1 /*
2  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
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, sublicense,
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 next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * 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 NONINFRINGEMENT.  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 FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21  * SOFTWARE.
22  *
23  * Authors:
24  *    Rob Clark <robclark@freedesktop.org>
25  */
26 
27 #include "pipe/p_state.h"
28 #include "util/u_dual_blend.h"
29 #include "util/u_helpers.h"
30 #include "util/u_memory.h"
31 #include "util/u_string.h"
32 
33 #include "freedreno_context.h"
34 #include "freedreno_gmem.h"
35 #include "freedreno_query_hw.h"
36 #include "freedreno_resource.h"
37 #include "freedreno_state.h"
38 #include "freedreno_texture.h"
39 #include "freedreno_util.h"
40 
41 #define get_safe(ptr, field) ((ptr) ? (ptr)->field : 0)
42 
43 /* All the generic state handling.. In case of CSO's that are specific
44  * to the GPU version, when the bind and the delete are common they can
45  * go in here.
46  */
47 
48 static void
update_draw_cost(struct fd_context * ctx)49 update_draw_cost(struct fd_context *ctx) assert_dt
50 {
51    struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
52 
53    ctx->draw_cost = pfb->nr_cbufs;
54    for (unsigned i = 0; i < pfb->nr_cbufs; i++)
55       if (fd_blend_enabled(ctx, i))
56          ctx->draw_cost++;
57    if (fd_depth_enabled(ctx))
58       ctx->draw_cost++;
59    if (fd_depth_write_enabled(ctx))
60       ctx->draw_cost++;
61 }
62 
63 static void
fd_set_blend_color(struct pipe_context * pctx,const struct pipe_blend_color * blend_color)64 fd_set_blend_color(struct pipe_context *pctx,
65                    const struct pipe_blend_color *blend_color) in_dt
66 {
67    struct fd_context *ctx = fd_context(pctx);
68    ctx->blend_color = *blend_color;
69    fd_context_dirty(ctx, FD_DIRTY_BLEND_COLOR);
70 }
71 
72 static void
fd_set_stencil_ref(struct pipe_context * pctx,const struct pipe_stencil_ref stencil_ref)73 fd_set_stencil_ref(struct pipe_context *pctx,
74                    const struct pipe_stencil_ref stencil_ref) in_dt
75 {
76    struct fd_context *ctx = fd_context(pctx);
77    ctx->stencil_ref = stencil_ref;
78    fd_context_dirty(ctx, FD_DIRTY_STENCIL_REF);
79 }
80 
81 static void
fd_set_clip_state(struct pipe_context * pctx,const struct pipe_clip_state * clip)82 fd_set_clip_state(struct pipe_context *pctx,
83                   const struct pipe_clip_state *clip) in_dt
84 {
85    struct fd_context *ctx = fd_context(pctx);
86    ctx->ucp = *clip;
87    fd_context_dirty(ctx, FD_DIRTY_UCP);
88 }
89 
90 static void
fd_set_sample_mask(struct pipe_context * pctx,unsigned sample_mask)91 fd_set_sample_mask(struct pipe_context *pctx, unsigned sample_mask) in_dt
92 {
93    struct fd_context *ctx = fd_context(pctx);
94    ctx->sample_mask = (uint16_t)sample_mask;
95    fd_context_dirty(ctx, FD_DIRTY_SAMPLE_MASK);
96 }
97 
98 static void
fd_set_min_samples(struct pipe_context * pctx,unsigned min_samples)99 fd_set_min_samples(struct pipe_context *pctx, unsigned min_samples) in_dt
100 {
101    struct fd_context *ctx = fd_context(pctx);
102    ctx->min_samples = min_samples;
103    fd_context_dirty(ctx, FD_DIRTY_MIN_SAMPLES);
104 }
105 
106 /* notes from calim on #dri-devel:
107  * index==0 will be non-UBO (ie. glUniformXYZ()) all packed together padded
108  * out to vec4's
109  * I should be able to consider that I own the user_ptr until the next
110  * set_constant_buffer() call, at which point I don't really care about the
111  * previous values.
