sync up t_vp_build.c brw_vs_tnl.c a bit

Bring over the optimizations for fog and normalized spot dir
from t_vp_build.c to brw_vs_tnl.c. Likewise, port a fix for point size
calc from brw_vs_tnl.c to t_vp_build.c (use ABS(eyez) instead of -eyez). Leave
the now differing point size calcs alone though, not sure what's better (it's
basically MOV, ABS, MUL, DP3 vs. ABS, MAD, MAD).

diff --git a/src/mesa/drivers/dri/i965/brw_vs_tnl.c b/src/mesa/drivers/dri/i965/brw_vs_tnl.c
index c05a9b5ea10e85af8f28608aaa129310ead68f7d..e22a26b291741ef7c33664f542009ea6f17ea110 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vs_tnl.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_tnl.c
@@ -850,14 +850,13 @@ static struct ureg calculate_light_attenuation( struct tnl_program *p,
    /* Calculate spot attenuation:
     */
    if (!p->state->unit[i].light_spotcutoff_is_180) {
-      struct ureg spot_dir = register_param3(p, STATE_LIGHT, i,
-                                            STATE_SPOT_DIRECTION);
+      struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
+                                                 STATE_SPOT_DIR_NORMALIZED, i);
       struct ureg spot = get_temp(p);
       struct ureg slt = get_temp(p);
-              
-      emit_normalize_vec3( p, spot, spot_dir ); /* XXX: precompute! */
-      emit_op2(p, OPCODE_DP3, spot, 0, ureg_negate(VPpli), spot);
-      emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir,W), spot);
+
+      emit_op2(p, OPCODE_DP3, spot, 0, ureg_negate(VPpli), spot_dir_norm);
+      emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
 
@@ -1166,7 +1165,8 @@ static void build_fog( struct tnl_program *p )
 
    if (p->state->fog_option &&
        p->state->tnl_do_vertex_fog) {
-      struct ureg params = register_param1(p, STATE_FOG_PARAMS);
+      struct ureg params = register_param2(p, STATE_INTERNAL,
+                                          STATE_FOG_PARAMS_OPTIMIZED);
       struct ureg tmp = get_temp(p);
       struct ureg id = get_identity_param(p);
 
@@ -1174,8 +1174,7 @@ static void build_fog( struct tnl_program *p )
 
       switch (p->state->fog_option) {
       case FOG_LINEAR: {
-        emit_op1(p, OPCODE_ABS, tmp, 0, input); 
-        emit_op2(p, OPCODE_SUB, tmp, 0, swizzle1(params,Z), tmp); 
+        emit_op3(p, OPCODE_MAD, tmp, 0, input, swizzle1(params,X), swizzle1(params,Y));
         emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,W)); 
         emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
         emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
@@ -1183,15 +1182,13 @@ static void build_fog( struct tnl_program *p )
       }
       case FOG_EXP:
         emit_op1(p, OPCODE_ABS, tmp, 0, input); 
-        emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,X)); 
-        emit_op2(p, OPCODE_POW, fog, WRITEMASK_X, 
-                 register_const1f(p, M_E), ureg_negate(tmp)); 
+        emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,Z));
+        emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp));
         break;
       case FOG_EXP2:
-        emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,X)); 
+        emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
         emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp); 
-        emit_op2(p, OPCODE_POW, fog, WRITEMASK_X, 
-                 register_const1f(p, M_E), ureg_negate(tmp)); 
+        emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp));
         break;
       }
       

diff --git a/src/mesa/tnl/t_vp_build.c b/src/mesa/tnl/t_vp_build.c
index 2a97a1c8c76693ea7633f5429292943f1446a533..2663d993049ffe77f820c27c680f311a4ed4b389 100644 (file)
--- a/src/mesa/tnl/t_vp_build.c
+++ b/src/mesa/tnl/t_vp_build.c
@@ -1329,14 +1329,16 @@ static void build_pointsize( struct tnl_program *p )
    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
    struct ureg ut = get_temp(p);
 
+   /* dist = |eyez| */
+   emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
    /* p1 + dist * (p2 + dist * p3); */
-   emit_op3(p, OPCODE_MAD, ut, 0, negate(swizzle1(eye, Z)),
+   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
                swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
-   emit_op3(p, OPCODE_MAD, ut, 0, negate(swizzle1(eye, Z)),
+   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
                ut, swizzle1(state_attenuation, X));
 
    /* 1 / sqrt(factor) */
-   emit_op1(p, OPCODE_RSQ, ut, 0, ut );
+   emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
 
 #if 1
    /* out = pointSize / sqrt(factor) */
@@ -1344,8 +1346,8 @@ static void build_pointsize( struct tnl_program *p )
 #else
    /* not sure, might make sense to do clamping here,
       but it's not done in t_vb_points neither */
-   emit_op2(p, OPCODE_MUL, ut, 0, ut, state_size);
-   emit_op2(p, OPCODE_MAX, ut, 0, ut, swizzle1(state_size, Y));
+   emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
+   emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
    emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
 #endif