remove code to deal with non-normalized texture coordinates for tex rect targets (swtcl tex coord translation stage, tcl tex matrix adaption) and use the chip's native handling of such coords instead (!!!). Seems noone noticed those bits in the se_coord_fmt reg, even though it works fairly similar to r200 (except it's set per-unit and always active, so only enable it for texture rectangles).

5 files changed, 42 insertions, 168 deletions

diff --git a/src/mesa/drivers/dri/radeon/radeon_context.c b/src/mesa/drivers/dri/radeon/radeon_context.c
index 21161d2f69d..2167e7afe82 100644
--- a/src/mesa/drivers/dri/radeon/radeon_context.c
+++ b/src/mesa/drivers/dri/radeon/radeon_context.c
@@ -157,7 +157,6 @@ const struct dri_extension card_extensions[] =
     { NULL,                                NULL }
 };
 
-extern const struct tnl_pipeline_stage _radeon_texrect_stage;
 extern const struct tnl_pipeline_stage _radeon_render_stage;
 extern const struct tnl_pipeline_stage _radeon_tcl_stage;
 
@@ -176,10 +175,6 @@ static const struct tnl_pipeline_stage *radeon_pipeline[] = {
    &_tnl_texgen_stage,
    &_tnl_texture_transform_stage,
 
-   /* Scale texture rectangle to 0..1.
-    */
-   &_radeon_texrect_stage,
-
    &_radeon_render_stage,
    &_tnl_render_stage,		/* FALLBACK:  */
    NULL,
diff --git a/src/mesa/drivers/dri/radeon/radeon_state.c b/src/mesa/drivers/dri/radeon/radeon_state.c
index 307c1f341b4..e3762f485b4 100644
--- a/src/mesa/drivers/dri/radeon/radeon_state.c
+++ b/src/mesa/drivers/dri/radeon/radeon_state.c
@@ -2017,26 +2017,6 @@ static void radeonLightingSpaceChange( GLcontext *ctx )
  * Deferred state management - matrices, textures, other?
  */
 
-static void texmat_set_texrect( radeonContextPtr rmesa,
-				struct gl_texture_object *tObj, GLuint unit )
-{
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   _math_matrix_set_identity( &rmesa->tmpmat[unit] );
-   rmesa->tmpmat[unit].m[0] = 1.0 / baseImage->Width;
-   rmesa->tmpmat[unit].m[5] = 1.0 / baseImage->Height;
-
-}
-
-static void texmat_fixup_texrect( radeonContextPtr rmesa,
-				  struct gl_texture_object *tObj, GLuint unit )
-{
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   GLuint i;
-   for (i = 0; i < 4; i++) {
-      rmesa->tmpmat[unit].m[i] = rmesa->tmpmat[unit].m[i] / baseImage->Width;
-      rmesa->tmpmat[unit].m[i+4] = rmesa->tmpmat[unit].m[i+4] / baseImage->Height;
-   }}
-
 
 void radeonUploadTexMatrix( radeonContextPtr rmesa,
 			    int unit, GLboolean swapcols )
@@ -2177,15 +2157,6 @@ static void update_texturematrix( GLcontext *ctx )
 	    _math_matrix_copy( &rmesa->tmpmat[unit], &rmesa->TexGenMatrix[unit] );
 	    needMatrix = GL_TRUE;
 	 }
-	 if (ctx->Texture.Unit[unit]._ReallyEnabled == TEXTURE_RECT_BIT) {
-	    texMatEnabled |= (RADEON_TEXGEN_TEXMAT_0_ENABLE |
-			      RADEON_TEXMAT_0_ENABLE) << unit;
-	    if (needMatrix)
-	       texmat_fixup_texrect( rmesa, ctx->Texture.Unit[unit]._Current, unit );
-	    else
-	       texmat_set_texrect( rmesa, ctx->Texture.Unit[unit]._Current, unit );
-	    needMatrix = GL_TRUE;
-	 }
 	 if (needMatrix) {
 	    rmesa->NeedTexMatrix |= 1 << unit;
 	    radeonUploadTexMatrix( rmesa, unit,
diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.c b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
index 4d5bbbd1f18..e36a710d33a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_swtcl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
@@ -246,7 +246,11 @@ void radeonChooseVertexState( GLcontext *ctx )
    radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
    TNLcontext *tnl = TNL_CONTEXT(ctx);
 