112  * index>0 will be UBO's.. well, I'll worry about that later
113  */
114 static void
fd_set_constant_buffer(struct pipe_context * pctx,enum pipe_shader_type shader,uint index,bool take_ownership,const struct pipe_constant_buffer * cb)115 fd_set_constant_buffer(struct pipe_context *pctx, enum pipe_shader_type shader,
116                        uint index, bool take_ownership,
117                        const struct pipe_constant_buffer *cb) in_dt
118 {
119    struct fd_context *ctx = fd_context(pctx);
120    struct fd_constbuf_stateobj *so = &ctx->constbuf[shader];
121 
122    util_copy_constant_buffer(&so->cb[index], cb, take_ownership);
123 
124    /* Note that gallium frontends can unbind constant buffers by
125     * passing NULL here.
126     */
127    if (unlikely(!cb)) {
128       so->enabled_mask &= ~(1 << index);
129       return;
130    }
131 
132    so->enabled_mask |= 1 << index;
133 
134    fd_context_dirty_shader(ctx, shader, FD_DIRTY_SHADER_CONST);
135    fd_resource_set_usage(cb->buffer, FD_DIRTY_CONST);
136 
137    if (index > 0) {
138       assert(!cb->user_buffer);
139       ctx->dirty |= FD_DIRTY_RESOURCE;
140    }
141 }
142 
143 static void
fd_set_shader_buffers(struct pipe_context * pctx,enum pipe_shader_type shader,unsigned start,unsigned count,const struct pipe_shader_buffer * buffers,unsigned writable_bitmask)144 fd_set_shader_buffers(struct pipe_context *pctx, enum pipe_shader_type shader,
145                       unsigned start, unsigned count,
146                       const struct pipe_shader_buffer *buffers,
147                       unsigned writable_bitmask) in_dt
148 {
149    struct fd_context *ctx = fd_context(pctx);
150    struct fd_shaderbuf_stateobj *so = &ctx->shaderbuf[shader];
151    const unsigned modified_bits = u_bit_consecutive(start, count);
152 
153    so->enabled_mask &= ~modified_bits;
154    so->writable_mask &= ~modified_bits;
155    so->writable_mask |= writable_bitmask << start;
156 
157    for (unsigned i = 0; i < count; i++) {
158       unsigned n = i + start;
159       struct pipe_shader_buffer *buf = &so->sb[n];
160 
161       if (buffers && buffers[i].buffer) {
162          if ((buf->buffer == buffers[i].buffer) &&
163              (buf->buffer_offset == buffers[i].buffer_offset) &&
164              (buf->buffer_size == buffers[i].buffer_size))
165             continue;
166 
167          buf->buffer_offset = buffers[i].buffer_offset;
168          buf->buffer_size = buffers[i].buffer_size;
169          pipe_resource_reference(&buf->buffer, buffers[i].buffer);
170 
171          fd_resource_set_usage(buffers[i].buffer, FD_DIRTY_SSBO);
172 
173          so->enabled_mask |= BIT(n);
174 
175          if (writable_bitmask & BIT(i)) {
176             struct fd_resource *rsc = fd_resource(buf->buffer);
177             util_range_add(&rsc->b.b, &rsc->valid_buffer_range,
178                            buf->buffer_offset,
179                            buf->buffer_offset + buf->buffer_size);
180          }
181       } else {
182          pipe_resource_reference(&buf->buffer, NULL);
183       }
184    }
185 
186    fd_context_dirty_shader(ctx, shader, FD_DIRTY_SHADER_SSBO);
187 }
188 
189 void
fd_set_shader_images(struct pipe_context * pctx,enum pipe_shader_type shader,unsigned start,unsigned count,unsigned unbind_num_trailing_slots,const struct pipe_image_view * images)190 fd_set_shader_images(struct pipe_context *pctx, enum pipe_shader_type shader,
