projects / mesa.git / blob
? search:


3fc46318150e1ab3f4983338d790e168c7eaafc4
[mesa.git] / src / intel / vulkan / anv_meta_blit.c
1 /*
2 * Copyright © 2015 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 struct blit_region {
28 VkOffset3D src_offset;
29 VkExtent3D src_extent;
30 VkOffset3D dest_offset;
31 VkExtent3D dest_extent;
32 };
33
34 static nir_shader *
35 build_nir_vertex_shader(void)
36 {
37 const struct glsl_type *vec4 = glsl_vec4_type();
38 nir_builder b;
39
40 nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
41 b.shader->info.name = ralloc_strdup(b.shader, "meta_blit_vs");
42
43 nir_variable *pos_in = nir_variable_create(b.shader, nir_var_shader_in,
44 vec4, "a_pos");
45 pos_in->data.location = VERT_ATTRIB_GENERIC0;
46 nir_variable *pos_out = nir_variable_create(b.shader, nir_var_shader_out,
47 vec4, "gl_Position");
48 pos_out->data.location = VARYING_SLOT_POS;
49 nir_copy_var(&b, pos_out, pos_in);
50
51 nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
52 vec4, "a_tex_pos");
53 tex_pos_in->data.location = VERT_ATTRIB_GENERIC1;
54 nir_variable *tex_pos_out = nir_variable_create(b.shader, nir_var_shader_out,
55 vec4, "v_tex_pos");
56 tex_pos_out->data.location = VARYING_SLOT_VAR0;
57 tex_pos_out->data.interpolation = INTERP_QUALIFIER_SMOOTH;
58 nir_copy_var(&b, tex_pos_out, tex_pos_in);
59
60 return b.shader;
61 }
62
63 static nir_shader *
64 build_nir_copy_fragment_shader(enum glsl_sampler_dim tex_dim)
65 {
66 const struct glsl_type *vec4 = glsl_vec4_type();
67 nir_builder b;
68
69 nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
70 b.shader->info.name = ralloc_strdup(b.shader, "meta_blit_fs");
71
72 nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
73 vec4, "v_tex_pos");
74 tex_pos_in->data.location = VARYING_SLOT_VAR0;
75
76 /* Swizzle the array index which comes in as Z coordinate into the right
77 * position.
78 */
79 unsigned swz[] = { 0, (tex_dim == GLSL_SAMPLER_DIM_1D ? 2 : 1), 2 };
80 nir_ssa_def *const tex_pos =
81 nir_swizzle(&b, nir_load_var(&b, tex_pos_in), swz,
82 (tex_dim == GLSL_SAMPLER_DIM_1D ? 2 : 3), false);
83
84 const struct glsl_type *sampler_type =
85 glsl_sampler_type(tex_dim, false, tex_dim != GLSL_SAMPLER_DIM_3D,
86 glsl_get_base_type(vec4));
87 nir_variable *sampler = nir_variable_create(b.shader, nir_var_uniform,
88 sampler_type, "s_tex");
89 sampler->data.descriptor_set = 0;
90 sampler->data.binding = 0;
91
92 nir_tex_instr *tex = nir_tex_instr_create(b.shader, 1);
93 tex->sampler_dim = tex_dim;
94 tex->op = nir_texop_tex;
95 tex->src[0].src_type = nir_tex_src_coord;
96 tex->src[0].src = nir_src_for_ssa(tex_pos);
97 tex->dest_type = nir_type_float; /* TODO */
98 tex->is_array = glsl_sampler_type_is_array(sampler_type);
99 tex->coord_components = tex_pos->num_components;
100 tex->texture = nir_deref_var_create(tex, sampler);
101 tex->sampler = nir_deref_var_create(tex, sampler);
102
103 nir_ssa_dest_init(&tex->instr, &tex->dest, 4, "tex");
104 nir_builder_instr_insert(&b, &tex->instr);
105
106 nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
107 vec4, "f_color");
108 color_out->data.location = FRAG_RESULT_DATA0;
109 nir_store_var(&b, color_out, &tex->dest.ssa, 4);
110
111 return b.shader;
112 }
113
114 static void
115 meta_prepare_blit(struct anv_cmd_buffer *cmd_buffer,
116 struct anv_meta_saved_state *saved_state)
117 {
118 anv_meta_save(saved_state, cmd_buffer,
119 (1 << VK_DYNAMIC_STATE_VIEWPORT));
120 }
121
122 void
123 anv_meta_begin_blit2d(struct anv_cmd_buffer *cmd_buffer,
124 struct anv_meta_saved_state *save)
125 {
126 meta_prepare_blit(cmd_buffer, save);
127 }
128
129
130 /* Returns the user-provided VkBufferImageCopy::imageOffset in units of
131 * elements rather than texels. One element equals one texel or one block
132 * if Image is uncompressed or compressed, respectively.
133 */
134 static struct VkOffset3D
135 meta_region_offset_el(const struct anv_image * image,
136 const struct VkOffset3D * offset)
137 {
138 const struct isl_format_layout * isl_layout = image->format->isl_layout;
139 return (VkOffset3D) {
140 .x = offset->x / isl_layout->bw,
141 .y = offset->y / isl_layout->bh,
142 .z = offset->z / isl_layout->bd,
143 };
144 }
145
146 /* Returns the user-provided VkBufferImageCopy::imageExtent in units of
147 * elements rather than texels. One element equals one texel or one block
148 * if Image is uncompressed or compressed, respectively.
