anv: Add anv_render_pass_attachment::store_op
[mesa.git] / src / intel / vulkan / anv_meta_blit2d.c
1 /*
2 * Copyright © 2016 Intel Corporation
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "anv_meta.h"
25 #include "nir/nir_builder.h"
26
27 enum blit2d_src_type {
28 /* We can make a "normal" image view of this source and just texture
29 * from it like you would in any other shader.
30 */
31 BLIT2D_SRC_TYPE_NORMAL,
32
33 /* The source is W-tiled and we need to detile manually in the shader.
34 * This will work on any platform but is needed for all W-tiled sources
35 * prior to Broadwell.
36 */
37 BLIT2D_SRC_TYPE_W_DETILE,
38
39 BLIT2D_NUM_SRC_TYPES,
40 };
41
42 enum blit2d_dst_type {
43 /* We can bind this destination as a "normal" render target and render
44 * to it just like you would anywhere else.
45 */
46 BLIT2D_DST_TYPE_NORMAL,
47
48 /* The destination is W-tiled and we need to do the tiling manually in
49 * the shader. This is required for all W-tiled destinations.
50 *
51 * Sky Lake adds a feature for providing explicit stencil values in the
52 * shader but mesa doesn't support that yet so neither do we.
53 */
54 BLIT2D_DST_TYPE_W_TILE,
55
56 /* The destination has a 3-channel RGB format. Since we can't render to
57 * non-power-of-two textures, we have to bind it as a red texture and
58 * select the correct component for the given red pixel in the shader.
59 */
60 BLIT2D_DST_TYPE_RGB,
61
62 BLIT2D_NUM_DST_TYPES,
63 };
64
65 static VkFormat
66 vk_format_for_size(int bs)
67 {
68 /* The choice of UNORM and UINT formats is very intentional here. Most of
69 * the time, we want to use a UINT format to avoid any rounding error in
70 * the blit. For stencil blits, R8_UINT is required by the hardware.
71 * (It's the only format allowed in conjunction with W-tiling.) Also we
72 * intentionally use the 4-channel formats whenever we can. This is so
73 * that, when we do a RGB <-> RGBX copy, the two formats will line up even
74 * though one of them is 3/4 the size of the other. The choice of UNORM
75 * vs. UINT is also very intentional because Haswell doesn't handle 8 or
76 * 16-bit RGB UINT formats at all so we have to use UNORM there.
77 * Fortunately, the only time we should ever use two different formats in
78 * the table below is for RGB -> RGBA blits and so we will never have any
79 * UNORM/UINT mismatch.
80 */
81 switch (bs) {
82 case 1: return VK_FORMAT_R8_UINT;
83 case 2: return VK_FORMAT_R8G8_UINT;
84 case 3: return VK_FORMAT_R8G8B8_UNORM;
85 case 4: return VK_FORMAT_R8G8B8A8_UNORM;
86 case 6: return VK_FORMAT_R16G16B16_UNORM;
87 case 8: return VK_FORMAT_R16G16B16A16_UNORM;
88 case 12: return VK_FORMAT_R32G32B32_UINT;
89 case 16: return VK_FORMAT_R32G32B32A32_UINT;
90 default:
91 unreachable("Invalid format block size");
92 }
93 }
94
95 static void
96 create_iview(struct anv_cmd_buffer *cmd_buffer,
97 struct anv_meta_blit2d_surf *surf,
98 uint64_t offset,
99 VkImageUsageFlags usage,
100 uint32_t width,
101 uint32_t height,
102 VkImage *img,
103 struct anv_image_view *iview)
104 {
105 const VkImageCreateInfo image_info = {
106 .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
107 .imageType = VK_IMAGE_TYPE_2D,
108 .format = vk_format_for_size(surf->bs),
109 .extent = {
110 .width = width,
111 .height = height,
112 .depth = 1,
113 },
114 .mipLevels = 1,
115 .arrayLayers = 1,
116 .samples = 1,
117 .tiling = surf->tiling == ISL_TILING_LINEAR ?
118 VK_IMAGE_TILING_LINEAR : VK_IMAGE_TILING_OPTIMAL,
119 .usage = usage,
120 };
121
122 /* Create the VkImage that is bound to the surface's memory. */
123 anv_image_create(anv_device_to_handle(cmd_buffer->device),
124 &(struct anv_image_create_info) {
125 .vk_info = &image_info,
126 .isl_tiling_flags = 1 << surf->tiling,
127 .stride = surf->pitch,
128 }, &cmd_buffer->pool->alloc, img);
129
130 /* We could use a vk call to bind memory, but that would require
131 * creating a dummy memory object etc. so there's really no point.
