softpipe_draw_arrays() function
authorBrian <brian.paul@tungstengraphics.com>
Thu, 16 Aug 2007 01:02:37 +0000 (19:02 -0600)
committerBrian <brian.paul@tungstengraphics.com>
Thu, 16 Aug 2007 01:04:45 +0000 (19:04 -0600)
Also includes:
  Temporary stand-in for vertex program execution.
  draw_prim() code
  post-transform vertex cache

src/mesa/pipe/softpipe/sp_draw_arrays.c [new file with mode: 0644]

diff --git a/src/mesa/pipe/softpipe/sp_draw_arrays.c b/src/mesa/pipe/softpipe/sp_draw_arrays.c
new file mode 100644 (file)
index 0000000..c466884
--- /dev/null
@@ -0,0 +1,663 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Author:
+ *    Brian Paul
+ *    Keith Whitwell
+ */
+
+
+#include "main/mtypes.h"
+#include "main/context.h"
+
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_winsys.h"
+
+#include "sp_context.h"
+#include "sp_state.h"
+
+#include "pipe/draw/draw_private.h"
+#include "pipe/draw/draw_context.h"
+
+
+
+#define RP_NONE  0
+#define RP_POINT 1
+#define RP_LINE  2
+#define RP_TRI   3
+
+static unsigned reduced_prim[GL_POLYGON + 1] = {
+   RP_POINT,
+   RP_LINE,
+   RP_LINE,
+   RP_LINE,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI,
+   RP_TRI
+};
+
+
+/**
+ * Stand-in for actual vertex program execution
+ * XXX this will probably live in a new file, like "sp_vs.c"
+ * \param draw  the drawing context
+ * \param vbuffer  the mapped vertex buffer pointer
+ * \param elem  which element of the vertex buffer to use as input
+ * \param vOut  the output vertex
+ */
+static void
+run_vertex_program(struct draw_context *draw,
+                   const void *vbuffer, GLuint elem,
+                   struct vertex_header *vOut)
+{
+   const float *vIn, *cIn;
+   const GLfloat *scale = draw->viewport.scale;
+   const GLfloat *trans = draw->viewport.translate;
+   const void *mapped = vbuffer;
+
+   GET_CURRENT_CONTEXT(ctx);
+   const GLfloat *m = ctx->_ModelProjectMatrix.m;
+
+   vIn = (const float *) ((const GLubyte *) mapped
+                          + draw->vertex_buffer[0].buffer_offset
+                          + draw->vertex_element[0].src_offset
+                          + elem * draw->vertex_buffer[0].pitch);
+
+   cIn = (const float *) ((const GLubyte *) mapped
+                          + draw->vertex_buffer[3].buffer_offset
+                          + draw->vertex_element[3].src_offset
+                          + elem * draw->vertex_buffer[3].pitch);
+
+   {
+      float x = vIn[0];
+      float y = vIn[1];
+      float z = vIn[2];
+      float w = 1.0;
+
+      vOut->clipmask = 0x0;
+      vOut->edgeflag = 0;
+      /* MVP */
+      vOut->clip[0] = m[0] * x + m[4] * y + m[ 8] * z + m[12] * w;
+      vOut->clip[1] = m[1] * x + m[5] * y + m[ 9] * z + m[13] * w;
+      vOut->clip[2] = m[2] * x + m[6] * y + m[10] * z + m[14] * w;
+      vOut->clip[3] = m[3] * x + m[7] * y + m[11] * z + m[15] * w;
+
+      /* divide by w */
+      x = vOut->clip[0] / vOut->clip[3];
+      y = vOut->clip[1] / vOut->clip[3];
+      z = vOut->clip[2] / vOut->clip[3];
+      w = 1.0 / vOut->clip[3];
+
+      /* Viewport */
+      vOut->data[0][0] = scale[0] * x + trans[0];
+      vOut->data[0][1] = scale[1] * y + trans[1];
+      vOut->data[0][2] = scale[2] * z + trans[2];
+      vOut->data[0][3] = w;
+
+      /* color */
+      vOut->data[1][0] = cIn[0];
+      vOut->data[1][1] = cIn[1];
+      vOut->data[1][2] = cIn[2];
+      vOut->data[1][3] = 1.0;
+   }
+}
+
+
+static void vs_flush( struct draw_context *draw )
+{
+   unsigned i;
+
+   /* We're not really running a vertex shader yet, so flushing the vs
+    * queue is just a matter of building the vertices and returning.
