Ditch unused code and features that arent in subject to get supported any near time.
[mesa.git] / src / mesa / drivers / dri / savage / savagetex.c
index 6641231371e6d3e2990d1e35be4bf9c0bb468746..5584e371e69e5d2cd7485941261c0763cc1765be 100644 (file)
 #include "texstore.h"
 #include "texobj.h"
 
+#include "convolve.h"
+#include "colormac.h"
+
 #include "swrast/swrast.h"
 
 #include "xmlpool.h"
 
+#define TILE_INDEX_DXT1 0
+#define TILE_INDEX_8    1
+#define TILE_INDEX_16   2
+#define TILE_INDEX_DXTn 3
+#define TILE_INDEX_32   4
+
 /* On Savage4 the texure LOD-bias needs an offset of ~ 0.3 to get
  * somewhere close to software rendering.
  */
 #define SAVAGE4_LOD_OFFSET 10
 
+/* Tile info for S3TC formats counts in 4x4 blocks instead of texels.
+ * In DXT1 each block is encoded in 64 bits. In DXT3 and 5 each block is
+ * encoded in 128 bits. */
+
 /* Size 1, 2 and 4 images are packed into the last subtile. Each image
  * is repeated to fill a 4x4 pixel area. The figure below shows the
  * layout of those 4x4 pixel areas in the 8x8 subtile.
  * Yuck! 8-bit texture formats use 4x8 subtiles. See below.
  */
 static const savageTileInfo tileInfo_pro[5] = {
-    {64, 64,  8, 8, 8, 8, {0x12, 0x02}}, /* 4-bit */
+    {16, 16, 16, 8, 1, 2, {0x18, 0x10}}, /* DXT1 */
     {64, 32, 16, 4, 4, 8, {0x30, 0x20}}, /* 8-bit */
     {64, 16,  8, 2, 8, 8, {0x48, 0x08}}, /* 16-bit */
-    { 0,  0,  0, 0, 0, 0, {0x00, 0x00}}, /* 24-bit */
+    {16,  8, 16, 4, 1, 2, {0x30, 0x20}}, /* DXT3, DXT5 */
     {32, 16,  4, 2, 8, 8, {0x90, 0x10}}, /* 32-bit */
 };
 
@@ -76,10 +89,10 @@ static const savageTileInfo tileInfo_pro[5] = {
  *                      x                 1
  */
 static const savageTileInfo tileInfo_s3d_s4[5] = {
-    {64, 64, 16, 8, 4, 8, {0x18, 0x10}}, /* 4-bit */
+    {16, 16, 16, 8, 1, 2, {0x18, 0x10}}, /* DXT1 */
     {64, 32, 16, 4, 4, 8, {0x30, 0x20}}, /* 8-bit */
     {64, 16, 16, 2, 4, 8, {0x60, 0x40}}, /* 16-bit */
-    { 0,  0,  0, 0, 0, 0, {0x00, 0x00}}, /* 24-bit */
+    {16,  8, 16, 4, 1, 2, {0x30, 0x20}}, /* DXT3, DXT5 */
     {32, 16,  8, 2, 4, 8, {0xc0, 0x80}}, /* 32-bit */
 };
 
@@ -106,6 +119,9 @@ SUBTILE_FUNC(8, 8)
 SUBTILE_FUNC(16, 8)
 SUBTILE_FUNC(32, 8) /* 4 bytes per pixel, 8 pixels wide */
 
