i965: Fix GS hang on Sandybridge

Don't use r0 for FF_SYNC dest reg on Sandybridge, which would
smash FFID field in GS payload, that cause later URB write fail.
Also not use r0 in any URB write requiring allocate.

4 files changed, 46 insertions, 14 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
index f9aa5f7efef..41286382d0e 100644
--- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
@@ -1848,7 +1848,8 @@ void brw_ff_sync(struct brw_compile *p,
    if (intel->gen >= 6) {
       brw_push_insn_state(p);
       brw_set_mask_control( p, BRW_MASK_DISABLE );
-      brw_MOV(p, brw_message_reg(msg_reg_nr), src0);
+      brw_MOV(p, retype(brw_message_reg(msg_reg_nr), BRW_REGISTER_TYPE_UD),
+	      retype(src0, BRW_REGISTER_TYPE_UD));
       brw_pop_insn_state(p);
       src0 = brw_message_reg(msg_reg_nr);
    }
diff --git a/src/mesa/drivers/dri/i965/brw_gs.c b/src/mesa/drivers/dri/i965/brw_gs.c
index ad178d84b65..cfcc8ea4d6a 100644
--- a/src/mesa/drivers/dri/i965/brw_gs.c
+++ b/src/mesa/drivers/dri/i965/brw_gs.c
@@ -96,8 +96,6 @@ static void compile_gs_prog( struct brw_context *brw,
       brw_gs_quad_strip( &c, key );
       break;
    case GL_LINE_LOOP:
-      /* XXX fix GS hang issue */
-      assert(intel->gen < 6);
       brw_gs_lines( &c );
       break;
    case GL_LINES:
diff --git a/src/mesa/drivers/dri/i965/brw_gs.h b/src/mesa/drivers/dri/i965/brw_gs.h
index 813b8d447ae..7e3531086f9 100644
--- a/src/mesa/drivers/dri/i965/brw_gs.h
+++ b/src/mesa/drivers/dri/i965/brw_gs.h
@@ -56,6 +56,7 @@ struct brw_gs_compile {
    struct {
       struct brw_reg R0;
       struct brw_reg vertex[MAX_GS_VERTS];
+      struct brw_reg temp;
    } reg;
 
    /* 3 different ways of expressing vertex size:
diff --git a/src/mesa/drivers/dri/i965/brw_gs_emit.c b/src/mesa/drivers/dri/i965/brw_gs_emit.c
index a01d5576f8c..e1f751fdaa4 100644
--- a/src/mesa/drivers/dri/i965/brw_gs_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_gs_emit.c
@@ -58,6 +58,8 @@ static void brw_gs_alloc_regs( struct brw_gs_compile *c,
       i += c->nr_regs;
    }
 
+   c->reg.temp = brw_vec8_grf(i, 0);
+
    c->prog_data.urb_read_length = c->nr_regs; 
    c->prog_data.total_grf = i;
 }
@@ -69,12 +71,22 @@ static void brw_gs_emit_vue(struct brw_gs_compile *c,
 			    GLuint header)
 {
    struct brw_compile *p = &c->func;
+   struct intel_context *intel = &c->func.brw->intel;
    GLboolean allocate = !last;
+   struct brw_reg temp;
+
+   if (intel->gen < 6)
+       temp = c->reg.R0;
+   else {
+       temp = c->reg.temp;
+       brw_MOV(p, retype(temp, BRW_REGISTER_TYPE_UD),
+	       retype(c->reg.R0, BRW_REGISTER_TYPE_UD));
+   }
 
    /* Overwrite PrimType and PrimStart in the message header, for
     * each vertex in turn:
     */
-   brw_MOV(p, get_element_ud(c->reg.R0, 2), brw_imm_ud(header));
+   brw_MOV(p, get_element_ud(temp, 2), brw_imm_ud(header));
 
    /* Copy the vertex from vertn into m1..mN+1:
     */
@@ -87,9 +99,9 @@ static void brw_gs_emit_vue(struct brw_gs_compile *c,
     * allocated each time.
     */
    brw_urb_WRITE(p, 
-		 allocate ? c->reg.R0 : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
+		 allocate ? temp : retype(brw_null_reg(), BRW_REGISTER_TYPE_UD),
 		 0,
-		 c->reg.R0,
+		 temp,
 		 allocate,
 		 1,		/* used */
 		 c->nr_regs + 1, /* msg length */
@@ -98,19 +110,39 @@ static void brw_gs_emit_vue(struct brw_gs_compile *c,
 		 1,		/* writes_complete */
 		 0,		/* urb offset */
 		 BRW_URB_SWIZZLE_NONE);
+
+   if (intel->gen >= 6 && allocate)
+       brw_MOV(p, get_element_ud(c->reg.R0, 0), get_element_ud(temp, 0));
 }
 
 static void brw_gs_ff_sync(struct brw_gs_compile *c, int num_prim)
 {
 	struct brw_compile *p = &c->func;
-	brw_MOV(p, get_element_ud(c->reg.R0, 1), brw_imm_ud(num_prim));
-	brw_ff_sync(p,
-		    c->reg.R0,
-		    0,
-		    c->reg.R0,
-		    1, /* allocate */
-		    1, /* response length */
-		    0 /* eot */);
+	struct intel_context *intel = &c->func.brw->intel;
+
+	if (intel->gen < 6) {
+	    brw_MOV(p, get_element_ud(c->reg.R0, 1), brw_imm_ud(num_prim));
+	    brw_ff_sync(p,
+		        c->reg.R0,
+			0,
+			c->reg.R0,
+			1, /* allocate */
+			1, /* response length */
+			0 /* eot */);
+	} else {
+	    brw_MOV(p, retype(c->reg.temp, BRW_REGISTER_TYPE_UD),
+		    retype(c->reg.R0, BRW_REGISTER_TYPE_UD));
+	    brw_MOV(p, get_element_ud(c->reg.temp, 1), brw_imm_ud(num_prim));
+	    brw_ff_sync(p,
+		        c->reg.temp,
+			0,
+			c->reg.temp,
+			1, /* allocate */
+			1, /* response length */
+			0 /* eot */);
+	    brw_MOV(p, get_element_ud(c->reg.R0, 0),
+		    get_element_ud(c->reg.temp, 0));
+	}
 }