Code reorganization: move files into their places.
[mesa.git] / src / gallium / drivers / i965simple / brw_clip_tri.c
diff --git a/src/gallium/drivers/i965simple/brw_clip_tri.c b/src/gallium/drivers/i965simple/brw_clip_tri.c
new file mode 100644 (file)
index 0000000..c5da7b8
--- /dev/null
@@ -0,0 +1,566 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice (including the
+ next paragraph) shall be included in all copies or substantial
+ portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#include "brw_defines.h"
+#include "brw_context.h"
+#include "brw_eu.h"
+#include "brw_util.h"
+#include "brw_clip.h"
+
+static struct brw_reg get_tmp( struct brw_clip_compile *c )
+{
+   struct brw_reg tmp = brw_vec4_grf(c->last_tmp, 0);
+
+   if (++c->last_tmp > c->prog_data.total_grf)
+      c->prog_data.total_grf = c->last_tmp;
+
+   return tmp;
+}
+
+static void release_tmps( struct brw_clip_compile *c )
+{
+   c->last_tmp = c->first_tmp;
+}
+
+
+void brw_clip_tri_alloc_regs( struct brw_clip_compile *c,
+                             unsigned nr_verts )
+{
+   unsigned i = 0,j;
+
+   /* Register usage is static, precompute here:
+    */
+   c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++;
+
+   if (c->key.nr_userclip) {
+      c->reg.fixed_planes = brw_vec4_grf(i, 0);
+      i += (6 + c->key.nr_userclip + 1) / 2;
+
+      c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2;
+   }
+   else
+      c->prog_data.curb_read_length = 0;
+
+
+   /* Payload vertices plus space for more generated vertices:
+    */
+   for (j = 0; j < nr_verts; j++) {
+      c->reg.vertex[j] = brw_vec4_grf(i, 0);
+      i += c->nr_regs;
+   }
+
+   if (c->nr_attrs & 1) {
+      for (j = 0; j < 3; j++) {
+        unsigned delta = c->nr_attrs*16 + 32;
+        brw_MOV(&c->func, byte_offset(c->reg.vertex[j], delta), brw_imm_f(0));
+      }
+   }
+
+   c->reg.t          = brw_vec1_grf(i, 0);
+   c->reg.loopcount  = retype(brw_vec1_grf(i, 1), BRW_REGISTER_TYPE_UD);
+   c->reg.nr_verts   = retype(brw_vec1_grf(i, 2), BRW_REGISTER_TYPE_UD);
+   c->reg.planemask  = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD);
+   c->reg.plane_equation = brw_vec4_grf(i, 4);
+   i++;
+
+   c->reg.dpPrev     = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */
+   c->reg.dp         = brw_vec1_grf(i, 4);
+   i++;
+
+   c->reg.inlist     = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
+   i++;
+
+   c->reg.outlist    = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
+   i++;
+
+   c->reg.freelist   = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, i, 0);
+   i++;
+
+   if (!c->key.nr_userclip) {
+      c->reg.fixed_planes = brw_vec8_grf(i, 0);
+      i++;
+   }
+
+   if (c->key.do_unfilled) {
+      c->reg.dir     = brw_vec4_grf(i, 0);
+      c->reg.offset  = brw_vec4_grf(i, 4);
+      i++;
+      c->reg.tmp0    = brw_vec4_grf(i, 0);
+      c->reg.tmp1    = brw_vec4_grf(i, 4);
+      i++;
+   }
+
+   c->first_tmp = i;
+   c->last_tmp = i;
+
+   c->prog_data.urb_read_length = c->nr_regs; /* ? */
+   c->prog_data.total_grf = i;
+}
+
+
+
+void brw_clip_tri_init_vertices( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
+   struct brw_instruction *is_rev;
+
+   /* Initial list of indices for incoming vertexes:
+    */
+   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
+   brw_CMP(p,
+          vec1(brw_null_reg()),
+          BRW_CONDITIONAL_EQ,
+          tmp0,
+          brw_imm_ud(_3DPRIM_TRISTRIP_REVERSE));
+
+   /* XXX: Is there an easier way to do this?  Need to reverse every
+    * second tristrip element:  Can ignore sometimes?