+SUBTILE_FUNC(8, 2) /* DXT1 */
+SUBTILE_FUNC(16, 2) /* DXT3 and DXT5 */
+
 /** \brief Upload a complete tile from src (srcStride) to dest
  *
  * \param tileInfo     Pointer to tiling information
@@ -135,8 +151,10 @@ static void savageUploadTile (const savageTileInfo *tileInfo,
     switch (subStride) {
     case  2: subtileFunc = savageUploadSubtile_2x8; break;
     case  4: subtileFunc = savageUploadSubtile_4x8; break;
-    case  8: subtileFunc = savageUploadSubtile_8x8; break;
-    case 16: subtileFunc = savageUploadSubtile_16x8; break;
+    case  8: subtileFunc = tileInfo->subHeight == 8 ?
+                savageUploadSubtile_8x8 : savageUploadSubtile_8x2; break;
+    case 16: subtileFunc = tileInfo->subHeight == 8 ?
+                savageUploadSubtile_16x8 : savageUploadSubtile_16x2; break;
     case 32: subtileFunc = savageUploadSubtile_32x8; break;
     default: assert(0);
     }
@@ -169,9 +187,10 @@ static void savageUploadTile (const savageTileInfo *tileInfo,
  * FIXME: Repeating inside this function would be more efficient.
  */
 static void savageUploadTiny (const savageTileInfo *tileInfo,
+                             GLuint pixWidth, GLuint pixHeight,
                              GLuint width, GLuint height, GLuint bpp,
                              GLubyte *src, GLubyte *dest) {
-    GLuint size = MAX2(width, height);
+    GLuint size = MAX2(pixWidth, pixHeight);
 
     if (width > tileInfo->subWidth) { /* assert: height <= subtile height */
        GLuint wInSub = width / tileInfo->subWidth;
@@ -194,14 +213,15 @@ static void savageUploadTiny (const savageTileInfo *tileInfo,
        GLuint srcStride = width * bpp;
        GLuint subStride = tileInfo->subWidth * bpp;
        /* if the subtile width is 4 we have to skip every other subtile */
-       GLuint subSkip = tileInfo->subWidth == 4 ?
+       GLuint subSkip = tileInfo->subWidth <= 4 ?
            subStride * tileInfo->subHeight : 0;
+       GLuint skipRemainder = tileInfo->subHeight - 1;
        GLuint y;
        for (y = 0; y < height; ++y) {
            memcpy (dest, src, srcStride);
            src += srcStride;
            dest += subStride;
-           if ((y & 7) == 7)
+           if ((y & skipRemainder) == skipRemainder)
                dest += subSkip;
        }
     } else { /* the last 3 mipmap levels */
@@ -223,8 +243,9 @@ static void savageUploadTexLevel( savageTexObjPtr t, int level )
 {
     const struct gl_texture_image *image = t->base.tObj->Image[0][level];
     const savageTileInfo *tileInfo = t->tileInfo;
-    GLuint width = image->Width2, height = image->Height2;
+    GLuint pixWidth = image->Width2, pixHeight = image->Height2;
     GLuint bpp = t->texelBytes;
+    GLuint width, height;
 
     /* FIXME: Need triangle (rather than pixel) fallbacks to simulate
      * this using normal textured triangles.
@@ -235,7 +256,16 @@ static void savageUploadTexLevel( savageTexObjPtr t, int level )
        fprintf (stderr, "Not supported texture border %d.\n",
                 (int) image->Border);
 
-    if (width >= 8 && height >= tileInfo->subHeight) {
+    if (t->hwFormat == TFT_S3TC4A4Bit || t->hwFormat == TFT_S3TC4CA4Bit ||
+       t->hwFormat == TFT_S3TC4Bit) {
+       width = (pixWidth+3) / 4;
+       height = (pixHeight+3) / 4;
+    } else {
+       width = pixWidth;
+       height = pixHeight;
+    }
+
+    if (pixWidth >= 8 && pixHeight >= 8) {
        GLuint *dirtyPtr = t->image[level].dirtyTiles;
        GLuint dirtyMask = 1;
 
