path: root/mm/src/code/PreRender.c

/**
 * @file PreRender.c
 *
 * This file implements various routines important to framebuffer effects, such as RDP accelerated color and depth
 * buffer copies and coverage drawing. Also contains software implementations of the Video Interface anti-aliasing and
 * divot filters.
 */

#include "z64prerender.h"
#include "global.h"
#include "PR/gs2dex.h"
#include "libc/alloca.h"
#include "libc/stdbool.h"
#include "color.h"
#include "macros.h"
#include "slowly.h"
#include "stack.h"
#include "stackcheck.h"
#include <stdlib.h>
#include <string.h>
#include <libultraship/bridge/gfxbridge.h>

/**
 * Assigns the "save" values in PreRender
 */
void PreRender_SetValuesSave(PreRender* this, u32 width, u32 height, void* fbuf, void* zbuf, void* cvg) {
    this->widthSave = width;
    this->heightSave = height;
    this->fbufSave = fbuf;
    this->cvgSave = cvg;
    this->zbufSave = zbuf;
    this->ulxSave = 0;
    this->ulySave = 0;
    this->lrxSave = width - 1;
    this->lrySave = height - 1;
}

void PreRender_Init(PreRender* this) {
    memset(this, 0, sizeof(PreRender));
    ListAlloc_Init(&this->alloc);
}

/**
 * Assigns the current values in PreRender
 */
void PreRender_SetValues(PreRender* this, u32 width, u32 height, void* fbuf, void* zbuf) {
    this->width = width;
    this->height = height;
    this->fbuf = fbuf;
    this->zbuf = zbuf;
    this->ulx = 0;
    this->uly = 0;
    this->lrx = width - 1;
    this->lry = height - 1;
}

void PreRender_Destroy(PreRender* this) {
    ListAlloc_FreeAll(&this->alloc);
}

void PreRender_CopyImage(PreRender* this, Gfx** gfxP, void* img, void* imgDst, u32 useThresholdAlphaCompare) {
    Gfx* gfx = *gfxP;
    u32 flags;

    gDPPipeSync(gfx++);
    gDPSetColorImage(gfx++, G_IM_FMT_RGBA, G_IM_SIZ_16b, this->width, imgDst);
    gDPSetScissor(gfx++, G_SC_NON_INTERLACE, 0, 0, this->width, this->height);

    flags = BG2D_FLAGS_LOAD_S2DEX2 | BG2D_FLAGS_COPY;
    if (useThresholdAlphaCompare == true) {
        flags |= BG2D_FLAGS_AC_THRESHOLD;
    }

    Prerender_DrawBackground2D(&gfx, img, NULL, this->width, this->height, G_IM_FMT_RGBA, G_IM_SIZ_16b, G_TT_NONE, 0,
                               0.0f, 0.0f, 1.0f, 1.0f, flags);
    gDPPipeSync(gfx++);
    gDPSetColorImage(gfx++, G_IM_FMT_RGBA, G_IM_SIZ_16b, this->width, this->fbuf);

    *gfxP = gfx;
}

void PreRender_RestoreBuffer(PreRender* this, Gfx** gfxP, void* buf, void* bufSave) {
    PreRender_CopyImage(this, gfxP, buf, bufSave, false);
}

void func_8016FF90(PreRender* this, Gfx** gfxP, void* buf, void* bufSave, s32 envR, s32 envG, s32 envB, s32 envA) {
    Gfx* gfx = *gfxP;

    gDPPipeSync(gfx++);

    if (envA == 255) {
        gDPSetOtherMode(gfx++,
                        G_AD_DISABLE | G_CD_DISABLE | G_CK_NONE | G_TC_FILT | G_TF_POINT | G_TT_NONE | G_TL_TILE |
                            G_TD_CLAMP | G_TP_NONE | G_CYC_1CYCLE | G_PM_NPRIMITIVE,
                        G_AC_NONE | G_ZS_PRIM | G_RM_OPA_SURF | G_RM_OPA_SURF2);
    } else {
        gDPSetOtherMode(gfx++,
                        G_AD_NOISE | G_CD_NOISE | G_CK_NONE | G_TC_FILT | G_TF_POINT | G_TT_NONE | G_TL_TILE |
                            G_TD_CLAMP | G_TP_NONE | G_CYC_1CYCLE | G_PM_NPRIMITIVE,
                        G_AC_NONE | G_ZS_PRIM | G_RM_CLD_SURF | G_RM_CLD_SURF2);
    }

    gDPSetEnvColor(gfx++, envR, envG, envB, envA);
    gDPSetCombineLERP(gfx++, TEXEL0, 0, ENVIRONMENT, 0, 0, 0, 0, ENVIRONMENT, TEXEL0, 0, ENVIRONMENT, 0, 0, 0, 0,
                      ENVIRONMENT);
    gDPSetColorImage(gfx++, G_IM_FMT_RGBA, G_IM_SIZ_16b, this->width, bufSave);

    gDPSetScissor(gfx++, G_SC_NON_INTERLACE, 0, 0, this->width, this->height);