+    */ 
+   /* Actually, I'm cheating even more and pre-building them still
+    * with the mesa/vf module.  So it's very easy...
+    */
+   for (i = 0; i < draw->vs.queue_nr; i++) {
+      /* Would do the following steps here:
+       *
+       * 1) Loop over vertex element descriptors, fetch data from each
+       *    to build the pre-tnl vertex.  This might require a new struct
+       *    to represent the pre-tnl vertex.
+       * 
+       * 2) Bundle groups of upto 4 pre-tnl vertices together and pass
+       *    to vertex shader.  
+       *
+       * 3) Do any necessary unswizzling, make sure vertex headers are
+       *    correctly populated, store resulting post-transformed
+       *    vertices in vcache.
+       *
+       * In this version, just do the last step:
+       */
+      const unsigned elt = draw->vs.queue[i].elt;
+      struct vertex_header *dest = draw->vs.queue[i].dest;
+
+      run_vertex_program(draw, draw->mapped_vbuffer, elt, dest);
+   }
+   draw->vs.queue_nr = 0;
+}
+
+
+static void draw_flush( struct draw_context *draw )
+{
+   struct draw_stage *first = draw->pipeline.first;
+   unsigned i;
+
+   /* Make sure all vertices are available:
+    */
+   vs_flush( draw );
+
+
+   switch (draw->reduced_prim) {
+   case RP_TRI:
+      for (i = 0; i < draw->pq.queue_nr; i++) {
+        if (draw->pq.queue[i].reset_line_stipple)
+           first->reset_stipple_counter( first );
+
+        first->tri( first, &draw->pq.queue[i] );
+      }
+      break;
+   case RP_LINE:
+      for (i = 0; i < draw->pq.queue_nr; i++) {
+        if (draw->pq.queue[i].reset_line_stipple)
+           first->reset_stipple_counter( first );
+
+        first->line( first, &draw->pq.queue[i] );
+      }
+      break;
+   case RP_POINT:
+      first->reset_stipple_counter( first );
+      for (i = 0; i < draw->pq.queue_nr; i++)
+        first->point( first, &draw->pq.queue[i] );
+      break;
+   }
+
+   draw->pq.queue_nr = 0;
+   draw->vcache.referenced = 0;
+   draw->vcache.overflow = 0;
+}
+
+
+static void draw_invalidate_vcache( struct draw_context *draw )
+{
+   unsigned i;
+
+   assert(draw->pq.queue_nr == 0);
+   assert(draw->vs.queue_nr == 0);
+   assert(draw->vcache.referenced == 0);
+   
+   for (i = 0; i < Elements( draw->vcache.idx ); i++)
+      draw->vcache.idx[i] = ~0;
+}
+
+
+/* Return a pointer to a freshly queued primitive header.  Ensure that
+ * there is room in the vertex cache for a maximum of "nr_verts" new
+ * vertices.  Flush primitive and/or vertex queues if necessary to
+ * make space.
+ */
+static struct prim_header *get_queued_prim( struct draw_context *draw,
+                                           GLuint nr_verts )
+{
+   if (draw->pq.queue_nr + 1 >= PRIM_QUEUE_LENGTH ||
+       draw->vcache.overflow + nr_verts >= VCACHE_OVERFLOW) 
+      draw_flush( draw );
+
+   /* The vs queue is sized so that this can never happen:
+    */
+   assert(draw->vs.queue_nr + nr_verts < VS_QUEUE_LENGTH);
+
+   return &draw->pq.queue[draw->pq.queue_nr++];
+}
+
+
+/* Check if vertex is in cache, otherwise add it.  It won't go through
+ * VS yet, not until there is a flush operation or the VS queue fills up.  