@@ -300,19 +330,22 @@ static void savageUploadTexLevel( savageTexObjPtr t, int level )
        }
     } else {
        GLuint minHeight, minWidth, hRepeat, vRepeat, x, y;
-       if (width > 4 || height > 4) {
+       if (t->hwFormat == TFT_S3TC4A4Bit || t->hwFormat == TFT_S3TC4CA4Bit ||
+           t->hwFormat == TFT_S3TC4Bit)
+           minWidth = minHeight = 1;
+       else
+           minWidth = minHeight = 4;
+       if (width > minWidth || height > minHeight) {
            minWidth = tileInfo->subWidth;
            minHeight = tileInfo->subHeight;
-       } else {
-           minWidth = 4;
-           minHeight = 4;
        }
        hRepeat = width  >= minWidth  ? 1 : minWidth  / width;
        vRepeat = height >= minHeight ? 1 : minHeight / height;
        for (y = 0; y < vRepeat; ++y) {
            GLuint offset = y * tileInfo->subWidth*height * bpp;
            for (x = 0; x < hRepeat; ++x) {
-               savageUploadTiny (tileInfo, width, height, bpp, image->Data,
+               savageUploadTiny (tileInfo, pixWidth, pixHeight,
+                                 width, height, bpp, image->Data,
                                  (GLubyte *)(t->bufAddr +
                                              t->image[level].offset+offset));
                offset += width * bpp;
@@ -342,6 +375,30 @@ static GLuint savageTexImageSize (GLuint width, GLuint height, GLuint bpp) {
        return 64 * bpp;
 }
 
+/** \brief Compute the destination size of a compressed texture image
+ */
+static GLuint savageCompressedTexImageSize (GLuint width, GLuint height,
+                                           GLuint bpp) {
+    width = (width+3) / 4;
+    height = (height+3) / 4;
+    /* full subtiles */
+    if (width >= 2 && height >= 2)
+       return width * height * bpp;
+    /* special case for the last three mipmap levels: the hardware computes
+     * the offset internally */
+    else if (width <= 1 && height <= 1)
+       return 0;
+    /* partially filled sub tiles waste memory
+     * on Savage3D and Savage4 with subtile width 4 every other subtile is
+     * skipped if width < 8 so we can assume a uniform subtile width of 8 */
+    else if (width >= 2)
+       return width * 2 * bpp;
+    else if (height >= 2)
+       return 2 * height * bpp;
+    else
+       return 4 * bpp;
+}
+
 /** \brief Compute the number of (partial) tiles of a texture image
  */
 static GLuint savageTexImageTiles (GLuint width, GLuint height,
@@ -462,6 +519,170 @@ savageAllocTexObj( struct gl_texture_object *texObj )
    return t;
 }
 
