Merge commit 'origin/master' into gallium-0.2

6 files changed, 603 insertions, 22 deletions

diff --git a/src/mesa/drivers/dri/Makefile b/src/mesa/drivers/dri/Makefile
index eef68825bc9..9e49fb16f53 100644
--- a/src/mesa/drivers/dri/Makefile
+++ b/src/mesa/drivers/dri/Makefile
@@ -25,7 +25,8 @@ pcedit = sed \
 	-e 's,@INSTALL_LIB_DIR@,$(INSTALL_LIB_DIR),' \
 	-e 's,@INSTALL_INC_DIR@,$(INSTALL_INC_DIR),' \
 	-e 's,@VERSION@,$(MESA_MAJOR).$(MESA_MINOR).$(MESA_TINY),' \
-	-e 's,@DRI_DRIVER_DIR@,$(DRI_DRIVER_SEARCH_DIR),'
+	-e 's,@DRI_DRIVER_DIR@,$(DRI_DRIVER_SEARCH_DIR),' \
+	-e 's,@DRI_PC_REQ_PRIV@,$(DRI_PC_REQ_PRIV),'
 
 dri.pc: dri.pc.in
 	$(pcedit) $< > $@
diff --git a/src/mesa/drivers/dri/dri.pc.in b/src/mesa/drivers/dri/dri.pc.in
index c47ee9c7e7d..695aa6cfd66 100644
--- a/src/mesa/drivers/dri/dri.pc.in
+++ b/src/mesa/drivers/dri/dri.pc.in
@@ -7,4 +7,5 @@ dridriverdir=@DRI_DRIVER_DIR@
 Name: dri
 Description: Direct Rendering Infrastructure
 Version: @VERSION@
+Requires.private: @DRI_PC_REQ_PRIV@
 Cflags: -I${includedir}
diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
index cb728190f5c..baecfdcb799 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
@@ -1095,7 +1095,7 @@ static void noise1_sub( struct brw_wm_compile *c ) {
     /* Arrange the two end coordinates into scalars (itmp0/itmp1) to
        be hashed.  Also compute the remainder (offset within the unit
        length), interleaved to reduce register dependency penalties. */
-    brw_RNDD( p, itmp[ 0 ], param );
+    brw_RNDD( p, retype( itmp[ 0 ], BRW_REGISTER_TYPE_D ), param );
     brw_FRC( p, param, param );
     brw_ADD( p, itmp[ 1 ], itmp[ 0 ], brw_imm_ud( 1 ) );
     brw_MOV( p, itmp[ 3 ], brw_imm_ud( 0x79D9 ) ); /* constant used later */
@@ -1220,8 +1220,8 @@ static void noise2_sub( struct brw_wm_compile *c ) {
     /* Arrange the four corner coordinates into scalars (itmp0..itmp3) to
        be hashed.  Also compute the remainders (offsets within the unit
        square), interleaved to reduce register dependency penalties. */
-    brw_RNDD( p, itmp[ 0 ], param0 );
-    brw_RNDD( p, itmp[ 1 ], param1 );
+    brw_RNDD( p, retype( itmp[ 0 ], BRW_REGISTER_TYPE_D ), param0 );
+    brw_RNDD( p, retype( itmp[ 1 ], BRW_REGISTER_TYPE_D ), param1 );
     brw_FRC( p, param0, param0 );
     brw_FRC( p, param1, param1 );
     brw_MOV( p, itmp[ 4 ], brw_imm_ud( 0xBA97 ) ); /* constant used later */
@@ -1400,21 +1400,19 @@ static void noise3_sub( struct brw_wm_compile *c ) {
     /* Arrange the eight corner coordinates into scalars (itmp0..itmp3) to
        be hashed.  Also compute the remainders (offsets within the unit
        cube), interleaved to reduce register dependency penalties. */
-    brw_RNDD( p, itmp[ 0 ], param0 );
-    brw_RNDD( p, itmp[ 1 ], param1 );
-    brw_RNDD( p, itmp[ 2 ], param2 );
-    brw_MOV( p, itmp[ 4 ], brw_imm_ud( 0xBC8F ) ); /* constant used later */
-    brw_MOV( p, itmp[ 5 ], brw_imm_ud( 0xD0BD ) ); /* constant used later */
-    brw_MOV( p, itmp[ 6 ], brw_imm_ud( 0x9B93 ) ); /* constant used later */
+    brw_RNDD( p, retype( itmp[ 0 ], BRW_REGISTER_TYPE_D ), param0 );
+    brw_RNDD( p, retype( itmp[ 1 ], BRW_REGISTER_TYPE_D ), param1 );
+    brw_RNDD( p, retype( itmp[ 2 ], BRW_REGISTER_TYPE_D ), param2 );
     brw_FRC( p, param0, param0 );
     brw_FRC( p, param1, param1 );
     brw_FRC( p, param2, param2 );
     /* Since we now have only 16 bits of precision in the hash, we must
        be more careful about thorough mixing to maintain entropy as we
        squash the input vector into a small scalar. */
-    brw_MUL( p, brw_acc_reg(), itmp[ 4 ], itmp[ 0 ] );
-    brw_MAC( p, brw_acc_reg(), itmp[ 5 ], itmp[ 1 ] );
-    brw_MAC( p, itmp[ 0 ], itmp[ 6 ], itmp[ 2 ] );
+    brw_MUL( p, brw_null_reg(), low_words( itmp[ 0 ] ), brw_imm_uw( 0xBC8F ) );
+    brw_MAC( p, brw_null_reg(), low_words( itmp[ 1 ] ), brw_imm_uw( 0xD0BD ) );
+    brw_MAC( p, low_words( itmp[ 0 ] ), low_words( itmp[ 2 ] ),
+	     brw_imm_uw( 0x9B93 ) );
     brw_ADD( p, high_words( itmp[ 0 ] ), low_words( itmp[ 0 ] ),
 	     brw_imm_uw( 0xBC8F ) );
 
