i965: fix EXT_provoking_vertex support

This didn't work for quad/quadstrips at all, and for all other primitive types
it only worked when they were unclipped.
Fix up the former in gs stage (could probably do without these changes and
instead set QuadsFollowProvokingVertexConvention to false), and the rest in
clip stage.

8 files changed, 61 insertions, 29 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_clip.c b/src/mesa/drivers/dri/i965/brw_clip.c
index 20a927cf386..f45dcf82820 100644
--- a/src/mesa/drivers/dri/i965/brw_clip.c
+++ b/src/mesa/drivers/dri/i965/brw_clip.c
@@ -156,6 +156,7 @@ static void upload_clip_prog(struct brw_context *brw)
    key.attrs = brw->vs.prog_data->outputs_written;
    /* _NEW_LIGHT */
    key.do_flat_shading = (ctx->Light.ShadeModel == GL_FLAT);
+   key.pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
    /* _NEW_TRANSFORM */
    key.nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
 
diff --git a/src/mesa/drivers/dri/i965/brw_clip.h b/src/mesa/drivers/dri/i965/brw_clip.h
index 957df441ab0..dc550ac7933 100644
--- a/src/mesa/drivers/dri/i965/brw_clip.h
+++ b/src/mesa/drivers/dri/i965/brw_clip.h
@@ -46,18 +46,17 @@ struct brw_clip_prog_key {
    GLuint primitive:4;
    GLuint nr_userclip:3;
    GLuint do_flat_shading:1;
+   GLuint pv_first:1;
    GLuint do_unfilled:1;
    GLuint fill_cw:2;		/* includes cull information */
    GLuint fill_ccw:2;		/* includes cull information */
    GLuint offset_cw:1;
    GLuint offset_ccw:1;
-   GLuint pad0:17;
-
    GLuint copy_bfc_cw:1;
    GLuint copy_bfc_ccw:1;
    GLuint clip_mode:3;
-   GLuint pad1:27;
-   
+   GLuint pad0:11;
+
    GLfloat offset_factor;
    GLfloat offset_units;
 };
diff --git a/src/mesa/drivers/dri/i965/brw_clip_line.c b/src/mesa/drivers/dri/i965/brw_clip_line.c
index 048ca620fab..fa9648f50f3 100644
--- a/src/mesa/drivers/dri/i965/brw_clip_line.c
+++ b/src/mesa/drivers/dri/i965/brw_clip_line.c
@@ -269,8 +269,12 @@ void brw_emit_line_clip( struct brw_clip_compile *c )
    brw_clip_line_alloc_regs(c);
    brw_clip_init_ff_sync(c);
 
-   if (c->key.do_flat_shading)
-      brw_clip_copy_colors(c, 0, 1);
+   if (c->key.do_flat_shading) {
+      if (c->key.pv_first)
+         brw_clip_copy_colors(c, 1, 0);
+      else
+         brw_clip_copy_colors(c, 0, 1);
+   }
                 
    clip_and_emit_line(c);
 }
diff --git a/src/mesa/drivers/dri/i965/brw_clip_tri.c b/src/mesa/drivers/dri/i965/brw_clip_tri.c
index 0efd77225e2..cf79224be43 100644
--- a/src/mesa/drivers/dri/i965/brw_clip_tri.c
+++ b/src/mesa/drivers/dri/i965/brw_clip_tri.c
@@ -188,14 +188,20 @@ void brw_clip_tri_flat_shade( struct brw_clip_compile *c )
 	   brw_imm_ud(_3DPRIM_POLYGON));
 
    is_poly = brw_IF(p, BRW_EXECUTE_1);
-   {   
+   {
       brw_clip_copy_colors(c, 1, 0);
       brw_clip_copy_colors(c, 2, 0);
    }
    is_poly = brw_ELSE(p, is_poly);
    {
-      brw_clip_copy_colors(c, 0, 2);
-      brw_clip_copy_colors(c, 1, 2);
+      if (c->key.pv_first) {
+         brw_clip_copy_colors(c, 1, 0);
+         brw_clip_copy_colors(c, 2, 0);
+      }
+      else {
+         brw_clip_copy_colors(c, 0, 2);
+         brw_clip_copy_colors(c, 1, 2);
+      }
    }
    brw_ENDIF(p, is_poly);
 }
diff --git a/src/mesa/drivers/dri/i965/brw_gs.c b/src/mesa/drivers/dri/i965/brw_gs.c
index 48c2b9a41ce..610b6c35e2a 100644
--- a/src/mesa/drivers/dri/i965/brw_gs.c
+++ b/src/mesa/drivers/dri/i965/brw_gs.c
@@ -85,10 +85,10 @@ static void compile_gs_prog( struct brw_context *brw,
     */
    switch (key->primitive) {
    case GL_QUADS:
-      brw_gs_quads( &c ); 
+      brw_gs_quads( &c, key );
       break;
    case GL_QUAD_STRIP:
-      brw_gs_quad_strip( &c );
+      brw_gs_quad_strip( &c, key );
       break;
    case GL_LINE_LOOP:
       brw_gs_lines( &c );
@@ -149,6 +149,7 @@ static const GLenum gs_prim[GL_POLYGON+1] = {
 static void populate_key( struct brw_context *brw,
 			  struct brw_gs_prog_key *key )
 {
+   GLcontext *ctx = &brw->intel.ctx;
    memset(key, 0, sizeof(*key));
 
