1/*
2 * Copyright (c) 2012-2015 Etnaviv Project
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sub license,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the
12 * next paragraph) shall be included in all copies or substantial portions
13 * of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 *    Wladimir J. van der Laan <laanwj@gmail.com>
25 */
26
27#include "etnaviv_shader.h"
28
29#include "etnaviv_compiler.h"
30#include "etnaviv_context.h"
31#include "etnaviv_debug.h"
32#include "etnaviv_disasm.h"
33#include "etnaviv_disk_cache.h"
34#include "etnaviv_screen.h"
35#include "etnaviv_util.h"
36
37#include "tgsi/tgsi_parse.h"
38#include "nir/tgsi_to_nir.h"
39#include "util/u_atomic.h"
40#include "util/u_math.h"
41#include "util/u_memory.h"
42
43/* Upload shader code to bo, if not already done */
44static bool etna_icache_upload_shader(struct etna_context *ctx, struct etna_shader_variant *v)
45{
46   if (v->bo)
47      return true;
48   v->bo = etna_bo_new(ctx->screen->dev, v->code_size*4, DRM_ETNA_GEM_CACHE_WC);
49   if (!v->bo)
50      return false;
51
52   void *buf = etna_bo_map(v->bo);
53   etna_bo_cpu_prep(v->bo, DRM_ETNA_PREP_WRITE);
54   memcpy(buf, v->code, v->code_size*4);
55   etna_bo_cpu_fini(v->bo);
56   DBG("Uploaded %s of %u words to bo %p", v->stage == MESA_SHADER_FRAGMENT ? "fs":"vs", v->code_size, v->bo);
57   return true;
58}
59
60extern const char *tgsi_swizzle_names[];
61void
62etna_dump_shader(const struct etna_shader_variant *shader)
63{
64   if (shader->stage == MESA_SHADER_VERTEX)
65      printf("VERT\n");
66   else
67      printf("FRAG\n");
68
69   etna_disasm(shader->code, shader->code_size, PRINT_RAW);
70
71   printf("num loops: %i\n", shader->num_loops);
72   printf("num temps: %i\n", shader->num_temps);
73   printf("immediates:\n");
74   for (int idx = 0; idx < shader->uniforms.count; ++idx) {
75      printf(" [%i].%s = %f (0x%08x) (%d)\n",
76             idx / 4,
77             tgsi_swizzle_names[idx % 4],
78             *((float *)&shader->uniforms.data[idx]),
79             shader->uniforms.data[idx],
80             shader->uniforms.contents[idx]);
81   }
82   printf("inputs:\n");
83   for (int idx = 0; idx < shader->infile.num_reg; ++idx) {
84      printf(" [%i] name=%s comps=%i\n", shader->infile.reg[idx].reg,
85               (shader->stage == MESA_SHADER_VERTEX) ?
86               gl_vert_attrib_name(shader->infile.reg[idx].slot) :
87               gl_varying_slot_name_for_stage(shader->infile.reg[idx].slot, shader->stage),
88               shader->infile.reg[idx].num_components);
89   }
90   printf("outputs:\n");
91   for (int idx = 0; idx < shader->outfile.num_reg; ++idx) {
92      printf(" [%i] name=%s comps=%i\n", shader->outfile.reg[idx].reg,
93               (shader->stage == MESA_SHADER_VERTEX) ?
94               gl_varying_slot_name_for_stage(shader->outfile.reg[idx].slot, shader->stage) :
95               gl_frag_result_name(shader->outfile.reg[idx].slot),
96               shader->outfile.reg[idx].num_components);
97   }
98   printf("special:\n");
99   if (shader->stage == MESA_SHADER_VERTEX) {
100      printf("  vs_pos_out_reg=%i\n", shader->vs_pos_out_reg);
101      printf("  vs_pointsize_out_reg=%i\n", shader->vs_pointsize_out_reg);
102      printf("  vs_load_balancing=0x%08x\n", shader->vs_load_balancing);
103   } else {
104      printf("  ps_color_out_reg=%i\n", shader->ps_color_out_reg);
105      printf("  ps_depth_out_reg=%i\n", shader->ps_depth_out_reg);
106   }
107   printf("  input_count_unk8=0x%08x\n", shader->input_count_unk8);
108}
109
110/* Link vs and fs together: fill in shader_state from vs and fs
111 * as this function is called every time a new fs or vs is bound, the goal is to
112 * do little processing as possible here, and to precompute as much as possible in
113 * the vs/fs shader_object.
114 *
115 * XXX we could cache the link result for a certain set of VS/PS; usually a pair
116 * of VS and PS will be used together anyway.
117 */
118static bool
119etna_link_shaders(struct etna_context *ctx, struct compiled_shader_state *cs,
120                  struct etna_shader_variant *vs, struct etna_shader_variant *fs)
121{
122   struct etna_shader_link_info link = { };
123   bool failed;
124
125   assert(vs->stage == MESA_SHADER_VERTEX);
126   assert(fs->stage == MESA_SHADER_FRAGMENT);
127
128#ifdef DEBUG
129   if (DBG_ENABLED(ETNA_DBG_DUMP_SHADERS)) {
130      etna_dump_shader(vs);
131      etna_dump_shader(fs);
132   }
133#endif
134
135   if (DBG_ENABLED(ETNA_DBG_NIR))
136      failed = etna_link_shader_nir(&link, vs, fs);
137   else
138      failed = etna_link_shader(&link, vs, fs);
139
140   if (failed) {
141      /* linking failed: some fs inputs do not have corresponding
142       * vs outputs */
143      assert(0);
144
145      return false;
146   }
147
148   if (DBG_ENABLED(ETNA_DBG_LINKER_MSGS)) {
149      debug_printf("link result:\n");
150      debug_printf("  vs  -> fs  comps use     pa_attr\n");
151
152      for (int idx = 0; idx < link.num_varyings; ++idx)
153         debug_printf("  t%-2u -> t%-2u %-5.*s %u,%u,%u,%u 0x%08x\n",
154                      link.varyings[idx].reg, idx + 1,
155                      link.varyings[idx].num_components, "xyzw",
156                      link.varyings[idx].use[0], link.varyings[idx].use[1],
157                      link.varyings[idx].use[2], link.varyings[idx].use[3],
158                      link.varyings[idx].pa_attributes);
159   }
160
161   /* set last_varying_2x flag if the last varying has 1 or 2 components */
162   bool last_varying_2x = false;
163   if (link.num_varyings > 0 && link.varyings[link.num_varyings - 1].num_components <= 2)
164      last_varying_2x = true;
165
166   cs->RA_CONTROL = VIVS_RA_CONTROL_UNK0 |
167                    COND(last_varying_2x, VIVS_RA_CONTROL_LAST_VARYING_2X);
168
169   cs->PA_ATTRIBUTE_ELEMENT_COUNT = VIVS_PA_ATTRIBUTE_ELEMENT_COUNT_COUNT(link.num_varyings);
170   for (int idx = 0; idx < link.num_varyings; ++idx)
171      cs->PA_SHADER_ATTRIBUTES[idx] = link.varyings[idx].pa_attributes;
172
173   cs->VS_END_PC = vs->code_size / 4;
174   cs->VS_OUTPUT_COUNT = 1 + link.num_varyings; /* position + varyings */
175
176   /* vs outputs (varyings) */
177   DEFINE_ETNA_BITARRAY(vs_output, 16, 8) = {0};
178   int varid = 0;
179   etna_bitarray_set(vs_output, 8, varid++, vs->vs_pos_out_reg);
180   for (int idx = 0; idx < link.num_varyings; ++idx)
181      etna_bitarray_set(vs_output, 8, varid++, link.varyings[idx].reg);
182   if (vs->vs_pointsize_out_reg >= 0)
183      etna_bitarray_set(vs_output, 8, varid++, vs->vs_pointsize_out_reg); /* pointsize is last */
184
185   for (int idx = 0; idx < ARRAY_SIZE(cs->VS_OUTPUT); ++idx)
186      cs->VS_OUTPUT[idx] = vs_output[idx];
187
188   if (vs->vs_pointsize_out_reg != -1) {
189      /* vertex shader outputs point coordinate, provide extra output and make
190       * sure PA config is
191       * not masked */
192      cs->PA_CONFIG = ~0;
193      cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT + 1;
194   } else {
195      /* vertex shader does not output point coordinate, make sure thate
