OK, one more time. Simplify the state-backup system by just storing the full

state in a ready-to-emit cmdbuf, which avoids the issue Nicolai Haehnle reported
where the check() could return differently during backup-and-emit than it should
have if it were called at the right time.  Move the lit emission before most of
the TCL state emission on r200, which fixes neverball issues.

Tested with:	r100/r200 with neverball, tuxracer, chromium, quake3, ipers

13 files changed, 95 insertions, 108 deletions

diff --git a/src/mesa/drivers/dri/r200/r200_cmdbuf.c b/src/mesa/drivers/dri/r200/r200_cmdbuf.c
index 384c1634ea..5cdb3657f9 100644
--- a/src/mesa/drivers/dri/r200/r200_cmdbuf.c
+++ b/src/mesa/drivers/dri/r200/r200_cmdbuf.c
@@ -92,18 +92,38 @@ void r200SetUpAtomList( r200ContextPtr rmesa )
        insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.tex[i] );
    for (i = 0; i < mtu; ++i)
        insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.cube[i] );
+   for (i = 0; i < 6; ++i)
+       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.pix[i] );
+
+   for (i = 0; i < 8; ++i)
+       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.lit[i] );
    for (i = 0; i < 3 + mtu; ++i)
        insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.mat[i] );
    insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.eye );
    insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.glt );
    for (i = 0; i < 2; ++i)
       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.mtl[i] );
-   for (i = 0; i < 8; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.lit[i] );
    for (i = 0; i < 6; ++i)
        insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.ucp[i] );
-   for (i = 0; i < 6; ++i)
-       insert_at_tail( &rmesa->hw.atomlist, &rmesa->hw.pix[i] );
+}
+
+static void r200SaveHwState( r200ContextPtr rmesa )
+{
+   struct r200_state_atom *atom;
+   char * dest = rmesa->backup_store.cmd_buf;
+
+   rmesa->backup_store.cmd_used = 0;
+
+   foreach( atom, &rmesa->hw.atomlist ) {
+      if ( atom->check( rmesa->glCtx, atom->idx ) ) {
+	 int size = atom->cmd_size * 4;
+	 memcpy( dest, atom->cmd, size);
+	 dest += size;
+	 rmesa->backup_store.cmd_used += size;
+      }
+   }
+
+   assert( rmesa->backup_store.cmd_used <= R200_CMD_BUF_SZ );
 }
 
 void r200EmitState( r200ContextPtr rmesa )
@@ -115,6 +135,11 @@ void r200EmitState( r200ContextPtr rmesa )
    if (R200_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
       fprintf(stderr, "%s\n", __FUNCTION__);
 
+   if (rmesa->save_on_next_emit) {
+      r200SaveHwState(rmesa);
+      rmesa->save_on_next_emit = GL_FALSE;
+   }
+
    if (!rmesa->hw.is_dirty && !rmesa->hw.all_dirty)
       return;
 
diff --git a/src/mesa/drivers/dri/r200/r200_context.c b/src/mesa/drivers/dri/r200/r200_context.c
index 3b8394f794..d5ca2f9b3a 100644
--- a/src/mesa/drivers/dri/r200/r200_context.c
+++ b/src/mesa/drivers/dri/r200/r200_context.c
@@ -62,7 +62,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_vtxfmt.h"
 #include "r200_maos.h"
 
-#define DRIVER_DATE	"20040924"
+#define DRIVER_DATE	"20040929"
 
 #include "vblank.h"
 #include "utils.h"
diff --git a/src/mesa/drivers/dri/r200/r200_context.h b/src/mesa/drivers/dri/r200/r200_context.h
index 25faec955c..140902c356 100644
--- a/src/mesa/drivers/dri/r200/r200_context.h
+++ b/src/mesa/drivers/dri/r200/r200_context.h
@@ -191,7 +191,6 @@ struct r200_state_atom {
    GLuint idx;
    int *cmd;			         /* one or more cmd's */
    int *lastcmd;			 /* one or more cmd's */
-   int *savedcmd;			 /* one or more cmd's */
    GLboolean dirty;
    GLboolean (*check)( GLcontext *, int );    /* is this state active? */
 };
@@ -825,8 +824,6 @@ struct r200_vbinfo {
 };
 
