diff options
author | nyorain <nyorain@gmail.com> | 2021-02-21 18:30:12 +0100 |
---|---|---|
committer | Simon Ser <contact@emersion.fr> | 2021-10-18 11:51:13 +0200 |
commit | 8e346922508aa3eaccd6e12f2917f6574f349843 (patch) | |
tree | 550742b8a086287b6d478db1ade14b2cc4a21294 /render | |
parent | 2edf468aeb7d4703aa211cea3b58f04cbc73298c (diff) |
render/vulkan: add Vulkan renderer
This new renderer is implemented with the existing wlr_renderer API
(which is known to be sub-optimal for some operations). It's not
used by default, but users can opt-in by setting WLR_RENDERER=vulkan.
The renderer depends on VK_EXT_image_drm_format_modifier and
VK_EXT_physical_device_drm.
Co-authored-by: Simon Ser <contact@emersion.fr>
Co-authored-by: Jan Beich <jbeich@FreeBSD.org>
Diffstat (limited to 'render')
-rw-r--r-- | render/meson.build | 6 | ||||
-rw-r--r-- | render/vulkan/meson.build | 38 | ||||
-rw-r--r-- | render/vulkan/pixel_format.c | 325 | ||||
-rw-r--r-- | render/vulkan/renderer.c | 1540 | ||||
-rw-r--r-- | render/vulkan/shaders/common.vert | 25 | ||||
-rw-r--r-- | render/vulkan/shaders/meson.build | 20 | ||||
-rw-r--r-- | render/vulkan/shaders/quad.frag | 10 | ||||
-rw-r--r-- | render/vulkan/shaders/texture.frag | 25 | ||||
-rw-r--r-- | render/vulkan/texture.c | 718 | ||||
-rw-r--r-- | render/vulkan/util.c | 93 | ||||
-rw-r--r-- | render/vulkan/vulkan.c | 550 | ||||
-rw-r--r-- | render/wlr_renderer.c | 9 |
12 files changed, 3358 insertions, 1 deletions
diff --git a/render/meson.build b/render/meson.build index b35ba7c9..2d651887 100644 --- a/render/meson.build +++ b/render/meson.build @@ -1,6 +1,6 @@ renderers = get_option('renderers') if 'auto' in renderers and get_option('auto_features').enabled() - renderers = ['gles2'] + renderers = ['gles2', 'vulkan'] elif 'auto' in renderers and get_option('auto_features').disabled() renderers = [] endif @@ -24,6 +24,10 @@ if 'gles2' in renderers or 'auto' in renderers subdir('gles2') endif +if 'vulkan' in renderers or 'auto' in renderers + subdir('vulkan') +endif + subdir('pixman') subdir('allocator') diff --git a/render/vulkan/meson.build b/render/vulkan/meson.build new file mode 100644 index 00000000..6b9ce840 --- /dev/null +++ b/render/vulkan/meson.build @@ -0,0 +1,38 @@ +msg = [] +if 'vulkan' in renderers + msg += 'Install "@0@" or pass "-Dvulkan=disabled" to disable it.' +else + msg += 'Required for vulkan renderer support.' +endif + +dep_vulkan = dependency('vulkan', + version: '>=1.2.182', + required: 'vulkan' in renderers, + not_found_message: '\n'.join(msg).format('vulkan') +) + +if not dep_vulkan.found() + subdir_done() +endif + +glslang = find_program('glslangValidator', native: true, required: false) +if not glslang.found() + if 'vulkan' in renderers + error('\n'.join(msg).format('glslang')) + else + subdir_done() + endif +endif + +wlr_files += files( + 'renderer.c', + 'texture.c', + 'vulkan.c', + 'util.c', + 'pixel_format.c', +) + +wlr_deps += dep_vulkan +features += { 'vulkan-renderer': true } + +subdir('shaders') diff --git a/render/vulkan/pixel_format.c b/render/vulkan/pixel_format.c new file mode 100644 index 00000000..cfab5ffe --- /dev/null +++ b/render/vulkan/pixel_format.c @@ -0,0 +1,325 @@ +#include <drm_fourcc.h> +#include <stdlib.h> +#include <vulkan/vulkan.h> +#include <wlr/util/log.h> +#include "render/vulkan.h" + +// Reversed endianess of shm and vulkan format names +static const struct wlr_vk_format formats[] = { + { + .drm_format = DRM_FORMAT_ARGB8888, + .vk_format = VK_FORMAT_B8G8R8A8_SRGB, + }, + { + .drm_format = DRM_FORMAT_XRGB8888, + .vk_format = VK_FORMAT_B8G8R8A8_SRGB, + }, + { + .drm_format = DRM_FORMAT_XBGR8888, + .vk_format = VK_FORMAT_R8G8B8A8_SRGB, + }, + { + .drm_format = DRM_FORMAT_ABGR8888, + .vk_format = VK_FORMAT_R8G8B8A8_SRGB, + }, +}; + +const struct wlr_vk_format *vulkan_get_format_list(size_t *len) { + *len = sizeof(formats) / sizeof(formats[0]); + return formats; +} + +const struct wlr_vk_format *vulkan_get_format_from_drm(uint32_t drm_format) { + for (unsigned i = 0; i < sizeof(formats) / sizeof(formats[0]); ++i) { + if (formats[i].drm_format == drm_format) { + return &formats[i]; + } + } + return NULL; +} + +static const VkImageUsageFlags render_usage = + VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; +static const VkImageUsageFlags tex_usage = + VK_IMAGE_USAGE_SAMPLED_BIT | + VK_IMAGE_USAGE_TRANSFER_DST_BIT; +static const VkImageUsageFlags dma_tex_usage = + VK_IMAGE_USAGE_SAMPLED_BIT; + +static const VkFormatFeatureFlags tex_features = + VK_FORMAT_FEATURE_TRANSFER_SRC_BIT | + VK_FORMAT_FEATURE_TRANSFER_DST_BIT | + VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT | + // NOTE: we don't strictly require this, we could create a NEAREST + // sampler for formats that need it, in case this ever makes problems. + VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT; +static const VkFormatFeatureFlags render_features = + VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT | + VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT; +static const VkFormatFeatureFlags dma_tex_features = + VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT | + // NOTE: we don't strictly require this, we could create a NEAREST + // sampler for formats that need it, in case this ever makes problems. + VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_LINEAR_BIT; + +static bool query_modifier_support(struct wlr_vk_device *dev, + struct wlr_vk_format_props *props, size_t modifier_count, + VkPhysicalDeviceImageFormatInfo2 fmti) { + VkResult res; + + VkFormatProperties2 fmtp = {0}; + fmtp.sType = VK_STRUCTURE_TYPE_FORMAT_PROPERTIES_2; + + VkDrmFormatModifierPropertiesListEXT modp = {0}; + modp.sType = VK_STRUCTURE_TYPE_DRM_FORMAT_MODIFIER_PROPERTIES_LIST_EXT; + modp.drmFormatModifierCount = modifier_count; + fmtp.pNext = &modp; + + // the first call to vkGetPhysicalDeviceFormatProperties2 did only + // retrieve the number of modifiers, we now have to retrieve + // the modifiers + modp.pDrmFormatModifierProperties = calloc(modifier_count, + sizeof(*modp.pDrmFormatModifierProperties)); + if (!modp.pDrmFormatModifierProperties) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + return false; + } + + vkGetPhysicalDeviceFormatProperties2(dev->phdev, + props->format.vk_format, &fmtp); + + props->render_mods = calloc(modp.drmFormatModifierCount, + sizeof(*props->render_mods)); + if (!props->render_mods) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + free(modp.pDrmFormatModifierProperties); + return false; + } + + props->texture_mods = calloc(modp.drmFormatModifierCount, + sizeof(*props->texture_mods)); + if (!props->texture_mods) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + free(modp.pDrmFormatModifierProperties); + free(props->render_mods); + return false; + } + + // detailed check + // format info + // only added if dmabuf/drm_fmt_ext supported + VkPhysicalDeviceExternalImageFormatInfo efmti = {0}; + efmti.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_IMAGE_FORMAT_INFO; + efmti.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT; + + fmti.tiling = VK_IMAGE_TILING_DRM_FORMAT_MODIFIER_EXT; + fmti.pNext = &efmti; + + VkPhysicalDeviceImageDrmFormatModifierInfoEXT modi = {0}; + modi.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGE_DRM_FORMAT_MODIFIER_INFO_EXT; + modi.sharingMode = VK_SHARING_MODE_EXCLUSIVE; + efmti.pNext = &modi; + + // format properties + VkExternalImageFormatProperties efmtp = {0}; + efmtp.sType = VK_STRUCTURE_TYPE_EXTERNAL_IMAGE_FORMAT_PROPERTIES; + + VkImageFormatProperties2 ifmtp = {0}; + ifmtp.sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_PROPERTIES_2; + ifmtp.pNext = &efmtp; + const VkExternalMemoryProperties *emp = &efmtp.externalMemoryProperties; + + bool found = false; + + for (unsigned i = 0u; i < modp.drmFormatModifierCount; ++i) { + VkDrmFormatModifierPropertiesEXT m = + modp.pDrmFormatModifierProperties[i]; + wlr_log(WLR_DEBUG, " modifier: 0x%"PRIx64 ": features 0x%"PRIx32", %d planes", + m.drmFormatModifier, m.drmFormatModifierTilingFeatures, + m.drmFormatModifierPlaneCount); + + // check that specific modifier for render usage + if ((m.drmFormatModifierTilingFeatures & render_features) == render_features) { + fmti.usage = render_usage; + + modi.drmFormatModifier = m.drmFormatModifier; + res = vkGetPhysicalDeviceImageFormatProperties2( + dev->phdev, &fmti, &ifmtp); + if (res != VK_SUCCESS) { + if (res != VK_ERROR_FORMAT_NOT_SUPPORTED) { + wlr_vk_error("vkGetPhysicalDeviceImageFormatProperties2", + res); + } + + wlr_log(WLR_DEBUG, " >> rendering: format not supported"); + } else if (emp->externalMemoryFeatures & + VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT) { + unsigned c = props->render_mod_count; + VkExtent3D me = ifmtp.imageFormatProperties.maxExtent; + VkExternalMemoryProperties emp = efmtp.externalMemoryProperties; + props->render_mods[c].props = m; + props->render_mods[c].max_extent.width = me.width; + props->render_mods[c].max_extent.height = me.height; + props->render_mods[c].dmabuf_flags = emp.externalMemoryFeatures; + props->render_mods[c].export_imported = + (emp.exportFromImportedHandleTypes & + VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT); + ++props->render_mod_count; + + found = true; + wlr_drm_format_set_add(&dev->dmabuf_render_formats, + props->format.drm_format, m.drmFormatModifier); + + wlr_log(WLR_DEBUG, " >> rendering: supported"); + } else { + wlr_log(WLR_DEBUG, " >> rendering: importing not supported"); + } + } else { + wlr_log(WLR_DEBUG, " >> rendering: format features not supported"); + } + + // check that specific modifier for texture usage + if ((m.drmFormatModifierTilingFeatures & dma_tex_features) == dma_tex_features) { + fmti.usage = dma_tex_usage; + + modi.drmFormatModifier = m.drmFormatModifier; + res = vkGetPhysicalDeviceImageFormatProperties2( + dev->phdev, &fmti, &ifmtp); + if (res != VK_SUCCESS) { + if (res != VK_ERROR_FORMAT_NOT_SUPPORTED) { + wlr_vk_error("vkGetPhysicalDeviceImageFormatProperties2", + res); + } + + wlr_log(WLR_DEBUG, " >> dmatex: format not supported"); + } else if (emp->externalMemoryFeatures & + VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT) { + unsigned c = props->texture_mod_count; + VkExtent3D me = ifmtp.imageFormatProperties.maxExtent; + VkExternalMemoryProperties emp = efmtp.externalMemoryProperties; + props->texture_mods[c].props = m; + props->texture_mods[c].max_extent.width = me.width; + props->texture_mods[c].max_extent.height = me.height; + props->texture_mods[c].dmabuf_flags = emp.externalMemoryFeatures; + props->texture_mods[c].export_imported = + (emp.exportFromImportedHandleTypes & + VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT); + ++props->texture_mod_count; + + found = true; + wlr_drm_format_set_add(&dev->dmabuf_texture_formats, + props->format.drm_format, m.drmFormatModifier); + + wlr_log(WLR_DEBUG, " >> dmatex: supported"); + } else { + wlr_log(WLR_DEBUG, " >> dmatex: importing not supported"); + } + } else { + wlr_log(WLR_DEBUG, " >> dmatex: format features not supported"); + } + } + + free(modp.pDrmFormatModifierProperties); + return found; +} + +void vulkan_format_props_query(struct wlr_vk_device *dev, + const struct wlr_vk_format *format) { + + wlr_log(WLR_DEBUG, "vulkan: Checking support for format %.4s (0x%" PRIx32 ")", + (const char *)&format->drm_format, format->drm_format); + VkResult res; + + // get general features and modifiers + VkFormatProperties2 fmtp = {0}; + fmtp.sType = VK_STRUCTURE_TYPE_FORMAT_PROPERTIES_2; + + VkDrmFormatModifierPropertiesListEXT modp = {0}; + modp.sType = VK_STRUCTURE_TYPE_DRM_FORMAT_MODIFIER_PROPERTIES_LIST_EXT; + fmtp.pNext = &modp; + + vkGetPhysicalDeviceFormatProperties2(dev->phdev, + format->vk_format, &fmtp); + + // detailed check + VkPhysicalDeviceImageFormatInfo2 fmti = {0}; + fmti.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGE_FORMAT_INFO_2; + fmti.type = VK_IMAGE_TYPE_2D; + fmti.format = format->vk_format; + + VkImageFormatProperties2 ifmtp = {0}; + ifmtp.sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_PROPERTIES_2; + + bool add_fmt_props = false; + struct wlr_vk_format_props props = {0}; + props.format = *format; + + wlr_log(WLR_DEBUG, " drmFormatModifierCount: %d", modp.drmFormatModifierCount); + if (modp.drmFormatModifierCount > 0) { + add_fmt_props |= query_modifier_support(dev, &props, + modp.drmFormatModifierCount, fmti); + } + + // non-dmabuf texture properties + if (fmtp.formatProperties.optimalTilingFeatures & tex_features) { + fmti.pNext = NULL; + ifmtp.pNext = NULL; + fmti.tiling = VK_IMAGE_TILING_OPTIMAL; + fmti.usage = tex_usage; + + res = vkGetPhysicalDeviceImageFormatProperties2( + dev->phdev, &fmti, &ifmtp); + if (res != VK_SUCCESS) { + if (res != VK_ERROR_FORMAT_NOT_SUPPORTED) { + wlr_vk_error("vkGetPhysicalDeviceImageFormatProperties2", + res); + } + + wlr_log(WLR_DEBUG, " >> shmtex: format not supported"); + } else { + VkExtent3D me = ifmtp.imageFormatProperties.maxExtent; + props.max_extent.width = me.width; + props.max_extent.height = me.height; + props.features = fmtp.formatProperties.optimalTilingFeatures; + + wlr_log(WLR_DEBUG, " >> shmtex: supported"); + + dev->shm_formats[dev->shm_format_count] = format->drm_format; + ++dev->shm_format_count; + + add_fmt_props = true; + } + } else { + wlr_log(WLR_DEBUG, " >> shmtex: format features not supported"); + } + + if (add_fmt_props) { + dev->format_props[dev->format_prop_count] = props; + ++dev->format_prop_count; + } +} + +void vulkan_format_props_finish(struct wlr_vk_format_props *props) { + free(props->texture_mods); + free(props->render_mods); +} + +struct wlr_vk_format_modifier_props *vulkan_format_props_find_modifier( + struct wlr_vk_format_props *props, uint64_t mod, bool render) { + if (render) { + for (unsigned i = 0u; i < props->render_mod_count; ++i) { + if (props->render_mods[i].props.drmFormatModifier == mod) { + return &props->render_mods[i]; + } + } + } else { + for (unsigned i = 0u; i < props->texture_mod_count; ++i) { + if (props->texture_mods[i].props.drmFormatModifier == mod) { + return &props->texture_mods[i]; + } + } + } + + return NULL; +} + diff --git a/render/vulkan/renderer.c b/render/vulkan/renderer.c new file mode 100644 index 00000000..4fae81a8 --- /dev/null +++ b/render/vulkan/renderer.c @@ -0,0 +1,1540 @@ +#define _POSIX_C_SOURCE 200809L +#include <assert.h> +#include <fcntl.h> +#include <math.h> +#include <stdlib.h> +#include <stdint.h> +#include <sys/types.h> +#include <unistd.h> +#include <drm_fourcc.h> +#include <vulkan/vulkan.h> +#include <wlr/render/interface.h> +#include <wlr/types/wlr_drm.h> +#include <wlr/types/wlr_matrix.h> +#include <wlr/util/box.h> +#include <wlr/util/log.h> +#include <wlr/render/vulkan.h> +#include <wlr/backend/interface.h> +#include <wlr/types/wlr_linux_dmabuf_v1.h> + +#include "render/pixel_format.h" +#include "render/vulkan.h" +#include "render/vulkan/shaders/common.vert.h" +#include "render/vulkan/shaders/texture.frag.h" +#include "render/vulkan/shaders/quad.frag.h" +#include "types/wlr_buffer.h" + +// TODO: +// - simplify stage allocation, don't track allocations but use ringbuffer-like +// - use a pipeline cache (not sure when to save though, after every pipeline +// creation?) +// - create pipelines as derivatives of each other +// - evaluate if creating VkDeviceMemory pools is a good idea. +// We can expect wayland client images to be fairly large (and shouldn't +// have more than 4k of those I guess) but pooling memory allocations +// might still be a good idea. + +static const VkDeviceSize min_stage_size = 1024 * 1024; // 1MB +static const VkDeviceSize max_stage_size = 64 * min_stage_size; // 64MB +static const size_t start_descriptor_pool_size = 256u; +static bool default_debug = true; + +static const struct wlr_renderer_impl renderer_impl; + +struct wlr_vk_renderer *vulkan_get_renderer(struct wlr_renderer *wlr_renderer) { + assert(wlr_renderer->impl == &renderer_impl); + return (struct wlr_vk_renderer *)wlr_renderer; +} + +static struct wlr_vk_render_format_setup *find_or_create_render_setup( + struct wlr_vk_renderer *renderer, VkFormat format); + +// vertex shader push constant range data +struct vert_pcr_data { + float mat4[4][4]; + float uv_off[2]; + float uv_size[2]; +}; + +// https://www.w3.org/Graphics/Color/srgb +static float color_to_linear(float non_linear) { + return (non_linear > 0.04045) ? + pow((non_linear + 0.055) / 1.055, 2.4) : + non_linear / 12.92; +} + +// renderer +// util +static void mat3_to_mat4(const float mat3[9], float mat4[4][4]) { + memset(mat4, 0, sizeof(float) * 16); + mat4[0][0] = mat3[0]; + mat4[0][1] = mat3[1]; + mat4[0][3] = mat3[2]; + + mat4[1][0] = mat3[3]; + mat4[1][1] = mat3[4]; + mat4[1][3] = mat3[5]; + + mat4[2][2] = 1.f; + mat4[3][3] = 1.f; +} + +struct wlr_vk_descriptor_pool *vulkan_alloc_texture_ds( + struct wlr_vk_renderer *renderer, VkDescriptorSet *ds) { + VkResult res; + VkDescriptorSetAllocateInfo ds_info = {0}; + ds_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO; + ds_info.descriptorSetCount = 1; + ds_info.pSetLayouts = &renderer->ds_layout; + + bool found = false; + struct wlr_vk_descriptor_pool *pool; + wl_list_for_each(pool, &renderer->descriptor_pools, link) { + if (pool->free > 0) { + found = true; + break; + } + } + + if (!found) { // create new pool + pool = calloc(1, sizeof(*pool)); + if (!pool) { + wlr_log_errno(WLR_ERROR, "allocation failed"); + return NULL; + } + + size_t count = renderer->last_pool_size; + if (!count) { + count = start_descriptor_pool_size; + } + + pool->free = count; + VkDescriptorPoolSize pool_size = {0}; + pool_size.descriptorCount = count; + pool_size.type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; + + VkDescriptorPoolCreateInfo dpool_info = {0}; + dpool_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO; + dpool_info.maxSets = count; + dpool_info.poolSizeCount = 1; + dpool_info.pPoolSizes = &pool_size; + dpool_info.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT; + + res = vkCreateDescriptorPool(renderer->dev->dev, &dpool_info, NULL, + &pool->pool); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateDescriptorPool", res); + free(pool); + return NULL; + } + + wl_list_insert(&renderer->descriptor_pools, &pool->link); + } + + ds_info.descriptorPool = pool->pool; + res = vkAllocateDescriptorSets(renderer->dev->dev, &ds_info, ds); + if (res != VK_SUCCESS) { + wlr_vk_error("vkAllocateDescriptorSets", res); + return NULL; + } + + --pool->free; + return pool; +} + +void vulkan_free_ds(struct wlr_vk_renderer *renderer, + struct wlr_vk_descriptor_pool *pool, VkDescriptorSet ds) { + vkFreeDescriptorSets(renderer->dev->dev, pool->pool, 1, &ds); + ++pool->free; +} + +static void destroy_render_format_setup(struct wlr_vk_renderer *renderer, + struct wlr_vk_render_format_setup *setup) { + if (!setup) { + return; + } + + VkDevice dev = renderer->dev->dev; + vkDestroyRenderPass(dev, setup->render_pass, NULL); + vkDestroyPipeline(dev, setup->tex_pipe, NULL); + vkDestroyPipeline(dev, setup->quad_pipe, NULL); +} + +static void shared_buffer_destroy(struct wlr_vk_renderer *r, + struct wlr_vk_shared_buffer *buffer) { + if (!buffer) { + return; + } + + if (buffer->allocs_size > 0) { + wlr_log(WLR_ERROR, "shared_buffer_finish: %d allocations left", + (unsigned) buffer->allocs_size); + } + + free(buffer->allocs); + if (buffer->buffer) { + vkDestroyBuffer(r->dev->dev, buffer->buffer, NULL); + } + if (buffer->memory) { + vkFreeMemory(r->dev->dev, buffer->memory, NULL); + } + + wl_list_remove(&buffer->link); + free(buffer); +} + +static void release_stage_allocations(struct wlr_vk_renderer *renderer) { + struct wlr_vk_shared_buffer *buf; + wl_list_for_each(buf, &renderer->stage.buffers, link) { + buf->allocs_size = 0u; + } +} + +struct wlr_vk_buffer_span vulkan_get_stage_span(struct wlr_vk_renderer *r, + VkDeviceSize size) { + // try to find free span + // simple greedy allocation algorithm - should be enough for this usecase + // since all allocations are freed together after the frame + struct wlr_vk_shared_buffer *buf; + wl_list_for_each_reverse(buf, &r->stage.buffers, link) { + VkDeviceSize start = 0u; + if (buf->allocs_size > 0) { + struct wlr_vk_allocation *last = &buf->allocs[buf->allocs_size - 1]; + start = last->start + last->size; + } + + assert(start <= buf->buf_size); + if (buf->buf_size - start < size) { + continue; + } + + ++buf->allocs_size; + if (buf->allocs_size > buf->allocs_capacity) { + buf->allocs_capacity = buf->allocs_size * 2; + void *allocs = realloc(buf->allocs, + buf->allocs_capacity * sizeof(*buf->allocs)); + if (!allocs) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + goto error_alloc; + } + + buf->allocs = allocs; + } + + struct wlr_vk_allocation *a = &buf->allocs[buf->allocs_size - 1]; + a->start = start; + a->size = size; + return (struct wlr_vk_buffer_span) { + .buffer = buf, + .alloc = *a, + }; + } + + // we didn't find a free buffer - create one + // size = clamp(max(size * 2, prev_size * 2), min_size, max_size) + VkDeviceSize bsize = size * 2; + bsize = bsize < min_stage_size ? min_stage_size : bsize; + if (!wl_list_empty(&r->stage.buffers)) { + struct wl_list *last_link = r->stage.buffers.prev; + struct wlr_vk_shared_buffer *prev = wl_container_of( + last_link, prev, link); + VkDeviceSize last_size = 2 * prev->buf_size; + bsize = bsize < last_size ? last_size : bsize; + } + + if (bsize > max_stage_size) { + wlr_log(WLR_INFO, "vulkan stage buffers have reached max size"); + bsize = max_stage_size; + } + + // create buffer + buf = calloc(1, sizeof(*buf)); + if (!buf) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + goto error_alloc; + } + + VkResult res; + VkBufferCreateInfo buf_info = {0}; + buf_info.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO; + buf_info.size = bsize; + buf_info.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | + VK_BUFFER_USAGE_TRANSFER_SRC_BIT; + buf_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE; + res = vkCreateBuffer(r->dev->dev, &buf_info, NULL, &buf->buffer); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateBuffer", res); + goto error; + } + + VkMemoryRequirements mem_reqs; + vkGetBufferMemoryRequirements(r->dev->dev, buf->buffer, &mem_reqs); + + VkMemoryAllocateInfo mem_info = {0}; + mem_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + mem_info.allocationSize = mem_reqs.size; + mem_info.memoryTypeIndex = vulkan_find_mem_type(r->dev, + VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | + VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, mem_reqs.memoryTypeBits); + res = vkAllocateMemory(r->dev->dev, &mem_info, NULL, &buf->memory); + if (res != VK_SUCCESS) { + wlr_vk_error("vkAllocatorMemory", res); + goto error; + } + + res = vkBindBufferMemory(r->dev->dev, buf->buffer, buf->memory, 0); + if (res != VK_SUCCESS) { + wlr_vk_error("vkBindBufferMemory", res); + goto error; + } + + size_t start_count = 8u; + buf->allocs = calloc(start_count, sizeof(*buf->allocs)); + if (!buf->allocs) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + goto error; + } + + wlr_log(WLR_DEBUG, "Created new vk staging buffer of size %" PRIu64, bsize); + buf->buf_size = bsize; + wl_list_insert(&r->stage.buffers, &buf->link); + + buf->allocs_capacity = start_count; + buf->allocs_size = 1u; + buf->allocs[0].start = 0u; + buf->allocs[0].size = size; + return (struct wlr_vk_buffer_span) { + .buffer = buf, + .alloc = buf->allocs[0], + }; + +error: + shared_buffer_destroy(r, buf); + +error_alloc: + return (struct wlr_vk_buffer_span) { + .buffer = NULL, + .alloc = (struct wlr_vk_allocation) {0, 0}, + }; +} + +VkCommandBuffer vulkan_record_stage_cb(struct wlr_vk_renderer *renderer) { + if (!renderer->stage.recording) { + VkCommandBufferBeginInfo begin_info = {0}; + begin_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; + vkBeginCommandBuffer(renderer->stage.cb, &begin_info); + renderer->stage.recording = true; + } + + return renderer->stage.cb; +} + +bool vulkan_submit_stage_wait(struct wlr_vk_renderer *renderer) { + if (!renderer->stage.recording) { + return false; + } + + vkEndCommandBuffer(renderer->stage.cb); + renderer->stage.recording = false; + + VkSubmitInfo submit_info = {0}; + submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO; + submit_info.commandBufferCount = 1u; + submit_info.pCommandBuffers = &renderer->stage.cb; + VkResult res = vkQueueSubmit(renderer->dev->queue, 1, + &submit_info, renderer->fence); + if (res != VK_SUCCESS) { + wlr_vk_error("vkQueueSubmit", res); + return false; + } + + res = vkWaitForFences(renderer->dev->dev, 1, &renderer->fence, true, + UINT64_MAX); + if (res != VK_SUCCESS) { + wlr_vk_error("vkWaitForFences", res); + return false; + } + + // NOTE: don't release stage allocations here since they may still be + // used for reading. Will be done next frame. + res = vkResetFences(renderer->dev->dev, 1, &renderer->fence); + if (res != VK_SUCCESS) { + wlr_vk_error("vkResetFences", res); + return false; + } + + return true; +} + +struct wlr_vk_format_props *vulkan_format_props_from_drm( + struct wlr_vk_device *dev, uint32_t drm_fmt) { + for (size_t i = 0u; i < dev->format_prop_count; ++i) { + if (dev->format_props[i].format.drm_format == drm_fmt) { + return &dev->format_props[i]; + } + } + return NULL; +} + +// buffer import +static void destroy_render_buffer(struct wlr_vk_render_buffer *buffer) { + wl_list_remove(&buffer->link); + wl_list_remove(&buffer->buffer_destroy.link); + + assert(buffer->renderer->current_render_buffer != buffer); + + VkDevice dev = buffer->renderer->dev->dev; + + vkDestroyFramebuffer(dev, buffer->framebuffer, NULL); + vkDestroyImageView(dev, buffer->image_view, NULL); + vkDestroyImage(dev, buffer->image, NULL); + + for (size_t i = 0u; i < buffer->mem_count; ++i) { + vkFreeMemory(dev, buffer->memories[i], NULL); + } + + free(buffer); +} + +static struct wlr_vk_render_buffer *get_render_buffer( + struct wlr_vk_renderer *renderer, struct wlr_buffer *wlr_buffer) { + struct wlr_vk_render_buffer *buffer; + wl_list_for_each(buffer, &renderer->render_buffers, link) { + if (buffer->wlr_buffer == wlr_buffer) { + return buffer; + } + } + return NULL; +} + +static void handle_render_buffer_destroy(struct wl_listener *listener, void *data) { + struct wlr_vk_render_buffer *buffer = + wl_container_of(listener, buffer, buffer_destroy); + destroy_render_buffer(buffer); +} + +static struct wlr_vk_render_buffer *create_render_buffer( + struct wlr_vk_renderer *renderer, struct wlr_buffer *wlr_buffer) { + VkResult res; + + struct wlr_vk_render_buffer *buffer = calloc(1, sizeof(*buffer)); + if (buffer == NULL) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + return NULL; + } + buffer->wlr_buffer = wlr_buffer; + buffer->renderer = renderer; + + struct wlr_dmabuf_attributes dmabuf = {0}; + if (!wlr_buffer_get_dmabuf(wlr_buffer, &dmabuf)) { + goto error_buffer; + } + + wlr_log(WLR_DEBUG, "vulkan create_render_buffer: %.4s, %dx%d", + (const char*) &dmabuf.format, dmabuf.width, dmabuf.height); + + // NOTE: we could at least support WLR_DMABUF_ATTRIBUTES_FLAGS_Y_INVERT + // if it is needed by anyone. Can be implemented using negative viewport + // height or flipping matrix. + if (dmabuf.flags != 0) { + wlr_log(WLR_ERROR, "dmabuf flags %x not supported/implemented on vulkan", + dmabuf.flags); + goto error_buffer; + } + + buffer->image = vulkan_import_dmabuf(renderer, &dmabuf, + buffer->memories, &buffer->mem_count, true); + if (!buffer->image) { + goto error_buffer; + } + + VkDevice dev = renderer->dev->dev; + const struct wlr_vk_format_props *fmt = vulkan_format_props_from_drm( + renderer->dev, dmabuf.format); + if (fmt == NULL) { + wlr_log(WLR_ERROR, "Unsupported pixel format %"PRIx32 " (%.4s)", + dmabuf.format, (const char*) &dmabuf.format); + goto error_buffer; + } + + VkImageViewCreateInfo view_info = {0}; + view_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + view_info.image = buffer->image; + view_info.viewType = VK_IMAGE_VIEW_TYPE_2D; + view_info.format = fmt->format.vk_format; + view_info.components.r = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.g = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.b = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.a = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.subresourceRange = (VkImageSubresourceRange) { + VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 + }; + + res = vkCreateImageView(dev, &view_info, NULL, &buffer->image_view); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateImageView failed", res); + goto error_view; + } + + buffer->render_setup = find_or_create_render_setup( + renderer, fmt->format.vk_format); + if (!buffer->render_setup) { + goto error_view; + } + + VkFramebufferCreateInfo fb_info = {0}; + fb_info.