196       * POINT_SIZE_ENABLE is masked
197       * and no extra output is given */
198      cs->PA_CONFIG = ~VIVS_PA_CONFIG_POINT_SIZE_ENABLE;
199      cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT;
200   }
201
202   /* if fragment shader doesn't read pointcoord, disable it */
203   if (link.pcoord_varying_comp_ofs == -1)
204      cs->PA_CONFIG &= ~VIVS_PA_CONFIG_POINT_SPRITE_ENABLE;
205
206   cs->VS_LOAD_BALANCING = vs->vs_load_balancing;
207   cs->VS_START_PC = 0;
208
209   cs->PS_END_PC = fs->code_size / 4;
210   cs->PS_OUTPUT_REG = fs->ps_color_out_reg;
211   cs->PS_INPUT_COUNT =
212      VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 1) | /* Number of inputs plus position */
213      VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
214   cs->PS_TEMP_REGISTER_CONTROL =
215      VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 1));
216   cs->PS_START_PC = 0;
217
218   /* Precompute PS_INPUT_COUNT and TEMP_REGISTER_CONTROL in the case of MSAA
219    * mode, avoids some fumbling in sync_context. */
220   cs->PS_INPUT_COUNT_MSAA =
221      VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 2) | /* MSAA adds another input */
222      VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
223   cs->PS_TEMP_REGISTER_CONTROL_MSAA =
224      VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 2));
225
226   uint32_t total_components = 0;
227   DEFINE_ETNA_BITARRAY(num_components, ETNA_NUM_VARYINGS, 4) = {0};
228   DEFINE_ETNA_BITARRAY(component_use, 4 * ETNA_NUM_VARYINGS, 2) = {0};
229   for (int idx = 0; idx < link.num_varyings; ++idx) {
230      const struct etna_varying *varying = &link.varyings[idx];
231
232      etna_bitarray_set(num_components, 4, idx, varying->num_components);
233      for (int comp = 0; comp < varying->num_components; ++comp) {
234         etna_bitarray_set(component_use, 2, total_components, varying->use[comp]);
235         total_components += 1;
236      }
237   }
238
239   cs->GL_VARYING_TOTAL_COMPONENTS =
240      VIVS_GL_VARYING_TOTAL_COMPONENTS_NUM(align(total_components, 2));
241   cs->GL_VARYING_NUM_COMPONENTS[0] = num_components[0];
242   cs->GL_VARYING_NUM_COMPONENTS[1] = num_components[1];
243   cs->GL_VARYING_COMPONENT_USE[0] = component_use[0];
244   cs->GL_VARYING_COMPONENT_USE[1] = component_use[1];
245
246   cs->GL_HALTI5_SH_SPECIALS =
247      0x7f7f0000 | /* unknown bits, probably other PS inputs */
248      /* pointsize is last (see above) */
249      VIVS_GL_HALTI5_SH_SPECIALS_VS_PSIZE_OUT((vs->vs_pointsize_out_reg != -1) ?
250                                              cs->VS_OUTPUT_COUNT * 4 : 0x00) |
251      VIVS_GL_HALTI5_SH_SPECIALS_PS_PCOORD_IN((link.pcoord_varying_comp_ofs != -1) ?
252                                              link.pcoord_varying_comp_ofs : 0x7f);
253
254   cs->writes_z = fs->ps_depth_out_reg >= 0;
255   cs->uses_discard = fs->uses_discard;
256
257   /* reference instruction memory */
258   cs->vs_inst_mem_size = vs->code_size;
259   cs->VS_INST_MEM = vs->code;
260
261   cs->ps_inst_mem_size = fs->code_size;
262   cs->PS_INST_MEM = fs->code;
263
264   if (vs->needs_icache || fs->needs_icache) {
265      /* If either of the shaders needs ICACHE, we use it for both. It is
266       * either switched on or off for the entire shader processor.