@@ -1668,6 +1666,430 @@ static void emit_noise3( struct brw_wm_compile *c,
     release_tmps( c, mark );
 }
     
+/* For the four-dimensional case, the little micro-optimisation benefits
+   we obtain by unrolling all the loops aren't worth the massive bloat it
+   now causes.  Instead, we loop twice around performing a similar operation
+   to noise3, once for the w=0 cube and once for the w=1, with a bit more
+   code to glue it all together. */
+static void noise4_sub( struct brw_wm_compile *c ) {
+
+    struct brw_compile *p = &c->func;
+    struct brw_reg param[ 4 ],
+	x0y0, x0y1, x1y0, x1y1, /* gradients at four of the corners */
+	w0, /* noise for the w=0 cube */
+	floors[ 2 ], /* integer coordinates of base corner of hypercube */
+	interp[ 4 ], /* interpolation coefficients */
+	t, tmp[ 8 ], /* float temporaries */
+	itmp[ 8 ], /* unsigned integer temporaries (aliases of floats above) */
+	wtmp[ 8 ]; /* 16-way unsigned word temporaries (aliases of above) */
+    int i, j;
+    int mark = mark_tmps( c );
+    GLuint loop, origin;
+    
+    x0y0 = alloc_tmp( c );
+    x0y1 = alloc_tmp( c );
+    x1y0 = alloc_tmp( c );
+    x1y1 = alloc_tmp( c );
+    t = alloc_tmp( c );
+    w0 = alloc_tmp( c );    
+    floors[ 0 ] = retype( alloc_tmp( c ), BRW_REGISTER_TYPE_UD );
+    floors[ 1 ] = retype( alloc_tmp( c ), BRW_REGISTER_TYPE_UD );
+
+    for( i = 0; i < 4; i++ ) {
+	param[ i ] = lookup_tmp( c, mark - 5 + i );
+	interp[ i ] = alloc_tmp( c );
+    }
+    
+    for( i = 0; i < 8; i++ ) {
+	tmp[ i ] = alloc_tmp( c );
+	itmp[ i ] = retype( tmp[ i ], BRW_REGISTER_TYPE_UD );
+	wtmp[ i ] = brw_uw16_grf( tmp[ i ].nr, 0 );
+    }
+
+    brw_set_access_mode( p, BRW_ALIGN_1 );
+
+    /* We only want 16 bits of precision from the integral part of each
+       co-ordinate, but unfortunately the RNDD semantics would saturate
+       at 16 bits if we performed the operation directly to a 16-bit
+       destination.  Therefore, we round to 32-bit temporaries where
+       appropriate, and then store only the lower 16 bits. */
+    brw_RNDD( p, retype( floors[ 0 ], BRW_REGISTER_TYPE_D ), param[ 0 ] );
+    brw_RNDD( p, retype( itmp[ 0 ], BRW_REGISTER_TYPE_D ), param[ 1 ] );
+    brw_RNDD( p, retype( floors[ 1 ], BRW_REGISTER_TYPE_D ), param[ 2 ] );
+    brw_RNDD( p, retype( itmp[ 1 ], BRW_REGISTER_TYPE_D ), param[ 3 ] );
+    brw_MOV( p, high_words( floors[ 0 ] ), low_words( itmp[ 0 ] ) );
+    brw_MOV( p, high_words( floors[ 1 ] ), low_words( itmp[ 1 ] ) );
+
+    /* Modify the flag register here, because the side effect is useful
+       later (see below).  We know for certain that all flags will be
+       cleared, since the FRC instruction cannot possibly generate
+       negative results.  Even for exceptional inputs (infinities, denormals,
+       NaNs), the architecture guarantees that the L conditional is false. */
+    brw_set_conditionalmod( p, BRW_CONDITIONAL_L );
+    brw_FRC( p, param[ 0 ], param[ 0 ] );
+    brw_set_predicate_control( p, BRW_PREDICATE_NONE );
+    for( i = 1; i < 4; i++ )	
+	brw_FRC( p, param[ i ], param[ i ] );
+    
+    /* Calculate the interpolation coefficients (6t^5 - 15t^4 + 10t^3) first
+       of all. */
+    for( i = 0; i < 4; i++ )
+	brw_MUL( p, interp[ i ], param[ i ], brw_imm_f( 6.0 ) );
+    for( i = 0; i < 4; i++ )
+	brw_ADD( p, interp[ i ], interp[ i ], brw_imm_f( -15.0 ) );
+    for( i = 0; i < 4; i++ )
+	brw_MUL( p, interp[ i ], interp[ i ], param[ i ] );
+    for( i = 0; i < 4; i++ )
+	brw_ADD( p, interp[ i ], interp[ i ], brw_imm_f( 10.0 ) );
+    for( j = 0; j < 3; j++ )
