Merge remote branch 'main/master' into radeon-rewrite
authorDave Airlie <airlied@redhat.com>
Fri, 20 Mar 2009 00:52:17 +0000 (10:52 +1000)
committerDave Airlie <airlied@redhat.com>
Fri, 20 Mar 2009 00:52:17 +0000 (10:52 +1000)
Conflicts:
src/mesa/drivers/dri/r300/r300_cmdbuf.c
src/mesa/drivers/dri/r300/r300_state.c
src/mesa/drivers/dri/r300/r300_swtcl.c
src/mesa/drivers/dri/r300/radeon_ioctl.c
src/mesa/drivers/dri/radeon/radeon_screen.c

15 files changed:
1  2 
configs/autoconf.in
configure.ac
src/mesa/drivers/dri/r200/r200_pixel.c
src/mesa/drivers/dri/r200/r200_texstate.c
src/mesa/drivers/dri/r200/r200_vertprog.c
src/mesa/drivers/dri/r300/r300_context.c
src/mesa/drivers/dri/r300/r300_emit.c
src/mesa/drivers/dri/r300/r300_reg.h
src/mesa/drivers/dri/r300/r300_state.c
src/mesa/drivers/dri/r300/r300_swtcl.c
src/mesa/drivers/dri/r300/radeon_program_pair.c
src/mesa/drivers/dri/radeon/radeon_common.c
src/mesa/drivers/dri/radeon/radeon_screen.c
src/mesa/drivers/dri/radeon/radeon_texstate.c
src/mesa/drivers/dri/radeon/server/radeon_reg.h

Simple merge
diff --cc configure.ac
@@@ -443,11 -453,9 +453,11 @@@ AC_SUBST([WINDOW_SYSTEM]
  AC_SUBST([GALLIUM_DIRS])
  AC_SUBST([GALLIUM_WINSYS_DIRS])
  AC_SUBST([GALLIUM_WINSYS_DRM_DIRS])
- AC_SUBST([GALLIUM_DRIVER_DIRS])
+ AC_SUBST([GALLIUM_DRIVERS_DIRS])
  AC_SUBST([GALLIUM_AUXILIARY_DIRS])
  AC_SUBST([GALLIUM_STATE_TRACKERS_DIRS])
 +AC_SUBST([RADEON_CFLAGS])
 +AC_SUBST([RADEON_LDFLAGS])
  
  dnl
  dnl User supplied program configuration
@@@ -1245,8 -1377,9 +1245,8 @@@ static GLboolean r200_validate_texgen( 
     } else {
        tgcm |= R200_TEXGEN_COMP_T << (unit * 4);
     }
 -
     if (texUnit->TexGenEnabled & R_BIT) {
-       if (texUnit->GenModeR != mode)
+       if (texUnit->GenR.Mode != mode)
         mixed_fallback = GL_TRUE;
     } else {
        tgcm |= R200_TEXGEN_COMP_R << (unit * 4);
Simple merge
Simple merge
@@@ -1654,31 -1589,27 +1525,27 @@@ static void r300SetupRSUnit(GLcontext 
                }
        }
  
-       if (InputsRead & FRAG_BIT_COL1) {
-               if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_COL1, _TNL_ATTRIB_COLOR1)) {
-                       r300->hw.rr.cmd[R300_RR_INST_1] |= R300_RS_INST_COL_ID(1) | R300_RS_INST_COL_CN_WRITE | (fp_reg++ << R300_RS_INST_COL_ADDR_SHIFT);
-                       InputsRead &= ~FRAG_BIT_COL1;
-                       if (high_rr < 1)
-                               high_rr = 1;
-                       col_interp_nr++;
-               } else {
-                       WARN_ONCE("fragprog wants col1, vp doesn't provide it\n");
-               }
+       if (InputsRead & FRAG_BIT_WPOS) {
+               r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |=  R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) |  R300_RS_TEX_PTR(rs_tex_count);
+               r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
+               InputsRead &= ~FRAG_BIT_WPOS;
+               rs_tex_count += 4;
+               ++tex_ip;
+               ++fp_reg;
        }
+       InputsRead &= ~FRAG_BIT_WPOS;
  
-       /* Need at least one. This might still lock as the values are undefined... */
-       if (rs_tex_count == 0 && col_interp_nr == 0) {
-               r300->hw.rr.cmd[R300_RR_INST_0] |= R300_RS_INST_COL_ID(0) | R300_RS_INST_COL_CN_WRITE | (fp_reg++ << R300_RS_INST_COL_ADDR_SHIFT);
-               col_interp_nr++;
+       /* Setup default color if no color or tex was set */
+       if (rs_tex_count == 0 && col_ip == 0) {
+               r300->hw.rr.cmd[R300_RR_INST_0] = R300_RS_INST_COL_ID(0) | R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(0) | R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
+               ++col_ip;
        }
  
-       r300->hw.rc.cmd[1] = 0 | (rs_tex_count << R300_IT_COUNT_SHIFT)
-         | (col_interp_nr << R300_IC_COUNT_SHIFT)
-         | R300_HIRES_EN;
+       high_rr = (col_ip > tex_ip) ? col_ip : tex_ip;
+       r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT)  | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN;
+       r300->hw.rc.cmd[2] |= high_rr - 1;
  
-       assert(high_rr >= 0);
-       r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_INST_0, high_rr + 1);
-       r300->hw.rc.cmd[2] = high_rr;
 -      r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R300_RS_INST_0, high_rr);
++       r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R300_RS_INST_0, high_rr);
  
        if (InputsRead)
                WARN_ONCE("Don't know how to satisfy InputsRead=0x%08x\n", InputsRead);
@@@ -1826,19 -1764,30 +1700,30 @@@ static void r500SetupRSUnit(GLcontext 
                }
        }
  
-       /* Need at least one. This might still lock as the values are undefined... */
-       if (in_texcoords == 0 && col_interp_nr == 0) {
-               r300->hw.rr.cmd[R300_RR_INST_0] |= 0 | R500_RS_INST_COL_CN_WRITE | (fp_reg++ << R500_RS_INST_COL_ADDR_SHIFT);
-               col_interp_nr++;
+       if (InputsRead & FRAG_BIT_WPOS) {
+               r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) |
+                               ((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) |
+                               ((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) |
+                               ((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT);
+               r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
+               InputsRead &= ~FRAG_BIT_WPOS;
+               rs_tex_count += 4;
+               ++tex_ip;
+               ++fp_reg;
        }
  
-       r300->hw.rc.cmd[1] = 0 | (in_texcoords << R300_IT_COUNT_SHIFT)
-         | (col_interp_nr << R300_IC_COUNT_SHIFT)
-         | R300_HIRES_EN;
+       /* Setup default color if no color or tex was set */
+       if (rs_tex_count == 0 && col_ip == 0) {
+               r300->hw.rr.cmd[R300_RR_INST_0] |= R500_RS_INST_COL_ID(0) | R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(0) | R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
+               ++col_ip;
+       }
+       high_rr = (col_ip > tex_ip) ? col_ip : tex_ip;
+       r300->hw.rc.cmd[1] |= (rs_tex_count << R300_IT_COUNT_SHIFT)  | (col_ip << R300_IC_COUNT_SHIFT) | R300_HIRES_EN;
+       r300->hw.rc.cmd[2] |= 0xC0 | (high_rr - 1);
  
-       assert(high_rr >= 0);
-       r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_INST_0, high_rr + 1);
-       r300->hw.rc.cmd[2] = 0xC0 | high_rr;
 -      r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(R500_RS_INST_0, high_rr);
++      r300->hw.rr.cmd[R300_RR_CMD_0] = cmdpacket0(r300->radeon.radeonScreen, R500_RS_INST_0, high_rr);
  
        if (InputsRead)
                WARN_ONCE("Don't know how to satisfy InputsRead=0x%08x\n", InputsRead);
@@@ -2579,11 -2534,25 +2446,24 @@@ void r300UpdateShaderStates(r300Context
        GLcontext *ctx;
        ctx = rmesa->radeon.glCtx;
  
