project64/Source/Project64-rsp-core/Hle/jpeg.cpp

// Project64 - A Nintendo 64 emulator
// https://www.pj64-emu.com/
// Copyright(C) 2001-2021 Project64
// Copyright(C) 2012 Bobby Smiles
// Copyright(C) 2009 Richard Goedeken
// Copyright(C) 2002 Hacktarux
// GNU/GPLv2 licensed: https://gnu.org/licenses/gpl-2.0.html

#include <stdint.h>
#include <stdlib.h>

#include "arithmetics.h"
#include "mem.h"

#define SUBBLOCK_SIZE 64

typedef void (*tile_line_emitter_t)(CHle * hle, const int16_t * y, const int16_t * u, uint32_t address);
typedef void (*subblock_transform_t)(int16_t * dst, const int16_t * src);

// Standard JPEG microcode decoder
static void jpeg_decode_std(CHle * hle,
                            const char * const version,
                            const subblock_transform_t transform_luma,
                            const subblock_transform_t transform_chroma,
                            const tile_line_emitter_t emit_line);

// Helper functions
static uint8_t clamp_u8(int16_t x);
static int16_t clamp_s12(int16_t x);
static uint16_t clamp_RGBA_component(int16_t x);

// Pixel conversion and formatting
static uint32_t GetUYVY(int16_t y1, int16_t y2, int16_t u, int16_t v);
static uint16_t GetRGBA(int16_t y, int16_t u, int16_t v);

// Tile line emitters
static void EmitYUVTileLine(CHle * hle, const int16_t * y, const int16_t * u, uint32_t address);
static void EmitRGBATileLine(CHle * hle, const int16_t * y, const int16_t * u, uint32_t address);

// Macroblocks operations
static void decode_macroblock_ob(int16_t * macroblock, int32_t * y_dc, int32_t * u_dc, int32_t * v_dc, const int16_t * qtable);
static void decode_macroblock_std(const subblock_transform_t transform_luma,
                                  const subblock_transform_t transform_chroma,
                                  int16_t * macroblock,
                                  unsigned int subblock_count,
                                  const int16_t qtables[3][SUBBLOCK_SIZE]);
static void EmitTilesMode0(CHle * hle, const tile_line_emitter_t emit_line, const int16_t * macroblock, uint32_t address);
static void EmitTilesMode2(CHle * hle, const tile_line_emitter_t emit_line, const int16_t * macroblock, uint32_t address);

// Sub blocks operations
static void TransposeSubBlock(int16_t * dst, const int16_t * src);
static void ZigZagSubBlock(int16_t * dst, const int16_t * src);
static void ReorderSubBlock(int16_t * dst, const int16_t * src, const unsigned int * table);
static void MultSubBlocks(int16_t * dst, const int16_t * src1, const int16_t * src2, unsigned int shift);
static void ScaleSubBlock(int16_t * dst, const int16_t * src, int16_t scale);
static void RShiftSubBlock(int16_t * dst, const int16_t * src, unsigned int shift);
static void InverseDCT1D(const float * const x, float * dst, unsigned int stride);
static void InverseDCTSubBlock(int16_t * dst, const int16_t * src);
static void RescaleYSubBlock(int16_t * dst, const int16_t * src);
static void RescaleUVSubBlock(int16_t * dst, const int16_t * src);

// Transposed dequantization table
static const int16_t DEFAULT_QTABLE[SUBBLOCK_SIZE] = {
    16, 12, 14, 14, 18, 24, 49, 72,
    11, 12, 13, 17, 22, 35, 64, 92,
    10, 14, 16, 22, 37, 55, 78, 95,
    16, 19, 24, 29, 56, 64, 87, 98,
    24, 26, 40, 51, 68, 81, 103, 112,
    40, 58, 57, 87, 109, 104, 121, 100,
    51, 60, 69, 80, 103, 113, 120, 103,
    61, 55, 56, 62, 77, 92, 101, 99};

