R300_PFS_NODE_LAST_NODE -> R300_PFS_NODE_OUTPUT_COLOR
[mesa.git] / src / mesa / drivers / dri / r300 / r300_render.c
index 4ecb2905b1081228450e696326d26c157f48af1e..fbd2135884ae201b7dfad54fcf8340c1cc1875b5 100644 (file)
@@ -44,6 +44,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "swrast_setup/swrast_setup.h"
 #include "array_cache/acache.h"
 #include "tnl/tnl.h"
+#include "tnl/t_vp_build.h"
 
 #include "radeon_reg.h"
 #include "radeon_macros.h"
@@ -58,6 +59,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_maos.h"
 #include "r300_emit.h"
 
+extern int future_hw_tcl_on;
+
 /**********************************************************************
 *                     Hardware rasterization
 *
@@ -67,10 +70,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 static int r300_get_primitive_type(r300ContextPtr rmesa, GLcontext *ctx, int prim)
 {
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   GLuint i;
-   int type=-1;
+       int type=-1;
 
        switch (prim & PRIM_MODE_MASK) {
        case GL_POINTS:
@@ -118,11 +118,8 @@ static int r300_get_num_verts(r300ContextPtr rmesa,
        int num_verts,
        int prim)
 {
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   GLuint i;
-   int type=-1, verts_off=0;
-   char *name="UNKNOWN";
+       int verts_off=0;
+       char *name="UNKNOWN";
 
        switch (prim & PRIM_MODE_MASK) {
        case GL_POINTS:
@@ -181,51 +178,20 @@ static int r300_get_num_verts(r300ContextPtr rmesa,
                break;
        }
 
-       if(num_verts - verts_off == 0){
-               WARN_ONCE("user error: Need more than %d vertices to draw primitive %s !\n", num_verts, name);
-               return 0;
-       }
+       if (RADEON_DEBUG & DEBUG_VERTS) {
+               if (num_verts - verts_off == 0) {
+                       WARN_ONCE("user error: Need more than %d vertices to draw primitive %s !\n", num_verts, name);
+                       return 0;
+               }
 
-       if(verts_off > 0){
-               WARN_ONCE("user error: %d is not a valid number of vertices for primitive %s !\n", num_verts, name);
+               if (verts_off > 0) {
+                       WARN_ONCE("user error: %d is not a valid number of vertices for primitive %s !\n", num_verts, name);
+               }
        }
 
        return num_verts - verts_off;
 }
 
-void dump_inputs(GLcontext *ctx, int render_inputs)
-{
-       int k;
-       fprintf(stderr, "inputs:");
-       fprintf(stderr, "%08x ", render_inputs);
-
-       if(render_inputs & _TNL_BIT_POS)
-               fprintf(stderr, "_TNL_BIT_POS ");
-       if(render_inputs & _TNL_BIT_NORMAL)
-               fprintf(stderr, "_TNL_BIT_NORMAL ");
-
-               /* color components */
-       if(render_inputs & _TNL_BIT_COLOR0)
-               fprintf(stderr, "_TNL_BIT_COLOR0 ");
-       if(render_inputs & _TNL_BIT_COLOR1)
-               fprintf(stderr, "_TNL_BIT_COLOR1 ");
-
-       if(render_inputs & _TNL_BIT_FOG)
-               fprintf(stderr, "_TNL_BIT_FOG ");
-
-               /* texture coordinates */
-       for(k=0;k < ctx->Const.MaxTextureUnits;k++)
-               if(render_inputs & (_TNL_BIT_TEX0<<k))
-                       fprintf(stderr, "_TNL_BIT_TEX%d ", k);
-
-       if(render_inputs & _TNL_BIT_INDEX)
-               fprintf(stderr, "_TNL_BIT_INDEX ");
-       if(render_inputs & _TNL_BIT_POINTSIZE)
-               fprintf(stderr, "_TNL_BIT_POINTSIZE ");
-
-       fprintf(stderr, "\n");
-}
-
 /* This function compiles GL context into state registers that
    describe data routing inside of R300 pipeline.
 
