1/**********************************************************
2 * Copyright 2008-2022 VMware, Inc.  All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26#include "util/format/u_format.h"
27#include "util/u_bitmask.h"
28#include "util/u_inlines.h"
29#include "util/u_memory.h"
30#include "pipe/p_defines.h"
31#include "util/u_upload_mgr.h"
32
33#include "svga_screen.h"
34#include "svga_context.h"
35#include "svga_state.h"
36#include "svga_cmd.h"
37#include "svga_tgsi.h"
38#include "svga_debug.h"
39#include "svga_resource_buffer.h"
40#include "svga_shader.h"
41
42#include "svga_hw_reg.h"
43
44
45static unsigned
46svga_get_image_size_constant(const struct svga_context *svga, float **dest,
47                             enum pipe_shader_type shader,
48                             unsigned num_image_views,
49                             const struct svga_image_view images[PIPE_SHADER_TYPES][SVGA3D_MAX_UAVIEWS])
50{
51   uint32_t *dest_u = (uint32_t *) *dest;
52
53   for (int i = 0; i < num_image_views; i++) {
54      if (images[shader][i].desc.resource) {
55         if (images[shader][i].desc.resource->target == PIPE_BUFFER) {
56            unsigned bytes_per_element = util_format_get_blocksize(images[shader][i].desc.format);
57            *dest_u++ = images[shader][i].desc.resource->width0 / bytes_per_element;
58         }
59         else
60            *dest_u++ = images[shader][i].desc.resource->width0;
61
62         if (images[shader][i].desc.resource->target == PIPE_TEXTURE_1D_ARRAY)
63            *dest_u++ = images[shader][i].desc.resource->array_size;
64         else
65            *dest_u++ = images[shader][i].desc.resource->height0;
66
67         if (images[shader][i].desc.resource->target == PIPE_TEXTURE_2D_ARRAY)
68            *dest_u++ = images[shader][i].desc.resource->array_size;
69         else if (images[shader][i].desc.resource->target == PIPE_TEXTURE_CUBE_ARRAY)
70            *dest_u++ = images[shader][i].desc.resource->array_size / 6;
71         else
72            *dest_u++ = images[shader][i].desc.resource->depth0;
73         *dest_u++ = 1; // Later this can be used for sample counts
74      }
75      else {
76         *dest_u += 4;
77      }
78   }
79   return num_image_views;
80}
81
82
83/*
84 * Don't try to send more than 4kb of successive constants.
85 */
86#define MAX_CONST_REG_COUNT 256  /**< number of float[4] constants */
87
88/**
89 * Extra space for svga-specific VS/PS constants (such as texcoord
90 * scale factors, vertex transformation scale/translation).
91 */
92#define MAX_EXTRA_CONSTS 32
93
94/** Guest-backed surface constant buffers must be this size */
95#define GB_CONSTBUF_SIZE (SVGA3D_CONSTREG_MAX)
96
97
98/**
99 * Emit any extra shader-type-independent shader constants into the buffer
100 * pointed to by 'dest'.
101 * \return number of float[4] constants put into the 'dest' buffer
102 */
103static unsigned
104svga_get_extra_constants_common(const struct svga_context *svga,
105                                const struct svga_shader_variant *variant,
106                                enum pipe_shader_type shader, float *dest)
107{
108   uint32_t *dest_u = (uint32_t *) dest;  // uint version of dest
109   unsigned i;
110   unsigned count = 0;
111
112   for (i = 0; i < variant->key.num_textures; i++) {
113      const struct pipe_sampler_view *sv = svga->curr.sampler_views[shader][i];
114      if (sv) {
115         const struct pipe_resource *tex = sv->texture;
116         /* Scaling factors needed for handling unnormalized texture coordinates
117          * for texture rectangles.
118          */
119         if (variant->key.tex[i].unnormalized) {
120            /* debug/sanity check */
121            assert(variant->key.tex[i].width_height_idx == count);
122
123            *dest++ = 1.0f / (float) tex->width0;
124            *dest++ = 1.0f / (float) tex->height0;
125            *dest++ = 1.0f;
126            *dest++ = 1.0f;
127
128            count++;
129         }
130
131         /* Store the sizes for texture buffers.
132         */
133         if (tex->target == PIPE_BUFFER) {
134            unsigned bytes_per_element = util_format_get_blocksize(sv->format);
135            *dest_u++ = tex->width0 / bytes_per_element;
136            *dest_u++ = 1;
137            *dest_u++ = 1;
138            *dest_u++ = 1;
139
140            count++;
141         }
142      }
143   }
144
145   /* image_size */
146   if (variant->key.image_size_used) {
147      count += svga_get_image_size_constant(svga, &dest, shader,
148                  svga->state.hw_draw.num_image_views[shader],
149                  svga->state.hw_draw.image_views);
150   }
151
152
153   return count;
154}
155
156
157/**
158 * Emit any extra fragment shader constants into the buffer pointed
159 * to by 'dest'.
160 * \return number of float[4] constants put into the dest buffer
161 */
162static unsigned
163svga_get_extra_fs_constants(const struct svga_context *svga, float *dest)
164{
165   const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
166   unsigned count = 0;
167
168   count += svga_get_extra_constants_common(svga, variant,
169                                            PIPE_SHADER_FRAGMENT, dest);
170
171   assert(count <= MAX_EXTRA_CONSTS);
172
173   return count;
174}
175
176/**
177 * Emit extra constants needed for prescale computation into the
178 * the buffer pointed to by '*dest'. The updated buffer pointer
179 * will be returned in 'dest'.
180 */
181static unsigned
182svga_get_prescale_constants(const struct svga_context *svga, float **dest,
183		            const struct svga_prescale *prescale)
184{
185   memcpy(*dest, prescale->scale, 4 * sizeof(float));
186   *dest += 4;
187
188   memcpy(*dest, prescale->translate, 4 * sizeof(float));
189   *dest += 4;
190
191   return 2;
192}
193
194/**
195 * Emit extra constants needed for point sprite emulation.
196 */
197static unsigned
198svga_get_pt_sprite_constants(const struct svga_context *svga, float **dest)
199{
200   const struct svga_screen *screen = svga_screen(svga->pipe.screen);
201   float *dst = *dest;
202
203   dst[0] = 1.0 / (svga->curr.viewport[0].scale[0] * 2);
204   dst[1] = 1.0 / (svga->curr.viewport[0].scale[1] * 2);
205   dst[2] = svga->curr.rast->pointsize;
206   dst[3] = screen->maxPointSize;
207   *dest = *dest + 4;
208   return 1;
209}
210
211/**
212 * Emit user-defined clip plane coefficients into the buffer pointed to
213 * by '*dest'. The updated buffer pointer will be returned in 'dest'.