    /* CACHE_NEW_VS_PROG */
@@ -158,6 +159,9 @@ static void populate_key( struct brw_context *brw,
    key->primitive = gs_prim[brw->primitive];
 
    key->hint_gs_always = 0;	/* debug code? */
+   
+   /* _NEW_LIGHT */
+   key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
 
    key->need_gs_prog = (key->hint_gs_always ||
 			brw->primitive == GL_QUADS ||
@@ -193,7 +197,7 @@ static void prepare_gs_prog(struct brw_context *brw)
 
 const struct brw_tracked_state brw_gs_prog = {
    .dirty = {
-      .mesa  = 0,
+      .mesa  = _NEW_LIGHT,
       .brw   = BRW_NEW_PRIMITIVE,
       .cache = CACHE_NEW_VS_PROG
    },
diff --git a/src/mesa/drivers/dri/i965/brw_gs.h b/src/mesa/drivers/dri/i965/brw_gs.h
index bbb991ea2e5..e0cf07256bc 100644
--- a/src/mesa/drivers/dri/i965/brw_gs.h
+++ b/src/mesa/drivers/dri/i965/brw_gs.h
@@ -43,8 +43,9 @@ struct brw_gs_prog_key {
    GLuint attrs:32;
    GLuint primitive:4;
    GLuint hint_gs_always:1;
+   GLuint pv_first:1;
    GLuint need_gs_prog:1;
-   GLuint pad:26;
+   GLuint pad:25;
 };
 
 struct brw_gs_compile {
@@ -67,8 +68,8 @@ struct brw_gs_compile {
 
 #define ATTR_SIZE  (4*4)
 
-void brw_gs_quads( struct brw_gs_compile *c );
-void brw_gs_quad_strip( struct brw_gs_compile *c );
+void brw_gs_quads( struct brw_gs_compile *c, struct brw_gs_prog_key *key );
+void brw_gs_quad_strip( struct brw_gs_compile *c, struct brw_gs_prog_key *key );
 void brw_gs_tris( struct brw_gs_compile *c );
 void brw_gs_lines( struct brw_gs_compile *c );
 void brw_gs_points( struct brw_gs_compile *c );
diff --git a/src/mesa/drivers/dri/i965/brw_gs_emit.c b/src/mesa/drivers/dri/i965/brw_gs_emit.c
index a9b2aa2eace..0fc5b02c619 100644
--- a/src/mesa/drivers/dri/i965/brw_gs_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_gs_emit.c
@@ -120,7 +120,7 @@ static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim)
 }
 
 
-void brw_gs_quads( struct brw_gs_compile *c )
+void brw_gs_quads( struct brw_gs_compile *c, struct brw_gs_prog_key *key )
 {
    brw_gs_alloc_regs(c, 4);
    
@@ -128,23 +128,39 @@ void brw_gs_quads( struct brw_gs_compile *c )
     * is the PV for quads, but vertex 0 for polygons:
     */
    if (c->need_ff_sync)
-	   brw_gs_ff_sync(c, 1);    
-   brw_gs_emit_vue(c, c->reg.vertex[3], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
-   brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2));
-   brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2)); 
-   brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+	   brw_gs_ff_sync(c, 1);
+   if (key->pv_first) {
+      brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
+      brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[2], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[3], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+   }
+   else {
+      brw_gs_emit_vue(c, c->reg.vertex[3], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
+      brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[2], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+   }
 }
 
-void brw_gs_quad_strip( struct brw_gs_compile *c )
+void brw_gs_quad_strip( struct brw_gs_compile *c, struct brw_gs_prog_key *key )
 {
    brw_gs_alloc_regs(c, 4);
    
    if (c->need_ff_sync)
 	   brw_gs_ff_sync(c, 1);      
-   brw_gs_emit_vue(c, c->reg.vertex[2], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
-   brw_gs_emit_vue(c, c->reg.vertex[3], 0, (_3DPRIM_POLYGON << 2));
-   brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2)); 
-   brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+   if (key->pv_first) {
+      brw_gs_emit_vue(c, c->reg.vertex[0], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
+      brw_gs_emit_vue(c, c->reg.vertex[1], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[2], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[3], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+   }
+   else {
+      brw_gs_emit_vue(c, c->reg.vertex[2], 0, ((_3DPRIM_POLYGON << 2) | R02_PRIM_START));
+      brw_gs_emit_vue(c, c->reg.vertex[3], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[0], 0, (_3DPRIM_POLYGON << 2));
+      brw_gs_emit_vue(c, c->reg.vertex[1], 1, ((_3DPRIM_POLYGON << 2) | R02_PRIM_END));
+   }
 }
 
 void brw_gs_tris( struct brw_gs_compile *c )
diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c
index bc0f0760738..79f37097d36 100644
--- a/src/mesa/drivers/dri/i965/brw_sf_state.c
+++ b/src/mesa/drivers/dri/i965/brw_sf_state.c
@@ -113,7 +113,8 @@ struct brw_sf_unit_key {
 
    unsigned int nr_urb_entries, urb_size, sfsize;
 
-   GLenum front_face, cull_face, provoking_vertex;
+   GLenum front_face, cull_face;
+   unsigned pv_first:1;
    unsigned scissor:1;
    unsigned line_smooth:1;
    unsigned point_sprite:1;
@@ -154,7 +155,7 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key)
    key->point_attenuated = ctx->Point._Attenuated;
 
    /* _NEW_LIGHT */
-   key->provoking_vertex = ctx->Light.ProvokingVertex;
+   key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
 
    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 }
@@ -287,7 +288,7 @@ sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
 
    /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
     */
-   if (key->provoking_vertex == GL_LAST_VERTEX_CONVENTION) {
+   if (!key->pv_first) {
       sf.sf7.trifan_pv = 2;
       sf.sf7.linestrip_pv = 1;
       sf.sf7.tristrip_pv = 2;