dusk/src/display/texture.c

/**
 * Copyright (c) 2025 Dominic Masters
 * 
 * This software is released under the MIT License.
 * https://opensource.org/licenses/MIT
 */

#include "texture.h"
#include "assert/assert.h"
#include "util/memory.h"
#include "util/math.h"
#include "display/palette/palettelist.h"

const texture_t *TEXTURE_BOUND = NULL;

void textureInit(
  texture_t *texture,
  const int32_t width,
  const int32_t height,
  const textureformat_t format,
  const texturedata_t data
) {
  assertNotNull(texture, "Texture cannot be NULL");
  assertTrue(width > 0 && height > 0, "width/height must be greater than 0");

  memoryZero(texture, sizeof(texture_t));
  texture->width = width;
  texture->height = height;

  #if PSP
    assertTrue(
      width == mathNextPowTwo(width),
      "Width must be powers of 2 for PSP"
    );
    assertTrue(
      height == mathNextPowTwo(height),
      "Height must be powers of 2 for PSP"
    );
  #endif

  #if DISPLAY_SDL2
    glGenTextures(1, &texture->id);
    glBindTexture(GL_TEXTURE_2D, texture->id);

    switch(format) {
      case TEXTURE_FORMAT_RGBA:
        glTexImage2D(
          GL_TEXTURE_2D, 0, format, width, height, 0,
          format, GL_UNSIGNED_BYTE, (void*)data.rgba.colors
        );
        break;

      case TEXTURE_FORMAT_ALPHA:
        glTexImage2D(
          GL_TEXTURE_2D, 0, format, width, height, 0,
          format, GL_UNSIGNED_BYTE, (void*)data.alpha.data
        );
        break;

      case TEXTURE_FORMAT_PALETTE:
        assertNotNull(data.palette.data, "Palette texture data cannot be NULL");
        assertTrue(
          data.palette.palette < PALETTE_LIST_COUNT,
          "Palette index out of range"
        );

        // Get and validate the palette.
        const palette_t *pal = PALETTE_LIST[data.palette.palette];
        assertNotNull(pal, "Palette cannot be NULL");
        GLenum err = glGetError();
        if(err != GL_NO_ERROR) {
          assertUnreachable("GL Error before setting color table");
        }

        // Do we support paletted textures?
        bool_t havePalTex = false;

        #if PSP
          havePalTex = true;
        #else
          GLint mask = 0;
          glGetIntegerv(GL_CONTEXT_PROFILE_MASK, &mask);
          if(mask & GL_CONTEXT_CORE_PROFILE_BIT) {
            GLint n=0; glGetIntegerv(GL_NUM_EXTENSIONS, &n);
            for(GLint i=0; i<n; ++i) {
              const char* e = (const char*)glGetStringi(GL_EXTENSIONS, i);
              if(e && strcmp(e, "GL_EXT_paletted_texture")==0) { havePalTex = true; break; }
            }
          } else {
            const char* ext = (const char*)glGetString(GL_EXTENSIONS);
            havePalTex = ext && strstr(ext, "GL_EXT_paletted_texture");
          }
        #endif

        if(!havePalTex) {
          // Not supported, convert to RGBA using lookup
          color_t formatted[width * height];
          for(int32_t i = 0; i < width * height; i++) {
            uint8_t index = data.palette.data[i];
            assertTrue(
              index < pal->colorCount,
              "Palette index out of range"
            );
            formatted[i] = pal->colors[index];
          }
          glTexImage2D(
            GL_TEXTURE_2D, 0, GL_RGBA, width, height, 0,
            GL_RGBA, GL_UNSIGNED_BYTE, (void*)formatted
          );
        } else {
          // Supported.
          #if PSP
            // PSP requires the color table itself to be a power of two
            assertTrue(
              pal->colorCount == mathNextPowTwo(pal->colorCount),
              "Palette color count must be a power of 2 for PSP"
            );
          #endif

          glTexImage2D(
            GL_TEXTURE_2D,
            0, GL_COLOR_INDEX8_EXT,
            width, height,
            0, GL_COLOR_INDEX8_EXT,
            GL_UNSIGNED_BYTE, (void*)data.palette.data
          );

          glColorTableEXT(
            GL_TEXTURE_2D, GL_RGBA, pal->colorCount, GL_RGBA,
            GL_UNSIGNED_BYTE, (const void*)pal->colors
          );
        }

        break;

      default:
        assertUnreachable("Unknown texture format");
        break;
    }

    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
    
    glBindTexture(GL_TEXTURE_2D, 0);
    texture->ready = true;
  
