gl_rasterizer_cache: Implement render to cubemap.

This commit is contained in:
bunnei 2018-09-20 23:41:25 -04:00
parent 15cc729ebd
commit f543b43fd0
3 changed files with 219 additions and 120 deletions

View file

@ -87,6 +87,9 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
params.size_in_bytes_total = params.SizeInBytesTotal(); params.size_in_bytes_total = params.SizeInBytesTotal();
params.size_in_bytes_2d = params.SizeInBytes2D(); params.size_in_bytes_2d = params.SizeInBytes2D();
params.max_mip_level = config.tic.max_mip_level + 1;
params.rt = {};
return params; return params;
} }
@ -106,6 +109,14 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
params.depth = 1; params.depth = 1;
params.size_in_bytes_total = params.SizeInBytesTotal(); params.size_in_bytes_total = params.SizeInBytesTotal();
params.size_in_bytes_2d = params.SizeInBytes2D(); params.size_in_bytes_2d = params.SizeInBytes2D();
params.max_mip_level = 0;
// Render target specific parameters, not used for caching
params.rt.index = static_cast<u32>(index);
params.rt.array_mode = config.array_mode;
params.rt.layer_stride = config.layer_stride;
params.rt.base_layer = config.base_layer;
return params; return params;
} }
@ -126,6 +137,9 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
params.depth = 1; params.depth = 1;
params.size_in_bytes_total = params.SizeInBytesTotal(); params.size_in_bytes_total = params.SizeInBytesTotal();
params.size_in_bytes_2d = params.SizeInBytes2D(); params.size_in_bytes_2d = params.SizeInBytes2D();
params.max_mip_level = 0;
params.rt = {};
return params; return params;
} }
@ -418,7 +432,8 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr),
}; };
static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface, static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
GLuint read_fb_handle, GLuint draw_fb_handle, std::size_t face = 0) { GLuint read_fb_handle, GLuint draw_fb_handle, GLenum src_attachment = 0,
GLenum dst_attachment = 0, std::size_t cubemap_face = 0) {
const auto& src_params{src_surface->GetSurfaceParams()}; const auto& src_params{src_surface->GetSurfaceParams()};
const auto& dst_params{dst_surface->GetSurfaceParams()}; const auto& dst_params{dst_surface->GetSurfaceParams()};
@ -436,34 +451,35 @@ static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
if (src_params.type == SurfaceType::ColorTexture) { if (src_params.type == SurfaceType::ColorTexture) {
switch (src_params.target) { switch (src_params.target) {
case SurfaceParams::SurfaceTarget::Texture2D: case SurfaceParams::SurfaceTarget::Texture2D:
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
src_surface->Texture().handle, 0); GL_TEXTURE_2D, src_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
0, 0); 0, 0);
break; break;
case SurfaceParams::SurfaceTarget::TextureCubemap: case SurfaceParams::SurfaceTarget::TextureCubemap:
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, glFramebufferTexture2D(
static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face), GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face),
src_surface->Texture().handle, 0); src_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, glFramebufferTexture2D(
static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face), 0, GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
0); static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face), 0, 0);
break; break;
case SurfaceParams::SurfaceTarget::Texture2DArray: case SurfaceParams::SurfaceTarget::Texture2DArray:
glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
src_surface->Texture().handle, 0, 0); src_surface->Texture().handle, 0, 0);
glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, 0, 0, 0); glFramebufferTextureLayer(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, 0, 0, 0);
break; break;
case SurfaceParams::SurfaceTarget::Texture3D: case SurfaceParams::SurfaceTarget::Texture3D:
glFramebufferTexture3D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, glFramebufferTexture3D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
SurfaceTargetToGL(src_params.target), SurfaceTargetToGL(src_params.target),
src_surface->Texture().handle, 0, 0); src_surface->Texture().handle, 0, 0);
glFramebufferTexture3D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, glFramebufferTexture3D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
SurfaceTargetToGL(src_params.target), 0, 0, 0); SurfaceTargetToGL(src_params.target), 0, 0, 0);
break; break;
default: default:
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
src_surface->Texture().handle, 0); GL_TEXTURE_2D, src_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
0, 0); 0, 0);
break; break;
@ -471,35 +487,36 @@ static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
switch (dst_params.target) { switch (dst_params.target) {
case SurfaceParams::SurfaceTarget::Texture2D: case SurfaceParams::SurfaceTarget::Texture2D:
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
dst_surface->Texture().handle, 0); GL_TEXTURE_2D, dst_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
0, 0); 0, 0);
break; break;
case SurfaceParams::SurfaceTarget::TextureCubemap: case SurfaceParams::SurfaceTarget::TextureCubemap:
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, glFramebufferTexture2D(
static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face), GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face),
dst_surface->Texture().handle, 0); dst_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, glFramebufferTexture2D(
static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + face), 0, GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
0); static_cast<GLenum>(GL_TEXTURE_CUBE_MAP_POSITIVE_X + cubemap_face), 0, 0);
break; break;
case SurfaceParams::SurfaceTarget::Texture2DArray: case SurfaceParams::SurfaceTarget::Texture2DArray:
glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
dst_surface->Texture().handle, 0, 0); dst_surface->Texture().handle, 0, 0);
glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, 0, 0, 0); glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, 0, 0, 0);
break; break;
case SurfaceParams::SurfaceTarget::Texture3D: case SurfaceParams::SurfaceTarget::Texture3D:
glFramebufferTexture3D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, glFramebufferTexture3D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
SurfaceTargetToGL(dst_params.target), SurfaceTargetToGL(dst_params.target),
dst_surface->Texture().handle, 0, 0); dst_surface->Texture().handle, 0, 0);
glFramebufferTexture3D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, glFramebufferTexture3D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
SurfaceTargetToGL(dst_params.target), 0, 0, 0); SurfaceTargetToGL(dst_params.target), 0, 0, 0);
break; break;
default: default:
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
dst_surface->Texture().handle, 0); GL_TEXTURE_2D, dst_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
0, 0); 0, 0);
break; break;
@ -507,23 +524,27 @@ static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
buffers = GL_COLOR_BUFFER_BIT; buffers = GL_COLOR_BUFFER_BIT;
} else if (src_params.type == SurfaceType::Depth) { } else if (src_params.type == SurfaceType::Depth) {
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
src_surface->Texture().handle, 0); src_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
dst_surface->Texture().handle, 0); dst_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
buffers = GL_DEPTH_BUFFER_BIT; buffers = GL_DEPTH_BUFFER_BIT;
} else if (src_params.type == SurfaceType::DepthStencil) { } else if (src_params.type == SurfaceType::DepthStencil) {
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + src_attachment,
GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
src_surface->Texture().handle, 0); src_surface->Texture().handle, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + dst_attachment,
GL_TEXTURE_2D, 0, 0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
dst_surface->Texture().handle, 0); dst_surface->Texture().handle, 0);
@ -538,6 +559,92 @@ static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
return true; return true;
} }
static void CopySurface(const Surface& src_surface, const Surface& dst_surface,
GLuint copy_pbo_handle, GLenum src_attachment = 0,
GLenum dst_attachment = 0, std::size_t cubemap_face = 0) {
ASSERT_MSG(dst_attachment == 0, "Unimplemented");
const auto& src_params{src_surface->GetSurfaceParams()};
const auto& dst_params{dst_surface->GetSurfaceParams()};
auto source_format = GetFormatTuple(src_params.pixel_format, src_params.component_type);
auto dest_format = GetFormatTuple(dst_params.pixel_format, dst_params.component_type);
std::size_t buffer_size =
std::max(src_params.size_in_bytes_total, dst_params.size_in_bytes_total);
glBindBuffer(GL_PIXEL_PACK_BUFFER, copy_pbo_handle);
glBufferData(GL_PIXEL_PACK_BUFFER, buffer_size, nullptr, GL_STREAM_DRAW_ARB);
if (source_format.compressed) {
glGetCompressedTextureImage(src_surface->Texture().handle, src_attachment,
static_cast<GLsizei>(src_params.size_in_bytes_total), nullptr);
} else {
glGetTextureImage(src_surface->Texture().handle, src_attachment, source_format.format,
source_format.type, static_cast<GLsizei>(src_params.size_in_bytes_total),
nullptr);
}
// If the new texture is bigger than the previous one, we need to fill in the rest with data
// from the CPU.
if (src_params.size_in_bytes_total < dst_params.size_in_bytes_total) {
// Upload the rest of the memory.
if (dst_params.is_tiled) {
// TODO(Subv): We might have to de-tile the subtexture and re-tile it with the rest
// of the data in this case. Games like Super Mario Odyssey seem to hit this case
// when drawing, it re-uses the memory of a previous texture as a bigger framebuffer
// but it doesn't clear it beforehand, the texture is already full of zeros.