@@ -242,16 +208,19 @@ void dump_inputs(GLcontext *ctx, int render_inputs)
 
 static GLfloat default_vector[4]={0.0, 0.0, 0.0, 1.0};
 
-#define output_vector(v, i) \
-       { \
+#define output_vector(v, i) { \
        int _i; \
        for(_i=0;_i<v->size;_i++){ \
-               efloat(VEC_ELT(v, GLfloat, i)[_i]); \
+               if(VB->Elts){ \
+                       efloat(VEC_ELT(v, GLfloat, VB->Elts[i])[_i]); \
+               }else{ \
+                       efloat(VEC_ELT(v, GLfloat, i)[_i]); \
                } \
+       } \
        for(_i=v->size;_i<4;_i++){ \
-                       efloat(default_vector[_i]); \
-                       } \
-       }
+               efloat(default_vector[_i]); \
+       } \
+}
 
 /* Immediate implementation - vertex data is sent via command stream */
 
@@ -305,7 +274,6 @@ static void r300_render_immediate_primitive(r300ContextPtr rmesa,
        return;
        }
 
-       //dump_inputs(ctx, render_inputs); return ;
 
    start_immediate_packet(num_verts, type, 4*rmesa->state.aos_count);
 
@@ -361,9 +329,6 @@ static GLboolean r300_run_immediate_render(GLcontext *ctx,
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    GLuint i;
-   /* Only do 2d textures */
-   struct gl_texture_object *to=ctx->Texture.Unit[0].Current2D;
-   r300TexObjPtr t=to->DriverData;
    LOCAL_VARS
 
 
@@ -414,23 +379,6 @@ static GLboolean r300_run_immediate_render(GLcontext *ctx,
 
    r300EmitState(rmesa);
 
-#if 0
-   reg_start(R300_RB3D_COLORMASK, 0);
-       e32(0xf);
-
-   vsf_start_fragment(0x406, 4);
-   efloat(0.0);
-   efloat(0.0);
-   efloat(0.0);
-   efloat(1.0);
-
-   vsf_start_fragment(0x400, 4);
-   efloat(0.0);
-   efloat(0.0);
-   efloat(0.0);
-   efloat(1.0);
-#endif
-
 /* Setup INPUT_ROUTE and INPUT_CNTL */
        r300EmitArrays(ctx, GL_TRUE);
 
@@ -459,29 +407,86 @@ static GLboolean r300_run_immediate_render(GLcontext *ctx,
    return GL_FALSE;
 }
 
+
 /* vertex buffer implementation */
 
-static void inline fire_EB(PREFIX unsigned long addr, int vertex_count, int type)
+static void inline fire_EB(PREFIX unsigned long addr, int vertex_count, int type, int elt_size)
 {
        LOCAL_VARS
        unsigned long addr_a;
+       unsigned long t_addr;
+       unsigned long magic_1, magic_2;
+       GLcontext *ctx;
+       ctx = rmesa->radeon.glCtx; 
+       
+       assert(elt_size == 2 || elt_size == 4);
        
-       if(addr & 1){
+       if(addr & (elt_size-1)){
                WARN_ONCE("Badly aligned buffer\n");
                return ;
        }
-       addr_a = 0; /*addr & 0x1c;*/
+#ifdef OPTIMIZE_ELTS
+       addr_a = 0;
+       
+       magic_1 = (addr % 32) / 4;
+       t_addr = addr & (~0x1d);
+       magic_2 = (vertex_count + 1 + (t_addr & 0x2)) / 2 + magic_1;
+       
+       check_space(6);
+       
+       start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0);
+       if(elt_size == 4){
+               e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+       } else {
+               e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type);
+       }
+
+       start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2);
+       if(elt_size == 4){
+               e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
+               e32(addr /*& 0xffffffe3*/);
+       } else {
+               e32(R300_EB_UNK1 | (magic_1 << 16) | R300_EB_UNK2);
+               e32(t_addr);
+       }
+       
+       if(elt_size == 4){
+               e32(vertex_count /*+ addr_a/4*/); /* Total number of dwords needed? */
+       } else {
+               e32(magic_2); /* Total number of dwords needed? */
+       }
+       //cp_delay(PASS_PREFIX 1);
+#if 0
+       fprintf(stderr, "magic_1 %d\n", magic_1);
+       fprintf(stderr, "t_addr %x\n", t_addr);
+       fprintf(stderr, "magic_2 %d\n", magic_2);
+       exit(1);
+#endif
+#else
+       (void)magic_2, (void)magic_1, (void)t_addr;
+       
+       addr_a = 0;
        