    Prerender_DrawBackground2D(&gfx, buf, NULL, this->width, this->height, G_IM_FMT_RGBA, G_IM_SIZ_16b, G_TT_NONE, 0,
                               0.0f, 0.0f, 1.0f, 1.0f, 0xB);
    gDPPipeSync(gfx++);
    gDPSetColorImage(gfx++, G_IM_FMT_RGBA, G_IM_SIZ_16b, this->width, this->fbuf);

    *gfxP = gfx;
}

void func_80170200(PreRender* this, Gfx** gfxP, void* buf, void* bufSave) {
    func_8016FF90(this, gfxP, buf, bufSave, 255, 255, 255, 255);
}

/**
 * Reads the coverage values stored in the RGBA16 format `img` with dimensions `this->width`, `this->height` and
 * converts it to an 8-bpp intensity image.
 *
 * @param gfxP      Display list pointer
 * @param img       Image to read coverage from
 * @param cvgDst    Buffer to store coverage into
 */
void PreRender_CoverageRgba16ToI8(PreRender* this, Gfx** gfxP, void* img, void* cvgDst) {
    // BENTODO:
    return;
    Gfx* gfx = *gfxP;
    s32 rowsRemaining;
    s32 curRow;
    s32 nRows;

    gDPPipeSync(gfx++);
    gDPSetOtherMode(gfx++,
                    G_AD_DISABLE | G_CD_DISABLE | G_CK_NONE | G_TC_FILT | G_TF_POINT | G_TT_NONE | G_TL_TILE |
                        G_TD_CLAMP | G_TP_NONE | G_CYC_1CYCLE | G_PM_NPRIMITIVE,
                    G_AC_NONE | G_ZS_PRIM | G_RM_PASS | G_RM_OPA_CI2);

    // Set the combiner to draw the texture as-is, discarding alpha channel
    gDPSetCombineLERP(gfx++, 0, 0, 0, TEXEL0, 0, 0, 0, 0, 0, 0, 0, TEXEL0, 0, 0, 0, 0);
    // Set the destination color image to the provided address
    gDPSetColorImage(gfx++, G_IM_FMT_I, G_IM_SIZ_8b, this->width, cvgDst);
    // Set up a scissor based on the source image
    gDPSetScissor(gfx++, G_SC_NON_INTERLACE, 0, 0, this->width, this->height);

    // Calculate the max number of rows that can fit into TMEM at once
    nRows = TMEM_SIZE / (this->width * G_IM_SIZ_16b_BYTES);

    // Set up the number of remaining rows
    rowsRemaining = this->height;
    curRow = 0;
    while (rowsRemaining > 0) {
        s32 uls = 0;
        s32 lrs = this->width - 1;
        s32 ult;
        s32 lrt;

        // Make sure that we don't load past the end of the source image
        if (nRows > rowsRemaining) {
            nRows = rowsRemaining;
        }

        // Determine the upper and lower bounds of the rect to draw
        ult = curRow;
        lrt = curRow + nRows - 1;

        // Load a horizontal strip of the source image in IA16 format. Since the source image is stored in memory as
        // RGBA16, the bits are reinterpreted into IA16:
        //
        // r     g      b     a
        // 11111 111 11 11111 1
        // i         a
        // 11111 111 11 11111 1
        //
        // I = (r << 3) | (g >> 2)
        // A = (g << 6) | (b << 1) | a
        //
        // Since it is expected that r = g = b = cvg in the source image, this results in
        //  I = (cvg << 3) | (cvg >> 2)
        // This expands the 5-bit coverage into an 8-bit value
        gDPLoadTextureTile(gfx++, img, G_IM_FMT_IA, G_IM_SIZ_16b, this->width, this->height, uls, ult, lrs, lrt, 0,
                           G_TX_NOMIRROR | G_TX_WRAP, G_TX_NOMIRROR | G_TX_WRAP, G_TX_NOMASK, G_TX_NOMASK, G_TX_NOLOD,
                           G_TX_NOLOD);

        // Draw that horizontal strip to the destination image. With the combiner and blender configuration set above,
        // the intensity (I) channel of the loaded IA16 texture will be written as-is to the I8 color image, each pixel
        // in the final image is
        //  I = (cvg << 3) | (cvg >> 2)
        gSPTextureRectangle(gfx++, uls << 2, ult << 2, (lrs + 1) << 2, (lrt + 1) << 2, G_TX_RENDERTILE, uls << 5,
                            ult << 5, 1 << 10, 1 << 10);