-   GLuint se_coord_fmt;
+   GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
+   
+   se_coord_fmt &= ~(RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
+		     RADEON_VTX_Z_PRE_MULT_1_OVER_W0 |
+		     RADEON_VTX_W0_IS_NOT_1_OVER_W0);
 
    /* We must ensure that we don't do _tnl_need_projected_coords while in a
     * rasterization fallback.  As this function will be called again when we
@@ -263,14 +267,12 @@ void radeonChooseVertexState( GLcontext *ctx )
        !RENDERINPUTS_TEST( tnl->render_inputs_bitset, _TNL_ATTRIB_COLOR1 ))
        || (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
       rmesa->swtcl.needproj = GL_TRUE;
-      se_coord_fmt = (RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
-		      RADEON_VTX_Z_PRE_MULT_1_OVER_W0 |
-		      RADEON_TEX1_W_ROUTING_USE_Q1);
+      se_coord_fmt |= (RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
+		      RADEON_VTX_Z_PRE_MULT_1_OVER_W0);
    }
    else {
       rmesa->swtcl.needproj = GL_FALSE;
-      se_coord_fmt = (RADEON_VTX_W0_IS_NOT_1_OVER_W0 |
-		      RADEON_TEX1_W_ROUTING_USE_Q1);
+      se_coord_fmt |= (RADEON_VTX_W0_IS_NOT_1_OVER_W0);
    }
 
    _tnl_need_projected_coords( ctx, rmesa->swtcl.needproj );
@@ -458,7 +460,6 @@ static GLboolean radeon_run_render( GLcontext *ctx,
 
 
 
-
 const struct tnl_pipeline_stage _radeon_render_stage =
 {
    "radeon render",
@@ -472,113 +473,6 @@ const struct tnl_pipeline_stage _radeon_render_stage =
 
 /**************************************************************************/
 
-/* Radeon 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);
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   GLuint i;
-
-   if (rmesa->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 _radeon_texrect_stage =
-{
-   "radeon texrect stage",			/* name */
-   NULL,
-   alloc_texrect_data,
-   free_texrect_data,
-   NULL,
-   run_texrect_stage
-};
-
-
-/**************************************************************************/
-
 
 static const GLuint reduced_hw_prim[GL_POLYGON+1] = {
    RADEON_CP_VC_CNTL_PRIM_TYPE_POINT,
diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.c b/src/mesa/drivers/dri/radeon/radeon_tcl.c
index cd421567b61..ffd49b525b8 100644
--- a/src/mesa/drivers/dri/radeon/radeon_tcl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_tcl.c
@@ -493,8 +493,12 @@ static void transition_to_hwtnl( GLcontext *ctx )
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
-   GLuint se_coord_fmt = (RADEON_VTX_W0_IS_NOT_1_OVER_W0 |
-			  RADEON_TEX1_W_ROUTING_USE_Q1);
+   GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
+
+   se_coord_fmt &= ~(RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
+		     RADEON_VTX_Z_PRE_MULT_1_OVER_W0 |
+		     RADEON_VTX_W0_IS_NOT_1_OVER_W0);
+   se_coord_fmt |= RADEON_VTX_W0_IS_NOT_1_OVER_W0;
 
    if ( se_coord_fmt != rmesa->hw.set.cmd[SET_SE_COORDFMT] ) {
       RADEON_STATECHANGE( rmesa, set );
diff --git a/src/mesa/drivers/dri/radeon/radeon_texstate.c b/src/mesa/drivers/dri/radeon/radeon_texstate.c
index 0a3c20f7071..b6a472ffa73 100644
--- a/src/mesa/drivers/dri/radeon/radeon_texstate.c
+++ b/src/mesa/drivers/dri/radeon/radeon_texstate.c
@@ -848,6 +848,7 @@ static void import_tex_obj_state( radeonContextPtr rmesa,
 				  radeonTexObjPtr texobj )
 {
    GLuint *cmd = RADEON_DB_STATE( tex[unit] );
+   GLuint se_coord_fmt = rmesa->hw.set.cmd[SET_SE_COORDFMT];
 
