Implement the conversion and do the initial execution pass.

9 files changed, 242 insertions, 10 deletions

diff --git a/src/mesa/pipe/draw/draw_private.h b/src/mesa/pipe/draw/draw_private.h
index b7811e891e..b672cc2e8b 100644
--- a/src/mesa/pipe/draw/draw_private.h
+++ b/src/mesa/pipe/draw/draw_private.h
@@ -251,6 +251,7 @@ extern void draw_vertex_cache_reset_vertex_ids( struct draw_context *draw );
 
 
 extern void draw_vertex_shader_queue_flush( struct draw_context *draw );
+extern void draw_vertex_shader_queue_flush_llvm( struct draw_context *draw );
 
 struct tgsi_exec_machine;
 
diff --git a/src/mesa/pipe/draw/draw_vertex_fetch.c b/src/mesa/pipe/draw/draw_vertex_fetch.c
index de1cd06da6..1b093b7342 100644
--- a/src/mesa/pipe/draw/draw_vertex_fetch.c
+++ b/src/mesa/pipe/draw/draw_vertex_fetch.c
@@ -96,7 +96,7 @@ void draw_vertex_fetch( struct draw_context *draw,
    for (j = 0; j < count; j++) {
       uint attr;
 
-      /*printf("fetch vertex %u: \n", j);*/
+      printf("fetch vertex %u: \n", j);
 
       /* loop over vertex attributes (vertex shader inputs) */
       for (attr = 0; attr < draw->vertex_shader->state->num_inputs; attr++) {
@@ -111,7 +111,7 @@ void draw_vertex_fetch( struct draw_context *draw,
 
          fetch_attrib4(src, draw->vertex_element[attr].src_format, p);
 
-         /*printf("  %u: %f %f %f %f\n", attr, p[0], p[1], p[2], p[3]);*/
+         printf(">  %u: %f %f %f %f\n", attr, p[0], p[1], p[2], p[3]);
 
          /* Transform to AoS xxxx/yyyy/zzzz/wwww representation:
           */
diff --git a/src/mesa/pipe/draw/draw_vertex_shader.c b/src/mesa/pipe/draw/draw_vertex_shader.c
index fd0183f8cd..874259e727 100644
--- a/src/mesa/pipe/draw/draw_vertex_shader.c
+++ b/src/mesa/pipe/draw/draw_vertex_shader.c
@@ -91,6 +91,8 @@ run_vertex_program(struct draw_context *draw,
    assert(draw->vertex_shader->state->output_semantic_name[0]
           == TGSI_SEMANTIC_POSITION);
 
+   fprintf(stderr, "------ run_vertex\n");
+
    /* Consts does not require 16 byte alignment. */
    machine->Consts = (float (*)[4]) draw->mapped_constants;
 
@@ -161,10 +163,6 @@ run_vertex_program(struct draw_context *draw,
          vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 #if DBG
          printf("output[%d][%d]: %f %f %f %f\n", j, slot,
-                vOut[j]->data[slot][0],
-                vOut[j]->data[slot][1],
-                vOut[j]->data[slot][2],
-                vOut[j]->data[slot][3]);
 #endif
       }
    } /* loop over vertices */
@@ -179,7 +177,12 @@ void draw_vertex_shader_queue_flush( struct draw_context *draw )
 {
    unsigned i, j;
 
-//   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
+   fprintf(stderr, "XX q(%d) ", draw->vs.queue_nr );
+
+   if (draw->vertex_shader->state->llvm_prog) {
+      draw_vertex_shader_queue_flush_llvm(draw);
+      return;
+   }
 