132 */
133 anv_image_from_handle(*img)->bo = surf->bo;
134 anv_image_from_handle(*img)->offset = surf->base_offset + offset;
135
136 anv_image_view_init(iview, cmd_buffer->device,
137 &(VkImageViewCreateInfo) {
138 .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
139 .image = *img,
140 .viewType = VK_IMAGE_VIEW_TYPE_2D,
141 .format = image_info.format,
142 .subresourceRange = {
143 .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
144 .baseMipLevel = 0,
145 .levelCount = 1,
146 .baseArrayLayer = 0,
147 .layerCount = 1
148 },
149 }, cmd_buffer, usage);
150 }
151
152 struct blit2d_src_temps {
153 VkImage image;
154 struct anv_image_view iview;
155
156 struct anv_buffer buffer;
157 struct anv_buffer_view bview;
158
159 VkDescriptorPool desc_pool;
160 VkDescriptorSet set;
161 };
162
163 static void
164 blit2d_bind_src(struct anv_cmd_buffer *cmd_buffer,
165 struct anv_meta_blit2d_surf *src,
166 enum blit2d_src_type src_type,
167 struct anv_meta_blit2d_rect *rect,
168 struct blit2d_src_temps *tmp)
169 {
170 struct anv_device *device = cmd_buffer->device;
171 VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
172
173 if (src_type == BLIT2D_SRC_TYPE_NORMAL) {
174 uint32_t offset = 0;
175 isl_tiling_get_intratile_offset_el(&cmd_buffer->device->isl_dev,
176 src->tiling, src->bs, src->pitch,
177 rect->src_x, rect->src_y,
178 &offset, &rect->src_x, &rect->src_y);
179
180 create_iview(cmd_buffer, src, offset, VK_IMAGE_USAGE_SAMPLED_BIT,
181 rect->src_x + rect->width, rect->src_y + rect->height,
182 &tmp->image, &tmp->iview);
183
184 anv_CreateDescriptorPool(vk_device,
185 &(const VkDescriptorPoolCreateInfo) {
186 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
187 .pNext = NULL,
188 .flags = 0,
189 .maxSets = 1,
190 .poolSizeCount = 1,
191 .pPoolSizes = (VkDescriptorPoolSize[]) {
192 {
193 .type = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
194 .descriptorCount = 1
195 },
196 }
197 }, &cmd_buffer->pool->alloc, &tmp->desc_pool);
198
199 anv_AllocateDescriptorSets(vk_device,
200 &(VkDescriptorSetAllocateInfo) {
201 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
202 .descriptorPool = tmp->desc_pool,
203 .descriptorSetCount = 1,
204 .pSetLayouts = &device->meta_state.blit2d.img_ds_layout
205 }, &tmp->set);
206
207 anv_UpdateDescriptorSets(vk_device,
208 1, /* writeCount */
209 (VkWriteDescriptorSet[]) {
210 {
211 .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
212 .dstSet = tmp->set,
213 .dstBinding = 0,
214 .dstArrayElement = 0,
215 .descriptorCount = 1,
216 .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
217 .pImageInfo = (VkDescriptorImageInfo[]) {
218 {
219 .sampler = NULL,
220 .imageView = anv_image_view_to_handle(&tmp->iview),
221 .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
222 },
223 }
224 }
225 }, 0, NULL);
226
227 anv_CmdBindDescriptorSets(anv_cmd_buffer_to_handle(cmd_buffer),
228 VK_PIPELINE_BIND_POINT_GRAPHICS,
229 device->meta_state.blit2d.img_p_layout, 0, 1,
230 &tmp->set, 0, NULL);
231 } else {
232 assert(src_type == BLIT2D_SRC_TYPE_W_DETILE);
233 assert(src->tiling == ISL_TILING_W);
234 assert(src->bs == 1);
235
236 uint32_t tile_offset = 0;
237 isl_tiling_get_intratile_offset_el(&cmd_buffer->device->isl_dev,
238 ISL_TILING_W, 1, src->pitch,
239 rect->src_x, rect->src_y,
240 &tile_offset,
241 &rect->src_x, &rect->src_y);
242
243 tmp->buffer = (struct anv_buffer) {
244 .device = device,
245 .size = align_u32(rect->src_y + rect->height, 64) * src->pitch,
246 .usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT,
247 .bo = src->bo,
248 .offset = src->base_offset + tile_offset,
249 };
250
251 anv_buffer_view_init(&tmp->bview, device,
252 &(VkBufferViewCreateInfo) {
253 .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
254 .buffer = anv_buffer_to_handle(&tmp->buffer),
255 .format = VK_FORMAT_R8_UINT,
256 .offset = 0,
257 .range = VK_WHOLE_SIZE,
258 }, cmd_buffer);
259
260 anv_CreateDescriptorPool(vk_device,
261 &(const VkDescriptorPoolCreateInfo) {
262 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
263 .pNext = NULL,
264 .flags = 0,
265 .maxSets = 1,
266 .poolSizeCount = 1,
267 .pPoolSizes = (VkDescriptorPoolSize[]) {
268 {
269 .type = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER,
270 .descriptorCount = 1
271 },
272 }
273 }, &cmd_buffer->pool->alloc, &tmp->desc_pool);
274
275 anv_AllocateDescriptorSets(vk_device,
276 &(VkDescriptorSetAllocateInfo) {
277 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
278 .descriptorPool = tmp->desc_pool,
279 .descriptorSetCount = 1,
280 .pSetLayouts = &device->meta_state.blit2d.buf_ds_layout
281 }, &tmp->set);
282
283 anv_UpdateDescriptorSets(vk_device,
284 1, /* writeCount */
285 (VkWriteDescriptorSet[]) {
286 {
287 .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
288 .dstSet = tmp->set,
289 .dstBinding = 0,
290 .dstArrayElement = 0,
291 .descriptorCount = 1,
292 .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER,
293 .pTexelBufferView = (VkBufferView[]) {
294 anv_buffer_view_to_handle(&tmp->bview),
295 },
296 }
297 }, 0, NULL);
298
299 anv_CmdBindDescriptorSets(anv_cmd_buffer_to_handle(cmd_buffer),
300 VK_PIPELINE_BIND_POINT_GRAPHICS,
301 device->meta_state.blit2d.buf_p_layout, 0, 1,
302 &tmp->set, 0, NULL);
303 }
304 }
305
306 static void
307 blit2d_unbind_src(struct anv_cmd_buffer *cmd_buffer,
308 enum blit2d_src_type src_type,
309 struct blit2d_src_temps *tmp)
310 {
311 anv_DestroyDescriptorPool(anv_device_to_handle(cmd_buffer->device),
312 tmp->desc_pool, &cmd_buffer->pool->alloc);
313 if (src_type == BLIT2D_SRC_TYPE_NORMAL) {
314 anv_DestroyImage(anv_device_to_handle(cmd_buffer->device),
315 tmp->image, &cmd_buffer->pool->alloc);
316 }
317 }
318
319 struct blit2d_dst_temps {
320 VkImage image;
321 struct anv_image_view iview;
322 VkFramebuffer fb;
323 };
324
325 static void
326 blit2d_bind_dst(struct anv_cmd_buffer *cmd_buffer,
327 struct anv_meta_blit2d_surf *dst,
328 uint64_t offset,
329 uint32_t width,
330 uint32_t height,
331 struct blit2d_dst_temps *tmp)
