mirror of https://github.com/RPCS3/rpcs3.git
vk: Batch compute jobs when doing texture upload
- Reduces overall number of invocations
This commit is contained in:
kd-11
kd-11
parent
6aa0b49dbc
commit
440d58f2ff
|
|
@ -110,7 +110,7 @@ namespace vk
|
|||
VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
|
||||
VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT);
|
||||
|
||||
job->run(cmd, dst, (u32)region.bufferOffset, packed_length, z_offset, s_offset);
|
||||
job->run(cmd, dst, (u32)region.bufferOffset, packed_length, (u32)z_offset, (u32)s_offset);
|
||||
|
||||
vk::insert_buffer_memory_barrier(cmd, dst->value, region.bufferOffset, packed_length,
|
||||
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
|
||||
|
|
@ -166,7 +166,7 @@ namespace vk
|
|||
job = vk::get_compute_task<vk::cs_scatter_d32x8>();
|
||||
}
|
||||
|
||||
job->run(cmd, src, (u32)region.bufferOffset, packed_length, z_offset, s_offset);
|
||||
job->run(cmd, src, (u32)region.bufferOffset, packed_length, (u32)z_offset, (u32)s_offset);
|
||||
|
||||
vk::insert_buffer_memory_barrier(cmd, src->value, z_offset, in_depth_size + in_stencil_size,
|
||||
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
|
||||
|
|
@ -517,10 +517,17 @@ namespace vk
|
|||
u8 block_size_in_bytes = get_format_block_size_in_bytes(format);
|
||||
|
||||
texture_uploader_capabilities caps{ true, false, heap_align };
|
||||
texture_memory_info opt{};
|
||||
bool check_caps = true;
|
||||
|
||||
vk::buffer* scratch_buf = nullptr;
|
||||
u32 scratch_offset = 0;
|
||||
u32 row_pitch, image_linear_size;
|
||||
|
||||
std::vector<VkBufferImageCopy> copy_regions;
|
||||
std::vector<VkBufferCopy> buffer_copies;
|
||||
copy_regions.reserve(subresource_layout.size());
|
||||
|
||||
for (const rsx_subresource_layout &layout : subresource_layout)
|
||||
{
|
||||
if (LIKELY(!heap_align))
|
||||
|
|
@ -539,16 +546,20 @@ namespace vk
|
|||
// Map with extra padding bytes in case of realignment
|
||||
size_t offset_in_buffer = upload_heap.alloc<512>(image_linear_size + 8);
|
||||
void *mapped_buffer = upload_heap.map(offset_in_buffer, image_linear_size + 8);
|
||||
VkBuffer buffer_handle = upload_heap.heap->value;
|
||||
|
||||
// Only do GPU-side conversion if occupancy is good
|
||||
caps.supports_byteswap = (image_linear_size >= 1024);
|
||||
if (check_caps)
|
||||
{
|
||||
caps.supports_byteswap = (image_linear_size >= 1024);
|
||||
check_caps = false;
|
||||
}
|
||||
|
||||
gsl::span<gsl::byte> mapped{ (gsl::byte*)mapped_buffer, ::narrow<int>(image_linear_size) };
|
||||
auto opt = upload_texture_subresource(mapped, layout, format, is_swizzled, caps);
|
||||
opt = upload_texture_subresource(mapped, layout, format, is_swizzled, caps);
|
||||
upload_heap.unmap();
|
||||
|
||||
VkBufferImageCopy copy_info = {};
|
||||
copy_regions.push_back({});
|
||||
auto& copy_info = copy_regions.back();
|
||||
copy_info.bufferOffset = offset_in_buffer;
|
||||
copy_info.imageExtent.height = layout.height_in_block * block_in_pixel;
|
||||
copy_info.imageExtent.width = layout.width_in_block * block_in_pixel;
|
||||
|
|
@ -564,63 +575,72 @@ namespace vk
|
|||
if (!scratch_buf)
|
||||
{
|
||||
scratch_buf = vk::get_scratch_buffer();
|
||||
}
|
||||
else if ((scratch_offset + image_linear_size) > scratch_buf->size())
|
||||
{
|
||||
scratch_offset = 0;
|
||||
insert_buffer_memory_barrier(cmd, scratch_buf->value, 0, scratch_buf->size(), VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
|
||||
VK_ACCESS_TRANSFER_READ_BIT, VK_ACCESS_TRANSFER_WRITE_BIT);
|
||||
buffer_copies.reserve(subresource_layout.size());
|
||||
}
|
||||
|
||||
VkBufferCopy copy = {};
|
||||
// Copy from upload heap to scratch mem
|
||||
buffer_copies.push_back({});
|
||||
auto& copy = buffer_copies.back();
|
||||
copy.srcOffset = offset_in_buffer;
|
||||
copy.dstOffset = scratch_offset;
|
||||
copy.size = image_linear_size;
|
||||
|
||||
vkCmdCopyBuffer(cmd, buffer_handle, scratch_buf->value, 1, ©);
|
||||
|
||||