        check_space(6);
        
        start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0);
-       /* TODO: R300_VAP_VF_CNTL__INDEX_SIZE_32bit . */
-       e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type);
+       if(elt_size == 4){
+               e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+       } else {
+               e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type);
+       }
 
        start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2);
-       e32(R300_EB_UNK1 | (addr_a << 16) | R300_EB_UNK2);
+       e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
        e32(addr /*& 0xffffffe3*/);
-       e32((vertex_count+1)/2 /*+ addr_a/4*/); /* Total number of dwords needed? */
+       
+       if(elt_size == 4){
+               e32(vertex_count /*+ addr_a/4*/); /* Total number of dwords needed? */
+       } else {
+               e32((vertex_count+1)/2 /*+ addr_a/4*/); /* Total number of dwords needed? */
+       }
+       //cp_delay(PASS_PREFIX 1);
+#endif 
 }
 
 static void r300_render_vb_primitive(r300ContextPtr rmesa,
@@ -491,11 +496,7 @@ static void r300_render_vb_primitive(r300ContextPtr rmesa,
        int prim)
 {
    int type, num_verts;
-   radeonScreenPtr rsp=rmesa->radeon.radeonScreen;
    LOCAL_VARS
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   int i;
 
    type=r300_get_primitive_type(rmesa, ctx, prim);
    num_verts=r300_get_num_verts(rmesa, ctx, end-start, prim);
@@ -505,11 +506,11 @@ static void r300_render_vb_primitive(r300ContextPtr rmesa,
    if(rmesa->state.Elts){
        r300EmitAOS(rmesa, rmesa->state.aos_count, 0);
 #if 0
+       int i;
        start_index32_packet(num_verts, type);
        for(i=0; i < num_verts; i++)
                e32(rmesa->state.Elts[start+i]); /* start ? */
 #else
-       WARN_ONCE("Rendering with elt buffers\n");
        if(num_verts == 1){
                start_index32_packet(num_verts, type);
                e32(rmesa->state.Elts[start]);
@@ -520,8 +521,8 @@ static void r300_render_vb_primitive(r300ContextPtr rmesa,
                WARN_ONCE("Too many elts\n");
                return;
        }
-       r300EmitElts(ctx, rmesa->state.Elts+start, num_verts);
-       fire_EB(PASS_PREFIX GET_START(&(rmesa->state.elt_dma)), num_verts, type);
+       r300EmitElts(ctx, rmesa->state.Elts+start, num_verts, 4);
+       fire_EB(PASS_PREFIX GET_START(&(rmesa->state.elt_dma)), num_verts, type, 4);
 #endif
    }else{
           r300EmitAOS(rmesa, rmesa->state.aos_count, start);
@@ -532,20 +533,22 @@ static void r300_render_vb_primitive(r300ContextPtr rmesa,
 static GLboolean r300_run_vb_render(GLcontext *ctx,
                                 struct tnl_pipeline_stage *stage)
 {
-   r300ContextPtr rmesa = R300_CONTEXT(ctx);
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   int i, j;
-   LOCAL_VARS
+       r300ContextPtr rmesa = R300_CONTEXT(ctx);
+       TNLcontext *tnl = TNL_CONTEXT(ctx);
+       struct vertex_buffer *VB = &tnl->vb;
+       int i;
+       LOCAL_VARS
    
        if (RADEON_DEBUG & DEBUG_PRIMS)
                fprintf(stderr, "%s\n", __FUNCTION__);
        
-
-       r300ReleaseArrays(ctx);
+       
+       r300UpdateShaders(rmesa);
+       
+       r300ReleaseArrays(ctx);
        r300EmitArrays(ctx, GL_FALSE);
-       //dump_inputs(ctx, rmesa->state.render_inputs);
-
+       
+       r300UpdateShaderStates(rmesa);
 //     LOCK_HARDWARE(&(rmesa->radeon));
 
        reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
@@ -554,7 +557,7 @@ static GLboolean r300_run_vb_render(GLcontext *ctx,
        reg_start(0x4f18,0);
        e32(0x00000003);
        r300EmitState(rmesa);
-
+       
        rmesa->state.Elts = VB->Elts;
 
        for(i=0; i < VB->PrimitiveCount; i++){
@@ -571,6 +574,9 @@ static GLboolean r300_run_vb_render(GLcontext *ctx,
        reg_start(0x4f18,0);
        e32(0x00000003);
 
