Merge branch 'master' into r500test

Conflicts:

	src/mesa/drivers/dri/r300/r300_cmdbuf.c
	src/mesa/drivers/dri/r300/r300_ioctl.c

11 files changed, 96 insertions, 47 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_clip.c b/src/mesa/drivers/dri/i965/brw_clip.c
index 24b0288f88..ce34da165c 100644
--- a/src/mesa/drivers/dri/i965/brw_clip.c
+++ b/src/mesa/drivers/dri/i965/brw_clip.c
@@ -168,12 +168,10 @@ static void upload_clip_prog( struct brw_context *brw )
 	       offset_front = 0;
 	       break;
 	    case GL_LINE:
-	       key.do_unfilled = 1;
 	       fill_front = CLIP_LINE;
 	       offset_front = brw->attribs.Polygon->OffsetLine;
 	       break;
 	    case GL_POINT:
-	       key.do_unfilled = 1;
 	       fill_front = CLIP_POINT;
 	       offset_front = brw->attribs.Polygon->OffsetPoint;
 	       break;
@@ -188,26 +186,23 @@ static void upload_clip_prog( struct brw_context *brw )
 	       offset_back = 0;
 	       break;
 	    case GL_LINE:
-	       key.do_unfilled = 1;
 	       fill_back = CLIP_LINE;
 	       offset_back = brw->attribs.Polygon->OffsetLine;
 	       break;
 	    case GL_POINT:
-	       key.do_unfilled = 1;
 	       fill_back = CLIP_POINT;
 	       offset_back = brw->attribs.Polygon->OffsetPoint;
 	       break;
 	    }
 	 }
 
-    if (brw->attribs.Polygon->BackMode != GL_FILL ||
-        brw->attribs.Polygon->FrontMode != GL_FILL)
-        key.do_unfilled = 1;
+	 if (brw->attribs.Polygon->BackMode != GL_FILL ||
+	     brw->attribs.Polygon->FrontMode != GL_FILL) {
+	    key.do_unfilled = 1;
 
-	 /* Most cases the fixed function units will handle.  Cases where
-	  * one or more polygon faces are unfilled will require help:
-	  */
-	 if (key.do_unfilled) {
+	    /* Most cases the fixed function units will handle.  Cases where
+	     * one or more polygon faces are unfilled will require help:
+	     */
 	    key.clip_mode = BRW_CLIPMODE_CLIP_NON_REJECTED;
 
 	    if (offset_back || offset_front) {
diff --git a/src/mesa/drivers/dri/i965/brw_clip_unfilled.c b/src/mesa/drivers/dri/i965/brw_clip_unfilled.c
index 918e000187..57ebf388f5 100644
--- a/src/mesa/drivers/dri/i965/brw_clip_unfilled.c
+++ b/src/mesa/drivers/dri/i965/brw_clip_unfilled.c
@@ -220,8 +220,8 @@ static void apply_one_offset( struct brw_clip_compile *c,
 			  struct brw_indirect vert )
 {
    struct brw_compile *p = &c->func;
-   struct brw_reg pos = deref_4f(vert, c->offset[VERT_RESULT_HPOS]);
-   struct brw_reg z = get_element(pos, 2);
+   struct brw_reg z = deref_1f(vert, c->header_position_offset +
+			       2 * type_sz(BRW_REGISTER_TYPE_F));
 
    brw_ADD(p, z, z, vec1(c->reg.offset));
 }
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 4e6b471fdd..56021fa209 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -146,12 +146,12 @@ struct brw_context;
 struct brw_state_flags {
    /** State update flags signalled by mesa internals */
    GLuint mesa;
-   /** State update flags signalled by brw_state_cache.c searches */
-   GLuint cache;
    /**
     * State update flags signalled as the result of brw_tracked_state updates
     */
    GLuint brw;
+   /** State update flags signalled by brw_state_cache.c searches */
+   GLuint cache;
 };
 
 struct brw_vertex_program {
diff --git a/src/mesa/drivers/dri/i965/brw_program.c b/src/mesa/drivers/dri/i965/brw_program.c
index 8e8fea48e9..f717b6f6c3 100644
--- a/src/mesa/drivers/dri/i965/brw_program.c
+++ b/src/mesa/drivers/dri/i965/brw_program.c
@@ -124,6 +124,9 @@ static void brwProgramStringNotify( GLcontext *ctx,
       struct brw_vertex_program *vp = (struct brw_vertex_program *)brw->vertex_program;
       if (p == vp)
 	 brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
+      if (p->program.IsPositionInvariant) {
+	 _mesa_insert_mvp_code(ctx, &p->program);
+      }
       p->id = brw->program_id++;      
       p->param_state = p->program.Base.Parameters->StateFlags;
 
diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c
index 2b6087d691..398048429b 100644
--- a/src/mesa/drivers/dri/i965/brw_sf_state.c
+++ b/src/mesa/drivers/dri/i965/brw_sf_state.c
@@ -224,11 +224,11 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
        sf.sf6.line_width = 0;
 