+	for( i = 0; i < 4; i++ )
+	    brw_MUL( p, interp[ i ], interp[ i ], param[ i ] );
+
+    /* Mark the current address, as it will be a jump destination.  The
+       following code will be executed twice: first, with the flag
+       register clear indicating the w=0 case, and second with flags
+       set for w=1. */
+    loop = p->nr_insn;
+    
+    /* Arrange the eight corner coordinates into scalars (itmp0..itmp3) to
+       be hashed.  Since we have only 16 bits of precision in the hash, we
+       must be careful about thorough mixing to maintain entropy as we
+       squash the input vector into a small scalar. */
+    brw_MUL( p, brw_null_reg(), low_words( floors[ 0 ] ),
+	     brw_imm_uw( 0xBC8F ) );
+    brw_MAC( p, brw_null_reg(), high_words( floors[ 0 ] ),
+	     brw_imm_uw( 0xD0BD ) );
+    brw_MAC( p, brw_null_reg(), low_words( floors[ 1 ] ),
+	     brw_imm_uw( 0x9B93 ) );
+    brw_MAC( p, low_words( itmp[ 0 ] ), high_words( floors[ 1 ] ),
+	     brw_imm_uw( 0xA359 ) );
+    brw_ADD( p, high_words( itmp[ 0 ] ), low_words( itmp[ 0 ] ),
+	     brw_imm_uw( 0xBC8F ) );
+
+    /* Temporarily disable the execution mask while we work with ExecSize=16
+       channels (the mask is set for ExecSize=8 and is probably incorrect).
+       Although this might cause execution of unwanted channels, the code
+       writes only to temporary registers and has no side effects, so
+       disabling the mask is harmless. */
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_ADD( p, wtmp[ 1 ], wtmp[ 0 ], brw_imm_uw( 0xD0BD ) );
+    brw_ADD( p, wtmp[ 2 ], wtmp[ 0 ], brw_imm_uw( 0x9B93 ) );
+    brw_ADD( p, wtmp[ 3 ], wtmp[ 1 ], brw_imm_uw( 0x9B93 ) );
+
+    /* We're now ready to perform the hashing.  The eight hashes are
+       interleaved for performance.  The hash function used is
+       designed to rapidly achieve avalanche and require only 16x16
+       bit multiplication, and 8-bit swizzles (which we get for
+       free). */
+    for( i = 0; i < 4; i++ )
+	brw_MUL( p, wtmp[ i ], wtmp[ i ], brw_imm_uw( 0x28D9 ) );
+    for( i = 0; i < 4; i++ )
+	brw_XOR( p, even_bytes( wtmp[ i ] ), even_bytes( wtmp[ i ] ),
+		 odd_bytes( wtmp[ i ] ) );
+    for( i = 0; i < 4; i++ )
+	brw_MUL( p, wtmp[ i ], wtmp[ i ], brw_imm_uw( 0xC6D5 ) );
+    for( i = 0; i < 4; i++ )
+	brw_XOR( p, even_bytes( wtmp[ i ] ), even_bytes( wtmp[ i ] ),
+		 odd_bytes( wtmp[ i ] ) );
+    brw_pop_insn_state( p );
+
+    /* Now we want to initialise the four rear gradients based on the
+       hashes.  Format conversion from signed integer to float leaves
+       everything scaled too high by a factor of pow( 2, 15 ), but
+       we correct for that right at the end. */
+    /* x component */
+    brw_ADD( p, t, param[ 0 ], brw_imm_f( -1.0 ) );
+    brw_MOV( p, x0y0, low_words( tmp[ 0 ] ) );
+    brw_MOV( p, x0y1, low_words( tmp[ 1 ] ) );
+    brw_MOV( p, x1y0, high_words( tmp[ 0 ] ) );
+    brw_MOV( p, x1y1, high_words( tmp[ 1 ] ) );
+
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_SHL( p, wtmp[ 0 ], wtmp[ 0 ], brw_imm_uw( 4 ) );
+    brw_SHL( p, wtmp[ 1 ], wtmp[ 1 ], brw_imm_uw( 4 ) );
+    brw_pop_insn_state( p );
+    
+    brw_MUL( p, x1y0, x1y0, t );
+    brw_MUL( p, x1y1, x1y1, t );
+    brw_ADD( p, t, param[ 1 ], brw_imm_f( -1.0 ) );
+    brw_MUL( p, x0y0, x0y0, param[ 0 ] );
+    brw_MUL( p, x0y1, x0y1, param[ 0 ] );
+
+    /* y component */
+    brw_MOV( p, tmp[ 5 ], low_words( tmp[ 1 ] ) );
+    brw_MOV( p, tmp[ 7 ], high_words( tmp[ 1 ] ) );
+    brw_MOV( p, tmp[ 4 ], low_words( tmp[ 0 ] ) );
+    brw_MOV( p, tmp[ 6 ], high_words( tmp[ 0 ] ) );