214 */
215static unsigned
216svga_get_clip_plane_constants(const struct svga_context *svga,
217                              const struct svga_shader_variant *variant,
218                              float **dest)
219{
220   unsigned count = 0;
221
222   /* SVGA_NEW_CLIP */
223   if (svga_have_vgpu10(svga)) {
224      /* append user-defined clip plane coefficients onto constant buffer */
225      unsigned clip_planes = variant->key.clip_plane_enable;
226      while (clip_planes) {
227         int i = u_bit_scan(&clip_planes);
228         COPY_4V(*dest, svga->curr.clip.ucp[i]);
229         *dest += 4;
230         count += 1;
231      }
232   }
233   return count;
234}
235
236
237/**
238 * Emit any extra vertex shader constants into the buffer pointed
239 * to by 'dest'.
240 * In particular, these would be the scale and bias factors computed
241 * from the framebuffer size which are used to copy with differences in
242 * GL vs D3D coordinate spaces.  See svga_tgsi_insn.c for more info.
243 * \return number of float[4] constants put into the dest buffer
244 */
245static unsigned
246svga_get_extra_vs_constants(const struct svga_context *svga, float *dest)
247{
248   const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
249   unsigned count = 0;
250
251   /* SVGA_NEW_VS_VARIANT
252    */
253   if (variant->key.vs.need_prescale) {
254      count += svga_get_prescale_constants(svga, &dest,
255		                           &svga->state.hw_clear.prescale[0]);
256   }
257
258   if (variant->key.vs.undo_viewport) {
259      /* Used to convert window coords back to NDC coords */
260      dest[0] = 1.0f / svga->curr.viewport[0].scale[0];
261      dest[1] = 1.0f / svga->curr.viewport[0].scale[1];
262      dest[2] = -svga->curr.viewport[0].translate[0];
263      dest[3] = -svga->curr.viewport[0].translate[1];
264      dest += 4;
265      count += 1;
266   }
267
268   /* Bias to be added to VertexID */
269   if (variant->key.vs.need_vertex_id_bias) {
270      uint32_t *dest_u = (uint32_t *) dest;  // uint version of dest
271      dest_u[0] = svga->curr.vertex_id_bias;
272      dest_u[1] = 1;
273      dest_u[2] = 1;
274      dest_u[3] = 1;
275      dest+=4;
276      count++;
277   }
278
279   /* SVGA_NEW_CLIP */
280   count += svga_get_clip_plane_constants(svga, variant, &dest);
281
282   /* common constants */
283   count += svga_get_extra_constants_common(svga, variant,
284                                            PIPE_SHADER_VERTEX, dest);
285
286   assert(count <= MAX_EXTRA_CONSTS);
287
288   return count;
289}
290
291/**
292 * Emit any extra geometry shader constants into the buffer pointed
293 * to by 'dest'.
294 */
295static unsigned
296svga_get_extra_gs_constants(const struct svga_context *svga, float *dest)
297{
298   const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
299   unsigned count = 0;
300
301   /* SVGA_NEW_GS_VARIANT
302    */
303
304   /* Constants for point sprite
305    * These are used in the transformed gs that supports point sprite.
306    * They need to be added before the prescale constants.
307    */
308   if (variant->key.gs.wide_point) {
309      count += svga_get_pt_sprite_constants(svga, &dest);
310   }
311
312   if (variant->key.gs.need_prescale) {
313      unsigned i, num_prescale = 1;
314
315      /* If prescale is needed and the geometry shader writes to viewport
316       * index, then prescale for all viewports will be added to the
317       * constant buffer.
318       */
319      if (variant->key.gs.writes_viewport_index)
320         num_prescale = svga->state.hw_clear.num_prescale;
321
322      for (i = 0; i < num_prescale; i++) {
323         count +=
324            svga_get_prescale_constants(svga, &dest,
325			                &svga->state.hw_clear.prescale[i]);
326      }
327   }
328
329   /* SVGA_NEW_CLIP */
330   count += svga_get_clip_plane_constants(svga, variant, &dest);
331
332   /* common constants */
333   count += svga_get_extra_constants_common(svga, variant,
334                                            PIPE_SHADER_GEOMETRY, dest);
335
336   assert(count <= MAX_EXTRA_CONSTS);
337   return count;
338}
339
340
341/**
342 * Emit any extra tessellation control shader constants into the
343 * buffer pointed to by 'dest'.
344 */
345static unsigned
346svga_get_extra_tcs_constants(struct svga_context *svga, float *dest)
347{
348   const struct svga_shader_variant *variant = svga->state.hw_draw.tcs;
349   unsigned count = 0;
350
351   /* SVGA_NEW_CLIP */
352   count += svga_get_clip_plane_constants(svga, variant, &dest);
353
354   /* common constants */
355   count += svga_get_extra_constants_common(svga, variant,
356                                            PIPE_SHADER_TESS_CTRL,
357                                            dest);
358
359   assert(count <= MAX_EXTRA_CONSTS);
360   return count;
361}
362
363
364/**
365 * Emit any extra tessellation evaluation shader constants into
366 * the buffer pointed to by 'dest'.
367 */
368static unsigned
369svga_get_extra_tes_constants(struct svga_context *svga, float *dest)
370{
371   const struct svga_shader_variant *variant = svga->state.hw_draw.tes;
372   unsigned count = 0;
373
374   if (variant->key.tes.need_prescale) {
375      count += svga_get_prescale_constants(svga, &dest,
376		                           &svga->state.hw_clear.prescale[0]);
377   }
378
379   /* SVGA_NEW_CLIP */
380   count += svga_get_clip_plane_constants(svga, variant, &dest);
381
382   /* common constants */
383   count += svga_get_extra_constants_common(svga, variant,
384                                            PIPE_SHADER_TESS_EVAL,
385                                            dest);
386
387   assert(count <= MAX_EXTRA_CONSTS);
388   return count;
389}
390
391
392/**
393 * Emit any extra compute shader constants into
394 * the buffer pointed to by 'dest'.
395 */
396static unsigned
397svga_get_extra_cs_constants(struct svga_context *svga, float *dest)
398{
399   const struct svga_shader_variant *variant = svga->state.hw_draw.cs;
400   unsigned count = 0;
401
402   /* common constants */
403   count += svga_get_extra_constants_common(svga, variant,
404                                            PIPE_SHADER_COMPUTE,
405                                            dest);
406
407   assert(count <= MAX_EXTRA_CONSTS);
408   return count;
409}
410
411
412/*
413 * Check and emit a range of shader constant registers, trying to coalesce
414 * successive shader constant updates in a single command in order to save
415 * space on the command buffer.  This is a HWv8 feature.
