Merge branch 'r500-support'

Bring in all the r500 3D code into master should be as stable as the r300
code it is landing on top off.

Conflicts:

	src/mesa/drivers/dri/r300/r300_reg.h
	src/mesa/drivers/dri/r300/r300_state.c
	src/mesa/drivers/dri/radeon/radeon_screen.c

diff --cc src/mesa/drivers/dri/r300/r300_reg.h
index d2a81750fb4c59e84c4023f1a611ef56eccd1afd,8b0da0db4e078dbdc4257abc2064a61ebb76ce6a..acdd72b74b40fb261cfd2918706420c1e7bcf6b6
--- 1/src/mesa/drivers/dri/r300/r300_reg.h
--- 2/src/mesa/drivers/dri/r300/r300_reg.h
+++ b/src/mesa/drivers/dri/r300/r300_reg.h
@@@ -67,15 -67,16 +67,15 @@@ USE OR OTHER DEALINGS IN THE SOFTWARE
  
  /*
   * Vertex Array Processing (VAP) Control
 - * Stolen from r200 code from Christoph Brill (It's a guess!)
   */
  #define R300_VAP_CNTL 0x2080
- #     define  R300_VAP_CNTL__PVS_NUM_SLOTS__SHIFT             0
- #     define  R300_VAP_CNTL__PVS_NUM_CNTRLS__SHIFT            4
- #     define  R300_VAP_CNTL__PVS_NUM_FPUS__SHIFT              8
- #     define  R300_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT            18
- #     define  R500_VAP_CNTL__VAP_NO_RENDER                    (1<<17)
- #     define  R300_VAP_CNTL__DX_CLIP_SPACE_DEF                (1<<22)
- #     define  R500_VAP_CNTL__TCL_STATE_OPTIMIZATION           (1<<23)
+ #       define R300_PVS_NUM_SLOTS_SHIFT                 0
+ #       define R300_PVS_NUM_CNTLRS_SHIFT                4
+ #       define R300_PVS_NUM_FPUS_SHIFT                  8
+ #       define R300_VF_MAX_VTX_NUM_SHIFT                18
+ #       define R300_GL_CLIP_SPACE_DEF                   (0 << 22)
+ #       define R300_DX_CLIP_SPACE_DEF                   (1 << 22)
+ #       define R500_TCL_STATE_OPTIMIZATION              (1 << 23)
  
  /* This register is written directly and also starts data section
   * in many 3d CP_PACKET3's

diff --cc src/mesa/drivers/dri/r300/r300_state.c
index aa0de4e7838d0bd48642d13022c73b0abb43f4a1,6f1f4aac74a3c83749f58d5776651f1667cec998..cb39ce01106a4b8bf77448a52b62b468869170d7
--- 1/src/mesa/drivers/dri/r300/r300_state.c
--- 2/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@@ -1648,44 -1891,58 +1891,60 @@@ static inline void r300SetupVertexProgr
        }
  }
  
- /* FIXME: move near the MIN2 define. */
 -#define MIN3(a,b,c) ((a)<(b) ? MIN2(a, c): MIN2(b, c))
 +#define MIN3(a, b, c) ((a) < (b) ? MIN2(a, c) : MIN2(b, c))
  
- /* FIXME: need to add a structure for per-card/chipset values; they are
-  * currently hard-coded. */
 -static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count, GLuint output_count, GLuint temp_count)
++
 +static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count,
 +                      GLuint output_count, GLuint temp_count)
  {
-       int cmd_reserved = 0;
-       int cmd_written = 0;
-       drm_radeon_cmd_header_t *cmd = NULL;
- 
-       int vtx_mem_size = 72;  /* FIXME: R3XX vs R5XX */
- 
-       /* Flush PVS engine before changing PVS_NUM_SLOTS, PVS_NUM_CNTRLS.
-        * See r500 docs 6.5.2 */
-       reg_start(R300_VAP_PVS_WAITIDLE, 0);
-       e32(0x00000000);
- 
-       /* avoid division by zero */
-       if (input_count == 0)
-               input_count = 1;
-       if (output_count == 0)
-               output_count = 1;
-       if (temp_count == 0)
-               temp_count = 1;
- 
-       int pvs_num_slots =
-           MIN3(10, vtx_mem_size / input_count, vtx_mem_size / output_count);
-       int pvs_num_cntrls = MIN2(6, vtx_mem_size / temp_count);
- 
-       R300_STATECHANGE(rmesa, vap_cntl);
-       rmesa->hw.vap_cntl.cmd[1] =
-           (pvs_num_slots << R300_VAP_CNTL__PVS_NUM_SLOTS__SHIFT) |
-           (pvs_num_cntrls << R300_VAP_CNTL__PVS_NUM_CNTRLS__SHIFT) |
-           (4 << R300_VAP_CNTL__PVS_NUM_FPUS__SHIFT) |
-           (12 << R300_VAP_CNTL__VF_MAX_VTX_NUM__SHIFT) |
-           R500_VAP_CNTL__TCL_STATE_OPTIMIZATION;
+     int vtx_mem_size;
+     int pvs_num_slots;
+     int pvs_num_cntrls;
+ 
+     /* Flush PVS engine before changing PVS_NUM_SLOTS, PVS_NUM_CNTRLS.
+      * See r500 docs 6.5.2 - done in emit */
+ 
+     /* avoid division by zero */
+     if (input_count == 0) input_count = 1;
+     if (output_count == 0) output_count = 1;
+     if (temp_count == 0) temp_count = 1;
+ 
+     if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
+       vtx_mem_size = 128;
+     else
+       vtx_mem_size = 72;
+ 
+     pvs_num_slots = MIN3(10, vtx_mem_size/input_count, vtx_mem_size/output_count);
+     pvs_num_cntrls = MIN2(6, vtx_mem_size/temp_count);
+ 
+     R300_STATECHANGE(rmesa, vap_cntl);
+     if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
+       rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] = 
+           (pvs_num_slots << R300_PVS_NUM_SLOTS_SHIFT) |
+           (pvs_num_cntrls << R300_PVS_NUM_CNTLRS_SHIFT) |
+           (12 << R300_VF_MAX_VTX_NUM_SHIFT);
+       if (rmesa->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515)
+           rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] |= R500_TCL_STATE_OPTIMIZATION;
+     } else
+       /* not sure about non-tcl */
+       rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] = ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
+                                   (5 << R300_PVS_NUM_CNTLRS_SHIFT) |
+                                   (5 << R300_VF_MAX_VTX_NUM_SHIFT));
+ 
+     if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV515)
+       rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] |= (2 << R300_PVS_NUM_FPUS_SHIFT);
+     else if ((rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV530) ||
+            (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV560))
+       rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] |= (5 << R300_PVS_NUM_FPUS_SHIFT);
+     else if (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R420)
+       rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] |= (6 << R300_PVS_NUM_FPUS_SHIFT);
+     else if ((rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R520) ||
+            (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_R580) ||
+            (rmesa->radeon.radeonScreen->chip_family == CHIP_FAMILY_RV570))
+       rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] |= (8 << R300_PVS_NUM_FPUS_SHIFT);
+     else
+       rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] |= (4 << R300_PVS_NUM_FPUS_SHIFT);
+ 
  }
  
  static void r300SetupDefaultVertexProgram(r300ContextPtr rmesa)