        // Update the number of rows remaining and index of the row being drawn
        curRow += nRows;
        rowsRemaining -= nRows;
    }

    // Reset the color image to the current framebuffer
    gDPPipeSync(gfx++);
    gDPSetColorImage(gfx++, G_IM_FMT_RGBA, G_IM_SIZ_16b, this->width, this->fbuf);

    *gfxP = gfx;
}

/**
 * Saves zbuf to zbufSave
 */
void PreRender_SaveZBuffer(PreRender* this, Gfx** gfxP) {
    if ((this->zbufSave != NULL) && (this->zbuf != NULL)) {
        PreRender_RestoreBuffer(this, gfxP, this->zbuf, this->zbufSave);
    }
}

/**
 * Saves fbuf to fbufSave
 */
void PreRender_SaveFramebuffer(PreRender* this, Gfx** gfxP) {
    if ((this->fbufSave != NULL) && (this->fbuf != NULL)) {
        func_80170200(this, gfxP, this->fbuf, this->fbufSave);
    }
}

/**
 * Fetches the coverage of the current framebuffer into an image of the same format as the current color image, storing
 * it over the framebuffer in memory.
 */
void PreRender_FetchFbufCoverage(PreRender* this, Gfx** gfxP) {
    Gfx* gfx = *gfxP;

    gDPPipeSync(gfx++);
    // Set the blend color to full white and set maximum depth
    gDPSetBlendColor(gfx++, 255, 255, 255, 8);
    gDPSetPrimDepth(gfx++, 0xFFFF, 0xFFFF);

    // Uses G_RM_VISCVG to blit the coverage values to the framebuffer
    //
    // G_RM_VISCVG is the following special render mode:
    //  IM_RD    : Allow read-modify-write operations on the framebuffer
    //  FORCE_BL : Apply the blender to all pixels rather than just edges, skip the division step of the blend formula
    //  (G_BL_CLR_IN * G_BL_0 + G_BL_CLR_BL * G_BL_A_MEM) = G_BL_CLR_BL * G_BL_A_MEM
    //
    // G_BL_A_MEM ("memory alpha") is coverage, therefore this blender configuration emits only the coverage (up to a
    // constant factor determined by blend color) and discards any pixel colors. For an RGBA16 framebuffer, each of the
    // three color channels r,g,b will receive the coverage value individually.
    //
    // Also disables other modes such as alpha compare and texture perspective correction
    gDPSetOtherMode(gfx++,
                    G_AD_DISABLE | G_CD_DISABLE | G_CK_NONE | G_TC_FILT | G_TF_POINT | G_TT_NONE | G_TL_TILE |
                        G_TD_CLAMP | G_TP_NONE | G_CYC_1CYCLE | G_PM_NPRIMITIVE,
                    G_AC_NONE | G_ZS_PRIM | G_RM_VISCVG | G_RM_VISCVG2);
    // Set up a scissor with the same dimensions as the framebuffer
    gDPSetScissor(gfx++, G_SC_NON_INTERLACE, 0, 0, this->width, this->height);
    // Fill rectangle to obtain the coverage values as an RGBA16 image
    gDPFillRectangle(gfx++, 0, 0, this->width, this->height);
    gDPPipeSync(gfx++);

    *gfxP = gfx;
}

/**
 * Draws the coverage of the current framebuffer `this->fbuf` to an I8 image at `this->cvgSave`. Overwrites
 * `this->fbuf` in the process.
 */
void PreRender_DrawCoverage(PreRender* this, Gfx** gfxP) {
    PreRender_FetchFbufCoverage(this, gfxP);

    if (this->cvgSave != NULL) {
        PreRender_CoverageRgba16ToI8(this, gfxP, this->fbuf, this->cvgSave);
    }
}

/**
 * Restores zbufSave to zbuf
 */
void PreRender_RestoreZBuffer(PreRender* this, Gfx** gfxP) {
    PreRender_RestoreBuffer(this, gfxP, this->zbufSave, this->zbuf);
}

