Fix r300 rectangular texture upload and swtcl coordinate fixing same as radeon

sw tcl

src/mesa/drivers/dri/r300/r300_cmdbuf.c

@@ -428,6 +428,9 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	ALLOC_STATE( tex.format, variable, mtu+1, "tex_format", 0 );
 		r300->hw.tex.format.cmd[R300_TEX_CMD_0] = cmdpacket0(R300_TX_FORMAT_0, 0);
 
+	ALLOC_STATE( tex.pitch, variable, mtu+1, "tex_pitch", 0 );
+		r300->hw.tex.pitch.cmd[R300_TEX_CMD_0] = cmdpacket0(R300_TX_PITCH_0, 0);
+
 	ALLOC_STATE( tex.offset, variable, mtu+1, "tex_offset", 0 );
 		r300->hw.tex.offset.cmd[R300_TEX_CMD_0] = cmdpacket0(R300_TX_OFFSET_0, 0);
 
@@ -513,6 +516,7 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.unknown1);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.size);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.format);
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.pitch);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.offset);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.unknown4);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.border_color);

src/mesa/drivers/dri/r300/r300_context.c

@@ -123,6 +123,7 @@ const struct dri_extension card_extensions[] = {
 
 extern struct tnl_pipeline_stage _r300_render_stage;
 extern struct tnl_pipeline_stage _r300_tcl_stage;
+extern const struct tnl_pipeline_stage _r300_texrect_stage;
 
 static const struct tnl_pipeline_stage *r300_pipeline[] = {
 
@@ -153,6 +154,8 @@ static const struct tnl_pipeline_stage *r300_pipeline[] = {
 
 	/* Else do them here.
 	 */
+	/* scale texture rectangle to 0..1. */
+	&_r300_texrect_stage,
 	&_r300_render_stage,
 	&_tnl_render_stage,	/* FALLBACK  */
 	0,

src/mesa/drivers/dri/r300/r300_context.h

@@ -177,10 +177,11 @@ struct r300_tex_obj {
 	/* Six, for the cube faces */
 
 
+	GLuint pitch; /* this isn't sent to hardware just used in calculations */
 	/* hardware register values */
 	/* Note that R200 has 8 registers per texture and R300 only 7 */
 	GLuint filter;
-	GLuint pitch; /* one of the unknown registers.. unknown 1 ?*/
+	GLuint pitch_reg;
 	GLuint size;	/* npot only */
 	GLuint format;
 	GLuint offset;	/* Image location in the card's address space.
@@ -481,6 +482,7 @@ struct r300_hw_state {
 		struct r300_state_atom unknown1;
 		struct r300_state_atom size;
 		struct r300_state_atom format;
+		struct r300_state_atom pitch;
 		struct r300_state_atom offset;
 		struct r300_state_atom unknown4;
 		struct r300_state_atom border_color;

src/mesa/drivers/dri/r300/r300_reg.h

@@ -793,6 +793,7 @@ I am fairly certain that they are correct unless stated otherwise in comments.
 
 #	define R300_TX_FORMAT_YUV_MODE		0x00800000
 
+#define R300_TX_PITCH_0			    0x4500 /* obvious missing in gap */
 #define R300_TX_OFFSET_0                    0x4540
 /* BEGIN: Guess from R200 */
 #       define R300_TXO_ENDIAN_NO_SWAP           (0 << 0)

