lpnf
/
lovr
mirror of https://github.com/bjornbytes/lovr.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Improve OpenXR layout transitions a bit;

This commit is contained in:
bjorn 2022-08-06 23:25:49 -07:00
parent 8aa1cf91b2
commit 2ebb4bb415
2 changed files with 24 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
src/core/gpu_vk.c
View File

@ -708,33 +708,45 @@ gpu_texture* gpu_surface_acquire(void) {
return &state.surfaceTextures[state.currentSurfaceTexture];
}
// The barriers here are a bit lazy (oversynchronized) and can be improved
void gpu_xr_acquire(gpu_stream* stream, gpu_texture* texture) {
if (texture->layout == VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL) { // TODO depth
VkImageLayout attachmentLayout = texture->aspect == VK_IMAGE_ASPECT_COLOR_BIT ?
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL :
VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
// If the texture only has the RENDER usage, its natural layout matches the layout that OpenXR
// gives us the texture in, so no layout transition is needed.
if (texture->layout == attachmentLayout) {
return;
}
VkImageMemoryBarrier transition = {
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.srcAccessMask = 0,
.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
.dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
.oldLayout = attachmentLayout,
.newLayout = texture->layout,
.image = texture->handle,
.subresourceRange.aspectMask = texture->aspect,
.subresourceRange.baseMipLevel = 0,
.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS,
.subresourceRange.baseArrayLayer = 0,
.subresourceRange.layerCount = VK_REMAINING_ARRAY_LAYERS
};
VkPipelineStageFlags prev = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
VkPipelineStageFlags next = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
VkPipelineStageFlags next = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
vkCmdPipelineBarrier(stream->commands, prev, next, 0, 0, NULL, 0, NULL, 1, &transition);
}
// The barriers here are a bit lazy (oversynchronized) and can be improved
void gpu_xr_release(gpu_stream* stream, gpu_texture* texture) {
if (texture->layout == VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL) { // TODO depth
VkImageLayout attachmentLayout = texture->aspect == VK_IMAGE_ASPECT_COLOR_BIT ?
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL :
VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
// If the texture only has the RENDER usage, its natural layout matches the layout that OpenXR
// expects the texture to be in, so no layout transition is needed.
if (texture->layout == attachmentLayout) {
return;
}
@ -743,12 +755,10 @@ void gpu_xr_release(gpu_stream* stream, gpu_texture* texture) {
.srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
.dstAccessMask = 0,
.oldLayout = texture->layout,
.newLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL,
.newLayout = attachmentLayout,
.image = texture->handle,
.subresourceRange.aspectMask = texture->aspect,
.subresourceRange.baseMipLevel = 0,
.subresourceRange.levelCount = VK_REMAINING_MIP_LEVELS,
.subresourceRange.baseArrayLayer = 0,
.subresourceRange.layerCount = VK_REMAINING_ARRAY_LAYERS
};

8
src/modules/graphics/graphics.c
View File

@ -959,6 +959,10 @@ void lovrGraphicsSubmit(Pass** passes, uint32_t count) {
}
}
for (uint32_t i = 0; i < count; i++) {
gpu_sync(streams[i], &barriers[i], 1);
}
for (uint32_t i = 0; i < count; i++) {
for (uint32_t j = 0; j < passes[i]->access.length; j++) {
passes[i]->access.data[j].sync->lastWriteIndex = 0;
@ -975,10 +979,6 @@ void lovrGraphicsSubmit(Pass** passes, uint32_t count) {
}
}
for (uint32_t i = 0; i < count; i++) {
gpu_sync(streams[i], &barriers[i], 1);
}
for (uint32_t i = 0; i < total; i++) {
gpu_stream_end(streams[i]);
}