416 */
417static enum pipe_error
418emit_const_range(struct svga_context *svga,
419                 enum pipe_shader_type shader,
420                 unsigned offset,
421                 unsigned count,
422                 const float (*values)[4])
423{
424   unsigned i, j;
425   enum pipe_error ret;
426
427   assert(shader == PIPE_SHADER_VERTEX ||
428          shader == PIPE_SHADER_FRAGMENT);
429   assert(!svga_have_vgpu10(svga));
430
431#ifdef DEBUG
432   if (offset + count > SVGA3D_CONSTREG_MAX) {
433      debug_printf("svga: too many constants (offset %u + count %u = %u (max = %u))\n",
434                   offset, count, offset + count, SVGA3D_CONSTREG_MAX);
435   }
436#endif
437
438   if (offset > SVGA3D_CONSTREG_MAX) {
439      /* This isn't OK, but if we propagate an error all the way up we'll
440       * just get into more trouble.
441       * XXX note that offset is always zero at this time so this is moot.
442       */
443      return PIPE_OK;
444   }
445
446   if (offset + count > SVGA3D_CONSTREG_MAX) {
447      /* Just drop the extra constants for now.
448       * Ideally we should not have allowed the app to create a shader
449       * that exceeds our constant buffer size but there's no way to
450       * express that in gallium at this time.
451       */
452      count = SVGA3D_CONSTREG_MAX - offset;
453   }
454
455   i = 0;
456   while (i < count) {
457      if (memcmp(svga->state.hw_draw.cb[shader][offset + i],
458                 values[i],
459                 4 * sizeof(float)) != 0) {
460         /* Found one dirty constant
461          */
462         if (SVGA_DEBUG & DEBUG_CONSTS)
463            debug_printf("%s %s %d: %f %f %f %f\n",
464                         __FUNCTION__,
465                         shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
466                         offset + i,
467                         values[i][0],
468                         values[i][1],
469                         values[i][2],
470                         values[i][3]);
471
472         /* Look for more consecutive dirty constants.
473          */
474         j = i + 1;
475         while (j < count &&
476                j < i + MAX_CONST_REG_COUNT &&
477                memcmp(svga->state.hw_draw.cb[shader][offset + j],
478                       values[j],
479                       4 * sizeof(float)) != 0) {
480
481            if (SVGA_DEBUG & DEBUG_CONSTS)
482               debug_printf("%s %s %d: %f %f %f %f\n",
483                            __FUNCTION__,
484                            shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
485                            offset + j,
486                            values[j][0],
487                            values[j][1],
488                            values[j][2],
489                            values[j][3]);
490
491            ++j;
492         }
493
494         assert(j >= i + 1);
495
496         /* Send them all together.
497          */
498         if (svga_have_gb_objects(svga)) {
499            ret = SVGA3D_SetGBShaderConstsInline(svga->swc,
500                                                 offset + i, /* start */
501                                                 j - i,  /* count */
502                                                 svga_shader_type(shader),
503                                                 SVGA3D_CONST_TYPE_FLOAT,
504                                                 values + i);
505         }
506         else {
507            ret = SVGA3D_SetShaderConsts(svga->swc,
508                                         offset + i, j - i,
509                                         svga_shader_type(shader),
510                                         SVGA3D_CONST_TYPE_FLOAT,
511                                         values + i);
512         }
513         if (ret != PIPE_OK) {
514            return ret;
515         }
516
517         /*
518          * Local copy of the hardware state.
519          */
520         memcpy(svga->state.hw_draw.cb[shader][offset + i],
521                values[i],
522                (j - i) * 4 * sizeof(float));
523
524         i = j + 1;
525
526         svga->hud.num_const_updates++;
527
528      } else {
529         ++i;
530      }
531   }
532
533   return PIPE_OK;
534}
535
536
537/**
538 * Emit all the constants in a constant buffer for a shader stage.
539 * On VGPU10, emit_consts_vgpu10 is used instead.
540 */
541static enum pipe_error
542emit_consts_vgpu9(struct svga_context *svga, enum pipe_shader_type shader)
543{
544   const struct pipe_constant_buffer *cbuf;
545   struct pipe_transfer *transfer = NULL;
546   unsigned count;
547   const float (*data)[4] = NULL;
548   enum pipe_error ret = PIPE_OK;
549   const unsigned offset = 0;
550
551   assert(shader < PIPE_SHADER_TYPES);
552   assert(!svga_have_vgpu10(svga));
553   /* Only one constant buffer per shader is supported before VGPU10.
554    * This is only an approximate check against that.
555    */
556   assert(svga->curr.constbufs[shader][1].buffer == NULL);
557
558   cbuf = &svga->curr.constbufs[shader][0];
559
560   if (svga->curr.constbufs[shader][0].buffer) {
561      /* emit user-provided constants */
562      data = (const float (*)[4])
563         pipe_buffer_map(&svga->pipe, svga->curr.constbufs[shader][0].buffer,
564                         PIPE_MAP_READ, &transfer);
565      if (!data) {
566         return PIPE_ERROR_OUT_OF_MEMORY;
567      }
568
569      /* sanity check */
570      assert(cbuf->buffer->width0 >= cbuf->buffer_size);
571
572      /* Use/apply the constant buffer size and offsets here */
573      count = cbuf->buffer_size / (4 * sizeof(float));
574      data += cbuf->buffer_offset / (4 * sizeof(float));
575
576      ret = emit_const_range( svga, shader, offset, count, data );
577
578      pipe_buffer_unmap(&svga->pipe, transfer);
579
580      if (ret != PIPE_OK) {
581         return ret;
582      }
583   }
584
585   /* emit extra shader constants */
586   {
587      const struct svga_shader_variant *variant = NULL;
588      unsigned offset;
589      float extras[MAX_EXTRA_CONSTS][4];
590      unsigned count;
591
592      switch (shader) {
593      case PIPE_SHADER_VERTEX:
594         variant = svga->state.hw_draw.vs;
595         count = svga_get_extra_vs_constants(svga, (float *) extras);
596         break;
597      case PIPE_SHADER_FRAGMENT:
598         variant = svga->state.hw_draw.fs;
599         count = svga_get_extra_fs_constants(svga, (float *) extras);
600         break;
601      default:
602         assert(!"Unexpected shader type");
603         count = 0;
604      }
605
606      assert(variant);
607      offset = variant->shader->info.constbuf0_num_uniforms;
608      assert(count <= ARRAY_SIZE(extras));
609
610      if (count > 0) {
611         ret = emit_const_range(svga, shader, offset, count,
612                                (const float (*) [4])extras);
613      }
614   }
615
616   return ret;
617}
618
619
620/**
621 * A helper function to destroy any pending unused srv.