267       */
268      if (!etna_icache_upload_shader(ctx, vs) ||
269          !etna_icache_upload_shader(ctx, fs)) {
270         assert(0);
271         return false;
272      }
273
274      cs->VS_INST_ADDR.bo = vs->bo;
275      cs->VS_INST_ADDR.offset = 0;
276      cs->VS_INST_ADDR.flags = ETNA_RELOC_READ;
277      cs->PS_INST_ADDR.bo = fs->bo;
278      cs->PS_INST_ADDR.offset = 0;
279      cs->PS_INST_ADDR.flags = ETNA_RELOC_READ;
280   } else {
281      /* clear relocs */
282      memset(&cs->VS_INST_ADDR, 0, sizeof(cs->VS_INST_ADDR));
283      memset(&cs->PS_INST_ADDR, 0, sizeof(cs->PS_INST_ADDR));
284   }
285
286   return true;
287}
288
289bool
290etna_shader_link(struct etna_context *ctx)
291{
292   if (!ctx->shader.vs || !ctx->shader.fs)
293      return false;
294
295   /* re-link vs and fs if needed */
296   return etna_link_shaders(ctx, &ctx->shader_state, ctx->shader.vs, ctx->shader.fs);
297}
298
299void
300etna_destroy_shader(struct etna_shader_variant *shader)
301{
302   assert(shader);
303
304   FREE(shader->code);
305   FREE(shader->uniforms.data);
306   FREE(shader->uniforms.contents);
307   FREE(shader);
308}
309
310static bool
311etna_shader_update_vs_inputs(struct compiled_shader_state *cs,
312                             const struct etna_shader_variant *vs,
313                             const struct compiled_vertex_elements_state *ves)
314{
315   unsigned num_temps, cur_temp, num_vs_inputs;
316
317   if (!vs)
318      return false;
319
320   /* Number of vertex elements determines number of VS inputs. Otherwise,
321    * the GPU crashes. Allocate any unused vertex elements to VS temporary
322    * registers. */
323   num_vs_inputs = MAX2(ves->num_elements, vs->infile.num_reg);
324   if (num_vs_inputs != ves->num_elements) {
325      BUG("Number of elements %u does not match the number of VS inputs %zu",
326          ves->num_elements, vs->infile.num_reg);
327      return false;
328   }
329
330   cur_temp = vs->num_temps;
331   num_temps = num_vs_inputs - vs->infile.num_reg + cur_temp;
332
333   cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs) |
334                        VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8);
335   cs->VS_TEMP_REGISTER_CONTROL =
336      VIVS_VS_TEMP_REGISTER_CONTROL_NUM_TEMPS(num_temps);
337
338   /* vs inputs (attributes) */
339   DEFINE_ETNA_BITARRAY(vs_input, 16, 8) = {0};
340   for (int idx = 0; idx < num_vs_inputs; ++idx) {
341      if (idx < vs->infile.num_reg)
342         etna_bitarray_set(vs_input, 8, idx, vs->infile.reg[idx].reg);
343      else
344         etna_bitarray_set(vs_input, 8, idx, cur_temp++);
345   }
346
347   if (vs->vs_id_in_reg >= 0) {
348      cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs + 1) |
349                           VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8) |
350                           VIVS_VS_INPUT_COUNT_ID_ENABLE;
351
352      etna_bitarray_set(vs_input, 8, num_vs_inputs, vs->vs_id_in_reg);
353
354      cs->FE_HALTI5_ID_CONFIG =
355         VIVS_FE_HALTI5_ID_CONFIG_VERTEX_ID_ENABLE |
356         VIVS_FE_HALTI5_ID_CONFIG_INSTANCE_ID_ENABLE |
357         VIVS_FE_HALTI5_ID_CONFIG_VERTEX_ID_REG(vs->vs_id_in_reg * 4) |
358         VIVS_FE_HALTI5_ID_CONFIG_INSTANCE_ID_REG(vs->vs_id_in_reg * 4 + 1);
359   }
360
361   for (int idx = 0; idx < ARRAY_SIZE(cs->VS_INPUT); ++idx)
362      cs->VS_INPUT[idx] = vs_input[idx];
363
364   return true;
365}
366
367static inline const char *
368etna_shader_stage(struct etna_shader_variant *shader)
369{
370   switch (shader->stage) {
371   case MESA_SHADER_VERTEX:     return "VERT";
372   case MESA_SHADER_FRAGMENT:   return "FRAG";
373   case MESA_SHADER_COMPUTE:    return "CL";
374   default:
375      unreachable("invalid type");
376      return NULL;
377   }
378}
379
380static void
381dump_shader_info(struct etna_shader_variant *v, struct pipe_debug_callback *debug)
382{
383   if (!unlikely(etna_mesa_debug & ETNA_DBG_SHADERDB))
384      return;
385
386   pipe_debug_message(debug, SHADER_INFO,
387         "%s shader: %u instructions, %u temps, "
388         "%u immediates, %u loops",
389         etna_shader_stage(v),