// Zig-zag indices
static const unsigned int ZIGZAG_TABLE[SUBBLOCK_SIZE] = {
    0, 1, 5, 6, 14, 15, 27, 28,
    2, 4, 7, 13, 16, 26, 29, 42,
    3, 8, 12, 17, 25, 30, 41, 43,
    9, 11, 18, 24, 31, 40, 44, 53,
    10, 19, 23, 32, 39, 45, 52, 54,
    20, 22, 33, 38, 46, 51, 55, 60,
    21, 34, 37, 47, 50, 56, 59, 61,
    35, 36, 48, 49, 57, 58, 62, 63};

// Transposition indices
static const unsigned int TRANSPOSE_TABLE[SUBBLOCK_SIZE] = {
    0, 8, 16, 24, 32, 40, 48, 56,
    1, 9, 17, 25, 33, 41, 49, 57,
    2, 10, 18, 26, 34, 42, 50, 58,
    3, 11, 19, 27, 35, 43, 51, 59,
    4, 12, 20, 28, 36, 44, 52, 60,
    5, 13, 21, 29, 37, 45, 53, 61,
    6, 14, 22, 30, 38, 46, 54, 62,
    7, 15, 23, 31, 39, 47, 55, 63};

/* IDCT related constants
 * Cn = alpha * cos(n * PI / 16) (alpha is chosen such as C4 = 1) */
static const float IDCT_C3 = 1.175875602f;
static const float IDCT_C6 = 0.541196100f;
static const float IDCT_K[10] = {
    0.765366865f,  //  C2-C6
    -1.847759065f, // -C2-C6
    -0.390180644f, //  C5-C3
    -1.961570561f, // -C5-C3
    1.501321110f,  //  C1+C3-C5-C7
    2.053119869f,  //  C1+C3-C5+C7
    3.072711027f,  //  C1+C3+C5-C7
    0.298631336f,  // -C1+C3+C5-C7
    -0.899976223f, //  C7-C3
    -2.562915448f  // -C1-C3
};

// Global functions

// JPEG decoding microcode found in Japanese-exclusive version of Pokémon Stadium

void jpeg_decode_PS0(CHle * hle)
{
    jpeg_decode_std(hle, "PS0", RescaleYSubBlock, RescaleUVSubBlock, EmitYUVTileLine);
}

// JPEG decoding microcode found in Ocarina of Time, Pokémon Stadium 1, and Pokémon Stadium 2

void jpeg_decode_PS(CHle * hle)
{
    jpeg_decode_std(hle, "PS", NULL, NULL, EmitRGBATileLine);
}

// JPEG decoding microcode found in Ogre Battle and Bottom of the 9th

void jpeg_decode_OB(CHle * hle)
{
    int16_t qtable[SUBBLOCK_SIZE];
    unsigned int mb;

    int32_t y_dc = 0;
    int32_t u_dc = 0;
    int32_t v_dc = 0;

    uint32_t address = *dmem_u32(hle, TASK_DATA_PTR);
    const unsigned int macroblock_count = *dmem_u32(hle, TASK_DATA_SIZE);
    const int qscale = *dmem_u32(hle, TASK_YIELD_DATA_SIZE);

    hle->VerboseMessage("jpeg_decode_OB: *buffer=%x, #MB=%d, qscale=%d", address, macroblock_count, qscale);

    if (qscale != 0)
    {
        if (qscale > 0)
        {
            ScaleSubBlock(qtable, DEFAULT_QTABLE, qscale);
        }
        else
        {
            RShiftSubBlock(qtable, DEFAULT_QTABLE, -qscale);
        }
    }

    for (mb = 0; mb < macroblock_count; ++mb)
    {
        int16_t macroblock[6 * SUBBLOCK_SIZE];

        dram_load_u16(hle, (uint16_t *)macroblock, address, 6 * SUBBLOCK_SIZE);
        decode_macroblock_ob(macroblock, &y_dc, &u_dc, &v_dc, (qscale != 0) ? qtable : NULL);
        EmitTilesMode2(hle, EmitYUVTileLine, macroblock, address);