 -      r300UpdateTextureState(ctx);
        r300SetEarlyZState(ctx);
  
-       GLuint fgdepthsrc = R300_FG_DEPTH_SRC_SCAN;
-       if (current_fragment_program_writes_depth(ctx))
+       /* w_fmt value is set to get best performance
+        * see p.130 R5xx 3D acceleration guide v1.3 */
+       GLuint w_fmt, fgdepthsrc;
+       if (current_fragment_program_writes_depth(ctx)) {
                fgdepthsrc = R300_FG_DEPTH_SRC_SHADER;
+               w_fmt = R300_W_FMT_W24 | R300_W_SRC_US;
+       } else {
+               fgdepthsrc = R300_FG_DEPTH_SRC_SCAN;
+               w_fmt = R300_W_FMT_W0 | R300_W_SRC_US;
+       }
+       if (w_fmt != rmesa->hw.us_out_fmt.cmd[5]) {
+               R300_STATECHANGE(rmesa, us_out_fmt);
+               rmesa->hw.us_out_fmt.cmd[5] = w_fmt;
+       }
        if (fgdepthsrc != rmesa->hw.fg_depth_src.cmd[1]) {
                R300_STATECHANGE(rmesa, fg_depth_src);
                rmesa->hw.fg_depth_src.cmd[1] = fgdepthsrc;
@@@ -203,27 -266,15 +263,28 @@@ static void r300SetVertexFormat( GLcont
        }
  
        R300_NEWPRIM(rmesa);
 -      R300_STATECHANGE(rmesa, vir[0]);
 -      ((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count =
 +      if (rmesa->radeon.radeonScreen->kernel_mm) {
 +              R300_STATECHANGE(rmesa, vir[0]);
 +              rmesa->hw.vir[0].cmd[0] &= 0xC000FFFF;
 +              rmesa->hw.vir[1].cmd[0] &= 0xC000FFFF;
 +              rmesa->hw.vir[0].cmd[0] |=
 +                      (r300VAPInputRoute0(&rmesa->hw.vir[0].cmd[R300_VIR_CNTL_0],
 +                                          VB->AttribPtr, inputs, tab, nr) & 0x3FFF) << 16;
 +              R300_STATECHANGE(rmesa, vir[1]);
 +              rmesa->hw.vir[1].cmd[0] |=
 +                      (r300VAPInputRoute1(&rmesa->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle,
 +                                          nr) & 0x3FFF) << 16;
 +      } else {
 +              R300_STATECHANGE(rmesa, vir[0]);
 +              ((drm_r300_cmd_header_t *) rmesa->hw.vir[0].cmd)->packet0.count =
                r300VAPInputRoute0(&rmesa->hw.vir[0].cmd[R300_VIR_CNTL_0],
                                   VB->AttribPtr, inputs, tab, nr);
 -      R300_STATECHANGE(rmesa, vir[1]);
 -      ((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count =
 +              R300_STATECHANGE(rmesa, vir[1]);
 +              ((drm_r300_cmd_header_t *) rmesa->hw.vir[1].cmd)->packet0.count =
                r300VAPInputRoute1(&rmesa->hw.vir[1].cmd[R300_VIR_CNTL_0], swizzle,
                                   nr);
 +      }
        R300_STATECHANGE(rmesa, vic);
        rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead);
        rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead);
@@@ -520,9 -638,11 +581,12 @@@ static void r300RenderStart(GLcontext *
        r300UpdateShaderStates(rmesa);
  
        r300EmitCacheFlush(rmesa);
 -      if (rmesa->dma.flush != 0 &&
 -          rmesa->dma.flush != flush_last_swtcl_prim)
 -              rmesa->dma.flush( rmesa );
++      /* investigate if we can put back flush optimisation if needed */
 +      if (rmesa->radeon.dma.flush != NULL) {
 +              rmesa->radeon.dma.flush(ctx);
 +      }
  }
  
  static void r300RenderFinish(GLcontext *ctx)
@@@ -628,29 -751,6 +692,28 @@@ void r300EmitVbufPrim(r300ContextPtr rm
        type = r300PrimitiveType(rmesa, primitive);
        num_verts = r300NumVerts(rmesa, vertex_nr, primitive);
  