149 */
150 static struct VkExtent3D
151 meta_region_extent_el(const VkFormat format,
152 const struct VkExtent3D * extent)
153 {
154 const struct isl_format_layout * isl_layout =
155 anv_format_for_vk_format(format)->isl_layout;
156 return (VkExtent3D) {
157 .width = DIV_ROUND_UP(extent->width , isl_layout->bw),
158 .height = DIV_ROUND_UP(extent->height, isl_layout->bh),
159 .depth = DIV_ROUND_UP(extent->depth , isl_layout->bd),
160 };
161 }
162
163 static void
164 meta_emit_blit(struct anv_cmd_buffer *cmd_buffer,
165 struct anv_image *src_image,
166 struct anv_image_view *src_iview,
167 VkOffset3D src_offset,
168 VkExtent3D src_extent,
169 struct anv_image *dest_image,
170 struct anv_image_view *dest_iview,
171 VkOffset3D dest_offset,
172 VkExtent3D dest_extent,
173 VkFilter blit_filter)
174 {
175 struct anv_device *device = cmd_buffer->device;
176
177 struct blit_vb_data {
178 float pos[2];
179 float tex_coord[3];
180 } *vb_data;
181
182 assert(src_image->samples == dest_image->samples);
183
184 unsigned vb_size = sizeof(struct anv_vue_header) + 3 * sizeof(*vb_data);
185
186 struct anv_state vb_state =
187 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, vb_size, 16);
188 memset(vb_state.map, 0, sizeof(struct anv_vue_header));
189 vb_data = vb_state.map + sizeof(struct anv_vue_header);
190
191 vb_data[0] = (struct blit_vb_data) {
192 .pos = {
193 dest_offset.x + dest_extent.width,
194 dest_offset.y + dest_extent.height,
195 },
196 .tex_coord = {
197 (float)(src_offset.x + src_extent.width) / (float)src_iview->extent.width,
198 (float)(src_offset.y + src_extent.height) / (float)src_iview->extent.height,
199 (float)src_offset.z / (float)src_iview->extent.depth,
200 },
201 };
202
203 vb_data[1] = (struct blit_vb_data) {
204 .pos = {
205 dest_offset.x,
206 dest_offset.y + dest_extent.height,
207 },
208 .tex_coord = {
209 (float)src_offset.x / (float)src_iview->extent.width,
210 (float)(src_offset.y + src_extent.height) / (float)src_iview->extent.height,
211 (float)src_offset.z / (float)src_iview->extent.depth,
212 },
213 };
214
215 vb_data[2] = (struct blit_vb_data) {
216 .pos = {
217 dest_offset.x,
218 dest_offset.y,
219 },
220 .tex_coord = {
221 (float)src_offset.x / (float)src_iview->extent.width,
222 (float)src_offset.y / (float)src_iview->extent.height,
223 (float)src_offset.z / (float)src_iview->extent.depth,
224 },
225 };
226
227 anv_state_clflush(vb_state);
228
229 struct anv_buffer vertex_buffer = {
230 .device = device,
231 .size = vb_size,
232 .bo = &device->dynamic_state_block_pool.bo,
233 .offset = vb_state.offset,
234 };
235
236 anv_CmdBindVertexBuffers(anv_cmd_buffer_to_handle(cmd_buffer), 0, 2,
237 (VkBuffer[]) {
238 anv_buffer_to_handle(&vertex_buffer),
239 anv_buffer_to_handle(&vertex_buffer)
240 },
241 (VkDeviceSize[]) {
242 0,
243 sizeof(struct anv_vue_header),
244 });
245
246 VkSampler sampler;
247 ANV_CALL(CreateSampler)(anv_device_to_handle(device),
248 &(VkSamplerCreateInfo) {
249 .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
250 .magFilter = blit_filter,
251 .minFilter = blit_filter,
252 }, &cmd_buffer->pool->alloc, &sampler);
253
254 VkDescriptorPool desc_pool;
255 anv_CreateDescriptorPool(anv_device_to_handle(device),
256 &(const VkDescriptorPoolCreateInfo) {
257 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
258 .pNext = NULL,
259 .flags = 0,
260 .maxSets = 1,
261 .poolSizeCount = 1,
262 .pPoolSizes = (VkDescriptorPoolSize[]) {
263 {
264 .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
265 .descriptorCount = 1
266 },
267 }
268 }, &cmd_buffer->pool->alloc, &desc_pool);
269
270 VkDescriptorSet set;
271 anv_AllocateDescriptorSets(anv_device_to_handle(device),
272 &(VkDescriptorSetAllocateInfo) {
273 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
274 .descriptorPool = desc_pool,
275 .descriptorSetCount = 1,
276 .pSetLayouts = &device->meta_state.blit.ds_layout
277 }, &set);
278
279 anv_UpdateDescriptorSets(anv_device_to_handle(device),
280 1, /* writeCount */
281 (VkWriteDescriptorSet[]) {
282 {
283 .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
284 .dstSet = set,
285 .dstBinding = 0,
286 .dstArrayElement = 0,
287 .descriptorCount = 1,
288 .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
289 .pImageInfo = (VkDescriptorImageInfo[]) {
290 {
291 .sampler = sampler,
292 .imageView = anv_image_view_to_handle(src_iview),
293 .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
294 },
295 }
296 }
297 }, 0, NULL);
298
299 VkFramebuffer fb;
300 anv_CreateFramebuffer(anv_device_to_handle(device),
301 &(VkFramebufferCreateInfo) {
302 .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
303 .attachmentCount = 1,
304 .pAttachments = (VkImageView[]) {
305 anv_image_view_to_handle(dest_iview),
306 },
307 .width = dest_iview->extent.width,
308 .height = dest_iview->extent.height,
309 .layers = 1
310 }, &cmd_buffer->pool->alloc, &fb);
311
312 ANV_CALL(CmdBeginRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer),
313 &(VkRenderPassBeginInfo) {
314 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
315 .renderPass = device->meta_state.blit.render_pass,
316 .framebuffer = fb,
317 .renderArea = {
318 .offset = { dest_offset.x, dest_offset.y },
319 .extent = { dest_extent.width, dest_extent.height },
320 },
321 .clearValueCount = 0,
322 .pClearValues = NULL,
323 }, VK_SUBPASS_CONTENTS_INLINE);
324
325 VkPipeline pipeline;
326
327 switch (src_image->type) {
328 case VK_IMAGE_TYPE_1D:
329 pipeline = device->meta_state.blit.pipeline_1d_src;
330 break;
331 case VK_IMAGE_TYPE_2D:
332 pipeline = device->meta_state.blit.pipeline_2d_src;
333 break;
334 case VK_IMAGE_TYPE_3D:
335 pipeline = device->meta_state.blit.pipeline_3d_src;
336 break;
337 default:
338 unreachable(!"bad VkImageType");
339 }
340
341 if (cmd_buffer->state.pipeline != anv_pipeline_from_handle(pipeline)) {
342 anv_CmdBindPipeline(anv_cmd_buffer_to_handle(cmd_buffer),
343 VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
344 }
345
346 anv_CmdSetViewport(anv_cmd_buffer_to_handle(cmd_buffer), 0, 1,
347 &(VkViewport) {
348 .x = 0.0f,
349 .y = 0.0f,
350 .width = dest_iview->extent.width,
351 .height = dest_iview->extent.height,
352 .minDepth = 0.0f,
353 .maxDepth = 1.0f,
354 });
355
356 anv_CmdBindDescriptorSets(anv_cmd_buffer_to_handle(cmd_buffer),
357 VK_PIPELINE_BIND_POINT_GRAPHICS,
358 device->meta_state.blit.pipeline_layout, 0, 1,
359 &set, 0, NULL);
360
361 ANV_CALL(CmdDraw)(anv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
362
363 ANV_CALL(CmdEndRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer));
364
365 /* At the point where we emit the draw call, all data from the
366 * descriptor sets, etc. has been used. We are free to delete it.
367 */
368 anv_DestroyDescriptorPool(anv_device_to_handle(device),
369 desc_pool, &cmd_buffer->pool->alloc);
370 anv_DestroySampler(anv_device_to_handle(device), sampler,
371 &cmd_buffer->pool->alloc);
372 anv_DestroyFramebuffer(anv_device_to_handle(device), fb,
373 &cmd_buffer->pool->alloc);
374 }
375
376 static void
377 meta_finish_blit(struct anv_cmd_buffer *cmd_buffer,
378 const struct anv_meta_saved_state *saved_state)
379 {
380 anv_meta_restore(saved_state, cmd_buffer);
381 }
382
383 void
384 anv_meta_end_blit2d(struct anv_cmd_buffer *cmd_buffer,
385 struct anv_meta_saved_state *save)
386 {
387 meta_finish_blit(cmd_buffer, save);
388 }
389
390 static VkFormat
391 vk_format_for_size(int bs)
392 {
393 /* The choice of UNORM and UINT formats is very intentional here. Most of
394 * the time, we want to use a UINT format to avoid any rounding error in
395 * the blit. For stencil blits, R8_UINT is required by the hardware.