191                      unsigned start, unsigned count,
192                      unsigned unbind_num_trailing_slots,
193                      const struct pipe_image_view *images) in_dt
194 {
195    struct fd_context *ctx = fd_context(pctx);
196    struct fd_shaderimg_stateobj *so = &ctx->shaderimg[shader];
197 
198    unsigned mask = 0;
199 
200    if (images) {
201       for (unsigned i = 0; i < count; i++) {
202          unsigned n = i + start;
203          struct pipe_image_view *buf = &so->si[n];
204 
205          if ((buf->resource == images[i].resource) &&
206              (buf->format == images[i].format) &&
207              (buf->access == images[i].access) &&
208              !memcmp(&buf->u, &images[i].u, sizeof(buf->u)))
209             continue;
210 
211          mask |= BIT(n);
212          util_copy_image_view(buf, &images[i]);
213 
214          if (buf->resource) {
215             fd_resource_set_usage(buf->resource, FD_DIRTY_IMAGE);
216             so->enabled_mask |= BIT(n);
217 
218             if ((buf->access & PIPE_IMAGE_ACCESS_WRITE) &&
219                 (buf->resource->target == PIPE_BUFFER)) {
220 
221                struct fd_resource *rsc = fd_resource(buf->resource);
222                util_range_add(&rsc->b.b, &rsc->valid_buffer_range,
223                               buf->u.buf.offset,
224                               buf->u.buf.offset + buf->u.buf.size);
225             }
226          } else {
227             so->enabled_mask &= ~BIT(n);
228          }
229       }
230    } else {
231       mask = (BIT(count) - 1) << start;
232 
233       for (unsigned i = 0; i < count; i++) {
234          unsigned n = i + start;
235          struct pipe_image_view *img = &so->si[n];
236 
237          pipe_resource_reference(&img->resource, NULL);
238       }
239 
240       so->enabled_mask &= ~mask;
241    }
242 
243    for (unsigned i = 0; i < unbind_num_trailing_slots; i++)
244       pipe_resource_reference(&so->si[i + start + count].resource, NULL);
245 
246    so->enabled_mask &=
247       ~(BITFIELD_MASK(unbind_num_trailing_slots) << (start + count));
248 
249    fd_context_dirty_shader(ctx, shader, FD_DIRTY_SHADER_IMAGE);
250 }
251 
252 void
fd_set_framebuffer_state(struct pipe_context * pctx,const struct pipe_framebuffer_state * framebuffer)253 fd_set_framebuffer_state(struct pipe_context *pctx,
254                          const struct pipe_framebuffer_state *framebuffer)
255 {
256    struct fd_context *ctx = fd_context(pctx);
257    struct pipe_framebuffer_state *cso;
258 
259    DBG("%ux%u, %u layers, %u samples", framebuffer->width, framebuffer->height,
260        framebuffer->layers, framebuffer->samples);
261 
262    cso = &ctx->framebuffer;
263 
264    if (util_framebuffer_state_equal(cso, framebuffer))
265       return;
266 
267    /* Do this *after* checking that the framebuffer state is actually
268     * changing.  In the fd_blitter_clear() path, we get a pfb update
269     * to restore the current pfb state, which should not trigger us
270     * to flush (as that can cause the batch to be freed at a point
271     * before fd_clear() returns, but after the point where it expects
272     * flushes to potentially happen.