 
-
-
 struct r200_context {
    GLcontext *glCtx;			/* Mesa context */
 
@@ -854,6 +851,10 @@ struct r200_context {
    struct r200_ioctl ioctl;
    struct r200_dma dma;
    struct r200_store store;
+   /* A full state emit as of the first state emit in the main store, in case
+    * the context is lost.
+    */
+   struct r200_store backup_store;
 
    /* Page flipping
     */
@@ -876,7 +877,7 @@ struct r200_context {
    drm_clip_rect_t *pClipRects;
    unsigned int lastStamp;
    GLboolean lost_context;
-   GLboolean save_on_next_unlock;
+   GLboolean save_on_next_emit;
    r200ScreenPtr r200Screen;	/* Screen private DRI data */
    drm_radeon_sarea_t *sarea;		/* Private SAREA data */
 
diff --git a/src/mesa/drivers/dri/r200/r200_ioctl.c b/src/mesa/drivers/dri/r200/r200_ioctl.c
index 74dd7d38eb..d2fde39a1c 100644
--- a/src/mesa/drivers/dri/r200/r200_ioctl.c
+++ b/src/mesa/drivers/dri/r200/r200_ioctl.c
@@ -58,59 +58,34 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 static void r200WaitForIdle( r200ContextPtr rmesa );
 
-void r200SaveHwState( r200ContextPtr rmesa )
-{
-   struct r200_state_atom *atom;
-
-   foreach( atom, &rmesa->hw.atomlist )
-      memcpy(atom->savedcmd, atom->cmd, atom->cmd_size * 4);
-}
-
-static void r200SwapHwState( r200ContextPtr rmesa )
-{
-   int *temp;
-   struct r200_state_atom *atom;
-
-   foreach( atom, &rmesa->hw.atomlist ) {
-      temp = atom->cmd;
-      atom->cmd = atom->savedcmd;
-      atom->savedcmd = temp;
-   }
-}
 
 /* At this point we were in FlushCmdBufLocked but we had lost our context, so
- * we need to unwire our current cmdbuf and hook a new one in, emit that, then
- * wire the old cmdbuf back in so that FlushCmdBufLocked can continue and the
- * buffer can depend on the state not being lost across lock/unlock.
+ * we need to unwire our current cmdbuf, hook the one with the saved state in
+ * it, flush it, and then put the current one back.  This is so commands at the
+ * start of a cmdbuf can rely on the state being kept from the previous one.
  */
 static void r200BackUpAndEmitLostStateLocked( r200ContextPtr rmesa )
 {
-   GLuint nr_released_bufs;
-   struct r200_store store;
+   GLuint nr_released_bufs, saved_cmd_used;
+   struct r200_store saved_store;
+
+   if (rmesa->backup_store.cmd_used == 0)
+      return;
+
+   if (R200_DEBUG & DEBUG_STATE)
+      fprintf(stderr, "Emitting backup state on lost context\n");
 
    rmesa->lost_context = GL_FALSE;
 
    nr_released_bufs = rmesa->dma.nr_released_bufs;
-   store = rmesa->store;
-   rmesa->store.statenr = 0;
-   rmesa->store.primnr = 0;
-   rmesa->store.cmd_used = 0;
-   rmesa->store.elts_start = 0;
-   rmesa->hw.all_dirty = GL_TRUE;
-   r200SwapHwState( rmesa );
-   /* In this case it's okay to EmitState while locked because we won't exhaust
-    * our (empty) cmdbuf.
-    */
-   r200EmitState( rmesa );
+   saved_store = rmesa->store;
+   rmesa->dma.nr_released_bufs = 0;
+   rmesa->store = rmesa->backup_store;
+   saved_cmd_used = rmesa->backup_store.cmd_used;
    r200FlushCmdBufLocked( rmesa, __FUNCTION__ );
-
-   r200SwapHwState( rmesa );
-   /* We've just cleared out the dirty flags, so we don't remember what 
-    * actually needed to be emitted for the next state emit.
-    */
-   rmesa->hw.all_dirty = GL_TRUE;
+   rmesa->backup_store.cmd_used = saved_cmd_used;
    rmesa->dma.nr_released_bufs = nr_released_bufs;
-   rmesa->store = store;
+   rmesa->store = saved_store;
 }
 
