gallium: Make PIPE_CAP_xxx enums.
[mesa.git] / src / gallium / drivers / i965 / brw_draw.c
index 741537309ad8610fec9c7bb65b1be60469657ccb..4625c2048f91e5744fe40448fa7c7848e38a11f8 100644 (file)
  **************************************************************************/
 
 
+#include "util/u_inlines.h"
+#include "util/u_prim.h"
+#include "util/u_upload_mgr.h"
+
 #include "brw_draw.h"
 #include "brw_defines.h"
 #include "brw_context.h"
 #include "brw_state.h"
+#include "brw_debug.h"
 
 #include "brw_batchbuffer.h"
-#include "intel_buffer_objects.h"
 
-#define FILE_DEBUG_FLAG DEBUG_BATCH
 
 static uint32_t prim_to_hw_prim[PIPE_PRIM_POLYGON+1] = {
    _3DPRIM_POINTLIST,
@@ -56,26 +59,21 @@ static uint32_t prim_to_hw_prim[PIPE_PRIM_POLYGON+1] = {
  * programs be immune to the active primitive (ie. cope with all
  * possibilities).  That may not be realistic however.
  */
-static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
+static int brw_set_prim(struct brw_context *brw, unsigned prim )
 {
 
-   if (INTEL_DEBUG & DEBUG_PRIMS)
-      _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
+   if (BRW_DEBUG & DEBUG_PRIMS)
+      debug_printf("PRIM: %s\n", u_prim_name(prim));
    
-   /* Slight optimization to avoid the GS program when not needed:
-    */
-   if (prim == GL_QUAD_STRIP &&
-       ctx->Light.ShadeModel != GL_FLAT &&
-       ctx->Polygon.FrontMode == GL_FILL &&
-       ctx->Polygon.BackMode == GL_FILL)
-      prim = GL_TRIANGLE_STRIP;
-
    if (prim != brw->primitive) {
+      unsigned reduced_prim;
+
       brw->primitive = prim;
       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 
-      if (reduced_prim[prim] != brw->intel.reduced_primitive) {
-        brw->intel.reduced_primitive = reduced_prim[prim];
+      reduced_prim = u_reduced_prim(prim);
+      if (reduced_prim != brw->reduced_primitive) {
+        brw->reduced_primitive = reduced_prim;
         brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
       }
    }
@@ -84,134 +82,82 @@ static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
 }
 
 
-static GLuint trim(GLenum prim, GLuint length)
-{
-   if (prim == GL_QUAD_STRIP)
-      return length > 3 ? (length - length % 2) : 0;
-   else if (prim == GL_QUADS)
-      return length - length % 4;
-   else 
-      return length;
-}
-
 
-static void brw_emit_prim(struct brw_context *brw,
-                         const struct _mesa_prim *prim,
-                         uint32_t hw_prim)
+static int brw_emit_prim(struct brw_context *brw,
+                        unsigned start,
+                        unsigned count,
+                        boolean indexed,
+                        uint32_t hw_prim)
 {
    struct brw_3d_primitive prim_packet;
+   int ret;
 
-   if (INTEL_DEBUG & DEBUG_PRIMS)
-      _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode), 
-                  prim->start, prim->count);
+   if (BRW_DEBUG & DEBUG_PRIMS)
+      debug_printf("%s start %d count %d indexed %d hw_prim %d\n",
+                   __FUNCTION__, start, count, indexed, hw_prim); 
 
    prim_packet.header.opcode = CMD_3D_PRIM;
    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
    prim_packet.header.pad = 0;
    prim_packet.header.topology = hw_prim;
-   prim_packet.header.indexed = prim->indexed;
+   prim_packet.header.indexed = indexed;
 
-   prim_packet.verts_per_instance = trim(prim->mode, prim->count);
-   prim_packet.start_vert_location = prim->start;
-   if (prim->indexed)
+   prim_packet.verts_per_instance = count;
+   prim_packet.start_vert_location = start;
+   if (indexed)
       prim_packet.start_vert_location += brw->ib.start_vertex_offset;
    prim_packet.instance_count = 1;
    prim_packet.start_instance_location = 0;
-   prim_packet.base_vert_location = prim->basevertex;
+   prim_packet.base_vert_location = 0; /* prim->basevertex; XXX: add this to gallium */
 
-   /* Can't wrap here, since we rely on the validated state. */
-   brw->no_batch_wrap = GL_TRUE;
 
