1 files changed, 227 insertions, 133 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_texstate.c b/src/mesa/drivers/dri/r300/r300_texstate.c
index 1d2909fd21..e2329f04ec 100644
--- a/src/mesa/drivers/dri/r300/r300_texstate.c
+++ b/src/mesa/drivers/dri/r300/r300_texstate.c
@@ -35,14 +35,14 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  * \todo Enable R300 texture tiling code?
  */
 
-#include "glheader.h"
-#include "imports.h"
-#include "context.h"
-#include "macros.h"
-#include "texformat.h"
-#include "teximage.h"
-#include "texobj.h"
-#include "enums.h"
+#include "main/glheader.h"
+#include "main/imports.h"
+#include "main/context.h"
+#include "main/macros.h"
+#include "main/texformat.h"
+#include "main/teximage.h"
+#include "main/texobj.h"
+#include "main/enums.h"
 
 #include "r300_context.h"
 #include "r300_state.h"
@@ -115,11 +115,186 @@ static const struct tx_table {
 	_ASSIGN(LUMINANCE_ALPHA_FLOAT16, R300_EASY_TX_FORMAT(X, X, X, Y, FL_I16A16)),
 	_ASSIGN(INTENSITY_FLOAT32, R300_EASY_TX_FORMAT(X, X, X, X, FL_I32)),
 	_ASSIGN(INTENSITY_FLOAT16, R300_EASY_TX_FORMAT(X, X, X, X, FL_I16)),
+	_ASSIGN(Z16, R300_EASY_TX_FORMAT(X, X, X, X, X16)),
+	_ASSIGN(Z24_S8, R300_EASY_TX_FORMAT(X, X, X, X, X24_Y8)),
+	_ASSIGN(Z32, R300_EASY_TX_FORMAT(X, X, X, X, X32)),
 	/* *INDENT-ON* */
 };
 