332 {
333 create_iview(cmd_buffer, dst, offset, VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
334 width, height, &tmp->image, &tmp->iview);
335
336 anv_CreateFramebuffer(anv_device_to_handle(cmd_buffer->device),
337 &(VkFramebufferCreateInfo) {
338 .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
339 .attachmentCount = 1,
340 .pAttachments = (VkImageView[]) {
341 anv_image_view_to_handle(&tmp->iview),
342 },
343 .width = width,
344 .height = height,
345 .layers = 1
346 }, &cmd_buffer->pool->alloc, &tmp->fb);
347 }
348
349 static void
350 blit2d_unbind_dst(struct anv_cmd_buffer *cmd_buffer,
351 struct blit2d_dst_temps *tmp)
352 {
353 VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
354 anv_DestroyFramebuffer(vk_device, tmp->fb, &cmd_buffer->pool->alloc);
355 anv_DestroyImage(vk_device, tmp->image, &cmd_buffer->pool->alloc);
356 }
357
358 void
359 anv_meta_end_blit2d(struct anv_cmd_buffer *cmd_buffer,
360 struct anv_meta_saved_state *save)
361 {
362 anv_meta_restore(save, cmd_buffer);
363 }
364
365 void
366 anv_meta_begin_blit2d(struct anv_cmd_buffer *cmd_buffer,
367 struct anv_meta_saved_state *save)
368 {
369 anv_meta_save(save, cmd_buffer, 0);
370 }
371
372 static void
373 bind_pipeline(struct anv_cmd_buffer *cmd_buffer,
374 enum blit2d_src_type src_type,
375 enum blit2d_dst_type dst_type)
376 {
377 VkPipeline pipeline =
378 cmd_buffer->device->meta_state.blit2d.pipelines[src_type][dst_type];
379
380 if (cmd_buffer->state.pipeline != anv_pipeline_from_handle(pipeline)) {
381 anv_CmdBindPipeline(anv_cmd_buffer_to_handle(cmd_buffer),
382 VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
383 }
384 }
385
386 static void
387 anv_meta_blit2d_normal_dst(struct anv_cmd_buffer *cmd_buffer,
388 struct anv_meta_blit2d_surf *src,
389 enum blit2d_src_type src_type,
390 struct anv_meta_blit2d_surf *dst,
391 unsigned num_rects,
392 struct anv_meta_blit2d_rect *rects)
393 {
394 struct anv_device *device = cmd_buffer->device;
395
396 for (unsigned r = 0; r < num_rects; ++r) {
397 struct blit2d_src_temps src_temps;
398 blit2d_bind_src(cmd_buffer, src, src_type, &rects[r], &src_temps);
399
400 uint32_t offset = 0;
401 isl_tiling_get_intratile_offset_el(&cmd_buffer->device->isl_dev,
402 dst->tiling, dst->bs, dst->pitch,
403 rects[r].dst_x, rects[r].dst_y,
404 &offset,
405 &rects[r].dst_x, &rects[r].dst_y);
406
407 struct blit2d_dst_temps dst_temps;
408 blit2d_bind_dst(cmd_buffer, dst, offset, rects[r].dst_x + rects[r].width,
409 rects[r].dst_y + rects[r].height, &dst_temps);
410
411 struct blit_vb_data {
412 float pos[2];
413 float tex_coord[3];
414 } *vb_data;
415
416 unsigned vb_size = sizeof(struct anv_vue_header) + 3 * sizeof(*vb_data);
417
418 struct anv_state vb_state =
419 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, vb_size, 16);
420 memset(vb_state.map, 0, sizeof(struct anv_vue_header));
421 vb_data = vb_state.map + sizeof(struct anv_vue_header);
422
423 vb_data[0] = (struct blit_vb_data) {
424 .pos = {
425 rects[r].dst_x + rects[r].width,
426 rects[r].dst_y + rects[r].height,
427 },
428 .tex_coord = {
429 rects[r].src_x + rects[r].width,
430 rects[r].src_y + rects[r].height,
431 src->pitch,
432 },
433 };
434
435 vb_data[1] = (struct blit_vb_data) {
436 .pos = {
437 rects[r].dst_x,
438 rects[r].dst_y + rects[r].height,
439 },
440 .tex_coord = {
441 rects[r].src_x,
442 rects[r].src_y + rects[r].height,
443 src->pitch,
444 },
445 };
446
447 vb_data[2] = (struct blit_vb_data) {
448 .pos = {
449 rects[r].dst_x,
450 rects[r].dst_y,
451 },
452 .tex_coord = {
453 rects[r].src_x,
454 rects[r].src_y,
455 src->pitch,
456 },
457 };
458
459 if (!device->info.has_llc)
460 anv_state_clflush(vb_state);
461
462 struct anv_buffer vertex_buffer = {
463 .device = device,
464 .size = vb_size,
465 .bo = &device->dynamic_state_block_pool.bo,
466 .offset = vb_state.offset,
467 };
468
469 anv_CmdBindVertexBuffers(anv_cmd_buffer_to_handle(cmd_buffer), 0, 2,
470 (VkBuffer[]) {
471 anv_buffer_to_handle(&vertex_buffer),
472 anv_buffer_to_handle(&vertex_buffer)
473 },
474 (VkDeviceSize[]) {
475 0,
476 sizeof(struct anv_vue_header),
477 });
478
479 ANV_CALL(CmdBeginRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer),
480 &(VkRenderPassBeginInfo) {
481 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
482 .renderPass = device->meta_state.blit2d.render_pass,
483 .framebuffer = dst_temps.fb,
484 .renderArea = {
485 .offset = { rects[r].dst_x, rects[r].dst_y, },
486 .extent = { rects[r].width, rects[r].height },
487 },
488 .clearValueCount = 0,
489 .pClearValues = NULL,
490 }, VK_SUBPASS_CONTENTS_INLINE);
491
492 bind_pipeline(cmd_buffer, src_type, BLIT2D_DST_TYPE_NORMAL);
493
494 ANV_CALL(CmdDraw)(anv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
495
496 ANV_CALL(CmdEndRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer));
497
498 /* At the point where we emit the draw call, all data from the
499 * descriptor sets, etc. has been used. We are free to delete it.
500 */
501 blit2d_unbind_src(cmd_buffer, src_type, &src_temps);
502 blit2d_unbind_dst(cmd_buffer, &dst_temps);
503 }
504 }
505
506 static void
507 anv_meta_blit2d_w_tiled_dst(struct anv_cmd_buffer *cmd_buffer,
508 struct anv_meta_blit2d_surf *src,
509 enum blit2d_src_type src_type,
510 struct anv_meta_blit2d_surf *dst,
511 unsigned num_rects,
512 struct anv_meta_blit2d_rect *rects)
513 {
514 struct anv_device *device = cmd_buffer->device;
515
516 for (unsigned r = 0; r < num_rects; ++r) {
517 struct blit2d_src_temps src_temps;
518 blit2d_bind_src(cmd_buffer, src, src_type, &rects[r], &src_temps);
519
520 assert(dst->bs == 1);
521 uint32_t offset;
522 isl_tiling_get_intratile_offset_el(&cmd_buffer->device->isl_dev,
523 ISL_TILING_W, 1, dst->pitch,
524 rects[r].dst_x, rects[r].dst_y,
525 &offset,
526 &rects[r].dst_x, &rects[r].dst_y);
527
528 /* The original coordinates were in terms of an actual W-tiled offset
529 * but we are binding this image as Y-tiled. We need to adjust our
530 * rectangle accordingly.
531 */
532 uint32_t xmin_Y, xmax_Y, ymin_Y, ymax_Y;
533 xmin_Y = (rects[r].dst_x / 8) * 16;
534 xmax_Y = DIV_ROUND_UP(rects[r].dst_x + rects[r].width, 8) * 16;
535 ymin_Y = (rects[r].dst_y / 4) * 2;
536 ymax_Y = DIV_ROUND_UP(rects[r].dst_y + rects[r].height, 4) * 2;
537
538 struct anv_meta_blit2d_surf dst_Y = {
539 .bo = dst->bo,
540 .tiling = ISL_TILING_Y0,
541 .base_offset = dst->base_offset,
542 .bs = 1,
543 .pitch = dst->pitch * 2,
544 };
545
546 struct blit2d_dst_temps dst_temps;
547 blit2d_bind_dst(cmd_buffer, &dst_Y, offset, xmax_Y, ymax_Y, &dst_temps);
548
549 struct blit_vb_header {
550 struct anv_vue_header vue;
551 int32_t tex_offset[2];
552 uint32_t tex_pitch;
553 uint32_t bounds[4];
554 } *vb_header;
555
556 struct blit_vb_data {
557 float pos[2];
558 } *vb_data;
559
560 unsigned vb_size = sizeof(*vb_header) + 3 * sizeof(*vb_data);
561
562 struct anv_state vb_state =
563 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, vb_size, 16);
564 vb_header = vb_state.map;
565
566 *vb_header = (struct blit_vb_header) {
567 .tex_offset = {
568 rects[r].src_x - rects[r].dst_x,
569 rects[r].src_y - rects[r].dst_y,
570 },
571 .tex_pitch = src->pitch,
572 .bounds = {
573 rects[r].dst_x,
574 rects[r].dst_y,
575 rects[r].dst_x + rects[r].width,
576 rects[r].dst_y + rects[r].height,
577 },
578 };
579
580 vb_data = (void *)(vb_header + 1);
581
582 vb_data[0] = (struct blit_vb_data) {
583 .pos = {
584 xmax_Y,
585 ymax_Y,
586 },
587 };
588
589 vb_data[1] = (struct blit_vb_data) {
590 .pos = {
591 xmin_Y,
592 ymax_Y,
593 },
594 };
595
596 vb_data[2] = (struct blit_vb_data) {
597 .pos = {
598 xmin_Y,
599 ymin_Y,
600 },
601 };
602
603 if (!device->info.has_llc)
604 anv_state_clflush(vb_state);
605
606 struct anv_buffer vertex_buffer = {
607 .device = device,
608 .size = vb_size,
609 .bo = &device->dynamic_state_block_pool.bo,
610 .offset = vb_state.offset,
611 };
612
613 anv_CmdBindVertexBuffers(anv_cmd_buffer_to_handle(cmd_buffer), 0, 2,
614 (VkBuffer[]) {
615 anv_buffer_to_handle(&vertex_buffer),
616 anv_buffer_to_handle(&vertex_buffer)
617 },
618 (VkDeviceSize[]) {
619 0,
620 (void *)vb_data - vb_state.map,
621 });
622
623 ANV_CALL(CmdBeginRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer),
624 &(VkRenderPassBeginInfo) {
625 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
626 .renderPass = device->meta_state.blit2d.render_pass,
627 .framebuffer = dst_temps.fb,
628 .renderArea = {
629 .offset = { xmin_Y, ymin_Y, },
630 .extent = { xmax_Y - xmin_Y, ymax_Y - ymin_Y },
631 },
632 .clearValueCount = 0,
633 .pClearValues = NULL,
634 }, VK_SUBPASS_CONTENTS_INLINE);
635
636 bind_pipeline(cmd_buffer, src_type, BLIT2D_DST_TYPE_W_TILE);
637
638 ANV_CALL(CmdDraw)(anv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
639
640 ANV_CALL(CmdEndRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer));
641
642 /* At the point where we emit the draw call, all data from the
643 * descriptor sets, etc. has been used. We are free to delete it.
644 */
645 blit2d_unbind_src(cmd_buffer, src_type, &src_temps);
646 blit2d_unbind_dst(cmd_buffer, &dst_temps);
647 }
648 }
649
650 void
651 anv_meta_blit2d(struct anv_cmd_buffer *cmd_buffer,
652 struct anv_meta_blit2d_surf *src,
653 struct anv_meta_blit2d_surf *dst,
654 unsigned num_rects,
655 struct anv_meta_blit2d_rect *rects)
656 {
657 enum blit2d_src_type src_type;
658 if (src->tiling == ISL_TILING_W && cmd_buffer->device->info.gen < 8) {
659 src_type = BLIT2D_SRC_TYPE_W_DETILE;
660 } else {
661 src_type = BLIT2D_SRC_TYPE_NORMAL;
662 }
663
664 if (dst->tiling == ISL_TILING_W) {
665 anv_meta_blit2d_w_tiled_dst(cmd_buffer, src, src_type, dst,
666 num_rects, rects);
667 return;
668 } else if (dst->bs % 3 == 0) {
669 anv_finishme("Blitting to RGB destinations not yet supported");
670 return;
671 } else {
672 assert(util_is_power_of_two(dst->bs));
673 anv_meta_blit2d_normal_dst(cmd_buffer, src, src_type, dst,
674 num_rects, rects);
675 }
676 }
677
678 static nir_shader *
679 build_nir_vertex_shader(void)
680 {
681 const struct glsl_type *vec4 = glsl_vec4_type();
682 nir_builder b;
683
684 nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
685 b.shader->info.name = ralloc_strdup(b.shader, "meta_blit_vs");
686
687 nir_variable *pos_in = nir_variable_create(b.shader, nir_var_shader_in,
688 vec4, "a_pos");
689 pos_in->data.location = VERT_ATTRIB_GENERIC0;
690 nir_variable *pos_out = nir_variable_create(b.shader, nir_var_shader_out,
691 vec4, "gl_Position");
692 pos_out->data.location = VARYING_SLOT_POS;
693 nir_copy_var(&b, pos_out, pos_in);
694
695 nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
696 vec4, "a_tex_pos");
697 tex_pos_in->data.location = VERT_ATTRIB_GENERIC1;
698 nir_variable *tex_pos_out = nir_variable_create(b.shader, nir_var_shader_out,
699 vec4, "v_tex_pos");
700 tex_pos_out->data.location = VARYING_SLOT_VAR0;
701 tex_pos_out->data.interpolation = INTERP_QUALIFIER_SMOOTH;
702 nir_copy_var(&b, tex_pos_out, tex_pos_in);
703
704 nir_variable *other_in = nir_variable_create(b.shader, nir_var_shader_in,
705 vec4, "a_other");
706 other_in->data.location = VERT_ATTRIB_GENERIC2;
707 nir_variable *other_out = nir_variable_create(b.shader, nir_var_shader_out,
708 vec4, "v_other");
709 other_out->data.location = VARYING_SLOT_VAR1;
710 other_out->data.interpolation = INTERP_QUALIFIER_FLAT;
711 nir_copy_var(&b, other_out, other_in);
712
713 return b.shader;
714 }
715
716 typedef nir_ssa_def* (*texel_fetch_build_func)(struct nir_builder *,
717 struct anv_device *,
718 nir_ssa_def *, nir_ssa_def *);
719
720 static nir_ssa_def *
721 nir_copy_bits(struct nir_builder *b, nir_ssa_def *dst, unsigned dst_offset,
722 nir_ssa_def *src, unsigned src_offset, unsigned num_bits)
723 {
724 unsigned src_mask = (~1u >> (32 - num_bits)) << src_offset;
725 nir_ssa_def *masked = nir_iand(b, src, nir_imm_int(b, src_mask));
726
727 nir_ssa_def *shifted;
728 if (dst_offset > src_offset) {
729 shifted = nir_ishl(b, masked, nir_imm_int(b, dst_offset - src_offset));
730 } else if (dst_offset < src_offset) {
731 shifted = nir_ushr(b, masked, nir_imm_int(b, src_offset - dst_offset));
732 } else {
733 assert(dst_offset == src_offset);
734 shifted = masked;
735 }
736
737 return nir_ior(b, dst, shifted);
738 }
739
740 static nir_ssa_def *
741 build_nir_w_tiled_fetch(struct nir_builder *b, struct anv_device *device,
742 nir_ssa_def *tex_pos, nir_ssa_def *tex_pitch)
743 {
744 nir_ssa_def *x = nir_channel(b, tex_pos, 0);
745 nir_ssa_def *y = nir_channel(b, tex_pos, 1);
746
747 /* First, compute the block-aligned offset */
748 nir_ssa_def *x_major = nir_ushr(b, x, nir_imm_int(b, 6));
749 nir_ssa_def *y_major = nir_ushr(b, y, nir_imm_int(b, 6));
750 nir_ssa_def *offset =
751 nir_iadd(b, nir_imul(b, y_major,
752 nir_imul(b, tex_pitch, nir_imm_int(b, 64))),
753 nir_imul(b, x_major, nir_imm_int(b, 4096)));
754
755 /* Compute the bottom 12 bits of the offset */
756 offset = nir_copy_bits(b, offset, 0, x, 0, 1);
757 offset = nir_copy_bits(b, offset, 1, y, 0, 1);
758 offset = nir_copy_bits(b, offset, 2, x, 1, 1);
759 offset = nir_copy_bits(b, offset, 3, y, 1, 1);
760 offset = nir_copy_bits(b, offset, 4, x, 2, 1);
761 offset = nir_copy_bits(b, offset, 5, y, 2, 4);
762 offset = nir_copy_bits(b, offset, 9, x, 3, 3);
763
764 if (device->isl_dev.has_bit6_swizzling) {
765 offset = nir_ixor(b, offset,
766 nir_ushr(b, nir_iand(b, offset, nir_imm_int(b, 0x0200)),
767 nir_imm_int(b, 3)));
768 }
769
770 const struct glsl_type *sampler_type =
771 glsl_sampler_type(GLSL_SAMPLER_DIM_BUF, false, false, GLSL_TYPE_FLOAT);
772 nir_variable *sampler = nir_variable_create(b->shader, nir_var_uniform,
773 sampler_type, "s_tex");
774 sampler->data.descriptor_set = 0;
775 sampler->data.binding = 0;
776
777 nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
778 tex->sampler_dim = GLSL_SAMPLER_DIM_BUF;
779 tex->op = nir_texop_txf;
780 tex->src[0].src_type = nir_tex_src_coord;
781 tex->src[0].src = nir_src_for_ssa(offset);
782 tex->dest_type = nir_type_float; /* TODO */
783 tex->is_array = false;
784 tex->coord_components = 1;
785 tex->texture = nir_deref_var_create(tex, sampler);
786 tex->sampler = NULL;
787
788 nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
789 nir_builder_instr_insert(b, &tex->instr);
790
791 return &tex->dest.ssa;
792 }
793
794 static nir_ssa_def *
795 build_nir_texel_fetch(struct nir_builder *b, struct anv_device *device,
796 nir_ssa_def *tex_pos, nir_ssa_def *tex_pitch)
797 {
798 const struct glsl_type *sampler_type =
799 glsl_sampler_type(GLSL_SAMPLER_DIM_2D, false, false, GLSL_TYPE_FLOAT);
800 nir_variable *sampler = nir_variable_create(b->shader, nir_var_uniform,
801 sampler_type, "s_tex");
802 sampler->data.descriptor_set = 0;
803 sampler->data.binding = 0;
804
805 nir_tex_instr *tex = nir_tex_instr_create(b->shader, 2);
806 tex->sampler_dim = GLSL_SAMPLER_DIM_2D;
807 tex->op = nir_texop_txf;
808 tex->src[0].src_type = nir_tex_src_coord;
809 tex->src[0].src = nir_src_for_ssa(tex_pos);
810 tex->src[1].src_type = nir_tex_src_lod;
811 tex->src[1].src = nir_src_for_ssa(nir_imm_int(b, 0));
812 tex->dest_type = nir_type_float; /* TODO */
813 tex->is_array = false;
814 tex->coord_components = 2;
815 tex->texture = nir_deref_var_create(tex, sampler);
