMap_LCDC |= bankmask;
            break;

        case 1: // BBG
            VRAMMap_BBG[0] |= bankmask;
            VRAMMap_BBG[1] |= bankmask;
            VRAMMap_BBG[4] |= bankmask;
            VRAMMap_BBG[5] |= bankmask;
            VRAMPtr_BBG[0] = GetUniqueBankPtr(VRAMMap_BBG[0], 0 << 14);
            VRAMPtr_BBG[1] = GetUniqueBankPtr(VRAMMap_BBG[1], 1 << 14);
            VRAMPtr_BBG[4] = GetUniqueBankPtr(VRAMMap_BBG[4], 4 << 14);
            VRAMPtr_BBG[5] = GetUniqueBankPtr(VRAMMap_BBG[5], 5 << 14);
            break;

        case 2: // BBG ext palette
            MAP_RANGE(BBGExtPal, 0, 4);
            break;
        }
    }
}

void GPU::MapVRAM_I(u32 bank, u8 cnt) noexcept
{
    cnt &= 0x83;

    u8 oldcnt = VRAMCNT[bank];
    VRAMCNT[bank] = cnt;

    if (oldcnt == cnt) return;

    u32 bankmask = 1 << bank;

    if (oldcnt & (1<<7))
    {
        switch (oldcnt & 0x3)
        {
        case 0: // LCDC
            VRAMMap_LCDC &= ~bankmask;
            break;

        case 1: // BBG
            VRAMMap_BBG[2] &= ~bankmask;
            VRAMMap_BBG[3] &= ~bankmask;
            VRAMMap_BBG[6] &= ~bankmask;
            VRAMMap_BBG[7] &= ~bankmask;
            VRAMPtr_BBG[2] = GetUniqueBankPtr(VRAMMap_BBG[2], 2 << 14);
            VRAMPtr_BBG[3] = GetUniqueBankPtr(VRAMMap_BBG[3], 3 << 14);
            VRAMPtr_BBG[6] = GetUniqueBankPtr(VRAMMap_BBG[6], 6 << 14);
            VRAMPtr_BBG[7] = GetUniqueBankPtr(VRAMMap_BBG[7], 7 << 14);
            break;

        case 2: // BOBJ
            UNMAP_RANGE_PTR(BOBJ, 0, 8);
            break;

        case 3: // BOBJ ext palette
            VRAMMap_BOBJExtPal &= ~bankmask;
            break;
        }
    }

    if (cnt & (1<<7))
    {
        switch (cnt & 0x3)
        {
        case 0: // LCDC
            VRAMMap_LCDC |= bankmask;
            break;

        case 1: // BBG
            VRAMMap_BBG[2] |= bankmask;
            VRAMMap_BBG[3] |= bankmask;
            VRAMMap_BBG[6] |= bankmask;
            VRAMMap_BBG[7] |= bankmask;
            VRAMPtr_BBG[2] = GetUniqueBankPtr(VRAMMap_BBG[2], 2 << 14);
            VRAMPtr_BBG[3] = GetUniqueBankPtr(VRAMMap_BBG[3], 3 << 14);
            VRAMPtr_BBG[6] = GetUniqueBankPtr(VRAMMap_BBG[6], 6 << 14);
            VRAMPtr_BBG[7] = GetUniqueBankPtr(VRAMMap_BBG[7], 7 << 14);
            break;

        case 2: // BOBJ
            MAP_RANGE_PTR(BOBJ, 0, 8);
            break;

        case 3: // BOBJ ext palette
            VRAMMap_BOBJExtPal |= bankmask;
            break;
        }
    }
}


void GPU::SetPowerCnt(u32 val) noexcept
{
    // POWCNT1 effects:
    // * bit0: asplodes hardware??? not tested.
    // * bit1: disables engine A palette and OAM (zero-filled) (TODO: affects mem timings???)
    // * bit2: disables rendering engine, resets its internal state (polygons and registers)
    // * bit3: disables geometry engine
    // * bit9: disables engine B palette, OAM and rendering (screen turns white)
    // * bit15: screen swap

    if (!(val & (1<<0))) Log(LogLevel::Warn, "!!! CLEARING POWCNT BIT0. DANGER\n");

    GPU2D_A.SetEnabled(val & (1<<1));
    GPU2D_B.SetEnabled(val & (1<<9));
    GPU3D.SetEnabled(val & (1<<3), val & (1<<2));