+    
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_SHL( p, wtmp[ 0 ], wtmp[ 0 ], brw_imm_uw( 4 ) );
+    brw_SHL( p, wtmp[ 1 ], wtmp[ 1 ], brw_imm_uw( 4 ) );
+    brw_pop_insn_state( p );
+
+    brw_MUL( p, tmp[ 5 ], tmp[ 5 ], t );
+    brw_MUL( p, tmp[ 7 ], tmp[ 7 ], t );    
+    /* prepare t for the w component (used below): w the first time through
+       the loop; w - 1 the second time) */
+    brw_set_predicate_control( p, BRW_PREDICATE_NORMAL );
+    brw_ADD( p, t, param[ 3 ], brw_imm_f( -1.0 ) );
+    p->current->header.predicate_inverse = 1;
+    brw_MOV( p, t, param[ 3 ] );
+    p->current->header.predicate_inverse = 0;
+    brw_set_predicate_control( p, BRW_PREDICATE_NONE );
+    brw_MUL( p, tmp[ 4 ], tmp[ 4 ], param[ 1 ] );
+    brw_MUL( p, tmp[ 6 ], tmp[ 6 ], param[ 1 ] );
+    
+    brw_ADD( p, x0y1, x0y1, tmp[ 5 ] );
+    brw_ADD( p, x1y1, x1y1, tmp[ 7 ] );
+    brw_ADD( p, x0y0, x0y0, tmp[ 4 ] );
+    brw_ADD( p, x1y0, x1y0, tmp[ 6 ] );
+    
+    /* z component */
+    brw_MOV( p, tmp[ 4 ], low_words( tmp[ 0 ] ) );
+    brw_MOV( p, tmp[ 5 ], low_words( tmp[ 1 ] ) );
+    brw_MOV( p, tmp[ 6 ], high_words( tmp[ 0 ] ) );
+    brw_MOV( p, tmp[ 7 ], high_words( tmp[ 1 ] ) );
+
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_SHL( p, wtmp[ 0 ], wtmp[ 0 ], brw_imm_uw( 4 ) );
+    brw_SHL( p, wtmp[ 1 ], wtmp[ 1 ], brw_imm_uw( 4 ) );
+    brw_pop_insn_state( p );
+
+    brw_MUL( p, tmp[ 4 ], tmp[ 4 ], param[ 2 ] );
+    brw_MUL( p, tmp[ 5 ], tmp[ 5 ], param[ 2 ] );
+    brw_MUL( p, tmp[ 6 ], tmp[ 6 ], param[ 2 ] );
+    brw_MUL( p, tmp[ 7 ], tmp[ 7 ], param[ 2 ] );
+    
+    brw_ADD( p, x0y0, x0y0, tmp[ 4 ] );
+    brw_ADD( p, x0y1, x0y1, tmp[ 5 ] );
+    brw_ADD( p, x1y0, x1y0, tmp[ 6 ] );
+    brw_ADD( p, x1y1, x1y1, tmp[ 7 ] );
+
+    /* w component */
+    brw_MOV( p, tmp[ 4 ], low_words( tmp[ 0 ] ) );
+    brw_MOV( p, tmp[ 5 ], low_words( tmp[ 1 ] ) );
+    brw_MOV( p, tmp[ 6 ], high_words( tmp[ 0 ] ) );
+    brw_MOV( p, tmp[ 7 ], high_words( tmp[ 1 ] ) );
+
+    brw_MUL( p, tmp[ 4 ], tmp[ 4 ], t );
+    brw_MUL( p, tmp[ 5 ], tmp[ 5 ], t );
+    brw_MUL( p, tmp[ 6 ], tmp[ 6 ], t );
+    brw_MUL( p, tmp[ 7 ], tmp[ 7 ], t );
+    brw_ADD( p, t, param[ 0 ], brw_imm_f( -1.0 ) );
+    
+    brw_ADD( p, x0y0, x0y0, tmp[ 4 ] );
+    brw_ADD( p, x0y1, x0y1, tmp[ 5 ] );
+    brw_ADD( p, x1y0, x1y0, tmp[ 6 ] );
+    brw_ADD( p, x1y1, x1y1, tmp[ 7 ] );
+
+    /* Here we interpolate in the y dimension... */
+    brw_ADD( p, x0y1, x0y1, negate( x0y0 ) );
+    brw_ADD( p, x1y1, x1y1, negate( x1y0 ) );
+    brw_MUL( p, x0y1, x0y1, interp[ 1 ] );
+    brw_MUL( p, x1y1, x1y1, interp[ 1 ] );
+    brw_ADD( p, x0y0, x0y0, x0y1 );
+    brw_ADD( p, x1y0, x1y0, x1y1 );
+
+    /* And now in x.  Leave the result in tmp[ 0 ] (see below)... */
+    brw_ADD( p, x1y0, x1y0, negate( x0y0 ) );
+    brw_MUL( p, x1y0, x1y0, interp[ 0 ] );
+    brw_ADD( p, tmp[ 0 ], x0y0, x1y0 );
+
+    /* Now do the same thing for the front four gradients... */
+    /* x component */
+    brw_MOV( p, x0y0, low_words( tmp[ 2 ] ) );
+    brw_MOV( p, x0y1, low_words( tmp[ 3 ] ) );
+    brw_MOV( p, x1y0, high_words( tmp[ 2 ] ) );
+    brw_MOV( p, x1y1, high_words( tmp[ 3 ] ) );
+
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_SHL( p, wtmp[ 2 ], wtmp[ 2 ], brw_imm_uw( 4 ) );
+    brw_SHL( p, wtmp[ 3 ], wtmp[ 3 ], brw_imm_uw( 4 ) );
+    brw_pop_insn_state( p );
+
+    brw_MUL( p, x1y0, x1y0, t );
+    brw_MUL( p, x1y1, x1y1, t );
+    brw_ADD( p, t, param[ 1 ], brw_imm_f( -1.0 ) );
+    brw_MUL( p, x0y0, x0y0, param[ 0 ] );
