vbo: Use a bitmask to track the active arrays in vbo_exec*.

The use of a bitmask makes functions iterating only active
attributes less visible in profiles.

v2: Use _mesa_bit_scan{,64} instead of open coding.
v3: Use u_bit_scan{,64} instead of _mesa_bit_scan{,64}.

Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>

src/mesa/vbo/vbo_exec.h

@@ -101,6 +101,7 @@ struct vbo_exec_context
       GLuint max_vert;     /**< Max number of vertices allowed in buffer */
       struct vbo_exec_copied_vtx copied;
 
+      GLbitfield64 enabled;             /**< mask of enabled vbo arrays. */
       GLubyte attrsz[VBO_ATTRIB_MAX];   /**< nr. of attrib components (1..4) */
       GLenum attrtype[VBO_ATTRIB_MAX];  /**< GL_FLOAT, GL_DOUBLE, GL_INT, etc */
       GLubyte active_sz[VBO_ATTRIB_MAX];  /**< attrib size (nr. 32-bit words) */

src/mesa/vbo/vbo_exec_api.c

@@ -42,6 +42,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/api_arrayelt.h"
 #include "main/api_validate.h"
 #include "main/dispatch.h"
+#include "util/bitscan.h"
 
 #include "vbo_context.h"
 #include "vbo_noop.h"
@@ -167,10 +168,11 @@ static void vbo_exec_copy_to_current( struct vbo_exec_context *exec )
 {
    struct gl_context *ctx = exec->ctx;
    struct vbo_context *vbo = vbo_context(ctx);
-   GLuint i;
+   GLbitfield64 enabled = exec->vtx.enabled & (~BITFIELD64_BIT(VBO_ATTRIB_POS));
+
+   while (enabled) {
+      const int i = u_bit_scan64(&enabled);
 
-   for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) {
-      if (exec->vtx.attrsz[i]) {
       /* Note: the exec->vtx.current[i] pointers point into the
        * ctx->Current.Attrib and ctx->Light.Material.Attrib arrays.
        */
@@ -178,6 +180,8 @@ static void vbo_exec_copy_to_current( struct vbo_exec_context *exec )
       fi_type tmp[8]; /* space for doubles */
       int dmul = exec->vtx.attrtype[i] == GL_DOUBLE ? 2 : 1;
 
+      assert(exec->vtx.attrsz[i]);
+
       if (exec->vtx.attrtype[i] == GL_DOUBLE) {
          memset(tmp, 0, sizeof(tmp));
          memcpy(tmp, exec->vtx.attrptr[i], exec->vtx.attrsz[i] * sizeof(GLfloat));
@@ -216,7 +220,6 @@ static void vbo_exec_copy_to_current( struct vbo_exec_context *exec )
          ctx->NewState |= _NEW_CURRENT_ATTRIB;
       }
    }
-   }
 
    /* Colormaterial -- this kindof sucks.
     */
@@ -311,6 +314,7 @@ vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec,
    exec->vtx.max_vert = vbo_compute_max_verts(exec);
    exec->vtx.vert_count = 0;
    exec->vtx.buffer_ptr = exec->vtx.buffer_map;
+   exec->vtx.enabled |= BITFIELD64_BIT(attr);
 
    if (unlikely(oldSize)) {
       /* Size changed, recalculate all the attrptr[] values
@@ -345,18 +349,19 @@ vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec,
    if (unlikely(exec->vtx.copied.nr)) {
       fi_type *data = exec->vtx.copied.buffer;
       fi_type *dest = exec->vtx.buffer_ptr;
-      GLuint j;
 
       assert(exec->vtx.buffer_ptr == exec->vtx.buffer_map);
 
       for (i = 0 ; i < exec->vtx.copied.nr ; i++) {
-	 for (j = 0 ; j < VBO_ATTRIB_MAX ; j++) {
+         GLbitfield64 enabled = exec->vtx.enabled;
+         while (enabled) {
+            const int j = u_bit_scan64(&enabled);
 	    GLuint sz = exec->vtx.attrsz[j];
-
-	    if (sz) {
             GLint old_offset = old_attrptr[j] - exec->vtx.vertex;
             GLint new_offset = exec->vtx.attrptr[j] - exec->vtx.vertex;
 
+            assert(sz);
+
             if (j == attr) {
                if (oldSize) {
                   fi_type tmp[4];
@@ -373,7 +378,6 @@ vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec,
                COPY_SZ_4V(dest + new_offset, sz, data + old_offset);
             }
 	 }
-	 }
 
 	 data += old_vtx_size;
 	 dest += exec->vtx.vertex_size;
@@ -1145,6 +1149,7 @@ void vbo_exec_vtx_init( struct vbo_exec_context *exec )
    vbo_exec_vtxfmt_init( exec );
    _mesa_noop_vtxfmt_init(&exec->vtxfmt_noop);
 
+   exec->vtx.enabled = 0;
    for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
       assert(i < ARRAY_SIZE(exec->vtx.attrsz));
       exec->vtx.attrsz[i] = 0;
@@ -1273,9 +1278,10 @@ void vbo_exec_FlushVertices( struct gl_context *ctx, GLuint flags )
 
 static void reset_attrfv( struct vbo_exec_context *exec )
 {   
-   GLuint i;
+   while (exec->vtx.enabled) {
+      const int i = u_bit_scan64(&exec->vtx.enabled);
+      assert(exec->vtx.attrsz[i]);
 
-   for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
       exec->vtx.attrsz[i] = 0;
       exec->vtx.attrtype[i] = GL_FLOAT;
       exec->vtx.active_sz[i] = 0;

src/mesa/vbo/vbo_exec_draw.c

@@ -214,6 +214,8 @@ vbo_exec_bind_arrays( struct gl_context *ctx )
          exec->vtx.attrsz[VERT_ATTRIB_GENERIC0] = exec->vtx.attrsz[0];
          exec->vtx.attrptr[VERT_ATTRIB_GENERIC0] = exec->vtx.attrptr[0];
          exec->vtx.attrsz[0] = 0;
+         exec->vtx.enabled &= (~BITFIELD64_BIT(VBO_ATTRIB_POS));
+         exec->vtx.enabled |= BITFIELD64_BIT(VBO_ATTRIB_GENERIC0);
       }
       break;
    default: