From 73af91e938eb27b001404f11195fb06ff9b08903 Mon Sep 17 00:00:00 2001
From: José Fonseca <jfonseca@vmware.com>
Date: Fri, 14 Aug 2009 10:27:32 +0100
Subject: llvmpipe: Eliminate non-LLVM fs execution paths.

---
 src/gallium/drivers/llvmpipe/lp_quad_fs.c | 119 ++++++++++++++++++++++++++++--
 1 file changed, 113 insertions(+), 6 deletions(-)

(limited to 'src/gallium/drivers/llvmpipe/lp_quad_fs.c')

diff --git a/src/gallium/drivers/llvmpipe/lp_quad_fs.c b/src/gallium/drivers/llvmpipe/lp_quad_fs.c
index 1da50e493b..338a6be80c 100644
--- a/src/gallium/drivers/llvmpipe/lp_quad_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_quad_fs.c
@@ -1,8 +1,8 @@
 /**************************************************************************
  * 
+ * Copyright 2008-2009 VMware, Inc.
  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
- * Copyright 2008 VMware, Inc.  All rights reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the
@@ -65,6 +65,114 @@ quad_shade_stage(struct quad_stage *qs)
 }
 
 
+static void
+shader_prepare( const struct lp_fragment_shader *shader,
+                struct tgsi_exec_machine *machine,
+                struct tgsi_sampler **samplers )
+{
+   /*
+    * Bind tokens/shader to the interpreter's machine state.
+    * Avoid redundant binding.
+    */
+   if (machine->Tokens != shader->base.tokens) {
+      tgsi_exec_machine_bind_shader( machine,
+                                     shader->base.tokens,
+                                     PIPE_MAX_SAMPLERS,
+                                     samplers );
+   }
+}
+
+
+
+static void
+setup_pos_vector(struct lp_fragment_shader *shader,
+                 const struct tgsi_interp_coef *coef,
+                 float x, float y)
+{
+   uint chan;
+
+   /* do X */
+   shader->pos[0].f[0] = x;
+   shader->pos[0].f[1] = x + 1;
+   shader->pos[0].f[2] = x;
+   shader->pos[0].f[3] = x + 1;
+
+   /* do Y */
+   shader->pos[1].f[0] = y;
+   shader->pos[1].f[1] = y;
+   shader->pos[1].f[2] = y + 1;
+   shader->pos[1].f[3] = y + 1;
+
+   /* do Z and W for all fragments in the quad */
+   for (chan = 2; chan < 4; chan++) {
+      const float dadx = coef->dadx[chan];
+      const float dady = coef->dady[chan];
+      const float a0 = coef->a0[chan] + dadx * x + dady * y;
+      shader->pos[chan].f[0] = a0;
+      shader->pos[chan].f[1] = a0 + dadx;
+      shader->pos[chan].f[2] = a0 + dady;
+      shader->pos[chan].f[3] = a0 + dadx + dady;
+   }
+}
+
+
+static void
+setup_coef_vector(struct lp_fragment_shader *shader,
+                  const struct tgsi_interp_coef *coef)
+{
+   unsigned attrib, chan, i;
+
+   for (attrib = 0; attrib < PIPE_MAX_SHADER_INPUTS; ++attrib) {
+      for (chan = 0; chan < NUM_CHANNELS; ++chan) {
+         for( i = 0; i < QUAD_SIZE; ++i ) {
+            shader->a0[attrib][chan].f[i] = coef[attrib].a0[chan];
+            shader->dadx[attrib][chan].f[i] = coef[attrib].dadx[chan];
+            shader->dady[attrib][chan].f[i] = coef[attrib].dady[chan];
+         }
+      }
+   }
+}
+
+
+/* TODO: codegenerate the whole run function, skip this wrapper.
+ * TODO: break dependency on tgsi_exec_machine struct
+ * TODO: push Position calculation into the generated shader
+ * TODO: process >1 quad at a time
+ */
+static unsigned
+shader_run( struct lp_fragment_shader *shader,
+            struct tgsi_exec_machine *machine,
+            struct quad_header *quad )
+{
+   unsigned mask;
+
+   /* Compute X, Y, Z, W vals for this quad */
+   setup_pos_vector(shader,
+                    quad->posCoef,
+                    (float)quad->input.x0, (float)quad->input.y0);
+
+   setup_coef_vector(shader,
+                     quad->coef);
+
+   /* init kill mask */
+   tgsi_set_kill_mask(machine, 0x0);
+   tgsi_set_exec_mask(machine, 1, 1, 1, 1);
+
+   memset(machine->Outputs, 0, sizeof machine->Outputs);
+
+   shader->jit_function( shader->pos,
+                         shader->a0, shader->dadx, shader->dady,
+                         machine->Consts,
+                         machine->Outputs,
+                         machine->Samplers);
+
+   /* FIXME */
+   mask = ~0;
+
+   return mask;
+}
+
+
 /**
  * Execute fragment shader for the four fragments in the quad.
  */
@@ -77,7 +185,7 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad)
    boolean z_written;
 
    /* run shader */
-   quad->inout.mask &= llvmpipe->fs->run( llvmpipe->fs, machine, quad );
+   quad->inout.mask &= shader_run( llvmpipe->fs, machine, quad );
    if (quad->inout.mask == 0)
       return FALSE;
 
@@ -177,10 +285,9 @@ shade_begin(struct quad_stage *qs)
    struct quad_shade_stage *qss = quad_shade_stage(qs);
    struct llvmpipe_context *llvmpipe = qs->llvmpipe;
 
-   llvmpipe->fs->prepare( llvmpipe->fs, 
-			  qss->machine,
-			  (struct tgsi_sampler **)
-                             llvmpipe->tgsi.frag_samplers_list );
+   shader_prepare( llvmpipe->fs,
+                   qss->machine,
+                   (struct tgsi_sampler **)llvmpipe->tgsi.frag_samplers_list );
 
    qs->next->begin(qs->next);
 }
-- 
cgit v1.2.3