+    brw_MUL( p, x0y1, x0y1, param[ 0 ] );
+
+    /* y component */
+    brw_MOV( p, tmp[ 5 ], low_words( tmp[ 3 ] ) );
+    brw_MOV( p, tmp[ 7 ], high_words( tmp[ 3 ] ) );
+    brw_MOV( p, tmp[ 4 ], low_words( tmp[ 2 ] ) );
+    brw_MOV( p, tmp[ 6 ], high_words( tmp[ 2 ] ) );
+    
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_SHL( p, wtmp[ 2 ], wtmp[ 2 ], brw_imm_uw( 4 ) );
+    brw_SHL( p, wtmp[ 3 ], wtmp[ 3 ], brw_imm_uw( 4 ) );
+    brw_pop_insn_state( p );
+
+    brw_MUL( p, tmp[ 5 ], tmp[ 5 ], t );
+    brw_MUL( p, tmp[ 7 ], tmp[ 7 ], t );
+    brw_ADD( p, t, param[ 2 ], brw_imm_f( -1.0 ) );
+    brw_MUL( p, tmp[ 4 ], tmp[ 4 ], param[ 1 ] );
+    brw_MUL( p, tmp[ 6 ], tmp[ 6 ], param[ 1 ] );
+    
+    brw_ADD( p, x0y1, x0y1, tmp[ 5 ] );
+    brw_ADD( p, x1y1, x1y1, tmp[ 7 ] );
+    brw_ADD( p, x0y0, x0y0, tmp[ 4 ] );
+    brw_ADD( p, x1y0, x1y0, tmp[ 6 ] );
+    
+    /* z component */
+    brw_MOV( p, tmp[ 4 ], low_words( tmp[ 2 ] ) );
+    brw_MOV( p, tmp[ 5 ], low_words( tmp[ 3 ] ) );
+    brw_MOV( p, tmp[ 6 ], high_words( tmp[ 2 ] ) );
+    brw_MOV( p, tmp[ 7 ], high_words( tmp[ 3 ] ) );
+
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_SHL( p, wtmp[ 2 ], wtmp[ 2 ], brw_imm_uw( 4 ) );
+    brw_SHL( p, wtmp[ 3 ], wtmp[ 3 ], brw_imm_uw( 4 ) );
+    brw_pop_insn_state( p );
+
+    brw_MUL( p, tmp[ 4 ], tmp[ 4 ], t );
+    brw_MUL( p, tmp[ 5 ], tmp[ 5 ], t );
+    brw_MUL( p, tmp[ 6 ], tmp[ 6 ], t );
+    brw_MUL( p, tmp[ 7 ], tmp[ 7 ], t );
+    /* prepare t for the w component (used below): w the first time through
+       the loop; w - 1 the second time) */
+    brw_set_predicate_control( p, BRW_PREDICATE_NORMAL );
+    brw_ADD( p, t, param[ 3 ], brw_imm_f( -1.0 ) );
+    p->current->header.predicate_inverse = 1;
+    brw_MOV( p, t, param[ 3 ] );
+    p->current->header.predicate_inverse = 0;
+    brw_set_predicate_control( p, BRW_PREDICATE_NONE );
+    
+    brw_ADD( p, x0y0, x0y0, tmp[ 4 ] );
+    brw_ADD( p, x0y1, x0y1, tmp[ 5 ] );
+    brw_ADD( p, x1y0, x1y0, tmp[ 6 ] );
+    brw_ADD( p, x1y1, x1y1, tmp[ 7 ] );
+
+    /* w component */
+    brw_MOV( p, tmp[ 4 ], low_words( tmp[ 2 ] ) );
+    brw_MOV( p, tmp[ 5 ], low_words( tmp[ 3 ] ) );
+    brw_MOV( p, tmp[ 6 ], high_words( tmp[ 2 ] ) );
+    brw_MOV( p, tmp[ 7 ], high_words( tmp[ 3 ] ) );
+
+    brw_MUL( p, tmp[ 4 ], tmp[ 4 ], t );
+    brw_MUL( p, tmp[ 5 ], tmp[ 5 ], t );
+    brw_MUL( p, tmp[ 6 ], tmp[ 6 ], t );
+    brw_MUL( p, tmp[ 7 ], tmp[ 7 ], t );
+    
+    brw_ADD( p, x0y0, x0y0, tmp[ 4 ] );
+    brw_ADD( p, x0y1, x0y1, tmp[ 5 ] );
+    brw_ADD( p, x1y0, x1y0, tmp[ 6 ] );
+    brw_ADD( p, x1y1, x1y1, tmp[ 7 ] );
+
+    /* Interpolate in the y dimension: */
+    brw_ADD( p, x0y1, x0y1, negate( x0y0 ) );
+    brw_ADD( p, x1y1, x1y1, negate( x1y0 ) );
+    brw_MUL( p, x0y1, x0y1, interp[ 1 ] );
+    brw_MUL( p, x1y1, x1y1, interp[ 1 ] );
+    brw_ADD( p, x0y0, x0y0, x0y1 );
+    brw_ADD( p, x1y0, x1y0, x1y1 );
+
+    /* And now in x.  The rear face is in tmp[ 0 ] (see above), so this
+       time put the front face in tmp[ 1 ] and we're nearly there... */
+    brw_ADD( p, x1y0, x1y0, negate( x0y0 ) );
+    brw_MUL( p, x1y0, x1y0, interp[ 0 ] );
+    brw_ADD( p, tmp[ 1 ], x0y0, x1y0 );
+
+    /* Another interpolation, in the z dimension: */
+    brw_ADD( p, tmp[ 1 ], tmp[ 1 ], negate( tmp[ 0 ] ) );    
+    brw_MUL( p, tmp[ 1 ], tmp[ 1 ], interp[ 2 ] );
+    brw_ADD( p, tmp[ 0 ], tmp[ 0 ], tmp[ 1 ] );
+
+    /* Exit the loop if we've computed both cubes... */
+    origin = p->nr_insn;
+    brw_push_insn_state( p );