src/mesa/drivers/dri/r300/r300_render.c

@@ -882,3 +882,108 @@ const struct tnl_pipeline_stage _r300_tcl_stage = {
 	NULL,
 	r300_run_tcl_render	/* run */
 };
+
+/* R300 texture rectangle expects coords in 0..1 range, not 0..dimension
+ * as in the extension spec.  Need to translate here.
+ *
+ * Note that swrast expects 0..dimension, so if a fallback is active,
+ * don't do anything.  (Maybe need to configure swrast to match hw)
+ */
+struct texrect_stage_data {
+   GLvector4f texcoord[MAX_TEXTURE_UNITS];
+};
+
+#define TEXRECT_STAGE_DATA(stage) ((struct texrect_stage_data *)stage->privatePtr)
+
+
+static GLboolean run_texrect_stage( GLcontext *ctx,
+				    struct tnl_pipeline_stage *stage )
+{
+   struct texrect_stage_data *store = TEXRECT_STAGE_DATA(stage);
+   r300ContextPtr rmesa = R300_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &tnl->vb;
+   GLuint i;
+
+   if (rmesa->radeon.Fallback)
+      return GL_TRUE;
+
+   for (i = 0 ; i < ctx->Const.MaxTextureUnits ; i++) {
+      if (ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_RECT_BIT) {
+	 struct gl_texture_object *texObj = ctx->Texture.Unit[i].CurrentRect;
+	 struct gl_texture_image *texImage = texObj->Image[0][texObj->BaseLevel];
+	 const GLfloat iw = 1.0/texImage->Width;
+	 const GLfloat ih = 1.0/texImage->Height;
+	 GLfloat *in = (GLfloat *)VB->TexCoordPtr[i]->data;
+	 GLint instride = VB->TexCoordPtr[i]->stride;
+	 GLfloat (*out)[4] = store->texcoord[i].data;
+	 GLint j;
+
+	 store->texcoord[i].size = VB->TexCoordPtr[i]->size;
+	 for (j = 0 ; j < VB->Count ; j++) {
+	    switch (VB->TexCoordPtr[i]->size) {
+	    case 4:
+	       out[j][3] = in[3];
+	    /* fallthrough */
+	    case 3:
+	       out[j][2] = in[2];
+	    /* fallthrough */
+	    default:
+	       out[j][0] = in[0] * iw;
+	       out[j][1] = in[1] * ih;
+	    }
+	    in = (GLfloat *)((GLubyte *)in + instride);
+	 }
+
+	 VB->AttribPtr[VERT_ATTRIB_TEX0+i] = VB->TexCoordPtr[i] = &store->texcoord[i];
+      }
+   }
+
+   return GL_TRUE;
+}
+
+
+/* Called the first time stage->run() is invoked.
+ */
+static GLboolean alloc_texrect_data( GLcontext *ctx,
+				     struct tnl_pipeline_stage *stage )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   struct texrect_stage_data *store;
+   GLuint i;
+
+   stage->privatePtr = CALLOC(sizeof(*store));
+   store = TEXRECT_STAGE_DATA(stage);
+   if (!store)
+      return GL_FALSE;
+
+   for (i = 0 ; i < ctx->Const.MaxTextureUnits ; i++)
+      _mesa_vector4f_alloc( &store->texcoord[i], 0, VB->Size, 32 );
+
+   return GL_TRUE;
+}
+
+static void free_texrect_data( struct tnl_pipeline_stage *stage )
+{
+   struct texrect_stage_data *store = TEXRECT_STAGE_DATA(stage);
+   GLuint i;
+
+   if (store) {
+      for (i = 0 ; i < MAX_TEXTURE_UNITS ; i++)
+	 if (store->texcoord[i].data)
+	    _mesa_vector4f_free( &store->texcoord[i] );
+      FREE( store );
+      stage->privatePtr = NULL;
+   }
+}
+
+const struct tnl_pipeline_stage _r300_texrect_stage =
+{
+   "r300 texrect stage",			/* name */
+   NULL,
+   alloc_texrect_data,
+   free_texrect_data,
+   NULL,
+   run_texrect_stage
+};
+