+#ifdef USER_BUFFERS
+       r300UseArrays(ctx);
+#endif
 //     end_3d(PASS_PREFIX_VOID);
 
    /* Flush state - we are done drawing.. */
@@ -581,187 +587,280 @@ static GLboolean r300_run_vb_render(GLcontext *ctx,
        return GL_FALSE;
 }
 
-/**
- * Called by the pipeline manager to render a batch of primitives.
- * We can return true to pass on to the next stage (i.e. software
- * rasterization) or false to indicate that the pipeline has finished
- * after we render something.
- */
-static GLboolean r300_run_render(GLcontext *ctx,
-                                struct tnl_pipeline_stage *stage)
+#ifdef RADEON_VTXFMT_A
+
+static void r300_render_vb_primitive_vtxfmt_a(r300ContextPtr rmesa,
+       GLcontext *ctx,
+       int start,
+       int end,
+       int prim)
 {
-   r300ContextPtr rmesa = R300_CONTEXT(ctx);
+   int type, num_verts;
+   radeonScreenPtr rsp=rmesa->radeon.radeonScreen;
+   LOCAL_VARS
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
-   GLuint i;
-
-       if (RADEON_DEBUG & DEBUG_PRIMS)
-               fprintf(stderr, "%s\n", __FUNCTION__);
+   int i;
 
+   type=r300_get_primitive_type(rmesa, ctx, prim);
+   num_verts=r300_get_num_verts(rmesa, ctx, end-start, prim);
 
-#if 1
+   if(type<0 || num_verts <= 0)return;
 
+   if(rmesa->state.VB.Elts){
+       r300EmitAOS(rmesa, rmesa->state.aos_count, /*0*/start);
 #if 0
-               return r300_run_immediate_render(ctx, stage);
-#else
-               return r300_run_vb_render(ctx, stage);
-#endif
+       start_index32_packet(num_verts, type);
+       for(i=0; i < num_verts; i++)
+               e32(((unsigned long *)rmesa->state.VB.Elts)[i]/*rmesa->state.Elts[start+i]*/); /* start ? */
 #else
-       return GL_TRUE;
+       if(num_verts == 1){
+               //start_index32_packet(num_verts, type);
+               //e32(rmesa->state.Elts[start]);
+               return;
+       }
+       
+       if(num_verts > 65535){ /* not implemented yet */
+               WARN_ONCE("Too many elts\n");
+               return;
+       }
+       
+       r300EmitElts(ctx, rmesa->state.VB.Elts, num_verts, rmesa->state.VB.elt_size);
+       fire_EB(PASS_PREFIX rmesa->state.elt_dma.aos_offset, num_verts, type, rmesa->state.VB.elt_size);
 #endif
+   }else{
+          r300EmitAOS(rmesa, rmesa->state.aos_count, start);
+          fire_AOS(PASS_PREFIX num_verts, type);
+   }
+}
 
-#if 0
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   GLuint i;
+void dump_array(struct r300_dma_region *rvb, int count)
+{
+       int *out = (int *)(rvb->address + rvb->start);
+       int i, ci;
+       
+       for (i=0; i < count; i++) {
+               fprintf(stderr, "{");
+               if (rvb->aos_format == AOS_FORMAT_FLOAT)
+                       for (ci=0; ci < rvb->aos_size; ci++)
+                               fprintf(stderr, "%f ", ((float *)out)[ci]);
+               else
+                       for (ci=0; ci < rvb->aos_size; ci++)
+                               fprintf(stderr, "%d ", ((unsigned char *)out)[ci]);
+               fprintf(stderr, "}");
+               
+               out += rvb->aos_stride;
+       }
 