+ */
+static struct vertex_header *get_vertex( struct draw_context *draw,
+                                        GLuint i )
+{
+   unsigned slot = (i + (i>>5)) & 31;
+   
+   /* Cache miss?
+    */
+   if (draw->vcache.idx[slot] != i) {
+
+      /* If slot is in use, use the overflow area:
+       */
+      if (draw->vcache.referenced & (1<<slot)) 
+        slot = draw->vcache.overflow++;
+
+      draw->vcache.idx[slot] = i;
+
+      /* Add to vertex shader queue:
+       */
+      draw->vs.queue[draw->vs.queue_nr].dest = draw->vcache.vertex[slot];
+      draw->vs.queue[draw->vs.queue_nr].elt = i;
+      draw->vs.queue_nr++;
+   }
+
+   /* Mark slot as in-use:
+    */
+   draw->vcache.referenced |= (1<<slot);
+   return draw->vcache.vertex[slot];
+}
+
+
+
+static void draw_set_prim( struct draw_context *draw,
+                          GLenum prim )
+{
+   if (reduced_prim[prim] != draw->reduced_prim) {
+      draw_flush( draw );
+      draw->reduced_prim = reduced_prim[prim];
+   }
+
+   draw->prim = prim;
+}
+
+
+static void do_point( struct draw_context *draw,
+                     GLuint i0 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 1 );
+   
+   prim->reset_line_stipple = 0;
+   prim->edgeflags = 1;
+   prim->pad = 0;
+   prim->v[0] = draw->get_vertex( draw, i0 );
+}
+
+
+static void do_line( struct draw_context *draw,
+                    GLboolean reset_stipple,
+                    GLuint i0,
+                    GLuint i1 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 2 );
+   
+   prim->reset_line_stipple = reset_stipple;
+   prim->edgeflags = 1;
+   prim->pad = 0;
+   prim->v[0] = draw->get_vertex( draw, i0 );
+   prim->v[1] = draw->get_vertex( draw, i1 );
+}
+
+static void do_triangle( struct draw_context *draw,
+                        GLuint i0,
+                        GLuint i1,
+                        GLuint i2 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 3 );
+   
+   prim->reset_line_stipple = 1;
+   prim->edgeflags = ~0;
+   prim->pad = 0;
+   prim->v[0] = draw->get_vertex( draw, i0 );
+   prim->v[1] = draw->get_vertex( draw, i1 );
+   prim->v[2] = draw->get_vertex( draw, i2 );
+}
+                         
+static void do_ef_triangle( struct draw_context *draw,
+                           GLboolean reset_stipple,
+                           GLuint ef_mask,
+                           GLuint i0,
+                           GLuint i1,
+                           GLuint i2 )
+{
+   struct prim_header *prim = get_queued_prim( draw, 3 );
+   struct vertex_header *v0 = draw->get_vertex( draw, i0 );
+   struct vertex_header *v1 = draw->get_vertex( draw, i1 );
+   struct vertex_header *v2 = draw->get_vertex( draw, i2 );
+   
+   prim->reset_line_stipple = reset_stipple;
+
+   prim->edgeflags = ef_mask & ((v0->edgeflag << 0) | 
+                               (v1->edgeflag << 1) | 
+                               (v2->edgeflag << 2));
+   prim->pad = 0;
+   prim->v[0] = v0;
+   prim->v[1] = v1;
+   prim->v[2] = v2;
+}
+
+
+static void do_quad( struct draw_context *draw,
+                    unsigned v0,
+                    unsigned v1,
+                    unsigned v2,
+                    unsigned v3 )
+{
+   do_ef_triangle( draw, 1, ~(1<<0), v0, v1, v3 );
+   do_ef_triangle( draw, 0, ~(1<<1), v1, v2, v3 );
+}
+
+
+static void draw_prim( struct draw_context *draw,
+                      GLuint start,
+                      GLuint count )
+{
+   GLuint i;
+
+//   _mesa_printf("%s (%d) %d/%d\n", __FUNCTION__, draw->prim, start, count );