 -      start_packet3(CP_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0), 0);
 -      e32(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type);
 +      BEGIN_BATCH(3);
 +      OUT_BATCH_PACKET3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 +      OUT_BATCH(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (num_verts << 16) | type);
 +      END_BATCH();
 +}
 +
 +void r300_swtcl_flush(GLcontext *ctx, uint32_t current_offset)
 +{
 +  r300ContextPtr rmesa = R300_CONTEXT(ctx);
 +
 +  rcommonEnsureCmdBufSpace(&rmesa->radeon,
 +                         rmesa->radeon.hw.max_state_size + (12*sizeof(int)),
 +                         __FUNCTION__);
 +  radeonEmitState(&rmesa->radeon);
 +  r300EmitVertexAOS(rmesa,
 +                  rmesa->radeon.swtcl.vertex_size,
 +                  rmesa->radeon.dma.current,
 +                  current_offset);
 +  
 +  r300EmitVbufPrim(rmesa,
 +                 rmesa->radeon.swtcl.hw_primitive,
 +                 rmesa->radeon.swtcl.numverts);
 +  r300EmitCacheFlush(rmesa);
 +  COMMIT_BATCH();
  }
index 5c34ca8,0000000..840233f
mode 100644,000000..100644
--- /dev/null
@@@ -1,1107 -1,0 +1,1107 @@@
-       uint32_t frame;
 +/**************************************************************************
 +
 +Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
 +
 +The Weather Channel (TM) funded Tungsten Graphics to develop the
 +initial release of the Radeon 8500 driver under the XFree86 license.
 +This notice must be preserved.
 +
 +Permission is hereby granted, free of charge, to any person obtaining
 +a copy of this software and associated documentation files (the
 +"Software"), to deal in the Software without restriction, including
 +without limitation the rights to use, copy, modify, merge, publish,
 +distribute, sublicense, and/or sell copies of the Software, and to
 +permit persons to whom the Software is furnished to do so, subject to
 +the following conditions:
 +
 +The above copyright notice and this permission notice (including the
 +next paragraph) shall be included in all copies or substantial
 +portions of the Software.
 +
 +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
 +IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
 +LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
 +OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
 +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 +
 +**************************************************************************/
 +
 +/*
 + * Authors:
 + *   Keith Whitwell <keith@tungstengraphics.com>
 + */
 +
 +/*
 +   - Scissor implementation
 +   - buffer swap/copy ioctls
 +   - finish/flush
 +   - state emission
 +   - cmdbuffer management
 +*/
 +
 +#include <errno.h>
 +#include "main/glheader.h"
 +#include "main/imports.h"
 +#include "main/context.h"
 +#include "main/api_arrayelt.h"
 +#include "main/enums.h"
 +#include "main/colormac.h"
 +#include "main/light.h"
 +#include "main/framebuffer.h"
 +#include "main/simple_list.h"
 +#include "main/renderbuffer.h"
 +#include "swrast/swrast.h"
 +#include "vbo/vbo.h"
 +#include "tnl/tnl.h"
 +#include "tnl/t_pipeline.h"
 +#include "swrast_setup/swrast_setup.h"
 +
 +#include "dri_util.h"
 +#include "vblank.h"
 +
 +#include "radeon_common.h"
 +#include "radeon_bocs_wrapper.h"
 +#include "radeon_lock.h"
 +#include "radeon_drm.h"
 +#include "radeon_mipmap_tree.h"
 +
 +#define DEBUG_CMDBUF         0
 +
 +/* =============================================================
 + * Scissoring
 + */
 +
 +static GLboolean intersect_rect(drm_clip_rect_t * out,
 +                              drm_clip_rect_t * a, drm_clip_rect_t * b)
 +{
 +      *out = *a;
 +      if (b->x1 > out->x1)
 +              out->x1 = b->x1;
 +      if (b->y1 > out->y1)
 +              out->y1 = b->y1;
 +      if (b->x2 < out->x2)
 +              out->x2 = b->x2;
 +      if (b->y2 < out->y2)
 +              out->y2 = b->y2;
 +      if (out->x1 >= out->x2)
 +              return GL_FALSE;
 +      if (out->y1 >= out->y2)
 +              return GL_FALSE;
 +      return GL_TRUE;
 +}
 +
 +void radeonRecalcScissorRects(radeonContextPtr radeon)
 +{
 +      drm_clip_rect_t *out;
 +      int i;
 +
 +      /* Grow cliprect store?
 +       */
 +      if (radeon->state.scissor.numAllocedClipRects < radeon->numClipRects) {
 +              while (radeon->state.scissor.numAllocedClipRects <
 +                     radeon->numClipRects) {
 +                      radeon->state.scissor.numAllocedClipRects += 1; /* zero case */
 +                      radeon->state.scissor.numAllocedClipRects *= 2;
 +              }
 +
 +              if (radeon->state.scissor.pClipRects)
 +                      FREE(radeon->state.scissor.pClipRects);
 +
 +              radeon->state.scissor.pClipRects =
 +                      MALLOC(radeon->state.scissor.numAllocedClipRects *
 +                             sizeof(drm_clip_rect_t));
 +
 +              if (radeon->state.scissor.pClipRects == NULL) {
 +                      radeon->state.scissor.numAllocedClipRects = 0;
 +                      return;
 +              }
 +      }
 +
 +      out = radeon->state.scissor.pClipRects;
 +      radeon->state.scissor.numClipRects = 0;
 +
 +      for (i = 0; i < radeon->numClipRects; i++) {
 +              if (intersect_rect(out,
 +                                 &radeon->pClipRects[i],
 +                                 &radeon->state.scissor.rect)) {
 +                      radeon->state.scissor.numClipRects++;
 +                      out++;
 +              }
 +      }
 +}
 +
 +static void radeon_get_cliprects(radeonContextPtr radeon,
 +                               struct drm_clip_rect **cliprects,
 +                               unsigned int *num_cliprects,
 +                               int *x_off, int *y_off)
 +{
 +      __DRIdrawablePrivate *dPriv = radeon->dri.drawable;
 +      struct radeon_framebuffer *rfb = dPriv->driverPrivate;
 +
 +      if (radeon->constant_cliprect) {
 +              radeon->fboRect.x1 = 0;
 +              radeon->fboRect.y1 = 0;
 +              radeon->fboRect.x2 = radeon->glCtx->DrawBuffer->Width;
 +              radeon->fboRect.y2 = radeon->glCtx->DrawBuffer->Height;
 +
 +              *cliprects = &radeon->fboRect;
 +              *num_cliprects = 1;
 +              *x_off = 0;
 +              *y_off = 0;
 +      } else if (radeon->front_cliprects ||
 +                 rfb->pf_active || dPriv->numBackClipRects == 0) {
 +              *cliprects = dPriv->pClipRects;
 +              *num_cliprects = dPriv->numClipRects;
 +              *x_off = dPriv->x;
 +              *y_off = dPriv->y;
 +      } else {
 +              *num_cliprects = dPriv->numBackClipRects;
 +              *cliprects = dPriv->pBackClipRects;
 +              *x_off = dPriv->backX;
 +              *y_off = dPriv->backY;
 +      }
 +}
 +
 +/**
 + * Update cliprects and scissors.
 + */
 +void radeonSetCliprects(radeonContextPtr radeon)
 +{
 +      __DRIdrawablePrivate *const drawable = radeon->dri.drawable;
 +      __DRIdrawablePrivate *const readable = radeon->dri.readable;
 +      struct radeon_framebuffer *const draw_rfb = drawable->driverPrivate;
 +      struct radeon_framebuffer *const read_rfb = readable->driverPrivate;
 +      int x_off, y_off;
 +
 +      radeon_get_cliprects(radeon, &radeon->pClipRects,
 +                           &radeon->numClipRects, &x_off, &y_off);
 +      
 +      if ((draw_rfb->base.Width != drawable->w) ||
 +          (draw_rfb->base.Height != drawable->h)) {
 +              _mesa_resize_framebuffer(radeon->glCtx, &draw_rfb->base,
 +                                       drawable->w, drawable->h);
 +              draw_rfb->base.Initialized = GL_TRUE;
 +      }
 +
 +      if (drawable != readable) {
 +              if ((read_rfb->base.Width != readable->w) ||
 +                  (read_rfb->base.Height != readable->h)) {
 +                      _mesa_resize_framebuffer(radeon->glCtx, &read_rfb->base,
 +                                               readable->w, readable->h);
 +                      read_rfb->base.Initialized = GL_TRUE;
 +              }
 +      }