        address += (2 * 6 * SUBBLOCK_SIZE);
    }
}

// Local functions

static void jpeg_decode_std(CHle * hle, const char * const version, const subblock_transform_t transform_luma, const subblock_transform_t transform_chroma, const tile_line_emitter_t emit_line)
{
    int16_t qtables[3][SUBBLOCK_SIZE];
    unsigned int mb;
    uint32_t address;
    uint32_t macroblock_count;
    uint32_t mode;
    uint32_t qtableY_ptr;
    uint32_t qtableU_ptr;
    uint32_t qtableV_ptr;
    unsigned int subblock_count;
    unsigned int macroblock_size;
    // Macroblock contains at most 6 sub blocks
    int16_t macroblock[6 * SUBBLOCK_SIZE];
    uint32_t data_ptr;

    if (*dmem_u32(hle, TASK_FLAGS) & 0x1)
    {
        hle->WarnMessage("jpeg_decode_%s: task yielding not implemented", version);
        return;
    }

    data_ptr = *dmem_u32(hle, TASK_DATA_PTR);
    address = *dram_u32(hle, data_ptr);
    macroblock_count = *dram_u32(hle, data_ptr + 4);
    mode = *dram_u32(hle, data_ptr + 8);
    qtableY_ptr = *dram_u32(hle, data_ptr + 12);
    qtableU_ptr = *dram_u32(hle, data_ptr + 16);
    qtableV_ptr = *dram_u32(hle, data_ptr + 20);

    hle->VerboseMessage("jpeg_decode_%s: *buffer=%x, #MB=%d, mode=%d, *Qy=%x, *Qu=%x, *Qv=%x", version, address, macroblock_count, mode, qtableY_ptr, qtableU_ptr, qtableV_ptr);

    if (mode != 0 && mode != 2)
    {
        hle->WarnMessage("jpeg_decode_%s: invalid mode %d", version, mode);
        return;
    }

    subblock_count = mode + 4;
    macroblock_size = subblock_count * SUBBLOCK_SIZE;

    dram_load_u16(hle, (uint16_t *)qtables[0], qtableY_ptr, SUBBLOCK_SIZE);
    dram_load_u16(hle, (uint16_t *)qtables[1], qtableU_ptr, SUBBLOCK_SIZE);
    dram_load_u16(hle, (uint16_t *)qtables[2], qtableV_ptr, SUBBLOCK_SIZE);

    for (mb = 0; mb < macroblock_count; ++mb)
    {
        dram_load_u16(hle, (uint16_t *)macroblock, address, macroblock_size);
        decode_macroblock_std(transform_luma, transform_chroma, macroblock, subblock_count, (const int16_t(*)[SUBBLOCK_SIZE])qtables);

        if (mode == 0)
        {
            EmitTilesMode0(hle, emit_line, macroblock, address);
        }
        else
        {
            EmitTilesMode2(hle, emit_line, macroblock, address);
        }
        address += 2 * macroblock_size;
    }
}

static uint8_t clamp_u8(int16_t x)
{
    return (x & (0xff00)) ? ((-x) >> 15) & 0xff : x;
}

static int16_t clamp_s12(int16_t x)
{
    if (x < -0x800)
    {
        x = -0x800;
    }
    else if (x > 0x7f0)
    {
        x = 0x7f0;
    }
    return x;
}

static uint16_t clamp_RGBA_component(int16_t x)
{
    if (x > 0xff0)
    {
        x = 0xff0;
    }
    else if (x < 0)
    {
        x = 0;
    }
    return (x & 0xf80);
}

static uint32_t GetUYVY(int16_t y1, int16_t y2, int16_t u, int16_t v)
{
    return (uint32_t)clamp_u8(u) << 24 |
           (uint32_t)clamp_u8(y1) << 16 |
           (uint32_t)clamp_u8(v) << 8 |
           (uint32_t)clamp_u8(y2);
}