  #elif DOLPHIN
    texture->format = format;

    switch(format) {
      case TEXTURE_FORMAT_RGBA:
        assertTrue(
          (width % 4) == 0 && (height % 4) == 0,
          "RGB5A3 requires w/h multiple of 4 (or pad)"
        );

        // Convert to RGB5A3 format
        size_t rgbaSize = width * height * sizeof(u16);
        texture->rgba = (u16*)memalign(32, rgbaSize);
        assertNotNull(texture->rgba, "Failed to allocate texture RGBA data");

        for(uint32_t y = 0; y < height; ++y) {
          for(uint32_t x = 0; x < width; ++x) {
            const int src = y * width + x;

            const int tileX = x >> 2;
            const int tileY = y >> 2;
            const int tilesPerRow = width >> 2;
            const int tileIndex = tileY * tilesPerRow + tileX;
            const int tileBaseWords = tileIndex * 16;
            const int inTile = ((y & 3) << 2) + (x & 3);
            const int dest = tileBaseWords + inTile;
            
            color4b_t col = data.rgba.colors[src];

            u16 outCol;
            if(col.a < 255) {
              // 0AAA RRRR GGGG BBBB
              outCol = (
                (0u << 15) |
                ((u16)(col.a >> 5) << 12) |
                ((u16)(col.r >> 4) << 8) |
                ((u16)(col.g >> 4) << 4) |
                ((u16)(col.b >> 4) << 0)
              );
            } else {
              // 1RRRR RRGG GGGB BBBB
              outCol = (
                (1u << 15) |
                ((u16)(col.r >> 3) << 10) |
                ((u16)(col.g >> 3) << 5) |
                ((u16)(col.b >> 3) << 0)
              );
            }
            texture->rgba[dest] = outCol;
          }
        }

        DCFlushRange(texture->rgba, rgbaSize);
        GX_InitTexObj(
          &texture->texObj,
          texture->rgba,
          width, height,
          GX_TF_RGB5A3,
          GX_CLAMP, GX_CLAMP,
          GX_FALSE
        );

        DCFlushRange(texture->rgba, rgbaSize);
        GX_InvalidateTexAll();

        GX_InitTexObjLOD(
          &texture->texObj,
          GX_NEAR, GX_NEAR,
          0.0f, 0.0f, 0.0f,
          GX_FALSE,
          GX_FALSE,
          GX_ANISO_1
        );
        break;

      case TEXTURE_FORMAT_ALPHA:
        assertTrue(
          (width % 4) == 0 && (height % 4) == 0,
          "GX_TF_I8 requires w/h multiple of 4 (or pad)"
        );

        // 1 byte per pixel (I8), GX expects 4x4 tiled layout
        const size_t alphaSize = (size_t)width * (size_t)height;

        texture->alpha = (u8*)memalign(32, alphaSize);
        assertNotNull(texture->alpha, "Failed to allocate alpha texture data");

        const u32 tilesPerRow = ((u32)width) >> 3; // /8

        for (u32 y = 0; y < (u32)height; ++y) {
          const u32 tileY   = y >> 2;        // /4
          const u32 inTileY = (y & 3) << 3;  // (y%4)*8

          for (u32 x = 0; x < (u32)width; ++x) {
            const u32 srcI = y * (u32)width + x;
            const u8  srcA = data.alpha.data[srcI]; // linear input

            const u32 tileX = x >> 3;               // /8
            const u32 tileIndex = tileY * tilesPerRow + tileX;

            const u32 tileBase = tileIndex * 32;    // 8*4*1 = 32 bytes per tile
            const u32 inTile   = inTileY + (x & 7); // (y%4)*8 + (x%8)

            texture->alpha[tileBase + inTile] = 0xFF - srcA;// Fixes inverted alpha.
          }
        }

        // Flush CPU cache so GX sees the swizzled I8 texture data
        DCFlushRange(texture->alpha, alphaSize);

        // Initialize GX texture object with swizzled data
        GX_InitTexObj(
          &texture->texObj,
          texture->alpha,
          width, height,
          GX_TF_I8,
          GX_CLAMP, GX_CLAMP,
          GX_FALSE
        );

        GX_InitTexObjLOD(
          &texture->texObj,
          GX_NEAR, GX_NEAR,
          0.0f, 0.0f, 0.0f,
          GX_FALSE,
          GX_FALSE,
          GX_ANISO_1
        );
        break;

      default:
        assertUnreachable("Unsupported texture format for Dolphin");
        break;
    }

    texture->ready = true;
  #endif
}

void textureBind(texture_t *texture) {
  if(TEXTURE_BOUND == texture) return;

  if(texture == NULL) {
    #if DISPLAY_SDL2
      glDisable(GL_TEXTURE_2D);
    #endif
    TEXTURE_BOUND = NULL;
    return;
  }

  assertTrue(texture->ready, "Texture ID must be ready");
  assertTrue(
    texture->width > 0 && texture->height > 0,
    "Texture width and height must be greater than 0"
  );