-   /* Don't handle clipping or indexed vertices or vertex manipulations.
-    */
-   if (mmesa->RenderIndex != 0 ||
-       !mga_validate_render( ctx, VB )) {
-      return GL_TRUE;
-   }
+       fprintf(stderr, "\n");
+}
+               
+void dump_dt(struct dt *dt, int count)
+{
+       int *out = dt->data;
+       int i, ci;
+       
+       fprintf(stderr, "base at %p ", out);
+       
+       for (i=0; i < count; i++){
+               fprintf(stderr, "{");
+               if (dt->type == GL_FLOAT)
+                       for (ci=0; ci < dt->size; ci++)
+                               fprintf(stderr, "%f ", ((float *)out)[ci]);
+               else
+                       for (ci=0; ci < dt->size; ci++)
+                               fprintf(stderr, "%d ", ((unsigned char *)out)[ci]);
+               fprintf(stderr, "}");
+               
+               out = (char *)out + dt->stride;
+       }
+       
+       fprintf(stderr, "\n");
+}
 
-   tnl->Driver.Render.Start( ctx );
-   mmesa->SetupNewInputs = ~0;
+/*static */GLboolean r300_run_vb_render_vtxfmt_a(GLcontext *ctx,
+                                struct tnl_pipeline_stage *stage)
+{
+   r300ContextPtr rmesa = R300_CONTEXT(ctx);
+   //TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct radeon_vertex_buffer *VB = &rmesa->state.VB; //&tnl->vb;
+   int i, j;
+   LOCAL_VARS
+   
+       if (RADEON_DEBUG & DEBUG_PRIMS)
+               fprintf(stderr, "%s\n", __FUNCTION__);
+       
+       if (rmesa->state.VB.LockCount == 0) {
+               r300ReleaseArrays(ctx);
+               r300EmitArraysVtx(ctx, GL_FALSE);
+               
+               r300UpdateShaderStates(rmesa);
+       } else {
+               /* TODO: Figure out why do we need these. */
+               R300_STATECHANGE(rmesa, vir[0]);
+               R300_STATECHANGE(rmesa, vir[1]);
+               R300_STATECHANGE(rmesa, vic);
+               R300_STATECHANGE(rmesa, vof);
+               
+#if 0          
+               fprintf(stderr, "dt:\n");
+               for(i=0; i < VERT_ATTRIB_MAX; i++){
+                       fprintf(stderr, "dt %d:", i);
+                       dump_dt(&rmesa->state.VB.AttribPtr[i], VB->Count);
+               }
+               
+               fprintf(stderr, "before:\n");
+               for(i=0; i < rmesa->state.aos_count; i++){
+                       fprintf(stderr, "aos %d:", i);
+                       dump_array(&rmesa->state.aos[i], VB->Count);
+               }
+#endif
+#if 0
+               r300ReleaseArrays(ctx);
+               r300EmitArraysVtx(ctx, GL_FALSE);
+                       
+               fprintf(stderr, "after:\n");
+               for(i=0; i < rmesa->state.aos_count; i++){
+                       fprintf(stderr, "aos %d:", i);
+                       dump_array(&rmesa->state.aos[i], VB->Count);
+               }
+#endif
+       }
+       
+//     LOCK_HARDWARE(&(rmesa->radeon));
 
-   for (i = 0 ; i < VB->PrimitiveCount ; i++)
-   {
-      GLuint prim = VB->Primitive[i].mode;
-      GLuint start = VB->Primitive[i].start;
-      GLuint length = VB->Primitive[i].count;
+       reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+       e32(0x0000000a);
 
-      if (!length)
-        continue;
+       reg_start(0x4f18,0);
+       e32(0x00000003);
+#if 0
+       reg_start(R300_VAP_PVS_WAITIDLE,0);
+               e32(0x00000000);
+#endif
+       r300EmitState(rmesa);
+       
+       for(i=0; i < VB->PrimitiveCount; i++){
+               GLuint prim = VB->Primitive[i].mode;
+               GLuint start = VB->Primitive[i].start;
+               GLuint length = VB->Primitive[i].count;
+               
+               r300_render_vb_primitive_vtxfmt_a(rmesa, ctx, start, start + length, prim);
+       }
 