+    */
+   is_rev = brw_IF(p, BRW_EXECUTE_1);
+   {
+      brw_MOV(p, get_element(c->reg.inlist, 0),  brw_address(c->reg.vertex[1]) );
+      brw_MOV(p, get_element(c->reg.inlist, 1),  brw_address(c->reg.vertex[0]) );
+      if (c->need_direction)
+        brw_MOV(p, c->reg.dir, brw_imm_f(-1));
+   }
+   is_rev = brw_ELSE(p, is_rev);
+   {
+      brw_MOV(p, get_element(c->reg.inlist, 0),  brw_address(c->reg.vertex[0]) );
+      brw_MOV(p, get_element(c->reg.inlist, 1),  brw_address(c->reg.vertex[1]) );
+      if (c->need_direction)
+        brw_MOV(p, c->reg.dir, brw_imm_f(1));
+   }
+   brw_ENDIF(p, is_rev);
+
+   brw_MOV(p, get_element(c->reg.inlist, 2),  brw_address(c->reg.vertex[2]) );
+   brw_MOV(p, brw_vec8_grf(c->reg.outlist.nr, 0), brw_imm_f(0));
+   brw_MOV(p, c->reg.nr_verts, brw_imm_ud(3));
+}
+
+
+
+void brw_clip_tri_flat_shade( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *is_poly;
+   struct brw_reg tmp0 = c->reg.loopcount; /* handy temporary */
+
+   brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
+   brw_CMP(p,
+          vec1(brw_null_reg()),
+          BRW_CONDITIONAL_EQ,
+          tmp0,
+          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);
+   }
+   brw_ENDIF(p, is_poly);
+}
+
+
+
+/* Use mesa's clipping algorithms, translated to GEN4 assembly.
+ */
+void brw_clip_tri( struct brw_clip_compile *c )
+{
+#if 0
+   struct brw_compile *p = &c->func;
+   struct brw_indirect vtx = brw_indirect(0, 0);
+   struct brw_indirect vtxPrev = brw_indirect(1, 0);
+   struct brw_indirect vtxOut = brw_indirect(2, 0);
+   struct brw_indirect plane_ptr = brw_indirect(3, 0);
+   struct brw_indirect inlist_ptr = brw_indirect(4, 0);
+   struct brw_indirect outlist_ptr = brw_indirect(5, 0);
+   struct brw_indirect freelist_ptr = brw_indirect(6, 0);
+   struct brw_instruction *plane_loop;
+   struct brw_instruction *plane_active;
+   struct brw_instruction *vertex_loop;
+   struct brw_instruction *next_test;
+   struct brw_instruction *prev_test;
+
+   brw_MOV(p, get_addr_reg(vtxPrev),     brw_address(c->reg.vertex[2]) );
+   brw_MOV(p, get_addr_reg(plane_ptr),   brw_clip_plane0_address(c));
+   brw_MOV(p, get_addr_reg(inlist_ptr),  brw_address(c->reg.inlist));
+   brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
+
+   brw_MOV(p, get_addr_reg(freelist_ptr), brw_address(c->reg.vertex[3]) );
+
+   plane_loop = brw_DO(p, BRW_EXECUTE_1);
+   {
+      /* if (planemask & 1)
+       */
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_AND(p, vec1(brw_null_reg()), c->reg.planemask, brw_imm_ud(1));
+
+      plane_active = brw_IF(p, BRW_EXECUTE_1);
+      {
+        /* vtxOut = freelist_ptr++
+         */
+        brw_MOV(p, get_addr_reg(vtxOut),       get_addr_reg(freelist_ptr) );
+        brw_ADD(p, get_addr_reg(freelist_ptr), get_addr_reg(freelist_ptr), brw_imm_uw(c->nr_regs * REG_SIZE));
+
+        if (c->key.nr_userclip)
+           brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
+        else
+           brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
+
+        brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
+        brw_MOV(p, c->reg.nr_verts, brw_imm_ud(0));
+
+        vertex_loop = brw_DO(p, BRW_EXECUTE_1);
+        {
+           /* vtx = *input_ptr;
+            */
+           brw_MOV(p, get_addr_reg(vtx), deref_1uw(inlist_ptr, 0));
+
+           /* IS_NEGATIVE(prev) */
+           brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
+           brw_DP4(p, vec4(c->reg.dpPrev), deref_4f(vtxPrev, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+           prev_test = brw_IF(p, BRW_EXECUTE_1);
+           {
+              /* IS_POSITIVE(next)
+               */
+              brw_set_conditionalmod(p, BRW_CONDITIONAL_GE);
+              brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+              next_test = brw_IF(p, BRW_EXECUTE_1);
+              {
+
+                 /* Coming back in.