  #if DISPLAY_SDL2
    glEnable(GL_TEXTURE_2D);
    glBindTexture(GL_TEXTURE_2D, texture->id);
    glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);
  #elif DOLPHIN
    GX_LoadTexObj(&texture->texObj, GX_TEXMAP0);
  #endif
  TEXTURE_BOUND = texture;
}

void textureDispose(texture_t *texture) {
  assertNotNull(texture, "Texture cannot be NULL");
  assertTrue(texture->ready, "Texture ID must be ready");

  if(TEXTURE_BOUND == texture) {
    textureBind(NULL);
  }

  #if DISPLAY_SDL2
    glDeleteTextures(1, &texture->id);
  #elif DOLPHIN
    switch(texture->format) {
      case TEXTURE_FORMAT_RGBA:
        free(texture->rgba);
        break;

      case TEXTURE_FORMAT_ALPHA:
        free(texture->alpha);
        break;

      default:
        break;
    }

    memoryZero(texture, sizeof(texture_t));
  #endif
}

#if DOLPHIN
  void textureDolphinUploadTEV() {
    // Add channel for vertex color
    GX_SetNumChans(1);
    GX_SetChanCtrl(
      GX_COLOR0A0,// Store in color channel 0
      GX_DISABLE,// Lighting disabled
      GX_SRC_REG,// Ambient color?
      GX_SRC_VTX,// Material color?
      GX_LIGHTNULL,// Light Mask
      GX_DF_NONE,// Diffuse function
      GX_AF_NONE// Attenuation function
    );

    if(!TEXTURE_BOUND) {
      GX_SetNumTexGens(0);
      GX_SetNumTevStages(1);
      GX_SetTevOp(GX_TEVSTAGE0, GX_PASSCLR);
      return;
    }

    // One set of UVs
    GX_SetNumTexGens(1);
    GX_SetTexCoordGen(
      GX_TEXCOORD0,
      GX_TG_MTX2x4,
      GX_TG_TEX0,
      GX_IDENTITY
    );
  
    // Basically the shader setup
    switch(TEXTURE_BOUND->format) {
      case TEXTURE_FORMAT_RGBA:
        // One TEV stage: vertex color * texture color
        GX_SetNumTevStages(1);
        GX_SetTevOrder(
          GX_TEVSTAGE0,
          GX_TEXCOORD0,
          GX_TEXMAP0,
          GX_COLOR0A0
        );
        GX_SetTevOp(GX_TEVSTAGE0, GX_MODULATE);
        break;

      case TEXTURE_FORMAT_ALPHA:
        // One TEV stage: vertex color * texture color
        GX_SetNumTevStages(1);
        GX_SetTevOrder(
          GX_TEVSTAGE0,
          GX_TEXCOORD0,
          GX_TEXMAP0,
          GX_COLOR0A0
        );

        // Color = vertex color
        GX_SetTevColorIn(
          GX_TEVSTAGE0,
          GX_CC_RASC,
          GX_CC_ZERO,
          GX_CC_ZERO,
          GX_CC_ZERO
        );
        GX_SetTevColorOp(
          GX_TEVSTAGE0,
          GX_TEV_ADD,
          GX_TB_ZERO,
          GX_CS_SCALE_1,
          GX_TRUE,
          GX_TEVPREV
        );

        // Alpha = vertex alpha * I8 intensity
        GX_SetTevAlphaIn(
          GX_TEVSTAGE0,
          GX_CA_RASA,
          GX_CA_ZERO,
          GX_CA_TEXA,
          GX_CA_ZERO
        );
        GX_SetTevAlphaOp(
          GX_TEVSTAGE0,
          GX_TEV_ADD,
          GX_TB_ZERO,
          GX_CS_SCALE_1,
          GX_TRUE,
          GX_TEVPREV
        );

        GX_SetBlendMode(
          GX_BM_BLEND,
          GX_BL_SRCALPHA,
          GX_BL_INVSRCALPHA,
          GX_LO_CLEAR
        );

        GX_SetColorUpdate(GX_TRUE);
        GX_SetAlphaUpdate(GX_TRUE);

        break;

      default:
        assertUnreachable("Unknown texture format in meshDraw");
        break;
    }
  }
#endif