 +
 +      if (radeon->state.scissor.enabled)
 +              radeonRecalcScissorRects(radeon);
 +
 +}
 +
 +
 +
 +void radeonUpdateScissor( GLcontext *ctx )
 +{
 +      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 +
 +      if ( rmesa->dri.drawable ) {
 +              __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
 +      
 +              int x = ctx->Scissor.X;
 +              int y = dPriv->h - ctx->Scissor.Y - ctx->Scissor.Height;
 +              int w = ctx->Scissor.X + ctx->Scissor.Width - 1;
 +              int h = dPriv->h - ctx->Scissor.Y - 1;
 +
 +              rmesa->state.scissor.rect.x1 = x + dPriv->x;
 +              rmesa->state.scissor.rect.y1 = y + dPriv->y;
 +              rmesa->state.scissor.rect.x2 = w + dPriv->x + 1;
 +              rmesa->state.scissor.rect.y2 = h + dPriv->y + 1;
 +
 +              radeonRecalcScissorRects( rmesa );
 +      }
 +}
 +
 +/* =============================================================
 + * Scissoring
 + */
 +
 +void radeonScissor(GLcontext* ctx, GLint x, GLint y, GLsizei w, GLsizei h)
 +{
 +      radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 +      if (ctx->Scissor.Enabled) {
 +              /* We don't pipeline cliprect changes */
 +              radeon_firevertices(radeon);
 +              radeonUpdateScissor(ctx);
 +      }
 +}
 +
 +
 +/* ================================================================
 + * SwapBuffers with client-side throttling
 + */
 +
 +static uint32_t radeonGetLastFrame(radeonContextPtr radeon)
 +{
 +      drm_radeon_getparam_t gp;
 +      int ret;
++      uint32_t frame = 0;
 +
 +      gp.param = RADEON_PARAM_LAST_FRAME;
 +      gp.value = (int *)&frame;
 +      ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM,
 +                                &gp, sizeof(gp));
 +      if (ret) {
 +              fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__,
 +                      ret);
 +              exit(1);
 +      }
 +
 +      return frame;
 +}
 +
 +uint32_t radeonGetAge(radeonContextPtr radeon)
 +{
 +      drm_radeon_getparam_t gp;
 +      int ret;
 +      uint32_t age;
 +
 +      gp.param = RADEON_PARAM_LAST_CLEAR;
 +      gp.value = (int *)&age;
 +      ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_GETPARAM,
 +                                &gp, sizeof(gp));
 +      if (ret) {
 +              fprintf(stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__,
 +                      ret);
 +              exit(1);
 +      }
 +
 +      return age;
 +}
 +
 +static void radeonEmitIrqLocked(radeonContextPtr radeon)
 +{
 +      drm_radeon_irq_emit_t ie;
 +      int ret;
 +
 +      ie.irq_seq = &radeon->iw.irq_seq;
 +      ret = drmCommandWriteRead(radeon->dri.fd, DRM_RADEON_IRQ_EMIT,
 +                                &ie, sizeof(ie));
 +      if (ret) {
 +              fprintf(stderr, "%s: drmRadeonIrqEmit: %d\n", __FUNCTION__,
 +                      ret);
 +              exit(1);
 +      }
 +}
 +
 +static void radeonWaitIrq(radeonContextPtr radeon)
 +{
 +      int ret;
 +
 +      do {
 +              ret = drmCommandWrite(radeon->dri.fd, DRM_RADEON_IRQ_WAIT,
 +                                    &radeon->iw, sizeof(radeon->iw));
 +      } while (ret && (errno == EINTR || errno == EBUSY));
 +
 +      if (ret) {
 +              fprintf(stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__,
 +                      ret);
 +              exit(1);
 +      }
 +}
 +
 +static void radeonWaitForFrameCompletion(radeonContextPtr radeon)
 +{
 +      drm_radeon_sarea_t *sarea = radeon->sarea;
 +
 +      if (radeon->do_irqs) {
 +              if (radeonGetLastFrame(radeon) < sarea->last_frame) {
 +                      if (!radeon->irqsEmitted) {
 +                              while (radeonGetLastFrame(radeon) <
 +                                     sarea->last_frame) ;
 +                      } else {
 +                              UNLOCK_HARDWARE(radeon);
 +                              radeonWaitIrq(radeon);
 +                              LOCK_HARDWARE(radeon);
 +                      }
 +                      radeon->irqsEmitted = 10;
 +              }
 +
 +              if (radeon->irqsEmitted) {
 +                      radeonEmitIrqLocked(radeon);
 +                      radeon->irqsEmitted--;
 +              }
 +      } else {
 +              while (radeonGetLastFrame(radeon) < sarea->last_frame) {
 +                      UNLOCK_HARDWARE(radeon);
 +                      if (radeon->do_usleeps)
 +                              DO_USLEEP(1);
 +                      LOCK_HARDWARE(radeon);
 +              }
 +      }
 +}
 +
 +/* wait for idle */
 +void radeonWaitForIdleLocked(radeonContextPtr radeon)
 +{
 +      int ret;
 +      int i = 0;
 +
 +      do {
 +              ret = drmCommandNone(radeon->dri.fd, DRM_RADEON_CP_IDLE);
 +              if (ret)
 +                      DO_USLEEP(1);
 +      } while (ret && ++i < 100);
 +
 +      if (ret < 0) {
 +              UNLOCK_HARDWARE(radeon);
 +              fprintf(stderr, "Error: R300 timed out... exiting\n");
 +              exit(-1);
 +      }
 +}
 +
 +static void radeonWaitForIdle(radeonContextPtr radeon)
 +{
 +      LOCK_HARDWARE(radeon);
 +      radeonWaitForIdleLocked(radeon);
 +      UNLOCK_HARDWARE(radeon);
 +}
 +
 +static void radeon_flip_renderbuffers(struct radeon_framebuffer *rfb)
 +{
 +      int current_page = rfb->pf_current_page;
 +      int next_page = (current_page + 1) % rfb->pf_num_pages;
 +      struct gl_renderbuffer *tmp_rb;
 +
 +      /* Exchange renderbuffers if necessary but make sure their
 +       * reference counts are preserved.
 +       */
 +      if (rfb->color_rb[current_page] &&
 +          rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer !=
 +          &rfb->color_rb[current_page]->base) {
 +              tmp_rb = NULL;
 +              _mesa_reference_renderbuffer(&tmp_rb,
 +                                           rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer);
 +              tmp_rb = &rfb->color_rb[current_page]->base;
 +              _mesa_reference_renderbuffer(&rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer, tmp_rb);
 +              _mesa_reference_renderbuffer(&tmp_rb, NULL);
 +      }
 +
 +      if (rfb->color_rb[next_page] &&
 +          rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer !=
 +          &rfb->color_rb[next_page]->base) {
 +              tmp_rb = NULL;
 +              _mesa_reference_renderbuffer(&tmp_rb,
 +                                           rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer);
 +              tmp_rb = &rfb->color_rb[next_page]->base;
 +              _mesa_reference_renderbuffer(&rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer, tmp_rb);
 +              _mesa_reference_renderbuffer(&tmp_rb, NULL);
 +      }
 +}
 +
 +/* Copy the back color buffer to the front color buffer.
 + */
 +void radeonCopyBuffer( __DRIdrawablePrivate *dPriv,
 +                     const drm_clip_rect_t      *rect)
 +{
 +      radeonContextPtr rmesa;
 +      struct radeon_framebuffer *rfb;
 +      GLint nbox, i, ret;
 +   
 +      assert(dPriv);
 +      assert(dPriv->driContextPriv);
 +      assert(dPriv->driContextPriv->driverPrivate);
 +   
 +      rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
 +
 +      LOCK_HARDWARE(rmesa);
 +
 +      rfb = dPriv->driverPrivate;
 +
 +      if ( RADEON_DEBUG & DEBUG_IOCTL ) {
 +              fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, (void *) rmesa->glCtx );
 +      }
 +
 +      nbox = dPriv->numClipRects; /* must be in locked region */
 +
 +      for ( i = 0 ; i < nbox ; ) {
 +              GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS , nbox );
 +              drm_clip_rect_t *box = dPriv->pClipRects;
 +              drm_clip_rect_t *b = rmesa->sarea->boxes;
 +              GLint n = 0;
 +
 +              for ( ; i < nr ; i++ ) {
 +
 +                      *b = box[i];
 +
 +                      if (rect)
 +                      {
 +                              if (rect->x1 > b->x1)
 +                                      b->x1 = rect->x1;