LOG_DEBUG(HW_GPU, "Trying to upload extra texture data from the CPU during "
"reinterpretation but the texture is tiled.");
}
std::size_t remaining_size =
dst_params.size_in_bytes_total - src_params.size_in_bytes_total;
std::vector<u8> data(remaining_size);
Memory::ReadBlock(dst_params.addr + src_params.size_in_bytes_total, data.data(),
data.size());
glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes_total, remaining_size,
data.data());
}
glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
const GLsizei width{static_cast<GLsizei>(
std::min(src_params.GetRect().GetWidth(), dst_params.GetRect().GetWidth()))};
const GLsizei height{static_cast<GLsizei>(
std::min(src_params.GetRect().GetHeight(), dst_params.GetRect().GetHeight()))};
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, copy_pbo_handle);
if (dest_format.compressed) {
LOG_CRITICAL(HW_GPU, "Compressed copy is unimplemented!");
UNREACHABLE();
} else {
switch (dst_params.target) {
case SurfaceParams::SurfaceTarget::Texture1D:
glTextureSubImage1D(dst_surface->Texture().handle, 0, 0, width, dest_format.format,
dest_format.type, nullptr);
break;
case SurfaceParams::SurfaceTarget::Texture2D:
glTextureSubImage2D(dst_surface->Texture().handle, 0, 0, 0, width, height,
dest_format.format, dest_format.type, nullptr);
break;
case SurfaceParams::SurfaceTarget::Texture3D:
case SurfaceParams::SurfaceTarget::Texture2DArray:
glTextureSubImage3D(dst_surface->Texture().handle, 0, 0, 0, 0, width, height,
static_cast<GLsizei>(dst_params.depth), dest_format.format,
dest_format.type, nullptr);
break;
case SurfaceParams::SurfaceTarget::TextureCubemap:
glTextureSubImage3D(dst_surface->Texture().handle, 0, 0, 0,
static_cast<GLint>(cubemap_face), width, height, 1,
dest_format.format, dest_format.type, nullptr);
break;
default:
LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}",
static_cast<u32>(dst_params.target));
UNREACHABLE();
}
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
}
}
CachedSurface::CachedSurface(const SurfaceParams& params) CachedSurface::CachedSurface(const SurfaceParams& params)
: params(params), gl_target(SurfaceTargetToGL(params.target)) { : params(params), gl_target(SurfaceTargetToGL(params.target)) {
texture.Create(); texture.Create();
@ -929,107 +1036,63 @@ Surface RasterizerCacheOpenGL::GetUncachedSurface(const SurfaceParams& params) {
Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface, Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
const SurfaceParams& new_params) { const SurfaceParams& new_params) {
// Verify surface is compatible for blitting // Verify surface is compatible for blitting
const auto& old_params{old_surface->GetSurfaceParams()}; auto old_params{old_surface->GetSurfaceParams()};
// Get a new surface with the new parameters, and blit the previous surface to it // Get a new surface with the new parameters, and blit the previous surface to it
Surface new_surface{GetUncachedSurface(new_params)}; Surface new_surface{GetUncachedSurface(new_params)};
if (old_params.pixel_format == new_params.pixel_format ||
!Settings::values.use_accurate_framebuffers) {
// If the format is the same, just do a framebuffer blit. This is significantly faster than // If the format is the same, just do a framebuffer blit. This is significantly faster than
// using PBOs. The is also likely less accurate, as textures will be converted rather than // using PBOs. The is also likely less accurate, as textures will be converted rather than
// reinterpreted. // reinterpreted. When use_accurate_framebuffers setting is enabled, perform a more accurate
// surface copy, where pixels are reinterpreted as a new format (without conversion). This
// code path uses OpenGL PBOs and is quite slow.
