1/*
2 Copyright (C) Intel Corp.  2006.  All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28  * Authors:
29  *   Keith Whitwell <keith@tungstengraphics.com>
30  */
31
32
33#include "main/macros.h"
34#include "brw_context.h"
35#include "brw_vs.h"
36
37/* Component is active if it may diverge from [0,0,0,1].  Undef values
38 * are promoted to [0,0,0,1] for the purposes of this analysis.
39 */
40struct tracker {
41   bool twoside;
42   GLubyte active[PROGRAM_OUTPUT+1][MAX_PROGRAM_TEMPS];
43   GLbitfield size_masks[4];  /**< one bit per fragment program input attrib */
44};
45
46
47static void set_active_component( struct tracker *t,
48				  GLuint file,
49				  GLuint index,
50				  GLubyte active )
51{
52   switch (file) {
53   case PROGRAM_TEMPORARY:
54   case PROGRAM_INPUT:
55   case PROGRAM_OUTPUT:
56      assert(file < PROGRAM_OUTPUT + 1);
57      assert(index < Elements(t->active[0]));
58      t->active[file][index] |= active;
59      break;
60   default:
61      break;
62   }
63}
64
65static void set_active( struct tracker *t,
66			struct prog_dst_register dst,
67			GLuint active )
68{
69   set_active_component( t, dst.File, dst.Index, active & dst.WriteMask );
70}
71
72
73static GLubyte get_active_component( struct tracker *t,
74				     GLuint file,
75				     GLuint index,
76				     GLuint component,
77				     GLubyte swz )
78{
79   switch (swz) {
80   case SWIZZLE_ZERO:
81      return component < 3 ? 0 : (1<<component);
82   case SWIZZLE_ONE:
83      return component == 3 ? 0 : (1<<component);
84   default:
85      switch (file) {
86      case PROGRAM_TEMPORARY:
87      case PROGRAM_INPUT:
88      case PROGRAM_OUTPUT:
89	 return t->active[file][index] & (1<<component);
90      default:
91	 return 1 << component;
92      }
93   }
94}
95
96
97static GLubyte get_active( struct tracker *t,
98			   struct prog_src_register src )
99{
100   GLuint i;
101   GLubyte active = src.Negate; /* NOTE! */
102
103   if (src.RelAddr)
104      return 0xf;
105
106   for (i = 0; i < 4; i++)
107      active |= get_active_component(t, src.File, src.Index, i,
108				     GET_SWZ(src.Swizzle, i));
109
110   return active;
111}
112
113/**
114 * Return the size (1,2,3 or 4) of the output/result for VERT_RESULT_idx.
115 */
116static GLubyte get_output_size( struct tracker *t,
117				GLuint idx )
118{
119   GLubyte active;
120   assert(idx < VERT_RESULT_MAX);
121   active = t->active[PROGRAM_OUTPUT][idx];
122   if (active & (1<<3)) return 4;
123   if (active & (1<<2)) return 3;
124   if (active & (1<<1)) return 2;
125   if (active & (1<<0)) return 1;
126   return 0;
127}
128
129/* Note the potential copying that occurs in the setup program:
130 */
131static void calc_sizes( struct tracker *t )
132{
133   GLint vertRes;
134
135   if (t->twoside) {
136      t->active[PROGRAM_OUTPUT][VERT_RESULT_COL0] |=
137	 t->active[PROGRAM_OUTPUT][VERT_RESULT_BFC0];
138
139      t->active[PROGRAM_OUTPUT][VERT_RESULT_COL1] |=
140	 t->active[PROGRAM_OUTPUT][VERT_RESULT_BFC1];
141   }
142
143   /* Examine vertex program output sizes to set the size_masks[] info
144    * which describes the fragment program input sizes.