    /* run vertex shader on vertex cache entries, four per invokation */
    for (i = 0; i < draw->vs.queue_nr; i += 4) {
diff --git a/src/mesa/pipe/draw/draw_vertex_shader_llvm.c b/src/mesa/pipe/draw/draw_vertex_shader_llvm.c
new file mode 100644
index 0000000000..b38498efc5
--- /dev/null
+++ b/src/mesa/pipe/draw/draw_vertex_shader_llvm.c
@@ -0,0 +1,129 @@
+/**************************************************************************
+ *
+ * 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.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin zack@tungstengraphics.com
+  */
+
+#include "pipe/p_util.h"
+#include "draw_private.h"
+#include "draw_context.h"
+#include "draw_vertex.h"
+
+#include "pipe/llvm/llvmtgsi.h"
+#include "pipe/tgsi/exec/tgsi_core.h"
+
+static INLINE void
+fetch_attrib4(const void *ptr, unsigned format, float attrib[4])
+{
+   /* defaults */
+   attrib[1] = 0.0;
+   attrib[2] = 0.0;
+   attrib[3] = 1.0;
+   switch (format) {
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      attrib[3] = ((float *) ptr)[3];
+      /* fall-through */
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      attrib[2] = ((float *) ptr)[2];
+      /* fall-through */
+   case PIPE_FORMAT_R32G32_FLOAT:
+      attrib[1] = ((float *) ptr)[1];
+      /* fall-through */
+   case PIPE_FORMAT_R32_FLOAT:
+      attrib[0] = ((float *) ptr)[0];
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+/**
+ * Fetch vertex attributes for 'count' vertices.
+ */
+static INLINE
+void vertex_fetch(struct draw_context *draw,
+                  const unsigned elt,
+                  float (*inputs)[4])
+{
+   uint attr;
+
+   printf("fetch vertex %u: \n", elt);
+
+   /* loop over vertex attributes (vertex shader inputs) */
+   for (attr = 0; attr < draw->vertex_shader->state->num_inputs; attr++) {
+
+      unsigned buf = draw->vertex_element[attr].vertex_buffer_index;
+      const void *src
+         = (const void *) ((const ubyte *) draw->mapped_vbuffer[buf]
+                           + draw->vertex_buffer[buf].buffer_offset
+                           + draw->vertex_element[attr].src_offset
+                           + elt * draw->vertex_buffer[buf].pitch);
+      float p[4];
+
+      fetch_attrib4(src, draw->vertex_element[attr].src_format, p);
+
+      printf(">  %u: %f %f %f %f\n", attr, p[0], p[1], p[2], p[3]);
+
+      inputs[attr][0] = p[0]; /*X*/
+      inputs[attr][1] = p[1]; /*Y*/
+      inputs[attr][2] = p[2]; /*Z*/
+      inputs[attr][3] = p[3]; /*W*/
+   }
+}
+
+
+/**
+ * Called by the draw module when the vertx cache needs to be flushed.
+ * This involves running the vertex shader.
+ */
+void draw_vertex_shader_queue_flush_llvm(struct draw_context *draw)
+{
+   unsigned i;
+
+   struct vertex_header *dests[VS_QUEUE_LENGTH];
+   float                 inputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4];
+   float (*consts)[4] = (float (*)[4]) draw->mapped_constants;
+   struct ga_llvm_prog *prog = draw->vertex_shader->state->llvm_prog;
+
+   fprintf(stderr, "XX q(%d) ", draw->vs.queue_nr);
+
+   /* fetch the inputs */
+   for (i = 0; i < draw->vs.queue_nr; ++i) {
+      unsigned elt = draw->vs.queue[i].elt;
+      dests[i] = draw->vs.queue[i].dest;
+      vertex_fetch(draw, elt, inputs[i]);
+   }
+
+   /* batch execute the shaders on all the vertices */
+   ga_llvm_prog_exec(prog, inputs, dests, consts,
+                     draw->vs.queue_nr);
+
+   draw->vs.queue_nr = 0;
+}
diff --git a/src/mesa/pipe/llvm/llvm_builtins.c b/src/mesa/pipe/llvm/llvm_builtins.c
new file mode 100644
index 0000000000..0f0efeb303
--- /dev/null
+++ b/src/mesa/pipe/llvm/llvm_builtins.c
@@ -0,0 +1,90 @@
+
+
+inline float4 compute_clip(float4 vec, float4 scale, float4 trans)
+{
+   return vec*scale + trans;
+}
+
+
+inline float
+dot4(const float4 a, const float4 b)
+{
+   float4 c = a*b;
+   return c.x + c.y + c.z + c.w;
+}
+
+inline unsigned
+compute_clipmask(float4 clip, const float4 (*plane), unsigned nr)
+{
+   unsigned mask = 0;
+   unsigned i;
+
+   for (i = 0; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0)
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+inline void collect_results(float4 *results, struct vertex_header *vOut,
+                            float4 *planes, int nr_planes,
+                            float4 scale, float4 trans)
+{
+   /* store results */
+   unsigned slot;
+   float x, y, z, w;
+
+   /* Handle attr[0] (position) specially:
+    */
+   float4 res0 = results[0];
+   x = vOut->clip[0] = clip.x;
+   y = vOut->clip[1] = clip.y;
+   z = vOut->clip[2] = clip.z;
+   w = vOut->clip[3] = clip.w;
+
+   vOut[i]->clipmask = compute_clipmask(res0, planes, nr_planes);
+   vOut[i]->edgeflag = 1;
+
+   /* divide by w */
+   w = 1.0f / w;
+   x *= w;
+   y *= w;
+   z *= w;
+   res0.x = x; res0.y = y; res0.z = z; res0.w = 1;
+
+   /* Viewport mapping */
+   res = res * scale + trans;
+   vOut->data[0][0] = res.x;
+   vOut->data[0][1] = res.y;
+   vOut->data[0][2] = res.z;
+   vOut->data[0][3] = w;
+
+   /* Remaining attributes are packed into sequential post-transform
+    * vertex attrib slots.
+    * Skip 0 since we just did it above.
+    * Subtract two because of the VERTEX_HEADER, CLIP_POS attribs.
+    */
+   for (slot = 1; slot < draw->vertex_info.num_attribs - 2; slot++) {
+      float4 vec = results[slot];
+      vOut->data[slot][0] = vec.x;
+      vOut->data[slot][1] = vec.y;
+      vOut->data[slot][2] = vec.z;
+      vOut->data[slot][3] = vec.w;
+
+      printf("output %d: %f %f %f %f\n", slot,
+             vOut->data[slot][0],
+             vOut->data[slot][1],
+             vOut->data[slot][2],
+             vOut->data[slot][3]);
+   }
+}
+
+void run_vertex_shader(float ainputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS][4],
+                       struct vertex_header *dests[VS_QUEUE_LENGTH],
+                       float *aconsts[4]
+                       int count)
+{
+   float4  inputs[VS_QUEUE_LENGTH][PIPE_MAX_SHADER_INPUTS];
+   float4 *consts;
+}
diff --git a/src/mesa/pipe/llvm/llvmtgsi.cpp b/src/mesa/pipe/llvm/llvmtgsi.cpp
index 2221a2c353..46b7561b5e 100644
--- a/src/mesa/pipe/llvm/llvmtgsi.cpp
+++ b/src/mesa/pipe/llvm/llvmtgsi.cpp
@@ -426,7 +426,11 @@ void ga_llvm_prog_delete(struct ga_llvm_prog *prog)
    free(prog);
 }
 