396 * (It's the only format allowed in conjunction with W-tiling.) Also we
397 * intentionally use the 4-channel formats whenever we can. This is so
398 * that, when we do a RGB <-> RGBX copy, the two formats will line up even
399 * though one of them is 3/4 the size of the other. The choice of UNORM
400 * vs. UINT is also very intentional because Haswell doesn't handle 8 or
401 * 16-bit RGB UINT formats at all so we have to use UNORM there.
402 * Fortunately, the only time we should ever use two different formats in
403 * the table below is for RGB -> RGBA blits and so we will never have any
404 * UNORM/UINT mismatch.
405 */
406 switch (bs) {
407 case 1: return VK_FORMAT_R8_UINT;
408 case 2: return VK_FORMAT_R8G8_UINT;
409 case 3: return VK_FORMAT_R8G8B8_UNORM;
410 case 4: return VK_FORMAT_R8G8B8A8_UNORM;
411 case 6: return VK_FORMAT_R16G16B16_UNORM;
412 case 8: return VK_FORMAT_R16G16B16A16_UNORM;
413 case 12: return VK_FORMAT_R32G32B32_UINT;
414 case 16: return VK_FORMAT_R32G32B32A32_UINT;
415 default:
416 unreachable("Invalid format block size");
417 }
418 }
419
420 static struct anv_meta_blit2d_surf
421 blit_surf_for_image(const struct anv_image* image,
422 const struct isl_surf *img_isl_surf)
423 {
424 return (struct anv_meta_blit2d_surf) {
425 .bo = image->bo,
426 .tiling = img_isl_surf->tiling,
427 .base_offset = image->offset,
428 .bs = isl_format_get_layout(img_isl_surf->format)->bs,
429 .pitch = isl_surf_get_row_pitch(img_isl_surf),
430 };
431 }
432
433 void
434 anv_meta_blit2d(struct anv_cmd_buffer *cmd_buffer,
435 struct anv_meta_blit2d_surf *src,
436 struct anv_meta_blit2d_surf *dst,
437 unsigned num_rects,
438 struct anv_meta_blit2d_rect *rects)
439 {
440 VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
441 VkFormat src_format = vk_format_for_size(src->bs);
442 VkFormat dst_format = vk_format_for_size(dst->bs);
443
444 for (unsigned r = 0; r < num_rects; ++r) {
445
446 /* Create VkImages */
447 VkImageCreateInfo image_info = {
448 .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
449 .imageType = VK_IMAGE_TYPE_2D,
450 .format = 0, /* TEMPLATE */
451 .extent = {
452 .width = 0, /* TEMPLATE */
453 /* Pad to highest tile height to compensate for a vertical intratile offset */
454 .height = MIN(rects[r].height + 64, 1 << 14),
455 .depth = 1,
456 },
457 .mipLevels = 1,
458 .arrayLayers = 1,
459 .samples = 1,
460 .tiling = 0, /* TEMPLATE */
461 .usage = 0, /* TEMPLATE */
462 };
463 struct anv_image_create_info anv_image_info = {
464 .vk_info = &image_info,
465 .isl_tiling_flags = 0, /* TEMPLATE */
466 };
467
468 anv_image_info.isl_tiling_flags = 1 << src->tiling;
469 image_info.tiling = anv_image_info.isl_tiling_flags == ISL_TILING_LINEAR_BIT ?
470 VK_IMAGE_TILING_LINEAR : VK_IMAGE_TILING_OPTIMAL;
471 image_info.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
472 image_info.format = src_format,
473 image_info.extent.width = src->pitch / src->bs;
474 VkImage src_image;
475 anv_image_create(vk_device, &anv_image_info,
476 &cmd_buffer->pool->alloc, &src_image);
477
478 anv_image_info.isl_tiling_flags = 1 << dst->tiling;
479 image_info.tiling = anv_image_info.isl_tiling_flags == ISL_TILING_LINEAR_BIT ?
480 VK_IMAGE_TILING_LINEAR : VK_IMAGE_TILING_OPTIMAL;
481 image_info.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
482 image_info.format = dst_format,
483 image_info.extent.width = dst->pitch / dst->bs;
484 VkImage dst_image;
485 anv_image_create(vk_device, &anv_image_info,
486 &cmd_buffer->pool->alloc, &dst_image);
487
488 /* We could use a vk call to bind memory, but that would require
489 * creating a dummy memory object etc. so there's really no point.
490 */
491 anv_image_from_handle(src_image)->bo = src->bo;
492 anv_image_from_handle(src_image)->offset = src->base_offset;
493 anv_image_from_handle(dst_image)->bo = dst->bo;
494 anv_image_from_handle(dst_image)->offset = dst->base_offset;
495
496 /* Create VkImageViews */
497 VkImageViewCreateInfo iview_info = {
498 .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
499 .image = 0, /* TEMPLATE */
500 .viewType = VK_IMAGE_VIEW_TYPE_2D,
501 .format = 0, /* TEMPLATE */
502 .subresourceRange = {
503 .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
504 .baseMipLevel = 0,
505 .levelCount = 1,
506 .baseArrayLayer = 0,
507 .layerCount = 1
508 },
509 };
510 uint32_t img_o = 0;
511
512 iview_info.image = src_image;
513 iview_info.format = src_format;
514 VkOffset3D src_offset_el = {0};
515 isl_surf_get_image_intratile_offset_el_xy(&cmd_buffer->device->isl_dev,
516 &anv_image_from_handle(src_image)->
517 color_surface.isl,
518 rects[r].src_x,
519 rects[r].src_y,
520 &img_o,
521 (uint32_t*)&src_offset_el.x,
522 (uint32_t*)&src_offset_el.y);
523
524 struct anv_image_view src_iview;
525 anv_image_view_init(&src_iview, cmd_buffer->device,
526 &iview_info, cmd_buffer, img_o, VK_IMAGE_USAGE_SAMPLED_BIT);
527
528 iview_info.image = dst_image;
529 iview_info.format = dst_format;
530 VkOffset3D dst_offset_el = {0};
531 isl_surf_get_image_intratile_offset_el_xy(&cmd_buffer->device->isl_dev,
532 &anv_image_from_handle(dst_image)->
533 color_surface.isl,
534 rects[r].dst_x,
535 rects[r].dst_y,
536 &img_o,
537 (uint32_t*)&dst_offset_el.x,
538 (uint32_t*)&dst_offset_el.y);
539 struct anv_image_view dst_iview;
540 anv_image_view_init(&dst_iview, cmd_buffer->device,
541 &iview_info, cmd_buffer, img_o, VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT);
542
543 /* Perform blit */
544 meta_emit_blit(cmd_buffer,
545 anv_image_from_handle(src_image),
546 &src_iview,
547 src_offset_el,
548 (VkExtent3D){rects[r].width, rects[r].height, 1},
549 anv_image_from_handle(dst_image),
550 &dst_iview,
551 dst_offset_el,
552 (VkExtent3D){rects[r].width, rects[r].height, 1},