 #undef _ASSIGN
 
+void r300SetDepthTexMode(struct gl_texture_object *tObj)
+{
+	static const GLuint formats[3][3] = {
+		{
+			R300_EASY_TX_FORMAT(X, X, X, ONE, X16),
+			R300_EASY_TX_FORMAT(X, X, X, X, X16),
+			R300_EASY_TX_FORMAT(ZERO, ZERO, ZERO, X, X16),
+		},
+		{
+			R300_EASY_TX_FORMAT(X, X, X, ONE, X24_Y8),
+			R300_EASY_TX_FORMAT(X, X, X, X, X24_Y8),
+			R300_EASY_TX_FORMAT(ZERO, ZERO, ZERO, X, X24_Y8),
+		},
+		{
+			R300_EASY_TX_FORMAT(X, X, X, ONE, X32),
+			R300_EASY_TX_FORMAT(X, X, X, X, X32),
+			R300_EASY_TX_FORMAT(ZERO, ZERO, ZERO, X, X32),
+		},
+	};
+	const GLuint *format;
+	r300TexObjPtr t;
+
+	if (!tObj)
+		return;
+
+	t = (r300TexObjPtr) tObj->DriverData;
+
+
+	switch (tObj->Image[0][tObj->BaseLevel]->TexFormat->MesaFormat) {
+	case MESA_FORMAT_Z16:
+		format = formats[0];
+		break;
+	case MESA_FORMAT_Z24_S8:
+		format = formats[1];
+		break;
+	case MESA_FORMAT_Z32:
+		format = formats[2];
+		break;
+	default:
+		/* Error...which should have already been caught by higher
+		 * levels of Mesa.
+		 */
+		ASSERT(0);
+		return;
+	}
+
+	switch (tObj->DepthMode) {
+	case GL_LUMINANCE:
+		t->format = format[0];
+		break;
+	case GL_INTENSITY:
+		t->format = format[1];
+		break;
+	case GL_ALPHA:
+		t->format = format[2];
+		break;
+	default:
+		/* Error...which should have already been caught by higher
+		 * levels of Mesa.
+		 */
+		ASSERT(0);
+		return;
+	}
+}
+
+
+/**
+ * Compute sizes and fill in offset and blit information for the given
+ * image (determined by \p face and \p level).
+ *
+ * \param curOffset points to the offset at which the image is to be stored
+ * and is updated by this function according to the size of the image.
+ */
+static void compute_tex_image_offset(
+	struct gl_texture_object *tObj,
+	GLuint face,
+	GLint level,
+	GLint* curOffset)
+{
+	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
+	const struct gl_texture_image* texImage;
+	GLuint blitWidth = R300_BLIT_WIDTH_BYTES;
+	GLuint texelBytes;
+	GLuint size;
+
+	texImage = tObj->Image[0][level + t->base.firstLevel];
+	if (!texImage)
+		return;
+
+	texelBytes = texImage->TexFormat->TexelBytes;
+
+	/* find image size in bytes */
+	if (texImage->IsCompressed) {
+		if ((t->format & R300_TX_FORMAT_DXT1) ==
+			R300_TX_FORMAT_DXT1) {
+			// fprintf(stderr,"DXT 1 %d %08X\n", texImage->Width, t->format);
+			if ((texImage->Width + 3) < 8)	/* width one block */
+				size = texImage->CompressedSize * 4;
+			else if ((texImage->Width + 3) < 16)
+				size = texImage->CompressedSize * 2;
+			else
+				size = texImage->CompressedSize;
+		} else {
+			/* DXT3/5, 16 bytes per block */
+			WARN_ONCE
+				("DXT 3/5 suffers from multitexturing problems!\n");
+			// fprintf(stderr,"DXT 3/5 %d\n", texImage->Width);
+			if ((texImage->Width + 3) < 8)
+				size = texImage->CompressedSize * 2;
+			else
+				size = texImage->CompressedSize;
+		}
+	} else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+		size =
+			((texImage->Width * texelBytes +
+			63) & ~63) * texImage->Height;
+		blitWidth = 64 / texelBytes;
+	} else if (t->tile_bits & R300_TXO_MICRO_TILE) {
+		/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
+			though the actual offset may be different (if texture is less than
+			32 bytes width) to the untiled case */
+		int w = (texImage->Width * texelBytes * 2 + 31) & ~31;
+		size =
+			(w * ((texImage->Height + 1) / 2)) *
+			texImage->Depth;
+		blitWidth = MAX2(texImage->Width, 64 / texelBytes);
+	} else {
+		int w = (texImage->Width * texelBytes + 31) & ~31;
+		size = w * texImage->Height * texImage->Depth;
+		blitWidth = MAX2(texImage->Width, 64 / texelBytes);
+	}
+	assert(size > 0);
+
+	if (RADEON_DEBUG & DEBUG_TEXTURE)
+		fprintf(stderr, "w=%d h=%d d=%d tb=%d intFormat=%d\n",
+			texImage->Width, texImage->Height,
+			texImage->Depth,
+			texImage->TexFormat->TexelBytes,
+			texImage->InternalFormat);
+
+	/* All images are aligned to a 32-byte offset */
+	*curOffset = (*curOffset + 0x1f) & ~0x1f;
+
+	if (texelBytes) {
+		/* fix x and y coords up later together with offset */
+		t->image[face][level].x = *curOffset;
+		t->image[face][level].y = 0;
+		t->image[face][level].width =
+			MIN2(size / texelBytes, blitWidth);
+		t->image[face][level].height =
+			(size / texelBytes) / t->image[face][level].width;
+	} else {
+		t->image[face][level].x = *curOffset % R300_BLIT_WIDTH_BYTES;
+		t->image[face][level].y = *curOffset / R300_BLIT_WIDTH_BYTES;
+		t->image[face][level].width =
+			MIN2(size, R300_BLIT_WIDTH_BYTES);
+		t->image[face][level].height = size / t->image[face][level].width;
+	}
+
+	if (RADEON_DEBUG & DEBUG_TEXTURE)
+		fprintf(stderr,
+			"level %d, face %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
+			level, face, texImage->Width, texImage->Height,
+			t->image[face][level].x, t->image[face][level].y,
+			t->image[face][level].width, t->image[face][level].height,
+			size, *curOffset);
+
+	*curOffset += size;
+}
+
+
+
 /**
  * This function computes the number of bytes of storage needed for
  * the given texture object (all mipmap levels, all cube faces).
@@ -137,7 +312,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
 	const struct gl_texture_image *baseImage =
 	    tObj->Image[0][tObj->BaseLevel];
-	GLint curOffset, blitWidth;
+	GLint curOffset;
 	GLint i, texelBytes;
 	GLint numLevels;
 	GLint log2Width, log2Height, log2Depth;
@@ -146,7 +321,12 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	 */
 	if (!t->image_override
 	    && VALID_FORMAT(baseImage->TexFormat->MesaFormat)) {
-		t->format = tx_table[baseImage->TexFormat->MesaFormat].format;
+		if (baseImage->TexFormat->BaseFormat == GL_DEPTH_COMPONENT) {
+			r300SetDepthTexMode(tObj);
+		} else {
+			t->format = tx_table[baseImage->TexFormat->MesaFormat].format;
+		}
+
 		t->filter |= tx_table[baseImage->TexFormat->MesaFormat].filter;
 	} else if (!t->image_override) {
 		_mesa_problem(NULL, "unexpected texture format in %s",
@@ -171,8 +351,6 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	 * The idea is that we lay out the mipmap levels within a block of
 	 * memory organized as a rectangle of width BLIT_WIDTH_BYTES.
 	 */
-	curOffset = 0;
-	blitWidth = R300_BLIT_WIDTH_BYTES;
 	t->tile_bits = 0;
 
