i965: fix for RHW workaround

It is possible that an object whose vertices all are outside of a
view plane is passed to clip thread due to the RHW workaround. This
object should be rejected by clip thread. Fix bug #19879

2 files changed, 99 insertions, 43 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_clip_line.c b/src/mesa/drivers/dri/i965/brw_clip_line.c
index c45d48dff8e..d830e49e50a 100644
--- a/src/mesa/drivers/dri/i965/brw_clip_line.c
+++ b/src/mesa/drivers/dri/i965/brw_clip_line.c
@@ -181,34 +181,54 @@ static void clip_and_emit_line( struct brw_clip_compile *c )
 	 brw_DP4(p, vec4(c->reg.dp1), deref_4f(vtx1, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
 	 is_negative = brw_IF(p, BRW_EXECUTE_1);
 	 {
-	    brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0));
-	    brw_math_invert(p, c->reg.t, c->reg.t);
-	    brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1);
-
-	    brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 );
-	    brw_MOV(p, c->reg.t1, c->reg.t);
-	    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+             /*
+              * Both can be negative on GM965/G965 due to RHW workaround
+              * if so, this object should be rejected.
+              */
+             if (!BRW_IS_G4X(p->brw)) {
+                 brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE, c->reg.dp0, brw_imm_f(0.0));
+                 is_neg2 = brw_IF(p, BRW_EXECUTE_1);
+                 {
+                     brw_clip_kill_thread(c);
+                 }
+                 brw_ENDIF(p, is_neg2);
+             }
+
+             brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0));
+             brw_math_invert(p, c->reg.t, c->reg.t);
+             brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1);
+
+             brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 );
+             brw_MOV(p, c->reg.t1, c->reg.t);
+             brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 	 } 
 	 is_negative = brw_ELSE(p, is_negative);
 	 {
-	    /* Coming back in.  We know that both cannot be negative
-	     * because the line would have been culled in that case.
-	     */
+             /* Coming back in.  We know that both cannot be negative
+              * because the line would have been culled in that case.
+              */
+
+             /* If both are positive, do nothing */
+             /* Only on GM965/G965 */
+             if (!BRW_IS_G4X(p->brw)) {
+                 brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0));
+                 is_neg2 = brw_IF(p, BRW_EXECUTE_1);
+             }
 
-	    /* If both are positive, do nothing */
-             brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0));
-             is_neg2 = brw_IF(p, BRW_EXECUTE_1);
              {
-		brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1));
-		brw_math_invert(p, c->reg.t, c->reg.t);
-		brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0);
-
-		brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 );
-		brw_MOV(p, c->reg.t0, c->reg.t);
-		brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-	     }
-	     brw_ENDIF(p, is_neg2);
-	 }
+                 brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1));
+                 brw_math_invert(p, c->reg.t, c->reg.t);
+                 brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0);
+
+                 brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 );
+                 brw_MOV(p, c->reg.t0, c->reg.t);
+                 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+             }
+
+             if (!BRW_IS_G4X(p->brw)) {
+                 brw_ENDIF(p, is_neg2);
+             }
+         }
 	 brw_ENDIF(p, is_negative);	 
       }
       brw_ENDIF(p, plane_active);
diff --git a/src/mesa/drivers/dri/i965/brw_clip_tri.c b/src/mesa/drivers/dri/i965/brw_clip_tri.c
index 1dbba37fe7e..7fd37bd05ff 100644
--- a/src/mesa/drivers/dri/i965/brw_clip_tri.c
+++ b/src/mesa/drivers/dri/i965/brw_clip_tri.c
@@ -455,6 +455,8 @@ static void brw_clip_test( struct brw_clip_compile *c )
     struct brw_indirect vt2 = brw_indirect(2, 0);
 
     struct brw_compile *p = &c->func;
+    struct brw_instruction *is_outside;
+    struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
 
     brw_MOV(p, get_addr_reg(vt0), brw_address(c->reg.vertex[0]));
     brw_MOV(p, get_addr_reg(vt1), brw_address(c->reg.vertex[1]));
@@ -462,53 +464,87 @@ static void brw_clip_test( struct brw_clip_compile *c )
     brw_MOV(p, v0, deref_4f(vt0, c->offset[VERT_RESULT_HPOS]));
     brw_MOV(p, v1, deref_4f(vt1, c->offset[VERT_RESULT_HPOS]));
     brw_MOV(p, v2, deref_4f(vt2, c->offset[VERT_RESULT_HPOS]));
+    brw_AND(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(~0x3f));
 