 +                              if (rect->y1 > b->y1)
 +                                      b->y1 = rect->y1;
 +                              if (rect->x2 < b->x2)
 +                                      b->x2 = rect->x2;
 +                              if (rect->y2 < b->y2)
 +                                      b->y2 = rect->y2;
 +
 +                              if (b->x1 >= b->x2 || b->y1 >= b->y2)
 +                                      continue;
 +                      }
 +
 +                      b++;
 +                      n++;
 +              }
 +              rmesa->sarea->nbox = n;
 +
 +              if (!n)
 +                      continue;
 +
 +              ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP );
 +
 +              if ( ret ) {
 +                      fprintf( stderr, "DRM_RADEON_SWAP_BUFFERS: return = %d\n", ret );
 +                      UNLOCK_HARDWARE( rmesa );
 +                      exit( 1 );
 +              }
 +      }
 +
 +      UNLOCK_HARDWARE( rmesa );
 +}
 +
 +static int radeonScheduleSwap(__DRIdrawablePrivate *dPriv, GLboolean *missed_target)
 +{
 +      radeonContextPtr rmesa;
 +
 +      rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
 +      radeon_firevertices(rmesa);
 +
 +      LOCK_HARDWARE( rmesa );
 +
 +      if (!dPriv->numClipRects) {
 +              UNLOCK_HARDWARE(rmesa);
 +              usleep(10000);  /* throttle invisible client 10ms */
 +              return 0;
 +      }
 +
 +      radeonWaitForFrameCompletion(rmesa);
 +
 +      UNLOCK_HARDWARE(rmesa);
 +      driWaitForVBlank(dPriv, missed_target);
 +      
 +      return 0;
 +}
 +
 +static GLboolean radeonPageFlip( __DRIdrawablePrivate *dPriv )
 +{
 +      radeonContextPtr radeon;
 +      GLint ret;
 +      __DRIscreenPrivate *psp;
 +      struct radeon_renderbuffer *rrb;
 +      struct radeon_framebuffer *rfb;
 +
 +      assert(dPriv);
 +      assert(dPriv->driContextPriv);
 +      assert(dPriv->driContextPriv->driverPrivate);
 +
 +      radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
 +      rfb = dPriv->driverPrivate;
 +      rrb = (void *)rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer;
 +
 +      psp = dPriv->driScreenPriv;
 +
 +      LOCK_HARDWARE(radeon);
 +
 +      if ( RADEON_DEBUG & DEBUG_IOCTL ) {
 +              fprintf(stderr, "%s: pfCurrentPage: %d %d\n", __FUNCTION__,
 +                      radeon->sarea->pfCurrentPage, radeon->sarea->pfState);
 +      }
 +      drm_clip_rect_t *box = dPriv->pClipRects;
 +      drm_clip_rect_t *b = radeon->sarea->boxes;
 +      b[0] = box[0];
 +      radeon->sarea->nbox = 1;
 +
 +      ret = drmCommandNone( radeon->dri.fd, DRM_RADEON_FLIP );
 +      
 +      UNLOCK_HARDWARE(radeon);
 +
 +      if ( ret ) {
 +              fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret );
 +              return GL_FALSE;
 +      }
 +
 +      if (!rfb->pf_active)
 +              return GL_FALSE;
 +
 +      rfb->pf_current_page = radeon->sarea->pfCurrentPage;
 +      radeon_flip_renderbuffers(rfb);
 +      radeon_draw_buffer(radeon->glCtx, &rfb->base);
 +
 +      return GL_TRUE;
 +}
 +
 +
 +/**
 + * Swap front and back buffer.
 + */
 +void radeonSwapBuffers(__DRIdrawablePrivate * dPriv)
 +{
 +      int64_t ust;
 +      __DRIscreenPrivate *psp;
 +
 +      if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
 +              radeonContextPtr radeon;
 +              GLcontext *ctx;
 +
 +              radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
 +              ctx = radeon->glCtx;
 +
 +              if (ctx->Visual.doubleBufferMode) {
 +                      GLboolean missed_target;
 +                      struct radeon_framebuffer *rfb = dPriv->driverPrivate;
 +                      _mesa_notifySwapBuffers(ctx);/* flush pending rendering comands */
 +
 +                      radeonScheduleSwap(dPriv, &missed_target);
 +
 +                      if (rfb->pf_active) {
 +                              radeonPageFlip(dPriv);
 +                      } else {
 +                              radeonCopyBuffer(dPriv, NULL);
 +                      }
 +
 +                      psp = dPriv->driScreenPriv;
 +
 +                      rfb->swap_count++;
 +                      (*psp->systemTime->getUST)( & ust );
 +                      if ( missed_target ) {
 +                              rfb->swap_missed_count++;
 +                              rfb->swap_missed_ust = ust - rfb->swap_ust;
 +                      }
 +
 +                      rfb->swap_ust = ust;
 +                      radeon->hw.all_dirty = GL_TRUE;
 +              }
 +      } else {
 +              /* XXX this shouldn't be an error but we can't handle it for now */
 +              _mesa_problem(NULL, "%s: drawable has no context!",
 +                            __FUNCTION__);
 +      }
 +}
 +
 +void radeonCopySubBuffer(__DRIdrawablePrivate * dPriv,
 +                       int x, int y, int w, int h )
 +{
 +      if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
 +              radeonContextPtr radeon;
 +              GLcontext *ctx;
 +
 +              radeon = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
 +              ctx = radeon->glCtx;
 +
 +              if (ctx->Visual.doubleBufferMode) {
 +                      drm_clip_rect_t rect;
 +                      rect.x1 = x + dPriv->x;
 +                      rect.y1 = (dPriv->h - y - h) + dPriv->y;
 +                      rect.x2 = rect.x1 + w;
 +                      rect.y2 = rect.y1 + h;
 +                      _mesa_notifySwapBuffers(ctx);   /* flush pending rendering comands */
 +                      radeonCopyBuffer(dPriv, &rect);
 +              }
 +      } else {
 +              /* XXX this shouldn't be an error but we can't handle it for now */
 +              _mesa_problem(NULL, "%s: drawable has no context!",
 +                            __FUNCTION__);
 +      }
 +}
 +
 +void radeon_draw_buffer(GLcontext *ctx, struct gl_framebuffer *fb)
 +{
 +      radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 +      struct radeon_renderbuffer *rrbDepth = NULL, *rrbStencil = NULL,
 +              *rrbColor = NULL;
 +       
 +
 +      if (!fb) {
 +              /* this can happen during the initial context initialization */
 +              return;
 +      }
 +
 +      /* radeons only handle 1 color draw so far */
 +      if (fb->_NumColorDrawBuffers != 1) {
 +              radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE);
 +              return;
 +      }
 +              
 +      /* Do this here, note core Mesa, since this function is called from
 +       * many places within the driver.
 +       */
 +      if (ctx->NewState & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) {
 +              /* this updates the DrawBuffer->_NumColorDrawBuffers fields, etc */
 +              _mesa_update_framebuffer(ctx);
 +              /* this updates the DrawBuffer's Width/Height if it's a FBO */
 +              _mesa_update_draw_buffer_bounds(ctx);
 +      }
 +
 +      if (fb->_Status != GL_FRAMEBUFFER_COMPLETE_EXT) {
 +              /* this may occur when we're called by glBindFrameBuffer() during
 +               * the process of someone setting up renderbuffers, etc.
 +               */
 +              /*_mesa_debug(ctx, "DrawBuffer: incomplete user FBO\n");*/
 +              return;
 +      }
 +
 +      if (fb->Name)
 +              ;/* do something depthy/stencily TODO */
 +
 +
 +              /* none */
 +      if (fb->Name == 0) {
 +              if (fb->_ColorDrawBufferIndexes[0] == BUFFER_FRONT_LEFT) {
 +                      rrbColor = (void *)fb->Attachment[BUFFER_FRONT_LEFT].Renderbuffer;
 +                      radeon->front_cliprects = GL_TRUE;
 +              } else {
 +                      rrbColor = (void *)fb->Attachment[BUFFER_BACK_LEFT].Renderbuffer;
 +                      radeon->front_cliprects = GL_FALSE;
 +              }
 +      } else {
 +              /* user FBO in theory */
 +              struct radeon_renderbuffer *rrb;
 +              rrb = (void *)fb->_ColorDrawBuffers[0];
 +              rrbColor = rrb;