553 VK_FILTER_NEAREST);
554
555 anv_DestroyImage(vk_device, src_image, &cmd_buffer->pool->alloc);
556 anv_DestroyImage(vk_device, dst_image, &cmd_buffer->pool->alloc);
557 }
558 }
559
560 static void
561 do_buffer_copy(struct anv_cmd_buffer *cmd_buffer,
562 struct anv_bo *src, uint64_t src_offset,
563 struct anv_bo *dest, uint64_t dest_offset,
564 int width, int height, int bs)
565 {
566 struct anv_meta_blit2d_surf b_src = {
567 .bo = src,
568 .tiling = ISL_TILING_LINEAR,
569 .base_offset = src_offset,
570 .bs = bs,
571 .pitch = width * bs,
572 };
573 struct anv_meta_blit2d_surf b_dst = {
574 .bo = dest,
575 .tiling = ISL_TILING_LINEAR,
576 .base_offset = dest_offset,
577 .bs = bs,
578 .pitch = width * bs,
579 };
580 struct anv_meta_blit2d_rect rect = {
581 .width = width,
582 .height = height,
583 };
584 anv_meta_blit2d(cmd_buffer,
585 &b_src,
586 &b_dst,
587 1,
588 &rect);
589 }
590
591 void anv_CmdCopyBuffer(
592 VkCommandBuffer commandBuffer,
593 VkBuffer srcBuffer,
594 VkBuffer destBuffer,
595 uint32_t regionCount,
596 const VkBufferCopy* pRegions)
597 {
598 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
599 ANV_FROM_HANDLE(anv_buffer, src_buffer, srcBuffer);
600 ANV_FROM_HANDLE(anv_buffer, dest_buffer, destBuffer);
601
602 struct anv_meta_saved_state saved_state;
603
604 anv_meta_begin_blit2d(cmd_buffer, &saved_state);
605
606 for (unsigned r = 0; r < regionCount; r++) {
607 uint64_t src_offset = src_buffer->offset + pRegions[r].srcOffset;
608 uint64_t dest_offset = dest_buffer->offset + pRegions[r].dstOffset;
609 uint64_t copy_size = pRegions[r].size;
610
611 /* First, we compute the biggest format that can be used with the
612 * given offsets and size.
613 */
614 int bs = 16;
615
616 int fs = ffs(src_offset) - 1;
617 if (fs != -1)
618 bs = MIN2(bs, 1 << fs);
619 assert(src_offset % bs == 0);
620
621 fs = ffs(dest_offset) - 1;
622 if (fs != -1)
623 bs = MIN2(bs, 1 << fs);
624 assert(dest_offset % bs == 0);
625
626 fs = ffs(pRegions[r].size) - 1;
627 if (fs != -1)
628 bs = MIN2(bs, 1 << fs);
629 assert(pRegions[r].size % bs == 0);
630
631 /* This is maximum possible width/height our HW can handle */
632 uint64_t max_surface_dim = 1 << 14;
633
634 /* First, we make a bunch of max-sized copies */
635 uint64_t max_copy_size = max_surface_dim * max_surface_dim * bs;
636 while (copy_size >= max_copy_size) {
637 do_buffer_copy(cmd_buffer, src_buffer->bo, src_offset,
638 dest_buffer->bo, dest_offset,
639 max_surface_dim, max_surface_dim, bs);
640 copy_size -= max_copy_size;
641 src_offset += max_copy_size;
642 dest_offset += max_copy_size;
643 }
644
645 uint64_t height = copy_size / (max_surface_dim * bs);
646 assert(height < max_surface_dim);
647 if (height != 0) {
648 uint64_t rect_copy_size = height * max_surface_dim * bs;
649 do_buffer_copy(cmd_buffer, src_buffer->bo, src_offset,
650 dest_buffer->bo, dest_offset,
651 max_surface_dim, height, bs);
652 copy_size -= rect_copy_size;
653 src_offset += rect_copy_size;
654 dest_offset += rect_copy_size;
655 }
656
657 if (copy_size != 0) {
658 do_buffer_copy(cmd_buffer, src_buffer->bo, src_offset,
659 dest_buffer->bo, dest_offset,
660 copy_size / bs, 1, bs);
661 }
662 }
663
664 anv_meta_end_blit2d(cmd_buffer, &saved_state);
665 }
666
667 void anv_CmdUpdateBuffer(
668 VkCommandBuffer commandBuffer,
669 VkBuffer dstBuffer,
670 VkDeviceSize dstOffset,
671 VkDeviceSize dataSize,
672 const uint32_t* pData)
673 {
674 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
675 ANV_FROM_HANDLE(anv_buffer, dst_buffer, dstBuffer);
676 struct anv_meta_saved_state saved_state;
677
678 anv_meta_begin_blit2d(cmd_buffer, &saved_state);
679
680 /* We can't quite grab a full block because the state stream needs a
681 * little data at the top to build its linked list.
682 */
683 const uint32_t max_update_size =
684 cmd_buffer->device->dynamic_state_block_pool.block_size - 64;
685
686 assert(max_update_size < (1 << 14) * 4);
687
688 while (dataSize) {
689 const uint32_t copy_size = MIN2(dataSize, max_update_size);
690
691 struct anv_state tmp_data =
692 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, copy_size, 64);
693
694 memcpy(tmp_data.map, pData, copy_size);
695
696 int bs;
697 if ((copy_size & 15) == 0 && (dstOffset & 15) == 0) {
698 bs = 16;
699 } else if ((copy_size & 7) == 0 && (dstOffset & 7) == 0) {
700 bs = 8;
701 } else {
702 assert((copy_size & 3) == 0 && (dstOffset & 3) == 0);
703 bs = 4;
704 }
705
706 do_buffer_copy(cmd_buffer,
707 &cmd_buffer->device->dynamic_state_block_pool.bo,
708 tmp_data.offset,
709 dst_buffer->bo, dst_buffer->offset + dstOffset,
710 copy_size / bs, 1, bs);
711
712 dataSize -= copy_size;
713 dstOffset += copy_size;
714 pData = (void *)pData + copy_size;
715 }
716
717 anv_meta_end_blit2d(cmd_buffer, &saved_state);
718 }
719
720 static VkFormat
721 choose_iview_format(struct anv_image *image, VkImageAspectFlagBits aspect)
722 {
723 assert(__builtin_popcount(aspect) == 1);
724
725 struct isl_surf *surf =
726 &anv_image_get_surface_for_aspect_mask(image, aspect)->isl;
727
728 /* vkCmdCopyImage behaves like memcpy. Therefore we choose identical UINT
729 * formats for the source and destination image views.
730 *
731 * From the Vulkan spec (2015-12-30):
732 *
733 * vkCmdCopyImage performs image copies in a similar manner to a host
734 * memcpy. It does not perform general-purpose conversions such as
735 * scaling, resizing, blending, color-space conversion, or format
736 * conversions. Rather, it simply copies raw image data. vkCmdCopyImage
737 * can copy between images with different formats, provided the formats
738 * are compatible as defined below.
739 *
740 * [The spec later defines compatibility as having the same number of
741 * bytes per block].