+/* Mesa texture formats for alpha-images on Savage3D/IX/MX
+ *
+ * Promoting texture images to ARGB888 or ARGB4444 doesn't work
+ * because we can't tell the hardware to ignore the color components
+ * and only use the alpha component. So we define our own texture
+ * formats that promote to ARGB8888 or ARGB4444 and set the color
+ * components to white. This way we get the correct result. */
+static GLboolean
+_savage_texstore_a1114444 (GLcontext *ctx, GLuint dims,
+                          GLenum baseInternalFormat,
+                          const struct gl_texture_format *dstFormat,
+                          GLvoid *dstAddr,
+                          GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
+                          GLint dstRowStride, GLint dstImageStride,
+                          GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                          GLenum srcFormat, GLenum srcType,
+                          const GLvoid *srcAddr,
+                          const struct gl_pixelstore_attrib *srcPacking);
+static GLboolean
+_savage_texstore_a1118888 (GLcontext *ctx, GLuint dims,
+                          GLenum baseInternalFormat,
+                          const struct gl_texture_format *dstFormat,
+                          GLvoid *dstAddr,
+                          GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
+                          GLint dstRowStride, GLint dstImageStride,
+                          GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                          GLenum srcFormat, GLenum srcType,
+                          const GLvoid *srcAddr,
+                          const struct gl_pixelstore_attrib *srcPacking);
+
+static struct gl_texture_format _savage_texformat_a1114444 = {
+    MESA_FORMAT_ARGB4444,              /* MesaFormat */
+    GL_RGBA,                           /* BaseFormat */
+    GL_UNSIGNED_NORMALIZED_ARB,                /* DataType */
+    4,                                 /* RedBits */
+    4,                                 /* GreenBits */
+    4,                                 /* BlueBits */
+    4,                                 /* AlphaBits */
+    0,                                 /* LuminanceBits */
+    0,                                 /* IntensityBits */
+    0,                                 /* IndexBits */
+    0,                                 /* DepthBits */
+    2,                                 /* TexelBytes */
+    _savage_texstore_a1114444,         /* StoreTexImageFunc */
+    NULL, NULL, NULL, NULL, NULL, NULL  /* FetchTexel* filled in by 
+                                        * savageDDInitTextureFuncs */
+};
+static struct gl_texture_format _savage_texformat_a1118888 = {
+    MESA_FORMAT_ARGB8888,              /* MesaFormat */
+    GL_RGBA,                           /* BaseFormat */
+    GL_UNSIGNED_NORMALIZED_ARB,                /* DataType */
+    8,                                 /* RedBits */
+    8,                                 /* GreenBits */
+    8,                                 /* BlueBits */
+    8,                                 /* AlphaBits */
+    0,                                 /* LuminanceBits */
+    0,                                 /* IntensityBits */
+    0,                                 /* IndexBits */
+    0,                                 /* DepthBits */
+    4,                                 /* TexelBytes */
+    _savage_texstore_a1118888,         /* StoreTexImageFunc */
+    NULL, NULL, NULL, NULL, NULL, NULL  /* FetchTexel* filled in by 
+                                        * savageDDInitTextureFuncs */
+};
+
+static GLboolean
+_savage_texstore_a1114444 (GLcontext *ctx, GLuint dims,
+                          GLenum baseInternalFormat,
+                          const struct gl_texture_format *dstFormat,
+                          GLvoid *dstAddr,
+                          GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
+                          GLint dstRowStride, GLint dstImageStride,
+                          GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                          GLenum srcFormat, GLenum srcType,
+                          const GLvoid *srcAddr,
+                          const struct gl_pixelstore_attrib *srcPacking)
+{
+    /* general path */
+    const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 baseInternalFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+    const GLchan *src = tempImage;
+    GLubyte *dstImage = (GLubyte *) dstAddr
+       + dstZoffset * dstImageStride
+       + dstYoffset * dstRowStride
+       + dstXoffset * dstFormat->TexelBytes;
+    GLint img, row, col;
+
+    ASSERT(dstFormat == &_savage_texformat_a1114444);
+    ASSERT(baseInternalFormat == GL_ALPHA);
+
+    if (!tempImage)
+       return GL_FALSE;
+    _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+    for (img = 0; img < srcDepth; img++) {
+       GLubyte *dstRow = dstImage;
+       for (row = 0; row < srcHeight; row++) {
+            GLushort *dstUI = (GLushort *) dstRow;
+           for (col = 0; col < srcWidth; col++) {
+               dstUI[col] = PACK_COLOR_4444( CHAN_TO_UBYTE(src[0]),
+                                             255, 255, 255 );
+               src += 1;
+            }
+            dstRow += dstRowStride;
+       }
+       dstImage += dstImageStride;
+    }
+    _mesa_free((void *) tempImage);
+
+    return GL_TRUE;
+}
+static GLboolean
+_savage_texstore_a1118888 (GLcontext *ctx, GLuint dims,
+                          GLenum baseInternalFormat,
+                          const struct gl_texture_format *dstFormat,
+                          GLvoid *dstAddr,
+                          GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
+                          GLint dstRowStride, GLint dstImageStride,
+                          GLint srcWidth, GLint srcHeight, GLint srcDepth,
+                          GLenum srcFormat, GLenum srcType,
+                          const GLvoid *srcAddr,
+                          const struct gl_pixelstore_attrib *srcPacking)
+{
+    /* general path */
+    const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
+                                                 baseInternalFormat,
+                                                 baseInternalFormat,
+                                                 srcWidth, srcHeight, srcDepth,
+                                                 srcFormat, srcType, srcAddr,
+                                                 srcPacking);
+    const GLchan *src = tempImage;
+    GLubyte *dstImage = (GLubyte *) dstAddr
+       + dstZoffset * dstImageStride
+       + dstYoffset * dstRowStride
+       + dstXoffset * dstFormat->TexelBytes;
+    GLint img, row, col;
+
+    ASSERT(dstFormat == &_savage_texformat_a1118888);
+    ASSERT(baseInternalFormat == GL_ALPHA);
+
+    if (!tempImage)
+       return GL_FALSE;
+    _mesa_adjust_image_for_convolution(ctx, dims, &srcWidth, &srcHeight);
+    for (img = 0; img < srcDepth; img++) {
+       GLubyte *dstRow = dstImage;
+       for (row = 0; row < srcHeight; row++) {
+            GLuint *dstUI = (GLuint *) dstRow;
+           for (col = 0; col < srcWidth; col++) {
+               dstUI[col] = PACK_COLOR_8888( CHAN_TO_UBYTE(src[0]),
+                                             255, 255, 255 );
+               src += 1;
+            }
+            dstRow += dstRowStride;
+       }
+       dstImage += dstImageStride;
+    }
+    _mesa_free((void *) tempImage);
+
+    return GL_TRUE;
+}
+
 /* Called by the _mesa_store_teximage[123]d() functions. */
 static const struct gl_texture_format *
 savageChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
@@ -541,14 +762,14 @@ savageChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
    case GL_ALPHA:
    case GL_COMPRESSED_ALPHA:
       return isSavage4 ? &_mesa_texformat_a8 : (
-        do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444);
+        do32bpt ? &_savage_texformat_a1118888 : &_savage_texformat_a1114444);
    case GL_ALPHA4:
-      return isSavage4 ? &_mesa_texformat_a8 : &_mesa_texformat_argb4444;
+      return isSavage4 ? &_mesa_texformat_a8 : &_savage_texformat_a1114444;
    case GL_ALPHA8:
    case GL_ALPHA12:
    case GL_ALPHA16:
       return isSavage4 ? &_mesa_texformat_a8 : (
-        !force16bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444);
+        !force16bpt ? &_savage_texformat_a1118888 : &_savage_texformat_a1114444);
 