+                  */
+                 brw_ADD(p, c->reg.t, c->reg.dpPrev, negate(c->reg.dp));
+                 brw_math_invert(p, c->reg.t, c->reg.t);
+                 brw_MUL(p, c->reg.t, c->reg.t, c->reg.dpPrev);
+
+                 /* If (vtxOut == 0) vtxOut = vtxPrev
+                  */
+                 brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
+                 brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtxPrev) );
+                 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+                 brw_clip_interp_vertex(c, vtxOut, vtxPrev, vtx, c->reg.t, FALSE);
+
+                 /* *outlist_ptr++ = vtxOut;
+                  * nr_verts++;
+                  * vtxOut = 0;
+                  */
+                 brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
+                 brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
+                 brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
+                 brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
+              }
+              brw_ENDIF(p, next_test);
+
+           }
+           prev_test = brw_ELSE(p, prev_test);
+           {
+              /* *outlist_ptr++ = vtxPrev;
+               * nr_verts++;
+               */
+              brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxPrev));
+              brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
+              brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
+
+              /* IS_NEGATIVE(next)
+               */
+              brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
+              brw_DP4(p, vec4(c->reg.dp), deref_4f(vtx, c->offset[VERT_RESULT_HPOS]), c->reg.plane_equation);
+              next_test = brw_IF(p, BRW_EXECUTE_1);
+              {
+                 /* Going out of bounds.  Avoid division by zero as we
+                  * know dp != dpPrev from DIFFERENT_SIGNS, above.
+                  */
+                 brw_ADD(p, c->reg.t, c->reg.dp, negate(c->reg.dpPrev));
+                 brw_math_invert(p, c->reg.t, c->reg.t);
+                 brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp);
+
+                 /* If (vtxOut == 0) vtxOut = vtx
+                  */
+                 brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_EQ, get_addr_reg(vtxOut), brw_imm_uw(0) );
+                 brw_MOV(p, get_addr_reg(vtxOut), get_addr_reg(vtx) );
+                 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+                 brw_clip_interp_vertex(c, vtxOut, vtx, vtxPrev, c->reg.t, TRUE);
+
+                 /* *outlist_ptr++ = vtxOut;
+                  * nr_verts++;
+                  * vtxOut = 0;
+                  */
+                 brw_MOV(p, deref_1uw(outlist_ptr, 0), get_addr_reg(vtxOut));
+                 brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_uw(sizeof(short)));
+                 brw_ADD(p, c->reg.nr_verts, c->reg.nr_verts, brw_imm_ud(1));
+                 brw_MOV(p, get_addr_reg(vtxOut), brw_imm_uw(0) );
+              }
+              brw_ENDIF(p, next_test);
+           }
+           brw_ENDIF(p, prev_test);
+
+           /* vtxPrev = vtx;
+            * inlist_ptr++;
+            */
+           brw_MOV(p, get_addr_reg(vtxPrev), get_addr_reg(vtx));
+           brw_ADD(p, get_addr_reg(inlist_ptr), get_addr_reg(inlist_ptr), brw_imm_uw(sizeof(short)));
+
+           /* while (--loopcount != 0)
+            */
+           brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+           brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
+        }
+        brw_WHILE(p, vertex_loop);
+
+        /* vtxPrev = *(outlist_ptr-1)  OR: outlist[nr_verts-1]
+         * inlist = outlist
+         * inlist_ptr = &inlist[0]
+         * outlist_ptr = &outlist[0]
+         */
+        brw_ADD(p, get_addr_reg(outlist_ptr), get_addr_reg(outlist_ptr), brw_imm_w(-2));
+        brw_MOV(p, get_addr_reg(vtxPrev), deref_1uw(outlist_ptr, 0));
+        brw_MOV(p, brw_vec8_grf(c->reg.inlist.nr, 0), brw_vec8_grf(c->reg.outlist.nr, 0));
+        brw_MOV(p, get_addr_reg(inlist_ptr), brw_address(c->reg.inlist));
+        brw_MOV(p, get_addr_reg(outlist_ptr), brw_address(c->reg.outlist));
+      }
+      brw_ENDIF(p, plane_active);
+
+      /* plane_ptr++;
+       */
+      brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
+
+      /* nr_verts >= 3
+       */
+      brw_CMP(p,
+             vec1(brw_null_reg()),
+             BRW_CONDITIONAL_GE,
+             c->reg.nr_verts,
+             brw_imm_ud(3));
+
+      /* && (planemask>>=1) != 0
+       */
+      brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+      brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
+   }
+   brw_WHILE(p, plane_loop);
+#else
+         #warning "disabled"
+#endif
+}
+
+
+
+void brw_clip_tri_emit_polygon(struct brw_clip_compile *c)