273     */
274    fd_context_switch_from(ctx);
275 
276    util_copy_framebuffer_state(cso, framebuffer);
277 
278    cso->samples = util_framebuffer_get_num_samples(cso);
279 
280    if (ctx->screen->reorder) {
281       struct fd_batch *old_batch = NULL;
282 
283       fd_batch_reference(&old_batch, ctx->batch);
284 
285       if (likely(old_batch))
286          fd_batch_finish_queries(old_batch);
287 
288       fd_batch_reference(&ctx->batch, NULL);
289       fd_context_all_dirty(ctx);
290       ctx->update_active_queries = true;
291 
292       fd_batch_reference(&old_batch, NULL);
293    } else if (ctx->batch) {
294       DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx->batch->needs_flush,
295           framebuffer->cbufs[0], framebuffer->zsbuf);
296       fd_batch_flush(ctx->batch);
297    }
298 
299    fd_context_dirty(ctx, FD_DIRTY_FRAMEBUFFER);
300 
301    ctx->disabled_scissor.minx = 0;
302    ctx->disabled_scissor.miny = 0;
303    ctx->disabled_scissor.maxx = cso->width;
304    ctx->disabled_scissor.maxy = cso->height;
305 
306    fd_context_dirty(ctx, FD_DIRTY_SCISSOR);
307    update_draw_cost(ctx);
308 }
309 
310 static void
fd_set_polygon_stipple(struct pipe_context * pctx,const struct pipe_poly_stipple * stipple)311 fd_set_polygon_stipple(struct pipe_context *pctx,
312                        const struct pipe_poly_stipple *stipple) in_dt
313 {
314    struct fd_context *ctx = fd_context(pctx);
315    ctx->stipple = *stipple;
316    fd_context_dirty(ctx, FD_DIRTY_STIPPLE);
317 }
318 
319 static void
fd_set_scissor_states(struct pipe_context * pctx,unsigned start_slot,unsigned num_scissors,const struct pipe_scissor_state * scissor)320 fd_set_scissor_states(struct pipe_context *pctx, unsigned start_slot,
321                       unsigned num_scissors,
322                       const struct pipe_scissor_state *scissor) in_dt
323 {
324    struct fd_context *ctx = fd_context(pctx);
325 
326    ctx->scissor = *scissor;
327    fd_context_dirty(ctx, FD_DIRTY_SCISSOR);
328 }
329 
330 static void
fd_set_viewport_states(struct pipe_context * pctx,unsigned start_slot,unsigned num_viewports,const struct pipe_viewport_state * viewport)331 fd_set_viewport_states(struct pipe_context *pctx, unsigned start_slot,
332                        unsigned num_viewports,
333                        const struct pipe_viewport_state *viewport) in_dt
334 {
335    struct fd_context *ctx = fd_context(pctx);
336    struct pipe_scissor_state *scissor = &ctx->viewport_scissor;
337    float minx, miny, maxx, maxy;
338 
339    ctx->viewport = *viewport;
340 
341    /* see si_get_scissor_from_viewport(): */
342 
343    /* Convert (-1, -1) and (1, 1) from clip space into window space. */
344    minx = -viewport->scale[0] + viewport->translate[0];
345    miny = -viewport->scale[1] + viewport->translate[1];
346    maxx = viewport->scale[0] + viewport->translate[0];
347    maxy = viewport->scale[1] + viewport->translate[1];
348 
349    /* Handle inverted viewports. */
350    if (minx > maxx) {
351       swap(minx, maxx);
352    }
353    if (miny > maxy) {
354       swap(miny, maxy);
355    }
356 
357    const float max_dims = ctx->screen->gen >= 4 ? 16384.f : 4096.f;
358 
359    /* Clamp, convert to integer and round up the max bounds. */
360    scissor->minx = CLAMP(minx, 0.f, max_dims);
361    scissor->miny = CLAMP(miny, 0.f, max_dims);
362    scissor->maxx = CLAMP(ceilf(maxx), 0.f, max_dims);
363    scissor->maxy = CLAMP(ceilf(maxy), 0.f, max_dims);