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO; + fb_info.attachmentCount = 1u; + fb_info.pAttachments = &buffer->image_view; + fb_info.flags = 0u; + fb_info.width = dmabuf.width; + fb_info.height = dmabuf.height; + fb_info.layers = 1u; + fb_info.renderPass = buffer->render_setup->render_pass; + + res = vkCreateFramebuffer(dev, &fb_info, NULL, &buffer->framebuffer); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateFramebuffer", res); + goto error_view; + } + + buffer->buffer_destroy.notify = handle_render_buffer_destroy; + wl_signal_add(&wlr_buffer->events.destroy, &buffer->buffer_destroy); + wl_list_insert(&renderer->render_buffers, &buffer->link); + + return buffer; + +error_view: + vkDestroyFramebuffer(dev, buffer->framebuffer, NULL); + vkDestroyImageView(dev, buffer->image_view, NULL); + vkDestroyImage(dev, buffer->image, NULL); + for (size_t i = 0u; i < buffer->mem_count; ++i) { + vkFreeMemory(dev, buffer->memories[i], NULL); + } +error_buffer: + wlr_dmabuf_attributes_finish(&dmabuf); + free(buffer); + return NULL; +} + +// interface implementation +static bool vulkan_bind_buffer(struct wlr_renderer *wlr_renderer, + struct wlr_buffer *wlr_buffer) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + + if (renderer->current_render_buffer) { + wlr_buffer_unlock(renderer->current_render_buffer->wlr_buffer); + renderer->current_render_buffer = NULL; + } + + if (!wlr_buffer) { + return true; + } + + struct wlr_vk_render_buffer *buffer = get_render_buffer(renderer, wlr_buffer); + if (!buffer) { + buffer = create_render_buffer(renderer, wlr_buffer); + if (!buffer) { + return false; + } + } + + wlr_buffer_lock(wlr_buffer); + renderer->current_render_buffer = buffer; + return true; +} + +static void vulkan_begin(struct wlr_renderer *wlr_renderer, + uint32_t width, uint32_t height) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + assert(renderer->current_render_buffer); + + VkCommandBuffer cb = renderer->cb; + VkCommandBufferBeginInfo begin_info = {0}; + begin_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; + vkBeginCommandBuffer(cb, &begin_info); + + // begin render pass + VkFramebuffer fb = renderer->current_render_buffer->framebuffer; + + VkRect2D rect = {{0, 0}, {width, height}}; + renderer->scissor = rect; + + VkRenderPassBeginInfo rp_info = {0}; + rp_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO; + rp_info.renderArea = rect; + rp_info.renderPass = renderer->current_render_buffer->render_setup->render_pass; + rp_info.framebuffer = fb; + rp_info.clearValueCount = 0; + vkCmdBeginRenderPass(cb, &rp_info, VK_SUBPASS_CONTENTS_INLINE); + + VkViewport vp = {0.f, 0.f, (float) width, (float) height, 0.f, 1.f}; + vkCmdSetViewport(cb, 0, 1, &vp); + vkCmdSetScissor(cb, 0, 1, &rect); + + // Refresh projection matrix. + // wlr_matrix_projection assumes a GL corrdinate system so we need + // to pass WL_OUTPUT_TRANSFORM_FLIPPED_180 to adjust it for vulkan. + wlr_matrix_projection(renderer->projection, width, height, + WL_OUTPUT_TRANSFORM_FLIPPED_180); + + renderer->render_width = width; + renderer->render_height = height; + renderer->bound_pipe = VK_NULL_HANDLE; +} + +static void vulkan_end(struct wlr_renderer *wlr_renderer) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + assert(renderer->current_render_buffer); + + VkCommandBuffer render_cb = renderer->cb; + VkCommandBuffer pre_cb = vulkan_record_stage_cb(renderer); + + renderer->render_width = 0u; + renderer->render_height = 0u; + renderer->bound_pipe = VK_NULL_HANDLE; + + vkCmdEndRenderPass(render_cb); + + // insert acquire and release barriers for dmabuf-images + unsigned barrier_count = wl_list_length(&renderer->foreign_textures) + 1; + VkImageMemoryBarrier* acquire_barriers = calloc(barrier_count, sizeof(VkImageMemoryBarrier)); + VkImageMemoryBarrier* release_barriers = calloc(barrier_count, sizeof(VkImageMemoryBarrier)); + + struct wlr_vk_texture *texture, *tmp_tex; + unsigned idx = 0; + + wl_list_for_each_safe(texture, tmp_tex, &renderer->foreign_textures, foreign_link) { + VkImageLayout src_layout = VK_IMAGE_LAYOUT_GENERAL; + if (!texture->transitioned) { + src_layout = VK_IMAGE_LAYOUT_PREINITIALIZED; + texture->transitioned = true; + } + + // acquire + acquire_barriers[idx].sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; + acquire_barriers[idx].srcQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT; + acquire_barriers[idx].dstQueueFamilyIndex = renderer->dev->queue_family; + acquire_barriers[idx].image = texture->image; + acquire_barriers[idx].oldLayout = src_layout; + acquire_barriers[idx].newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + acquire_barriers[idx].srcAccessMask = 0u; // ignored anyways + acquire_barriers[idx].dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + acquire_barriers[idx].subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + acquire_barriers[idx].subresourceRange.layerCount = 1; + acquire_barriers[idx].subresourceRange.levelCount = 1; + + // releaes + release_barriers[idx].sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; + release_barriers[idx].srcQueueFamilyIndex = renderer->dev->queue_family; + release_barriers[idx].dstQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT; + release_barriers[idx].image = texture->image; + release_barriers[idx].oldLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + release_barriers[idx].newLayout = VK_IMAGE_LAYOUT_GENERAL; + release_barriers[idx].srcAccessMask = VK_ACCESS_SHADER_READ_BIT; + release_barriers[idx].dstAccessMask = 0u; // ignored anyways + release_barriers[idx].subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + release_barriers[idx].subresourceRange.layerCount = 1; + release_barriers[idx].subresourceRange.levelCount = 1; + ++idx; + + wl_list_remove(&texture->foreign_link); + texture->owned = false; + } + + // also add acquire/release barriers for the current render buffer + VkImageLayout src_layout = VK_IMAGE_LAYOUT_GENERAL; + if (!renderer->current_render_buffer->transitioned) { + src_layout = VK_IMAGE_LAYOUT_PREINITIALIZED; + renderer->current_render_buffer->transitioned = true; + } + + // acquire render buffer before rendering + acquire_barriers[idx].sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; + acquire_barriers[idx].srcQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT; + acquire_barriers[idx].dstQueueFamilyIndex = renderer->dev->queue_family; + acquire_barriers[idx].image = renderer->current_render_buffer->image; + acquire_barriers[idx].oldLayout = src_layout; + acquire_barriers[idx].newLayout = VK_IMAGE_LAYOUT_GENERAL; + acquire_barriers[idx].srcAccessMask = 0u; // ignored anyways + acquire_barriers[idx].dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + acquire_barriers[idx].subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + acquire_barriers[idx].subresourceRange.layerCount = 1; + acquire_barriers[idx].subresourceRange.levelCount = 1; + + // release render buffer after rendering + release_barriers[idx].sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; + release_barriers[idx].srcQueueFamilyIndex = renderer->dev->queue_family; + release_barriers[idx].dstQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT; + release_barriers[idx].image = renderer->current_render_buffer->image; + release_barriers[idx].oldLayout = VK_IMAGE_LAYOUT_GENERAL; + release_barriers[idx].newLayout = VK_IMAGE_LAYOUT_GENERAL; + release_barriers[idx].srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + release_barriers[idx].dstAccessMask = 0u; // ignored anyways + release_barriers[idx].subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + release_barriers[idx].subresourceRange.layerCount = 1; + release_barriers[idx].subresourceRange.levelCount = 1; + ++idx; + + vkCmdPipelineBarrier(pre_cb, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, + 0, 0, NULL, 0, NULL, barrier_count, acquire_barriers); + + vkCmdPipelineBarrier(render_cb, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, + VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, 0, NULL, 0, NULL, + barrier_count, release_barriers); + + free(acquire_barriers); + free(release_barriers); + + vkEndCommandBuffer(renderer->cb); + + unsigned submit_count = 0u; + VkSubmitInfo submit_infos[2] = {0}; + + // No semaphores needed here. + // We don't need a semaphore from the stage/transfer submission + // to the render submissions since they are on the same queue + // and we have a renderpass dependency for that. + if (renderer->stage.recording) { + vkEndCommandBuffer(renderer->stage.cb); + renderer->stage.recording = false; + + VkSubmitInfo *stage_sub = &submit_infos[submit_count]; + stage_sub->sType = VK_STRUCTURE_TYPE_SUBMIT_INFO; + stage_sub->commandBufferCount = 1u; + stage_sub->pCommandBuffers = &pre_cb; + ++submit_count; + } + + VkSubmitInfo *render_sub = &submit_infos[submit_count]; + render_sub->sType = VK_STRUCTURE_TYPE_SUBMIT_INFO; + render_sub->pCommandBuffers = &render_cb; + render_sub->commandBufferCount = 1u; + ++submit_count; + + VkResult res = vkQueueSubmit(renderer->dev->queue, submit_count, + submit_infos, renderer->fence); + if (res != VK_SUCCESS) { + wlr_vk_error("vkQueueSubmit", res); + return; + } + + // sadly this is required due to the current api/rendering model of wlr + // ideally we could use gpu and cpu in parallel (_without_ the + // implicit synchronization overhead and mess of opengl drivers) + res = vkWaitForFences(renderer->dev->dev, 1, &renderer->fence, true, + UINT64_MAX); + if (res != VK_SUCCESS) { + wlr_vk_error("vkWaitForFences", res); + return; + } + + ++renderer->frame; + release_stage_allocations(renderer); + + // destroy pending textures + wl_list_for_each_safe(texture, tmp_tex, &renderer->destroy_textures, destroy_link) { + wlr_texture_destroy(&texture->wlr_texture); + } + + wl_list_init(&renderer->destroy_textures); // reset the list + res = vkResetFences(renderer->dev->dev, 1, &renderer->fence); + if (res != VK_SUCCESS) { + wlr_vk_error("vkResetFences", res); + return; + } +} + +static bool vulkan_render_subtexture_with_matrix(struct wlr_renderer *wlr_renderer, + struct wlr_texture *wlr_texture, const struct wlr_fbox *box, + const float matrix[static 9], float alpha) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + VkCommandBuffer cb = renderer->cb; + + struct wlr_vk_texture *texture = vulkan_get_texture(wlr_texture); + assert(texture->renderer == renderer); + if (texture->dmabuf_imported && !texture->owned) { + // Store this texture in the list of textures that need to be + // acquired before rendering and released after rendering. + // We don't do it here immediately since barriers inside + // a renderpass are suboptimal (would require additional renderpass + // dependency and potentially multiple barriers) and it's + // better to issue one barrier for all used textures anyways. + texture->owned = true; + assert(texture->foreign_link.prev == NULL); + assert(texture->foreign_link.next == NULL); + wl_list_insert(&renderer->foreign_textures, &texture->foreign_link); + } + + VkPipeline pipe = renderer->current_render_buffer->render_setup->tex_pipe; + if (pipe != renderer->bound_pipe) { + vkCmdBindPipeline(cb, VK_PIPELINE_BIND_POINT_GRAPHICS, pipe); + renderer->bound_pipe = pipe; + } + + vkCmdBindDescriptorSets(cb, VK_PIPELINE_BIND_POINT_GRAPHICS, + renderer->pipe_layout, 0, 1, &texture->ds, 0, NULL); + + float final_matrix[9]; + wlr_matrix_multiply(final_matrix, renderer->projection, matrix); + + struct vert_pcr_data vert_pcr_data; + mat3_to_mat4(final_matrix, vert_pcr_data.mat4); + + vert_pcr_data.uv_off[0] = box->x / wlr_texture->width; + vert_pcr_data.uv_off[1] = box->y / wlr_texture->height; + vert_pcr_data.uv_size[0] = box->width / wlr_texture->width; + vert_pcr_data.uv_size[1] = box->height / wlr_texture->height; + + if (texture->invert_y) { + vert_pcr_data.uv_off[1] += vert_pcr_data.uv_size[1]; + vert_pcr_data.uv_size[1] = -vert_pcr_data.uv_size[1]; + } + + // When the texture itself does not have alpha information we want + // to ignore the sampled value and just use the alpha passed here, + // we pass a negative value to communicate that. + // See the texture.frag shader for more details. + const struct wlr_pixel_format_info *format_info = drm_get_pixel_format_info( + texture->format->drm_format); + assert(format_info); + if (!format_info->has_alpha) { + alpha *= -1; + } + + vkCmdPushConstants(cb, renderer->pipe_layout, + VK_SHADER_STAGE_VERTEX_BIT, 0, sizeof(vert_pcr_data), &vert_pcr_data); + vkCmdPushConstants(cb, renderer->pipe_layout, + VK_SHADER_STAGE_FRAGMENT_BIT, sizeof(vert_pcr_data), sizeof(float), + &alpha); + vkCmdDraw(cb, 4, 1, 0, 0); + texture->last_used = renderer->frame; + + return true; +} + +static void vulkan_clear(struct wlr_renderer *wlr_renderer, + const float color[static 4]) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + VkCommandBuffer cb = renderer->cb; + + VkClearAttachment att = {0}; + att.