    /* _NEW_POINT */
-   sf.sf6.point_rast_rule = 1;	/* opengl conventions */
+   sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT;	/* opengl conventions */
    /* XXX clamp max depends on AA vs. non-AA */
 
    sf.sf7.sprite_point = key->point_sprite;
-   sf.sf7.point_size = CLAMP(key->point_size, 1.0, 255.0) * (1<<3);
+   sf.sf7.point_size = CLAMP(nearbyint(key->point_size), 1, 255) * (1<<3);
    sf.sf7.use_point_size_state = !key->point_attenuated;
    sf.sf7.aa_line_distance_mode = 0;
 
diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c
index 7b5eff4f2d..d617650fad 100644
--- a/src/mesa/drivers/dri/i965/brw_state_cache.c
+++ b/src/mesa/drivers/dri/i965/brw_state_cache.c
@@ -85,7 +85,7 @@ static GLuint hash_key( const void *key, GLuint key_size,
    }
 
    /* Include the BO pointers as key data as well */
-   ikey = (void *)reloc_bufs;
+   ikey = (GLuint *)reloc_bufs;
    key_size = nr_reloc_bufs * sizeof(dri_bo *);
    for (i = 0; i < key_size/4; i++) {
       hash ^= ikey[i];
@@ -102,6 +102,9 @@ static void
 update_cache_last(struct brw_cache *cache, enum brw_cache_id cache_id,
 		  dri_bo *bo)
 {
+   if (bo == cache->last_bo[cache_id])
+      return; /* no change */
+
    dri_bo_unreference(cache->last_bo[cache_id]);
    cache->last_bo[cache_id] = bo;
    dri_bo_reference(cache->last_bo[cache_id]);
@@ -255,7 +258,7 @@ brw_upload_cache( struct brw_cache *cache,
    if (INTEL_DEBUG & DEBUG_STATE)
       _mesa_printf("upload %s: %d bytes to cache id %d\n",
 		   cache->name[cache_id],
-		   data_size);
+		   data_size, cache_id);
 
    /* Copy data to the buffer */
    dri_bo_subdata(bo, 0, data_size, data);
@@ -282,6 +285,7 @@ brw_cache_data_sz(struct brw_cache *cache,
    item = search_cache(cache, cache_id, hash, data, data_size,
 		       reloc_bufs, nr_reloc_bufs);
    if (item) {
+      update_cache_last(cache, cache_id, item->bo);
       dri_bo_reference(item->bo);
       return item->bo;
    }
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index 5ca2a634c0..297617ee2d 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -199,6 +199,7 @@ struct brw_wm_compile {
    GLuint nr_fp_insns;
    GLuint fp_temp;
    GLuint fp_interp_emitted;
+   GLuint fp_fragcolor_emitted;
    GLuint fp_deriv_emitted;
 
    struct prog_src_register pixel_xy;
diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c
index 83a0d12d90..a02f70a50c 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c
@@ -137,7 +137,7 @@ static void emit_wpos_xy(struct brw_wm_compile *c,
       brw_ADD(p,
 	      dst[0],
 	      retype(arg0[0], BRW_REGISTER_TYPE_W),
-	      brw_imm_d(- c->key.origin_x));
+	      brw_imm_d(0 - c->key.origin_x));
    }
 
    if (mask & WRITEMASK_Y) {
@@ -223,6 +223,10 @@ static void emit_pinterp( struct brw_compile *p,
       if (mask & (1<<i)) {
 	 brw_LINE(p, brw_null_reg(), interp[i], deltas[0]);
 	 brw_MAC(p, dst[i], suboffset(interp[i],1), deltas[1]);
+      }
+   }
+   for(i = 0; i < 4; i++ ) {
+      if (mask & (1<<i)) {
 	 brw_MUL(p, dst[i], dst[i], w[3]);
       }
    }
@@ -500,6 +504,9 @@ static void emit_dp3( struct brw_compile *p,
 		      const struct brw_reg *arg0,
 		      const struct brw_reg *arg1 )
 {
+   if (!(mask & WRITEMASK_XYZW))
+      return; /* Do not emit dead code*/
+
    assert((mask & WRITEMASK_XYZW) == WRITEMASK_X);
 
    brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]);
@@ -517,6 +524,9 @@ static void emit_dp4( struct brw_compile *p,
 		      const struct brw_reg *arg0,
 		      const struct brw_reg *arg1 )
 {
+   if (!(mask & WRITEMASK_XYZW))
+      return; /* Do not emit dead code*/
+
    assert((mask & WRITEMASK_XYZW) == WRITEMASK_X);
 
    brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]);
@@ -535,6 +545,9 @@ static void emit_dph( struct brw_compile *p,
 		      const struct brw_reg *arg0,
 		      const struct brw_reg *arg1 )
 {
+   if (!(mask & WRITEMASK_XYZW))
+      return; /* Do not emit dead code*/
+
    assert((mask & WRITEMASK_XYZW) == WRITEMASK_X);
 
    brw_MUL(p, brw_null_reg(), arg0[0], arg1[0]);
@@ -578,6 +591,9 @@ static void emit_math1( struct brw_compile *p,
 			GLuint mask,
 			const struct brw_reg *arg0 )
 {
+   if (!(mask & WRITEMASK_XYZW))
+      return; /* Do not emit dead code*/
+
    //assert((mask & WRITEMASK_XYZW) == WRITEMASK_X ||
    //	  function == BRW_MATH_FUNCTION_SINCOS);
    
@@ -602,6 +618,9 @@ static void emit_math2( struct brw_compile *p,
 			const struct brw_reg *arg0,
 			const struct brw_reg *arg1)
 {
+   if (!(mask & WRITEMASK_XYZW))
+      return; /* Do not emit dead code*/
+
    assert((mask & WRITEMASK_XYZW) == WRITEMASK_X);
 
    brw_push_insn_state(p);
diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c
index 682b16d74d..7e80724130 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_fp.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c
@@ -144,7 +144,7 @@ static struct prog_dst_register dst_undef( void )
 
 static struct prog_dst_register get_temp( struct brw_wm_compile *c )
 {
-   int bit = ffs( ~c->fp_temp );
+   int bit = _mesa_ffs( ~c->fp_temp );
 
    if (!bit) {
       _mesa_printf("%s: out of temporaries\n", __FILE__);
@@ -158,7 +158,7 @@ static struct prog_dst_register get_temp( struct brw_wm_compile *c )
 
 static void release_temp( struct brw_wm_compile *c, struct prog_dst_register temp )
 {
-   c->fp_temp &= ~1<<(temp.Index + 1 - FIRST_INTERNAL_TEMP);
+   c->fp_temp &= ~(1 << (temp.Index - FIRST_INTERNAL_TEMP));
 }
 
 
@@ -494,17 +494,20 @@ static void precalc_dst( struct brw_wm_compile *c,
 
 
    if (dst.WriteMask & WRITEMASK_XZ) {
+      struct prog_instruction *swz;
       GLuint z = GET_SWZ(src0.Swizzle, Z);
 
       /* dst.xz = swz src0.1zzz
        */
-      emit_op(c,
-	      OPCODE_SWZ,
-	      dst_mask(dst, WRITEMASK_XZ),
-	      inst->SaturateMode, 0, 0,
-	      src_swizzle(src0, SWIZZLE_ONE, z, z, z),
-	      src_undef(),
-	      src_undef());
+      swz = emit_op(c,
+		    OPCODE_SWZ,
+		    dst_mask(dst, WRITEMASK_XZ),
+		    inst->SaturateMode, 0, 0,
+		    src_swizzle(src0, SWIZZLE_ONE, z, z, z),
+		    src_undef(),
+		    src_undef());
+      /* Avoid letting negation flag of src0 affect our 1 constant. */
+      swz->SrcReg[0].NegateBase &= ~NEGATE_X;
    }
    if (dst.WriteMask & WRITEMASK_W) {
       /* dst.w = mov src1.w
@@ -527,15 +530,19 @@ static void precalc_lit( struct brw_wm_compile *c,
    struct prog_dst_register dst = inst->DstReg;
    
    if (dst.WriteMask & WRITEMASK_XW) {
+      struct prog_instruction *swz;
+
       /* dst.xw = swz src0.1111
        */
-      emit_op(c,
-	      OPCODE_SWZ,
-	      dst_mask(dst, WRITEMASK_XW),
-	      0, 0, 0,
-	      src_swizzle1(src0, SWIZZLE_ONE),
-	      src_undef(),
-	      src_undef());
+      swz = emit_op(c,
+		    OPCODE_SWZ,
+		    dst_mask(dst, WRITEMASK_XW),
+		    0, 0, 0,
+		    src_swizzle1(src0, SWIZZLE_ONE),
+		    src_undef(),
+		    src_undef());
+      /* Avoid letting the negation flag of src0 affect our 1 constant. */
+      swz->SrcReg[0].NegateBase = 0;
    }
 
 
@@ -863,25 +870,31 @@ static void emit_fb_write( struct brw_wm_compile *c )
    struct prog_src_register outdepth = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DEPR);
    GLuint i;
 
-   struct prog_instruction *inst;
+   struct prog_instruction *inst, *last_inst;
    struct brw_context *brw = c->func.brw;
 
    /* inst->Sampler is not used by backend, 
       use it for fb write target and eot */
 
-   inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0),
-           0, 0, 0, outcolor, payload_r0_depth, outdepth);
-   inst->Sampler = (brw->state.nr_draw_regions > 1 ? 0: 1)|(0<<1);
-
    if (brw->state.nr_draw_regions > 1) {
        for (i = 0 ; i < brw->state.nr_draw_regions; i++) {
 	   outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_DATA0 + i);
-	   inst = emit_op(c,
+	   last_inst = inst = emit_op(c,
 		   WM_FB_WRITE, dst_mask(dst_undef(),0), 0, 0, 0,
 		   outcolor, payload_r0_depth, outdepth);
-	   inst->Sampler = ((i == brw->state.nr_draw_regions - 1) ? 1: 0);
-	   inst->Sampler |= (i<<1);
+	   inst->Sampler = (i<<1);
+	   if (c->fp_fragcolor_emitted) {
+	       outcolor = src_reg(PROGRAM_OUTPUT, FRAG_RESULT_COLR);
+	       last_inst = inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0),
+		       0, 0, 0, outcolor, payload_r0_depth, outdepth);
+	       inst->Sampler = (i<<1);
+	   }
        }
+       last_inst->Sampler |= 1; //eot
+   }else {
+       inst = emit_op(c, WM_FB_WRITE, dst_mask(dst_undef(),0),
+	       0, 0, 0, outcolor, payload_r0_depth, outdepth);
+       inst->Sampler = 1|(0<<1);
    }
 }
 
@@ -908,7 +921,15 @@ static void validate_src_regs( struct brw_wm_compile *c,
    }
 }
 	 
-
+static void validate_dst_regs( struct brw_wm_compile *c,
+			       const struct prog_instruction *inst )
+{
+   if (inst->DstReg.File == PROGRAM_OUTPUT) {
+       GLuint idx = inst->DstReg.Index;
+       if (idx == FRAG_RESULT_COLR)
+	   c->fp_fragcolor_emitted = 1;
+   }
+}
 
 static void print_insns( const struct prog_instruction *insn,
 			 GLuint nr )
@@ -953,12 +974,16 @@ void brw_wm_pass_fp( struct brw_wm_compile *c )
 
    for (insn = 0; insn < fp->program.Base.NumInstructions; insn++) {
       const struct prog_instruction *inst = &fp->program.Base.Instructions[insn];
+      validate_src_regs(c, inst);
+      validate_dst_regs(c, inst);
+   }
+   for (insn = 0; insn < fp->program.Base.NumInstructions; insn++) {
+      const struct prog_instruction *inst = &fp->program.Base.Instructions[insn];
       struct prog_instruction *out;
 
       /* Check for INPUT values, emit INTERP instructions where
        * necessary:
        */
-      validate_src_regs(c, inst);
 
 
       switch (inst->Opcode) {
diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass1.c b/src/mesa/drivers/dri/i965/brw_wm_pass1.c
index 26c044d400..f6f3a38e9e 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_pass1.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_pass1.c
@@ -150,6 +150,7 @@ void brw_wm_pass1( struct brw_wm_compile *c )
       case OPCODE_FLR:
       case OPCODE_FRC:
       case OPCODE_MOV:
+      case OPCODE_SWZ:
 	 read0 = writemask;
 	 break;
 
@@ -257,7 +258,6 @@ void brw_wm_pass1( struct brw_wm_compile *c )
 	 read0 = WRITEMASK_XYW;
 	 break;
 
-      case OPCODE_SWZ:
       case OPCODE_DST:
       case OPCODE_TXP:
       default:
diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
index 86566df388..c5c944f781 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
@@ -257,6 +257,8 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region,
       GLboolean tiled, color_blend;
    } key;
 
+   memset(&key, 0, sizeof(key));
+
    if (region != NULL) {
       region_bo = region->buffer;