145    */
146   for (vertRes = 0; vertRes < VERT_RESULT_MAX; vertRes++) {
147
148      /* map vertex program output index to fragment program input index */
149      GLint fragAttrib = _mesa_vert_result_to_frag_attrib(vertRes);
150      if (fragAttrib < 0)
151         continue;
152
153      switch (get_output_size(t, vertRes)) {
154      case 4: t->size_masks[4-1] |= 1 << fragAttrib;
155      case 3: t->size_masks[3-1] |= 1 << fragAttrib;
156      case 2: t->size_masks[2-1] |= 1 << fragAttrib;
157      case 1: t->size_masks[1-1] |= 1 << fragAttrib;
158	 break;
159      }
160   }
161}
162
163static GLubyte szflag[4+1] = {
164   0,
165   0x1,
166   0x3,
167   0x7,
168   0xf
169};
170
171/* Pull a size out of the packed array:
172 */
173static GLuint get_input_size(struct brw_context *brw,
174			     GLuint attr)
175{
176   GLuint sizes_dword = brw->vb.info.sizes[attr/16];
177   GLuint sizes_bits = (sizes_dword>>((attr%16)*2)) & 0x3;
178   return sizes_bits + 1;
179/*    return brw->vb.inputs[attr].glarray->Size; */
180}
181
182/* Calculate sizes of vertex program outputs.  Size is the largest
183 * component index which might vary from [0,0,0,1]
184 */
185static void calc_wm_input_sizes( struct brw_context *brw )
186{
187   struct gl_context *ctx = &brw->intel.ctx;
188   /* BRW_NEW_VERTEX_PROGRAM */
189   const struct brw_vertex_program *vp =
190      brw_vertex_program_const(brw->vertex_program);
191   /* BRW_NEW_INPUT_DIMENSIONS */
192   struct tracker t;
193   GLuint insn;
194   GLuint i;
195
196   /* Mesa IR is not generated for GLSL vertex shaders.  If there's no Mesa
197    * IR, the code below cannot determine which output components are
198    * written.  So, skip it and assume everything is written.  This
199    * circumvents some optimizations in the fragment shader, but it guarantees
200    * that correct code is generated.
201    */
202   if (vp->program.Base.NumInstructions == 0) {
203      brw->wm.input_size_masks[0] = ~0;
204      brw->wm.input_size_masks[1] = ~0;
205      brw->wm.input_size_masks[2] = ~0;
206      brw->wm.input_size_masks[3] = ~0;
207      return;
208   }
209
210
211   memset(&t, 0, sizeof(t));
212
213   /* _NEW_LIGHT | _NEW_PROGRAM */
214   if (ctx->VertexProgram._TwoSideEnabled)
215      t.twoside = 1;
216
217   for (i = 0; i < VERT_ATTRIB_MAX; i++)
218      if (vp->program.Base.InputsRead & BITFIELD64_BIT(i))
219	 set_active_component(&t, PROGRAM_INPUT, i,
220			      szflag[get_input_size(brw, i)]);
221
222   for (insn = 0; insn < vp->program.Base.NumInstructions; insn++) {
223      struct prog_instruction *inst = &vp->program.Base.Instructions[insn];
224
225      switch (inst->Opcode) {
226      case OPCODE_ARL:
227	 break;
228
229      case OPCODE_MOV:
230	 set_active(&t, inst->DstReg, get_active(&t, inst->SrcReg[0]));
231	 break;
232
233      default:
234	 set_active(&t, inst->DstReg, 0xf);
235	 break;
236      }
237   }
238
239   calc_sizes(&t);
240
241   if (memcmp(brw->wm.input_size_masks, t.size_masks, sizeof(t.size_masks)) != 0) {
242      memcpy(brw->wm.input_size_masks, t.size_masks, sizeof(t.size_masks));
243      brw->state.dirty.brw |= BRW_NEW_WM_INPUT_DIMENSIONS;
244   }
245}
246
247const struct brw_tracked_state brw_wm_input_sizes = {
248   .dirty = {
249      .mesa  = _NEW_LIGHT | _NEW_PROGRAM,
250      .brw   = BRW_NEW_VERTEX_PROGRAM | BRW_NEW_INPUT_DIMENSIONS,
251      .cache = 0
252   },
253   .emit = calc_wm_input_sizes
254};
255
256