    /* If we're set to always flush, do it before and after the primitive emit.
     * We want to catch both missed flushes that hurt instruction/state cache
     * and missed flushes of the render cache as it heads to other parts of
     * the besides the draw code.
     */
-   if (intel->always_flush_cache) {
+   if (0) {
       BEGIN_BATCH(1, IGNORE_CLIPRECTS);
-      OUT_BATCH(intel->vtbl.flush_cmd());
+      OUT_BATCH((CMD_MI_FLUSH << 16) | BRW_FLUSH_STATE_CACHE);
       ADVANCE_BATCH();
    }
    if (prim_packet.verts_per_instance) {
-      brw_batchbuffer_data( brw->intel.batch, &prim_packet,
-                             sizeof(prim_packet), LOOP_CLIPRECTS);
+      ret = brw_batchbuffer_data( brw->batch, &prim_packet,
+                                 sizeof(prim_packet), LOOP_CLIPRECTS);
+      if (ret)
+        return ret;
    }
-   if (intel->always_flush_cache) {
+   if (0) {
       BEGIN_BATCH(1, IGNORE_CLIPRECTS);
-      OUT_BATCH(intel->vtbl.flush_cmd());
+      OUT_BATCH((CMD_MI_FLUSH << 16) | BRW_FLUSH_STATE_CACHE);
       ADVANCE_BATCH();
    }
 
-   brw->no_batch_wrap = GL_FALSE;
+   return 0;
 }
 
-static void brw_merge_inputs( struct brw_context *brw,
-                      const struct gl_client_array *arrays[])
-{
-   struct brw_vertex_info old = brw->vb.info;
-   GLuint i;
-
-   for (i = 0; i < VERT_ATTRIB_MAX; i++)
-      brw->sws->bo_unreference(brw->vb.inputs[i].bo);
-
-   memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
-   memset(&brw->vb.info, 0, sizeof(brw->vb.info));
-
-   for (i = 0; i < VERT_ATTRIB_MAX; i++) {
-      brw->vb.inputs[i].glarray = arrays[i];
-      brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
-
-      if (arrays[i]->StrideB != 0)
-        brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
-           ((i%16) * 2);
-   }
-
-   /* Raise statechanges if input sizes have changed. */
-   if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
-      brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
-}
 
 /* May fail if out of video memory for texture or vbo upload, or on
  * fallback conditions.
  */
-static GLboolean brw_try_draw_prims( struct brw_context *brw,
-                                    const struct gl_client_array *arrays[],
-                                    const struct _mesa_prim *prim,
-                                    GLuint nr_prims,
-                                    const struct _mesa_index_buffer *ib,
-                                    GLuint min_index,
-                                    GLuint max_index )
+static int
+try_draw_range_elements(struct brw_context *brw,
+                       struct pipe_resource *index_buffer,
+                       unsigned hw_prim, 
+                       unsigned start, unsigned count)
 {
-   struct brw_context *brw = brw_context(ctx);
-   GLboolean retval = GL_FALSE;
-   GLboolean warn = GL_FALSE;
-   GLboolean first_time = GL_TRUE;
-   uint32_t hw_prim;
-   GLuint i;
-
-   if (ctx->NewState)
-      _mesa_update_state( ctx );
+   int ret;
 
-   /* Bind all inputs, derive varying and size information:
-    */
-   brw_merge_inputs( brw, arrays );
-
-   brw->ib.ib = ib;
-   brw->state.dirty.brw |= BRW_NEW_INDICES;
-
-   brw->vb.min_index = min_index;
-   brw->vb.max_index = max_index;
-   brw->state.dirty.brw |= BRW_NEW_VERTICES;
-
-   hw_prim = brw_set_prim(brw, prim[i].mode);
-
-   brw_validate_state(brw);
+   ret = brw_validate_state(brw);
+   if (ret)
+      return ret;
 
    /* Check that we can fit our state in with our existing batchbuffer, or
     * flush otherwise.
     */
-   ret = dri_bufmgr_check_aperture_space(brw->state.validated_bos,
-                                        brw->state.validated_bo_count);
+   ret = brw->sws->check_aperture_space(brw->sws,
+                                       brw->state.validated_bos,
+                                       brw->state.validated_bo_count);
    if (ret)
       return ret;
 
@@ -219,68 +165,127 @@ static GLboolean brw_try_draw_prims( struct brw_context *brw,
    if (ret)
       return ret;
    
-   ret = brw_emit_prim(brw, &prim[i], hw_prim);
+   ret = brw_emit_prim(brw, start, count, index_buffer != NULL, hw_prim);
    if (ret)
       return ret;
 