insert_buffer_memory_barrier(cmd, scratch_buf->value, scratch_offset, image_linear_size, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
|
||||
VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);
|
||||
}
|
||||
|
||||
if (opt.require_swap)
|
||||
{
|
||||
if (opt.element_size == 4)
|
||||
{
|
||||
vk::get_compute_task<vk::cs_shuffle_32>()->run(cmd, scratch_buf, image_linear_size, scratch_offset);
|
||||
}
|
||||
else if (opt.element_size == 2)
|
||||
{
|
||||
vk::get_compute_task<vk::cs_shuffle_16>()->run(cmd, scratch_buf, image_linear_size, scratch_offset);
|
||||
}
|
||||
else
|
||||
{
|
||||
fmt::throw_exception("Unreachable" HERE);
|
||||
}
|
||||
}
|
||||
|
||||
if (dst_image->aspect() & VK_IMAGE_ASPECT_STENCIL_BIT)
|
||||
{
|
||||
// Point data source to scratch mem
|
||||
copy_info.bufferOffset = scratch_offset;
|
||||
scratch_offset = align(scratch_offset + image_linear_size, 512);
|
||||
vk::copy_buffer_to_image(cmd, scratch_buf, dst_image, copy_info);
|
||||
}
|
||||
else if (opt.require_swap)
|
||||
{
|
||||
insert_buffer_memory_barrier(cmd, scratch_buf->value, scratch_offset, image_linear_size, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
|
||||
VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT);
|
||||
|
||||
copy_info.bufferOffset = scratch_offset;
|
||||
scratch_offset = align(scratch_offset + image_linear_size, 512);
|
||||
vkCmdCopyBufferToImage(cmd, scratch_buf->value, dst_image->value, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, ©_info);
|
||||
}
|
||||
else
|
||||
{
|
||||
vkCmdCopyBufferToImage(cmd, buffer_handle, dst_image->value, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, ©_info);
|
||||
scratch_offset += image_linear_size;
|
||||
verify("Out of scratch memory" HERE), (scratch_offset + image_linear_size) <= scratch_buf->size();
|
||||
}
|
||||
|
||||
mipmap_level++;
|
||||
}
|
||||
|
||||
if (opt.require_swap || dst_image->aspect() & VK_IMAGE_ASPECT_STENCIL_BIT)
|
||||
{
|
||||
verify(HERE), scratch_buf;
|
||||
vkCmdCopyBuffer(cmd, upload_heap.heap->value, scratch_buf->value, (u32)buffer_copies.size(), buffer_copies.data());
|
||||
|
||||
insert_buffer_memory_barrier(cmd, scratch_buf->value, 0, scratch_offset, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
|
||||
VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);
|
||||
}
|
||||
|
||||
// Swap if requested
|
||||
if (opt.require_swap)
|
||||
{
|
||||
if (opt.element_size == 4)
|
||||
{
|
||||
vk::get_compute_task<vk::cs_shuffle_32>()->run(cmd, scratch_buf, scratch_offset);
|
||||
}
|
||||
else if (opt.element_size == 2)
|
||||
{
|
||||
vk::get_compute_task<vk::cs_shuffle_16>()->run(cmd, scratch_buf, scratch_offset);
|
||||
}
|
||||
else
|
||||
{
|
||||
fmt::throw_exception("Unreachable" HERE);
|
||||
}
|
||||
}
|
||||
|
||||
// CopyBufferToImage routines
|
||||
if (dst_image->aspect() & VK_IMAGE_ASPECT_STENCIL_BIT)
|
||||
{
|
||||
// Upload in reverse to avoid polluting data in lower space
|
||||
for (auto rIt = copy_regions.crbegin(); rIt != copy_regions.crend(); ++rIt)
|
||||
{
|
||||
vk::copy_buffer_to_image(cmd, scratch_buf, dst_image, *rIt);
|
||||
}
|
||||
}
|
||||
else if (opt.require_swap)
|
||||
{
|
||||
insert_buffer_memory_barrier(cmd, scratch_buf->value, 0, scratch_offset, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
|
||||
VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_TRANSFER_READ_BIT);
|
||||
|
||||
vkCmdCopyBufferToImage(cmd, scratch_buf->value, dst_image->value, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, (u32)copy_regions.size(), copy_regions.data());
|
||||
}
|
||||
else
|
||||
{
|
||||
vkCmdCopyBufferToImage(cmd, upload_heap.heap->value, dst_image->value, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, (u32)copy_regions.size(), copy_regions.data());
|
||||
}
|
||||
}
|
||||
|
||||
VkComponentMapping apply_swizzle_remap(const std::array<VkComponentSwizzle, 4>& base_remap, const std::pair<std::array<u8, 4>, std::array<u8, 4>>& remap_vector)
|
||||
|
|
|
|||
Loading…
Reference in New Issue