 	/* figure out if this texture is suitable for tiling. */
@@ -202,94 +380,23 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	}
 #endif
 
-	for (i = 0; i < numLevels; i++) {
-		const struct gl_texture_image *texImage;
-		GLuint size;
-
-		texImage = tObj->Image[0][i + t->base.firstLevel];
-		if (!texImage)
-			break;
-
-		/* find image size in bytes */
-		if (texImage->IsCompressed) {
-			if ((t->format & R300_TX_FORMAT_DXT1) ==
-			    R300_TX_FORMAT_DXT1) {
-				// fprintf(stderr,"DXT 1 %d %08X\n", texImage->Width, t->format);
-				if ((texImage->Width + 3) < 8)	/* width one block */
-					size = texImage->CompressedSize * 4;
-				else if ((texImage->Width + 3) < 16)
-					size = texImage->CompressedSize * 2;
-				else
-					size = texImage->CompressedSize;
-			} else {
-				/* DXT3/5, 16 bytes per block */
-				WARN_ONCE
-				    ("DXT 3/5 suffers from multitexturing problems!\n");
-				// fprintf(stderr,"DXT 3/5 %d\n", texImage->Width);
-				if ((texImage->Width + 3) < 8)
-					size = texImage->CompressedSize * 2;
-				else
-					size = texImage->CompressedSize;
-			}
-		} else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-			size =
-			    ((texImage->Width * texelBytes +
-			      63) & ~63) * texImage->Height;
-			blitWidth = 64 / texelBytes;
-		} else if (t->tile_bits & R300_TXO_MICRO_TILE) {
-			/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
-			   though the actual offset may be different (if texture is less than
-			   32 bytes width) to the untiled case */
-			int w = (texImage->Width * texelBytes * 2 + 31) & ~31;
-			size =
-			    (w * ((texImage->Height + 1) / 2)) *
-			    texImage->Depth;
-			blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-		} else {
-			int w = (texImage->Width * texelBytes + 31) & ~31;
-			size = w * texImage->Height * texImage->Depth;
-			blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-		}
-		assert(size > 0);
-
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr, "w=%d h=%d d=%d tb=%d intFormat=%d\n",
-				texImage->Width, texImage->Height,
-				texImage->Depth,
-				texImage->TexFormat->TexelBytes,
-				texImage->InternalFormat);
-
-		/* Align to 32-byte offset.  It is faster to do this unconditionally
-		 * (no branch penalty).
-		 */
+	curOffset = 0;
 