+    brw_set_predicate_control( p, BRW_PREDICATE_NORMAL );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_ADD( p, brw_ip_reg(), brw_ip_reg(), brw_imm_d( 0 ) );
+    brw_pop_insn_state( p );
+
+    /* Save the result for the w=0 case, and increment the w coordinate: */
+    brw_MOV( p, w0, tmp[ 0 ] );
+    brw_ADD( p, high_words( floors[ 1 ] ), high_words( floors[ 1 ] ),
+	     brw_imm_uw( 1 ) );
+
+    /* Loop around for the other cube.  Explicitly set the flag register
+       (unfortunately we must spend an extra instruction to do this: we
+       can't rely on a side effect of the previous MOV or ADD because
+       conditional modifiers which are normally true might be false in
+       exceptional circumstances, e.g. given a NaN input; the add to
+       brw_ip_reg() is not suitable because the IP is not an 8-vector). */
+    brw_push_insn_state( p );
+    brw_set_mask_control( p, BRW_MASK_DISABLE );
+    brw_MOV( p, brw_flag_reg(), brw_imm_uw( 0xFF ) );
+    brw_ADD( p, brw_ip_reg(), brw_ip_reg(),
+	     brw_imm_d( ( loop - p->nr_insn ) << 4 ) );
+    brw_pop_insn_state( p );
+
+    /* Patch the previous conditional branch now that we know the
+       destination address. */
+    brw_set_src1( p->store + origin,
+		  brw_imm_d( ( p->nr_insn - origin ) << 4 ) );
+
+    /* The very last interpolation. */
+    brw_ADD( p, tmp[ 0 ], tmp[ 0 ], negate( w0 ) );    
+    brw_MUL( p, tmp[ 0 ], tmp[ 0 ], interp[ 3 ] );
+    brw_ADD( p, tmp[ 0 ], tmp[ 0 ], w0 );
+
+    /* scale by pow( 2, -15 ), as described above */
+    brw_MUL( p, param[ 0 ], tmp[ 0 ], brw_imm_f( 0.000030517578125 ) );
+
+    release_tmps( c, mark );
+}
+
+static void emit_noise4( struct brw_wm_compile *c,
+			 struct prog_instruction *inst )
+{
+    struct brw_compile *p = &c->func;
+    struct brw_reg src0, src1, src2, src3, param0, param1, param2, param3, dst;
+    GLuint mask = inst->DstReg.WriteMask;
+    int i;
+    int mark = mark_tmps( c );
+
+    assert( mark == 0 );
+    
+    src0 = get_src_reg( c, inst->SrcReg, 0, 1 );
+    src1 = get_src_reg( c, inst->SrcReg, 1, 1 );
+    src2 = get_src_reg( c, inst->SrcReg, 2, 1 );
+    src3 = get_src_reg( c, inst->SrcReg, 3, 1 );
+
+    param0 = alloc_tmp( c );
+    param1 = alloc_tmp( c );
+    param2 = alloc_tmp( c );
+    param3 = alloc_tmp( c );
+
+    brw_MOV( p, param0, src0 );
+    brw_MOV( p, param1, src1 );
+    brw_MOV( p, param2, src2 );
+    brw_MOV( p, param3, src3 );
+
+    invoke_subroutine( c, SUB_NOISE4, noise4_sub );
+    
+    /* Fill in the result: */
+    brw_set_saturate( p, inst->SaturateMode == SATURATE_ZERO_ONE );
+    for (i = 0 ; i < 4; i++) {
+	if (mask & (1<<i)) {
+	    dst = get_dst_reg(c, inst, i, 1);
+	    brw_MOV( p, dst, param0 );
+	}
+    }
+    if( inst->SaturateMode == SATURATE_ZERO_ONE )
+	brw_set_saturate( p, 0 );
+    
+    release_tmps( c, mark );
+}
+    
 static void emit_wpos_xy(struct brw_wm_compile *c,
                 struct prog_instruction *inst)
 {
@@ -1996,8 +2418,9 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 	    case OPCODE_NOISE3:
 		emit_noise3(c, inst);
 		break;
-	    /* case OPCODE_NOISE4: */
-		/* not yet implemented */
+	    case OPCODE_NOISE4:
+		emit_noise4(c, inst);
+		break;
 	    case OPCODE_TEX:
 		emit_tex(c, inst);
 		break;
diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.h b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
index 8129996979f..51579df09e7 100644
--- a/src/mesa/drivers/dri/intel/intel_batchbuffer.h
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
@@ -55,6 +55,12 @@ struct intel_batchbuffer
 