742 */
743 return vk_format_for_size(isl_format_layouts[surf->format].bs);
744 }
745
746 static VkFormat
747 choose_buffer_format(VkFormat format, VkImageAspectFlagBits aspect)
748 {
749 assert(__builtin_popcount(aspect) == 1);
750
751 /* vkCmdCopy* commands behave like memcpy. Therefore we choose
752 * compatable UINT formats for the source and destination image views.
753 *
754 * For the buffer, we go back to the original image format and get a
755 * the format as if it were linear. This way, for RGB formats, we get
756 * an RGB format here even if the tiled image is RGBA. XXX: This doesn't
757 * work if the buffer is the destination.
758 */
759 enum isl_format linear_format = anv_get_isl_format(format, aspect,
760 VK_IMAGE_TILING_LINEAR,
761 NULL);
762
763 return vk_format_for_size(isl_format_layouts[linear_format].bs);
764 }
765
766 void anv_CmdCopyImage(
767 VkCommandBuffer commandBuffer,
768 VkImage srcImage,
769 VkImageLayout srcImageLayout,
770 VkImage destImage,
771 VkImageLayout destImageLayout,
772 uint32_t regionCount,
773 const VkImageCopy* pRegions)
774 {
775 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
776 ANV_FROM_HANDLE(anv_image, src_image, srcImage);
777 ANV_FROM_HANDLE(anv_image, dest_image, destImage);
778 struct anv_meta_saved_state saved_state;
779
780 /* From the Vulkan 1.0 spec:
781 *
782 * vkCmdCopyImage can be used to copy image data between multisample
783 * images, but both images must have the same number of samples.
784 */
785 assert(src_image->samples == dest_image->samples);
786
787 anv_meta_begin_blit2d(cmd_buffer, &saved_state);
788
789 for (unsigned r = 0; r < regionCount; r++) {
790 assert(pRegions[r].srcSubresource.aspectMask ==
791 pRegions[r].dstSubresource.aspectMask);
792
793 VkImageAspectFlags aspect = pRegions[r].srcSubresource.aspectMask;
794
795 /* Create blit surfaces */
796 struct isl_surf *src_isl_surf =
797 &anv_image_get_surface_for_aspect_mask(src_image, aspect)->isl;
798 struct isl_surf *dst_isl_surf =
799 &anv_image_get_surface_for_aspect_mask(dest_image, aspect)->isl;
800 struct anv_meta_blit2d_surf b_src = blit_surf_for_image(src_image, src_isl_surf);
801 struct anv_meta_blit2d_surf b_dst = blit_surf_for_image(dest_image, dst_isl_surf);
802
803 /* Start creating blit rect */
804 const VkOffset3D dst_offset_el = meta_region_offset_el(dest_image, &pRegions[r].dstOffset);
805 const VkOffset3D src_offset_el = meta_region_offset_el(src_image, &pRegions[r].srcOffset);
806 const VkExtent3D img_extent_el = meta_region_extent_el(src_image->vk_format,
807 &pRegions[r].extent);
808 struct anv_meta_blit2d_rect rect = {
809 .width = img_extent_el.width,
810 .height = img_extent_el.height,
811 };
812
813 /* Loop through each 3D or array slice */
814 unsigned num_slices_3d = pRegions[r].extent.depth;
815 unsigned num_slices_array = pRegions[r].dstSubresource.layerCount;
816 unsigned slice_3d = 0;
817 unsigned slice_array = 0;
818 while (slice_3d < num_slices_3d && slice_array < num_slices_array) {
819
820 /* Finish creating blit rect */
821 isl_surf_get_image_offset_el(dst_isl_surf,
822 pRegions[r].dstSubresource.mipLevel,
823 pRegions[r].dstSubresource.baseArrayLayer + slice_array,
824 pRegions[r].dstOffset.z + slice_3d,
825 &rect.dst_x,
826 &rect.dst_y);
827 isl_surf_get_image_offset_el(src_isl_surf,
828 pRegions[r].srcSubresource.mipLevel,
829 pRegions[r].srcSubresource.baseArrayLayer + slice_array,
830 pRegions[r].srcOffset.z + slice_3d,
831 &rect.src_x,
832 &rect.src_y);
833 rect.dst_x += dst_offset_el.x;
834 rect.dst_y += dst_offset_el.y;
835 rect.src_x += src_offset_el.x;
836 rect.src_y += src_offset_el.y;
837
838 /* Perform Blit */
839 anv_meta_blit2d(cmd_buffer,
840 &b_src,
841 &b_dst,
842 1,
843 &rect);
844
845 if (dest_image->type == VK_IMAGE_TYPE_3D)
846 slice_3d++;
847 else
848 slice_array++;
849 }
850 }
851
852 anv_meta_end_blit2d(cmd_buffer, &saved_state);
853 }
854
855 void anv_CmdBlitImage(
856 VkCommandBuffer commandBuffer,
857 VkImage srcImage,
858 VkImageLayout srcImageLayout,
859 VkImage destImage,
860 VkImageLayout destImageLayout,
861 uint32_t regionCount,
862 const VkImageBlit* pRegions,
863 VkFilter filter)
864
865 {
866 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
867 ANV_FROM_HANDLE(anv_image, src_image, srcImage);
868 ANV_FROM_HANDLE(anv_image, dest_image, destImage);
869 struct anv_meta_saved_state saved_state;
870
871 /* From the Vulkan 1.0 spec:
872 *
873 * vkCmdBlitImage must not be used for multisampled source or
874 * destination images. Use vkCmdResolveImage for this purpose.