    AssignFramebuffers();
}


void GPU::DisplayFIFO(u32 x) noexcept
{
    // sample the FIFO
    // as this starts 16 cycles (~3 pixels) before display start,
    // we aren't aligned to the 8-pixel grid
    if (x > 0)
    {
        if (x == 8)
            GPU2D_A.SampleFIFO(0, 5);
        else
            GPU2D_A.SampleFIFO(x-11, 8);
    }

    if (x < 256)
    {
        // transfer the next 8 pixels
        NDS.CheckDMAs(0, 0x04);
        NDS.ScheduleEvent(Event_DisplayFIFO, true, 6*8, 0, x+8);
    }
    else
        GPU2D_A.SampleFIFO(253, 3); // sample the remaining pixels
}

void GPU::StartFrame() noexcept
{
    // only run the display FIFO if needed:
    // * if it is used for display or capture
    // * if we have display FIFO DMA
    RunFIFO = GPU2D_A.UsesFIFO() || NDS.DMAsInMode(0, 0x04);

    TotalScanlines = 0;
    StartScanline(0);
}

void GPU::StartHBlank(u32 line) noexcept
{
    DispStat[0] |= (1<<1);
    DispStat[1] |= (1<<1);

    if (VCount < 192)
    {
        // draw
        // note: this should start 48 cycles after the scanline start
        if (line < 192)
        {
            GPU2D_Renderer->DrawScanline(line, &GPU2D_A);
            GPU2D_Renderer->DrawScanline(line, &GPU2D_B);
        }

        // sprites are pre-rendered one scanline in advance
        if (line < 191)
        {
            GPU2D_Renderer->DrawSprites(line+1, &GPU2D_A);
            GPU2D_Renderer->DrawSprites(line+1, &GPU2D_B);
        }

        NDS.CheckDMAs(0, 0x02);
    }
    else if (VCount == 215)
    {
        GPU3D.VCount215();
    }
    else if (VCount == 262)
    {
        GPU2D_Renderer->DrawSprites(0, &GPU2D_A);
        GPU2D_Renderer->DrawSprites(0, &GPU2D_B);
    }

    if (DispStat[0] & (1<<4)) NDS.SetIRQ(0, IRQ_HBlank);
    if (DispStat[1] & (1<<4)) NDS.SetIRQ(1, IRQ_HBlank);

    if (VCount < 262)
        NDS.ScheduleEvent(Event_LCD, true, (LINE_CYCLES - HBLANK_CYCLES), LCD_StartScanline, line+1);
    else
        NDS.ScheduleEvent(Event_LCD, true, (LINE_CYCLES - HBLANK_CYCLES), LCD_FinishFrame, line+1);
}

void GPU::FinishFrame(u32 lines) noexcept
{
    FrontBuffer = FrontBuffer ? 0 : 1;
    AssignFramebuffers();

    TotalScanlines = lines;

    if (GPU3D.AbortFrame)
    {
        GPU3D.RestartFrame();
        GPU3D.AbortFrame = false;
    }
}

void GPU::BlankFrame() noexcept
{
    int backbuf = FrontBuffer ? 0 : 1;
    int fbsize;
    if (GPU3D.IsRendererAccelerated())
        fbsize = (256*3 + 1) * 192;
    else
        fbsize = 256 * 192;

    memset(Framebuffer[backbuf][0].get(), 0, fbsize*4);
    memset(Framebuffer[backbuf][1].get(), 0, fbsize*4);

    FrontBuffer = backbuf;
    AssignFramebuffers();

    TotalScanlines = 263;
}

void GPU::StartScanline(u32 line) noexcept
{
    if (line == 0)
        VCount = 0;
    else if (NextVCount != 0xFFFFFFFF)
        VCount = NextVCount;
    else
        VCount++;

    NextVCount = -1;

    DispStat[0] &= ~(1<<1);
    DispStat[1] &= ~(1<<1);

    if (VCount == VMatch[0])
    {
        DispStat[0] |= (1<<2);

        if (DispStat[0] & (1<<5)) NDS.SetIRQ(0, IRQ_VCount);
    }
    else
        DispStat[0] &= ~(1<<2);

    if (VCount == VMatch[1])
    {
        DispStat[1] |= (1<<2);

        if (DispStat[1] & (1<<5)) NDS.SetIRQ(1, IRQ_VCount);
    }
    else
        DispStat[1] &= ~(1<<2);