 +              radeon->constant_cliprect = GL_TRUE;
 +      }
 +
 +      if (rrbColor == NULL)
 +              radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE);
 +      else
 +              radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DRAW_BUFFER, GL_FALSE);
 +
 +
 +
 +      if (fb->_DepthBuffer && fb->_DepthBuffer->Wrapped) {
 +              rrbDepth = (struct radeon_renderbuffer *)fb->_DepthBuffer->Wrapped;
 +              if (rrbDepth && rrbDepth->bo) {
 +                      radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_FALSE);
 +              } else {
 +                      radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_TRUE);
 +              }
 +      } else {
 +              radeon->vtbl.fallback(ctx, RADEON_FALLBACK_DEPTH_BUFFER, GL_FALSE);
 +              rrbDepth = NULL;
 +      }
 +
 +      /* TODO stencil things */
 +      if (fb->_StencilBuffer && fb->_StencilBuffer->Wrapped) {
 +              rrbStencil = (struct radeon_renderbuffer *)fb->_DepthBuffer->Wrapped;
 +              if (rrbStencil && rrbStencil->bo) {
 +                      radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_FALSE);
 +                      /* need to re-compute stencil hw state */
 +                      if (ctx->Driver.Enable != NULL)
 +                              ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
 +                      else
 +                              ctx->NewState |= _NEW_STENCIL;
 +                      if (!rrbDepth)
 +                              rrbDepth = rrbStencil;
 +              } else {
 +                      radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_TRUE);
 +              }
 +      } else {
 +              radeon->vtbl.fallback(ctx, RADEON_FALLBACK_STENCIL_BUFFER, GL_FALSE);
 +              if (ctx->Driver.Enable != NULL)
 +                      ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
 +              else
 +                      ctx->NewState |= _NEW_STENCIL;
 +      }
 +
 +      /* Update culling direction which changes depending on the
 +       * orientation of the buffer:
 +       */
 +      if (ctx->Driver.FrontFace)
 +              ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace);
 +      else
 +              ctx->NewState |= _NEW_POLYGON;
 +      
 +      /*
 +       * Update depth test state
 +       */
 +      if (ctx->Driver.Enable) {
 +              if (ctx->Depth.Test && fb->Visual.depthBits > 0) {
 +                      ctx->Driver.Enable(ctx, GL_DEPTH_TEST, GL_TRUE);
 +              } else {
 +                      ctx->Driver.Enable(ctx, GL_DEPTH_TEST, GL_FALSE);
 +              }
 +      } else {
 +              ctx->NewState |= _NEW_DEPTH;
 +      }
 +      
 +      radeon->state.depth.rrb = rrbDepth;
 +
 +      radeon->state.color.rrb = rrbColor;
 +
 +      /* update viewport since it depends on window size */
 +      if (ctx->Driver.Viewport) {
 +              ctx->Driver.Viewport(ctx, ctx->Viewport.X, ctx->Viewport.Y,
 +                                   ctx->Viewport.Width, ctx->Viewport.Height);
 +      } else {
 +              ctx->NewState |= _NEW_VIEWPORT;
 +      }
 +
 +      /* Set state we know depends on drawable parameters:
 +       */
 +      if (ctx->Driver.Scissor)
 +              ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y,
 +                                  ctx->Scissor.Width, ctx->Scissor.Height);
 +      radeon->NewGLState |= _NEW_SCISSOR;
 +}
 +
 +/**
 + * Called via glDrawBuffer.
 + */
 +void radeonDrawBuffer( GLcontext *ctx, GLenum mode )
 +{
 +      radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 +      
 +      if (RADEON_DEBUG & DEBUG_DRI)
 +              fprintf(stderr, "%s %s\n", __FUNCTION__,
 +                      _mesa_lookup_enum_by_nr( mode ));
 +      
 +      radeon_firevertices(radeon);    /* don't pipeline cliprect changes */
 +      
 +      radeon_draw_buffer(ctx, ctx->DrawBuffer);
 +}
 +
 +void radeonReadBuffer( GLcontext *ctx, GLenum mode )
 +{
 +      /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
 +      if (ctx->ReadBuffer == ctx->DrawBuffer) {
 +              /* This will update FBO completeness status.
 +               * A framebuffer will be incomplete if the GL_READ_BUFFER setting
 +               * refers to a missing renderbuffer.  Calling glReadBuffer can set
 +               * that straight and can make the drawing buffer complete.
 +               */
 +              radeon_draw_buffer(ctx, ctx->DrawBuffer);
 +      }
 +}
 +
 +
 +/* Turn on/off page flipping according to the flags in the sarea:
 + */
 +void radeonUpdatePageFlipping(radeonContextPtr radeon)
 +{
 +      struct radeon_framebuffer *rfb = radeon->dri.drawable->driverPrivate;
 +
 +      rfb->pf_active = radeon->sarea->pfState;
 +      rfb->pf_current_page = radeon->sarea->pfCurrentPage;
 +      rfb->pf_num_pages = 2;
 +      radeon_flip_renderbuffers(rfb);
 +      radeon_draw_buffer(radeon->glCtx, radeon->glCtx->DrawBuffer);
 +}
 +
 +void radeon_window_moved(radeonContextPtr radeon)
 +{
 +      GLcontext *ctx = radeon->glCtx;
 +      __DRIdrawablePrivate *dPriv = radeon->dri.drawable;
 +      struct radeon_framebuffer *rfb = dPriv->driverPrivate;
 +
 +      if (!radeon->radeonScreen->driScreen->dri2.enabled) {
 +              radeonUpdatePageFlipping(radeon);
 +      }
 +      radeonSetCliprects(radeon);
 +}
 +
 +void radeon_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei width, GLsizei height)
 +{
 +      radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 +      __DRIcontext *driContext = radeon->dri.context;
 +      void (*old_viewport)(GLcontext *ctx, GLint x, GLint y,
 +                           GLsizei w, GLsizei h);
 +
 +      if (!driContext->driScreenPriv->dri2.enabled)
 +              return;
 +
 +      radeon_update_renderbuffers(driContext, driContext->driDrawablePriv);
 +      if (driContext->driDrawablePriv != driContext->driReadablePriv)
 +              radeon_update_renderbuffers(driContext, driContext->driReadablePriv);
 +
 +      old_viewport = ctx->Driver.Viewport;
 +      ctx->Driver.Viewport = NULL;
 +      radeon->dri.drawable = driContext->driDrawablePriv;
 +      radeon_window_moved(radeon);
 +      radeon_draw_buffer(ctx, radeon->glCtx->DrawBuffer);
 +      ctx->Driver.Viewport = old_viewport;
 +
 +
 +}
 +static void radeon_print_state_atom(radeonContextPtr radeon, struct radeon_state_atom *state )
 +{
 +      int i;
 +      int dwords = (*state->check)(radeon->glCtx, state);
 +
 +      fprintf(stderr, "emit %s %d/%d\n", state->name, state->cmd_size, dwords);
 +
 +      if (RADEON_DEBUG & DEBUG_VERBOSE) 
 +              for (i = 0 ; i < dwords; i++) 
 +                      fprintf(stderr, "\t%s[%d]: %x\n", state->name, i, state->cmd[i]);
 +
 +}
 +
 +static INLINE void radeonEmitAtoms(radeonContextPtr radeon, GLboolean dirty)
 +{
 +      BATCH_LOCALS(radeon);
 +      struct radeon_state_atom *atom;
 +      int dwords;
 +
 +      if (radeon->vtbl.pre_emit_atoms)
 +              radeon->vtbl.pre_emit_atoms(radeon);
 +
 +      /* Emit actual atoms */
 +      foreach(atom, &radeon->hw.atomlist) {
 +              if ((atom->dirty || radeon->hw.all_dirty) == dirty) {
 +                      dwords = (*atom->check) (radeon->glCtx, atom);
 +                      if (dwords) {
 +                              if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) {
 +                                      radeon_print_state_atom(radeon, atom);
 +                              }
 +                              if (atom->emit) {
 +                                      (*atom->emit)(radeon->glCtx, atom);
 +                              } else {
 +                                      BEGIN_BATCH_NO_AUTOSTATE(dwords);
 +                                      OUT_BATCH_TABLE(atom->cmd, dwords);
 +                                      END_BATCH();
 +                              }
 +                              atom->dirty = GL_FALSE;
 +                      } else {
 +                              if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_STATE) {