-      mga_render_tab_verts[prim & PRIM_MODE_MASK]( ctx, start, start + length,
-                                                  prim);
-   }
+       reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+       e32(0x0000000a/*0x2*/);
 
-   tnl->Driver.Render.Finish( ctx );
+       reg_start(0x4f18,0);
+       e32(0x00000003/*0x1*/);
 
-   return GL_FALSE;            /* finished the pipe */
+#ifdef USER_BUFFERS
+       r300UseArrays(ctx);
 #endif
-}
+//     end_3d(PASS_PREFIX_VOID);
 
+   /* Flush state - we are done drawing.. */
+//     r300FlushCmdBufLocked(rmesa, __FUNCTION__);
+//     radeonWaitForIdleLocked(&(rmesa->radeon));
+
+//     UNLOCK_HARDWARE(&(rmesa->radeon));
+       return GL_FALSE;
+}
+#endif
 
-/**
- * Called by the pipeline manager once before rendering.
- * We check the GL state here to
- *  a) decide whether we can do the current state in hardware and
- *  b) update hardware registers
- */
 #define FALLBACK_IF(expr) \
 do {                                                                           \
        if (expr) {                                                             \
-               if (1 || RADEON_DEBUG & DEBUG_FALLBACKS)                                \
-                       fprintf(stderr, "%s: fallback:%s\n",                    \
-                               __FUNCTION__, #expr);                           \
-               /*stage->active = GL_FALSE*/;                                   \
-               return;                                                         \
+               if (1 || RADEON_DEBUG & DEBUG_FALLBACKS)                        \
+                       WARN_ONCE("fallback:%s\n", #expr);                      \
+               return GL_TRUE;                                                 \
        }                                                                       \
 } while(0)
 
-static void r300_check_render(GLcontext *ctx, struct tnl_pipeline_stage *stage)
+GLboolean r300Fallback(GLcontext *ctx)
 {
-       r300ContextPtr r300 = R300_CONTEXT(ctx);
-       int i;
 
-       if (RADEON_DEBUG & DEBUG_STATE)
-               fprintf(stderr, "%s\n", __FUNCTION__);
-
-       /* We only support rendering in hardware for now */
-       if (ctx->RenderMode != GL_RENDER) {
-               //stage->active = GL_FALSE;
-               return;
-       }
-               
-
-       /* I'm almost certain I forgot something here */
+       FALLBACK_IF(ctx->RenderMode != GL_RENDER);  // We do not do SELECT or FEEDBACK (yet ?)
+       
 #if 0 /* These should work now.. */
        FALLBACK_IF(ctx->Color.DitherFlag);
        FALLBACK_IF(ctx->Color.AlphaEnabled); // GL_ALPHA_TEST
        FALLBACK_IF(ctx->Color.BlendEnabled); // GL_BLEND
        FALLBACK_IF(ctx->Polygon.OffsetFill); // GL_POLYGON_OFFSET_FILL
 #endif
-       //FALLBACK_IF(ctx->Polygon.OffsetPoint); // GL_POLYGON_OFFSET_POINT
-       //FALLBACK_IF(ctx->Polygon.OffsetLine); // GL_POLYGON_OFFSET_LINE
+       FALLBACK_IF(ctx->Polygon.OffsetPoint); // GL_POLYGON_OFFSET_POINT
+       FALLBACK_IF(ctx->Polygon.OffsetLine); // GL_POLYGON_OFFSET_LINE
        //FALLBACK_IF(ctx->Stencil.Enabled); // GL_STENCIL_TEST
        
        //FALLBACK_IF(ctx->Fog.Enabled); // GL_FOG disable as swtcl doesnt seem to support this
        //FALLBACK_IF(ctx->Polygon.SmoothFlag); // GL_POLYGON_SMOOTH disabling to get blender going
        FALLBACK_IF(ctx->Polygon.StippleFlag); // GL_POLYGON_STIPPLE
        FALLBACK_IF(ctx->Multisample.Enabled); // GL_MULTISAMPLE_ARB
-
-#if 0 /* ut2k3 fails to start if this is on */
-       /* One step at a time - let one texture pass.. */
-       for (i = 1; i < ctx->Const.MaxTextureUnits; i++)
-               FALLBACK_IF(ctx->Texture.Unit[i].Enabled);
-#endif 
        