    case 1:
    case GL_LUMINANCE:
@@ -603,6 +824,26 @@ savageChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
       return !force16bpt ? &_mesa_texformat_argb8888 :
          &_mesa_texformat_argb4444;
 #endif
+
+   case GL_RGB_S3TC:
+   case GL_RGB4_S3TC:
+   case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
+      return &_mesa_texformat_rgb_dxt1;
+   case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
+      return &_mesa_texformat_rgba_dxt1;
+
+   case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
+      return &_mesa_texformat_rgba_dxt3;
+
+   case GL_RGBA_S3TC:
+   case GL_RGBA4_S3TC:
+      if (!isSavage4)
+        /* Not the best choice but Savage3D/MX/IX don't support DXT3 or DXT5. */
+        return &_mesa_texformat_rgba_dxt1;
+      /* fall through */
+   case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
+      return &_mesa_texformat_rgba_dxt5;
+
 /*
    case GL_COLOR_INDEX:
    case GL_COLOR_INDEX1_EXT:
@@ -624,7 +865,7 @@ static void savageSetTexImages( savageContextPtr imesa,
 {
    savageTexObjPtr t = (savageTexObjPtr) tObj->DriverData;
    struct gl_texture_image *image = tObj->Image[0][tObj->BaseLevel];
-   GLuint offset, i, textureFormat, size;
+   GLuint offset, i, textureFormat, tileIndex, size;
    GLint firstLevel, lastLevel;
 