src/mesa/drivers/dri/r300/r300_state.c

@@ -1019,6 +1019,7 @@ void r300_setup_textures(GLcontext *ctx)
 	R300_STATECHANGE(r300, tex.unknown1);
 	R300_STATECHANGE(r300, tex.size);
 	R300_STATECHANGE(r300, tex.format);
+	R300_STATECHANGE(r300, tex.pitch);
 	R300_STATECHANGE(r300, tex.offset);
 	R300_STATECHANGE(r300, tex.unknown4);
 	R300_STATECHANGE(r300, tex.border_color);
@@ -1070,6 +1071,7 @@ void r300_setup_textures(GLcontext *ctx)
 			r300->hw.tex.size.cmd[R300_TEX_VALUE_0+i]=t->size;
 			r300->hw.tex.format.cmd[R300_TEX_VALUE_0+i]=t->format;
 			//fprintf(stderr, "t->format=%08x\n", t->format);
+			r300->hw.tex.pitch.cmd[R300_TEX_VALUE_0+i]=t->pitch_reg;
 			r300->hw.tex.offset.cmd[R300_TEX_VALUE_0+i]=t->offset;
 			r300->hw.tex.unknown4.cmd[R300_TEX_VALUE_0+i]=0x0;
 			r300->hw.tex.border_color.cmd[R300_TEX_VALUE_0+i]=t->pp_border_color;
@@ -1080,6 +1082,7 @@ void r300_setup_textures(GLcontext *ctx)
 	((drm_r300_cmd_header_t*)r300->hw.tex.unknown1.cmd)->packet0.count = max_texture_unit+1;
 	((drm_r300_cmd_header_t*)r300->hw.tex.size.cmd)->packet0.count = max_texture_unit+1;
 	((drm_r300_cmd_header_t*)r300->hw.tex.format.cmd)->packet0.count = max_texture_unit+1;
+	((drm_r300_cmd_header_t*)r300->hw.tex.pitch.cmd)->packet0.count = max_texture_unit+1;
 	((drm_r300_cmd_header_t*)r300->hw.tex.offset.cmd)->packet0.count = max_texture_unit+1;
 	((drm_r300_cmd_header_t*)r300->hw.tex.unknown4.cmd)->packet0.count = max_texture_unit+1;
 	((drm_r300_cmd_header_t*)r300->hw.tex.border_color.cmd)->packet0.count = max_texture_unit+1;

src/mesa/drivers/dri/r300/r300_texmem.c

@@ -183,7 +183,7 @@ static void r300UploadRectSubImage(r300ContextPtr rmesa,
 	 */
 	width = texImage->Width;
 	height = texImage->Height;
-	dstPitch = t->pitch + 32;
+	dstPitch = t->pitch;
 
 	if (rmesa->prefer_gart_client_texturing && texImage->IsClientData) {
 		/* In this case, could also use GART texturing.  This is

src/mesa/drivers/dri/r300/r300_texstate.c

@@ -243,6 +243,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	    
 	  } else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
 	    size = ((texImage->Width * texelBytes + 63) & ~63) * texImage->Height;
+	    blitWidth = 64 / texelBytes;
 	  } else {
 	    int w = (texImage->Width * texelBytes + 31) & ~31;
 	    size = w * texImage->Height * texImage->Depth;
@@ -363,8 +364,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	}
 
 	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)
-		        |((log2Width>log2Height)?log2Width:log2Height)<<R300_TX_SIZE_SHIFT);
+			|((tObj->Image[0][t->base.firstLevel]->Height - 1) << R300_TX_HEIGHTMASK_SHIFT));
 
 	/* Only need to round to nearest 32 for textures, but the blitter
 	 * requires 64-byte aligned pitches, and we may/may not need the
@@ -373,11 +373,19 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	if (baseImage->IsCompressed)
 		t->pitch =
 		    (tObj->Image[0][t->base.firstLevel]->Width + 63) & ~(63);
-	else
+	else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+		unsigned int align = blitWidth - 1;
+		t->pitch = ((tObj->Image[0][t->base.firstLevel]->Width *
+		      texelBytes) + 63) & ~(63);
+		t->size |= R300_TX_SIZE_TXPITCH_EN;
+		t->pitch_reg = (((tObj->Image[0][t->base.firstLevel]->Width) + align) & ~align) - 1;
+	}
+	else {
+		t->size |= ((log2Width>log2Height)?log2Width:log2Height)<<R300_TX_SIZE_SHIFT;
 		t->pitch =
 		    ((tObj->Image[0][t->base.firstLevel]->Width *
 		      texelBytes) + 63) & ~(63);
-	t->pitch -= 32;
+	}
 
 	t->dirty_state = TEX_ALL;