 +                                      fprintf(stderr, "  skip state %s\n",
 +                                              atom->name);
 +                              }
 +                      }
 +              }
 +      }
 +   
 +      COMMIT_BATCH();
 +}
 +
 +void radeonEmitState(radeonContextPtr radeon)
 +{
 +      if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
 +              fprintf(stderr, "%s\n", __FUNCTION__);
 +
 +      if (radeon->vtbl.pre_emit_state)
 +              radeon->vtbl.pre_emit_state(radeon);
 +
 +      /* this code used to return here but now it emits zbs */
 +      if (radeon->cmdbuf.cs->cdw && !radeon->hw.is_dirty && !radeon->hw.all_dirty)
 +              return;
 +
 +      /* To avoid going across the entire set of states multiple times, just check
 +       * for enough space for the case of emitting all state, and inline the
 +       * radeonAllocCmdBuf code here without all the checks.
 +       */
 +      rcommonEnsureCmdBufSpace(radeon, radeon->hw.max_state_size, __FUNCTION__);
 +
 +      if (!radeon->cmdbuf.cs->cdw) {
 +              if (RADEON_DEBUG & DEBUG_STATE)
 +                      fprintf(stderr, "Begin reemit state\n");
 +              
 +              radeonEmitAtoms(radeon, GL_FALSE);
 +      }
 +
 +      if (RADEON_DEBUG & DEBUG_STATE)
 +              fprintf(stderr, "Begin dirty state\n");
 +
 +      radeonEmitAtoms(radeon, GL_TRUE);
 +      radeon->hw.is_dirty = GL_FALSE;
 +      radeon->hw.all_dirty = GL_FALSE;
 +
 +}
 +
 +
 +void radeonFlush(GLcontext *ctx)
 +{
 +      radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 +      if (RADEON_DEBUG & DEBUG_IOCTL)
 +              fprintf(stderr, "%s %d\n", __FUNCTION__, radeon->cmdbuf.cs->cdw);
 +
 +      if (radeon->dma.flush)
 +              radeon->dma.flush( ctx );
 +
 +      radeonEmitState(radeon);
 +   
 +      if (radeon->cmdbuf.cs->cdw)
 +              rcommonFlushCmdBuf(radeon, __FUNCTION__);
 +}
 +
 +/* Make sure all commands have been sent to the hardware and have
 + * completed processing.
 + */
 +void radeonFinish(GLcontext * ctx)
 +{
 +      radeonContextPtr radeon = RADEON_CONTEXT(ctx);
 +      struct gl_framebuffer *fb = ctx->DrawBuffer;
 +      int i;
 +
 +      radeonFlush(ctx);
 +
 +      if (radeon->radeonScreen->kernel_mm) {
 +              for (i = 0; i < fb->_NumColorDrawBuffers; i++) {
 +                      struct radeon_renderbuffer *rrb;
 +                      rrb = (struct radeon_renderbuffer *)fb->_ColorDrawBuffers[i];
 +                      if (rrb->bo)
 +                              radeon_bo_wait(rrb->bo);
 +              }
 +      } else if (radeon->do_irqs) {
 +              LOCK_HARDWARE(radeon);
 +              radeonEmitIrqLocked(radeon);
 +              UNLOCK_HARDWARE(radeon);
 +              radeonWaitIrq(radeon);
 +      } else {
 +              radeonWaitForIdle(radeon);
 +      }
 +}
 +
 +/* cmdbuffer */
 +/**
 + * Send the current command buffer via ioctl to the hardware.
 + */
 +int rcommonFlushCmdBufLocked(radeonContextPtr rmesa, const char *caller)
 +{
 +      int ret = 0;
 +
 +      if (rmesa->cmdbuf.flushing) {
 +              fprintf(stderr, "Recursive call into r300FlushCmdBufLocked!\n");
 +              exit(-1);
 +      }
 +      rmesa->cmdbuf.flushing = 1;
 +
 +      if (RADEON_DEBUG & DEBUG_IOCTL) {
 +              fprintf(stderr, "%s from %s - %i cliprects\n",
 +                      __FUNCTION__, caller, rmesa->numClipRects);
 +      }
 +
 +      if (rmesa->cmdbuf.cs->cdw) {
 +              ret = radeon_cs_emit(rmesa->cmdbuf.cs);
 +              rmesa->hw.all_dirty = GL_TRUE;
 +      }
 +      radeon_cs_erase(rmesa->cmdbuf.cs);
 +      rmesa->cmdbuf.flushing = 0;
 +      return ret;
 +}
 +
 +int rcommonFlushCmdBuf(radeonContextPtr rmesa, const char *caller)
 +{
 +      int ret;
 +
 +      radeonReleaseDmaRegion(rmesa);
 +      
 +      LOCK_HARDWARE(rmesa);
 +      ret = rcommonFlushCmdBufLocked(rmesa, caller);
 +      UNLOCK_HARDWARE(rmesa);
 +
 +      if (ret) {
 +              fprintf(stderr, "drmRadeonCmdBuffer: %d\n", ret);
 +              _mesa_exit(ret);
 +      }
 +
 +      return ret;
 +}
 +
 +/**
 + * Make sure that enough space is available in the command buffer
 + * by flushing if necessary.
 + *
 + * \param dwords The number of dwords we need to be free on the command buffer
 + */
 +void rcommonEnsureCmdBufSpace(radeonContextPtr rmesa, int dwords, const char *caller)
 +{
 +      if ((rmesa->cmdbuf.cs->cdw + dwords + 128) > rmesa->cmdbuf.size ||
 +          radeon_cs_need_flush(rmesa->cmdbuf.cs)) {
 +              rcommonFlushCmdBuf(rmesa, caller);
 +      }
 +}
 +
 +void rcommonInitCmdBuf(radeonContextPtr rmesa)
 +{
 +      GLuint size;
 +      /* Initialize command buffer */
 +      size = 256 * driQueryOptioni(&rmesa->optionCache,
 +                                   "command_buffer_size");
 +      if (size < 2 * rmesa->hw.max_state_size) {
 +              size = 2 * rmesa->hw.max_state_size + 65535;
 +      }
 +      if (size > 64 * 256)
 +              size = 64 * 256;
 +
 +      if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA)) {
 +              fprintf(stderr, "sizeof(drm_r300_cmd_header_t)=%zd\n",
 +                      sizeof(drm_r300_cmd_header_t));
 +              fprintf(stderr, "sizeof(drm_radeon_cmd_buffer_t)=%zd\n",
 +                      sizeof(drm_radeon_cmd_buffer_t));
 +              fprintf(stderr,
 +                      "Allocating %d bytes command buffer (max state is %d bytes)\n",
 +                      size * 4, rmesa->hw.max_state_size * 4);
 +      }
 +
 +      if (rmesa->radeonScreen->kernel_mm) {
 +              int fd = rmesa->radeonScreen->driScreen->fd;
 +              rmesa->cmdbuf.csm = radeon_cs_manager_gem_ctor(fd);
 +      } else {
 +              rmesa->cmdbuf.csm = radeon_cs_manager_legacy_ctor(rmesa);
 +      }
 +      if (rmesa->cmdbuf.csm == NULL) {
 +              /* FIXME: fatal error */
 +              return;
 +      }
 +      rmesa->cmdbuf.cs = radeon_cs_create(rmesa->cmdbuf.csm, size);
 +      assert(rmesa->cmdbuf.cs != NULL);
 +      rmesa->cmdbuf.size = size;
 +      
 +      if (!rmesa->radeonScreen->kernel_mm) {
 +              radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, rmesa->radeonScreen->texSize[0]);
 +              radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, rmesa->radeonScreen->gartTextures.size);
 +      } else {
 +              struct drm_radeon_gem_info mminfo;
 +
 +              if (!drmCommandWriteRead(rmesa->dri.fd, DRM_RADEON_GEM_INFO, &mminfo, sizeof(mminfo)))
 +              {
 +                      radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_VRAM, mminfo.vram_size);
 +                      radeon_cs_set_limit(rmesa->cmdbuf.cs, RADEON_GEM_DOMAIN_GTT, mminfo.gart_size);
 +              }
 +      }
 +
 +}
 +/**
 + * Destroy the command buffer
 + */
 +void rcommonDestroyCmdBuf(radeonContextPtr rmesa)
 +{
 +      radeon_cs_destroy(rmesa->cmdbuf.cs);
 +      if (rmesa->radeonScreen->driScreen->dri2.enabled || rmesa->radeonScreen->kernel_mm) {
 +              radeon_cs_manager_gem_dtor(rmesa->cmdbuf.csm);
 +      } else {
 +              radeon_cs_manager_legacy_dtor(rmesa->cmdbuf.csm);
 +      }
 +}
 +
 +void rcommonBeginBatch(radeonContextPtr rmesa, int n,
 +                     int dostate,
 +                     const char *file,
 +                     const char *function,
 +                     int line)
 +{
 +      rcommonEnsureCmdBufSpace(rmesa, n, function);
 +      if (!rmesa->cmdbuf.cs->cdw && dostate) {
 +              if (RADEON_DEBUG & DEBUG_IOCTL)
 +                      fprintf(stderr, "Reemit state after flush (from %s)\n", function);
 +              radeonEmitState(rmesa);
 +      }
 +      radeon_cs_begin(rmesa->cmdbuf.cs, n, file, function, line);
 +
 +        if (DEBUG_CMDBUF && RADEON_DEBUG & DEBUG_IOCTL)
 +                fprintf(stderr, "BEGIN_BATCH(%d) at %d, from %s:%i\n",
 +                        n, rmesa->cmdbuf.cs->cdw, function, line);
 +
 +}
 +
 +
 +
@@@ -714,162 -819,9 +714,162 @@@ static int radeon_set_screen_flags(rade
  