    assert(t);
@@ -633,31 +874,51 @@ static void savageSetTexImages( savageContextPtr imesa,
    switch (image->TexFormat->MesaFormat) {
    case MESA_FORMAT_ARGB8888:
       textureFormat = TFT_ARGB8888;
-      t->texelBytes = 4;
+      t->texelBytes = tileIndex = 4;
       break;
    case MESA_FORMAT_ARGB1555:
       textureFormat = TFT_ARGB1555;
-      t->texelBytes = 2;
+      t->texelBytes = tileIndex = 2;
       break;
    case MESA_FORMAT_ARGB4444:
       textureFormat = TFT_ARGB4444;
-      t->texelBytes = 2;
+      t->texelBytes = tileIndex = 2;
       break;
    case MESA_FORMAT_RGB565:
       textureFormat = TFT_RGB565;
-      t->texelBytes = 2;
+      t->texelBytes = tileIndex = 2;
       break;
    case MESA_FORMAT_L8:
       textureFormat = TFT_L8;
-      t->texelBytes = 1;
+      t->texelBytes = tileIndex = 1;
       break;
    case MESA_FORMAT_I8:
       textureFormat = TFT_I8;
-      t->texelBytes = 1;
+      t->texelBytes = tileIndex = 1;
       break;
    case MESA_FORMAT_A8:
       textureFormat = TFT_A8;
-      t->texelBytes = 1;
+      t->texelBytes = tileIndex = 1;
+      break;
+   case MESA_FORMAT_RGB_DXT1:
+      textureFormat = TFT_S3TC4Bit;
+      tileIndex = TILE_INDEX_DXT1;
+      t->texelBytes = 8;
+      break;
+   case MESA_FORMAT_RGBA_DXT1:
+      textureFormat = TFT_S3TC4Bit;
+      tileIndex = TILE_INDEX_DXT1;
+      t->texelBytes = 8;
+      break;
+   case MESA_FORMAT_RGBA_DXT3:
+      textureFormat =  TFT_S3TC4A4Bit;
+      tileIndex = TILE_INDEX_DXTn;
+      t->texelBytes = 16;
+      break;
+   case MESA_FORMAT_RGBA_DXT5:
+      textureFormat = TFT_S3TC4CA4Bit;
+      tileIndex = TILE_INDEX_DXTn;
+      t->texelBytes = 16;
       break;
    default:
       _mesa_problem(imesa->glCtx, "Bad texture format in %s", __FUNCTION__);
@@ -665,11 +926,11 @@ static void savageSetTexImages( savageContextPtr imesa,
    }
    t->hwFormat = textureFormat;
 
-   /* Select tiling format depending on the chipset and bytes per texel */
+   /* Select tiling format depending on the chipset and texture format */
    if (imesa->savageScreen->chipset <= S3_SAVAGE4)
-       t->tileInfo = &tileInfo_s3d_s4[t->texelBytes];
+       t->tileInfo = &tileInfo_s3d_s4[tileIndex];
    else
-       t->tileInfo = &tileInfo_pro[t->texelBytes];
+       t->tileInfo = &tileInfo_pro[tileIndex];
 
    /* Compute which mipmap levels we really want to send to the hardware.
     */
@@ -701,8 +962,12 @@ static void savageSetTexImages( savageContextPtr imesa,
       t->image[i].offset = offset;
 
       image = tObj->Image[0][i];
-      size = savageTexImageSize (image->Width2, image->Height2,
-                                t->texelBytes);
+      if (t->texelBytes >= 8)
+        size = savageCompressedTexImageSize (image->Width2, image->Height2,
+                                             t->texelBytes);
+      else
+        size = savageTexImageSize (image->Width2, image->Height2,
+                                   t->texelBytes);
       offset += size;
    }
 
@@ -711,7 +976,7 @@ static void savageSetTexImages( savageContextPtr imesa,
    /* the last three mipmap levels don't add to the offset. They are packed
     * into 64 pixels. */
    if (size == 0)
-       t->base.totalSize += 64 * t->texelBytes;
+       t->base.totalSize += (t->texelBytes >= 8 ? 4 : 64) * t->texelBytes;
    /* 2k-aligned (really needed?) */
    t->base.totalSize = (t->base.totalSize + 2047UL) & ~2047UL;
 }
@@ -1321,13 +1586,12 @@ static void savageUpdateTexState_s3d( GLcontext *ctx )
     GLuint format;
 