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + att.colorAttachment = 0u; + + // Input color values are given in srgb space, vulkan expects + // them in linear space. We explicitly import argb8 render buffers + // as srgb, vulkan will convert the input values we give here to + // srgb first. + // But in other parts of wlroots we just always assume + // srgb so that's why we have to convert here. + att.clearValue.color.float32[0] = color_to_linear(color[0]); + att.clearValue.color.float32[1] = color_to_linear(color[1]); + att.clearValue.color.float32[2] = color_to_linear(color[2]); + att.clearValue.color.float32[3] = color[3]; // no conversion for alpha + + VkClearRect rect = {0}; + rect.rect = renderer->scissor; + rect.layerCount = 1; + vkCmdClearAttachments(cb, 1, &att, 1, &rect); +} + +static void vulkan_scissor(struct wlr_renderer *wlr_renderer, + struct wlr_box *box) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + VkCommandBuffer cb = renderer->cb; + + uint32_t w = renderer->render_width; + uint32_t h = renderer->render_height; + struct wlr_box dst = {0, 0, w, h}; + if (box && !wlr_box_intersection(&dst, box, &dst)) { + dst = (struct wlr_box) {0, 0, 0, 0}; // empty + } + + VkRect2D rect = (VkRect2D) {{dst.x, dst.y}, {dst.width, dst.height}}; + renderer->scissor = rect; + vkCmdSetScissor(cb, 0, 1, &rect); +} + +static const uint32_t *vulkan_get_shm_texture_formats( + struct wlr_renderer *wlr_renderer, size_t *len) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + *len = renderer->dev->shm_format_count; + return renderer->dev->shm_formats; +} + +static void vulkan_render_quad_with_matrix(struct wlr_renderer *wlr_renderer, + const float color[static 4], const float matrix[static 9]) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + VkCommandBuffer cb = renderer->cb; + + VkPipeline pipe = renderer->current_render_buffer->render_setup->quad_pipe; + if (pipe != renderer->bound_pipe) { + vkCmdBindPipeline(cb, VK_PIPELINE_BIND_POINT_GRAPHICS, pipe); + renderer->bound_pipe = pipe; + } + + float final_matrix[9]; + wlr_matrix_multiply(final_matrix, renderer->projection, matrix); + + struct vert_pcr_data vert_pcr_data; + mat3_to_mat4(final_matrix, vert_pcr_data.mat4); + vert_pcr_data.uv_off[0] = 0.f; + vert_pcr_data.uv_off[1] = 0.f; + vert_pcr_data.uv_size[0] = 1.f; + vert_pcr_data.uv_size[1] = 1.f; + + // Input color values are given in srgb space, shader expects + // them in linear space. The shader does all computation in linear + // space and expects in inputs in linear space since it outputs + // colors in linear space as well (and vulkan then automatically + // does the conversion for out SRGB render targets). + // But in other parts of wlroots we just always assume + // srgb so that's why we have to convert here. + float linear_color[4]; + linear_color[0] = color_to_linear(color[0]); + linear_color[1] = color_to_linear(color[1]); + linear_color[2] = color_to_linear(color[2]); + linear_color[3] = color[3]; // no conversion for alpha + + vkCmdPushConstants(cb, renderer->pipe_layout, + VK_SHADER_STAGE_VERTEX_BIT, 0, sizeof(vert_pcr_data), &vert_pcr_data); + vkCmdPushConstants(cb, renderer->pipe_layout, + VK_SHADER_STAGE_FRAGMENT_BIT, sizeof(vert_pcr_data), sizeof(float) * 4, + linear_color); + vkCmdDraw(cb, 4, 1, 0, 0); +} + +static const struct wlr_drm_format_set *vulkan_get_dmabuf_texture_formats( + struct wlr_renderer *wlr_renderer) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + return &renderer->dev->dmabuf_texture_formats; +} + +static const struct wlr_drm_format_set *vulkan_get_render_formats( + struct wlr_renderer *wlr_renderer) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + return &renderer->dev->dmabuf_render_formats; +} + +static uint32_t vulkan_preferred_read_format( + struct wlr_renderer *wlr_renderer) { + // TODO: implement! + wlr_log(WLR_ERROR, "vulkan_preferred_read_format not implemented"); + return DRM_FORMAT_XBGR8888; +} + +static void vulkan_destroy(struct wlr_renderer *wlr_renderer) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + struct wlr_vk_device *dev = renderer->dev; + if (!dev) { + free(renderer); + return; + } + + assert(!renderer->current_render_buffer); + + // stage.cb automatically freed with command pool + struct wlr_vk_shared_buffer *buf, *tmp_buf; + wl_list_for_each_safe(buf, tmp_buf, &renderer->stage.buffers, link) { + shared_buffer_destroy(renderer, buf); + } + + struct wlr_vk_texture *tex, *tex_tmp; + wl_list_for_each_safe(tex, tex_tmp, &renderer->textures, link) { + vulkan_texture_destroy(tex); + } + + struct wlr_vk_render_buffer *render_buffer, *render_buffer_tmp; + wl_list_for_each_safe(render_buffer, render_buffer_tmp, + &renderer->render_buffers, link) { + destroy_render_buffer(render_buffer); + } + + struct wlr_vk_render_format_setup *setup, *tmp_setup; + wl_list_for_each_safe(setup, tmp_setup, + &renderer->render_format_setups, link) { + destroy_render_format_setup(renderer, setup); + } + + struct wlr_vk_descriptor_pool *pool, *tmp_pool; + wl_list_for_each_safe(pool, tmp_pool, &renderer->descriptor_pools, link) { + vkDestroyDescriptorPool(dev->dev, pool->pool, NULL); + free(pool); + } + + vkDestroyShaderModule(dev->dev, renderer->vert_module, NULL); + vkDestroyShaderModule(dev->dev, renderer->tex_frag_module, NULL); + vkDestroyShaderModule(dev->dev, renderer->quad_frag_module, NULL); + + vkDestroyFence(dev->dev, renderer->fence, NULL); + vkDestroyPipelineLayout(dev->dev, renderer->pipe_layout, NULL); + vkDestroyDescriptorSetLayout(dev->dev, renderer->ds_layout, NULL); + vkDestroySampler(dev->dev, renderer->sampler, NULL); + vkDestroyCommandPool(dev->dev, renderer->command_pool, NULL); + + struct wlr_vk_instance *ini = dev->instance; + vulkan_device_destroy(dev); + vulkan_instance_destroy(ini); + free(renderer); +} + +static bool vulkan_read_pixels(struct wlr_renderer *wlr_renderer, + uint32_t drm_format, uint32_t *flags, uint32_t stride, + uint32_t width, uint32_t height, uint32_t src_x, uint32_t src_y, + uint32_t dst_x, uint32_t dst_y, void *data) { + // TODO: implement! + wlr_log(WLR_ERROR, "vulkan_read_pixels not implemented"); + return false; +} + +static int vulkan_get_drm_fd(struct wlr_renderer *wlr_renderer) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + return renderer->dev->drm_fd; +} + +static uint32_t vulkan_get_render_buffer_caps(struct wlr_renderer *wlr_renderer) { + return WLR_BUFFER_CAP_DMABUF; +} + +static const struct wlr_renderer_impl renderer_impl = { + .bind_buffer = vulkan_bind_buffer, + .begin = vulkan_begin, + .end = vulkan_end, + .clear = vulkan_clear, + .scissor = vulkan_scissor, + .render_subtexture_with_matrix = vulkan_render_subtexture_with_matrix, + .render_quad_with_matrix = vulkan_render_quad_with_matrix, + .get_shm_texture_formats = vulkan_get_shm_texture_formats, + .get_dmabuf_texture_formats = vulkan_get_dmabuf_texture_formats, + .get_render_formats = vulkan_get_render_formats, + .preferred_read_format = vulkan_preferred_read_format, + .read_pixels = vulkan_read_pixels, + .destroy = vulkan_destroy, + .get_drm_fd = vulkan_get_drm_fd, + .get_render_buffer_caps = vulkan_get_render_buffer_caps, + .texture_from_buffer = vulkan_texture_from_buffer, +}; + +// Initializes the VkDescriptorSetLayout and VkPipelineLayout needed +// for the texture rendering pipeline using the given VkSampler. +static bool init_tex_layouts(struct wlr_vk_renderer *renderer, + VkSampler tex_sampler, VkDescriptorSetLayout *out_ds_layout, + VkPipelineLayout *out_pipe_layout) { + VkResult res; + VkDevice dev = renderer->dev->dev; + + // layouts + // descriptor set + VkDescriptorSetLayoutBinding ds_bindings[1] = {{ + 0, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1, + VK_SHADER_STAGE_FRAGMENT_BIT, &tex_sampler, + }}; + + VkDescriptorSetLayoutCreateInfo ds_info = {0}; + ds_info.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO; + ds_info.bindingCount = 1; + ds_info.pBindings = ds_bindings; + + res = vkCreateDescriptorSetLayout(dev, &ds_info, NULL, out_ds_layout); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateDescriptorSetLayout", res); + return false; + } + + // pipeline layout + VkPushConstantRange pc_ranges[2] = {0}; + pc_ranges[0].size = sizeof(struct vert_pcr_data); + pc_ranges[0].stageFlags = VK_SHADER_STAGE_VERTEX_BIT; + + pc_ranges[1].offset = pc_ranges[0].size; + pc_ranges[1].size = sizeof(float) * 4; // alpha or color + pc_ranges[1].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT; + + VkPipelineLayoutCreateInfo pl_info = {0}; + pl_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO; + pl_info.setLayoutCount = 1; + pl_info.pSetLayouts = out_ds_layout; + pl_info.pushConstantRangeCount = 2; + pl_info.pPushConstantRanges = pc_ranges; + + res = vkCreatePipelineLayout(dev, &pl_info, NULL, out_pipe_layout); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreatePipelineLayout", res); + return false; + } + + return true; +} + +// Initializes the pipeline for rendering textures and using the given +// VkRenderPass and VkPipelineLayout. +static bool init_tex_pipeline(struct wlr_vk_renderer *renderer, + VkRenderPass rp, VkPipelineLayout pipe_layout, VkPipeline *pipe) { + VkResult res; + VkDevice dev = renderer->dev->dev; + + // shaders + VkPipelineShaderStageCreateInfo vert_stage = { + VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + NULL, 0, VK_SHADER_STAGE_VERTEX_BIT, renderer->vert_module, + "main", NULL + }; + + VkPipelineShaderStageCreateInfo tex_stages[2] = {vert_stage, { + VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + NULL, 0, VK_SHADER_STAGE_FRAGMENT_BIT, renderer->tex_frag_module, + "main", NULL + }}; + + // info + VkPipelineInputAssemblyStateCreateInfo assembly = {0}; + assembly.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO; + assembly.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN; + + VkPipelineRasterizationStateCreateInfo rasterization = {0}; + rasterization.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO; + rasterization.polygonMode = VK_POLYGON_MODE_FILL; + rasterization.cullMode = VK_CULL_MODE_NONE; + rasterization.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE; + rasterization.lineWidth = 1.f; + + VkPipelineColorBlendAttachmentState blend_attachment = {0}; + blend_attachment.blendEnable = true; + // we generally work with pre-multiplied alpha + blend_attachment.srcColorBlendFactor = VK_BLEND_FACTOR_ONE; + blend_attachment.dstColorBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA; + blend_attachment.colorBlendOp = VK_BLEND_OP_ADD; + blend_attachment.srcAlphaBlendFactor = VK_BLEND_FACTOR_ONE; + blend_attachment.dstAlphaBlendFactor = VK_BLEND_FACTOR_ONE; + blend_attachment.alphaBlendOp = VK_BLEND_OP_ADD; + blend_attachment.colorWriteMask = + VK_COLOR_COMPONENT_R_BIT | + VK_COLOR_COMPONENT_G_BIT | + VK_COLOR_COMPONENT_B_BIT | + VK_COLOR_COMPONENT_A_BIT; + + VkPipelineColorBlendStateCreateInfo blend = {0}; + blend.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO; + blend.attachmentCount = 1; + blend.pAttachments = &blend_attachment; + + VkPipelineMultisampleStateCreateInfo multisample = {0}; + multisample.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO; + multisample.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT; + + VkPipelineViewportStateCreateInfo viewport = {0}; + viewport.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO; + viewport.viewportCount = 1; + viewport.scissorCount = 1; + + VkDynamicState dynStates[2] = { + VK_DYNAMIC_STATE_VIEWPORT, + VK_DYNAMIC_STATE_SCISSOR, + }; + VkPipelineDynamicStateCreateInfo dynamic = {0}; + dynamic.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO; + dynamic.pDynamicStates = dynStates; + dynamic.dynamicStateCount = 2; + + VkPipelineVertexInputStateCreateInfo vertex = {0}; + vertex.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO; + + VkGraphicsPipelineCreateInfo pinfo = {0}; + pinfo.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO; + pinfo.layout = pipe_layout; + pinfo.renderPass = rp; + pinfo.subpass = 0; + pinfo.stageCount = 2; + pinfo.pStages = tex_stages; + + pinfo.pInputAssemblyState = &assembly; + pinfo.pRasterizationState = &rasterization; + pinfo.pColorBlendState = &blend; + pinfo.pMultisampleState = &multisample; + pinfo.pViewportState = &viewport; + pinfo.pDynamicState = &dynamic; + pinfo.pVertexInputState = &vertex; + + // NOTE: use could use a cache here for faster loading + // store it somewhere like $XDG_CACHE_HOME/wlroots/vk_pipe_cache + VkPipelineCache cache = VK_NULL_HANDLE; + res = vkCreateGraphicsPipelines(dev, cache, 1, &pinfo, NULL, pipe); + if (res != VK_SUCCESS) { + wlr_vk_error("failed to create vulkan pipelines:", res); + return false; + } + + return true; +} + +// Creates static render data, such as sampler, layouts and shader modules +// for the given rednerer. +// Cleanup is done by destroying the renderer. +static bool init_static_render_data(struct wlr_vk_renderer *renderer) { + VkResult res; + VkDevice dev = renderer->dev->dev; + + // default sampler (non ycbcr) + VkSamplerCreateInfo sampler_info = {0}; + sampler_info.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO; + sampler_info.magFilter = VK_FILTER_LINEAR; + sampler_info.minFilter = VK_FILTER_LINEAR; + sampler_info.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST; + sampler_info.addressModeU = VK_SAMPLER_ADDRESS_MODE_REPEAT; + sampler_info.addressModeV = VK_SAMPLER_ADDRESS_MODE_REPEAT; + sampler_info.addressModeW = VK_SAMPLER_ADDRESS_MODE_REPEAT; + sampler_info.maxAnisotropy = 1.f; + sampler_info.minLod = 0.f; + sampler_info.maxLod = 0.25f; + sampler_info.borderColor = VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK; + + res = vkCreateSampler(dev, &sampler_info, NULL, &renderer->sampler); + if (res != VK_SUCCESS) { + wlr_vk_error("Failed to create sampler", res); + return false; + } + + if (!init_tex_layouts(renderer, renderer->sampler, + &renderer->ds_layout, &renderer->pipe_layout)) { + return false; + } + + // load vert module and tex frag module since they are needed to + // initialize the tex pipeline + VkShaderModuleCreateInfo sinfo = {0}; + sinfo.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO; + sinfo.codeSize = sizeof(common_vert_data); + sinfo.pCode = common_vert_data; + res = vkCreateShaderModule(dev, &sinfo, NULL, &renderer->vert_module); + if (res != VK_SUCCESS) { + wlr_vk_error("Failed to create vertex shader module", res); + return false; + } + + // tex frag + sinfo.codeSize = sizeof(texture_frag_data); + sinfo.pCode = texture_frag_data; + res = vkCreateShaderModule(dev, &sinfo, NULL, &renderer->tex_frag_module); + if (res != VK_SUCCESS) { + wlr_vk_error("Failed to create tex fragment shader module", res); + return false; + } + + // quad frag + sinfo.codeSize = sizeof(quad_frag_data); + sinfo.pCode = quad_frag_data; + res = vkCreateShaderModule(dev, &sinfo, NULL, &renderer->quad_frag_module); + if (res != VK_SUCCESS) { + wlr_vk_error("Failed to create quad fragment shader module", res); + return false; + } + + return true; +} + +static struct wlr_vk_render_format_setup *find_or_create_render_setup( + struct wlr_vk_renderer *renderer, VkFormat format) { + struct wlr_vk_render_format_setup *setup; + wl_list_for_each(setup, &renderer->render_format_setups, link) { + if (setup->render_format == format) { + return setup; + } + } + + setup = calloc(1u, sizeof(*setup)); + if (!setup) { + wlr_log(WLR_ERROR, "Allocation failed"); + return NULL; + } + + setup->render_format = format; + + // util + VkDevice dev = renderer->dev->dev; + VkResult res; + + VkAttachmentDescription attachment = {0}; + attachment.format = format; + attachment.samples = VK_SAMPLE_COUNT_1_BIT; + attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD; + attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE; + attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE; + attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE; + attachment.initialLayout = VK_IMAGE_LAYOUT_GENERAL; + attachment.finalLayout = VK_IMAGE_LAYOUT_GENERAL; + + VkAttachmentReference color_ref = {0}; + color_ref.attachment = 0u; + color_ref.