    GPU2D_A.CheckWindows(VCount);
    GPU2D_B.CheckWindows(VCount);

    if (VCount >= 2 && VCount < 194)
        NDS.CheckDMAs(0, 0x03);
    else if (VCount == 194)
        NDS.StopDMAs(0, 0x03);

    if (line < 192)
    {
        if (line == 0)
        {
            GPU2D_Renderer->VBlankEnd(&GPU2D_A, &GPU2D_B);
            GPU2D_A.VBlankEnd();
            GPU2D_B.VBlankEnd();
        }

        if (RunFIFO)
            NDS.ScheduleEvent(Event_DisplayFIFO, false, 32, 0, 0);
    }

    if (VCount == 262)
    {
        // frame end

        DispStat[0] &= ~(1<<0);
        DispStat[1] &= ~(1<<0);
    }
    else
    {
        if (VCount == 192)
        {
            // in reality rendering already finishes at line 144
            // and games might already start to modify texture memory.
            // That doesn't matter for us because we cache the entire
            // texture memory anyway and only update it before the start
            //of the next frame.
            // So we can give the rasteriser a bit more headroom
            GPU3D.VCount144();

            // VBlank
            DispStat[0] |= (1<<0);
            DispStat[1] |= (1<<0);

            NDS.StopDMAs(0, 0x04);

            NDS.CheckDMAs(0, 0x01);
            NDS.CheckDMAs(1, 0x11);

            if (DispStat[0] & (1<<3)) NDS.SetIRQ(0, IRQ_VBlank);
            if (DispStat[1] & (1<<3)) NDS.SetIRQ(1, IRQ_VBlank);

            GPU2D_A.VBlank();
            GPU2D_B.VBlank();
            GPU3D.VBlank();

            // Need a better way to identify the openGL renderer in particular
            if (GPU3D.IsRendererAccelerated())
                GPU3D.Blit();
        }
    }

    NDS.ScheduleEvent(Event_LCD, true, HBLANK_CYCLES, LCD_StartHBlank, line);
}


void GPU::SetDispStat(u32 cpu, u16 val) noexcept
{
    val &= 0xFFB8;
    DispStat[cpu] &= 0x0047;
    DispStat[cpu] |= val;

    VMatch[cpu] = (val >> 8) | ((val & 0x80) << 1);
}

void GPU::SetVCount(u16 val) noexcept
{
    // VCount write is delayed until the next scanline

    // TODO: how does the 3D engine react to VCount writes while it's rendering?
    // 3D engine seems to give up on the current frame in that situation, repeating the last two scanlines
    // TODO: also check the various DMA types that can be involved

    GPU3D.AbortFrame |= NextVCount != val;
    NextVCount = val;
}

template <u32 Size, u32 MappingGranularity>
NonStupidBitField<Size/VRAMDirtyGranularity> VRAMTrackingSet<Size, MappingGranularity>::DeriveState(u32* currentMappings, GPU& gpu)
{
    NonStupidBitField<Size/VRAMDirtyGranularity> result;
    u16 banksToBeZeroed = 0;
    for (u32 i = 0; i < Size / MappingGranularity; i++)
    {
        if (currentMappings[i] != Mapping[i])
        {
            result.SetRange(i*VRAMBitsPerMapping, VRAMBitsPerMapping);
            banksToBeZeroed |= currentMappings[i];
            Mapping[i] = currentMappings[i];
        }
        else
        {
            u32 mapping = Mapping[i];

            banksToBeZeroed |= mapping;

            while (mapping != 0)
            {
                u32 num = __builtin_ctz(mapping);
                mapping &= ~(1 << num);

                // hack for **speed**
                // this could probably be done less ugly but then we would rely
                // on the compiler for vectorisation
                static_assert(VRAMDirtyGranularity == 512, "");
                if (MappingGranularity == 16*1024)
                {
                    u32 dirty = ((u32*)gpu.VRAMDirty[num].Data)[i & (gpu.VRAMMask[num] >> 14)];
                    result.Data[i / 2] |= (u64)dirty << ((i&1)*32);
                }
                else if (MappingGranularity == 8*1024)
                {
                    u16 dirty = ((u16*)gpu.VRAMDirty[num].Data)[i & (gpu.VRAMMask[num] >> 13)];
                    result.Data[i / 4] |= (u64)dirty << ((i&3)*16);
                }
                else if (MappingGranularity == 128*1024)
                {
                    result.Data[i * 4 + 0] |= gpu.VRAMDirty[num].Data[0];
                    result.Data[i * 4 + 1] |= gpu.VRAMDirty[num].Data[1];
                    result.Data[i * 4 + 2] |= gpu.VRAMDirty[num].Data[2];
                    result.Data[i * 4 + 3] |= gpu.VRAMDirty[num].Data[3];
                }
                else
                {
                    // welp
                    abort();
                }
            }
        }
    }