/**
 * Draws a full-screen image to the current framebuffer, that sources the rgb channel from `this->fbufSave` and
 * the alpha channel from `this->cvgSave` modulated by environment color.
 */
void func_80170798(PreRender* this, Gfx** gfxP) {
    Gfx* gfx;
    s32 rowsRemaining;
    s32 curRow;
    s32 nRows;
    s32 rtile = 1;

    if (this->cvgSave != NULL) {
        gfx = *gfxP;

        gDPPipeSync(gfx++);
        gDPSetEnvColor(gfx++, 255, 255, 255, 32);
        // Effectively disable blending in both cycles. It's 2-cycle so that TEXEL1 can be used to point to a different
        // texture tile.
        gDPSetOtherMode(gfx++,
                        G_AD_DISABLE | G_CD_DISABLE | G_CK_NONE | G_TC_FILT | G_TF_POINT | G_TT_NONE | G_TL_TILE |
                            G_TD_CLAMP | G_TP_NONE | G_CYC_2CYCLE | G_PM_NPRIMITIVE,
                        G_AC_NONE | G_ZS_PRIM | AA_EN | CVG_DST_CLAMP | ZMODE_OPA | CVG_X_ALPHA |
                            GBL_c1(G_BL_CLR_IN, G_BL_0, G_BL_CLR_IN, G_BL_1) |
                            GBL_c2(G_BL_CLR_IN, G_BL_0, G_BL_CLR_IN, G_BL_1));

        // Set up the color combiner: first cycle: TEXEL0, TEXEL1 + ENVIRONMENT; second cycle: G_CC_PASS2
        gDPSetCombineLERP(gfx++, 0, 0, 0, TEXEL0, 1, 0, TEXEL1, ENVIRONMENT, 0, 0, 0, COMBINED, 0, 0, 0, COMBINED);

        nRows = (this->width > SCREEN_WIDTH) ? 2 : 4;

        rowsRemaining = this->height;
        curRow = 0;

        while (rowsRemaining > 0) {
            s32 uls = 0;
            s32 lrs = this->width - 1;
            s32 ult;
            s32 lrt;

            // Make sure that we don't load past the end of the source image
            if (nRows > rowsRemaining) {
                nRows = rowsRemaining;
            }

            // Determine the upper and lower bounds of the rect to draw
            ult = curRow;
            lrt = curRow + nRows - 1;

            // Load the frame buffer line
            gDPLoadMultiTile(gfx++, this->fbufSave, 0x0000, G_TX_RENDERTILE, G_IM_FMT_RGBA, G_IM_SIZ_16b, this->width,
                             this->height, uls, ult, lrs, lrt, 0, G_TX_NOMIRROR | G_TX_WRAP, G_TX_NOMIRROR | G_TX_WRAP,
                             G_TX_NOMASK, G_TX_NOMASK, G_TX_NOLOD, G_TX_NOLOD);

            // Load the coverage line
            gDPLoadMultiTile(gfx++, this->cvgSave, 0x0160, rtile, G_IM_FMT_I, G_IM_SIZ_8b, this->width, this->height,
                             uls, ult, lrs, lrt, 0, G_TX_NOMIRROR | G_TX_WRAP, G_TX_NOMIRROR | G_TX_WRAP, G_TX_NOMASK,
                             G_TX_NOMASK, G_TX_NOLOD, G_TX_NOLOD);

            // Draw a texture for which the rgb channels come from the framebuffer and the alpha channel comes from
            // coverage, modulated by env color
            gSPTextureRectangle(gfx++, uls << 2, ult << 2, (lrs + 1) << 2, (lrt + 1) << 2, G_TX_RENDERTILE, uls << 5,
                                ult << 5, 1 << 10, 1 << 10);

            curRow += nRows;
            rowsRemaining -= nRows;
        }

        gDPPipeSync(gfx++);
        *gfxP = gfx;
    }
}

void func_80170AE0(PreRender* this, Gfx** gfxP, s32 alpha) {
    func_8016FF90(this, gfxP, this->fbufSave, this->fbuf, 255, 255, 255, alpha);
}

/**
 * Copies fbufSave to fbuf
 */
void PreRender_RestoreFramebuffer(PreRender* this, Gfx** gfxP) {
    PreRender_RestoreBuffer(this, gfxP, this->fbufSave, this->fbuf);
}

/**
 * Applies the Video Interface anti-aliasing of silhouette edges to an image.
 *
 * This filter performs a linear interpolation on partially covered pixels between the current pixel color (called
 * foreground color) and a "background" pixel color obtained by sampling fully covered pixels at the six highlighted
 * points in the following 5x3 neighborhood:
 *    _ _ _ _ _
 *  |   o   o   |
 *  | o   X   o |
 *  |   o   o   |
 *    ‾ ‾ ‾ ‾ ‾
 * Whether a pixel is partially covered is determined by reading the coverage values associated with the image.
 * Coverage is a measure of how many subpixels the last drawn primitive covered. A fully covered pixel is one with a
 * full coverage value, the entire pixel was covered by the primitive.
 * The background color is calculated as the average of the "penultimate" minimum and maximum colors in the 5x3
 * neighborhood.
 *
 * The final color is calculated by interpolating the foreground and background color weighted by the coverage:
 *      OutputColor = cvg * ForeGround + (1.0 - cvg) * BackGround
 *
 * This is a software implementation of the same algorithm used in the Video Interface hardware when Anti-Aliasing is
 * enabled in the VI Control Register.
 *
 * Patent describing the algorithm:
 *
 * Gossett, C. P., & van Hook, T. J. (Filed 1995, Published 1998)
 * Antialiasing of silhouette edges (USOO5742277A)
 * U.S. Patent and Trademark Office
 * Expired 2015-10-06
 * https://patents.google.com/patent/US5742277A/en
 *
 * @param this  PreRender instance
 * @param x     Center pixel x
 * @param y     Center pixel y
 */
void PreRender_AntiAliasFilterPixel(PreRender* this, s32 x, s32 y) {
    s32 i;
    s32 j;
    s32 buffCvg[3 * 5];
    s32 buffR[3 * 5];
    s32 buffG[3 * 5];
    s32 buffB[3 * 5];
    s32 xi;
    s32 yi;
    s32 temp;
    s32 pmaxR;
    s32 pmaxG;
    s32 pmaxB;
    s32 pminR;
    s32 pminG;
    s32 pminB;
    Color_RGBA16 pxIn;
    Color_RGBA16 pxOut;
    u32 outR;
    u32 outG;
    u32 outB;