364 
365    fd_context_dirty(ctx, FD_DIRTY_VIEWPORT);
366 }
367 
368 static void
fd_set_vertex_buffers(struct pipe_context * pctx,unsigned start_slot,unsigned count,unsigned unbind_num_trailing_slots,bool take_ownership,const struct pipe_vertex_buffer * vb)369 fd_set_vertex_buffers(struct pipe_context *pctx, unsigned start_slot,
370                       unsigned count, unsigned unbind_num_trailing_slots,
371                       bool take_ownership,
372                       const struct pipe_vertex_buffer *vb) in_dt
373 {
374    struct fd_context *ctx = fd_context(pctx);
375    struct fd_vertexbuf_stateobj *so = &ctx->vtx.vertexbuf;
376    int i;
377 
378    /* on a2xx, pitch is encoded in the vtx fetch instruction, so
379     * we need to mark VTXSTATE as dirty as well to trigger patching
380     * and re-emitting the vtx shader:
381     */
382    if (ctx->screen->gen < 3) {
383       for (i = 0; i < count; i++) {
384          bool new_enabled = vb && vb[i].buffer.resource;
385          bool old_enabled = so->vb[start_slot + i].buffer.resource != NULL;
386          uint32_t new_stride = vb ? vb[i].stride : 0;
387          uint32_t old_stride = so->vb[start_slot + i].stride;
388          if ((new_enabled != old_enabled) || (new_stride != old_stride)) {
389             fd_context_dirty(ctx, FD_DIRTY_VTXSTATE);
390             break;
391          }
392       }
393    }
394 
395    util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb, start_slot,
396                                 count, unbind_num_trailing_slots,
397                                 take_ownership);
398    so->count = util_last_bit(so->enabled_mask);
399 
400    if (!vb)
401       return;
402 
403    fd_context_dirty(ctx, FD_DIRTY_VTXBUF);
404 
405    for (unsigned i = 0; i < count; i++) {
406       assert(!vb[i].is_user_buffer);
407       fd_resource_set_usage(vb[i].buffer.resource, FD_DIRTY_VTXBUF);
408 
409       /* Robust buffer access: Return undefined data (the start of the buffer)
410        * instead of process termination or a GPU hang in case of overflow.
411        */
412       if (vb[i].buffer.resource &&
413           unlikely(vb[i].buffer_offset >= vb[i].buffer.resource->width0)) {
414          so->vb[start_slot + i].buffer_offset = 0;
415       }
416    }
417 }
418 
419 static void
fd_blend_state_bind(struct pipe_context * pctx,void * hwcso)420 fd_blend_state_bind(struct pipe_context *pctx, void *hwcso) in_dt
421 {
422    struct fd_context *ctx = fd_context(pctx);
423    struct pipe_blend_state *cso = hwcso;
424    bool old_is_dual = ctx->blend ? ctx->blend->rt[0].blend_enable &&
425                                       util_blend_state_is_dual(ctx->blend, 0)
426                                  : false;
427    bool new_is_dual =
428       cso ? cso->rt[0].blend_enable && util_blend_state_is_dual(cso, 0) : false;
429    ctx->blend = hwcso;
430    fd_context_dirty(ctx, FD_DIRTY_BLEND);
431    if (old_is_dual != new_is_dual)
432       fd_context_dirty(ctx, FD_DIRTY_BLEND_DUAL);
433    update_draw_cost(ctx);
434 }
435 
436 static void
fd_blend_state_delete(struct pipe_context * pctx,void * hwcso)437 fd_blend_state_delete(struct pipe_context *pctx, void *hwcso) in_dt
438 {
439    FREE(hwcso);
440 }
441 
442 static void
fd_rasterizer_state_bind(struct pipe_context * pctx,void * hwcso)443 fd_rasterizer_state_bind(struct pipe_context *pctx, void *hwcso) in_dt
444 {
445    struct fd_context *ctx = fd_context(pctx);