+
+   switch (draw->prim) {
+   case GL_POINTS:
+      for (i = 0; i < count; i ++) {
+        do_point( draw,
+                  start + i );
+      }
+      break;
+
+   case GL_LINES:
+      for (i = 0; i+1 < count; i += 2) {
+        do_line( draw, 
+                 TRUE,
+                 start + i + 0,
+                 start + i + 1);
+      }
+      break;
+
+   case GL_LINE_LOOP:  
+      if (count >= 2) {
+        for (i = 1; i < count; i++) {
+           do_line( draw, 
+                    i == 1,    /* XXX: only if vb not split */
+                    start + i - 1,
+                    start + i );
+        }
+
+        do_line( draw, 
+                 0,
+                 start + count - 1,
+                 start + 0 );
+      }
+      break;
+
+   case GL_LINE_STRIP:
+      if (count >= 2) {
+        for (i = 1; i < count; i++) {
+           do_line( draw,
+                    i == 1,
+                    start + i - 1,
+                    start + i );
+        }
+      }
+      break;
+
+   case GL_TRIANGLES:
+      for (i = 0; i+2 < count; i += 3) {
+        do_ef_triangle( draw,
+                        1, 
+                        ~0,
+                        start + i + 0,
+                        start + i + 1,
+                        start + i + 2 );
+      }
+      break;
+
+   case GL_TRIANGLE_STRIP:
+      for (i = 0; i+2 < count; i++) {
+        if (i & 1) {
+           do_triangle( draw,
+                        start + i + 1,
+                        start + i + 0,
+                        start + i + 2 );
+        }
+        else {
+           do_triangle( draw,
+                        start + i + 0,
+                        start + i + 1,
+                        start + i + 2 );
+        }
+      }
+      break;
+
+   case GL_TRIANGLE_FAN:
+      if (count >= 3) {
+        for (i = 0; i+2 < count; i++) {
+           do_triangle( draw,
+                        start + 0,
+                        start + i + 1,
+                        start + i + 2 );
+        }
+      }
+      break;
+
+
+   case GL_QUADS:
+      for (i = 0; i+3 < count; i += 4) {
+        do_quad( draw,
+                 start + i + 0,
+                 start + i + 1,
+                 start + i + 2,
+                 start + i + 3);
+      }
+      break;
+
+   case GL_QUAD_STRIP:
+      for (i = 0; i+3 < count; i += 2) {
+        do_quad( draw,
+                 start + i + 2,
+                 start + i + 0,
+                 start + i + 1,
+                 start + i + 3);
+      }
+      break;
+
+   case GL_POLYGON:
+      if (count >= 3) {
+        unsigned ef_mask = (1<<2) | (1<<0);
+
+        for (i = 0; i+2 < count; i++) {
+
+            if (i + 3 >= count)
+              ef_mask |= (1<<1);
+
+           do_ef_triangle( draw,
+                           i == 0,
+                           ef_mask,
+                           start + i + 1,
+                           start + i + 2,
+                           start + i + 0);
+
+           ef_mask &= ~(1<<2);
+        }
+      }
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+}
+
+
+
+static GLuint draw_prim_info(GLenum mode, GLuint *first, GLuint *incr)
+{
+   switch (mode) {
+   case GL_POINTS:
+      *first = 1;
+      *incr = 1;
+      return 0;
+   case GL_LINES:
+      *first = 2;
+      *incr = 2;
+      return 0;
+   case GL_LINE_STRIP:
+      *first = 2;
+      *incr = 1;
+      return 0;
+   case GL_LINE_LOOP:
+      *first = 2;
+      *incr = 1;
+      return 1;
+   case GL_TRIANGLES:
+      *first = 3;
+      *incr = 3;
+      return 0;
+   case GL_TRIANGLE_STRIP:
+      *first = 3;
+      *incr = 1;
+      return 0;
+   case GL_TRIANGLE_FAN:
+   case GL_POLYGON:
+      *first = 3;
+      *incr = 1;
+      return 1;
+   case GL_QUADS:
+      *first = 4;
+      *incr = 4;
+      return 0;
+   case GL_QUAD_STRIP:
+      *first = 4;
+      *incr = 2;
+      return 0;
+   default:
+      assert(0);
+      *first = 1;
+      *incr = 1;
+      return 0;
+   }
+}
+
+
+static GLuint trim( GLuint count, GLuint first, GLuint incr )
+{
+   if (count < first)
+      return 0;
+   else
+      return count - (count - first) % incr; 
+}
+
+
+
+void
+softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode,
+                     unsigned start, unsigned count)
+{
+   struct softpipe_context *sp = softpipe_context(pipe);
+   struct draw_context *draw = sp->draw;
+   struct pipe_buffer_handle *buf;
+
+   softpipe_map_surfaces(sp);
+
+   /*
+    * Map vertex buffers
+    */
+   buf = sp->vertex_buffer[0].buffer;
+   draw->mapped_vbuffer
+      = pipe->winsys->buffer_map(pipe->winsys, buf, PIPE_BUFFER_FLAG_READ);
+
+
+   /* tell drawing pipeline we're beginning drawing */
+   draw->pipeline.first->begin( draw->pipeline.first );
+
+   draw_invalidate_vcache( draw );
+
+#if 0
+   if (VB->Elts) 
+      draw->get_vertex = get_uint_elt_vertex;
+   else
+#endif
+      draw->get_vertex = get_vertex;
+
+   draw_set_prim( draw, mode );
+
+   /* XXX draw_prim_info() and TRIM here */
+   draw_prim(draw, start, count);
+
+   /* draw any left-over buffered prims */
+   draw_flush(draw);
+
+   /* tell drawing pipeline we're done drawing */
+   draw->pipeline.first->end( draw->pipeline.first );
+
+#if 0
+   draw->verts = NULL;
+   draw->in_vb = 0;
+   draw->elts = NULL;
+#endif
+
+   pipe->winsys->buffer_unmap(pipe->winsys, buf);
+
+   softpipe_unmap_surfaces(sp);
+}
+
+
+
+#define EMIT_ATTR( VF_ATTR, STYLE, SIZE )                      \
+do {                                                           \
+   if (draw->nr_attrs >= 2)                                    \
+      draw->vf_attr_to_slot[VF_ATTR] = draw->nr_attrs - 2;     \
+   draw->attrs[draw->nr_attrs].attrib = VF_ATTR;               \
+   draw->attrs[draw->nr_attrs].format = STYLE;                 \
+   draw->nr_attrs++;                                           \
+   draw->vertex_size += SIZE;                                  \
+} while (0)
+
+
+void draw_set_vertex_attributes2( struct draw_context *draw,
+                                const GLuint *slot_to_vf_attr,
+                                GLuint nr_attrs )
+{
+   GLuint i;
+
+   memset(draw->vf_attr_to_slot, 0, sizeof(draw->vf_attr_to_slot));
+   draw->nr_attrs = 0;
+   draw->vertex_size = 0;
+
+   /*
+    * First three attribs are always the same: header, clip pos, winpos
+    */
+   EMIT_ATTR(VF_ATTRIB_VERTEX_HEADER, EMIT_1F, 1);
+   EMIT_ATTR(VF_ATTRIB_CLIP_POS, EMIT_4F, 4);
+
+   assert(slot_to_vf_attr[0] == VF_ATTRIB_POS);
+   EMIT_ATTR(slot_to_vf_attr[0], EMIT_4F_VIEWPORT, 4);
+
+   /*
+    * Remaining attribs (color, texcoords, etc)
+    */
+   for (i = 1; i < nr_attrs; i++) 
+      EMIT_ATTR(slot_to_vf_attr[i], EMIT_4F, 4);
+
+#if 0
+   /* tell the vertex format module how to construct vertices for us */
+   draw->vertex_size = vf_set_vertex_attributes( draw->vf, draw->attrs,
+                                                 draw->nr_attrs, 0 );
+#endif
+
+   draw->vertex_size *= 4; /* floats to bytes */
+}
+                           
+