390         v->code_size,
391         v->num_temps,
392         v->uniforms.count,
393         v->num_loops);
394}
395
396bool
397etna_shader_update_vertex(struct etna_context *ctx)
398{
399   return etna_shader_update_vs_inputs(&ctx->shader_state, ctx->shader.vs,
400                                       ctx->vertex_elements);
401}
402
403static struct etna_shader_variant *
404create_variant(struct etna_shader *shader, struct etna_shader_key key)
405{
406   struct etna_shader_variant *v = CALLOC_STRUCT(etna_shader_variant);
407   int ret;
408
409   if (!v)
410      return NULL;
411
412   v->shader = shader;
413   v->key = key;
414   v->id = ++shader->variant_count;
415
416   if (etna_disk_cache_retrieve(shader->compiler, v))
417      return v;
418
419   ret = etna_compile_shader(v);
420   if (!ret) {
421      debug_error("compile failed!");
422      goto fail;
423   }
424
425   etna_disk_cache_store(shader->compiler, v);
426
427   return v;
428
429fail:
430   FREE(v);
431   return NULL;
432}
433
434struct etna_shader_variant *
435etna_shader_variant(struct etna_shader *shader, struct etna_shader_key key,
436                   struct pipe_debug_callback *debug)
437{
438   struct etna_shader_variant *v;
439
440   for (v = shader->variants; v; v = v->next)
441      if (etna_shader_key_equal(&key, &v->key))
442         return v;
443
444   /* compile new variant if it doesn't exist already */
445   v = create_variant(shader, key);
446   if (v) {
447      v->next = shader->variants;
448      shader->variants = v;
449      dump_shader_info(v, debug);
450   }
451
452   return v;
453}
454
455static void *
456etna_create_shader_state(struct pipe_context *pctx,
457                         const struct pipe_shader_state *pss)
458{
459   struct etna_context *ctx = etna_context(pctx);
460   struct etna_screen *screen = ctx->screen;
461   struct etna_compiler *compiler = screen->compiler;
462   struct etna_shader *shader = CALLOC_STRUCT(etna_shader);
463
464   if (!shader)
465      return NULL;
466
467   shader->id = p_atomic_inc_return(&compiler->shader_count);
468   shader->specs = &screen->specs;
469   shader->compiler = screen->compiler;
470
471   if (DBG_ENABLED(ETNA_DBG_NIR))
472      shader->nir = (pss->type == PIPE_SHADER_IR_NIR) ? pss->ir.nir :
473                     tgsi_to_nir(pss->tokens, pctx->screen, false);
474   else
475      shader->tokens = tgsi_dup_tokens(pss->tokens);
476
477   etna_disk_cache_init_shader_key(compiler, shader);
478
479   if (etna_mesa_debug & ETNA_DBG_SHADERDB) {
480      /* if shader-db run, create a standard variant immediately
481       * (as otherwise nothing will trigger the shader to be
482       * actually compiled).
483       */
484      struct etna_shader_key key = {};
485      etna_shader_variant(shader, key, &ctx->debug);
486   }
487
488   return shader;
489}
490
491static void
492etna_delete_shader_state(struct pipe_context *pctx, void *ss)
493{
494   struct etna_shader *shader = ss;
495   struct etna_shader_variant *v, *t;
496
497   v = shader->variants;
498   while (v) {
499      t = v;
500      v = v->next;
501      if (t->bo)
502         etna_bo_del(t->bo);
503
504      etna_destroy_shader(t);
505   }
506
507   tgsi_free_tokens(shader->tokens);
508   ralloc_free(shader->nir);
509   FREE(shader);
510}
511
512static void
513etna_bind_fs_state(struct pipe_context *pctx, void *hwcso)
514{
515   struct etna_context *ctx = etna_context(pctx);
516
517   ctx->shader.bind_fs = hwcso;
518   ctx->dirty |= ETNA_DIRTY_SHADER;
519}
520
521static void
522etna_bind_vs_state(struct pipe_context *pctx, void *hwcso)
523{
524   struct etna_context *ctx = etna_context(pctx);
525
526   ctx->shader.bind_vs = hwcso;
527   ctx->dirty |= ETNA_DIRTY_SHADER;
528}
529
530void
531etna_shader_init(struct pipe_context *pctx)
532{
533   pctx->create_fs_state = etna_create_shader_state;
534   pctx->bind_fs_state = etna_bind_fs_state;
535   pctx->delete_fs_state = etna_delete_shader_state;
536   pctx->create_vs_state = etna_create_shader_state;
537   pctx->bind_vs_state = etna_bind_vs_state;
538   pctx->delete_vs_state = etna_delete_shader_state;
539}
540