summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
blob: f2432f4ff520f0aa8da1b556b48b1987cb1a05be (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
/*
 * (C) Copyright IBM Corporation 2008
 * 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
 * on 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
 * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR 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.
 */

#include <inttypes.h>
#include "pipe/p_defines.h"
#include "pipe/p_context.h"
#include "pipe/p_format.h"

#include "pipe/draw/draw_context.h"
#include "pipe/draw/draw_private.h"

#include "pipe/cell/ppu/cell_context.h"
#include "ppc/rtasm/spe_asm.h"

typedef uint64_t register_mask;

int allocate_available_register(register_mask *m)
{
   unsigned i;
   for (i = 0; i < 64; i++) {
      const uint64_t mask = (1ULL << i);

      if ((m[0] & mask) != 0) {
	 m[0] &= ~mask;
	 return i;
      }
   }

   return -1;
}


int allocate_register(register_mask *m, unsigned reg)
{
   assert((m[0] & (1ULL << reg)) != 0);

   m[0] &= ~(1ULL << reg);
   return reg;
}


void release_register(register_mask *m, unsigned reg)
{
   assert((m[0] & (1ULL << reg)) == 0);

   m[0] |= (1ULL << reg);
}


/**
 * Emit a 4x4 matrix transpose operation
 *
 * \param p         Function that the transpose operation is to be appended to
 * \param m         Live register mask
 * \param row0      Register containing row 0 of the source matrix
 * \param row1      Register containing row 1 of the source matrix
 * \param row2      Register containing row 2 of the source matrix
 * \param row3      Register containing row 3 of the source matrix
 * \param dest_ptr  Register containing the address of the destination matrix
 * \param shuf_ptr  Register containing the address of the shuffled data
 * \param count     Number of colums to actually be written to the destination
 *
 * \note
 * This function assumes that the registers named by \c row0, \c row1,
 * \c row2, and \c row3 are scratch and can be modified by the generated code.
 * Furthermore, these registers will be released, via calls to
 * \c release_register, by this function.
 * 
 * \note
 * This function requires that four temporary are available on entry.
 */
static void
emit_matrix_transpose(struct spe_function *p, register_mask *m,
		      unsigned row0, unsigned row1, unsigned row2,
		      unsigned row3, unsigned dest_ptr,
		      unsigned shuf_ptr, unsigned count)
{
   int shuf_hi = allocate_available_register(m);
   int shuf_lo = allocate_available_register(m);
   int t1 = allocate_available_register(m);
   int t2 = allocate_available_register(m);
   int t3;
   int t4;
   int col0;
   int col1;
   int col2;
   int col3;


   spe_lqd(p, shuf_hi, shuf_ptr, 3);
   spe_lqd(p, shuf_lo, shuf_ptr, 4);
   spe_shufb(p, t1, row0, row2, shuf_hi);
   spe_shufb(p, t2, row0, row2, shuf_lo);


   /* row0 and row2 are now no longer needed.  Re-use those registers as
    * temporaries.
    */
   t3 = row0;
   t4 = row2;

   spe_shufb(p, t3, row1, row3, shuf_hi);
   spe_shufb(p, t4, row1, row3, shuf_lo);


   /* row1 and row3 are now no longer needed.  Re-use those registers as
    * temporaries.
    */
   col0 = row1;
   col1 = row3;

   spe_shufb(p, col0, t1, t3, shuf_hi);
   if (count > 1) {
      spe_shufb(p, col1, t1, t3, shuf_lo);
   }

   /* t1 and t3 are now no longer needed.  Re-use those registers as
    * temporaries.
    */
   col2 = t1;
   col3 = t3;

   if (count > 2) {
      spe_shufb(p, col2, t2, t4, shuf_hi);
   }

   if (count > 3) {
      spe_shufb(p, col3, t2, t4, shuf_lo);
   }


   /* Store the results.  Remember that the stqd instruction is encoded using
    * the qword offset (stand-alone assemblers to the byte-offset to
    * qword-offset conversion for you), so the byte-offset needs be divided by
    * 16.
    */
   switch (count) {
   case 4:
      spe_stqd(p, col3, dest_ptr, 3);
   case 3:
      spe_stqd(p, col2, dest_ptr, 2);
   case 2:
      spe_stqd(p, col1, dest_ptr, 1);
   case 1:
      spe_stqd(p, col0, dest_ptr, 0);
   }


   /* Release all of the temporary registers used.
    */
   release_register(m, col0);
   release_register(m, col1);
   release_register(m, col2);
   release_register(m, col3);
   release_register(m, shuf_hi);
   release_register(m, shuf_lo);
   release_register(m, t2);
   release_register(m, t4);
}


static void
emit_fetch(struct spe_function *p, register_mask *m,
	   unsigned in_ptr, unsigned *offset,
	   unsigned out_ptr, unsigned shuf_ptr,
	   enum pipe_format format)
{
   const unsigned count = (pf_size_x(format) != 0) + (pf_size_y(format) != 0)
       + (pf_size_z(format) != 0) + (pf_size_w(format) != 0);
   const unsigned type = pf_type(format);
   const unsigned bytes = pf_size_x(format);

   int v0 = allocate_available_register(m);
   int v1 = allocate_available_register(m);
   int v2 = allocate_available_register(m);
   int v3 = allocate_available_register(m);
   int tmp = allocate_available_register(m);
   int float_zero = -1;
   int float_one = -1;
   float scale_signed = 0.0;
   float scale_unsigned = 0.0;

   spe_lqd(p, v0, in_ptr, 0 + offset[0]);
   spe_lqd(p, v1, in_ptr, 1 + offset[0]);
   spe_lqd(p, v2, in_ptr, 2 + offset[0]);
   spe_lqd(p, v3, in_ptr, 3 + offset[0]);
   offset[0] += 4;
   
   switch (bytes) {
   case 1:
      scale_signed = 1.0f / 127.0f;
      scale_unsigned = 1.0f / 255.0f;
      spe_lqd(p, tmp, shuf_ptr, 1);
      spe_shufb(p, v0, v0, v0, tmp);
      spe_shufb(p, v1, v1, v1, tmp);
      spe_shufb(p, v2, v2, v2, tmp);
      spe_shufb(p, v3, v3, v3, tmp);
      break;
   case 2:
      scale_signed = 1.0f / 32767.0f;
      scale_unsigned = 1.0f / 65535.0f;
      spe_lqd(p, tmp, shuf_ptr, 2);
      spe_shufb(p, v0, v0, v0, tmp);
      spe_shufb(p, v1, v1, v1, tmp);
      spe_shufb(p, v2, v2, v2, tmp);
      spe_shufb(p, v3, v3, v3, tmp);
      break;
   case 4:
      scale_signed = 1.0f / 2147483647.0f;
      scale_unsigned = 1.0f / 4294967295.0f;
      break;
   default:
      assert(0);
      break;
   }

   switch (type) {
   case PIPE_FORMAT_TYPE_FLOAT:
      break;
   case PIPE_FORMAT_TYPE_UNORM:
      spe_ilhu(p, tmp, ((unsigned) scale_unsigned) >> 16);
      spe_iohl(p, tmp, ((unsigned) scale_unsigned) & 0x0ffff);
      spe_cuflt(p, v0, v0, 0);
      spe_fm(p, v0, v0, tmp);
      break;
   case PIPE_FORMAT_TYPE_SNORM:
      spe_ilhu(p, tmp, ((unsigned) scale_signed) >> 16);
      spe_iohl(p, tmp, ((unsigned) scale_signed) & 0x0ffff);
      spe_csflt(p, v0, v0, 0);
      spe_fm(p, v0, v0, tmp);
      break;
   case PIPE_FORMAT_TYPE_USCALED:
      spe_cuflt(p, v0, v0, 0);
      break;
   case PIPE_FORMAT_TYPE_SSCALED:
      spe_csflt(p, v0, v0, 0);
      break;
   }


   if (count < 4) {
      float_one = allocate_available_register(m);
      spe_il(p, float_one, 1);
      spe_cuflt(p, float_one, float_one, 0);
      
      if (count < 3) {
	 float_zero = allocate_available_register(m);
	 spe_il(p, float_zero, 0);
      }
   }

   release_register(m, tmp);

   emit_matrix_transpose(p, m, v0, v1, v2, v3, out_ptr, shuf_ptr, count);

   switch (count) {
   case 1:
      spe_stqd(p, float_zero, out_ptr, 1);
   case 2:
      spe_stqd(p, float_zero, out_ptr, 2);
   case 3:
      spe_stqd(p, float_one, out_ptr, 3);
   }

   if (float_zero != -1) {
      release_register(m, float_zero);
   }

   if (float_one != -1) {
      release_register(m, float_one);
   }
}


void cell_update_vertex_fetch(struct draw_context *draw)
{
   struct cell_context *const cell =
       (struct cell_context *) draw->driver_private;
   register_mask m = ~0;
   struct spe_function *p = &cell->attrib_fetch;
   unsigned function_index[PIPE_ATTRIB_MAX];
   unsigned unique_attr_formats;
   int out_ptr;
   int in_ptr;
   int shuf_ptr;
   unsigned i;
   unsigned j;


   /* Determine how many unique input attribute formats there are.  At the
    * same time, store the index of the lowest numbered attribute that has
    * the same format as any non-unique format.
    */
   unique_attr_formats = 1;
   function_index[0] = 0;
   for (i = 1; i < draw->vertex_fetch.nr_attrs; i++) {
      const enum pipe_format curr_fmt = draw->vertex_element[i].src_format;

      for (j = 0; j < i; j++) {
	 if (curr_fmt == draw->vertex_element[j].src_format) {
	    break;
	 }
      }
      
      if (j == i) {
	 unique_attr_formats++;
      }

      function_index[i] = j;
   }


   /* Each fetch function can be a maximum of 34 instructions (note: this is
    * actually a slight over-estimate).  That means (34 * 4) = 136 bytes
    * each maximum.
    */
   spe_init_func(p, 136 * unique_attr_formats);


   /* Registers 0, 1, and 2 are reserved by the ABI.
    */
   allocate_register(&m, 0);
   allocate_register(&m, 1);
   allocate_register(&m, 2);


   /* Allocate registers for the function's input parameters.
    */
   out_ptr = allocate_register(&m, 3);
   in_ptr = allocate_register(&m, 4);
   shuf_ptr = allocate_register(&m, 5);


   /* Generate code for the individual attribute fetch functions.
    */
   for (i = 0; i < draw->vertex_fetch.nr_attrs; i++) {
      unsigned offset;

      if (function_index[i] == i) {
	 cell->attrib_fetch_offsets[i] = (unsigned) ((void *) p->csr 
						     - (void *) p->store);

	 offset = 0;
	 emit_fetch(p, & m, in_ptr, &offset, out_ptr, shuf_ptr,
		    draw->vertex_element[i].src_format);
	 spe_bi(p, 0, 0, 0);

	 /* Round up to the next 16-byte boundary.
	  */
	 if ((((unsigned) p->store) & 0x0f) != 0) {
	    const unsigned align = ((unsigned) p->store) & 0x0f;
	    p->store = (uint32_t *) (((void *) p->store) + align);
	 }
      } else {
	 /* Use the same function entry-point as a previously seen attribute
	  * with the same format.
	  */
	 cell->attrib_fetch_offsets[i] = 
	     cell->attrib_fetch_offsets[function_index[i]];
      }
   }
   
   static first_time = 1;
   if (first_time) {
      first_time = 0;
      const unsigned instructions = p->csr - p->store;
      for (i = 0; i < instructions; i++) {
	 printf("\t.long\t0x%08x\n", p->store[i]);
      }
   }
}