446    struct pipe_scissor_state *old_scissor = fd_context_get_scissor(ctx);
447    bool discard = get_safe(ctx->rasterizer, rasterizer_discard);
448    unsigned clip_plane_enable = get_safe(ctx->rasterizer, clip_plane_enable);
449 
450    ctx->rasterizer = hwcso;
451    fd_context_dirty(ctx, FD_DIRTY_RASTERIZER);
452 
453    if (ctx->rasterizer && ctx->rasterizer->scissor) {
454       ctx->current_scissor = &ctx->scissor;
455    } else {
456       ctx->current_scissor = &ctx->disabled_scissor;
457    }
458 
459    /* if scissor enable bit changed we need to mark scissor
460     * state as dirty as well:
461     * NOTE: we can do a shallow compare, since we only care
462     * if it changed to/from &ctx->disable_scissor
463     */
464    if (old_scissor != fd_context_get_scissor(ctx))
465       fd_context_dirty(ctx, FD_DIRTY_SCISSOR);
466 
467    if (discard != get_safe(ctx->rasterizer, rasterizer_discard))
468       fd_context_dirty(ctx, FD_DIRTY_RASTERIZER_DISCARD);
469 
470    if (clip_plane_enable != get_safe(ctx->rasterizer, clip_plane_enable))
471       fd_context_dirty(ctx, FD_DIRTY_RASTERIZER_CLIP_PLANE_ENABLE);
472 }
473 
474 static void
fd_rasterizer_state_delete(struct pipe_context * pctx,void * hwcso)475 fd_rasterizer_state_delete(struct pipe_context *pctx, void *hwcso) in_dt
476 {
477    FREE(hwcso);
478 }
479 
480 static void
fd_zsa_state_bind(struct pipe_context * pctx,void * hwcso)481 fd_zsa_state_bind(struct pipe_context *pctx, void *hwcso) in_dt
482 {
483    struct fd_context *ctx = fd_context(pctx);
484    ctx->zsa = hwcso;
485    fd_context_dirty(ctx, FD_DIRTY_ZSA);
486    update_draw_cost(ctx);
487 }
488 
489 static void
fd_zsa_state_delete(struct pipe_context * pctx,void * hwcso)490 fd_zsa_state_delete(struct pipe_context *pctx, void *hwcso) in_dt
491 {
492    FREE(hwcso);
493 }
494 
495 static void *
fd_vertex_state_create(struct pipe_context * pctx,unsigned num_elements,const struct pipe_vertex_element * elements)496 fd_vertex_state_create(struct pipe_context *pctx, unsigned num_elements,
497                        const struct pipe_vertex_element *elements)
498 {
499    struct fd_vertex_stateobj *so = CALLOC_STRUCT(fd_vertex_stateobj);
500 
501    if (!so)
502       return NULL;
503 
504    memcpy(so->pipe, elements, sizeof(*elements) * num_elements);
505    so->num_elements = num_elements;
506 
507    return so;
508 }
509 
510 static void
fd_vertex_state_delete(struct pipe_context * pctx,void * hwcso)511 fd_vertex_state_delete(struct pipe_context *pctx, void *hwcso) in_dt
512 {
513    FREE(hwcso);
514 }
515 
516 static void
fd_vertex_state_bind(struct pipe_context * pctx,void * hwcso)517 fd_vertex_state_bind(struct pipe_context *pctx, void *hwcso) in_dt
518 {
519    struct fd_context *ctx = fd_context(pctx);
520    ctx->vtx.vtx = hwcso;
521    fd_context_dirty(ctx, FD_DIRTY_VTXSTATE);
522 }
523 
524 static struct pipe_stream_output_target *
fd_create_stream_output_target(struct pipe_context * pctx,struct pipe_resource * prsc,unsigned buffer_offset,unsigned buffer_size)525 fd_create_stream_output_target(struct pipe_context *pctx,
526                                struct pipe_resource *prsc,
527                                unsigned buffer_offset, unsigned buffer_size)
528 {
529    struct fd_stream_output_target *target;
530    struct fd_resource *rsc = fd_resource(prsc);
531 
532    target = CALLOC_STRUCT(fd_stream_output_target);
533    if (!target)
534       return NULL;
535 
536    pipe_reference_init(&target->base.reference, 1);