-		curOffset = (curOffset + 0x1f) & ~0x1f;
+	if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
+		ASSERT(log2Width == log2Height);
+		t->format |= R300_TX_FORMAT_CUBIC_MAP;
 
-		if (texelBytes) {
-			/* fix x and y coords up later together with offset */
-			t->image[0][i].x = curOffset;
-			t->image[0][i].y = 0;
-			t->image[0][i].width =
-			    MIN2(size / texelBytes, blitWidth);
-			t->image[0][i].height =
-			    (size / texelBytes) / t->image[0][i].width;
-		} else {
-			t->image[0][i].x = curOffset % R300_BLIT_WIDTH_BYTES;
-			t->image[0][i].y = curOffset / R300_BLIT_WIDTH_BYTES;
-			t->image[0][i].width =
-			    MIN2(size, R300_BLIT_WIDTH_BYTES);
-			t->image[0][i].height = size / t->image[0][i].width;
+		for(i = 0; i < numLevels; i++) {
+			GLuint face;
+			for(face = 0; face < 6; face++)
+				compute_tex_image_offset(tObj, face, i, &curOffset);
 		}
+	} else {
+		if (tObj->Target == GL_TEXTURE_3D)
+                	t->format |= R300_TX_FORMAT_3D;
 
-		if (RADEON_DEBUG & DEBUG_TEXTURE)
-			fprintf(stderr,
-				"level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
-				i, texImage->Width, texImage->Height,
-				t->image[0][i].x, t->image[0][i].y,
-				t->image[0][i].width, t->image[0][i].height,
-				size, curOffset);
-
-		curOffset += size;
+		for (i = 0; i < numLevels; i++)
+			compute_tex_image_offset(tObj, 0, i, &curOffset);
 	}
 
 	/* Align the total size of texture memory block.
@@ -297,43 +404,27 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	t->base.totalSize =
 	    (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
 
-	/* Setup remaining cube face blits, if needed */
-	if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
-		GLuint face;
-		for (face = 1; face < 6; face++) {
-			for (i = 0; i < numLevels; i++) {
-				t->image[face][i].x = t->image[0][i].x;
-				t->image[face][i].y = t->image[0][i].y;
-				t->image[face][i].width = t->image[0][i].width;
-				t->image[face][i].height =
-				    t->image[0][i].height;
-			}
-		}
-		t->base.totalSize *= 6;	/* total texmem needed */
-	}
-
-	if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
-		ASSERT(log2Width == log2Height);
-		t->format |= R300_TX_FORMAT_CUBIC_MAP;
-	}
-
 	t->size =
 	    (((tObj->Image[0][t->base.firstLevel]->Width -
 	       1) << R300_TX_WIDTHMASK_SHIFT)
 	     | ((tObj->Image[0][t->base.firstLevel]->Height - 1) <<
-		R300_TX_HEIGHTMASK_SHIFT))
+		R300_TX_HEIGHTMASK_SHIFT)
+	     | ((tObj->Image[0][t->base.firstLevel]->DepthLog2) <<
+		R300_TX_DEPTHMASK_SHIFT))
 	    | ((numLevels - 1) << R300_TX_MAX_MIP_LEVEL_SHIFT);
 