static uint16_t GetRGBA(int16_t y, int16_t u, int16_t v)
{
    const float fY = (float)y + 2048.0f;
    const float fU = (float)u;
    const float fV = (float)v;

    const uint16_t r = clamp_RGBA_component((int16_t)(fY + 1.4025 * fV));
    const uint16_t g = clamp_RGBA_component((int16_t)(fY - 0.3443 * fU - 0.7144 * fV));
    const uint16_t b = clamp_RGBA_component((int16_t)(fY + 1.7729 * fU));

    return (r << 4) | (g >> 1) | (b >> 6) | 1;
}

static void EmitYUVTileLine(CHle * hle, const int16_t * y, const int16_t * u, uint32_t address)
{
    uint32_t uyvy[8];

    const int16_t * const v = u + SUBBLOCK_SIZE;
    const int16_t * const y2 = y + SUBBLOCK_SIZE;

    uyvy[0] = GetUYVY(y[0], y[1], u[0], v[0]);
    uyvy[1] = GetUYVY(y[2], y[3], u[1], v[1]);
    uyvy[2] = GetUYVY(y[4], y[5], u[2], v[2]);
    uyvy[3] = GetUYVY(y[6], y[7], u[3], v[3]);
    uyvy[4] = GetUYVY(y2[0], y2[1], u[4], v[4]);
    uyvy[5] = GetUYVY(y2[2], y2[3], u[5], v[5]);
    uyvy[6] = GetUYVY(y2[4], y2[5], u[6], v[6]);
    uyvy[7] = GetUYVY(y2[6], y2[7], u[7], v[7]);

    dram_store_u32(hle, uyvy, address, 8);
}

static void EmitRGBATileLine(CHle * hle, const int16_t * y, const int16_t * u, uint32_t address)
{
    uint16_t rgba[16];

    const int16_t * const v = u + SUBBLOCK_SIZE;
    const int16_t * const y2 = y + SUBBLOCK_SIZE;

    rgba[0] = GetRGBA(y[0], u[0], v[0]);
    rgba[1] = GetRGBA(y[1], u[0], v[0]);
    rgba[2] = GetRGBA(y[2], u[1], v[1]);
    rgba[3] = GetRGBA(y[3], u[1], v[1]);
    rgba[4] = GetRGBA(y[4], u[2], v[2]);
    rgba[5] = GetRGBA(y[5], u[2], v[2]);
    rgba[6] = GetRGBA(y[6], u[3], v[3]);
    rgba[7] = GetRGBA(y[7], u[3], v[3]);
    rgba[8] = GetRGBA(y2[0], u[4], v[4]);
    rgba[9] = GetRGBA(y2[1], u[4], v[4]);
    rgba[10] = GetRGBA(y2[2], u[5], v[5]);
    rgba[11] = GetRGBA(y2[3], u[5], v[5]);
    rgba[12] = GetRGBA(y2[4], u[6], v[6]);
    rgba[13] = GetRGBA(y2[5], u[6], v[6]);
    rgba[14] = GetRGBA(y2[6], u[7], v[7]);
    rgba[15] = GetRGBA(y2[7], u[7], v[7]);

    dram_store_u16(hle, rgba, address, 16);
}

static void EmitTilesMode0(CHle * hle, const tile_line_emitter_t emit_line, const int16_t * macroblock, uint32_t address)
{
    unsigned int i;

    unsigned int y_offset = 0;
    unsigned int u_offset = 2 * SUBBLOCK_SIZE;

    for (i = 0; i < 8; ++i)
    {
        emit_line(hle, &macroblock[y_offset], &macroblock[u_offset], address);

        y_offset += 8;
        u_offset += 8;
        address += 32;
    }
}

static void EmitTilesMode2(CHle * hle, const tile_line_emitter_t emit_line, const int16_t * macroblock, uint32_t address)
{
    unsigned int i;