    // Extract pixels in the 5x3 neighborhood
    for (i = 0; i < 5 * 3; i++) {
        xi = x + (i % 5) - 2;
        yi = y + (i / 5) - 1;

        // Clamp coordinates to the edges of the image
        if (xi < 0) {
            xi = 0;
        } else if (xi > (this->width - 1)) {
            xi = this->width - 1;
        }
        if (yi < 0) {
            yi = 0;
        } else if (yi > (this->height - 1)) {
            yi = this->height - 1;
        }

        // Extract color channels for each pixel, convert 5-bit color channels to 8-bit
        pxIn.rgba = this->fbufSave[xi + yi * this->width];
        buffR[i] = (pxIn.r << 3) | (pxIn.r >> 2);
        buffG[i] = (pxIn.g << 3) | (pxIn.g >> 2);
        buffB[i] = (pxIn.b << 3) | (pxIn.b >> 2);
        buffCvg[i] = this->cvgSave[xi + yi * this->width] >> 5;
    }

    pmaxR = pminR = buffR[7];
    pmaxG = pminG = buffG[7];
    pmaxB = pminB = buffB[7];

    // For each neighbor
    for (i = 1; i < 5 * 3; i += 2) {
        // Only sample fully covered pixels
        if (buffCvg[i] != 7) {
            continue;
        }
        // Determine "Penultimate Maximum" Value

        // If current maximum is less than this neighbor
        if (pmaxR < buffR[i]) {
            // For each neighbor (again)
            for (j = 1; j < 5 * 3; j += 2) {
                // If not the neighbor we were at before, and this neighbor has a larger value and this pixel is
                // fully covered, that means the neighbor at `i` is the "penultimate maximum"
                if ((i != j) && (buffR[j] >= buffR[i]) && (buffCvg[j] == 7)) {
                    pmaxR = buffR[i];
                }
            }
        }
        if (pmaxG < buffG[i]) {
            for (j = 1; j < 5 * 3; j += 2) {
                if ((i != j) && (buffG[j] >= buffG[i]) && (buffCvg[j] == 7)) {
                    pmaxG = buffG[i];
                }
            }
        }
        if (pmaxB < buffB[i]) {
            for (j = 1; j < 5 * 3; j += 2) {
                if ((i != j) && (buffB[j] >= buffB[i]) && (buffCvg[j] == 7)) {
                    pmaxB = buffB[i];
                }
            }
        }

        // Determine "Penultimate Minimum" Value

        // Same as above with inverted conditions
        if (pminR > buffR[i]) {
            for (j = 1; j < 5 * 3; j += 2) {
                if ((i != j) && (buffR[j] <= buffR[i]) && (buffCvg[j] == 7)) {
                    pminR = buffR[i];
                }
            }
        }
        if (pminG > buffG[i]) {
            for (j = 1; j < 5 * 3; j += 2) {
                if ((i != j) && (buffG[j] <= buffG[i]) && (buffCvg[j] == 7)) {
                    pminG = buffG[i];
                }
            }
        }
        if (pminB > buffB[i]) {
            for (j = 1; j < 5 * 3; j += 2) {
                if ((i != j) && (buffB[j] <= buffB[i]) && (buffCvg[j] == 7)) {
                    pminB = buffB[i];
                }
            }
        }
    }

    // The background color is determined by averaging the penultimate minimum and maximum pixels, and subtracting the
    // Foreground color:
    //      BackGround = (pMax + pMin) - (ForeGround) * 2