816 tex->sampler = NULL;
817
818 nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
819 nir_builder_instr_insert(b, &tex->instr);
820
821 return &tex->dest.ssa;
822 }
823
824 static const VkPipelineVertexInputStateCreateInfo normal_vi_create_info = {
825 .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
826 .vertexBindingDescriptionCount = 2,
827 .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
828 {
829 .binding = 0,
830 .stride = 0,
831 .inputRate = VK_VERTEX_INPUT_RATE_INSTANCE
832 },
833 {
834 .binding = 1,
835 .stride = 5 * sizeof(float),
836 .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
837 },
838 },
839 .vertexAttributeDescriptionCount = 3,
840 .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
841 {
842 /* VUE Header */
843 .location = 0,
844 .binding = 0,
845 .format = VK_FORMAT_R32G32B32A32_UINT,
846 .offset = 0
847 },
848 {
849 /* Position */
850 .location = 1,
851 .binding = 1,
852 .format = VK_FORMAT_R32G32_SFLOAT,
853 .offset = 0
854 },
855 {
856 /* Texture Coordinate */
857 .location = 2,
858 .binding = 1,
859 .format = VK_FORMAT_R32G32B32_SFLOAT,
860 .offset = 8
861 },
862 },
863 };
864
865 static nir_shader *
866 build_nir_copy_fragment_shader(struct anv_device *device,
867 texel_fetch_build_func txf_func)
868 {
869 const struct glsl_type *vec4 = glsl_vec4_type();
870 const struct glsl_type *vec3 = glsl_vector_type(GLSL_TYPE_FLOAT, 3);
871 nir_builder b;
872
873 nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
874 b.shader->info.name = ralloc_strdup(b.shader, "meta_blit2d_fs");
875
876 nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
877 vec3, "v_tex_pos");
878 tex_pos_in->data.location = VARYING_SLOT_VAR0;
879
880 nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
881 vec4, "f_color");
882 color_out->data.location = FRAG_RESULT_DATA0;
883
884 nir_ssa_def *pos_int = nir_f2i(&b, nir_load_var(&b, tex_pos_in));
885 unsigned swiz[4] = { 0, 1 };
886 nir_ssa_def *tex_pos = nir_swizzle(&b, pos_int, swiz, 2, false);
887 nir_ssa_def *tex_pitch = nir_channel(&b, pos_int, 2);
888
889 nir_ssa_def *color = txf_func(&b, device, tex_pos, tex_pitch);
890 nir_store_var(&b, color_out, color, 0xf);
891
892 return b.shader;
893 }
894
895 static const VkPipelineVertexInputStateCreateInfo w_tiled_vi_create_info = {
896 .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
897 .vertexBindingDescriptionCount = 2,
898 .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
899 {
900 .binding = 0,
901 .stride = 0,
902 .inputRate = VK_VERTEX_INPUT_RATE_INSTANCE
903 },
904 {
905 .binding = 1,
906 .stride = 2 * sizeof(float),
907 .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
908 },
909 },
910 .vertexAttributeDescriptionCount = 4,
911 .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
912 {
913 /* VUE Header */
914 .location = 0,
915 .binding = 0,
916 .format = VK_FORMAT_R32G32B32A32_UINT,
917 .offset = 0
918 },
919 {
920 /* Position */
921 .location = 1,
922 .binding = 1,
923 .format = VK_FORMAT_R32G32_SFLOAT,
924 .offset = 0
925 },
926 {
927 /* Texture Offset */
928 .location = 2,
929 .binding = 0,
930 .format = VK_FORMAT_R32G32B32_UINT,
931 .offset = 16
932 },
933 {
934 /* Destination bounds */
935 .location = 3,
936 .binding = 0,
937 .format = VK_FORMAT_R32G32B32A32_UINT,
938 .offset = 28
939 },
940 },
941 };
942
943 static nir_shader *
944 build_nir_w_tiled_fragment_shader(struct anv_device *device,
945 texel_fetch_build_func txf_func)
946 {
947 const struct glsl_type *vec4 = glsl_vec4_type();
948 const struct glsl_type *ivec3 = glsl_vector_type(GLSL_TYPE_INT, 3);
949 const struct glsl_type *uvec4 = glsl_vector_type(GLSL_TYPE_UINT, 4);
950 nir_builder b;
951
952 nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
953 b.shader->info.name = ralloc_strdup(b.shader, "meta_blit2d_fs");
954
955 /* We need gl_FragCoord so we know our Y-tiled position */
956 nir_variable *frag_coord_in = nir_variable_create(b.shader,
957 nir_var_shader_in,
958 vec4, "gl_FragCoord");
959 frag_coord_in->data.location = VARYING_SLOT_POS;
960 frag_coord_in->data.origin_upper_left = true;
961
962 /* In location 0 we have an ivec3 that has the offset from dest to
963 * source in the first two components and the stride in the third.
964 */
965 nir_variable *tex_off_in = nir_variable_create(b.shader, nir_var_shader_in,
966 ivec3, "v_tex_off");
967 tex_off_in->data.location = VARYING_SLOT_VAR0;
968 tex_off_in->data.interpolation = INTERP_QUALIFIER_FLAT;
969
970 /* In location 1 we have a uvec4 that gives us the bounds of the
971 * destination. We need to discard if we get outside this boundary.
972 */
973 nir_variable *bounds_in = nir_variable_create(b.shader, nir_var_shader_in,
974 uvec4, "v_bounds");
975 bounds_in->data.location = VARYING_SLOT_VAR1;
976 bounds_in->data.interpolation = INTERP_QUALIFIER_FLAT;
977
978 nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
979 vec4, "f_color");
980 color_out->data.location = FRAG_RESULT_DATA0;
981
982 nir_ssa_def *frag_coord_int = nir_f2i(&b, nir_load_var(&b, frag_coord_in));
983 nir_ssa_def *x_Y = nir_channel(&b, frag_coord_int, 0);
984 nir_ssa_def *y_Y = nir_channel(&b, frag_coord_int, 1);
985
986 /* Compute the W-tiled position from the Y-tiled position */
987 nir_ssa_def *x_W = nir_iand(&b, x_Y, nir_imm_int(&b, 0xffffff80));
988 x_W = nir_ushr(&b, x_W, nir_imm_int(&b, 1));
989 x_W = nir_copy_bits(&b, x_W, 0, x_Y, 0, 1);
990 x_W = nir_copy_bits(&b, x_W, 1, x_Y, 2, 1);
991 x_W = nir_copy_bits(&b, x_W, 2, y_Y, 0, 1);
992 x_W = nir_copy_bits(&b, x_W, 3, x_Y, 4, 3);
993
994 nir_ssa_def *y_W = nir_iand(&b, y_Y, nir_imm_int(&b, 0xffffffe0));
995 y_W = nir_ishl(&b, y_W, nir_imm_int(&b, 1));
996 y_W = nir_copy_bits(&b, y_W, 0, x_Y, 1, 1);
997 y_W = nir_copy_bits(&b, y_W, 1, x_Y, 3, 1);
998 y_W = nir_copy_bits(&b, y_W, 2, y_Y, 1, 4);
999
1000 /* Figure out if we are out-of-bounds and discard */
1001 nir_ssa_def *bounds = nir_load_var(&b, bounds_in);
1002 nir_ssa_def *oob =
1003 nir_ior(&b, nir_ult(&b, x_W, nir_channel(&b, bounds, 0)),
1004 nir_ior(&b, nir_ult(&b, y_W, nir_channel(&b, bounds, 1)),
1005 nir_ior(&b, nir_uge(&b, x_W, nir_channel(&b, bounds, 2)),
1006 nir_uge(&b, y_W, nir_channel(&b, bounds, 3)))));
1007
1008 nir_intrinsic_instr *discard =
1009 nir_intrinsic_instr_create(b.shader, nir_intrinsic_discard_if);
1010 discard->src[0] = nir_src_for_ssa(oob);
1011 nir_builder_instr_insert(&b, &discard->instr);
1012
1013 nir_ssa_def *tex_off = nir_channels(&b, nir_load_var(&b, tex_off_in), 0x3);
1014 nir_ssa_def *tex_pos = nir_iadd(&b, nir_vec2(&b, x_W, y_W), tex_off);
1015 nir_ssa_def *tex_pitch = nir_channel(&b, nir_load_var(&b, tex_off_in), 2);
1016
1017 nir_ssa_def *color = txf_func(&b, device, tex_pos, tex_pitch);
1018 nir_store_var(&b, color_out, color, 0xf);
1019
1020 return b.shader;
1021 }
1022
1023 void
1024 anv_device_finish_meta_blit2d_state(struct anv_device *device)
1025 {
1026 if (device->meta_state.blit2d.render_pass) {
1027 anv_DestroyRenderPass(anv_device_to_handle(device),
1028 device->meta_state.blit2d.render_pass,
1029 &device->meta_state.alloc);
1030 }
1031
1032 if (device->meta_state.blit2d.img_p_layout) {
1033 anv_DestroyPipelineLayout(anv_device_to_handle(device),
1034 device->meta_state.blit2d.img_p_layout,
1035 &device->meta_state.alloc);
1036 }
1037
1038 if (device->meta_state.blit2d.img_ds_layout) {
1039 anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
1040 device->meta_state.blit2d.img_ds_layout,
1041 &device->meta_state.alloc);
1042 }
1043
1044 if (device->meta_state.blit2d.buf_p_layout) {
1045 anv_DestroyPipelineLayout(anv_device_to_handle(device),
1046 device->meta_state.blit2d.buf_p_layout,
1047 &device->meta_state.alloc);
1048 }
1049
1050 if (device->meta_state.blit2d.buf_ds_layout) {
1051 anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
1052 device->meta_state.blit2d.buf_ds_layout,
1053 &device->meta_state.alloc);
1054 }
1055
1056 for (unsigned src = 0; src < BLIT2D_NUM_SRC_TYPES; src++) {
1057 for (unsigned dst = 0; dst < BLIT2D_NUM_DST_TYPES; dst++) {
1058 if (device->meta_state.blit2d.pipelines[src][dst]) {
1059 anv_DestroyPipeline(anv_device_to_handle(device),
1060 device->meta_state.blit2d.pipelines[src][dst],
1061 &device->meta_state.alloc);
1062 }
1063 }
1064 }
1065 }
1066
1067 static VkResult
1068 blit2d_init_pipeline(struct anv_device *device,
1069 enum blit2d_src_type src_type,
1070 enum blit2d_dst_type dst_type)
1071 {
1072 VkResult result;
1073
1074 texel_fetch_build_func src_func;
1075 switch (src_type) {
1076 case BLIT2D_SRC_TYPE_NORMAL:
1077 src_func = build_nir_texel_fetch;
1078 break;
1079 case BLIT2D_SRC_TYPE_W_DETILE:
1080 src_func = build_nir_w_tiled_fetch;
1081 break;
1082 default:
1083 unreachable("Invalid blit2d source type");
1084 }
1085
1086 const VkPipelineVertexInputStateCreateInfo *vi_create_info;
1087 struct anv_shader_module fs = { .nir = NULL };
1088 switch (dst_type) {
1089 case BLIT2D_DST_TYPE_NORMAL:
1090 fs.nir = build_nir_copy_fragment_shader(device, src_func);
1091 vi_create_info = &normal_vi_create_info;
1092 break;
1093 case BLIT2D_DST_TYPE_W_TILE:
1094 fs.nir = build_nir_w_tiled_fragment_shader(device, src_func);
1095 vi_create_info = &w_tiled_vi_create_info;
1096 break;
1097 case BLIT2D_DST_TYPE_RGB:
1098 /* Not yet supported */
1099 default:
1100 return VK_SUCCESS;
1101 }
1102
1103 /* We don't use a vertex shader for blitting, but instead build and pass
1104 * the VUEs directly to the rasterization backend. However, we do need
1105 * to provide GLSL source for the vertex shader so that the compiler
1106 * does not dead-code our inputs.