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *loop, *if_insn;
+
+   /* for (loopcount = nr_verts-2; loopcount > 0; loopcount--)
+    */
+   brw_set_conditionalmod(p, BRW_CONDITIONAL_G);
+   brw_ADD(p,
+          c->reg.loopcount,
+          c->reg.nr_verts,
+          brw_imm_d(-2));
+
+   if_insn = brw_IF(p, BRW_EXECUTE_1);
+   {
+      struct brw_indirect v0 = brw_indirect(0, 0);
+      struct brw_indirect vptr = brw_indirect(1, 0);
+
+      brw_MOV(p, get_addr_reg(vptr), brw_address(c->reg.inlist));
+      brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
+
+      brw_clip_emit_vue(c, v0, 1, 0, ((_3DPRIM_TRIFAN << 2) | R02_PRIM_START));
+
+      brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2));
+      brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
+
+      loop = brw_DO(p, BRW_EXECUTE_1);
+      {
+        brw_clip_emit_vue(c, v0, 1, 0, (_3DPRIM_TRIFAN << 2));
+
+        brw_ADD(p, get_addr_reg(vptr), get_addr_reg(vptr), brw_imm_uw(2));
+        brw_MOV(p, get_addr_reg(v0), deref_1uw(vptr, 0));
+
+        brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+        brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
+      }
+      brw_WHILE(p, loop);
+
+      brw_clip_emit_vue(c, v0, 0, 1, ((_3DPRIM_TRIFAN << 2) | R02_PRIM_END));
+   }
+   brw_ENDIF(p, if_insn);
+}
+
+static void do_clip_tri( struct brw_clip_compile *c )
+{
+   brw_clip_init_planes(c);
+
+   brw_clip_tri(c);
+}
+
+
+static void maybe_do_clip_tri( struct brw_clip_compile *c )
+{
+   struct brw_compile *p = &c->func;
+   struct brw_instruction *do_clip;
+
+   brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0));
+   do_clip = brw_IF(p, BRW_EXECUTE_1);
+   {
+      do_clip_tri(c);
+   }
+   brw_ENDIF(p, do_clip);
+}
+
+static void brw_clip_test( struct brw_clip_compile *c )
+{
+#if 0
+    struct brw_reg t = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+    struct brw_reg t1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+    struct brw_reg t2 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+    struct brw_reg t3 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
+
+    struct brw_reg v0 = get_tmp(c);
+    struct brw_reg v1 = get_tmp(c);
+    struct brw_reg v2 = get_tmp(c);
+
+    struct brw_indirect vt0 = brw_indirect(0, 0);
+    struct brw_indirect vt1 = brw_indirect(1, 0);
+    struct brw_indirect vt2 = brw_indirect(2, 0);
+
+    struct brw_compile *p = &c->func;
+
+    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]));
+    brw_MOV(p, get_addr_reg(vt2), brw_address(c->reg.vertex[2]));
+    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]));
+
+    /* test nearz, xmin, ymin plane */
+    brw_CMP(p, t1, BRW_CONDITIONAL_LE, negate(v0), 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_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, t3, BRW_CONDITIONAL_LE, negate(v2), get_element(v2, 3));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    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_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_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_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));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+    brw_CMP(p, t2, BRW_CONDITIONAL_L, 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_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    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_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_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_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud((1<<0)));
+    brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+
+    release_tmps(c);
+#else
+         #warning "disabled"
+#endif
+}
+
+
+void brw_emit_tri_clip( struct brw_clip_compile *c )
+{
+   struct brw_instruction *neg_rhw;
+   struct brw_compile *p = &c->func;
+   brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6);
+   brw_clip_tri_init_vertices(c);
+   brw_clip_init_clipmask(c);
+
+   /* if -ve rhw workaround bit is set,
+      do cliptest */
+   brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
+   brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2),
+          brw_imm_ud(1<<20));
+   neg_rhw = brw_IF(p, BRW_EXECUTE_1);
+   {
+       brw_clip_test(c);
+   }
+   brw_ENDIF(p, neg_rhw);
+
+   /* Can't push into do_clip_tri because with polygon (or quad)
+    * flatshading, need to apply the flatshade here because we don't
+    * respect the PV when converting to trifan for emit:
+    */
+   if (c->key.do_flat_shading)
+      brw_clip_tri_flat_shade(c);
+
+   if (c->key.clip_mode == BRW_CLIPMODE_NORMAL)
+      do_clip_tri(c);
+   else
+      maybe_do_clip_tri(c);
+
+   brw_clip_tri_emit_polygon(c);
+
+   /* Send an empty message to kill the thread:
+    */
+   brw_clip_kill_thread(c);
+}