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL; + + VkSubpassDescription subpass = {0}; + subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS; + subpass.colorAttachmentCount = 1; + subpass.pColorAttachments = &color_ref; + + VkSubpassDependency deps[2] = {0}; + deps[0].srcSubpass = VK_SUBPASS_EXTERNAL; + deps[0].srcStageMask = VK_PIPELINE_STAGE_HOST_BIT | + VK_PIPELINE_STAGE_TRANSFER_BIT | + VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT | + VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT; + deps[0].srcAccessMask = VK_ACCESS_HOST_WRITE_BIT | + VK_ACCESS_TRANSFER_WRITE_BIT | + VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + deps[0].dstSubpass = 0; + deps[0].dstStageMask = VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT; + deps[0].dstAccessMask = VK_ACCESS_UNIFORM_READ_BIT | + VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | + VK_ACCESS_INDIRECT_COMMAND_READ_BIT | + VK_ACCESS_SHADER_READ_BIT; + + deps[1].srcSubpass = 0; + deps[1].srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT; + deps[1].srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + deps[1].dstSubpass = VK_SUBPASS_EXTERNAL; + deps[1].dstStageMask = VK_PIPELINE_STAGE_TRANSFER_BIT | + VK_PIPELINE_STAGE_HOST_BIT | VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT; + deps[1].dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | + VK_ACCESS_MEMORY_READ_BIT; + + VkRenderPassCreateInfo rp_info = {0}; + rp_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO; + rp_info.attachmentCount = 1; + rp_info.pAttachments = &attachment; + rp_info.subpassCount = 1; + rp_info.pSubpasses = &subpass; + rp_info.dependencyCount = 2u; + rp_info.pDependencies = deps; + + res = vkCreateRenderPass(dev, &rp_info, NULL, &setup->render_pass); + if (res != VK_SUCCESS) { + wlr_vk_error("Failed to create render pass", res); + free(setup); + return NULL; + } + + if (!init_tex_pipeline(renderer, setup->render_pass, renderer->pipe_layout, + &setup->tex_pipe)) { + goto error; + } + + VkPipelineShaderStageCreateInfo vert_stage = { + VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + NULL, 0, VK_SHADER_STAGE_VERTEX_BIT, renderer->vert_module, + "main", NULL + }; + + VkPipelineShaderStageCreateInfo quad_stages[2] = {vert_stage, { + VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO, + NULL, 0, VK_SHADER_STAGE_FRAGMENT_BIT, + renderer->quad_frag_module, "main", NULL + }}; + + // info + VkPipelineInputAssemblyStateCreateInfo assembly = {0}; + assembly.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO; + assembly.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN; + + VkPipelineRasterizationStateCreateInfo rasterization = {0}; + rasterization.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO; + rasterization.polygonMode = VK_POLYGON_MODE_FILL; + rasterization.cullMode = VK_CULL_MODE_NONE; + rasterization.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE; + rasterization.lineWidth = 1.f; + + VkPipelineColorBlendAttachmentState blend_attachment = {0}; + blend_attachment.blendEnable = true; + blend_attachment.srcColorBlendFactor = VK_BLEND_FACTOR_ONE; + blend_attachment.dstColorBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA; + blend_attachment.colorBlendOp = VK_BLEND_OP_ADD; + blend_attachment.srcAlphaBlendFactor = VK_BLEND_FACTOR_ONE; + blend_attachment.dstAlphaBlendFactor = VK_BLEND_FACTOR_ONE; + blend_attachment.alphaBlendOp = VK_BLEND_OP_ADD; + blend_attachment.colorWriteMask = + VK_COLOR_COMPONENT_R_BIT | + VK_COLOR_COMPONENT_G_BIT | + VK_COLOR_COMPONENT_B_BIT | + VK_COLOR_COMPONENT_A_BIT; + + VkPipelineColorBlendStateCreateInfo blend = {0}; + blend.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO; + blend.attachmentCount = 1; + blend.pAttachments = &blend_attachment; + + VkPipelineMultisampleStateCreateInfo multisample = {0}; + multisample.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO; + multisample.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT; + + VkPipelineViewportStateCreateInfo viewport = {0}; + viewport.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO; + viewport.viewportCount = 1; + viewport.scissorCount = 1; + + VkDynamicState dynStates[2] = { + VK_DYNAMIC_STATE_VIEWPORT, + VK_DYNAMIC_STATE_SCISSOR, + }; + VkPipelineDynamicStateCreateInfo dynamic = {0}; + dynamic.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO; + dynamic.pDynamicStates = dynStates; + dynamic.dynamicStateCount = 2; + + VkPipelineVertexInputStateCreateInfo vertex = {0}; + vertex.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO; + + VkGraphicsPipelineCreateInfo pinfo = {0}; + pinfo.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO; + pinfo.layout = renderer->pipe_layout; + pinfo.renderPass = setup->render_pass; + pinfo.subpass = 0; + pinfo.stageCount = 2; + pinfo.pStages = quad_stages; + + pinfo.pInputAssemblyState = &assembly; + pinfo.pRasterizationState = &rasterization; + pinfo.pColorBlendState = &blend; + pinfo.pMultisampleState = &multisample; + pinfo.pViewportState = &viewport; + pinfo.pDynamicState = &dynamic; + pinfo.pVertexInputState = &vertex; + + // NOTE: use could use a cache here for faster loading + // store it somewhere like $XDG_CACHE_HOME/wlroots/vk_pipe_cache.bin + VkPipelineCache cache = VK_NULL_HANDLE; + res = vkCreateGraphicsPipelines(dev, cache, 1, &pinfo, NULL, &setup->quad_pipe); + if (res != VK_SUCCESS) { + wlr_log(WLR_ERROR, "failed to create vulkan quad pipeline: %d", res); + goto error; + } + + wl_list_insert(&renderer->render_format_setups, &setup->link); + return setup; + +error: + destroy_render_format_setup(renderer, setup); + return NULL; +} + +struct wlr_renderer *vulkan_renderer_create_for_device(struct wlr_vk_device *dev) { + struct wlr_vk_renderer *renderer; + VkResult res; + if (!(renderer = calloc(1, sizeof(*renderer)))) { + wlr_log_errno(WLR_ERROR, "failed to allocate wlr_vk_renderer"); + return NULL; + } + + renderer->dev = dev; + wlr_renderer_init(&renderer->wlr_renderer, &renderer_impl); + wl_list_init(&renderer->stage.buffers); + wl_list_init(&renderer->destroy_textures); + wl_list_init(&renderer->foreign_textures); + wl_list_init(&renderer->textures); + wl_list_init(&renderer->descriptor_pools); + wl_list_init(&renderer->render_format_setups); + wl_list_init(&renderer->render_buffers); + + if (!init_static_render_data(renderer)) { + goto error; + } + + // command pool + VkCommandPoolCreateInfo cpool_info = {0}; + cpool_info.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO; + cpool_info.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT; + cpool_info.queueFamilyIndex = dev->queue_family; + res = vkCreateCommandPool(dev->dev, &cpool_info, NULL, + &renderer->command_pool); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateCommandPool", res); + goto error; + } + + VkCommandBufferAllocateInfo cbai = {0}; + cbai.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; + cbai.commandBufferCount = 1u; + cbai.commandPool = renderer->command_pool; + cbai.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY; + res = vkAllocateCommandBuffers(dev->dev, &cbai, &renderer->cb); + if (res != VK_SUCCESS) { + wlr_vk_error("vkAllocateCommandBuffers", res); + goto error; + } + + VkFenceCreateInfo fence_info = {0}; + fence_info.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; + res = vkCreateFence(dev->dev, &fence_info, NULL, + &renderer->fence); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateFence", res); + goto error; + } + + // staging command buffer + VkCommandBufferAllocateInfo cmd_buf_info = {0}; + cmd_buf_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; + cmd_buf_info.commandPool = renderer->command_pool; + cmd_buf_info.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY; + cmd_buf_info.commandBufferCount = 1u; + res = vkAllocateCommandBuffers(dev->dev, &cmd_buf_info, + &renderer->stage.cb); + if (res != VK_SUCCESS) { + wlr_vk_error("vkAllocateCommandBuffers", res); + goto error; + } + + return &renderer->wlr_renderer; + +error: + vulkan_destroy(&renderer->wlr_renderer); + return NULL; +} + +struct wlr_renderer *wlr_vk_renderer_create_with_drm_fd(int drm_fd) { + wlr_log(WLR_INFO, "The vulkan renderer is only experimental and " + "not expected to be ready for daily use"); + + // NOTE: we could add functionality to allow the compositor passing its + // name and version to this function. Just use dummies until then, + // shouldn't be relevant to the driver anyways + struct wlr_vk_instance *ini = vulkan_instance_create(0, NULL, default_debug); + if (!ini) { + wlr_log(WLR_ERROR, "creating vulkan instance for renderer failed"); + return NULL; + } + + VkPhysicalDevice phdev = vulkan_find_drm_phdev(ini, drm_fd); + if (!phdev) { + // We rather fail here than doing some guesswork + wlr_log(WLR_ERROR, "Could not match drm and vulkan device"); + return NULL; + } + + // queue families + uint32_t qfam_count; + vkGetPhysicalDeviceQueueFamilyProperties(phdev, &qfam_count, NULL); + VkQueueFamilyProperties queue_props[qfam_count]; + vkGetPhysicalDeviceQueueFamilyProperties(phdev, &qfam_count, + queue_props); + + struct wlr_vk_device *dev = vulkan_device_create(ini, phdev, 0, NULL); + if (!dev) { + wlr_log(WLR_ERROR, "Failed to create vulkan device"); + vulkan_instance_destroy(ini); + return NULL; + } + + // We duplicate it so it's not closed while we still need it. + dev->drm_fd = fcntl(drm_fd, F_DUPFD_CLOEXEC, 0); + if (dev->drm_fd < 0) { + wlr_log_errno(WLR_ERROR, "fcntl(F_DUPFD_CLOEXEC) failed"); + vulkan_device_destroy(dev); + vulkan_instance_destroy(ini); + return NULL; + } + + return vulkan_renderer_create_for_device(dev); +} diff --git a/render/vulkan/shaders/common.vert b/render/vulkan/shaders/common.vert new file mode 100644 index 00000000..fa31d26c --- /dev/null +++ b/render/vulkan/shaders/common.vert @@ -0,0 +1,25 @@ +#version 450 + +// we use a mat4 since it uses the same size as mat3 due to +// alignment. Easier to deal with (tighly-packed) mat4 though. +layout(push_constant, row_major) uniform UBO { + mat4 proj; + vec2 uv_offset; + vec2 uv_size; +} data; + +layout(location = 0) out vec2 uv; + +// 4 outlining points and uv coords +const vec2[] values = { + {0, 0}, + {1, 0}, + {1, 1}, + {0, 1}, +}; + +void main() { + vec2 pos = values[gl_VertexIndex % 4]; + uv = data.uv_offset + pos * data.uv_size; + gl_Position = data.proj * vec4(pos, 0.0, 1.0); +} diff --git a/render/vulkan/shaders/meson.build b/render/vulkan/shaders/meson.build new file mode 100644 index 00000000..b183c46c --- /dev/null +++ b/render/vulkan/shaders/meson.build @@ -0,0 +1,20 @@ +vulkan_shaders_src = [ + 'common.vert', + 'texture.frag', + 'quad.frag', +] + +vulkan_shaders = [] +foreach shader : vulkan_shaders_src + name = shader.underscorify() + '_data' + args = [glslang, '-V', '@INPUT@', '-o', '@OUTPUT@', '--vn', name] + header = custom_target( + shader + '_spv', + output: shader + '.h', + input: shader, + command: args) + + vulkan_shaders += [header] +endforeach + +wlr_files += vulkan_shaders diff --git a/render/vulkan/shaders/quad.frag b/render/vulkan/shaders/quad.frag new file mode 100644 index 00000000..affd1f11 --- /dev/null +++ b/render/vulkan/shaders/quad.frag @@ -0,0 +1,10 @@ +#version 450 + +layout(location = 0) out vec4 out_color; +layout(push_constant) uniform UBO { + layout(offset = 80) vec4 color; +} data; + +void main() { + out_color = data.color; +} diff --git a/render/vulkan/shaders/texture.frag b/render/vulkan/shaders/texture.frag new file mode 100644 index 00000000..7a7b8c57 --- /dev/null +++ b/render/vulkan/shaders/texture.frag @@ -0,0 +1,25 @@ +#version 450 + +layout(set = 0, binding = 0) uniform sampler2D tex; + +layout(location = 0) in vec2 uv; +layout(location = 0) out vec4 out_color; + +layout(push_constant) uniform UBO { + layout(offset = 80) float alpha; +} data; + +void main() { + out_color = textureLod(tex, uv, 0); + + // We expect this shader to output pre-alpha-multiplied color values. + // alpha < 0.0 means that this shader should ignore the texture's alpha + // value. + if (data.alpha < 0.0) { + out_color.a = -data.alpha; + out_color.rgb *= -data.alpha; + } else { + out_color *= data.alpha; + } +} + diff --git a/render/vulkan/texture.c b/render/vulkan/texture.c new file mode 100644 index 00000000..b4ba67ed --- /dev/null +++ b/render/vulkan/texture.c @@ -0,0 +1,718 @@ +#define _POSIX_C_SOURCE 200809L +#include <assert.h> +#include <fcntl.h> +#include <stdbool.h> +#include <stdlib.h> +#include <string.h> +#include <sys/stat.h> +#include <unistd.h> +#include <wlr/render/wlr_texture.h> +#include <wlr/util/log.h> +#include "render/pixel_format.h" +#include "render/vulkan.h" + +static const struct wlr_texture_impl texture_impl; + +struct wlr_vk_texture *vulkan_get_texture(struct wlr_texture *wlr_texture) { + assert(wlr_texture->impl == &texture_impl); + return (struct wlr_vk_texture *)wlr_texture; +} + +static VkImageAspectFlagBits mem_plane_aspect(unsigned i) { + switch (i) { + case 0: return VK_IMAGE_ASPECT_MEMORY_PLANE_0_BIT_EXT; + case 1: return VK_IMAGE_ASPECT_MEMORY_PLANE_1_BIT_EXT; + case 2: return VK_IMAGE_ASPECT_MEMORY_PLANE_2_BIT_EXT; + case 3: return VK_IMAGE_ASPECT_MEMORY_PLANE_3_BIT_EXT; + default: abort(); // unreachable + } +} + +static bool vulkan_texture_is_opaque(struct wlr_texture *wlr_texture) { + struct wlr_vk_texture *texture = vulkan_get_texture(wlr_texture); + const struct wlr_pixel_format_info *format_info = drm_get_pixel_format_info( + texture->format->drm_format); + assert(format_info); + return !format_info->has_alpha; +} + +// Will transition the texture to shaderReadOnlyOptimal layout for