    GLuint size;
 
+   /** Tracking of BEGIN_BATCH()/OUT_BATCH()/ADVANCE_BATCH() debugging */
+   struct {
+      GLuint total;
+      GLubyte *start_ptr;
+   } emit;
+
    GLuint dirty_state;
 };
 
@@ -143,9 +149,12 @@ intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
 
 #define BEGIN_BATCH(n, cliprect_mode) do {				\
    intel_batchbuffer_require_space(intel->batch, (n)*4, cliprect_mode); \
+   assert(intel->batch->emit.start_ptr == NULL);			\
+   intel->batch->emit.total = (n) * 4;					\
+   intel->batch->emit.start_ptr = intel->batch->ptr;			\
 } while (0)
 
-#define OUT_BATCH(d)  intel_batchbuffer_emit_dword(intel->batch, d)
+#define OUT_BATCH(d) intel_batchbuffer_emit_dword(intel->batch, d)
 
 #define OUT_RELOC(buf, read_domains, write_domain, delta) do {		\
    assert((delta) >= 0);						\
@@ -153,7 +162,16 @@ intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
 				read_domains, write_domain, delta);	\
 } while (0)
 
-#define ADVANCE_BATCH() do { } while(0)
+#define ADVANCE_BATCH() do {						\
+   unsigned int _n = intel->batch->ptr - intel->batch->emit.start_ptr;	\
+   assert(intel->batch->emit.start_ptr != NULL);			\
+   if (_n != intel->batch->emit.total) {				\
+      fprintf(stderr, "ADVANCE_BATCH: %d of %d dwords emitted\n",	\
+	      _n, intel->batch->emit.total);				\
+      abort();								\
+   }									\
+   intel->batch->emit.start_ptr = NULL;					\
+} while(0)
 
 
 static INLINE void
diff --git a/src/mesa/drivers/dri/intel/intel_decode.c b/src/mesa/drivers/dri/intel/intel_decode.c
index 0e72ca08b22..0b8a287f6fe 100644
--- a/src/mesa/drivers/dri/intel/intel_decode.c
+++ b/src/mesa/drivers/dri/intel/intel_decode.c
@@ -836,10 +836,71 @@ get_965_depthformat(unsigned int depthformat)
     }
 }
 
+static const char *
+get_965_element_component(uint32_t data, int component)
+{
+    uint32_t component_control = (data >> (16 + (3 - component) * 4)) & 0x7;
+
+    switch (component_control) {
+    case 0:
+	return "nostore";
+    case 1:
+	switch (component) {
+	case 0: return "X";
+	case 1: return "Y";
+	case 2: return "Z";
+	case 3: return "W";
+	default: return "fail";
+	}
+    case 2:
+	return "0.0";
+    case 3:
+	return "1.0";
+    case 4:
+	return "0x1";
+    case 5:
+	return "VID";
+    default:
+	return "fail";
+    }
+}
+
+static const char *
+get_965_prim_type(uint32_t data)
+{
+    uint32_t primtype = (data >> 10) & 0x1f;
+
+    switch (primtype) {
+    case 0x01: return "point list";
+    case 0x02: return "line list";
+    case 0x03: return "line strip";
+    case 0x04: return "tri list";
+    case 0x05: return "tri strip";
+    case 0x06: return "tri fan";
+    case 0x07: return "quad list";
+    case 0x08: return "quad strip";
+    case 0x09: return "line list adj";
+    case 0x0a: return "line strip adj";
+    case 0x0b: return "tri list adj";
+    case 0x0c: return "tri strip adj";
+    case 0x0d: return "tri strip reverse";
+    case 0x0e: return "polygon";
+    case 0x0f: return "rect list";
+    case 0x10: return "line loop";
+    case 0x11: return "point list bf";
+    case 0x12: return "line strip cont";
+    case 0x13: return "line strip bf";
+    case 0x14: return "line strip cont bf";
+    case 0x15: return "tri fan no stipple";
+    default: return "fail";
+    }
+}
+
 static int
 decode_3d_965(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 {
     unsigned int opcode, len;
+    int i;
 
     struct {
 	uint32_t opcode;
@@ -860,8 +921,7 @@ decode_3d_965(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 	{ 0x780b, 1, 1, "3DSTATE_VF_STATISTICS" },
 	{ 0x7808, 5, 257, "3DSTATE_VERTEX_BUFFERS" },
 	{ 0x7809, 3, 256, "3DSTATE_VERTEX_ELEMENTS" },
-	/* 0x7808: 3DSTATE_VERTEX_BUFFERS */
-	/* 0x7809: 3DSTATE_VERTEX_ELEMENTS */
+	{ 0x780a, 3, 3, "3DSTATE_INDEX_BUFFER" },
 	{ 0x7900, 4, 4, "3DSTATE_DRAWING_RECTANGLE" },
 	{ 0x7901, 5, 5, "3DSTATE_CONSTANT_COLOR" },
 	{ 0x7905, 5, 7, "3DSTATE_DEPTH_BUFFER" },
@@ -947,6 +1007,64 @@ decode_3d_965(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 
 	return len;
 