-int ga_llvm_prog_exec(struct ga_llvm_prog *prog)
+int ga_llvm_prog_exec(struct ga_llvm_prog *prog,
+                      float (*inputs)[32][4],
+                      void *dests[16*32*4],
+                      float (*consts)[4],
+                      int count)
 {
    //std::cout << "START "<<std::endl;
    llvm::Module *mod = static_cast<llvm::Module*>(prog->module);
diff --git a/src/mesa/pipe/llvm/llvmtgsi.h b/src/mesa/pipe/llvm/llvmtgsi.h
index c0cee915b9..9fbb0ea8f9 100644
--- a/src/mesa/pipe/llvm/llvmtgsi.h
+++ b/src/mesa/pipe/llvm/llvmtgsi.h
@@ -18,7 +18,11 @@ ga_llvm_from_tgsi(const struct tgsi_token *tokens);
 
 void ga_llvm_prog_delete(struct ga_llvm_prog *prog);
 
-int ga_llvm_prog_exec(struct ga_llvm_prog *prog);
+int ga_llvm_prog_exec(struct ga_llvm_prog *prog,
+                      float (*inputs)[32][4],
+                      void *dests[16*32*4],
+                      float (*consts)[4],
+                      int count);
 
 #if defined __cplusplus
 } // extern "C"
diff --git a/src/mesa/pipe/softpipe/sp_quad_fs.c b/src/mesa/pipe/softpipe/sp_quad_fs.c
index 035f05170b..00afcd930a 100644
--- a/src/mesa/pipe/softpipe/sp_quad_fs.c
+++ b/src/mesa/pipe/softpipe/sp_quad_fs.c
@@ -109,7 +109,7 @@ shade_quad(
          machine->InterpCoefs );
    }
    else {
-      ga_llvm_prog_exec(softpipe->fs->llvm_prog);
+      //ga_llvm_prog_exec(softpipe->fs->llvm_prog);
       quad->mask &= tgsi_exec_machine_run( machine );
    }
 
diff --git a/src/mesa/sources b/src/mesa/sources
index 41c3ae387e..c2544ed322 100644
--- a/src/mesa/sources
+++ b/src/mesa/sources
@@ -172,6 +172,7 @@ DRAW_SOURCES = \
 	pipe/draw/draw_vertex_cache.c \
 	pipe/draw/draw_vertex_fetch.c \
 	pipe/draw/draw_vertex_shader.c \
+	pipe/draw/draw_vertex_shader_llvm.c \
 	pipe/draw/draw_wide_prims.c
 
 TGSIEXEC_SOURCES = \