    while (banksToBeZeroed != 0)
    {
        u32 num = __builtin_ctz(banksToBeZeroed);
        banksToBeZeroed &= ~(1 << num);
        gpu.VRAMDirty[num].Clear();
    }

    return result;
}

template NonStupidBitField<32*1024/VRAMDirtyGranularity> VRAMTrackingSet<32*1024, 8*1024>::DeriveState(u32*, GPU& gpu);
template NonStupidBitField<8*1024/VRAMDirtyGranularity> VRAMTrackingSet<8*1024, 8*1024>::DeriveState(u32*, GPU& gpu);
template NonStupidBitField<512*1024/VRAMDirtyGranularity> VRAMTrackingSet<512*1024, 128*1024>::DeriveState(u32*, GPU& gpu);
template NonStupidBitField<128*1024/VRAMDirtyGranularity> VRAMTrackingSet<128*1024, 16*1024>::DeriveState(u32*, GPU& gpu);
template NonStupidBitField<256*1024/VRAMDirtyGranularity> VRAMTrackingSet<256*1024, 16*1024>::DeriveState(u32*, GPU& gpu);
template NonStupidBitField<512*1024/VRAMDirtyGranularity> VRAMTrackingSet<512*1024, 16*1024>::DeriveState(u32*, GPU& gpu);



bool GPU::MakeVRAMFlat_TextureCoherent(NonStupidBitField<512*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<128*1024>(VRAMFlat_Texture, VRAMMap_Texture, dirty, &GPU::ReadVRAM_Texture<u64>);
}
bool GPU::MakeVRAMFlat_TexPalCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<16*1024>(VRAMFlat_TexPal, VRAMMap_TexPal, dirty, &GPU::ReadVRAM_TexPal<u64>);
}

bool GPU::MakeVRAMFlat_ABGCoherent(NonStupidBitField<512*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<16*1024>(VRAMFlat_ABG, VRAMMap_ABG, dirty, &GPU::ReadVRAM_ABG<u64>);
}
bool GPU::MakeVRAMFlat_BBGCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<16*1024>(VRAMFlat_BBG, VRAMMap_BBG, dirty, &GPU::ReadVRAM_BBG<u64>);
}

bool GPU::MakeVRAMFlat_AOBJCoherent(NonStupidBitField<256*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<16*1024>(VRAMFlat_AOBJ, VRAMMap_AOBJ, dirty, &GPU::ReadVRAM_AOBJ<u64>);
}
bool GPU::MakeVRAMFlat_BOBJCoherent(NonStupidBitField<128*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<16*1024>(VRAMFlat_BOBJ, VRAMMap_BOBJ, dirty, &GPU::ReadVRAM_BOBJ<u64>);
}

bool GPU::MakeVRAMFlat_ABGExtPalCoherent(NonStupidBitField<32*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<8*1024>(VRAMFlat_ABGExtPal, VRAMMap_ABGExtPal, dirty, &GPU::ReadVRAM_ABGExtPal<u64>);
}
bool GPU::MakeVRAMFlat_BBGExtPalCoherent(NonStupidBitField<32*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<8*1024>(VRAMFlat_BBGExtPal, VRAMMap_BBGExtPal, dirty, &GPU::ReadVRAM_BBGExtPal<u64>);
}

bool GPU::MakeVRAMFlat_AOBJExtPalCoherent(NonStupidBitField<8*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<8*1024>(VRAMFlat_AOBJExtPal, &VRAMMap_AOBJExtPal, dirty, &GPU::ReadVRAM_AOBJExtPal<u64>);
}
bool GPU::MakeVRAMFlat_BOBJExtPalCoherent(NonStupidBitField<8*1024/VRAMDirtyGranularity>& dirty) noexcept
{
    return CopyLinearVRAM<8*1024>(VRAMFlat_BOBJExtPal, &VRAMMap_BOBJExtPal, dirty, &GPU::ReadVRAM_BOBJExtPal<u64>);
}
}