622 */
623void
624svga_destroy_rawbuf_srv(struct svga_context *svga)
625{
626   unsigned index = 0;
627
628   while ((index = util_bitmask_get_next_index(
629                      svga->sampler_view_to_free_id_bm, index))
630           != UTIL_BITMASK_INVALID_INDEX) {
631
632      SVGA_RETRY(svga, SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc,
633                                                               index));
634      util_bitmask_clear(svga->sampler_view_id_bm, index);
635      util_bitmask_clear(svga->sampler_view_to_free_id_bm, index);
636   }
637}
638
639/**
640 * A helper function to emit constant buffer as srv raw buffer.
641 */
642static enum pipe_error
643emit_rawbuf(struct svga_context *svga,
644            unsigned slot,
645            enum pipe_shader_type shader,
646            unsigned buffer_offset,
647            unsigned buffer_size,
648            void *buffer)
649{
650   enum pipe_error ret = PIPE_OK;
651   struct svga_raw_buffer *rawbuf = &svga->state.hw_draw.rawbufs[shader][slot];
652   struct svga_winsys_surface *buf_handle = NULL;
653   unsigned srvid = SVGA3D_INVALID_ID;
654   unsigned enabled_rawbufs = svga->state.hw_draw.enabled_rawbufs[shader];
655
656   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_EMITRAWBUFFER);
657
658   if (buffer == NULL) {
659      if ((svga->state.hw_draw.enabled_rawbufs[shader] & (1 << slot)) == 0) {
660         goto done;
661      }
662      enabled_rawbufs &= ~(1 << slot);
663   }
664   else {
665      if ((rawbuf->buffer_offset != buffer_offset) ||
666          (rawbuf->buffer_size != buffer_size) ||
667          (rawbuf->buffer != buffer)) {
668
669         /* Add the current srvid to the delete list */
670         if (rawbuf->srvid != SVGA3D_INVALID_ID) {
671            util_bitmask_set(svga->sampler_view_to_free_id_bm, rawbuf->srvid);
672            rawbuf->srvid = SVGA3D_INVALID_ID;
673         }
674
675         buf_handle = svga_buffer_handle(svga, buffer,
676                                        PIPE_BIND_SAMPLER_VIEW);
677         if (!buf_handle) {
678            ret = PIPE_ERROR_OUT_OF_MEMORY;
679            goto done;
680         }
681
682         /* Create a srv for the constant buffer */
683         srvid = util_bitmask_add(svga->sampler_view_id_bm);
684
685         SVGA3dShaderResourceViewDesc viewDesc;
686         viewDesc.bufferex.firstElement = buffer_offset / 4;
687         viewDesc.bufferex.numElements = buffer_size / 4;
688         viewDesc.bufferex.flags = SVGA3D_BUFFEREX_SRV_RAW;
689
690         ret = SVGA3D_vgpu10_DefineShaderResourceView(svga->swc,
691                  srvid, buf_handle, SVGA3D_R32_TYPELESS,
692                  SVGA3D_RESOURCE_BUFFEREX, &viewDesc);
693
694         if (ret != PIPE_OK) {
695            util_bitmask_clear(svga->sampler_view_id_bm, srvid);
696            goto done;
697         }
698
699         /* Save the current raw buffer attributes in the slot */
700         rawbuf->srvid = srvid;
701         rawbuf->buffer_size = buffer_size;
702         rawbuf->buffer = buffer;
703         rawbuf->handle = buf_handle;
704
705         SVGA_STATS_COUNT_INC(svga_sws(svga), SVGA_STATS_COUNT_RAWBUFFERSRVIEW);
706      }
707      else {
708         /* Same buffer attributes in the slot. Can use the same SRV. */
709         assert(rawbuf->srvid != SVGA3D_INVALID_ID);
710         srvid = rawbuf->srvid;
711         buf_handle = rawbuf->handle;
712      }
713      enabled_rawbufs |= (1 << slot);
714   }
715
716   ret = SVGA3D_vgpu10_SetShaderResources(svga->swc,
717                                          svga_shader_type(shader),
718                                          slot + PIPE_MAX_SAMPLERS,
719                                          1,
720                                          &srvid,
721                                          &buf_handle);
722   if (ret != PIPE_OK) {
723      goto done;
724   }
725
726   /* Save the enabled rawbuf state */
727   svga->state.hw_draw.enabled_rawbufs[shader] = enabled_rawbufs;
728
729done:
730   SVGA_STATS_TIME_POP(svga_sws(svga));
731   return ret;
732}
733
734
735/**
736 * A helper function to emit a constant buffer binding at the
737 * specified slot for the specified shader type
738 */
739static enum pipe_error
740emit_constbuf(struct svga_context *svga,
741              unsigned slot,
742              enum pipe_shader_type shader,
743              unsigned buffer_offset,
744              unsigned buffer_size,
745              const void *buffer,
746              unsigned extra_buffer_offset,
747              unsigned extra_buffer_size,
748              const void *extra_buffer)
749{
750   struct svga_buffer *sbuf = svga_buffer((struct pipe_resource *)buffer);
751   struct pipe_resource *dst_buffer = NULL;
752   enum pipe_error ret = PIPE_OK;
753   struct pipe_transfer *src_transfer;
754   struct svga_winsys_surface *dst_handle = NULL;
755   unsigned new_buf_size = 0;
756   unsigned alloc_buf_size;
757   unsigned offset = 0;;
758   void *src_map = NULL, *dst_map;
759
760   if ((sbuf && sbuf->swbuf) || extra_buffer) {
761
762      /* buffer here is a user-space buffer so mapping it is really cheap. */
763      if (buffer_size > 0) {
764         src_map = pipe_buffer_map_range(&svga->pipe,
765                                         (struct pipe_resource *)buffer,
766                                         buffer_offset, buffer_size,
767                                         PIPE_MAP_READ, &src_transfer);
768         assert(src_map);
769         if (!src_map) {
770            return PIPE_ERROR_OUT_OF_MEMORY;
771         }
772      }
773
774      new_buf_size = MAX2(buffer_size, extra_buffer_offset) + extra_buffer_size;
775
776      /* According to the DX10 spec, the constant buffer size must be
777       * in multiples of 16.
778       */
779      new_buf_size = align(new_buf_size, 16);
780
781      /* Constant buffer size in the upload buffer must be in multiples of 256.
782       * In order to maximize the chance of merging the upload buffer chunks
783       * when svga_buffer_add_range() is called,
784       * the allocate buffer size needs to be in multiples of 256 as well.
785       * Otherwise, since there is gap between each dirty range of the upload buffer,
786       * each dirty range will end up in its own UPDATE_GB_IMAGE command.