    // OutputColor = cvg * ForeGround + (1.0 - cvg) * BackGround
    temp = 7 - buffCvg[7];
    outR = buffR[7] + ((s32)(temp * (pmaxR + pminR - (buffR[7] * 2)) + 4) >> 3);
    outG = buffG[7] + ((s32)(temp * (pmaxG + pminG - (buffG[7] * 2)) + 4) >> 3);
    outB = buffB[7] + ((s32)(temp * (pmaxB + pminB - (buffB[7] * 2)) + 4) >> 3);

    pxOut.r = outR >> 3;
    pxOut.g = outG >> 3;
    pxOut.b = outB >> 3;
    pxOut.a = 1;
    this->fbufSave[x + y * this->width] = pxOut.rgba;
}

/**
 * Applies the Video Interface anti-aliasing filter to `this->fbufSave` using `this->cvgSave`
 */
void PreRender_AntiAliasFilter(PreRender* this) {
    s32 x;
    s32 y;
    s32 cvg;

    // Apply AA filter
    for (y = 0; y < this->height; y++) {
        for (x = 0; x < this->width; x++) {
            cvg = this->cvgSave[x + y * this->width];
            cvg >>= 5;
            cvg++;

            if (cvg != 8) {
                // If this pixel has only partial coverage, perform the Video Filter interpolation on it
                PreRender_AntiAliasFilterPixel(this, x, y);
            }
        }
    }
}

/**
 * Selects the median value from 9 different 5-bit pixels:
 * px1[0], px1[1], px1[2], px2[0], px2[1], px2[2], px3[0], px3[1], px3[2]
 * all args are expected to be an array of 3 different 5-bit values
 */
u32 PreRender_Get5bMedian9(u8* px1, u8* px2, u8* px3) {
    u8 pxValCount[32]; // Stores the count for each of the possible 32 5-bit pixel values
    u32 pxCount;       // Pixel count
    s32 pxMed;         // Pixel median value

    // Initialize count to 0 in groups of 4 bits
    *(s32*)(&pxValCount[0]) = 0;
    *(s32*)(&pxValCount[4]) = 0;
    *(s32*)(&pxValCount[8]) = 0;
    *(s32*)(&pxValCount[12]) = 0;
    *(s32*)(&pxValCount[16]) = 0;
    *(s32*)(&pxValCount[20]) = 0;
    *(s32*)(&pxValCount[24]) = 0;
    *(s32*)(&pxValCount[28]) = 0;

    // Increment the count that contains the pixel values
    pxValCount[px1[0]]++;
    pxValCount[px1[1]]++;
    pxValCount[px1[2]]++;

    pxValCount[px2[0]]++;
    pxValCount[px2[1]]++;
    pxValCount[px2[2]]++;

    pxValCount[px3[0]]++;
    pxValCount[px3[1]]++;
    pxValCount[px3[2]]++;

    // Loop through the 32 bits until 5 values are found, then return that bit value.
    // Note that the median of 9 is the 5th sequential value.
    pxCount = 0;
    pxMed = 0;
    while (true) {
        pxCount += pxValCount[pxMed];
        if (pxCount >= 5) {
            // the median is found
            break;
        }
        pxMed++;
    }

    return pxMed;
}

// Despite the name, this function doesn't seem like an hardware-accurate divot filter
void PreRender_DivotFilter(PreRender* this) {
    u32 width = this->width;
    u32 height = this->height;
    u8* buffer = malloc(width * 10);
    u8* redRow[3];
    u8* greenRow[3];
    u8* blueRow[3];
    u8* cvgFull;
    Color_RGBA16 inPx;
    Color_RGBA16 outPx;
    u32 x;
    u32 y;
    s32 pad;

    redRow[0] = &buffer[width * 0];
    redRow[1] = &buffer[width * 1];
    redRow[2] = &buffer[width * 2];

    greenRow[0] = &buffer[width * 3];
    greenRow[1] = &buffer[width * 4];
    greenRow[2] = &buffer[width * 5];

    blueRow[0] = &buffer[width * 6];
    blueRow[1] = &buffer[width * 7];
    blueRow[2] = &buffer[width * 8];

    cvgFull = &buffer[width * 9];

    // Fill line buffers for first 2 rows
    for (y = 0; y < 2; y++) {
        for (x = 0; x < width; x++) {
            inPx.rgba = this->fbufSave[x + y * this->width];

            redRow[y][x] = inPx.r;
            greenRow[y][x] = inPx.g;
            blueRow[y][x] = inPx.b;
        }
    }

    // For each row in the image, except first and last
    for (y = 1; y < height - 1; y++) {
        // Find start of pixels and coverage for current line (bug? this should probably be fetching the NEXT line, but
        // really the divot filter only cares about individual lines so it's already wrong)
        u8* redRow2 = redRow[2];
        u8* greenRow2 = greenRow[2];
        u8* blueRow2 = blueRow[2];
        u8* lineCvg = &this->cvgSave[width * y];
        u16* linePx = &this->fbufSave[width * y];