    unsigned int y_offset = 0;
    unsigned int u_offset = 4 * SUBBLOCK_SIZE;

    for (i = 0; i < 8; ++i)
    {
        emit_line(hle, &macroblock[y_offset], &macroblock[u_offset], address);
        emit_line(hle, &macroblock[y_offset + 8], &macroblock[u_offset], address + 32);

        y_offset += (i == 3) ? SUBBLOCK_SIZE + 16 : 16;
        u_offset += 8;
        address += 64;
    }
}

static void decode_macroblock_ob(int16_t * macroblock, int32_t * y_dc, int32_t * u_dc, int32_t * v_dc, const int16_t * qtable)
{
    int sb;

    for (sb = 0; sb < 6; ++sb)
    {
        int16_t tmp_sb[SUBBLOCK_SIZE];

        // Update decode
        int32_t dc = (int32_t)macroblock[0];
        switch (sb)
        {
        case 0:
        case 1:
        case 2:
        case 3:
            *y_dc += dc;
            macroblock[0] = *y_dc & 0xffff;
            break;
        case 4:
            *u_dc += dc;
            macroblock[0] = *u_dc & 0xffff;
            break;
        case 5:
            *v_dc += dc;
            macroblock[0] = *v_dc & 0xffff;
            break;
        }

        ZigZagSubBlock(tmp_sb, macroblock);
        if (qtable != NULL)
        {
            MultSubBlocks(tmp_sb, tmp_sb, qtable, 0);
        }
        TransposeSubBlock(macroblock, tmp_sb);
        InverseDCTSubBlock(macroblock, macroblock);

        macroblock += SUBBLOCK_SIZE;
    }
}

static void decode_macroblock_std(const subblock_transform_t transform_luma,
                                  const subblock_transform_t transform_chroma,
                                  int16_t * macroblock,
                                  unsigned int subblock_count,
                                  const int16_t qtables[3][SUBBLOCK_SIZE])
{
    unsigned int sb;
    unsigned int q = 0;

    for (sb = 0; sb < subblock_count; ++sb)
    {
        int16_t tmp_sb[SUBBLOCK_SIZE];
        const int isChromaSubBlock = (subblock_count - sb <= 2);

        if (isChromaSubBlock)
        {
            ++q;
        }

        MultSubBlocks(macroblock, macroblock, qtables[q], 4);
        ZigZagSubBlock(tmp_sb, macroblock);
        InverseDCTSubBlock(macroblock, tmp_sb);

        if (isChromaSubBlock)
        {
            if (transform_chroma != NULL)
            {
                transform_chroma(macroblock, macroblock);
            }
        }
        else
        {
            if (transform_luma != NULL)
            {
                transform_luma(macroblock, macroblock);
            }
        }
        macroblock += SUBBLOCK_SIZE;
    }
}

static void TransposeSubBlock(int16_t * dst, const int16_t * src)
{
    ReorderSubBlock(dst, src, TRANSPOSE_TABLE);
}

static void ZigZagSubBlock(int16_t * dst, const int16_t * src)
{
    ReorderSubBlock(dst, src, ZIGZAG_TABLE);
}

static void ReorderSubBlock(int16_t * dst, const int16_t * src, const unsigned int * table)
{
    unsigned int i;

    // Source and destination sub blocks cannot overlap
    assert(abs(dst - src) > SUBBLOCK_SIZE);

    for (i = 0; i < SUBBLOCK_SIZE; ++i)
        dst[i] = src[table[i]];
}

static void MultSubBlocks(int16_t * dst, const int16_t * src1, const int16_t * src2, unsigned int shift)
{
    unsigned int i;

    for (i = 0; i < SUBBLOCK_SIZE; ++i)
    {
        int32_t v = src1[i] * src2[i];
        dst[i] = clamp_s16(v) << shift;
    }
}