 int r200FlushCmdBufLocked( r200ContextPtr rmesa, const char * caller )
@@ -189,7 +164,7 @@ int r200FlushCmdBufLocked( r200ContextPtr rmesa, const char * caller )
    rmesa->store.statenr = 0;
    rmesa->store.cmd_used = 0;
    rmesa->dma.nr_released_bufs = 0;
-   rmesa->save_on_next_unlock = 1;
+   rmesa->save_on_next_emit = 1;
 
    return ret;
 }
diff --git a/src/mesa/drivers/dri/r200/r200_ioctl.h b/src/mesa/drivers/dri/r200/r200_ioctl.h
index 57474d7a05..939dab3685 100644
--- a/src/mesa/drivers/dri/r200/r200_ioctl.h
+++ b/src/mesa/drivers/dri/r200/r200_ioctl.h
@@ -117,7 +117,6 @@ extern GLboolean r200IsGartMemory( r200ContextPtr rmesa, const GLvoid *pointer,
 extern GLuint r200GartOffsetFromVirtual( r200ContextPtr rmesa, 
 					 const GLvoid *pointer );
 
-void r200SaveHwState( r200ContextPtr rmesa );
 void r200SetUpAtomList( r200ContextPtr rmesa );
 
 /* ================================================================
diff --git a/src/mesa/drivers/dri/r200/r200_lock.h b/src/mesa/drivers/dri/r200/r200_lock.h
index 908052659c..587e4fe5cc 100644
--- a/src/mesa/drivers/dri/r200/r200_lock.h
+++ b/src/mesa/drivers/dri/r200/r200_lock.h
@@ -104,10 +104,6 @@ extern int prevLockLine;
 		  rmesa->dri.hwLock,					\
 		  rmesa->dri.hwContext );				\
       DEBUG_RESET();							\
-      if (rmesa->save_on_next_unlock) {					\
-	 r200SaveHwState( rmesa );					\
-	 rmesa->save_on_next_unlock = GL_FALSE;				\
-      }									\
    } while (0)
 
 #endif
diff --git a/src/mesa/drivers/dri/r200/r200_state_init.c b/src/mesa/drivers/dri/r200/r200_state_init.c
index e97a2f4d99..049820b3cb 100644
--- a/src/mesa/drivers/dri/r200/r200_state_init.c
+++ b/src/mesa/drivers/dri/r200/r200_state_init.c
@@ -208,7 +208,6 @@ void r200InitState( r200ContextPtr rmesa )
       rmesa->hw.ATOM.cmd_size = SZ;				\
       rmesa->hw.ATOM.cmd = (int *)CALLOC(SZ * sizeof(int));	\
       rmesa->hw.ATOM.lastcmd = (int *)CALLOC(SZ * sizeof(int));	\
-      rmesa->hw.ATOM.savedcmd = (int *)CALLOC(SZ * sizeof(int)); \
       rmesa->hw.ATOM.name = NM;					\
       rmesa->hw.ATOM.idx = IDX;					\
       rmesa->hw.ATOM.check = check_##CHK;			\
@@ -770,6 +769,5 @@ void r200InitState( r200ContextPtr rmesa )
 
    r200LightingSpaceChange( ctx );
    
-   r200SaveHwState( rmesa );
    rmesa->hw.all_dirty = GL_TRUE;
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_context.c b/src/mesa/drivers/dri/radeon/radeon_context.c
index dcc704cfd2..5f419347b4 100644
--- a/src/mesa/drivers/dri/radeon/radeon_context.c
+++ b/src/mesa/drivers/dri/radeon/radeon_context.c
@@ -62,7 +62,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_vtxfmt.h"
 #include "radeon_maos.h"
 
-#define DRIVER_DATE	"20040924"
+#define DRIVER_DATE	"20040929"
 
 #include "vblank.h"
 #include "utils.h"
diff --git a/src/mesa/drivers/dri/radeon/radeon_context.h b/src/mesa/drivers/dri/radeon/radeon_context.h
index f5bc0b0547..b44ea08dc8 100644
--- a/src/mesa/drivers/dri/radeon/radeon_context.h
+++ b/src/mesa/drivers/dri/radeon/radeon_context.h
@@ -185,7 +185,6 @@ struct radeon_state_atom {
    GLuint is_tcl;
    int *cmd;			         /* one or more cmd's */
    int *lastcmd;			 /* one or more cmd's */
-   int *savedcmd;			 /* one or more cmd's */
    GLboolean dirty;                      /* dirty-mark in emit_state_list */
    GLboolean (*check)( GLcontext * );    /* is this state active? */
 };
@@ -714,6 +713,10 @@ struct radeon_context {
    struct radeon_ioctl ioctl;
    struct radeon_dma dma;
    struct radeon_store store;
+   /* A full state emit as of the first state emit in the main store, in case
+    * the context is lost.
+    */
+   struct radeon_store backup_store;
 