        // Obtain next row from current line, current line becomes the bottom row?? (weird, you would expect this to
        // sample the NEXT line?)
        for (x = 0; x < width; x++) {
            inPx.rgba = linePx[x];

            redRow2[x] = inPx.r;
            greenRow2[x] = inPx.g;
            blueRow2[x] = inPx.b;

            // checking for full coverage
            cvgFull[x] = (lineCvg[x] >> 5) == 7;
        }

        for (x = 1; x < width - 1; x++) {
            // if the coverage of the three adjacent pixels on the current line are not all fully covered
            if (cvgFull[x - 1] && cvgFull[x] && cvgFull[x + 1]) {
                continue;
            }

            // find median value in 3x3 square for each (r,g,b), replaces the pixel marked by X:
            //  * * *
            //  * * *
            //  * X *
            outPx.r = PreRender_Get5bMedian9(&redRow[0][x - 1], &redRow[1][x - 1], &redRow[2][x - 1]);
            outPx.g = PreRender_Get5bMedian9(&greenRow[0][x - 1], &greenRow[1][x - 1], &greenRow[2][x - 1]);
            outPx.b = PreRender_Get5bMedian9(&blueRow[0][x - 1], &blueRow[1][x - 1], &blueRow[2][x - 1]);
            outPx.a = 1;

            this->fbufSave[x + y * this->width] = outPx.rgba;
        }

        // Shuffle row 1 -> 0
        redRow[0] = redRow[1];
        greenRow[0] = greenRow[1];
        blueRow[0] = blueRow[1];
        // Shuffle row 2 -> 1
        redRow[1] = redRow[2];
        greenRow[1] = greenRow[2];
        blueRow[1] = blueRow[2];
    }
    free(buffer);
}

/**
 * Applies filters to the framebuffer prerender to make it look smoother
 */
void PreRender_ApplyFilters(PreRender* this) {
    if ((this->cvgSave == NULL) || (this->fbufSave == NULL)) {
        this->filterState = PRERENDER_FILTER_STATE_NONE;
    } else {
        this->filterState = PRERENDER_FILTER_STATE_PROCESS;
        PreRender_AntiAliasFilter(this);
        PreRender_DivotFilter(this);
        this->filterState = PRERENDER_FILTER_STATE_DONE;
    }
}

SlowlyMgr sSlowlyMgr;
s32 sSlowlyRunning;
StackEntry sSlowlyStackInfo;
STACK(sSlowlyStack, 0x1000);

/**
 * Initializes `PreRender_ApplyFilters` onto a new "slowly" thread
 */
void PreRender_ApplyFiltersSlowlyInit(PreRender* this) {
    if ((this->cvgSave != NULL) && (this->fbufSave != NULL)) {
        if (sSlowlyRunning) {
            StackCheck_Cleanup(&sSlowlyStackInfo);
            Slowly_Destroy(&sSlowlyMgr);
        }

        this->filterState = PRERENDER_FILTER_STATE_PROCESS;
        StackCheck_Init(&sSlowlyStackInfo, sSlowlyStack, STACK_TOP(sSlowlyStack), 0, 0x100, "slowly");
        Slowly_Init(&sSlowlyMgr, STACK_TOP(sSlowlyStack), (void*)PreRender_ApplyFilters, this, NULL);
        sSlowlyRunning = true;
    }
}

/**
 * Destroys the "slowly" thread
 */
void PreRender_ApplyFiltersSlowlyDestroy(PreRender* this) {
    if (sSlowlyRunning) {
        StackCheck_Cleanup(&sSlowlyStackInfo);
        Slowly_Destroy(&sSlowlyMgr);
        sSlowlyRunning = false;
    }
}

// Unused, likely since `PreRender_ApplyFilters` already handles NULL checks
void func_801720C4(PreRender* this) {
    if ((this->cvgSave != NULL) && (this->fbufSave != NULL)) {
        PreRender_ApplyFilters(this);
    }
}

typedef struct {
    /* 0x00 */ void* timg;
    /* 0x04 */ void* tlut;
    /* 0x08 */ u16 width;
    /* 0x0A */ u16 height;
    /* 0x0C */ u8 fmt;
    /* 0x0D */ u8 siz;
    /* 0x0E */ u16 tt;
    /* 0x10 */ u16 tlutCount;
    /* 0x14 */ f32 x;
    /* 0x18 */ f32 y;
    /* 0x1C */ f32 xScale;
    /* 0x20 */ f32 yScale;
    /* 0x24 */ u32 flags;
} PreRenderBackground2DParams; // size = 0x28