537    pipe_resource_reference(&target->base.buffer, prsc);
538 
539    target->base.context = pctx;
540    target->base.buffer_offset = buffer_offset;
541    target->base.buffer_size = buffer_size;
542 
543    target->offset_buf = pipe_buffer_create(
544       pctx->screen, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, sizeof(uint32_t));
545 
546    assert(rsc->b.b.target == PIPE_BUFFER);
547    util_range_add(&rsc->b.b, &rsc->valid_buffer_range, buffer_offset,
548                   buffer_offset + buffer_size);
549 
550    return &target->base;
551 }
552 
553 static void
fd_stream_output_target_destroy(struct pipe_context * pctx,struct pipe_stream_output_target * target)554 fd_stream_output_target_destroy(struct pipe_context *pctx,
555                                 struct pipe_stream_output_target *target)
556 {
557    struct fd_stream_output_target *cso = fd_stream_output_target(target);
558 
559    pipe_resource_reference(&cso->base.buffer, NULL);
560    pipe_resource_reference(&cso->offset_buf, NULL);
561 
562    FREE(target);
563 }
564 
565 static void
fd_set_stream_output_targets(struct pipe_context * pctx,unsigned num_targets,struct pipe_stream_output_target ** targets,const unsigned * offsets)566 fd_set_stream_output_targets(struct pipe_context *pctx, unsigned num_targets,
567                              struct pipe_stream_output_target **targets,
568                              const unsigned *offsets) in_dt
569 {
570    struct fd_context *ctx = fd_context(pctx);
571    struct fd_streamout_stateobj *so = &ctx->streamout;
572    unsigned i;
573 
574    debug_assert(num_targets <= ARRAY_SIZE(so->targets));
575 
576    /* Older targets need sw stats enabled for streamout emulation in VS: */
577    if (ctx->screen->gen < 5) {
578       if (num_targets && !so->num_targets) {
579          ctx->stats_users++;
580       } else if (so->num_targets && !num_targets) {
581          ctx->stats_users--;
582       }
583    }
584 
585    for (i = 0; i < num_targets; i++) {
586       boolean changed = targets[i] != so->targets[i];
587       boolean reset = (offsets[i] != (unsigned)-1);
588 
589       so->reset |= (reset << i);
590 
591       if (!changed && !reset)
592          continue;
593 
594       /* Note that all SO targets will be reset at once at a
595        * BeginTransformFeedback().
596        */
597       if (reset) {
598          so->offsets[i] = offsets[i];
599          ctx->streamout.verts_written = 0;
600       }
601 
602       pipe_so_target_reference(&so->targets[i], targets[i]);
603    }
604 
605    for (; i < so->num_targets; i++) {
606       pipe_so_target_reference(&so->targets[i], NULL);
607    }
608 
609    so->num_targets = num_targets;
610 
611    fd_context_dirty(ctx, FD_DIRTY_STREAMOUT);
612 }
613 
614 static void
fd_bind_compute_state(struct pipe_context * pctx,void * state)615 fd_bind_compute_state(struct pipe_context *pctx, void *state) in_dt
616 {
617    struct fd_context *ctx = fd_context(pctx);
618    ctx->compute = state;
619    /* NOTE: Don't mark FD_DIRTY_PROG for compute specific state */
620    ctx->dirty_shader[PIPE_SHADER_COMPUTE] |= FD_DIRTY_SHADER_PROG;
621 }
622 
623 /* TODO pipe_context::set_compute_resources() should DIAF and clover
624  * should be updated to use pipe_context::set_constant_buffer() and
625  * pipe_context::set_shader_images().  Until then just directly frob
626  * the UBO/image state to avoid the rest of the driver needing to
627  * know about this bastard api..