875 */
876 assert(src_image->samples == 1);
877 assert(dest_image->samples == 1);
878
879 anv_finishme("respect VkFilter");
880
881 meta_prepare_blit(cmd_buffer, &saved_state);
882
883 for (unsigned r = 0; r < regionCount; r++) {
884 struct anv_image_view src_iview;
885 anv_image_view_init(&src_iview, cmd_buffer->device,
886 &(VkImageViewCreateInfo) {
887 .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
888 .image = srcImage,
889 .viewType = anv_meta_get_view_type(src_image),
890 .format = src_image->vk_format,
891 .subresourceRange = {
892 .aspectMask = pRegions[r].srcSubresource.aspectMask,
893 .baseMipLevel = pRegions[r].srcSubresource.mipLevel,
894 .levelCount = 1,
895 .baseArrayLayer = pRegions[r].srcSubresource.baseArrayLayer,
896 .layerCount = 1
897 },
898 },
899 cmd_buffer, 0, VK_IMAGE_USAGE_SAMPLED_BIT);
900
901 const VkOffset3D dest_offset = {
902 .x = pRegions[r].dstOffsets[0].x,
903 .y = pRegions[r].dstOffsets[0].y,
904 .z = 0,
905 };
906
907 if (pRegions[r].dstOffsets[1].x < pRegions[r].dstOffsets[0].x ||
908 pRegions[r].dstOffsets[1].y < pRegions[r].dstOffsets[0].y ||
909 pRegions[r].srcOffsets[1].x < pRegions[r].srcOffsets[0].x ||
910 pRegions[r].srcOffsets[1].y < pRegions[r].srcOffsets[0].y)
911 anv_finishme("FINISHME: Allow flipping in blits");
912
913 const VkExtent3D dest_extent = {
914 .width = pRegions[r].dstOffsets[1].x - pRegions[r].dstOffsets[0].x,
915 .height = pRegions[r].dstOffsets[1].y - pRegions[r].dstOffsets[0].y,
916 };
917
918 const VkExtent3D src_extent = {
919 .width = pRegions[r].srcOffsets[1].x - pRegions[r].srcOffsets[0].x,
920 .height = pRegions[r].srcOffsets[1].y - pRegions[r].srcOffsets[0].y,
921 };
922
923 const uint32_t dest_array_slice =
924 anv_meta_get_iview_layer(dest_image, &pRegions[r].dstSubresource,
925 &pRegions[r].dstOffsets[0]);
926
927 if (pRegions[r].srcSubresource.layerCount > 1)
928 anv_finishme("FINISHME: copy multiple array layers");
929
930 if (pRegions[r].srcOffsets[0].z + 1 != pRegions[r].srcOffsets[1].z ||
931 pRegions[r].dstOffsets[0].z + 1 != pRegions[r].dstOffsets[1].z)
932 anv_finishme("FINISHME: copy multiple depth layers");
933
934 struct anv_image_view dest_iview;
935 anv_image_view_init(&dest_iview, cmd_buffer->device,
936 &(VkImageViewCreateInfo) {
937 .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
938 .image = destImage,
939 .viewType = anv_meta_get_view_type(dest_image),
940 .format = dest_image->vk_format,
941 .subresourceRange = {
942 .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
943 .baseMipLevel = pRegions[r].dstSubresource.mipLevel,
944 .levelCount = 1,
945 .baseArrayLayer = dest_array_slice,
946 .layerCount = 1
947 },
948 },
949 cmd_buffer, 0, VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT);
950
951 meta_emit_blit(cmd_buffer,
952 src_image, &src_iview,
953 pRegions[r].srcOffsets[0], src_extent,
954 dest_image, &dest_iview,
955 dest_offset, dest_extent,
956 filter);
957 }
958
959 meta_finish_blit(cmd_buffer, &saved_state);
960 }
961
962 static void
963 meta_copy_buffer_to_image(struct anv_cmd_buffer *cmd_buffer,
964 struct anv_buffer* buffer,
965 struct anv_image* image,
966 uint32_t regionCount,
967 const VkBufferImageCopy* pRegions,
968 bool forward)
969 {
970 struct anv_meta_saved_state saved_state;
971
972 /* The Vulkan 1.0 spec says "dstImage must have a sample count equal to
973 * VK_SAMPLE_COUNT_1_BIT."
974 */
975 assert(image->samples == 1);
976
977 anv_meta_begin_blit2d(cmd_buffer, &saved_state);
978
979 for (unsigned r = 0; r < regionCount; r++) {
980
981 /* Start creating blit rect */
982 const VkOffset3D img_offset_el = meta_region_offset_el(image, &pRegions[r].imageOffset);
983 const VkExtent3D bufferExtent = {
984 .width = pRegions[r].bufferRowLength,
985 .height = pRegions[r].bufferImageHeight,
986 };
987 const VkExtent3D buf_extent_el = meta_region_extent_el(image->vk_format, &bufferExtent);
988 const VkExtent3D img_extent_el = meta_region_extent_el(image->vk_format,
989 &pRegions[r].imageExtent);
990 struct anv_meta_blit2d_rect rect = {
991 .width = MAX2(buf_extent_el.width, img_extent_el.width),
992 .height = MAX2(buf_extent_el.height, img_extent_el.height),
993 };
994
995 /* Create blit surfaces */
996 VkImageAspectFlags aspect = pRegions[r].imageSubresource.aspectMask;
997 const struct isl_surf *img_isl_surf =
998 &anv_image_get_surface_for_aspect_mask(image, aspect)->isl;
999 struct anv_meta_blit2d_surf img_bsurf = blit_surf_for_image(image, img_isl_surf);
1000 struct anv_meta_blit2d_surf buf_bsurf = {
1001 .bo = buffer->bo,
1002 .tiling = ISL_TILING_LINEAR,
1003 .base_offset = buffer->offset + pRegions[r].bufferOffset,
1004 .bs = forward ? image->format->isl_layout->bs : img_bsurf.bs,
1005 .pitch = rect.width * buf_bsurf.bs,
1006 };
1007
1008 /* Set direction-dependent variables */
1009 struct anv_meta_blit2d_surf *dst_bsurf = forward ? &img_bsurf : &buf_bsurf;
1010 struct anv_meta_blit2d_surf *src_bsurf = forward ? &buf_bsurf : &img_bsurf;
1011 uint32_t *x_offset = forward ? &rect.dst_x : &rect.src_x;
1012 uint32_t *y_offset = forward ? &rect.dst_y : &rect.src_y;
1013
1014 /* Loop through each 3D or array slice */
1015 unsigned num_slices_3d = pRegions[r].imageExtent.depth;
1016 unsigned num_slices_array = pRegions[r].imageSubresource.layerCount;
1017 unsigned slice_3d = 0;
1018 unsigned slice_array = 0;
1019 while (slice_3d < num_slices_3d && slice_array < num_slices_array) {
1020
1021 /* Finish creating blit rect */
1022 isl_surf_get_image_offset_el(img_isl_surf,
1023 pRegions[r].imageSubresource.mipLevel,
1024 pRegions[r].imageSubresource.baseArrayLayer + slice_array,
1025 pRegions[r].imageOffset.z + slice_3d,
1026 x_offset,
1027 y_offset);
1028 *x_offset += img_offset_el.x;
1029 *y_offset += img_offset_el.y;
1030
1031 /* Perform Blit */
1032 anv_meta_blit2d(cmd_buffer,
1033 src_bsurf,
1034 dst_bsurf,
1035 1,
1036 &rect);
1037
1038 /* Once we've done the blit, all of the actual information about
1039 * the image is embedded in the command buffer so we can just
1040 * increment the offset directly in the image effectively
1041 * re-binding it to different backing memory.
1042 */
1043 buf_bsurf.base_offset += rect.width * rect.height * buf_bsurf.bs;
1044
1045 if (image->type == VK_IMAGE_TYPE_3D)
1046 slice_3d++;
1047 else
1048 slice_array++;
1049 }
1050 }
1051 anv_meta_end_blit2d(cmd_buffer, &saved_state);
1052 }
1053
1054 static struct anv_image *
1055 make_image_for_buffer(VkDevice vk_device, VkBuffer vk_buffer, VkFormat format,
1056 VkImageUsageFlags usage,
1057 VkImageType image_type,
1058 const VkAllocationCallbacks *alloc,
1059 const VkBufferImageCopy *copy)
1060 {
1061 ANV_FROM_HANDLE(anv_buffer, buffer, vk_buffer);
1062
1063 VkExtent3D extent = copy->imageExtent;
1064 if (copy->bufferRowLength)
1065 extent.width = copy->bufferRowLength;
1066 if (copy->bufferImageHeight)
1067 extent.height = copy->bufferImageHeight;
1068 extent.depth = 1;
1069 extent = meta_region_extent_el(format, &extent);
1070
1071 VkImageAspectFlags aspect = copy->imageSubresource.aspectMask;
1072 VkFormat buffer_format = choose_buffer_format(format, aspect);
1073
1074 VkImage vk_image;
1075 VkResult result = anv_CreateImage(vk_device,
1076 &(VkImageCreateInfo) {
1077 .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
1078 .imageType = VK_IMAGE_TYPE_2D,
1079 .format = buffer_format,
1080 .extent = extent,
1081 .mipLevels = 1,
1082 .arrayLayers = 1,
1083 .samples = 1,
1084 .tiling = VK_IMAGE_TILING_LINEAR,
1085 .usage = usage,
1086 .flags = 0,
1087 }, alloc, &vk_image);
1088 assert(result == VK_SUCCESS);
1089
1090 ANV_FROM_HANDLE(anv_image, image, vk_image);
1091
1092 /* We could use a vk call to bind memory, but that would require
1093 * creating a dummy memory object etc. so there's really no point.