787       */
788      alloc_buf_size = align(new_buf_size, CONST0_UPLOAD_ALIGNMENT);
789
790      u_upload_alloc(svga->const0_upload, 0, alloc_buf_size,
791                     CONST0_UPLOAD_ALIGNMENT, &offset,
792                     &dst_buffer, &dst_map);
793
794      if (!dst_map) {
795         if (src_map)
796            pipe_buffer_unmap(&svga->pipe, src_transfer);
797         return PIPE_ERROR_OUT_OF_MEMORY;
798      }
799
800      /* Initialize the allocated buffer slot to 0 to ensure the padding is
801       * filled with 0.
802       */
803      memset(dst_map, 0, alloc_buf_size);
804
805      if (src_map) {
806         memcpy(dst_map, src_map, buffer_size);
807         pipe_buffer_unmap(&svga->pipe, src_transfer);
808      }
809
810      if (extra_buffer_size) {
811         assert(extra_buffer_offset + extra_buffer_size <= new_buf_size);
812         memcpy((char *) dst_map + extra_buffer_offset, extra_buffer,
813                extra_buffer_size);
814      }
815
816      /* Get winsys handle for the constant buffer */
817      if (svga->state.hw_draw.const0_buffer == dst_buffer &&
818          svga->state.hw_draw.const0_handle) {
819         /* re-reference already mapped buffer */
820         dst_handle = svga->state.hw_draw.const0_handle;
821      }
822      else {
823         /* we must unmap the buffer before getting the winsys handle */
824         u_upload_unmap(svga->const0_upload);
825
826         dst_handle = svga_buffer_handle(svga, dst_buffer,
827                                         PIPE_BIND_CONSTANT_BUFFER);
828         if (!dst_handle) {
829            pipe_resource_reference(&dst_buffer, NULL);
830            return PIPE_ERROR_OUT_OF_MEMORY;
831         }
832      }
833   }
834   else if (sbuf) {
835      dst_handle = svga_buffer_handle(svga, &sbuf->b, PIPE_BIND_CONSTANT_BUFFER);
836      new_buf_size = align(buffer_size, 16);
837      offset = buffer_offset;
838   }
839
840   assert(new_buf_size % 16 == 0);
841
842   /* clamp the buf size before sending the command */
843   new_buf_size = MIN2(new_buf_size, SVGA3D_DX_MAX_CONSTBUF_BINDING_SIZE);
844
845   const struct svga_screen *screen = svga_screen(svga->pipe.screen);
846   const struct svga_winsys_screen *sws = screen->sws;
847
848   /* Issue the SetSingleConstantBuffer command */
849   if (!sws->have_constant_buffer_offset_cmd ||
850       svga->state.hw_draw.constbufoffsets[shader][slot].handle != dst_handle ||
851       svga->state.hw_draw.constbufoffsets[shader][slot].size != new_buf_size) {
852      ret = SVGA3D_vgpu10_SetSingleConstantBuffer(svga->swc,
853                                                  slot, /* index */
854                                                  svga_shader_type(shader),
855                                                  dst_handle,
856                                                  offset,
857                                                  new_buf_size);
858   }
859   else if (dst_handle){
860      unsigned command = SVGA_3D_CMD_DX_SET_VS_CONSTANT_BUFFER_OFFSET + shader;
861      ret = SVGA3D_vgpu10_SetConstantBufferOffset(svga->swc,
862                                                  command,
863                                                  slot, /* index */
864                                                  offset);
865   }
866
867   if (ret != PIPE_OK) {
868      pipe_resource_reference(&dst_buffer, NULL);
869      return ret;
870   }
871
872   /* save the upload buffer / handle for next time */
873   if (dst_buffer != buffer && dst_buffer) {
874      pipe_resource_reference(&svga->state.hw_draw.const0_buffer, dst_buffer);
875      svga->state.hw_draw.const0_handle = dst_handle;
876   }
877
878   /* Save this const buffer until it's replaced in the future.
879    * Otherwise, all references to the buffer will go away after the
880    * command buffer is submitted, it'll get recycled and we will have
881    * incorrect constant buffer bindings.
882    */
883   pipe_resource_reference(&svga->state.hw_draw.constbuf[shader][slot], dst_buffer);
884   svga->state.hw_draw.constbufoffsets[shader][slot].handle = dst_handle;
885   svga->state.hw_draw.constbufoffsets[shader][slot].size = new_buf_size;
886
887   pipe_resource_reference(&dst_buffer, NULL);
888
889   return PIPE_OK;
890}
891
892
893/* For constbuf 0 */
894static enum pipe_error
895emit_consts_vgpu10(struct svga_context *svga, enum pipe_shader_type shader)
896{
897   const struct pipe_constant_buffer *cbuf;
898   enum pipe_error ret = PIPE_OK;
899   float extras[MAX_EXTRA_CONSTS][4];
900   unsigned extra_count, extra_size, extra_offset;
901   const struct svga_shader_variant *variant;
902
903   assert(shader == PIPE_SHADER_VERTEX ||
904          shader == PIPE_SHADER_GEOMETRY ||
905          shader == PIPE_SHADER_FRAGMENT ||
906          shader == PIPE_SHADER_TESS_CTRL ||
907          shader == PIPE_SHADER_TESS_EVAL ||
908          shader == PIPE_SHADER_COMPUTE);
909
910   cbuf = &svga->curr.constbufs[shader][0];
911
912   switch (shader) {
913   case PIPE_SHADER_VERTEX:
914      variant = svga->state.hw_draw.vs;
915      extra_count = svga_get_extra_vs_constants(svga, (float *) extras);
916      break;
917   case PIPE_SHADER_FRAGMENT:
918      variant = svga->state.hw_draw.fs;
919      extra_count = svga_get_extra_fs_constants(svga, (float *) extras);
920      break;
921   case PIPE_SHADER_GEOMETRY:
922      variant = svga->state.hw_draw.gs;
923      extra_count = svga_get_extra_gs_constants(svga, (float *) extras);
924      break;
925   case PIPE_SHADER_TESS_CTRL:
926      variant = svga->state.hw_draw.tcs;
927      extra_count = svga_get_extra_tcs_constants(svga, (float *) extras);
928      break;
929   case PIPE_SHADER_TESS_EVAL:
930      variant = svga->state.hw_draw.tes;
931      extra_count = svga_get_extra_tes_constants(svga, (float *) extras);
932      break;
933   case PIPE_SHADER_COMPUTE:
934      variant = svga->state.hw_draw.cs;
935      extra_count = svga_get_extra_cs_constants(svga, (float *) extras);
936      break;
937   default:
938      assert(!"Unexpected shader type");
939      /* Don't return an error code since we don't want to keep re-trying
940       * this function and getting stuck in an infinite loop.