reading +// from fragment shader later on +static bool write_pixels(struct wlr_texture *wlr_texture, + uint32_t stride, uint32_t width, uint32_t height, uint32_t src_x, + uint32_t src_y, uint32_t dst_x, uint32_t dst_y, const void *vdata, + VkImageLayout old_layout, VkPipelineStageFlags src_stage, + VkAccessFlags src_access) { + VkResult res; + struct wlr_vk_texture *texture = vulkan_get_texture(wlr_texture); + struct wlr_vk_renderer *renderer = texture->renderer; + VkDevice dev = texture->renderer->dev->dev; + + // make sure assumptions are met + assert(src_x + width <= texture->wlr_texture.width); + assert(src_y + height <= texture->wlr_texture.height); + assert(dst_x + width <= texture->wlr_texture.width); + assert(dst_y + height <= texture->wlr_texture.height); + + const struct wlr_pixel_format_info *format_info = drm_get_pixel_format_info( + texture->format->drm_format); + assert(format_info); + + // deferred upload by transfer; using staging buffer + // calculate maximum side needed + uint32_t bsize = 0; + unsigned bytespb = format_info->bpp / 8; + bsize += height * bytespb * width; + + // get staging buffer + struct wlr_vk_buffer_span span = vulkan_get_stage_span(renderer, bsize); + if (!span.buffer || span.alloc.size != bsize) { + wlr_log(WLR_ERROR, "Failed to retrieve staging buffer"); + return false; + } + + void *vmap; + res = vkMapMemory(dev, span.buffer->memory, span.alloc.start, + bsize, 0, &vmap); + if (res != VK_SUCCESS) { + wlr_vk_error("vkMapMemory", res); + return false; + } + char *map = (char *)vmap; + + // record staging cb + // will be executed before next frame + VkCommandBuffer cb = vulkan_record_stage_cb(renderer); + vulkan_change_layout(cb, texture->image, + old_layout, src_stage, src_access, + VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_ACCESS_TRANSFER_WRITE_BIT); + + // upload data + const char *pdata = vdata; // data iterator + + uint32_t packed_stride = bytespb * width; + uint32_t buf_off = span.alloc.start + (map - (char *)vmap); + + // write data into staging buffer span + pdata += stride * src_y; + pdata += bytespb * src_x; + if (src_x == 0 && width == texture->wlr_texture.width && + stride == packed_stride) { + memcpy(map, pdata, packed_stride * height); + map += packed_stride * height; + } else { + for (unsigned i = 0u; i < height; ++i) { + memcpy(map, pdata, packed_stride); + pdata += stride; + map += packed_stride; + } + } + + VkBufferImageCopy copy; + copy.imageExtent.width = width; + copy.imageExtent.height = height; + copy.imageExtent.depth = 1; + copy.imageOffset.x = dst_x; + copy.imageOffset.y = dst_y; + copy.imageOffset.z = 0; + copy.bufferOffset = buf_off; + copy.bufferRowLength = width; + copy.bufferImageHeight = height; + copy.imageSubresource.mipLevel = 0; + copy.imageSubresource.baseArrayLayer = 0; + copy.imageSubresource.layerCount = 1; + copy.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + + assert((uint32_t)(map - (char *)vmap) == bsize); + vkUnmapMemory(dev, span.buffer->memory); + + vkCmdCopyBufferToImage(cb, span.buffer->buffer, texture->image, + VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, ©); + vulkan_change_layout(cb, texture->image, + VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_PIPELINE_STAGE_TRANSFER_BIT, + VK_ACCESS_TRANSFER_WRITE_BIT, + VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, + VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, VK_ACCESS_SHADER_READ_BIT); + texture->last_used = renderer->frame; + + return true; +} + +static bool vulkan_texture_write_pixels(struct wlr_texture *wlr_texture, + uint32_t stride, uint32_t width, uint32_t height, uint32_t src_x, + uint32_t src_y, uint32_t dst_x, uint32_t dst_y, const void *vdata) { + return write_pixels(wlr_texture, stride, width, height, src_x, src_y, + dst_x, dst_y, vdata, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, + VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_ACCESS_SHADER_READ_BIT); +} + +void vulkan_texture_destroy(struct wlr_vk_texture *texture) { + if (!texture->renderer) { + free(texture); + return; + } + + // when we recorded a command to fill this image _this_ frame, + // it has to be executed before the texture can be destroyed. + // Add it to the renderer->destroy_textures list, destroying + // _after_ the stage command buffer has exectued + if (texture->last_used == texture->renderer->frame) { + assert(texture->destroy_link.next == NULL); // not already inserted + wl_list_insert(&texture->renderer->destroy_textures, + &texture->destroy_link); + return; + } + + wl_list_remove(&texture->link); + wl_list_remove(&texture->buffer_destroy.link); + + VkDevice dev = texture->renderer->dev->dev; + if (texture->ds && texture->ds_pool) { + vulkan_free_ds(texture->renderer, texture->ds_pool, texture->ds); + } + + vkDestroyImageView(dev, texture->image_view, NULL); + vkDestroyImage(dev, texture->image, NULL); + + for (unsigned i = 0u; i < texture->mem_count; ++i) { + vkFreeMemory(dev, texture->memories[i], NULL); + } + + free(texture); +} + +static void vulkan_texture_unref(struct wlr_texture *wlr_texture) { + struct wlr_vk_texture *texture = vulkan_get_texture(wlr_texture); + if (texture->buffer != NULL) { + // Keep the texture around, in case the buffer is re-used later. We're + // still listening to the buffer's destroy event. + wlr_buffer_unlock(texture->buffer); + } else { + vulkan_texture_destroy(texture); + } +} + +static const struct wlr_texture_impl texture_impl = { + .is_opaque = vulkan_texture_is_opaque, + .write_pixels = vulkan_texture_write_pixels, + .destroy = vulkan_texture_unref, +}; + +static struct wlr_vk_texture *vulkan_texture_create( + struct wlr_vk_renderer *renderer, uint32_t width, uint32_t height) { + struct wlr_vk_texture *texture = + calloc(1, sizeof(struct wlr_vk_texture)); + if (texture == NULL) { + wlr_log_errno(WLR_ERROR, "Allocation failed"); + return NULL; + } + wlr_texture_init(&texture->wlr_texture, &texture_impl, width, height); + texture->renderer = renderer; + wl_list_insert(&renderer->textures, &texture->link); + wl_list_init(&texture->buffer_destroy.link); + return texture; +} + +static struct wlr_texture *vulkan_texture_from_pixels(struct wlr_renderer *wlr_renderer, + uint32_t drm_fmt, uint32_t stride, uint32_t width, + uint32_t height, const void *data) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + + VkResult res; + VkDevice dev = renderer->dev->dev; + + wlr_log(WLR_DEBUG, "vulkan_texture_from_pixels: %.4s, %dx%d", + (const char*) &drm_fmt, width, height); + + const struct wlr_vk_format_props *fmt = + vulkan_format_props_from_drm(renderer->dev, drm_fmt); + if (fmt == NULL) { + wlr_log(WLR_ERROR, "Unsupported pixel format %"PRIx32 " (%.4s)", + drm_fmt, (const char*) &drm_fmt); + return NULL; + } + + struct wlr_vk_texture *texture = vulkan_texture_create(renderer, width, height); + if (texture == NULL) { + return NULL; + } + + texture->format = &fmt->format; + + // create image + unsigned mem_bits = 0xFFFFFFFF; + + VkImageCreateInfo img_info = {0}; + img_info.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO; + img_info.imageType = VK_IMAGE_TYPE_2D; + img_info.format = texture->format->vk_format; + img_info.mipLevels = 1; + img_info.arrayLayers = 1; + img_info.samples = VK_SAMPLE_COUNT_1_BIT; + img_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE; + img_info.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; + img_info.extent = (VkExtent3D) { width, height, 1 }; + img_info.usage = VK_IMAGE_USAGE_SAMPLED_BIT; + + img_info.tiling = VK_IMAGE_TILING_OPTIMAL; + img_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; + mem_bits = vulkan_find_mem_type(renderer->dev, + VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, mem_bits); + VkImageLayout layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + + res = vkCreateImage(dev, &img_info, NULL, &texture->image); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateImage failed", res); + goto error; + } + + // memory + VkMemoryRequirements mem_reqs; + vkGetImageMemoryRequirements(dev, texture->image, &mem_reqs); + + VkMemoryAllocateInfo mem_info = {0}; + mem_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + mem_info.allocationSize = mem_reqs.size; + mem_info.memoryTypeIndex = mem_bits & mem_reqs.memoryTypeBits; + res = vkAllocateMemory(dev, &mem_info, NULL, &texture->memories[0]); + if (res != VK_SUCCESS) { + wlr_vk_error("vkAllocatorMemory failed", res); + goto error; + } + + texture->mem_count = 1; + res = vkBindImageMemory(dev, texture->image, texture->memories[0], 0); + if (res != VK_SUCCESS) { + wlr_vk_error("vkBindMemory failed", res); + goto error; + } + + // view + VkImageViewCreateInfo view_info = {0}; + view_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + view_info.viewType = VK_IMAGE_VIEW_TYPE_2D; + view_info.format = texture->format->vk_format; + view_info.components.r = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.g = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.b = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.a = VK_COMPONENT_SWIZZLE_IDENTITY; + + view_info.subresourceRange = (VkImageSubresourceRange) { + VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 + }; + view_info.image = texture->image; + + res = vkCreateImageView(dev, &view_info, NULL, + &texture->image_view); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateImageView failed", res); + goto error; + } + + // descriptor + texture->ds_pool = vulkan_alloc_texture_ds(renderer, &texture->ds); + if (!texture->ds_pool) { + wlr_log(WLR_ERROR, "failed to allocate descriptor"); + goto error; + } + + VkDescriptorImageInfo ds_img_info = {0}; + ds_img_info.imageView = texture->image_view; + ds_img_info.imageLayout = layout; + + VkWriteDescriptorSet ds_write = {0}; + ds_write.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET; + ds_write.descriptorCount = 1; + ds_write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; + ds_write.dstSet = texture->ds; + ds_write.pImageInfo = &ds_img_info; + + vkUpdateDescriptorSets(dev, 1, &ds_write, 0, NULL); + + // write data + if (!write_pixels(&texture->wlr_texture, stride, + width, height, 0, 0, 0, 0, data, VK_IMAGE_LAYOUT_UNDEFINED, + VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0)) { + goto error; + } + + return &texture->wlr_texture; + +error: + vulkan_texture_destroy(texture); + return NULL; +} + +static bool is_dmabuf_disjoint(const struct wlr_dmabuf_attributes *attribs) { + if (attribs->n_planes == 1) { + return false; + } + + struct stat first_stat; + if (fstat(attribs->fd[0], &first_stat) != 0) { + wlr_log_errno(WLR_ERROR, "fstat failed"); + return true; + } + + for (int i = 1; i < attribs->n_planes; i++) { + struct stat plane_stat; + if (fstat(attribs->fd[i], &plane_stat) != 0) { + wlr_log_errno(WLR_ERROR, "fstat failed"); + return true; + } + + if (first_stat.st_ino != plane_stat.st_ino) { + return true; + } + } + + return false; +} + +VkImage vulkan_import_dmabuf(struct wlr_vk_renderer *renderer, + const struct wlr_dmabuf_attributes *attribs, + VkDeviceMemory mems[static WLR_DMABUF_MAX_PLANES], uint32_t *n_mems, + bool for_render) { + VkResult res; + VkDevice dev = renderer->dev->dev; + *n_mems = 0u; + + wlr_log(WLR_DEBUG, "vulkan_import_dmabuf: %.4s (mod %"PRIx64"), %dx%d, %d planes", + (const char *)&attribs->format, attribs->modifier, + attribs->width, attribs->height, attribs->n_planes); + + struct wlr_vk_format_props *fmt = vulkan_format_props_from_drm(renderer->dev, + attribs->format); + if (fmt == NULL) { + wlr_log(WLR_ERROR, "Unsupported pixel format %"PRIx32 " (%.4s)", + attribs->format, (const char*) &attribs->format); + return VK_NULL_HANDLE; + } + + uint32_t plane_count = attribs->n_planes; + assert(plane_count < WLR_DMABUF_MAX_PLANES); + struct wlr_vk_format_modifier_props *mod = + vulkan_format_props_find_modifier(fmt, attribs->modifier, for_render); + if (!mod || !(mod->dmabuf_flags & VK_EXTERNAL_MEMORY_FEATURE_IMPORTABLE_BIT)) { + wlr_log(WLR_ERROR, "Format %"PRIx32" (%.4s) can't be used with modifier " + "%"PRIx64, attribs->format, (const char*) &attribs->format, + attribs->modifier); + return VK_NULL_HANDLE; + } + + if ((uint32_t) attribs->width > mod->max_extent.width || + (uint32_t) attribs->height > mod->max_extent.height) { + wlr_log(WLR_ERROR, "dmabuf is too large to import"); + return VK_NULL_HANDLE; + } + + if (mod->props.drmFormatModifierPlaneCount != plane_count) { + wlr_log(WLR_ERROR, "Number of planes (%d) does not match format (%d)", + plane_count, mod->props.drmFormatModifierPlaneCount); + return VK_NULL_HANDLE; + } + + // check if we have to create the image disjoint + bool disjoint = is_dmabuf_disjoint(attribs); + if (disjoint && !(mod->props.drmFormatModifierTilingFeatures + & VK_FORMAT_FEATURE_DISJOINT_BIT)) { + wlr_log(WLR_ERROR, "Format/Modifier does not support disjoint images"); + return VK_NULL_HANDLE; + } + + // image + VkExternalMemoryHandleTypeFlagBits htype = + VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT; + + VkImageCreateInfo img_info = {0}; + img_info.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO; + img_info.imageType = VK_IMAGE_TYPE_2D; + img_info.format = fmt->format.vk_format; + img_info.mipLevels = 1; + img_info.arrayLayers = 1; + img_info.samples = VK_SAMPLE_COUNT_1_BIT; + img_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE; + img_info.initialLayout = VK_IMAGE_LAYOUT_PREINITIALIZED; + img_info.extent = (VkExtent3D) { attribs->width, attribs->height, 1 }; + img_info.usage = for_render ? + VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT : + VK_IMAGE_USAGE_SAMPLED_BIT; + if (disjoint) { + img_info.flags = VK_IMAGE_CREATE_DISJOINT_BIT; + } + + VkExternalMemoryImageCreateInfo eimg = {0}; + eimg.sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO; + eimg.