+	t->pitch = 0;
+
 	/* Only need to round to nearest 32 for textures, but the blitter
 	 * requires 64-byte aligned pitches, and we may/may not need the
 	 * blitter.   NPOT only!
 	 */
 	if (baseImage->IsCompressed) {
-		t->pitch =
+		t->pitch |=
 		    (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63);
 	} else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-		unsigned int align = blitWidth - 1;
-		t->pitch = ((tObj->Image[0][t->base.firstLevel]->Width *
+		unsigned int align = (64 / texelBytes) - 1;
+		t->pitch |= ((tObj->Image[0][t->base.firstLevel]->Width *
 			     texelBytes) + 63) & ~(63);
 		t->size |= R300_TX_SIZE_TXPITCH_EN;
 		if (!t->image_override)
@@ -341,14 +432,17 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 			    (((tObj->Image[0][t->base.firstLevel]->Width) +
 			      align) & ~align) - 1;
 	} else {
-		t->pitch =
+		t->pitch |=
 		    ((tObj->Image[0][t->base.firstLevel]->Width *
 		      texelBytes) + 63) & ~(63);
 	}
 
-	t->dirty_state = TEX_ALL;
-
-	/* FYI: r300UploadTexImages( rmesa, t ) used to be called here */
+	if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+	    if (tObj->Image[0][t->base.firstLevel]->Width > 2048)
+		t->pitch_reg |= R500_TXWIDTH_BIT11;
+	    if (tObj->Image[0][t->base.firstLevel]->Height > 2048)
+		t->pitch_reg |= R500_TXHEIGHT_BIT11;
+	}
 }
 
 /* ================================================================
@@ -485,7 +579,6 @@ static GLboolean r300UpdateTexture(GLcontext * ctx, int unit)
 
 		rmesa->state.texture.unit[unit].texobj = t;
 		t->base.bound |= (1 << unit);
-		t->dirty_state |= 1 << unit;
 		driUpdateTextureLRU((driTextureObject *) t);	/* XXX: should be locked! */
 	}
 
@@ -495,12 +588,11 @@ static GLboolean r300UpdateTexture(GLcontext * ctx, int unit)
 void r300SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 		      unsigned long long offset, GLint depth, GLuint pitch)
 {
-	r300ContextPtr rmesa =
-	    (r300ContextPtr) ((__DRIcontextPrivate *) pDRICtx->private)->
-	    driverPrivate;
+	r300ContextPtr rmesa = pDRICtx->driverPrivate;
 	struct gl_texture_object *tObj =
 	    _mesa_lookup_texture(rmesa->radeon.glCtx, texname);
 	r300TexObjPtr t;
+	uint32_t pitch_val;
 
 	if (!tObj)
 		return;
@@ -513,28 +605,30 @@ void r300SetTexOffset(__DRIcontext * pDRICtx, GLint texname,
 		return;
 
 	t->offset = offset;
-	t->pitch_reg = pitch;
+	t->pitch_reg &= (1 << 13) -1;
+	pitch_val = pitch;
 
 	switch (depth) {
 	case 32:
 		t->format = R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8);
 		t->filter |= tx_table[2].filter;
-		t->pitch_reg /= 4;
+		pitch_val /= 4;
 		break;
 	case 24:
 	default:
 		t->format = R300_EASY_TX_FORMAT(X, Y, Z, ONE, W8Z8Y8X8);
 		t->filter |= tx_table[4].filter;
-		t->pitch_reg /= 4;
+		pitch_val /= 4;
 		break;
 	case 16:
 		t->format = R300_EASY_TX_FORMAT(X, Y, Z, ONE, Z5Y6X5);
 		t->filter |= tx_table[5].filter;
-		t->pitch_reg /= 2;
+		pitch_val /= 2;
 		break;
 	}
+	pitch_val--;
 
-	t->pitch_reg--;
+	t->pitch_reg |= pitch_val;
 }
 
 static GLboolean r300UpdateTextureUnit(GLcontext * ctx, int unit)