    /* Page flipping
     */
@@ -732,7 +735,7 @@ struct radeon_context {
    drm_clip_rect_t *pClipRects;
    unsigned int lastStamp;
    GLboolean lost_context;
-   GLboolean save_on_next_unlock;
+   GLboolean save_on_next_emit;
    radeonScreenPtr radeonScreen;	/* Screen private DRI data */
    drm_radeon_sarea_t *sarea;		/* Private SAREA data */
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.c b/src/mesa/drivers/dri/radeon/radeon_ioctl.c
index 5ef14e34bb..50d5154016 100644
--- a/src/mesa/drivers/dri/radeon/radeon_ioctl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.c
@@ -62,59 +62,52 @@ static void radeonWaitForIdle( radeonContextPtr rmesa );
 static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, 
 				    const char * caller );
 
-void radeonSaveHwState( radeonContextPtr rmesa )
+static void radeonSaveHwState( radeonContextPtr rmesa )
 {
    struct radeon_state_atom *atom;
+   char * dest = rmesa->backup_store.cmd_buf;
 
-   foreach(atom, &rmesa->hw.atomlist)
-      memcpy(atom->savedcmd, atom->cmd, atom->cmd_size * 4);
-}
-
-static void radeonSwapHwState( radeonContextPtr rmesa )
-{
-   int *temp;
-   struct radeon_state_atom *atom;
+   rmesa->backup_store.cmd_used = 0;
 
-   foreach(atom, &rmesa->hw.atomlist) {
-      temp = atom->cmd;
-      atom->cmd = atom->savedcmd;
-      atom->savedcmd = temp;
+   foreach( atom, &rmesa->hw.atomlist ) {
+      if ( atom->check( rmesa->glCtx ) ) {
+	 int size = atom->cmd_size * 4;
+	 memcpy( dest, atom->cmd, size);
+	 dest += size;
+	 rmesa->backup_store.cmd_used += size;
+      }
    }
+
+   assert( rmesa->backup_store.cmd_used <= RADEON_CMD_BUF_SZ );
 }
 
 /* At this point we were in FlushCmdBufLocked but we had lost our context, so
- * we need to unwire our current cmdbuf and hook a new one in, emit that, then
- * wire the old cmdbuf back in so that FlushCmdBufLocked can continue and the
- * buffer can depend on the state not being lost across lock/unlock.
+ * we need to unwire our current cmdbuf, hook the one with the saved state in
+ * it, flush it, and then put the current one back.  This is so commands at the
+ * start of a cmdbuf can rely on the state being kept from the previous one.
  */
 static void radeonBackUpAndEmitLostStateLocked( radeonContextPtr rmesa )
 {
-   GLuint nr_released_bufs;
-   struct radeon_store store;
+   GLuint nr_released_bufs, saved_cmd_used;
+   struct radeon_store saved_store;
+
+   if (rmesa->backup_store.cmd_used == 0)
+      return;
+
+   if (RADEON_DEBUG & DEBUG_STATE)
+      fprintf(stderr, "Emitting backup state on lost context\n");
 
    rmesa->lost_context = GL_FALSE;
 
    nr_released_bufs = rmesa->dma.nr_released_bufs;
-   store = rmesa->store;
-   rmesa->store.statenr = 0;
-   rmesa->store.primnr = 0;
-   rmesa->store.cmd_used = 0;
-   rmesa->store.elts_start = 0;
-   rmesa->hw.all_dirty = GL_TRUE;
-   radeonSwapHwState( rmesa );
-   /* In this case it's okay to EmitState while locked because we won't exhaust
-    * our (empty) cmdbuf.
-    */
-   radeonEmitState(rmesa);
-   radeonFlushCmdBufLocked(rmesa, __FUNCTION__);
-
-   radeonSwapHwState(rmesa);
-   /* We've just cleared out the dirty flags, so we don't remember what 
-    * actually needed to be emitted for the next state emit.
-    */
-   rmesa->hw.all_dirty = GL_TRUE;
+   saved_store = rmesa->store;
+   rmesa->dma.nr_released_bufs = 0;
+   rmesa->store = rmesa->backup_store;
+   saved_cmd_used = rmesa->backup_store.cmd_used;
+   radeonFlushCmdBufLocked( rmesa, __FUNCTION__ );
+   rmesa->backup_store.cmd_used = saved_cmd_used;
    rmesa->dma.nr_released_bufs = nr_released_bufs;
-   rmesa->store = store;
+   rmesa->store = saved_store;
 }
 