    cmd[TEX_PP_TXFILTER] &= ~TEXOBJ_TXFILTER_MASK;
    cmd[TEX_PP_TXFILTER] |= texobj->pp_txfilter & TEXOBJ_TXFILTER_MASK;
@@ -856,31 +857,40 @@ static void import_tex_obj_state( radeonContextPtr rmesa,
    cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset;
    cmd[TEX_PP_BORDER_COLOR] = texobj->pp_border_color;
 
-   if (texobj->base.tObj->Target == GL_TEXTURE_CUBE_MAP) {
-      GLuint *cube_cmd = RADEON_DB_STATE( cube[unit] );
-      GLuint bytesPerFace = texobj->base.totalSize / 6;
-      ASSERT(texobj->base.totalSize % 6 == 0);
-
-      cube_cmd[CUBE_PP_CUBIC_FACES] = texobj->pp_cubic_faces;
-      /* dont know if this setup conforms to OpenGL.. 
-       * at least it matches the behavior of mesa software renderer
-       */
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_0] = texobj->pp_txoffset; /* right */
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_1] = texobj->pp_txoffset + 1 * bytesPerFace; /* left */
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_2] = texobj->pp_txoffset + 2 * bytesPerFace; /* top */
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_3] = texobj->pp_txoffset + 3 * bytesPerFace; /* bottom */
-      cube_cmd[CUBE_PP_CUBIC_OFFSET_4] = texobj->pp_txoffset + 4 * bytesPerFace; /* front */
-      RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.cube[unit] );
-      cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset + 5 * bytesPerFace; /* back */
-   }
-   else if (texobj->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+   if (texobj->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
       GLuint *txr_cmd = RADEON_DB_STATE( txr[unit] );
       txr_cmd[TXR_PP_TEX_SIZE] = texobj->pp_txsize; /* NPOT only! */
       txr_cmd[TXR_PP_TEX_PITCH] = texobj->pp_txpitch; /* NPOT only! */
       RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.txr[unit] );
+      se_coord_fmt |= RADEON_VTX_ST0_NONPARAMETRIC << unit;
+   }
+   else {
+      se_coord_fmt &= ~(RADEON_VTX_ST0_NONPARAMETRIC << unit);
+
+      if (texobj->base.tObj->Target == GL_TEXTURE_CUBE_MAP) {
+	 GLuint *cube_cmd = RADEON_DB_STATE( cube[unit] );
+	 GLuint bytesPerFace = texobj->base.totalSize / 6;
+	 ASSERT(texobj->base.totalSize % 6 == 0);
+
+	 cube_cmd[CUBE_PP_CUBIC_FACES] = texobj->pp_cubic_faces;
+	 /* dont know if this setup conforms to OpenGL.. 
+	  * at least it matches the behavior of mesa software renderer
+	  */
+	 cube_cmd[CUBE_PP_CUBIC_OFFSET_0] = texobj->pp_txoffset; /* right */
+	 cube_cmd[CUBE_PP_CUBIC_OFFSET_1] = texobj->pp_txoffset + 1 * bytesPerFace; /* left */
+	 cube_cmd[CUBE_PP_CUBIC_OFFSET_2] = texobj->pp_txoffset + 2 * bytesPerFace; /* top */
+	 cube_cmd[CUBE_PP_CUBIC_OFFSET_3] = texobj->pp_txoffset + 3 * bytesPerFace; /* bottom */
+	 cube_cmd[CUBE_PP_CUBIC_OFFSET_4] = texobj->pp_txoffset + 4 * bytesPerFace; /* front */
+	 RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.cube[unit] );
+	 cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset + 5 * bytesPerFace; /* back */
+      }
    }
 
    RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.tex[unit] );
+   if (se_coord_fmt != rmesa->hw.set.cmd[SET_SE_COORDFMT]) {
+      RADEON_STATECHANGE( rmesa, set );
+      rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt;
+   }
 
    texobj->dirty_state &= ~(1<<unit);
 }