1094 */
1095 image->bo = buffer->bo;
1096 image->offset = buffer->offset + copy->bufferOffset;
1097
1098 return image;
1099 }
1100
1101 void anv_CmdCopyBufferToImage(
1102 VkCommandBuffer commandBuffer,
1103 VkBuffer srcBuffer,
1104 VkImage destImage,
1105 VkImageLayout destImageLayout,
1106 uint32_t regionCount,
1107 const VkBufferImageCopy* pRegions)
1108 {
1109 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
1110 ANV_FROM_HANDLE(anv_image, dest_image, destImage);
1111 ANV_FROM_HANDLE(anv_buffer, src_buffer, srcBuffer);
1112
1113 meta_copy_buffer_to_image(cmd_buffer, src_buffer, dest_image,
1114 regionCount, pRegions, true);
1115 }
1116
1117 void anv_CmdCopyImageToBuffer(
1118 VkCommandBuffer commandBuffer,
1119 VkImage srcImage,
1120 VkImageLayout srcImageLayout,
1121 VkBuffer destBuffer,
1122 uint32_t regionCount,
1123 const VkBufferImageCopy* pRegions)
1124 {
1125 ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
1126 ANV_FROM_HANDLE(anv_image, src_image, srcImage);
1127 ANV_FROM_HANDLE(anv_buffer, dst_buffer, destBuffer);
1128
1129 meta_copy_buffer_to_image(cmd_buffer, dst_buffer, src_image,
1130 regionCount, pRegions, false);
1131 }
1132
1133 void
1134 anv_device_finish_meta_blit_state(struct anv_device *device)
1135 {
1136 anv_DestroyRenderPass(anv_device_to_handle(device),
1137 device->meta_state.blit.render_pass,
1138 &device->meta_state.alloc);
1139 anv_DestroyPipeline(anv_device_to_handle(device),
1140 device->meta_state.blit.pipeline_1d_src,
1141 &device->meta_state.alloc);
1142 anv_DestroyPipeline(anv_device_to_handle(device),
1143 device->meta_state.blit.pipeline_2d_src,
1144 &device->meta_state.alloc);
1145 anv_DestroyPipeline(anv_device_to_handle(device),
1146 device->meta_state.blit.pipeline_3d_src,
1147 &device->meta_state.alloc);
1148 anv_DestroyPipelineLayout(anv_device_to_handle(device),
1149 device->meta_state.blit.pipeline_layout,
1150 &device->meta_state.alloc);
1151 anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
1152 device->meta_state.blit.ds_layout,
1153 &device->meta_state.alloc);
1154 }
1155
1156 VkResult
1157 anv_device_init_meta_blit_state(struct anv_device *device)
1158 {
1159 VkResult result;
1160
1161 result = anv_CreateRenderPass(anv_device_to_handle(device),
1162 &(VkRenderPassCreateInfo) {
1163 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
1164 .attachmentCount = 1,
1165 .pAttachments = &(VkAttachmentDescription) {
1166 .format = VK_FORMAT_UNDEFINED, /* Our shaders don't care */
1167 .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
1168 .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
1169 .initialLayout = VK_IMAGE_LAYOUT_GENERAL,
1170 .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
1171 },
1172 .subpassCount = 1,
1173 .pSubpasses = &(VkSubpassDescription) {
1174 .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
1175 .inputAttachmentCount = 0,
1176 .colorAttachmentCount = 1,
1177 .pColorAttachments = &(VkAttachmentReference) {
1178 .attachment = 0,
1179 .layout = VK_IMAGE_LAYOUT_GENERAL,
1180 },
1181 .pResolveAttachments = NULL,
1182 .pDepthStencilAttachment = &(VkAttachmentReference) {
1183 .attachment = VK_ATTACHMENT_UNUSED,
1184 .layout = VK_IMAGE_LAYOUT_GENERAL,
1185 },
1186 .preserveAttachmentCount = 1,
1187 .pPreserveAttachments = (uint32_t[]) { 0 },
1188 },
1189 .dependencyCount = 0,
1190 }, &device->meta_state.alloc, &device->meta_state.blit.render_pass);
1191 if (result != VK_SUCCESS)
1192 goto fail;
1193
1194 /* We don't use a vertex shader for blitting, but instead build and pass
1195 * the VUEs directly to the rasterization backend. However, we do need
1196 * to provide GLSL source for the vertex shader so that the compiler
1197 * does not dead-code our inputs.