-   if (intel->always_flush_batch)
-      brw_batchbuffer_flush(intel->batch);
+   if (brw->flags.always_flush_batch)
+      brw_context_flush( brw );
 
    return 0;
 }
 
-void brw_draw_prims( struct brw_context *brw,
-                    const struct gl_client_array *arrays[],
-                    const struct _mesa_prim *prim,
-                    GLuint nr_prims,
-                    const struct _mesa_index_buffer *ib,
-                    GLboolean index_bounds_valid,
-                    GLuint min_index,
-                    GLuint max_index )
+
+static void
+brw_draw_range_elements(struct pipe_context *pipe,
+                       struct pipe_resource *index_buffer,
+                       unsigned index_size, int index_bias,
+                       unsigned min_index,
+                       unsigned max_index,
+                       unsigned mode, unsigned start, unsigned count)
 {
-   enum pipe_error ret;
+   struct brw_context *brw = brw_context(pipe);
+   int ret;
+   uint32_t hw_prim;
+
+   hw_prim = brw_set_prim(brw, mode);
+
+   if (BRW_DEBUG & DEBUG_PRIMS)
+      debug_printf("PRIM: %s start %d count %d index_buffer %p\n",
+                   u_prim_name(mode), start, count, (void *)index_buffer);
+
+   assert(index_bias == 0);
 
-   if (!vbo_all_varyings_in_vbos(arrays)) {
-      if (!index_bounds_valid)
-        vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
+   /* Potentially trigger upload of new index buffer.
+    *
+    * XXX: do we need to go through state validation to achieve this?
+    * Could just call upload code directly.
+    */
+   if (brw->curr.index_buffer != index_buffer ||
+       brw->curr.index_size != index_size) {
+      pipe_resource_reference( &brw->curr.index_buffer, index_buffer );
+      brw->curr.index_size = index_size;
+      brw->state.dirty.mesa |= PIPE_NEW_INDEX_BUFFER;
+   }
+
+   /* XXX: do we really care?
+    */
+   if (brw->curr.min_index != min_index ||
+       brw->curr.max_index != max_index) 
+   { 
+      brw->curr.min_index = min_index;
+      brw->curr.max_index = max_index;
+      brw->state.dirty.mesa |= PIPE_NEW_INDEX_RANGE;
    }
 
+
    /* Make a first attempt at drawing:
     */
-   ret = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+   ret = try_draw_range_elements(brw, index_buffer, hw_prim, start, count );
 
    /* Otherwise, flush and retry:
     */
    if (ret != 0) {
-      brw_batchbuffer_flush(intel->batch);
-      ret = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
+      brw_context_flush( brw );
+      ret = try_draw_range_elements(brw, index_buffer, hw_prim, start, count );
       assert(ret == 0);
    }
 }
 
-void brw_draw_init( struct brw_context *brw )
+static void
+brw_draw_elements(struct pipe_context *pipe,
+                 struct pipe_resource *index_buffer,
+                 unsigned index_size, int index_bias,
+                 unsigned mode, 
+                 unsigned start, unsigned count)
+{
+   brw_draw_range_elements( pipe, index_buffer,
+                            index_size, index_bias,
+                            0, 0xffffffff,
+                            mode, 
+                            start, count );
+}
+
+static void
+brw_draw_arrays(struct pipe_context *pipe, unsigned mode,
+                     unsigned start, unsigned count)
 {
-   struct vbo_context *vbo = vbo_context(ctx);
+   brw_draw_elements(pipe, NULL, 0, 0, mode, start, count);
+}
 
+
+
+boolean brw_draw_init( struct brw_context *brw )
+{
    /* Register our drawing function: 
     */
-   vbo->draw_prims = brw_draw_prims;
+   brw->base.draw_arrays = brw_draw_arrays;
+   brw->base.draw_elements = brw_draw_elements;
+   brw->base.draw_range_elements = brw_draw_range_elements;
+
+   /* Create helpers for uploading data in user buffers:
+    */
+   brw->vb.upload_vertex = u_upload_create( &brw->base,
+                                           128 * 1024,
+                                           64,
+                                           PIPE_BIND_VERTEX_BUFFER );
+   if (brw->vb.upload_vertex == NULL)
+      return FALSE;
+
+   brw->vb.upload_index = u_upload_create( &brw->base,
+                                          32 * 1024,
+                                          64,
+                                          PIPE_BIND_INDEX_BUFFER );
+   if (brw->vb.upload_index == NULL)
+      return FALSE;
+
+   return TRUE;
 }
 
-void brw_draw_destroy( struct brw_context *brw )
+void brw_draw_cleanup( struct brw_context *brw )
 {
-   int i;
-
-   if (brw->vb.upload.bo != NULL) {
-      brw->sws->bo_unreference(brw->vb.upload.bo);
-      brw->vb.upload.bo = NULL;
-   }
-
-   for (i = 0; i < VERT_ATTRIB_MAX; i++) {
-      brw->sws->bo_unreference(brw->vb.inputs[i].bo);
-      brw->vb.inputs[i].bo = NULL;
-   }
+   u_upload_destroy( brw->vb.upload_vertex );
+   u_upload_destroy( brw->vb.upload_index );
 
-   brw->sws->bo_unreference(brw->ib.bo);
-   brw->ib.bo = NULL;
+   bo_reference(&brw->ib.bo, NULL);
 }