1107 */
1108 struct anv_shader_module vs = {
1109 .nir = build_nir_vertex_shader(),
1110 };
1111
1112 VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
1113 {
1114 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
1115 .stage = VK_SHADER_STAGE_VERTEX_BIT,
1116 .module = anv_shader_module_to_handle(&vs),
1117 .pName = "main",
1118 .pSpecializationInfo = NULL
1119 }, {
1120 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
1121 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
1122 .module = anv_shader_module_to_handle(&fs),
1123 .pName = "main",
1124 .pSpecializationInfo = NULL
1125 },
1126 };
1127
1128 const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
1129 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
1130 .stageCount = ARRAY_SIZE(pipeline_shader_stages),
1131 .pStages = pipeline_shader_stages,
1132 .pVertexInputState = vi_create_info,
1133 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
1134 .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
1135 .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
1136 .primitiveRestartEnable = false,
1137 },
1138 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
1139 .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
1140 .viewportCount = 1,
1141 .scissorCount = 1,
1142 },
1143 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
1144 .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
1145 .rasterizerDiscardEnable = false,
1146 .polygonMode = VK_POLYGON_MODE_FILL,
1147 .cullMode = VK_CULL_MODE_NONE,
1148 .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE
1149 },
1150 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
1151 .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
1152 .rasterizationSamples = 1,
1153 .sampleShadingEnable = false,
1154 .pSampleMask = (VkSampleMask[]) { UINT32_MAX },
1155 },
1156 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
1157 .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
1158 .attachmentCount = 1,
1159 .pAttachments = (VkPipelineColorBlendAttachmentState []) {
1160 { .colorWriteMask =
1161 VK_COLOR_COMPONENT_A_BIT |
1162 VK_COLOR_COMPONENT_R_BIT |
1163 VK_COLOR_COMPONENT_G_BIT |
1164 VK_COLOR_COMPONENT_B_BIT },
1165 }
1166 },
1167 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
1168 .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
1169 .dynamicStateCount = 9,
1170 .pDynamicStates = (VkDynamicState[]) {
1171 VK_DYNAMIC_STATE_VIEWPORT,
1172 VK_DYNAMIC_STATE_SCISSOR,
1173 VK_DYNAMIC_STATE_LINE_WIDTH,
1174 VK_DYNAMIC_STATE_DEPTH_BIAS,
1175 VK_DYNAMIC_STATE_BLEND_CONSTANTS,
1176 VK_DYNAMIC_STATE_DEPTH_BOUNDS,
1177 VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
1178 VK_DYNAMIC_STATE_STENCIL_WRITE_MASK,
1179 VK_DYNAMIC_STATE_STENCIL_REFERENCE,
1180 },
1181 },
1182 .flags = 0,
1183 .layout = device->meta_state.blit2d.img_p_layout,
1184 .renderPass = device->meta_state.blit2d.render_pass,
1185 .subpass = 0,
1186 };
1187
1188 const struct anv_graphics_pipeline_create_info anv_pipeline_info = {
1189 .color_attachment_count = -1,
1190 .use_repclear = false,
1191 .disable_vs = true,
1192 .use_rectlist = true
1193 };
1194
1195 result = anv_graphics_pipeline_create(anv_device_to_handle(device),
1196 VK_NULL_HANDLE,
1197 &vk_pipeline_info, &anv_pipeline_info,
1198 &device->meta_state.alloc,
1199 &device->meta_state.blit2d.pipelines[src_type][dst_type]);
1200
1201 ralloc_free(vs.nir);
1202 ralloc_free(fs.nir);
1203
1204 return result;
1205 }
1206
1207 VkResult
1208 anv_device_init_meta_blit2d_state(struct anv_device *device)
1209 {
1210 VkResult result;
1211
1212 zero(device->meta_state.blit2d);
1213
1214 result = anv_CreateRenderPass(anv_device_to_handle(device),
1215 &(VkRenderPassCreateInfo) {
1216 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
1217 .attachmentCount = 1,
1218 .pAttachments = &(VkAttachmentDescription) {
1219 .format = VK_FORMAT_UNDEFINED, /* Our shaders don't care */
1220 .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
1221 .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
1222 .initialLayout = VK_IMAGE_LAYOUT_GENERAL,
1223 .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
1224 },
1225 .subpassCount = 1,
1226 .pSubpasses = &(VkSubpassDescription) {
1227 .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
1228 .inputAttachmentCount = 0,
1229 .colorAttachmentCount = 1,
1230 .pColorAttachments = &(VkAttachmentReference) {
1231 .attachment = 0,
1232 .layout = VK_IMAGE_LAYOUT_GENERAL,
1233 },
1234 .pResolveAttachments = NULL,
1235 .pDepthStencilAttachment = &(VkAttachmentReference) {
1236 .attachment = VK_ATTACHMENT_UNUSED,
1237 .layout = VK_IMAGE_LAYOUT_GENERAL,
1238 },
1239 .preserveAttachmentCount = 1,
1240 .pPreserveAttachments = (uint32_t[]) { 0 },
1241 },
1242 .dependencyCount = 0,
1243 }, &device->meta_state.alloc, &device->meta_state.blit2d.render_pass);
1244 if (result != VK_SUCCESS)
1245 goto fail;
1246
1247 result = anv_CreateDescriptorSetLayout(anv_device_to_handle(device),
1248 &(VkDescriptorSetLayoutCreateInfo) {
1249 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
1250 .bindingCount = 1,
1251 .pBindings = (VkDescriptorSetLayoutBinding[]) {
1252 {
1253 .binding = 0,
1254 .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
1255 .descriptorCount = 1,
1256 .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
1257 .pImmutableSamplers = NULL
1258 },
1259 }
1260 }, &device->meta_state.alloc, &device->meta_state.blit2d.img_ds_layout);
1261 if (result != VK_SUCCESS)
1262 goto fail;
1263
1264 result = anv_CreatePipelineLayout(anv_device_to_handle(device),
1265 &(VkPipelineLayoutCreateInfo) {
1266 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
1267 .setLayoutCount = 1,
1268 .pSetLayouts = &device->meta_state.blit2d.img_ds_layout,
1269 },
1270 &device->meta_state.alloc, &device->meta_state.blit2d.img_p_layout);
1271 if (result != VK_SUCCESS)
1272 goto fail;
1273
1274 result = anv_CreateDescriptorSetLayout(anv_device_to_handle(device),
1275 &(VkDescriptorSetLayoutCreateInfo) {
1276 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
1277 .bindingCount = 1,
1278 .pBindings = (VkDescriptorSetLayoutBinding[]) {
1279 {
1280 .binding = 0,
1281 .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER,
1282 .descriptorCount = 1,
1283 .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
1284 .pImmutableSamplers = NULL
1285 },
1286 }
1287 }, &device->meta_state.alloc, &device->meta_state.blit2d.buf_ds_layout);
1288 if (result != VK_SUCCESS)
1289 goto fail;
1290
1291 result = anv_CreatePipelineLayout(anv_device_to_handle(device),
1292 &(VkPipelineLayoutCreateInfo) {
1293 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
1294 .setLayoutCount = 1,
1295 .pSetLayouts = &device->meta_state.blit2d.buf_ds_layout,
1296 },
1297 &device->meta_state.alloc, &device->meta_state.blit2d.buf_p_layout);
1298 if (result != VK_SUCCESS)
1299 goto fail;
1300
1301 for (unsigned src = 0; src < BLIT2D_NUM_SRC_TYPES; src++) {
1302 for (unsigned dst = 0; dst < BLIT2D_NUM_DST_TYPES; dst++) {
1303 result = blit2d_init_pipeline(device, src, dst);
1304 if (result != VK_SUCCESS)
1305 goto fail;
1306 }
1307 }
1308
1309 return VK_SUCCESS;
1310
1311 fail:
1312 anv_device_finish_meta_blit2d_state(device);
1313 return result;
1314 }