1198 */
1199 struct anv_shader_module vs = {
1200 .nir = build_nir_vertex_shader(),
1201 };
1202
1203 struct anv_shader_module fs_1d = {
1204 .nir = build_nir_copy_fragment_shader(GLSL_SAMPLER_DIM_1D),
1205 };
1206
1207 struct anv_shader_module fs_2d = {
1208 .nir = build_nir_copy_fragment_shader(GLSL_SAMPLER_DIM_2D),
1209 };
1210
1211 struct anv_shader_module fs_3d = {
1212 .nir = build_nir_copy_fragment_shader(GLSL_SAMPLER_DIM_3D),
1213 };
1214
1215 VkPipelineVertexInputStateCreateInfo vi_create_info = {
1216 .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
1217 .vertexBindingDescriptionCount = 2,
1218 .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
1219 {
1220 .binding = 0,
1221 .stride = 0,
1222 .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
1223 },
1224 {
1225 .binding = 1,
1226 .stride = 5 * sizeof(float),
1227 .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
1228 },
1229 },
1230 .vertexAttributeDescriptionCount = 3,
1231 .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
1232 {
1233 /* VUE Header */
1234 .location = 0,
1235 .binding = 0,
1236 .format = VK_FORMAT_R32G32B32A32_UINT,
1237 .offset = 0
1238 },
1239 {
1240 /* Position */
1241 .location = 1,
1242 .binding = 1,
1243 .format = VK_FORMAT_R32G32_SFLOAT,
1244 .offset = 0
1245 },
1246 {
1247 /* Texture Coordinate */
1248 .location = 2,
1249 .binding = 1,
1250 .format = VK_FORMAT_R32G32B32_SFLOAT,
1251 .offset = 8
1252 }
1253 }
1254 };
1255
1256 VkDescriptorSetLayoutCreateInfo ds_layout_info = {
1257 .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
1258 .bindingCount = 1,
1259 .pBindings = (VkDescriptorSetLayoutBinding[]) {
1260 {
1261 .binding = 0,
1262 .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
1263 .descriptorCount = 1,
1264 .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
1265 .pImmutableSamplers = NULL
1266 },
1267 }
1268 };
1269 result = anv_CreateDescriptorSetLayout(anv_device_to_handle(device),
1270 &ds_layout_info,
1271 &device->meta_state.alloc,
1272 &device->meta_state.blit.ds_layout);
1273 if (result != VK_SUCCESS)
1274 goto fail_render_pass;
1275
1276 result = anv_CreatePipelineLayout(anv_device_to_handle(device),
1277 &(VkPipelineLayoutCreateInfo) {
1278 .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
1279 .setLayoutCount = 1,
1280 .pSetLayouts = &device->meta_state.blit.ds_layout,
1281 },
1282 &device->meta_state.alloc, &device->meta_state.blit.pipeline_layout);
1283 if (result != VK_SUCCESS)
1284 goto fail_descriptor_set_layout;
1285
1286 VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
1287 {
1288 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
1289 .stage = VK_SHADER_STAGE_VERTEX_BIT,
1290 .module = anv_shader_module_to_handle(&vs),
1291 .pName = "main",
1292 .pSpecializationInfo = NULL
1293 }, {
1294 .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
1295 .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
1296 .module = VK_NULL_HANDLE, /* TEMPLATE VALUE! FILL ME IN! */
1297 .pName = "main",
1298 .pSpecializationInfo = NULL
1299 },
1300 };
1301
1302 const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
1303 .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
1304 .stageCount = ARRAY_SIZE(pipeline_shader_stages),
1305 .pStages = pipeline_shader_stages,
1306 .pVertexInputState = &vi_create_info,
1307 .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
1308 .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
1309 .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
1310 .primitiveRestartEnable = false,
1311 },
1312 .pViewportState = &(VkPipelineViewportStateCreateInfo) {
1313 .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
1314 .viewportCount = 1,
1315 .scissorCount = 1,
1316 },
1317 .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
1318 .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
1319 .rasterizerDiscardEnable = false,
1320 .polygonMode = VK_POLYGON_MODE_FILL,
1321 .cullMode = VK_CULL_MODE_NONE,
1322 .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE
1323 },
1324 .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
1325 .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
1326 .rasterizationSamples = 1,
1327 .sampleShadingEnable = false,
1328 .pSampleMask = (VkSampleMask[]) { UINT32_MAX },
1329 },
1330 .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
1331 .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
1332 .attachmentCount = 1,
1333 .pAttachments = (VkPipelineColorBlendAttachmentState []) {
1334 { .colorWriteMask =
1335 VK_COLOR_COMPONENT_A_BIT |
1336 VK_COLOR_COMPONENT_R_BIT |
1337 VK_COLOR_COMPONENT_G_BIT |
1338 VK_COLOR_COMPONENT_B_BIT },
1339 }
1340 },
1341 .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
1342 .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
1343 .dynamicStateCount = 9,
1344 .pDynamicStates = (VkDynamicState[]) {
1345 VK_DYNAMIC_STATE_VIEWPORT,
1346 VK_DYNAMIC_STATE_SCISSOR,
1347 VK_DYNAMIC_STATE_LINE_WIDTH,
1348 VK_DYNAMIC_STATE_DEPTH_BIAS,
1349 VK_DYNAMIC_STATE_BLEND_CONSTANTS,
1350 VK_DYNAMIC_STATE_DEPTH_BOUNDS,
1351 VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
1352 VK_DYNAMIC_STATE_STENCIL_WRITE_MASK,
1353 VK_DYNAMIC_STATE_STENCIL_REFERENCE,
1354 },
1355 },
1356 .flags = 0,
1357 .layout = device->meta_state.blit.pipeline_layout,
1358 .renderPass = device->meta_state.blit.render_pass,
1359 .subpass = 0,
1360 };
1361
1362 const struct anv_graphics_pipeline_create_info anv_pipeline_info = {
1363 .color_attachment_count = -1,
1364 .use_repclear = false,
1365 .disable_viewport = true,
1366 .disable_scissor = true,
1367 .disable_vs = true,
1368 .use_rectlist = true
1369 };
1370
1371 pipeline_shader_stages[1].module = anv_shader_module_to_handle(&fs_1d);
1372 result = anv_graphics_pipeline_create(anv_device_to_handle(device),
1373 VK_NULL_HANDLE,
1374 &vk_pipeline_info, &anv_pipeline_info,
1375 &device->meta_state.alloc, &device->meta_state.blit.pipeline_1d_src);
1376 if (result != VK_SUCCESS)
1377 goto fail_pipeline_layout;
1378
1379 pipeline_shader_stages[1].module = anv_shader_module_to_handle(&fs_2d);
1380 result = anv_graphics_pipeline_create(anv_device_to_handle(device),
1381 VK_NULL_HANDLE,
1382 &vk_pipeline_info, &anv_pipeline_info,
1383 &device->meta_state.alloc, &device->meta_state.blit.pipeline_2d_src);
1384 if (result != VK_SUCCESS)
1385 goto fail_pipeline_1d;
1386
1387 pipeline_shader_stages[1].module = anv_shader_module_to_handle(&fs_3d);
1388 result = anv_graphics_pipeline_create(anv_device_to_handle(device),
1389 VK_NULL_HANDLE,
1390 &vk_pipeline_info, &anv_pipeline_info,
1391 &device->meta_state.alloc, &device->meta_state.blit.pipeline_3d_src);
1392 if (result != VK_SUCCESS)
1393 goto fail_pipeline_2d;
1394
1395 ralloc_free(vs.nir);
1396 ralloc_free(fs_1d.nir);
1397 ralloc_free(fs_2d.nir);
1398 ralloc_free(fs_3d.nir);
1399
1400 return VK_SUCCESS;
1401
1402 fail_pipeline_2d:
1403 anv_DestroyPipeline(anv_device_to_handle(device),
1404 device->meta_state.blit.pipeline_2d_src,
1405 &device->meta_state.alloc);
1406
1407 fail_pipeline_1d:
1408 anv_DestroyPipeline(anv_device_to_handle(device),
1409 device->meta_state.blit.pipeline_1d_src,
1410 &device->meta_state.alloc);
1411
1412 fail_pipeline_layout:
1413 anv_DestroyPipelineLayout(anv_device_to_handle(device),
1414 device->meta_state.blit.pipeline_layout,
1415 &device->meta_state.alloc);
1416 fail_descriptor_set_layout:
1417 anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
1418 device->meta_state.blit.ds_layout,
1419 &device->meta_state.alloc);
1420 fail_render_pass:
1421 anv_DestroyRenderPass(anv_device_to_handle(device),
1422 device->meta_state.blit.render_pass,
1423 &device->meta_state.alloc);
1424
1425 ralloc_free(vs.nir);
1426 ralloc_free(fs_1d.nir);
1427 ralloc_free(fs_2d.nir);
1428 ralloc_free(fs_3d.nir);
1429 fail:
1430 return result;
1431 }