draw_vs_ppc.c revision 4eb3225b38ce12cb34ab3d90804c9683bd7b4ed3 
1/**************************************************************************
2 *
3 * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * 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, sub license, 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 portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 /*
29  * Authors:
30  *   Keith Whitwell <keith@tungstengraphics.com>
31  *   Brian Paul
32  */
33
34#include "util/u_math.h"
35#include "util/u_memory.h"
36#include "pipe/p_config.h"
37
38#include "draw_vs.h"
39
40#if defined(PIPE_ARCH_PPC)
41
42#include "pipe/p_shader_tokens.h"
43
44#include "draw_private.h"
45#include "draw_context.h"
46
47#include "rtasm/rtasm_cpu.h"
48#include "rtasm/rtasm_ppc.h"
49#include "tgsi/tgsi_ppc.h"
50#include "tgsi/tgsi_parse.h"
51#include "tgsi/tgsi_exec.h"
52
53
54
55typedef void (PIPE_CDECL *codegen_function) (float (*inputs)[4][4],
56                                             float (*outputs)[4][4],
57                                             float (*temps)[4][4],
58                                             float (*immeds)[4],
59                                             float (*consts)[4],
60                                             const float *builtins);
61
62
63struct draw_ppc_vertex_shader {
64   struct draw_vertex_shader base;
65   struct ppc_function ppc_program;
66
67   codegen_function func;
68};
69
70
71static void
72vs_ppc_prepare( struct draw_vertex_shader *base,
73		struct draw_context *draw )
74{
75   /* nothing */
76}
77
78
79/**
80 * Simplified vertex shader interface for the pt paths.  Given the
81 * complexity of code-generating all the above operations together,
82 * it's time to try doing all the other stuff separately.
83 */
84static void
85vs_ppc_run_linear( struct draw_vertex_shader *base,
86		   const float (*input)[4],
87		   float (*output)[4],
88                  const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
89		   unsigned count,
90		   unsigned input_stride,
91		   unsigned output_stride )
92{
93   struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
94   unsigned int i;
95
96#define MAX_VERTICES 4
97
98   /* loop over verts */
99   for (i = 0; i < count; i += MAX_VERTICES) {
100      const uint max_vertices = MIN2(MAX_VERTICES, count - i);
101      PIPE_ALIGN_VAR(16) float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4];
102      PIPE_ALIGN_VAR(16) float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4];
103      PIPE_ALIGN_VAR(16) float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4];
104      uint attr;
105
106      /* convert (up to) four input verts to SoA format */
107      for (attr = 0; attr < base->info.num_inputs; attr++) {
108         const float *vIn = (const float *) input;
109         uint vert;
110         for (vert = 0; vert < max_vertices; vert++) {
111#if 0
112            if (attr==0)
113               printf("Input v%d a%d: %f %f %f %f\n",
114                      vert, attr, vIn[0], vIn[1], vIn[2], vIn[3]);
115#endif
116            inputs_soa[attr][0][vert] = vIn[attr * 4 + 0];
117            inputs_soa[attr][1][vert] = vIn[attr * 4 + 1];
118            inputs_soa[attr][2][vert] = vIn[attr * 4 + 2];
119            inputs_soa[attr][3][vert] = vIn[attr * 4 + 3];
120            vIn += input_stride / 4;
121         }
122      }
123
124      /* run compiled shader
125       */
126      shader->func(inputs_soa, outputs_soa, temps_soa,
127		   (float (*)[4]) shader->base.immediates,
128                   (float (*)[4])constants[0],
129                   ppc_builtin_constants);
130
131      /* convert (up to) four output verts from SoA back to AoS format */
132      for (attr = 0; attr < base->info.num_outputs; attr++) {
133         float *vOut = (float *) output;
134         uint vert;
135         for (vert = 0; vert < max_vertices; vert++) {
136            vOut[attr * 4 + 0] = outputs_soa[attr][0][vert];
137            vOut[attr * 4 + 1] = outputs_soa[attr][1][vert];
138            vOut[attr * 4 + 2] = outputs_soa[attr][2][vert];
139            vOut[attr * 4 + 3] = outputs_soa[attr][3][vert];
140#if 0
141            if (attr==0)
142               printf("Output v%d a%d: %f %f %f %f\n",
143                      vert, attr, vOut[0], vOut[1], vOut[2], vOut[3]);
144#endif
145            vOut += output_stride / 4;
146         }
147      }
148
149      /* advance to next group of four input/output verts */
150      input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
151      output = (float (*)[4])((char *)output + output_stride * max_vertices);
152   }
153}
154
155
156static void
157vs_ppc_delete( struct draw_vertex_shader *base )
158{
159   struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
160
161   ppc_release_func( &shader->ppc_program );
162
163   align_free( (void *) shader->base.immediates );
164
165   FREE( (void*) shader->base.state.tokens );
166   FREE( shader );
167}
168
169
170struct draw_vertex_shader *
171draw_create_vs_ppc(struct draw_context *draw,
172                   const struct pipe_shader_state *templ)
173{
174   struct draw_ppc_vertex_shader *vs;
175
176   vs = CALLOC_STRUCT( draw_ppc_vertex_shader );
177   if (vs == NULL)
178      return NULL;
179
180   /* we make a private copy of the tokens */
181   vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
182   if (!vs->base.state.tokens)
183      goto fail;
184
185   tgsi_scan_shader(templ->tokens, &vs->base.info);
186
187   vs->base.draw = draw;
188   vs->base.create_variant = draw_vs_create_variant_generic;
189   vs->base.prepare = vs_ppc_prepare;
190   vs->base.run_linear = vs_ppc_run_linear;
191   vs->base.delete = vs_ppc_delete;
192
193   vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
194                                      sizeof(float), 16);
195
196   ppc_init_func( &vs->ppc_program );
197
198#if 0
199   ppc_print_code(&vs->ppc_program, TRUE);
200   ppc_indent(&vs->ppc_program, 8);
201#endif
202
203   if (!tgsi_emit_ppc( (struct tgsi_token *) vs->base.state.tokens,
204			&vs->ppc_program,
205                       (float (*)[4]) vs->base.immediates,
206                        TRUE ))
207      goto fail;
208
209   vs->func = (codegen_function) ppc_get_func( &vs->ppc_program );
210   if (!vs->func) {
211      goto fail;
212   }
213
214   return &vs->base;
215
216fail:
217   /*
218   debug_error("tgsi_emit_ppc() failed, falling back to interpreter\n");
219   */
220
221   ppc_release_func( &vs->ppc_program );
222
223   FREE(vs);
224   return NULL;
225}
226
227
228
229#else /* PIPE_ARCH_PPC */
230
231
232struct draw_vertex_shader *
233draw_create_vs_ppc( struct draw_context *draw,
234		    const struct pipe_shader_state *templ )
235{
236   return (void *) 0;
237}
238
239
240#endif /* PIPE_ARCH_PPC */
241