     /* disable */
-    if (ctx->Texture.Unit[0]._ReallyEnabled == 0) {
-       imesa->regs.s3d.texCtrl.ui = 0;
-       imesa->regs.s3d.texCtrl.ni.texEn = GL_FALSE;
-       imesa->regs.s3d.texCtrl.ni.dBias = 0x08;
-       imesa->regs.s3d.texCtrl.ni.texXprEn = GL_TRUE;
+    imesa->regs.s3d.texCtrl.ui = 0;
+    imesa->regs.s3d.texCtrl.ni.texEn = GL_FALSE;
+    imesa->regs.s3d.texCtrl.ni.dBias = 0x08;
+    imesa->regs.s3d.texCtrl.ni.texXprEn = GL_TRUE;
+    if (ctx->Texture.Unit[0]._ReallyEnabled == 0)
        return;
-    }
 
     tObj = ctx->Texture.Unit[0]._Current;
     if ((ctx->Texture.Unit[0]._ReallyEnabled & ~(TEXTURE_1D_BIT|TEXTURE_2D_BIT))
@@ -1360,12 +1624,28 @@ static void savageUpdateTexState_s3d( GLcontext *ctx )
     /* FIXME: copied from utah-glx, probably needs some tuning */
     switch (ctx->Texture.Unit[0].EnvMode) {
     case GL_DECAL:
-       imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = SAVAGETBC_DECAL_S3D;
+       imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = SAVAGETBC_DECALALPHA_S3D;
        break;
     case GL_REPLACE:
-       imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = SAVAGETBC_COPY_S3D;
+       switch (format) {
+       case GL_ALPHA: /* FIXME */
+           imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = 1;
+           break;
+       case GL_LUMINANCE_ALPHA:
+       case GL_RGBA:
+           imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = 4;
+           break;
+       case GL_RGB:
+       case GL_LUMINANCE:
+           imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = SAVAGETBC_DECAL_S3D;
+           break;
+       case GL_INTENSITY:
+           imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = SAVAGETBC_COPY_S3D;
+       }
        break;
-    case GL_BLEND: /* FIXIT */
+    case GL_BLEND: /* hardware can't do GL_BLEND */
+       FALLBACK (ctx, SAVAGE_FALLBACK_TEXTURE, GL_TRUE);
+       return;
     case GL_MODULATE:
        imesa->regs.s3d.drawCtrl.ni.texBlendCtrl = SAVAGETBC_MODULATEALPHA_S3D;
        break;
@@ -1375,17 +1655,16 @@ static void savageUpdateTexState_s3d( GLcontext *ctx )
        break;                  
     }
 
-    imesa->regs.s3d.drawCtrl.ni.flushPdDestWrites = GL_TRUE;
-    imesa->regs.s3d.drawCtrl.ni.flushPdZbufWrites = GL_TRUE;
-
-    /* FIXME: this is how the utah-driver works. I doubt it's the ultimate 
-       truth. */
+    /* The Savage3D can't handle different wrapping modes in s and t.
+     * If they are not the same, fall back to software. */
+    if (t->setup.sWrapMode != t->setup.tWrapMode) {
+       FALLBACK (ctx, SAVAGE_FALLBACK_TEXTURE, GL_TRUE);
+       return;
+    }
     imesa->regs.s3d.texCtrl.ni.uWrapEn = 0;
     imesa->regs.s3d.texCtrl.ni.vWrapEn = 0;
-    if (t->setup.sWrapMode == GL_CLAMP)
-       imesa->regs.s3d.texCtrl.ni.wrapMode = TAM_Clamp;
-    else
-       imesa->regs.s3d.texCtrl.ni.wrapMode = TAM_Wrap;
+    imesa->regs.s3d.texCtrl.ni.wrapMode =
+       (t->setup.sWrapMode == GL_REPEAT) ? TAM_Wrap : TAM_Clamp;
 
     switch (t->setup.minFilter) {
     case GL_NEAREST:
@@ -1684,6 +1963,63 @@ static void savageTexSubImage2D( GLcontext *ctx,
    SAVAGE_CONTEXT(ctx)->new_state |= SAVAGE_NEW_TEXTURE;
 }
 