     /* test nearz, xmin, ymin plane */
-    brw_CMP(p, t1, BRW_CONDITIONAL_LE, negate(v0), get_element(v0, 3)); 
+    /* clip.xyz < -clip.w */
+    brw_CMP(p, t1, BRW_CONDITIONAL_L, v0, negate(get_element(v0, 3))); 
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, t2, BRW_CONDITIONAL_LE, negate(v1), get_element(v1, 3)); 
+    brw_CMP(p, t2, BRW_CONDITIONAL_L, v1, negate(get_element(v1, 3))); 
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, t3, BRW_CONDITIONAL_LE, negate(v2), get_element(v2, 3)); 
+    brw_CMP(p, t3, BRW_CONDITIONAL_L, v2, negate(get_element(v2, 3))); 
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    /* All vertices are outside of a plane, rejected */
+    brw_AND(p, t, t1, t2);
+    brw_AND(p, t, t, t3);
+    brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1));
+    brw_OR(p, tmp0, tmp0, get_element(t, 2));
+    brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+    brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1));
+    is_outside = brw_IF(p, BRW_EXECUTE_1);
+    {
+        brw_clip_kill_thread(c);
+    }
+    brw_ENDIF(p, is_outside);
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    /* some vertices are inside a plane, some are outside,need to clip */
     brw_XOR(p, t, t1, t2);
     brw_XOR(p, t1, t2, t3);
     brw_OR(p, t, t, t1);
-
-    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, 
-	    get_element(t, 0), brw_imm_ud(0));
+    brw_AND(p, t, t, brw_imm_ud(0x1));
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+            get_element(t, 0), brw_imm_ud(0));
     brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<5)));
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, 
-	    get_element(t, 1), brw_imm_ud(0));
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+            get_element(t, 1), brw_imm_ud(0));
     brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<3)));
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, 
-	    get_element(t, 2), brw_imm_ud(0));
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+            get_element(t, 2), brw_imm_ud(0));
     brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<1)));
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 
     /* test farz, xmax, ymax plane */
-    brw_CMP(p, t1, BRW_CONDITIONAL_L, v0, get_element(v0, 3)); 
+    /* clip.xyz > clip.w */
+    brw_CMP(p, t1, BRW_CONDITIONAL_G, v0, get_element(v0, 3)); 
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, t2, BRW_CONDITIONAL_L, v1, get_element(v1, 3)); 
+    brw_CMP(p, t2, BRW_CONDITIONAL_G, v1, get_element(v1, 3)); 
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, t3, BRW_CONDITIONAL_L, v2, get_element(v2, 3)); 
+    brw_CMP(p, t3, BRW_CONDITIONAL_G, v2, get_element(v2, 3)); 
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
 
+    /* All vertices are outside of a plane, rejected */
+    brw_AND(p, t, t1, t2);
+    brw_AND(p, t, t, t3);
+    brw_OR(p, tmp0, get_element(t, 0), get_element(t, 1));
+    brw_OR(p, tmp0, tmp0, get_element(t, 2));
+    brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+    brw_AND(p, brw_null_reg(), tmp0, brw_imm_ud(0x1));
+    is_outside = brw_IF(p, BRW_EXECUTE_1);
+    {
+        brw_clip_kill_thread(c);
+    }
+    brw_ENDIF(p, is_outside);
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    /* some vertices are inside a plane, some are outside,need to clip */
     brw_XOR(p, t, t1, t2);
     brw_XOR(p, t1, t2, t3);
     brw_OR(p, t, t, t1);
-
-    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, 
-	    get_element(t, 0), brw_imm_ud(0));
+    brw_AND(p, t, t, brw_imm_ud(0x1));
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+            get_element(t, 0), brw_imm_ud(0));
     brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<4)));
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, 
-	    get_element(t, 1), brw_imm_ud(0));
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+            get_element(t, 1), brw_imm_ud(0));
     brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<2)));
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ, 
-	    get_element(t, 2), brw_imm_ud(0));
+    brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_NZ,
+            get_element(t, 2), brw_imm_ud(0));
     brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<0)));
     brw_set_predicate_control(p, BRW_PREDICATE_NONE);