941       */
942      return PIPE_OK;
943   }
944
945   assert(variant);
946
947   cbuf = &svga->curr.constbufs[shader][0];
948
949   /* Compute extra constants size and offset in bytes */
950   extra_size = extra_count * 4 * sizeof(float);
951   extra_offset = 4 * sizeof(float) * variant->extra_const_start;
952
953   if (cbuf->buffer_size + extra_size == 0)
954      return PIPE_OK;  /* nothing to do */
955
956   ret = emit_constbuf(svga, 0, shader,
957                       cbuf->buffer_offset, cbuf->buffer_size, cbuf->buffer,
958                       extra_offset, extra_size, extras);
959   if (ret != PIPE_OK)
960      return ret;
961
962   svga->state.hw_draw.default_constbuf_size[shader] =
963      svga->state.hw_draw.constbufoffsets[shader][0].size;
964
965   svga->hud.num_const_updates++;
966
967   return ret;
968}
969
970
971static enum pipe_error
972emit_constbuf_vgpu10(struct svga_context *svga, enum pipe_shader_type shader)
973{
974   enum pipe_error ret = PIPE_OK;
975   unsigned dirty_constbufs;
976   unsigned enabled_constbufs;
977
978   enabled_constbufs = svga->state.hw_draw.enabled_constbufs[shader] | 1u;
979   dirty_constbufs = (svga->state.dirty_constbufs[shader]|enabled_constbufs) & ~1u;
980
981   while (dirty_constbufs) {
982      unsigned index = u_bit_scan(&dirty_constbufs);
983      unsigned offset = svga->curr.constbufs[shader][index].buffer_offset;
984      unsigned size = svga->curr.constbufs[shader][index].buffer_size;
985      struct svga_buffer *buffer =
986         svga_buffer(svga->curr.constbufs[shader][index].buffer);
987
988      if (buffer) {
989         enabled_constbufs |= 1 << index;
990      }
991      else {
992         enabled_constbufs &= ~(1 << index);
993         assert(offset == 0);
994         assert(size == 0);
995      }
996
997      if (size % 16 != 0) {
998         /* GL's buffer range sizes can be any number of bytes but the
999          * SVGA3D device requires a multiple of 16 bytes.
1000          */
1001         const unsigned total_size = buffer->b.width0;
1002
1003         if (offset + align(size, 16) <= total_size) {
1004            /* round up size to multiple of 16 */
1005            size = align(size, 16);
1006         }
1007         else {
1008            /* round down to multiple of 16 (this may cause rendering problems
1009             * but should avoid a device error).
1010             */
1011            size &= ~15;
1012         }
1013      }
1014
1015      assert(size % 16 == 0);
1016
1017      /**
1018       * If the buffer has been bound as an uav buffer, it will
1019       * need to be bound as a shader resource raw buffer.
1020       */
1021      if (svga->state.raw_constbufs[shader] & (1 << index)) {
1022         ret = emit_rawbuf(svga, index, shader, offset, size, buffer);
1023         if (ret != PIPE_OK) {
1024            return ret;
1025         }
1026
1027         ret = emit_constbuf(svga, index, shader, 0, 0, NULL,
1028                             0, 0, NULL);
1029         if (ret != PIPE_OK) {
1030            return ret;
1031         }
1032
1033         /* Remove the rawbuf from the to-be-enabled constbuf list
1034          * so the buffer will not be referenced again as constant buffer
1035          * at resource validation time.
1036          */
1037         enabled_constbufs &= ~(1 << index);
1038      }
1039      else {
1040         if (svga->state.hw_draw.enabled_rawbufs[shader] & (1 << index)) {
1041            ret = emit_rawbuf(svga, index, shader, offset, size, NULL);
1042            if (ret != PIPE_OK) {
1043               return ret;
1044            }
1045         }
1046
1047         ret = emit_constbuf(svga, index, shader, offset, size, buffer,
1048                          0, 0, NULL);
1049         if (ret != PIPE_OK) {
1050            return ret;
1051         }
1052      }
1053      svga->hud.num_const_buf_updates++;
1054   }
1055
1056   svga->state.hw_draw.enabled_constbufs[shader] = enabled_constbufs;
1057   svga->state.dirty_constbufs[shader] = 0;
1058
1059   return ret;
1060}
1061
1062static enum pipe_error
1063emit_fs_consts(struct svga_context *svga, uint64_t dirty)
1064{
1065   const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
1066   enum pipe_error ret = PIPE_OK;
1067
1068   /* SVGA_NEW_FS_VARIANT
1069    */
1070   if (!variant)
1071      return PIPE_OK;
1072
1073   /* SVGA_NEW_FS_CONSTS
1074    */
1075   if (svga_have_vgpu10(svga)) {
1076      ret = emit_consts_vgpu10(svga, PIPE_SHADER_FRAGMENT);
1077   }
1078   else {
1079      ret = emit_consts_vgpu9(svga, PIPE_SHADER_FRAGMENT);
1080   }
1081
1082   return ret;
1083}
1084
1085static enum pipe_error
1086emit_fs_constbuf(struct svga_context *svga, uint64_t dirty)
1087{
1088   const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
1089   enum pipe_error ret = PIPE_OK;
1090
1091   /* SVGA_NEW_FS_VARIANT
1092    */
1093   if (!variant)
1094      return PIPE_OK;
1095
1096   /* SVGA_NEW_FS_CONSTBUF
1097    */
1098   assert(svga_have_vgpu10(svga));
1099   ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_FRAGMENT);
1100
1101   return ret;
1102}
1103
1104struct svga_tracked_state svga_hw_fs_constants =
1105{
1106   "hw fs params",
1107   (SVGA_NEW_IMAGE_VIEW |
1108    SVGA_NEW_FS_CONSTS |
1109    SVGA_NEW_FS_VARIANT |
1110    SVGA_NEW_TEXTURE_CONSTS),
1111   emit_fs_consts
1112};
1113
1114
1115struct svga_tracked_state svga_hw_fs_constbufs =
1116{
1117   "hw fs params",
1118   SVGA_NEW_FS_CONST_BUFFER,
1119   emit_fs_constbuf
1120};
1121
1122
1123static enum pipe_error
1124emit_vs_consts(struct svga_context *svga, uint64_t dirty)
1125{
1126   const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
1127   enum pipe_error ret = PIPE_OK;
1128
1129   /* SVGA_NEW_VS_VARIANT
1130    */
1131   if (!variant)
1132      return PIPE_OK;
1133
1134   /* SVGA_NEW_VS_CONST_BUFFER
1135    */
1136   if (svga_have_vgpu10(svga)) {
1137      ret = emit_consts_vgpu10(svga, PIPE_SHADER_VERTEX);
1138   }
1139   else {
1140      ret = emit_consts_vgpu9(svga, PIPE_SHADER_VERTEX);
1141   }
1142
1143   return ret;
1144}
1145
1146
1147static enum pipe_error
1148emit_vs_constbuf(struct svga_context *svga, uint64_t dirty)
1149{
1150   const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
1151   enum pipe_error ret = PIPE_OK;
1152
1153   /* SVGA_NEW_FS_VARIANT
1154    */
1155   if (!variant)
1156      return PIPE_OK;
1157
1158   /* SVGA_NEW_FS_CONSTBUF
1159    */
1160   assert(svga_have_vgpu10(svga));
1161   ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_VERTEX);
1162
1163   return ret;
1164}
1165
1166
1167struct svga_tracked_state svga_hw_vs_constants =
1168{
1169   "hw vs params",
1170   (SVGA_NEW_PRESCALE |
1171    SVGA_NEW_IMAGE_VIEW |
1172    SVGA_NEW_VS_CONSTS |
1173    SVGA_NEW_VS_VARIANT |
1174    SVGA_NEW_TEXTURE_CONSTS),
1175   emit_vs_consts
1176};
1177
1178
1179struct svga_tracked_state svga_hw_vs_constbufs =
1180{
1181   "hw vs params",
1182   SVGA_NEW_VS_CONST_BUFFER,
1183   emit_vs_constbuf
1184};
1185
1186
1187static enum pipe_error
1188emit_gs_consts(struct svga_context *svga, uint64_t dirty)
1189{
1190   const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
1191   enum pipe_error ret = PIPE_OK;
1192
1193   /* SVGA_NEW_GS_VARIANT
1194    */
1195   if (!variant)
1196      return PIPE_OK;
1197
1198   /* SVGA_NEW_GS_CONST_BUFFER
1199    */
1200   assert(svga_have_vgpu10(svga));
1201
1202   /**
1203    * If only the rasterizer state has changed and the current geometry
1204    * shader does not emit wide points, then there is no reason to
1205    * re-emit the GS constants, so skip it.