 /* =============================================================
@@ -176,6 +169,11 @@ void radeonEmitState( radeonContextPtr rmesa )
    if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
       fprintf(stderr, "%s\n", __FUNCTION__);
 
+   if (rmesa->save_on_next_emit) {
+      radeonSaveHwState(rmesa);
+      rmesa->save_on_next_emit = GL_FALSE;
+   }
+
    if (!rmesa->hw.is_dirty && !rmesa->hw.all_dirty)
       return;
 
@@ -582,7 +580,7 @@ static int radeonFlushCmdBufLocked( radeonContextPtr rmesa,
    rmesa->store.statenr = 0;
    rmesa->store.cmd_used = 0;
    rmesa->dma.nr_released_bufs = 0;
-   rmesa->save_on_next_unlock = 1;
+   rmesa->save_on_next_emit = 1;
 
    return ret;
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.h b/src/mesa/drivers/dri/radeon/radeon_ioctl.h
index 7ad4f3a932..79b3b72c60 100644
--- a/src/mesa/drivers/dri/radeon/radeon_ioctl.h
+++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.h
@@ -103,8 +103,6 @@ extern void radeonWaitForIdleLocked( radeonContextPtr rmesa );
 extern void radeonWaitForVBlank( radeonContextPtr rmesa );
 extern void radeonInitIoctlFuncs( GLcontext *ctx );
 extern void radeonGetAllParams( radeonContextPtr rmesa );
-
-extern void radeonSaveHwState( radeonContextPtr rmesa );
 extern void radeonSetUpAtomList( radeonContextPtr rmesa );
 
 /* radeon_compat.c:
diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.h b/src/mesa/drivers/dri/radeon/radeon_lock.h
index c28cf23161..c2e0c3706b 100644
--- a/src/mesa/drivers/dri/radeon/radeon_lock.h
+++ b/src/mesa/drivers/dri/radeon/radeon_lock.h
@@ -105,10 +105,6 @@ extern int prevLockLine;
 		  rmesa->dri.hwLock,					\
 		  rmesa->dri.hwContext );				\
       DEBUG_RESET();							\
-      if ( rmesa->save_on_next_unlock ) {					\
-	 radeonSaveHwState( rmesa );					\
-	 rmesa->save_on_next_unlock = GL_FALSE;				\
-      }									\
    } while (0)
 
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_state_init.c b/src/mesa/drivers/dri/radeon/radeon_state_init.c
index 4df76603bb..0945fa803f 100644
--- a/src/mesa/drivers/dri/radeon/radeon_state_init.c
+++ b/src/mesa/drivers/dri/radeon/radeon_state_init.c
@@ -205,7 +205,6 @@ void radeonInitState( radeonContextPtr rmesa )
       rmesa->hw.ATOM.cmd_size = SZ;				\
       rmesa->hw.ATOM.cmd = (int *)CALLOC(SZ * sizeof(int));	\
       rmesa->hw.ATOM.lastcmd = (int *)CALLOC(SZ * sizeof(int));	\
-      rmesa->hw.ATOM.savedcmd = (int *)CALLOC(SZ * sizeof(int));	\
       rmesa->hw.ATOM.name = NM;					\
       rmesa->hw.ATOM.is_tcl = FLAG;					\
       rmesa->hw.ATOM.check = check_##CHK;				\
@@ -551,6 +550,5 @@ void radeonInitState( radeonContextPtr rmesa )
    rmesa->hw.eye.cmd[EYE_Z] = IEEE_ONE;
    rmesa->hw.eye.cmd[EYE_RESCALE_FACTOR] = IEEE_ONE;
    
-   radeonSaveHwState( rmesa );
    rmesa->hw.all_dirty = GL_TRUE;
 }