     default:
        fprintf(stderr, "unknown chip id 0x%x, can't guess.\n",
 -            dri_priv->deviceID);
 +            device_id);
 +      return -1;
 +   }
 +
 +   return 0;
 +}
 +
 +
 +/* Create the device specific screen private data struct.
 + */
 +static radeonScreenPtr
 +radeonCreateScreen( __DRIscreenPrivate *sPriv )
 +{
 +   radeonScreenPtr screen;
 +   RADEONDRIPtr dri_priv = (RADEONDRIPtr)sPriv->pDevPriv;
 +   unsigned char *RADEONMMIO = NULL;
 +   int i;
 +   int ret;
-    uint32_t temp;
++   uint32_t temp = 0;
 +
 +   if (sPriv->devPrivSize != sizeof(RADEONDRIRec)) {
 +      fprintf(stderr,"\nERROR!  sizeof(RADEONDRIRec) does not match passed size from device driver\n");
 +      return GL_FALSE;
 +   }
 +
 +   /* Allocate the private area */
 +   screen = (radeonScreenPtr) CALLOC( sizeof(*screen) );
 +   if ( !screen ) {
 +      __driUtilMessage("%s: Could not allocate memory for screen structure",
 +                     __FUNCTION__);
        return NULL;
     }
 +
 +#if DO_DEBUG && RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)
 +      RADEON_DEBUG = driParseDebugString(getenv("RADEON_DEBUG"), debug_control);
 +#endif
 +
 +   /* parse information in __driConfigOptions */
 +   driParseOptionInfo (&screen->optionCache,
 +                     __driConfigOptions, __driNConfigOptions);
 +
 +   /* This is first since which regions we map depends on whether or
 +    * not we are using a PCI card.
 +    */
 +   screen->card_type = (dri_priv->IsPCI ? RADEON_CARD_PCI : RADEON_CARD_AGP);
 +   {
 +      int ret;
 +
 +#ifdef RADEON_PARAM_KERNEL_MM
 +     ret = radeonGetParam( sPriv->fd, RADEON_PARAM_KERNEL_MM,
 +                            &screen->kernel_mm);
 +
 +      if (ret && ret != -EINVAL) {
 +         FREE( screen );
 +         fprintf(stderr, "drm_radeon_getparam_t (RADEON_OFFSET): %d\n", ret);
 +         return NULL;
 +      }
 +
 +      if (ret == -EINVAL)
 +          screen->kernel_mm = 0;
 +#endif
 +
 +      ret = radeonGetParam( sPriv->fd, RADEON_PARAM_GART_BUFFER_OFFSET,
 +                          &screen->gart_buffer_offset);
 +
 +      if (ret) {
 +       FREE( screen );
 +       fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BUFFER_OFFSET): %d\n", ret);
 +       return NULL;
 +      }
 +
 +      ret = radeonGetParam( sPriv->fd, RADEON_PARAM_GART_BASE,
 +                          &screen->gart_base);
 +      if (ret) {
 +       FREE( screen );
 +       fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_GART_BASE): %d\n", ret);
 +       return NULL;
 +      }
 +
 +      ret = radeonGetParam( sPriv->fd, RADEON_PARAM_IRQ_NR,
 +                          &screen->irq);
 +      if (ret) {
 +       FREE( screen );
 +       fprintf(stderr, "drm_radeon_getparam_t (RADEON_PARAM_IRQ_NR): %d\n", ret);
 +       return NULL;
 +      }
 +      screen->drmSupportsCubeMapsR200 = (sPriv->drm_version.minor >= 7);
 +      screen->drmSupportsBlendColor = (sPriv->drm_version.minor >= 11);
 +      screen->drmSupportsTriPerf = (sPriv->drm_version.minor >= 16);
 +      screen->drmSupportsFragShader = (sPriv->drm_version.minor >= 18);
 +      screen->drmSupportsPointSprites = (sPriv->drm_version.minor >= 13);
 +      screen->drmSupportsCubeMapsR100 = (sPriv->drm_version.minor >= 15);
 +      screen->drmSupportsVertexProgram = (sPriv->drm_version.minor >= 25);
 +   }
 +
 +   if (!screen->kernel_mm) {
 +     screen->mmio.handle = dri_priv->registerHandle;
 +     screen->mmio.size   = dri_priv->registerSize;
 +     if ( drmMap( sPriv->fd,
 +                screen->mmio.handle,
 +                screen->mmio.size,
 +                &screen->mmio.map ) ) {
 +       FREE( screen );
 +       __driUtilMessage("%s: drmMap failed\n", __FUNCTION__ );
 +       return NULL;
 +     }
 +
 +     RADEONMMIO = screen->mmio.map;
 +
 +     screen->status.handle = dri_priv->statusHandle;
 +     screen->status.size   = dri_priv->statusSize;
 +     if ( drmMap( sPriv->fd,
 +                screen->status.handle,
 +                screen->status.size,
 +                &screen->status.map ) ) {
 +       drmUnmap( screen->mmio.map, screen->mmio.size );
 +       FREE( screen );
 +       __driUtilMessage("%s: drmMap (2) failed\n", __FUNCTION__ );
 +       return NULL;
 +     }
 +     screen->scratch = (__volatile__ uint32_t *)
 +       ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET);
 +
 +     screen->buffers = drmMapBufs( sPriv->fd );
 +     if ( !screen->buffers ) {
 +       drmUnmap( screen->status.map, screen->status.size );
 +       drmUnmap( screen->mmio.map, screen->mmio.size );
 +       FREE( screen );
 +       __driUtilMessage("%s: drmMapBufs failed\n", __FUNCTION__ );
 +       return NULL;
 +     }
 +     
 +     if ( dri_priv->gartTexHandle && dri_priv->gartTexMapSize ) {
 +       screen->gartTextures.handle = dri_priv->gartTexHandle;
 +       screen->gartTextures.size   = dri_priv->gartTexMapSize;
 +       if ( drmMap( sPriv->fd,
 +                  screen->gartTextures.handle,
 +                  screen->gartTextures.size,
 +                  (drmAddressPtr)&screen->gartTextures.map ) ) {
 +       drmUnmapBufs( screen->buffers );
 +       drmUnmap( screen->status.map, screen->status.size );
 +       drmUnmap( screen->mmio.map, screen->mmio.size );
 +       FREE( screen );
 +       __driUtilMessage("%s: drmMap failed for GART texture area\n", __FUNCTION__);
 +       return NULL;
 +       }
 +       
 +       screen->gart_texture_offset = dri_priv->gartTexOffset + screen->gart_base;
 +     }
 +   }
 +
 +
 +   ret = radeon_set_screen_flags(screen, dri_priv->deviceID);
 +   if (ret == -1)
 +     return NULL;
 +
     if ((screen->chip_family == CHIP_FAMILY_R350 || screen->chip_family == CHIP_FAMILY_R300) &&
         sPriv->ddx_version.minor < 2) {
        fprintf(stderr, "xf86-video-ati-6.6.2 or newer needed for Radeon 9500/9700/9800 cards.\n");
@@@ -1340,26 -1117,7 +1340,26 @@@ radeonCreateBuffer( __DRIscreenPrivate 
  static void
  radeonDestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
  {
 -   _mesa_reference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)), NULL);
 +    struct radeon_renderbuffer *rb;
 +    struct radeon_framebuffer *rfb;
 +    
 +    rfb = (void*)driDrawPriv->driverPrivate;
 +    rb = (void *)rfb->base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer;
 +    if (rb && rb->bo) {
 +        radeon_bo_unref(rb->bo);
 +        rb->bo = NULL;
 +    }
 +    rb = (void *)rfb->base.Attachment[BUFFER_BACK_LEFT].Renderbuffer;
 +    if (rb && rb->bo) {
 +        radeon_bo_unref(rb->bo);
 +        rb->bo = NULL;
 +    }
 +    rb = (void *)rfb->base.Attachment[BUFFER_DEPTH].Renderbuffer;
 +    if (rb && rb->bo) {
 +        radeon_bo_unref(rb->bo);
 +        rb->bo = NULL;
 +    }
-    _mesa_unreference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)));
++    _mesa_reference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)), NULL);
  }
  
  #if RADEON_COMMON && defined(RADEON_COMMON_FOR_R300)