RPCS3/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2026-03-10 23:45:16 +01:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

gl: Implement proper, sectioned DMA transfers

Browse source 
- Allows to transfer partial contents when needed
This commit is contained in:
kd-11 2026-03-10 04:05:33 +03:00
parent d7af42635d
commit b49a2c5570
1 changed files with 53 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

61
rpcs3/Emu/RSX/GL/GLTextureCache.h
View file

@ -148,7 +148,7 @@ namespace gl
}
}
void dma_transfer(gl::command_context& cmd, gl::texture* src, const areai& src_area, const utils::address_range32& /*valid_range*/, u32 pitch)
void dma_transfer(gl::command_context& cmd, gl::texture* src, const areai& src_area, const utils::address_range32& valid_range, u32 pitch)
{
init_buffer(src);
glGetError();
@ -171,6 +171,14 @@ namespace gl
{ static_cast<u32>(src_area.width()), static_cast<u32>(src_area.height()), 1 }
};
u32 pbo_offset = 0;
if (valid_range.valid())
{
const u32 section_base = get_section_base();
pbo_offset = valid_range.start - section_base;
ensure(valid_range.start >= section_base && pbo_offset <= pbo.size());
}
bool use_driver_pixel_transform = true;
if (get_driver_caps().ARB_compute_shader_supported) [[likely]]
{
@ -188,9 +196,10 @@ namespace gl
pack_info.type = static_cast<GLenum>(type);
pack_info.block_size = (src->aspect() & image_aspect::stencil) ? 4 : 2;
pack_info.swap_bytes = true;
pack_info.row_length = rsx_pitch / pack_info.block_size;
mem_info.image_size_in_texels = src->width() * src->height();
mem_info.image_size_in_bytes = src->pitch() * src->height();
mem_info.image_size_in_texels = pack_info.row_length * src_area.height();
mem_info.image_size_in_bytes = rsx_pitch * src_area.height();
mem_info.memory_required = 0;
if (pack_info.type == GL_FLOAT_32_UNSIGNED_INT_24_8_REV)
@ -200,13 +209,15 @@ namespace gl
}
void* out_offset = copy_image_to_buffer(cmd, pack_info, src, &scratch_mem, 0, 0, src_rgn, &mem_info);
real_pitch = rsx_pitch;
glBindBuffer(GL_SHADER_STORAGE_BUFFER, GL_NONE);
glMemoryBarrier(GL_BUFFER_UPDATE_BARRIER_BIT);
real_pitch = pack_info.block_size * src->width();
const u64 data_length = pack_info.block_size * mem_info.image_size_in_texels;
scratch_mem.copy_to(&pbo, reinterpret_cast<u64>(out_offset), 0, data_length);
const u64 data_length = mem_info.image_size_in_bytes - rsx_pitch + (src_area.width() * pack_info.block_size);
ensure(data_length + pbo_offset <= static_cast<u64>(pbo.size()), "Memory allocation cannot fit image contents. Report to developers.");
scratch_mem.copy_to(&pbo, reinterpret_cast<u64>(out_offset), pbo_offset, data_length);
}
else
{
@ -225,11 +236,16 @@ namespace gl
pack_unpack_swap_bytes = false;
}
const auto bpp = src->pitch() / src->width();
real_pitch = rsx_pitch;
ensure((real_pitch % bpp) == 0);
pixel_pack_settings pack_settings;
pack_settings.alignment(1);
pack_settings.swap_bytes(pack_unpack_swap_bytes);
pack_settings.row_length(rsx_pitch / bpp);
src->copy_to(pbo, 0, format, type, 0, src_rgn, pack_settings);
src->copy_to(pbo, pbo_offset, format, type, 0, src_rgn, pack_settings);
}
if (auto error = glGetError())
@ -270,6 +286,7 @@ namespace gl
gl::texture* target_texture = vram_texture;
u32 transfer_width = width;
u32 transfer_height = height;
u32 transfer_x = 0, transfer_y = 0;
if (context == rsx::texture_upload_context::framebuffer_storage)
{
@ -315,7 +332,35 @@ namespace gl
}
}
dma_transfer(cmd, target_texture, {}, {}, rsx_pitch);
const auto valid_range = get_confirmed_range();
if (const auto section_range = get_section_range(); section_range != valid_range)
{
if (const auto offset = (valid_range.start - get_section_base()))
{
transfer_y = offset / rsx_pitch;
transfer_x = (offset % rsx_pitch) / rsx::get_format_block_size_in_bytes(gcm_format);
ensure(transfer_width >= transfer_x);
ensure(transfer_height >= transfer_y);
transfer_width -= transfer_x;
transfer_height -= transfer_y;
}
if (const auto tail = (section_range.end - valid_range.end))
{
const auto row_count = tail / rsx_pitch;
ensure(transfer_height >= row_count);
transfer_height -= row_count;
}
}
areai src_area;
src_area.x1 = static_cast<s32>(transfer_x);
src_area.y1 = static_cast<s32>(transfer_y);
src_area.x2 = s32(transfer_x + transfer_width);
src_area.y2 = s32(transfer_y + transfer_height);
dma_transfer(cmd, target_texture, src_area, valid_range, rsx_pitch);
}
/**