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT; + img_info.pNext = &eimg; + + VkSubresourceLayout plane_layouts[WLR_DMABUF_MAX_PLANES] = {0}; + VkImageDrmFormatModifierExplicitCreateInfoEXT mod_info = {0}; + + img_info.tiling = VK_IMAGE_TILING_DRM_FORMAT_MODIFIER_EXT; + for (unsigned i = 0u; i < plane_count; ++i) { + plane_layouts[i].offset = attribs->offset[i]; + plane_layouts[i].rowPitch = attribs->stride[i]; + plane_layouts[i].size = 0; + } + + mod_info.sType = VK_STRUCTURE_TYPE_IMAGE_DRM_FORMAT_MODIFIER_EXPLICIT_CREATE_INFO_EXT; + mod_info.drmFormatModifierPlaneCount = plane_count; + mod_info.drmFormatModifier = mod->props.drmFormatModifier; + mod_info.pPlaneLayouts = plane_layouts; + eimg.pNext = &mod_info; + + VkImage image; + res = vkCreateImage(dev, &img_info, NULL, &image); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateImage", res); + return VK_NULL_HANDLE; + } + + unsigned mem_count = disjoint ? plane_count : 1u; + VkBindImageMemoryInfo bindi[WLR_DMABUF_MAX_PLANES] = {0}; + VkBindImagePlaneMemoryInfo planei[WLR_DMABUF_MAX_PLANES] = {0}; + + for (unsigned i = 0u; i < mem_count; ++i) { + struct VkMemoryFdPropertiesKHR fdp = {0}; + fdp.sType = VK_STRUCTURE_TYPE_MEMORY_FD_PROPERTIES_KHR; + res = renderer->dev->api.getMemoryFdPropertiesKHR(dev, htype, + attribs->fd[i], &fdp); + if (res != VK_SUCCESS) { + wlr_vk_error("getMemoryFdPropertiesKHR", res); + goto error_image; + } + + VkImageMemoryRequirementsInfo2 memri = {0}; + memri.image = image; + memri.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_REQUIREMENTS_INFO_2; + + VkImagePlaneMemoryRequirementsInfo planeri = {0}; + if (disjoint) { + planeri.sType = VK_STRUCTURE_TYPE_IMAGE_PLANE_MEMORY_REQUIREMENTS_INFO; + planeri.planeAspect = mem_plane_aspect(i); + memri.pNext = &planeri; + } + + VkMemoryRequirements2 memr = {0}; + memr.sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2; + + vkGetImageMemoryRequirements2(dev, &memri, &memr); + int mem = vulkan_find_mem_type(renderer->dev, 0, + memr.memoryRequirements.memoryTypeBits & fdp.memoryTypeBits); + if (mem < 0) { + wlr_log(WLR_ERROR, "no valid memory type index"); + goto error_image; + } + + // Since importing transfers ownership of the FD to Vulkan, we have + // to duplicate it since this operation does not transfer ownership + // of the attribs to this texture. Will be closed by Vulkan on + // vkFreeMemory. + int dfd = fcntl(attribs->fd[i], F_DUPFD_CLOEXEC, 0); + if (dfd < 0) { + wlr_log_errno(WLR_ERROR, "fcntl(F_DUPFD_CLOEXEC) failed"); + goto error_image; + } + + VkMemoryAllocateInfo memi = {0}; + memi.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO; + memi.allocationSize = memr.memoryRequirements.size; + memi.memoryTypeIndex = mem; + + VkImportMemoryFdInfoKHR importi = {0}; + importi.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_FD_INFO_KHR; + importi.fd = dfd; + importi.handleType = htype; + memi.pNext = &importi; + + VkMemoryDedicatedAllocateInfo dedi = {0}; + dedi.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO; + dedi.image = image; + importi.pNext = &dedi; + + res = vkAllocateMemory(dev, &memi, NULL, &mems[i]); + if (res != VK_SUCCESS) { + close(dfd); + wlr_vk_error("vkAllocateMemory failed", res); + goto error_image; + } + + ++(*n_mems); + + // fill bind info + bindi[i].image = image; + bindi[i].memory = mems[i]; + bindi[i].memoryOffset = 0; + bindi[i].sType = VK_STRUCTURE_TYPE_BIND_IMAGE_MEMORY_INFO; + + if (disjoint) { + planei[i].sType = VK_STRUCTURE_TYPE_BIND_IMAGE_PLANE_MEMORY_INFO; + planei[i].planeAspect = planeri.planeAspect; + bindi[i].pNext = &planei[i]; + } + } + + res = vkBindImageMemory2(dev, mem_count, bindi); + if (res != VK_SUCCESS) { + wlr_vk_error("vkBindMemory failed", res); + goto error_image; + } + + return image; + +error_image: + vkDestroyImage(dev, image, NULL); + for (size_t i = 0u; i < *n_mems; ++i) { + vkFreeMemory(dev, mems[i], NULL); + mems[i] = VK_NULL_HANDLE; + } + + return VK_NULL_HANDLE; +} + +static struct wlr_texture *vulkan_texture_from_dmabuf(struct wlr_renderer *wlr_renderer, + struct wlr_dmabuf_attributes *attribs) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + + VkResult res; + VkDevice dev = renderer->dev->dev; + + const struct wlr_vk_format_props *fmt = vulkan_format_props_from_drm( + renderer->dev, attribs->format); + if (fmt == NULL) { + wlr_log(WLR_ERROR, "Unsupported pixel format %"PRIx32 " (%.4s)", + attribs->format, (const char*) &attribs->format); + return NULL; + } + + struct wlr_vk_texture *texture = vulkan_texture_create(renderer, + attribs->width, attribs->height); + if (texture == NULL) { + return NULL; + } + + texture->format = &fmt->format; + texture->image = vulkan_import_dmabuf(renderer, attribs, + texture->memories, &texture->mem_count, false); + if (!texture->image) { + goto error; + } + + uint32_t flags = attribs->flags; + if (flags & WLR_DMABUF_ATTRIBUTES_FLAGS_Y_INVERT) { + texture->invert_y = true; + flags &= ~WLR_DMABUF_ATTRIBUTES_FLAGS_Y_INVERT; + } + + if (flags != 0) { + wlr_log(WLR_ERROR, "dmabuf flags %x not supported/implemented on vulkan", + attribs->flags); + // NOTE: should probably make this a critical error in future + // return VK_NULL_HANDLE; + } + + // view + VkImageViewCreateInfo view_info = {0}; + view_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO; + view_info.viewType = VK_IMAGE_VIEW_TYPE_2D; + view_info.format = texture->format->vk_format; + view_info.components.r = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.g = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.b = VK_COMPONENT_SWIZZLE_IDENTITY; + view_info.components.a = VK_COMPONENT_SWIZZLE_IDENTITY; + + view_info.subresourceRange = (VkImageSubresourceRange) { + VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 + }; + view_info.image = texture->image; + + res = vkCreateImageView(dev, &view_info, NULL, &texture->image_view); + if (res != VK_SUCCESS) { + wlr_vk_error("vkCreateImageView failed", res); + goto error; + } + + // descriptor + texture->ds_pool = vulkan_alloc_texture_ds(renderer, &texture->ds); + if (!texture->ds_pool) { + wlr_log(WLR_ERROR, "failed to allocate descriptor"); + goto error; + } + + VkDescriptorImageInfo ds_img_info = {0}; + ds_img_info.imageView = texture->image_view; + ds_img_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + + VkWriteDescriptorSet ds_write = {0}; + ds_write.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET; + ds_write.descriptorCount = 1; + ds_write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER; + ds_write.dstSet = texture->ds; + ds_write.pImageInfo = &ds_img_info; + + vkUpdateDescriptorSets(dev, 1, &ds_write, 0, NULL); + texture->dmabuf_imported = true; + + return &texture->wlr_texture; + +error: + vulkan_texture_destroy(texture); + return NULL; +} + +static void texture_handle_buffer_destroy(struct wl_listener *listener, + void *data) { + struct wlr_vk_texture *texture = + wl_container_of(listener, texture, buffer_destroy); + vulkan_texture_destroy(texture); +} + +static struct wlr_texture *vulkan_texture_from_dmabuf_buffer( + struct wlr_vk_renderer *renderer, struct wlr_buffer *buffer, + struct wlr_dmabuf_attributes *dmabuf) { + struct wlr_vk_texture *texture; + wl_list_for_each(texture, &renderer->textures, link) { + if (texture->buffer == buffer) { + wlr_buffer_lock(texture->buffer); + return &texture->wlr_texture; + } + } + + struct wlr_texture *wlr_texture = + vulkan_texture_from_dmabuf(&renderer->wlr_renderer, dmabuf); + if (wlr_texture == NULL) { + return false; + } + + texture = vulkan_get_texture(wlr_texture); + texture->buffer = wlr_buffer_lock(buffer); + + texture->buffer_destroy.notify = texture_handle_buffer_destroy; + wl_signal_add(&buffer->events.destroy, &texture->buffer_destroy); + + return &texture->wlr_texture; +} + +struct wlr_texture *vulkan_texture_from_buffer( + struct wlr_renderer *wlr_renderer, + struct wlr_buffer *buffer) { + struct wlr_vk_renderer *renderer = vulkan_get_renderer(wlr_renderer); + + void *data; + uint32_t format; + size_t stride; + struct wlr_dmabuf_attributes dmabuf; + if (wlr_buffer_get_dmabuf(buffer, &dmabuf)) { + return vulkan_texture_from_dmabuf_buffer(renderer, buffer, &dmabuf); + } else if (wlr_buffer_begin_data_ptr_access(buffer, + WLR_BUFFER_DATA_PTR_ACCESS_READ, &data, &format, &stride)) { + struct wlr_texture *tex = vulkan_texture_from_pixels(wlr_renderer, + format, stride, buffer->width, buffer->height, data); + wlr_buffer_end_data_ptr_access(buffer); + return tex; + } else { + return NULL; + } +} diff --git a/render/vulkan/util.c b/render/vulkan/util.c new file mode 100644 index 00000000..b15c57eb --- /dev/null +++ b/render/vulkan/util.c @@ -0,0 +1,93 @@ +#include <vulkan/vulkan.h> +#include <wlr/util/log.h> +#include "render/vulkan.h" + +int vulkan_find_mem_type(struct wlr_vk_device *dev, + VkMemoryPropertyFlags flags, uint32_t req_bits) { + + VkPhysicalDeviceMemoryProperties props; + vkGetPhysicalDeviceMemoryProperties(dev->phdev, &props); + + for (unsigned i = 0u; i < props.memoryTypeCount; ++i) { + if (req_bits & (1 << i)) { + if ((props.memoryTypes[i].propertyFlags & flags) == flags) { + return i; + } + } + } + + return -1; +} + +const char *vulkan_strerror(VkResult err) { + #define ERR_STR(r) case VK_ ##r: return #r + switch (err) { + ERR_STR(SUCCESS); + ERR_STR(NOT_READY); + ERR_STR(TIMEOUT); + ERR_STR(EVENT_SET); + ERR_STR(EVENT_RESET); + ERR_STR(INCOMPLETE); + ERR_STR(SUBOPTIMAL_KHR); + ERR_STR(ERROR_OUT_OF_HOST_MEMORY); + ERR_STR(ERROR_OUT_OF_DEVICE_MEMORY); + ERR_STR(ERROR_INITIALIZATION_FAILED); + ERR_STR(ERROR_DEVICE_LOST); + ERR_STR(ERROR_MEMORY_MAP_FAILED); + ERR_STR(ERROR_LAYER_NOT_PRESENT); + ERR_STR(ERROR_EXTENSION_NOT_PRESENT); + ERR_STR(ERROR_FEATURE_NOT_PRESENT); + ERR_STR(ERROR_INCOMPATIBLE_DRIVER); + ERR_STR(ERROR_TOO_MANY_OBJECTS); + ERR_STR(ERROR_FORMAT_NOT_SUPPORTED); + ERR_STR(ERROR_SURFACE_LOST_KHR); + ERR_STR(ERROR_NATIVE_WINDOW_IN_USE_KHR); + ERR_STR(ERROR_OUT_OF_DATE_KHR); + ERR_STR(ERROR_FRAGMENTED_POOL); + ERR_STR(ERROR_INCOMPATIBLE_DISPLAY_KHR); + ERR_STR(ERROR_VALIDATION_FAILED_EXT); + ERR_STR(ERROR_INVALID_EXTERNAL_HANDLE); + ERR_STR(ERROR_OUT_OF_POOL_MEMORY); + ERR_STR(ERROR_INVALID_DRM_FORMAT_MODIFIER_PLANE_LAYOUT_EXT); + default: + return "<unknown>"; + } + #undef ERR_STR +} + +void vulkan_change_layout_queue(VkCommandBuffer cb, VkImage img, + VkImageLayout ol, VkPipelineStageFlags srcs, VkAccessFlags srca, + VkImageLayout nl, VkPipelineStageFlags dsts, VkAccessFlags dsta, + uint32_t src_family, uint32_t dst_family) { + VkImageMemoryBarrier barrier = {0}; + barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; + barrier.oldLayout = ol; + barrier.newLayout = nl; + barrier.image = img; + barrier.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; + barrier.subresourceRange.layerCount = 1; + barrier.subresourceRange.levelCount = 1; + barrier.srcAccessMask = srca; + barrier.dstAccessMask = dsta; + barrier.srcQueueFamilyIndex = src_family; + barrier.dstQueueFamilyIndex = dst_family; + + vkCmdPipelineBarrier(cb, srcs, dsts, 0, 0, NULL, 0, NULL, 1, &barrier); +} + +void vulkan_change_layout(VkCommandBuffer cb, VkImage img, + VkImageLayout ol, VkPipelineStageFlags srcs, VkAccessFlags srca, + VkImageLayout nl, VkPipelineStageFlags dsts, VkAccessFlags dsta) { + vulkan_change_layout_queue(cb, img, ol, srcs, srca, nl, dsts, dsta, + VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED); +} + +bool vulkan_has_extension(size_t count, const char **exts, const char *find) { + for (unsigned i = 0; i < count; ++i) { + if (strcmp(exts[i], find) == 0u) { + return true; + } + } + + return false; +} diff --git a/render/vulkan/vulkan.c b/render/vulkan/vulkan.c new file mode 100644 index 00000000..4932ec4d --- /dev/null +++ b/render/vulkan/vulkan.c @@ -0,0 +1,550 @@ +#include <assert.h> +#include <math.h> +#include <stdlib.h> +#include <stdint.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <unistd.h> +#include <xf86drm.h> +#include <vulkan/vulkan.h> +#include <wlr/util/log.h> +#include <wlr/version.h> +#include <wlr/config.h> +#include "render/vulkan.h" + +// Returns the name of the first extension that could not be found or NULL. +static const char *find_extensions(const VkExtensionProperties *avail, + unsigned availc, const char **req, unsigned reqc) { + // check if all required extensions are supported + for (size_t i = 0; i < reqc; ++i) { + bool found = false; + for (size_t j = 0; j < availc; ++j) { + if (!strcmp(avail[j].extensionName, req[i])) { + found = true; + break; + } + } + + if (!found) { + return req[i]; + } + } + + return NULL; +} + +static VkBool32 debug_callback(VkDebugUtilsMessageSeverityFlagBitsEXT severity, + VkDebugUtilsMessageTypeFlagsEXT type, + const VkDebugUtilsMessengerCallbackDataEXT *debug_data, + void *data) { + // we ignore some of the non-helpful warnings + static const char *const ignored[] = { + // notifies us that shader output is not consumed since + // we use the shared vertex buffer with uv output + "UNASSIGNED-CoreValidation-Shader-OutputNotConsumed", + }; + + if (debug_data->pMessageIdName) { + for (unsigned i = 0; i < sizeof(ignored) / sizeof(ignored[0]); ++i) { + if (strcmp(debug_data->pMessageIdName, ignored[i]) == 0) { + return false; + } + } + } + + enum wlr_log_importance importance; + switch (severity) { + case VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT: + case VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT: + importance = WLR_ERROR; + break; + default: + case VK_DEBUG_UTILS_MESSAGE_SEVERITY_INFO_BIT_EXT: + importance = WLR_INFO; + break; + } + + wlr_log(importance, "%s (%s)", debug_data->pMessage, + debug_data->pMessageIdName); + if (debug_data->queueLabelCount > 0) { + const char *name = debug_data->pQueueLabels[0].pLabelName; + if (name) { + wlr_log(importance, " last label '%s'", name); + } + } + + for (unsigned i = 0; i < debug_data->objectCount; ++i) { + if (debug_data->pObjects[i].pObjectName) { + wlr_log(importance, " involving '%s'", debug_data->pMessage); + } + } + + return false; +} + + +// instance +struct wlr_vk_instance *vulkan_instance_create(size_t ext_count, + const char **exts, bool debug) { + // we require vulkan 1.1 + PFN_vkEnumerateInstanceVersion pfEnumInstanceVersion = + (PFN_vkEnumerateInstanceVersion) + vkGetInstanceProcAddr(VK_NULL_HANDLE, "vkEnumerateInstanceVersion"); + if (!pfEnumInstanceVersion) { + wlr_log(WLR_ERROR, "wlroots requires vulkan 1.1 which is not available"); + return NULL; + } + + uint32_t ini_version; + if (pfEnumInstanceVersion(&ini_version) != VK_SUCCESS || + ini_version < VK_API_VERSION_1_1) { + wlr_log(WLR_ERROR, "wlroots requires vulkan 1.1 which is not available"); + return NULL; + } + + // query extension support + uint32_t avail_extc = 0; + VkResult res; + res = vkEnumerateInstanceExtensionProperties(NULL, &avail_extc, NULL); + if ((res != VK_SUCCESS) || (avail_extc == 0)) { + wlr_vk_error("Could not enumerate instance extensions (1)", res); + return NULL; + } + + VkExtensionProperties avail_ext_props[avail_extc + 1]; + res = vkEnumerateInstanceExtensionProperties(NULL, &avail_extc, + avail_ext_props); + if (res != VK_SUCCESS) { + wlr_vk_error("Could not enumerate instance extensions (2)", res); + return NULL; + } + + for (size_t j = 0; j < avail_extc; ++j) { + wlr_log(WLR_DEBUG, "Vulkan instance extension %s v%"PRIu32, + avail_ext_props[j].extensionName, avail_ext_props[j].specVersion); + } + + // create instance + struct wlr_vk_instance *ini = calloc(1, sizeof(*ini)); + if (!ini) { + wlr_log_errno(WLR_ERROR, "allocation failed"); + return NULL; + } + + bool debug_utils_found = false; + ini->extensions = calloc(1 + ext_count, sizeof(*ini->extensions)); + if (!ini->extensions) { + wlr_log_errno(WLR_ERROR, "allocation failed"); + goto error; + } + + // find extensions + for (unsigned i = 0; i < ext_count; ++i) { + if (find_extensions(avail_ext_props, avail_extc, &exts[i], 1)) { + wlr_log(WLR_DEBUG, "vulkan instance extension %s not found", + exts[i]); + continue; + } + + ini->extensions[ini->extension_count++] = exts[i]; + } + + if (debug) { + const char *name = VK_EXT_DEBUG_UTILS_EXTENSION_NAME; + if (find_extensions(avail_ext_props, avail_extc, &name, 1) == NULL) { + debug_utils_found = true; + ini->extensions[ini->extension_count++] = name; + } + } + + VkApplicationInfo application_info = {0}; + application_info.sType = VK_STRUCTURE_TYPE_APPLICATION_INFO; + application_info.pEngineName = "wlroots"; + application_info.engineVersion = WLR_VERSION_NUM; + application_info.apiVersion = VK_API_VERSION_1_1; + + const char *layers[] = { + "VK_LAYER_KHRONOS_validation", + // "VK_LAYER_RENDERDOC_Capture", + // "VK_LAYER_live_introspection", + }; + + unsigned layer_count = debug * (sizeof(layers) / sizeof(layers[0])); + + VkInstanceCreateInfo instance_info = {0}; + instance_info.