+static void
+savageCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+                           GLint internalFormat,
+                           GLint width, GLint height, GLint border,
+                           GLsizei imageSize, const GLvoid *data,
+                           struct gl_texture_object *texObj,
+                           struct gl_texture_image *texImage )
+{
+   savageTexObjPtr t = (savageTexObjPtr) texObj->DriverData;
+   if (t) {
+      savageTexImageChanged (t);
+   } else {
+      t = savageAllocTexObj(texObj);
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexImage2D");
+         return;
+      }
+   }
+   _mesa_store_compressed_teximage2d( ctx, target, level, internalFormat,
+                                     width, height, border, imageSize,
+                                     data, texObj, texImage );
+   t->base.dirty_images[0] |= (1 << level);
+   SAVAGE_CONTEXT(ctx)->new_state |= SAVAGE_NEW_TEXTURE;
+}
+
+static void
+savageCompressedTexSubImage2D( GLcontext *ctx, 
+                              GLenum target,
+                              GLint level,     
+                              GLint xoffset, GLint yoffset,
+                              GLsizei width, GLsizei height,
+                              GLenum format, GLsizei imageSize,
+                              const GLvoid *data,
+                              struct gl_texture_object *texObj,
+                              struct gl_texture_image *texImage )
+{
+   savageTexObjPtr t = (savageTexObjPtr) texObj->DriverData;
+   assert( t ); /* this _should_ be true */
+   if (t) {
+      savageTexImageChanged (t);
+      savageMarkDirtyTiles(t, level, texImage->Width2, texImage->Height2,
+                          xoffset, yoffset, width, height);
+   } else {
+      t = savageAllocTexObj(texObj);
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D");
+         return;
+      }
+      t->base.dirty_images[0] |= (1 << level);
+   }
+   _mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset, yoffset,
+                                       width, height, format, imageSize,
+                                       data, texObj, texImage);
+   t->dirtySubImages |= (1 << level);
+   SAVAGE_CONTEXT(ctx)->new_state |= SAVAGE_NEW_TEXTURE;
+}
+
 static void savageTexParameter( GLcontext *ctx, GLenum target,
                              struct gl_texture_object *tObj,
                              GLenum pname, const GLfloat *params )
@@ -1761,9 +2097,27 @@ void savageDDInitTextureFuncs( struct dd_function_table *functions )
    functions->TexSubImage1D = savageTexSubImage1D;
    functions->TexImage2D = savageTexImage2D;
    functions->TexSubImage2D = savageTexSubImage2D;
+   functions->CompressedTexImage2D = savageCompressedTexImage2D;
+   functions->CompressedTexSubImage2D = savageCompressedTexSubImage2D;
    functions->BindTexture = savageBindTexture;
    functions->NewTextureObject = savageNewTextureObject;
    functions->DeleteTexture = savageDeleteTexture;
    functions->IsTextureResident = driIsTextureResident;
    functions->TexParameter = savageTexParameter;
+
+   /* Texel fetching with our custom texture formats works just like
+    * the standard argb formats. */
+   _savage_texformat_a1114444.FetchTexel1D = _mesa_texformat_argb4444.FetchTexel1D;
+   _savage_texformat_a1114444.FetchTexel2D = _mesa_texformat_argb4444.FetchTexel2D;
+   _savage_texformat_a1114444.FetchTexel3D = _mesa_texformat_argb4444.FetchTexel3D;
+   _savage_texformat_a1114444.FetchTexel1Df= _mesa_texformat_argb4444.FetchTexel1Df;
+   _savage_texformat_a1114444.FetchTexel2Df= _mesa_texformat_argb4444.FetchTexel2Df;
+   _savage_texformat_a1114444.FetchTexel3Df= _mesa_texformat_argb4444.FetchTexel3Df;
+
+   _savage_texformat_a1118888.FetchTexel1D = _mesa_texformat_argb8888.FetchTexel1D;
+   _savage_texformat_a1118888.FetchTexel2D = _mesa_texformat_argb8888.FetchTexel2D;
+   _savage_texformat_a1118888.FetchTexel3D = _mesa_texformat_argb8888.FetchTexel3D;
+   _savage_texformat_a1118888.FetchTexel1Df= _mesa_texformat_argb8888.FetchTexel1Df;
+   _savage_texformat_a1118888.FetchTexel2Df= _mesa_texformat_argb8888.FetchTexel2Df;
+   _savage_texformat_a1118888.FetchTexel3Df= _mesa_texformat_argb8888.FetchTexel3Df;
 }