1206    */
1207   if (dirty == SVGA_NEW_RAST && !variant->key.gs.wide_point)
1208      return PIPE_OK;
1209
1210   ret = emit_consts_vgpu10(svga, PIPE_SHADER_GEOMETRY);
1211
1212   return ret;
1213}
1214
1215
1216static enum pipe_error
1217emit_gs_constbuf(struct svga_context *svga, uint64_t dirty)
1218{
1219   const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
1220   enum pipe_error ret = PIPE_OK;
1221
1222   /* SVGA_NEW_GS_VARIANT
1223    */
1224   if (!variant)
1225      return PIPE_OK;
1226
1227   /* SVGA_NEW_GS_CONSTBUF
1228    */
1229   assert(svga_have_vgpu10(svga));
1230   ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_GEOMETRY);
1231
1232   return ret;
1233}
1234
1235
1236struct svga_tracked_state svga_hw_gs_constants =
1237{
1238   "hw gs params",
1239   (SVGA_NEW_PRESCALE |
1240    SVGA_NEW_IMAGE_VIEW |
1241    SVGA_NEW_GS_CONSTS |
1242    SVGA_NEW_RAST |
1243    SVGA_NEW_GS_VARIANT |
1244    SVGA_NEW_TEXTURE_CONSTS),
1245   emit_gs_consts
1246};
1247
1248
1249struct svga_tracked_state svga_hw_gs_constbufs =
1250{
1251   "hw gs params",
1252   SVGA_NEW_GS_CONST_BUFFER,
1253   emit_gs_constbuf
1254};
1255
1256
1257/**
1258 * Emit constant buffer for tessellation control shader
1259 */
1260static enum pipe_error
1261emit_tcs_consts(struct svga_context *svga, uint64_t dirty)
1262{
1263   const struct svga_shader_variant *variant = svga->state.hw_draw.tcs;
1264   enum pipe_error ret = PIPE_OK;
1265
1266   assert(svga_have_sm5(svga));
1267
1268   /* SVGA_NEW_TCS_VARIANT */
1269   if (!variant)
1270      return PIPE_OK;
1271
1272   /* SVGA_NEW_TCS_CONST_BUFFER */
1273
1274   ret = emit_consts_vgpu10(svga, PIPE_SHADER_TESS_CTRL);
1275
1276   return ret;
1277}
1278
1279
1280static enum pipe_error
1281emit_tcs_constbuf(struct svga_context *svga, uint64_t dirty)
1282{
1283   const struct svga_shader_variant *variant = svga->state.hw_draw.tcs;
1284   enum pipe_error ret = PIPE_OK;
1285
1286   /* SVGA_NEW_TCS_VARIANT
1287    */
1288   if (!variant)
1289      return PIPE_OK;
1290
1291   /* SVGA_NEW_TCS_CONSTBUF
1292    */
1293   assert(svga_have_vgpu10(svga));
1294   ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_TESS_CTRL);
1295
1296   return ret;
1297}
1298
1299
1300struct svga_tracked_state svga_hw_tcs_constants =
1301{
1302   "hw tcs params",
1303   (SVGA_NEW_IMAGE_VIEW |
1304    SVGA_NEW_TCS_CONSTS |
1305    SVGA_NEW_TCS_VARIANT),
1306   emit_tcs_consts
1307};
1308
1309
1310struct svga_tracked_state svga_hw_tcs_constbufs =
1311{
1312   "hw tcs params",
1313   SVGA_NEW_TCS_CONST_BUFFER,
1314   emit_tcs_constbuf
1315};
1316
1317
1318/**
1319 * Emit constant buffer for tessellation evaluation shader
1320 */
1321static enum pipe_error
1322emit_tes_consts(struct svga_context *svga, uint64_t dirty)
1323{
1324   const struct svga_shader_variant *variant = svga->state.hw_draw.tes;
1325   enum pipe_error ret = PIPE_OK;
1326
1327   assert(svga_have_sm5(svga));
1328
1329   /* SVGA_NEW_TES_VARIANT */
1330   if (!variant)
1331      return PIPE_OK;
1332
1333   ret = emit_consts_vgpu10(svga, PIPE_SHADER_TESS_EVAL);
1334
1335   return ret;
1336}
1337
1338
1339static enum pipe_error
1340emit_tes_constbuf(struct svga_context *svga, uint64_t dirty)
1341{
1342   const struct svga_shader_variant *variant = svga->state.hw_draw.tes;
1343   enum pipe_error ret = PIPE_OK;
1344
1345   /* SVGA_NEW_TES_VARIANT
1346    */
1347   if (!variant)
1348      return PIPE_OK;
1349
1350   /* SVGA_NEW_TES_CONSTBUF
1351    */
1352   assert(svga_have_vgpu10(svga));
1353   ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_TESS_EVAL);
1354
1355   return ret;
1356}
1357
1358
1359struct svga_tracked_state svga_hw_tes_constants =
1360{
1361   "hw tes params",
1362   (SVGA_NEW_PRESCALE |
1363    SVGA_NEW_IMAGE_VIEW |
1364    SVGA_NEW_TES_CONSTS |
1365    SVGA_NEW_TES_VARIANT),
1366   emit_tes_consts
1367};
1368
1369
1370struct svga_tracked_state svga_hw_tes_constbufs =
1371{
1372   "hw gs params",
1373   SVGA_NEW_TES_CONST_BUFFER,
1374   emit_tes_constbuf
1375};
1376
1377
1378/**
1379 * Emit constant buffer for compute shader
1380 */
1381static enum pipe_error
1382emit_cs_consts(struct svga_context *svga, uint64_t dirty)
1383{
1384   const struct svga_shader_variant *variant = svga->state.hw_draw.cs;
1385   enum pipe_error ret = PIPE_OK;
1386
1387   assert(svga_have_sm5(svga));
1388
1389   /* SVGA_NEW_CS_VARIANT */
1390   if (!variant)
1391      return PIPE_OK;
1392
1393   /* SVGA_NEW_CS_CONST_BUFFER */