void Prerender_DrawBackground2DImpl(PreRenderBackground2DParams* bg2D, Gfx** gfxP) {
    Gfx* gfx;
    uObjBg* bg;
    u32 alphaCompare;
    Gfx* gfxTemp;
    u32 loadS2DEX2;

    loadS2DEX2 = (bg2D->flags & BG2D_FLAGS_LOAD_S2DEX2) != 0;
    alphaCompare = (bg2D->flags & BG2D_FLAGS_AC_THRESHOLD) ? G_AC_THRESHOLD : G_AC_NONE;

    gfxTemp = *gfxP;
    bg = Graph_DlistAlloc(&gfxTemp, sizeof(uObjBg));
    gfx = gfxTemp;

    bg->b.imageX = 0;
    bg->b.imageW = (bg2D->width * (1 << 2)) + 1;
    bg->b.frameX = bg2D->x * (1 << 2);

    bg->b.imageY = 0;
    bg->b.imageH = (bg2D->height * (1 << 2)) + 1;
    bg->b.frameY = bg2D->y * (1 << 2);

    bg->b.imagePtr = bg2D->timg;
    bg->b.imageLoad = G_BGLT_LOADTILE;
    bg->b.imageFmt = bg2D->fmt;
    bg->b.imageSiz = bg2D->siz;
    bg->b.imagePal = 0;
    bg->b.imageFlip = 0;

    if (loadS2DEX2) {
        gSPLoadUcodeL(gfx++, ucode_s2dex);
    }

    if ((bg2D->fmt == G_IM_FMT_CI) && (bg2D->tlut != NULL)) {
        gDPLoadTLUT(gfx++, bg2D->tlutCount, 256, bg2D->tlut);
    } else {
        gDPPipeSync(gfx++);
    }

    if (bg2D->flags & BG2D_FLAGS_COPY) {
        bg->b.frameW = bg2D->width * (1 << 2);
        bg->b.frameH = bg2D->height * (1 << 2);

        guS2DInitBg(bg);

        if (!(bg2D->flags & BG2D_FLAGS_1)) {
            gDPSetOtherMode(gfx++, bg2D->tt | G_CYC_COPY, alphaCompare);
        }

        gSPBgRectCopy(gfx++, bg);
    } else {
        bg->b.frameW = (u32)(bg2D->width * (1 << 2)) * bg2D->xScale;
        bg->b.frameH = (u32)(bg2D->height * (1 << 2)) * bg2D->yScale;
        bg->b.tmemW = (1 << 10) / bg2D->xScale;
        bg->b.tmemH = (1 << 10) / bg2D->yScale;
        bg->s.imageYorig = bg->b.imageY;

        if (!(bg2D->flags & BG2D_FLAGS_1)) {
            gDPSetOtherMode(gfx++, bg2D->tt | G_AD_DISABLE | G_CD_DISABLE | G_TC_FILT,
                            AA_EN | CVG_X_ALPHA | ALPHA_CVG_SEL |
                                GBL_c1(G_BL_CLR_IN, G_BL_A_IN, G_BL_CLR_BL, G_BL_1MA) |
                                GBL_c2(G_BL_CLR_IN, G_BL_A_IN, G_BL_CLR_BL, G_BL_1MA) | alphaCompare);
        }

        if (!(bg2D->flags & BG2D_FLAGS_2)) {
            gDPSetCombineLERP(gfx++, 0, 0, 0, TEXEL0, 0, 0, 0, 1, 0, 0, 0, TEXEL0, 0, 0, 0, 1);
        }

        gSPObjRenderMode(gfx++, G_OBJRM_ANTIALIAS | G_OBJRM_BILERP);
        gSPBgRect1Cyc(gfx++, bg);
    }

    gDPPipeSync(gfx++);

    if (loadS2DEX2) {
        gSPLoadUcode(gfx++, SysUcode_GetUCode());
    }

    *gfxP = gfx;
}

void Prerender_DrawBackground2D(Gfx** gfxP, void* timg, void* tlut, u16 width, u16 height, u8 fmt, u8 siz, u16 tt,
                                u16 tlutCount, f32 x, f32 y, f32 xScale, f32 yScale, u32 flags) {
    PreRenderBackground2DParams bg2D;
    PreRenderBackground2DParams* bg2DPtr = &bg2D;

    bg2D.timg = timg;
    bg2D.tlut = tlut;
    bg2D.width = width;
    bg2D.height = height;
    bg2D.fmt = fmt;
    bg2D.siz = siz;
    bg2D.tt = tt;
    bg2D.tlutCount = tlutCount;
    bg2D.x = x;
    bg2D.y = y;
    bg2D.xScale = xScale;
    bg2D.yScale = yScale;
    bg2D.flags = flags;

    Prerender_DrawBackground2DImpl(bg2DPtr, gfxP);
}