mesa: more complete fix for transform_invarient glitches

Add a new flag mvp_with_dp4 in the context, and use that to switch
both ffvertex.c and programopt.c vertex transformation code to
either DP4 or MUL/MAD implementations.

4 files changed, 30 insertions, 9 deletions

diff --git a/src/mesa/main/context.c b/src/mesa/main/context.c
index 016284de9aa..d780f91f048 100644
--- a/src/mesa/main/context.c
+++ b/src/mesa/main/context.c
@@ -1522,4 +1522,17 @@ _mesa_Flush(void)
 }
 
 
+/**
+ * Set mvp_with_dp4 flag.  If a driver has a preference for DP4 over
+ * MUL/MAD, or vice versa, call this function to register that.
+ * Otherwise we default to MUL/MAD.
+ */
+void
+_mesa_set_mvp_with_dp4( GLcontext *ctx,
+                        GLboolean flag )
+{
+   ctx->mvp_with_dp4 = flag;
+}
+
+
 /*@}*/
diff --git a/src/mesa/main/context.h b/src/mesa/main/context.h
index ecc1cec7799..5b57d88029c 100644
--- a/src/mesa/main/context.h
+++ b/src/mesa/main/context.h
@@ -151,6 +151,10 @@ extern struct _glapi_table *
 _mesa_get_dispatch(GLcontext *ctx);
 
 
+void
+_mesa_set_mvp_with_dp4( GLcontext *ctx,
+                        GLboolean flag );
+
 
 /** \name Miscellaneous */
 /*@{*/
diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c
index 82e1c4af665..43325b13529 100644
--- a/src/mesa/main/ffvertex_prog.c
+++ b/src/mesa/main/ffvertex_prog.c
@@ -315,12 +315,6 @@ static void make_state_key( GLcontext *ctx, struct state_key *key )
  */
 #define DISASSEM 0
 
-/* Should be tunable by the driver - do we want to do matrix
- * multiplications with DP4's or with MUL/MAD's?  SSE works better
- * with the latter, drivers may differ.
- */
-#define PREFER_DP4 1
-
 
 /* Use uregs to represent registers internally, translate to Mesa's
  * expected formats on emit.  
@@ -348,6 +342,7 @@ struct tnl_program {
    const struct state_key *state;
    struct gl_vertex_program *program;
    GLint max_inst;  /** number of instructions allocated for program */
+   GLboolean mvp_with_dp4;
    
    GLuint temp_in_use;
    GLuint temp_reserved;
@@ -775,7 +770,7 @@ static struct ureg get_eye_position( struct tnl_program *p )
 
       p->eye_position = reserve_temp(p);
 
-      if (PREFER_DP4) {
+      if (p->mvp_with_dp4) {
 	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
                                  0, modelview );
 
@@ -881,7 +876,7 @@ static void build_hpos( struct tnl_program *p )
    struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
    struct ureg mvp[4];
 
-   if (PREFER_DP4) {
+   if (p->mvp_with_dp4) {
       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3, 
 			      0, mvp );
       emit_matrix_transform_vec4( p, hpos, mvp, pos );
@@ -1574,7 +1569,7 @@ static void build_texture_transform( struct tnl_program *p )
 	    struct ureg in = (!is_undef(out_texgen) ? 
 			      out_texgen : 
 			      register_input(p, VERT_ATTRIB_TEX0+i));
-	    if (PREFER_DP4) {
+	    if (p->mvp_with_dp4) {
 	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
 				       0, texmat );
 	       emit_matrix_transform_vec4( p, out, texmat, in );
@@ -1708,6 +1703,7 @@ static void build_tnl_program( struct tnl_program *p )
 static void
 create_new_program( const struct state_key *key,
                     struct gl_vertex_program *program,
+                    GLboolean mvp_with_dp4,
                     GLuint max_temps)
 {
    struct tnl_program p;
@@ -1721,6 +1717,7 @@ create_new_program( const struct state_key *key,
    p.transformed_normal = undef;
    p.identity = undef;
    p.temp_in_use = 0;
+   p.mvp_with_dp4 = mvp_with_dp4;
    
    if (max_temps >= sizeof(int) * 8)
       p.temp_reserved = 0;
@@ -1776,6 +1773,7 @@ _mesa_get_fixed_func_vertex_program(GLcontext *ctx)
          return NULL;
 
       create_new_program( &key, prog,
+                          ctx->mvp_with_dp4,
                           ctx->Const.VertexProgram.MaxTemps );
 
 #if 0
diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
index cf710777286..587dc801466 100644
--- a/src/mesa/main/mtypes.h
+++ b/src/mesa/main/mtypes.h
@@ -2980,6 +2980,12 @@ struct __GLcontextRec
    /** software compression/decompression supported or not */
    GLboolean Mesa_DXTn;
 
+   /** 
+    * Use dp4 (rather than mul/mad) instructions for position
+    * transformation?
+    */
+   GLboolean mvp_with_dp4;
+
    /** Core tnl module support */
    struct gl_tnl_module TnlModule;