+    case 0x7808:
+	len = (data[0] & 0xff) + 2;
+	if ((len - 1) % 4 != 0)
+	    fprintf(out, "Bad count in 3DSTATE_VERTEX_BUFFERS\n");
+	if (count < len)
+	    BUFFER_FAIL(count, len, "3DSTATE_VERTEX_BUFFERS");
+	instr_out(data, hw_offset, 0, "3DSTATE_VERTEX_BUFFERS\n");
+
+	for (i = 1; i < len;) {
+	    instr_out(data, hw_offset, i, "buffer %d: %s, pitch %db\n",
+		      data[i] >> 27,
+		      data[i] & (1 << 26) ? "random" : "sequential",
+		      data[i] & 0x07ff);
+	    i++;
+	    instr_out(data, hw_offset, i++, "buffer address\n");
+	    instr_out(data, hw_offset, i++, "max index\n");
+	    instr_out(data, hw_offset, i++, "mbz\n");
+	}
+	return len;
+
+    case 0x7809:
+	len = (data[0] & 0xff) + 2;
+	if ((len + 1) % 2 != 0)
+	    fprintf(out, "Bad count in 3DSTATE_VERTEX_ELEMENTS\n");
+	if (count < len)
+	    BUFFER_FAIL(count, len, "3DSTATE_VERTEX_ELEMENTS");
+	instr_out(data, hw_offset, 0, "3DSTATE_VERTEX_ELEMENTS\n");
+
+	for (i = 1; i < len;) {
+	    instr_out(data, hw_offset, i, "buffer %d: %svalid, type 0x%04x, "
+		      "src offset 0x%04xd bytes\n",
+		      data[i] >> 27,
+		      data[i] & (1 << 26) ? "" : "in",
+		      (data[i] >> 16) & 0x1ff,
+		      data[i] & 0x07ff);
+	    i++;
+	    instr_out(data, hw_offset, i, "(%s, %s, %s, %s), "
+		      "dst offset 0x%02x bytes\n",
+		      get_965_element_component(data[i], 0),
+		      get_965_element_component(data[i], 1),
+		      get_965_element_component(data[i], 2),
+		      get_965_element_component(data[i], 3),
+		      (data[i] & 0xff) * 4);
+	    i++;
+	}
+	return len;
+
+    case 0x780a:
+	len = (data[0] & 0xff) + 2;
+	if (len != 3)
+	    fprintf(out, "Bad count in 3DSTATE_INDEX_BUFFER\n");
+	if (count < len)
+	    BUFFER_FAIL(count, len, "3DSTATE_INDEX_BUFFER");
+	instr_out(data, hw_offset, 0, "3DSTATE_INDEX_BUFFER\n");
+	instr_out(data, hw_offset, 1, "beginning buffer address\n");
+	instr_out(data, hw_offset, 2, "ending buffer address\n");
+	return len;
+
     case 0x7900:
 	if (len != 4)
 	    fprintf(out, "Bad count in 3DSTATE_DRAWING_RECTANGLE\n");
@@ -968,9 +1086,9 @@ decode_3d_965(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 	return len;
 
     case 0x7905:
-	if (len != 5)
+	if (len != 5 && len != 6)
 	    fprintf(out, "Bad count in 3DSTATE_DEPTH_BUFFER\n");
-	if (count < 5)
+	if (count < len)
 	    BUFFER_FAIL(count, len, "3DSTATE_DEPTH_BUFFER");
 
 	instr_out(data, hw_offset, 0,
@@ -985,7 +1103,27 @@ decode_3d_965(uint32_t *data, int count, uint32_t hw_offset, int *failures)
 		  ((data[3] & 0x0007ffc0) >> 6) + 1,
 		  ((data[3] & 0xfff80000) >> 19) + 1);
 	instr_out(data, hw_offset, 4, "volume depth\n");
+	if (len == 6)
+	    instr_out(data, hw_offset, 5, "\n");
+
+	return len;
 
+    case 0x7b00:
+	len = (data[0] & 0xff) + 2;
+	if (len != 6)
+	    fprintf(out, "Bad count in 3DPRIMITIVE\n");
+	if (count < len)
+	    BUFFER_FAIL(count, len, "3DPRIMITIVE");
+
+	instr_out(data, hw_offset, 0,
+		  "3DPRIMITIVE: %s %s\n",
+		  get_965_prim_type(data[0]),
+		  (data[0] & (1 << 15)) ? "random" : "sequential");
+	instr_out(data, hw_offset, 1, "primitive count\n");
+	instr_out(data, hw_offset, 2, "start vertex\n");
+	instr_out(data, hw_offset, 3, "instance count\n");
+	instr_out(data, hw_offset, 4, "start instance\n");
+	instr_out(data, hw_offset, 5, "index bias\n");
 	return len;
     }
 
diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
index c677ddd63c5..bf1c3f03f0e 100644
--- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
+++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
@@ -163,7 +163,7 @@ intel_miptree_create_for_region(struct intel_context *intel,
    mt->pitch = region->pitch;
 #endif
 
-   mt->region = region;
+   intel_region_reference(&mt->region, region);
 
    return mt;
  }