eliminate rhw divide under some circumstances

3 files changed, 32 insertions, 18 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_vs.c b/src/mesa/drivers/dri/i965/brw_vs.c
index 2a94ac64965..e5a28b96e32 100644
--- a/src/mesa/drivers/dri/i965/brw_vs.c
+++ b/src/mesa/drivers/dri/i965/brw_vs.c
@@ -105,6 +105,11 @@ static void brw_upload_vs_prog( struct brw_context *brw )
    key.copy_edgeflag = (brw->attribs.Polygon->FrontMode != GL_FILL ||
 			brw->attribs.Polygon->BackMode != GL_FILL);
 
+   /* BRW_NEW_METAOPS
+    */
+   if (brw->metaops.active)
+      key.know_w_is_one = 1;
+
    /* Make an early check for the key.
     */
    if (brw_search_cache(&brw->cache[BRW_VS_PROG], 
@@ -122,7 +127,7 @@ static void brw_upload_vs_prog( struct brw_context *brw )
 const struct brw_tracked_state brw_vs_prog = {
    .dirty = {
       .mesa  = _NEW_TRANSFORM | _NEW_POLYGON,
-      .brw   = BRW_NEW_VERTEX_PROGRAM,
+      .brw   = BRW_NEW_VERTEX_PROGRAM | BRW_NEW_METAOPS,
       .cache = 0
    },
    .update = brw_upload_vs_prog
diff --git a/src/mesa/drivers/dri/i965/brw_vs.h b/src/mesa/drivers/dri/i965/brw_vs.h
index d355681b5ea..fdb5785d67d 100644
--- a/src/mesa/drivers/dri/i965/brw_vs.h
+++ b/src/mesa/drivers/dri/i965/brw_vs.h
@@ -43,7 +43,8 @@ struct brw_vs_prog_key {
    GLuint program_string_id;
    GLuint nr_userclip:4;
    GLuint copy_edgeflag:1;
-   GLuint pad:27;
+   GLuint know_w_is_one:1;
+   GLuint pad:26;
 };
 
 
diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c
index a22740084d3..da9d3bacb0e 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c
@@ -797,13 +797,21 @@ static void emit_vertex_write( struct brw_vs_compile *c)
 
    /* Build ndc coords?   TODO: Shortcircuit when w is known to be one.
     */
-   ndc = get_tmp(c);
-   emit_math1(c, BRW_MATH_FUNCTION_INV, ndc, brw_swizzle1(pos, 3), BRW_MATH_PRECISION_FULL);
-   brw_MUL(p, brw_writemask(ndc, WRITEMASK_XYZ), pos, ndc);
+   if (!c->key.know_w_is_one) {
+      ndc = get_tmp(c);
+      emit_math1(c, BRW_MATH_FUNCTION_INV, ndc, brw_swizzle1(pos, 3), BRW_MATH_PRECISION_FULL);
+      brw_MUL(p, brw_writemask(ndc, WRITEMASK_XYZ), pos, ndc);
+   }
+   else {
+      ndc = pos;
+   }
 
    /* This includes the workaround for -ve rhw, so is no longer an
     * optional step:
     */
+   if ((c->prog_data.outputs_written & (1<<VERT_RESULT_PSIZ)) ||
+       c->key.nr_userclip ||
+       !c->key.know_w_is_one)
    {
       struct brw_reg header1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
       GLuint i;
@@ -836,20 +844,17 @@ static void emit_vertex_write( struct brw_vs_compile *c)
        * Later, clipping will detect ucp[6] and ensure the primitive is
        * clipped against all fixed planes.
        */
-      brw_CMP(p,
-	      vec8(brw_null_reg()),
-	      BRW_CONDITIONAL_L,
-	      brw_swizzle1(ndc, 3),
-	      brw_imm_f(0));
+      if (!c->key.know_w_is_one) {
+	 brw_CMP(p,
+		 vec8(brw_null_reg()),
+		 BRW_CONDITIONAL_L,
+		 brw_swizzle1(ndc, 3),
+		 brw_imm_f(0));
    
-      brw_OR(p, brw_writemask(header1, WRITEMASK_W), header1, brw_imm_ud(1<<6));
-      brw_MOV(p, ndc, brw_imm_f(0));
-      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
-
-
-
-
-
+	 brw_OR(p, brw_writemask(header1, WRITEMASK_W), header1, brw_imm_ud(1<<6));
+	 brw_MOV(p, ndc, brw_imm_f(0));
+	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
+      }
 
       brw_set_access_mode(p, BRW_ALIGN_1);	/* why? */
       brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), header1);
@@ -857,6 +862,9 @@ static void emit_vertex_write( struct brw_vs_compile *c)
 
       release_tmp(c, header1);
    }
+   else {
+      brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), brw_imm_ud(0));
+   }
 
 
    /* Emit the (interleaved) headers for the two vertices - an 8-reg