-       /* Assumed factor reg is found but pattern is still missing */
-       //FALLBACK_IF(ctx->Line.StippleFlag); // GL_LINE_STIPPLE disabling to get blender going
+       
+       FALLBACK_IF(ctx->Line.StippleFlag);
        
        /* HW doesnt appear to directly support these */
-       //FALLBACK_IF(ctx->Line.SmoothFlag); // GL_LINE_SMOOTH disabling to get blender going
+       FALLBACK_IF(ctx->Line.SmoothFlag); // GL_LINE_SMOOTH
        FALLBACK_IF(ctx->Point.SmoothFlag); // GL_POINT_SMOOTH
        /* Rest could be done with vertex fragments */
        if (ctx->Extensions.NV_point_sprite || ctx->Extensions.ARB_point_sprite)
                FALLBACK_IF(ctx->Point.PointSprite); // GL_POINT_SPRITE_NV
-       //GL_POINT_DISTANCE_ATTENUATION_ARB
-       //GL_POINT_FADE_THRESHOLD_SIZE_ARB
        
-       /* let r300_run_render do its job */
-#if 0
-       stage->active = GL_FALSE;
-#endif
+       return GL_FALSE;
 }
 
-
-static void dtr(struct tnl_pipeline_stage *stage)
+/**
+ * Called by the pipeline manager to render a batch of primitives.
+ * We can return true to pass on to the next stage (i.e. software
+ * rasterization) or false to indicate that the pipeline has finished
+ * after we render something.
+ */
+static GLboolean r300_run_render(GLcontext *ctx,
+                                struct tnl_pipeline_stage *stage)
 {
-       (void)stage;
-}
 
-GLboolean r300_create_render(GLcontext *ctx, struct tnl_pipeline_stage *stage){
-       return GL_TRUE; 
-}
+       if (RADEON_DEBUG & DEBUG_PRIMS)
+               fprintf(stderr, "%s\n", __FUNCTION__);
 
+       if (r300Fallback(ctx))
+               return GL_TRUE;
+#if 0
+               return r300_run_immediate_render(ctx, stage);
+#else
+               return r300_run_vb_render(ctx, stage);
+#endif
+}
 
 const struct tnl_pipeline_stage _r300_render_stage = {
        "r300 hw rasterize",
        NULL,
-       r300_create_render,
-       dtr,                    /* destructor */
-       r300_check_render,      /* check */
+       NULL,
+       NULL,
+       NULL,
        r300_run_render         /* run */
 };
-       
+
 static GLboolean r300_run_tcl_render(GLcontext *ctx,
                                 struct tnl_pipeline_stage *stage)
 {
-   r300ContextPtr rmesa = R300_CONTEXT(ctx);
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   GLuint i;
-
+       r300ContextPtr rmesa = R300_CONTEXT(ctx);
+   
+       hw_tcl_on=future_hw_tcl_on;
+   
        if (RADEON_DEBUG & DEBUG_PRIMS)
                fprintf(stderr, "%s\n", __FUNCTION__);
        if(hw_tcl_on == GL_FALSE)
                return GL_TRUE;
-       if(ctx->VertexProgram._Enabled == GL_FALSE){
-               _tnl_UpdateFixedFunctionProgram(ctx);
-               r300ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB, &ctx->_TnlProgram);
-       }
-
+       
+       //r300UpdateShaders(rmesa);
+       //r300UpdateShaderStates(rmesa);
+       
        return r300_run_vb_render(ctx, stage);
 }
 
 static void r300_check_tcl_render(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 {
-       r300ContextPtr r300 = R300_CONTEXT(ctx);
-       int i;
 
        if (RADEON_DEBUG & DEBUG_STATE)
                fprintf(stderr, "%s\n", __FUNCTION__);
@@ -776,8 +875,113 @@ static void r300_check_tcl_render(GLcontext *ctx, struct tnl_pipeline_stage *sta
 const struct tnl_pipeline_stage _r300_tcl_stage = {
        "r300 tcl",
        NULL,
-       r300_create_render,
-       dtr,                    /* destructor */
-       r300_check_tcl_render,  /* check */
+       NULL,
+       NULL,
+       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
+};
+