1394   ret = emit_consts_vgpu10(svga, PIPE_SHADER_COMPUTE);
1395
1396   return ret;
1397}
1398
1399
1400static enum pipe_error
1401emit_cs_constbuf(struct svga_context *svga, uint64_t dirty)
1402{
1403   const struct svga_shader_variant *variant = svga->state.hw_draw.cs;
1404   enum pipe_error ret = PIPE_OK;
1405
1406   /* SVGA_NEW_CS_VARIANT
1407    */
1408   if (!variant)
1409      return PIPE_OK;
1410
1411   /* SVGA_NEW_CS_CONSTBUF
1412    */
1413   assert(svga_have_vgpu10(svga));
1414   ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_COMPUTE);
1415
1416   return ret;
1417}
1418
1419
1420struct svga_tracked_state svga_hw_cs_constants =
1421{
1422   "hw cs params",
1423   (SVGA_NEW_IMAGE_VIEW |
1424    SVGA_NEW_CS_CONSTS |
1425    SVGA_NEW_CS_VARIANT |
1426    SVGA_NEW_TEXTURE_CONSTS),
1427   emit_cs_consts
1428};
1429
1430
1431struct svga_tracked_state svga_hw_cs_constbufs =
1432{
1433   "hw cs params",
1434   SVGA_NEW_CS_CONST_BUFFER,
1435   emit_cs_constbuf
1436};
1437
1438
1439/**
1440 * A helper function to update the rawbuf for constbuf mask
1441 */
1442static void
1443update_rawbuf_mask(struct svga_context *svga, enum pipe_shader_type shader)
1444{
1445   unsigned dirty_constbufs;
1446   unsigned enabled_constbufs;
1447
1448   enabled_constbufs = svga->state.hw_draw.enabled_constbufs[shader] | 1u;
1449   dirty_constbufs = (svga->state.dirty_constbufs[shader]|enabled_constbufs) & ~1u;
1450
1451   while (dirty_constbufs) {
1452      unsigned index = u_bit_scan(&dirty_constbufs);
1453      struct svga_buffer *sbuf =
1454         svga_buffer(svga->curr.constbufs[shader][index].buffer);
1455
1456      if (sbuf && sbuf->uav) {
1457         svga->state.raw_constbufs[shader] |= (1 << index);
1458      } else {
1459         svga->state.raw_constbufs[shader] &= ~(1 << index);
1460      }
1461   }
1462}
1463
1464
1465/**
1466 * update_rawbuf is called at hw state update time to determine
1467 * if any of the bound constant buffers need to be bound as
1468 * raw buffer srv. This function is called after uav state is
1469 * updated and before shader variants are bound.
1470 */
1471static enum pipe_error
1472update_rawbuf(struct svga_context *svga, uint64 dirty)
1473{
1474   uint64_t rawbuf_dirtybit[] = {
1475      SVGA_NEW_VS_RAW_BUFFER,       /* PIPE_SHADER_VERTEX */
1476      SVGA_NEW_FS_RAW_BUFFER,       /* PIPE_SHADER_FRAGMENT */
1477      SVGA_NEW_GS_RAW_BUFFER,       /* PIPE_SHADER_GEOMETRY */
1478      SVGA_NEW_TCS_RAW_BUFFER,      /* PIPE_SHADER_TESS_CTRL */
1479      SVGA_NEW_TES_RAW_BUFFER,      /* PIPE_SHADER_TESS_EVAL */
1480   };
1481
1482   for (enum pipe_shader_type shader = PIPE_SHADER_VERTEX;
1483        shader <= PIPE_SHADER_TESS_EVAL; shader++) {
1484      unsigned rawbuf_mask = svga->state.raw_constbufs[shader];
1485
1486      update_rawbuf_mask(svga, shader);
1487
1488      /* If the rawbuf state is different for the shader stage,
1489       * send SVGA_NEW_XX_RAW_BUFFER to trigger a new shader
1490       * variant that will use srv for ubo access.
1491       */
1492      if (svga->state.raw_constbufs[shader] != rawbuf_mask)
1493         svga->dirty |= rawbuf_dirtybit[shader];
1494   }
1495
1496   return PIPE_OK;
1497}
1498
1499
1500struct svga_tracked_state svga_need_rawbuf_srv =
1501{
1502   "raw buffer srv",
1503   (SVGA_NEW_IMAGE_VIEW |
1504    SVGA_NEW_SHADER_BUFFER |
1505    SVGA_NEW_CONST_BUFFER),
1506   update_rawbuf
1507};
1508
1509
1510/**
1511 * update_cs_rawbuf is called at compute dispatch time to determine
1512 * if any of the bound constant buffers need to be bound as
1513 * raw buffer srv. This function is called after uav state is
1514 * updated and before a compute shader variant is bound.
1515 */
1516static enum pipe_error
1517update_cs_rawbuf(struct svga_context *svga, uint64 dirty)
1518{
1519   unsigned rawbuf_mask = svga->state.raw_constbufs[PIPE_SHADER_COMPUTE];
1520
1521   update_rawbuf_mask(svga, PIPE_SHADER_COMPUTE);
1522
1523   /* if the rawbuf state is different for the shader stage,
1524    * send SVGA_NEW_RAW_BUFFER to trigger a new shader
1525    * variant to use srv for ubo access.
1526    */
1527   if (svga->state.raw_constbufs[PIPE_SHADER_COMPUTE] != rawbuf_mask)
1528      svga->dirty |= SVGA_NEW_CS_RAW_BUFFER;
1529
1530   return PIPE_OK;
1531}
1532
1533
1534struct svga_tracked_state svga_cs_need_rawbuf_srv =
1535{
1536   "raw buffer srv",
1537   (SVGA_NEW_IMAGE_VIEW |
1538    SVGA_NEW_SHADER_BUFFER |
1539    SVGA_NEW_CONST_BUFFER),
1540   update_cs_rawbuf
1541};
1542