628  */
629 static void
fd_set_compute_resources(struct pipe_context * pctx,unsigned start,unsigned count,struct pipe_surface ** prscs)630 fd_set_compute_resources(struct pipe_context *pctx, unsigned start,
631                          unsigned count, struct pipe_surface **prscs) in_dt
632 {
633    struct fd_context *ctx = fd_context(pctx);
634    struct fd_constbuf_stateobj *so = &ctx->constbuf[PIPE_SHADER_COMPUTE];
635 
636    for (unsigned i = 0; i < count; i++) {
637       const uint32_t index = i + start + 1;   /* UBOs start at index 1 */
638 
639       if (!prscs) {
640          util_copy_constant_buffer(&so->cb[index], NULL, false);
641          so->enabled_mask &= ~(1 << index);
642       } else if (prscs[i]->format == PIPE_FORMAT_NONE) {
643          struct pipe_constant_buffer cb = {
644                .buffer = prscs[i]->texture,
645          };
646          util_copy_constant_buffer(&so->cb[index], &cb, false);
647          so->enabled_mask |= (1 << index);
648       } else {
649          // TODO images
650          unreachable("finishme");
651       }
652    }
653 }
654 
655 /* used by clover to bind global objects, returning the bo address
656  * via handles[n]
657  */
658 static void
fd_set_global_binding(struct pipe_context * pctx,unsigned first,unsigned count,struct pipe_resource ** prscs,uint32_t ** handles)659 fd_set_global_binding(struct pipe_context *pctx, unsigned first, unsigned count,
660                       struct pipe_resource **prscs, uint32_t **handles) in_dt
661 {
662    struct fd_context *ctx = fd_context(pctx);
663    struct fd_global_bindings_stateobj *so = &ctx->global_bindings;
664    unsigned mask = 0;
665 
666    if (prscs) {
667       for (unsigned i = 0; i < count; i++) {
668          unsigned n = i + first;
669 
670          mask |= BIT(n);
671 
672          pipe_resource_reference(&so->buf[n], prscs[i]);
673 
674          if (so->buf[n]) {
675             struct fd_resource *rsc = fd_resource(so->buf[n]);
676             uint32_t offset = *handles[i];
677             uint64_t iova = fd_bo_get_iova(rsc->bo) + offset;
678 
679             /* Yes, really, despite what the type implies: */
680             memcpy(handles[i], &iova, sizeof(iova));
681          }
682 
683          if (prscs[i])
684             so->enabled_mask |= BIT(n);
685          else
686             so->enabled_mask &= ~BIT(n);
687       }
688    } else {
689       mask = (BIT(count) - 1) << first;
690 
691       for (unsigned i = 0; i < count; i++) {
692          unsigned n = i + first;
693          pipe_resource_reference(&so->buf[n], NULL);
694       }
695 
696       so->enabled_mask &= ~mask;
697    }
698 }
699 
700 void
fd_state_init(struct pipe_context * pctx)701 fd_state_init(struct pipe_context *pctx)
702 {
703    pctx->set_blend_color = fd_set_blend_color;
704    pctx->set_stencil_ref = fd_set_stencil_ref;
705    pctx->set_clip_state = fd_set_clip_state;
706    pctx->set_sample_mask = fd_set_sample_mask;
707    pctx->set_min_samples = fd_set_min_samples;
708    pctx->set_constant_buffer = fd_set_constant_buffer;
709    pctx->set_shader_buffers = fd_set_shader_buffers;
710    pctx->set_shader_images = fd_set_shader_images;
711    pctx->set_framebuffer_state = fd_set_framebuffer_state;
712    pctx->set_polygon_stipple = fd_set_polygon_stipple;
713    pctx->set_scissor_states = fd_set_scissor_states;
714    pctx->set_viewport_states = fd_set_viewport_states;
715 
716    pctx->set_vertex_buffers = fd_set_vertex_buffers;
717 
718    pctx->bind_blend_state = fd_blend_state_bind;
719    pctx->delete_blend_state = fd_blend_state_delete;
720 
721    pctx->bind_rasterizer_state = fd_rasterizer_state_bind;
722    pctx->delete_rasterizer_state = fd_rasterizer_state_delete;
723 
724    pctx->bind_depth_stencil_alpha_state = fd_zsa_state_bind;
725    pctx->delete_depth_stencil_alpha_state = fd_zsa_state_delete;
726 
727    if (!pctx->create_vertex_elements_state)
728       pctx->create_vertex_elements_state = fd_vertex_state_create;
729    pctx->delete_vertex_elements_state = fd_vertex_state_delete;
730    pctx->bind_vertex_elements_state = fd_vertex_state_bind;
731 
732    pctx->create_stream_output_target = fd_create_stream_output_target;
733    pctx->stream_output_target_destroy = fd_stream_output_target_destroy;
734    pctx->set_stream_output_targets = fd_set_stream_output_targets;
735 
736    if (has_compute(fd_screen(pctx->screen))) {
737       pctx->bind_compute_state = fd_bind_compute_state;
738       pctx->set_compute_resources = fd_set_compute_resources;
739       pctx->set_global_binding = fd_set_global_binding;
740    }
741 }
742