static void ScaleSubBlock(int16_t * dst, const int16_t * src, int16_t scale)
{
    unsigned int i;

    for (i = 0; i < SUBBLOCK_SIZE; ++i)
    {
        int32_t v = src[i] * scale;
        dst[i] = clamp_s16(v);
    }
}

static void RShiftSubBlock(int16_t * dst, const int16_t * src, unsigned int shift)
{
    unsigned int i;

    for (i = 0; i < SUBBLOCK_SIZE; ++i)
        dst[i] = src[i] >> shift;
}

/*
TODO: find a better, more general resource for this
Fast 2D IDCT using separable formulation and normalization
Computations use single precision floats
Implementation based on Wikipedia:
https://fr.wikipedia.org/wiki/Transform%C3%A9e_en_cosinus_discr%C3%A8te
*/

static void InverseDCT1D(const float * const x, float * dst, unsigned int stride)
{
    float e[4];
    float f[4];
    float x26, x1357, x15, x37, x17, x35;

    x15 = IDCT_K[2] * (x[1] + x[5]);
    x37 = IDCT_K[3] * (x[3] + x[7]);
    x17 = IDCT_K[8] * (x[1] + x[7]);
    x35 = IDCT_K[9] * (x[3] + x[5]);
    x1357 = IDCT_C3 * (x[1] + x[3] + x[5] + x[7]);
    x26 = IDCT_C6 * (x[2] + x[6]);

    f[0] = x[0] + x[4];
    f[1] = x[0] - x[4];
    f[2] = x26 + IDCT_K[0] * x[2];
    f[3] = x26 + IDCT_K[1] * x[6];

    e[0] = x1357 + x15 + IDCT_K[4] * x[1] + x17;
    e[1] = x1357 + x37 + IDCT_K[6] * x[3] + x35;
    e[2] = x1357 + x15 + IDCT_K[5] * x[5] + x35;
    e[3] = x1357 + x37 + IDCT_K[7] * x[7] + x17;

    *dst = f[0] + f[2] + e[0];
    dst += stride;
    *dst = f[1] + f[3] + e[1];
    dst += stride;
    *dst = f[1] - f[3] + e[2];
    dst += stride;
    *dst = f[0] - f[2] + e[3];
    dst += stride;
    *dst = f[0] - f[2] - e[3];
    dst += stride;
    *dst = f[1] - f[3] - e[2];
    dst += stride;
    *dst = f[1] + f[3] - e[1];
    dst += stride;
    *dst = f[0] + f[2] - e[0];
}

static void InverseDCTSubBlock(int16_t * dst, const int16_t * src)
{
    float x[8];
    float block[SUBBLOCK_SIZE];
    unsigned int i, j;

    // IDCT 1D on rows (+transposition)
    for (i = 0; i < 8; ++i)
    {
        for (j = 0; j < 8; ++j)
        {
            x[j] = (float)src[i * 8 + j];
        }
        InverseDCT1D(x, &block[i], 8);
    }

    // IDCT 1D on columns (thanks to previous transposition)
    for (i = 0; i < 8; ++i)
    {
        InverseDCT1D(&block[i * 8], x, 1);

        // C4 = 1 normalization implies a division by 8
        for (j = 0; j < 8; ++j)
        {
            dst[i + j * 8] = (int16_t)x[j] >> 3;
        }
    }
}
static void RescaleYSubBlock(int16_t * dst, const int16_t * src)
{
    unsigned int i;

    for (i = 0; i < SUBBLOCK_SIZE; ++i)
    {
        dst[i] = (((uint32_t)(clamp_s12(src[i]) + 0x800) * 0xdb0) >> 16) + 0x10;
    }
}
static void RescaleUVSubBlock(int16_t * dst, const int16_t * src)
{
    unsigned int i;

    for (i = 0; i < SUBBLOCK_SIZE; ++i)
    {
        dst[i] = (((int)clamp_s12(src[i]) * 0xe00) >> 16) + 0x80;
    }
}