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO; + instance_info.pApplicationInfo = &application_info; + instance_info.enabledExtensionCount = ini->extension_count; + instance_info.ppEnabledExtensionNames = ini->extensions; + instance_info.enabledLayerCount = layer_count; + instance_info.ppEnabledLayerNames = layers; + + VkDebugUtilsMessageSeverityFlagsEXT severity = + // VK_DEBUG_UTILS_MESSAGE_SEVERITY_INFO_BIT_EXT | + VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT | + VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT; + VkDebugUtilsMessageTypeFlagsEXT types = + // VK_DEBUG_UTILS_MESSAGE_TYPE_GENERAL_BIT_EXT | + VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT | + VK_DEBUG_UTILS_MESSAGE_TYPE_PERFORMANCE_BIT_EXT; + + VkDebugUtilsMessengerCreateInfoEXT debug_info = {0}; + debug_info.sType = VK_STRUCTURE_TYPE_DEBUG_UTILS_MESSENGER_CREATE_INFO_EXT; + debug_info.messageSeverity = severity; + debug_info.messageType = types; + debug_info.pfnUserCallback = &debug_callback; + debug_info.pUserData = ini; + + if (debug_utils_found) { + // already adding the debug utils messenger extension to + // instance creation gives us additional information during + // instance creation and destruction, can be useful for debugging + // layers/extensions not being found. + instance_info.pNext = &debug_info; + } + + res = vkCreateInstance(&instance_info, NULL, &ini->instance); + if (res != VK_SUCCESS) { + wlr_vk_error("Could not create instance", res); + goto error; + } + + // debug callback + if (debug_utils_found) { + ini->api.createDebugUtilsMessengerEXT = + (PFN_vkCreateDebugUtilsMessengerEXT) vkGetInstanceProcAddr( + ini->instance, "vkCreateDebugUtilsMessengerEXT"); + ini->api.destroyDebugUtilsMessengerEXT = + (PFN_vkDestroyDebugUtilsMessengerEXT) vkGetInstanceProcAddr( + ini->instance, "vkDestroyDebugUtilsMessengerEXT"); + + if (ini->api.createDebugUtilsMessengerEXT) { + ini->api.createDebugUtilsMessengerEXT(ini->instance, + &debug_info, NULL, &ini->messenger); + } else { + wlr_log(WLR_ERROR, "vkCreateDebugUtilsMessengerEXT not found"); + } + } + + return ini; + +error: + vulkan_instance_destroy(ini); + return NULL; +} + +void vulkan_instance_destroy(struct wlr_vk_instance *ini) { + if (!ini) { + return; + } + + if (ini->messenger && ini->api.destroyDebugUtilsMessengerEXT) { + ini->api.destroyDebugUtilsMessengerEXT(ini->instance, + ini->messenger, NULL); + } + + if (ini->instance) { + vkDestroyInstance(ini->instance, NULL); + } + + free(ini->extensions); + free(ini); +} + +// physical device matching +static void log_phdev(const VkPhysicalDeviceProperties *props) { + uint32_t vv_major = VK_VERSION_MAJOR(props->apiVersion); + uint32_t vv_minor = VK_VERSION_MINOR(props->apiVersion); + uint32_t vv_patch = VK_VERSION_PATCH(props->apiVersion); + + uint32_t dv_major = VK_VERSION_MAJOR(props->driverVersion); + uint32_t dv_minor = VK_VERSION_MINOR(props->driverVersion); + uint32_t dv_patch = VK_VERSION_PATCH(props->driverVersion); + + const char *dev_type = "unknown"; + switch (props->deviceType) { + case VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU: + dev_type = "integrated"; + break; + case VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU: + dev_type = "discrete"; + break; + case VK_PHYSICAL_DEVICE_TYPE_CPU: + dev_type = "cpu"; + break; + case VK_PHYSICAL_DEVICE_TYPE_VIRTUAL_GPU: + dev_type = "vgpu"; + break; + default: + break; + } + + wlr_log(WLR_INFO, "Vulkan device: '%s'", props->deviceName); + wlr_log(WLR_INFO, " Device type: '%s'", dev_type); + wlr_log(WLR_INFO, " Supported API version: %u.%u.%u", vv_major, vv_minor, vv_patch); + wlr_log(WLR_INFO, " Driver version: %u.%u.%u", dv_major, dv_minor, dv_patch); +} + +VkPhysicalDevice vulkan_find_drm_phdev(struct wlr_vk_instance *ini, int drm_fd) { + VkResult res; + uint32_t num_phdevs; + + res = vkEnumeratePhysicalDevices(ini->instance, &num_phdevs, NULL); + if (res != VK_SUCCESS) { + wlr_vk_error("Could not retrieve physical devices", res); + return VK_NULL_HANDLE; + } + + VkPhysicalDevice phdevs[1 + num_phdevs]; + res = vkEnumeratePhysicalDevices(ini->instance, &num_phdevs, phdevs); + if (res != VK_SUCCESS) { + wlr_vk_error("Could not retrieve physical devices", res); + return VK_NULL_HANDLE; + } + + struct stat drm_stat = {0}; + if (fstat(drm_fd, &drm_stat) != 0) { + wlr_log_errno(WLR_ERROR, "fstat failed"); + return VK_NULL_HANDLE; + } + + for (uint32_t i = 0; i < num_phdevs; ++i) { + VkPhysicalDevice phdev = phdevs[i]; + + // check whether device supports vulkan 1.1, needed for + // vkGetPhysicalDeviceProperties2 + VkPhysicalDeviceProperties phdev_props; + vkGetPhysicalDeviceProperties(phdev, &phdev_props); + + log_phdev(&phdev_props); + + if (phdev_props.apiVersion < VK_API_VERSION_1_1) { + // NOTE: we could additionaly check whether the + // VkPhysicalDeviceProperties2KHR extension is supported but + // implementations not supporting 1.1 are unlikely in future + continue; + } + + // check for extensions + uint32_t avail_extc = 0; + res = vkEnumerateDeviceExtensionProperties(phdev, NULL, + &avail_extc, NULL); + if ((res != VK_SUCCESS) || (avail_extc == 0)) { + wlr_vk_error(" Could not enumerate device extensions", res); + continue; + } + + VkExtensionProperties avail_ext_props[avail_extc + 1]; + res = vkEnumerateDeviceExtensionProperties(phdev, NULL, + &avail_extc, avail_ext_props); + if (res != VK_SUCCESS) { + wlr_vk_error(" Could not enumerate device extensions", res); + continue; + } + + const char *name = VK_EXT_PHYSICAL_DEVICE_DRM_EXTENSION_NAME; + if (find_extensions(avail_ext_props, avail_extc, &name, 1) != NULL) { + wlr_log(WLR_DEBUG, " Ignoring physical device \"%s\": " + "VK_EXT_physical_device_drm not supported", + phdev_props.deviceName); + continue; + } + + VkPhysicalDeviceDrmPropertiesEXT drm_props = {0}; + drm_props.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRM_PROPERTIES_EXT; + + VkPhysicalDeviceProperties2 props = {0}; + props.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2; + props.pNext = &drm_props; + + vkGetPhysicalDeviceProperties2(phdev, &props); + + dev_t primary_devid = makedev(drm_props.primaryMajor, drm_props.primaryMinor); + dev_t render_devid = makedev(drm_props.renderMajor, drm_props.renderMinor); + if (primary_devid == drm_stat.st_rdev || + render_devid == drm_stat.st_rdev) { + wlr_log(WLR_INFO, "Found matching Vulkan physical device: %s", + phdev_props.deviceName); + return phdev; + } + } + + return VK_NULL_HANDLE; +} + +struct wlr_vk_device *vulkan_device_create(struct wlr_vk_instance *ini, + VkPhysicalDevice phdev, size_t ext_count, const char **exts) { + VkResult res; + + // check for extensions + uint32_t avail_extc = 0; + res = vkEnumerateDeviceExtensionProperties(phdev, NULL, + &avail_extc, NULL); + if (res != VK_SUCCESS || avail_extc == 0) { + wlr_vk_error("Could not enumerate device extensions (1)", res); + return NULL; + } + + VkExtensionProperties avail_ext_props[avail_extc + 1]; + res = vkEnumerateDeviceExtensionProperties(phdev, NULL, + &avail_extc, avail_ext_props); + if (res != VK_SUCCESS) { + wlr_vk_error("Could not enumerate device extensions (2)", res); + return NULL; + } + + for (size_t j = 0; j < avail_extc; ++j) { + wlr_log(WLR_DEBUG, "Vulkan device extension %s v%"PRIu32, + avail_ext_props[j].extensionName, avail_ext_props[j].specVersion); + } + + // create device + struct wlr_vk_device *dev = calloc(1, sizeof(*dev)); + if (!dev) { + wlr_log_errno(WLR_ERROR, "allocation failed"); + return NULL; + } + + dev->phdev = phdev; + dev->instance = ini; + dev->drm_fd = -1; + dev->extensions = calloc(16 + ext_count, sizeof(*ini->extensions)); + if (!dev->extensions) { + wlr_log_errno(WLR_ERROR, "allocation failed"); + goto error; + } + + // find extensions + for (unsigned i = 0; i < ext_count; ++i) { + if (find_extensions(avail_ext_props, avail_extc, &exts[i], 1)) { + wlr_log(WLR_DEBUG, "vulkan device extension %s not found", + exts[i]); + continue; + } + + dev->extensions[dev->extension_count++] = exts[i]; + } + + // For dmabuf import we require at least the external_memory_fd, + // external_memory_dma_buf, queue_family_foreign and + // image_drm_format_modifier extensions. + const char *names[] = { + VK_KHR_EXTERNAL_MEMORY_FD_EXTENSION_NAME, + VK_KHR_IMAGE_FORMAT_LIST_EXTENSION_NAME, // or vulkan 1.2 + VK_EXT_EXTERNAL_MEMORY_DMA_BUF_EXTENSION_NAME, + VK_EXT_QUEUE_FAMILY_FOREIGN_EXTENSION_NAME, + VK_EXT_IMAGE_DRM_FORMAT_MODIFIER_EXTENSION_NAME, + }; + + unsigned nc = sizeof(names) / sizeof(names[0]); + const char *not_found = find_extensions(avail_ext_props, avail_extc, names, nc); + if (not_found) { + wlr_log(WLR_ERROR, "vulkan: required device extension %s not found", + not_found); + goto error; + } + + for (unsigned i = 0u; i < nc; ++i) { + dev->extensions[dev->extension_count++] = names[i]; + } + + // queue families + { + uint32_t qfam_count; + vkGetPhysicalDeviceQueueFamilyProperties(phdev, &qfam_count, NULL); + assert(qfam_count > 0); + VkQueueFamilyProperties queue_props[qfam_count]; + vkGetPhysicalDeviceQueueFamilyProperties(phdev, &qfam_count, + queue_props); + + bool graphics_found = false; + for (unsigned i = 0u; i < qfam_count; ++i) { + graphics_found = queue_props[i].queueFlags & VK_QUEUE_GRAPHICS_BIT; + if (graphics_found) { + dev->queue_family = i; + break; + } + } + + assert(graphics_found); + } + + const float prio = 1.f; + VkDeviceQueueCreateInfo qinfo = {}; + qinfo.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO; + qinfo.queueFamilyIndex = dev->queue_family; + qinfo.queueCount = 1; + qinfo.pQueuePriorities = &prio; + + VkDeviceCreateInfo dev_info = {0}; + dev_info.pNext = NULL; + dev_info.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO; + dev_info.queueCreateInfoCount = 1u; + dev_info.pQueueCreateInfos = &qinfo; + dev_info.enabledExtensionCount = dev->extension_count; + dev_info.ppEnabledExtensionNames = dev->extensions; + + res = vkCreateDevice(phdev, &dev_info, NULL, &dev->dev); + if (res != VK_SUCCESS) { + wlr_vk_error("Failed to create vulkan device", res); + goto error; + } + + + vkGetDeviceQueue(dev->dev, dev->queue_family, 0, &dev->queue); + + // load api + dev->api.getMemoryFdPropertiesKHR = (PFN_vkGetMemoryFdPropertiesKHR) + vkGetDeviceProcAddr(dev->dev, "vkGetMemoryFdPropertiesKHR"); + + if (!dev->api.getMemoryFdPropertiesKHR) { + wlr_log(WLR_ERROR, "Failed to retrieve required dev function pointers"); + goto error; + } + + // - check device format support - + size_t max_fmts; + const struct wlr_vk_format *fmts = vulkan_get_format_list(&max_fmts); + dev->shm_formats = calloc(max_fmts, sizeof(*dev->shm_formats)); + dev->format_props = calloc(max_fmts, sizeof(*dev->format_props)); + if (!dev->shm_formats || !dev->format_props) { + wlr_log_errno(WLR_ERROR, "allocation failed"); + goto error; + } + + for (unsigned i = 0u; i < max_fmts; ++i) { + vulkan_format_props_query(dev, &fmts[i]); + } + + return dev; + +error: + vulkan_device_destroy(dev); + return NULL; +} + +void vulkan_device_destroy(struct wlr_vk_device *dev) { + if (!dev) { + return; + } + + if (dev->dev) { + vkDestroyDevice(dev->dev, NULL); + } + + if (dev->drm_fd > 0) { + close(dev->drm_fd); + } + + wlr_drm_format_set_finish(&dev->dmabuf_render_formats); + wlr_drm_format_set_finish(&dev->dmabuf_texture_formats); + + for (unsigned i = 0u; i < dev->format_prop_count; ++i) { + vulkan_format_props_finish(&dev->format_props[i]); + } + + free(dev->extensions); + free(dev->shm_formats); + free(dev->format_props); + free(dev); +} diff --git a/render/wlr_renderer.c b/render/wlr_renderer.c index 5818cd4a..8a1f1e94 100644 --- a/render/wlr_renderer.c +++ b/render/wlr_renderer.c @@ -17,6 +17,10 @@ #include <wlr/render/gles2.h> #endif +#if WLR_HAS_VULKAN_RENDERER +#include <wlr/render/vulkan.h> +#endif // WLR_HAS_VULKAN_RENDERER + #include "util/signal.h" #include "render/pixel_format.h" #include "render/wlr_renderer.h" @@ -258,6 +262,11 @@ struct wlr_renderer *renderer_autocreate_with_drm_fd(int drm_fd) { return wlr_gles2_renderer_create_with_drm_fd(drm_fd); } #endif +#if WLR_HAS_VULKAN_RENDERER + if (strcmp(name, "vulkan") == 0) { + return wlr_vk_renderer_create_with_drm_fd(drm_fd); + } +#endif if (strcmp(name, "pixman") == 0) { return wlr_pixman_renderer_create(); } |