const bool is_blit{old_params.pixel_format == new_params.pixel_format ||
!Settings::values.use_accurate_framebuffers};
switch (new_params.target) {
case SurfaceParams::SurfaceTarget::Texture2D:
if (is_blit) {
BlitSurface(old_surface, new_surface, read_framebuffer.handle, draw_framebuffer.handle); BlitSurface(old_surface, new_surface, read_framebuffer.handle, draw_framebuffer.handle);
} else { } else {
// When use_accurate_framebuffers setting is enabled, perform a more accurate surface copy, CopySurface(old_surface, new_surface, copy_pbo.handle);
// where pixels are reinterpreted as a new format (without conversion). This code path uses }
// OpenGL PBOs and is quite slow. break;
case SurfaceParams::SurfaceTarget::TextureCubemap: {
const u32 byte_stride{old_params.rt.layer_stride *
(SurfaceParams::GetFormatBpp(old_params.pixel_format) / CHAR_BIT)};
auto source_format = GetFormatTuple(old_params.pixel_format, old_params.component_type); // This seems to be used for render-to-cubemap texture
auto dest_format = GetFormatTuple(new_params.pixel_format, new_params.component_type); const std::size_t size_with_mipmaps{new_params.SizeInBytes2DWithMipmap()};
ASSERT_MSG(size_with_mipmaps == byte_stride, "Unexpected");
ASSERT_MSG(old_params.target == SurfaceParams::SurfaceTarget::Texture2D, "Unexpected");
ASSERT_MSG(old_params.pixel_format == new_params.pixel_format, "Unexpected");
ASSERT_MSG(old_params.width == new_params.width, "Unexpected");
ASSERT_MSG(old_params.height == new_params.height, "Unexpected");
ASSERT_MSG(old_params.rt.array_mode == 1, "Unexpected");
ASSERT_MSG(old_params.rt.base_layer == 0, "Unimplemented");
std::size_t buffer_size = for (std::size_t index = 0; index < new_params.depth; ++index) {
std::max(old_params.size_in_bytes_total, new_params.size_in_bytes_total); Surface face_surface{TryGetReservedSurface(old_params)};
ASSERT_MSG(face_surface, "Unexpected");
glBindBuffer(GL_PIXEL_PACK_BUFFER, copy_pbo.handle); if (is_blit) {
glBufferData(GL_PIXEL_PACK_BUFFER, buffer_size, nullptr, GL_STREAM_DRAW_ARB); BlitSurface(face_surface, new_surface, read_framebuffer.handle,
if (source_format.compressed) { draw_framebuffer.handle, face_surface->GetSurfaceParams().rt.index,
glGetCompressedTextureImage(old_surface->Texture().handle, 0, new_params.rt.index, index);
static_cast<GLsizei>(old_params.size_in_bytes_total),
nullptr);
} else { } else {
glGetTextureImage(old_surface->Texture().handle, 0, source_format.format, CopySurface(face_surface, new_surface, copy_pbo.handle,
source_format.type, face_surface->GetSurfaceParams().rt.index, new_params.rt.index, index);
static_cast<GLsizei>(old_params.size_in_bytes_total), nullptr);
}
// If the new texture is bigger than the previous one, we need to fill in the rest with data
// from the CPU.
if (old_params.size_in_bytes_total < new_params.size_in_bytes_total) {
// Upload the rest of the memory.
if (new_params.is_tiled) {
// TODO(Subv): We might have to de-tile the subtexture and re-tile it with the rest
// of the data in this case. Games like Super Mario Odyssey seem to hit this case
// when drawing, it re-uses the memory of a previous texture as a bigger framebuffer
// but it doesn't clear it beforehand, the texture is already full of zeros.
LOG_DEBUG(HW_GPU, "Trying to upload extra texture data from the CPU during "
"reinterpretation but the texture is tiled.");
}
std::size_t remaining_size =
new_params.size_in_bytes_total - old_params.size_in_bytes_total;
std::vector<u8> data(remaining_size);
Memory::ReadBlock(new_params.addr + old_params.size_in_bytes_total, data.data(),
data.size());
glBufferSubData(GL_PIXEL_PACK_BUFFER, old_params.size_in_bytes_total, remaining_size,
data.data());
} }
glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); old_params.addr += size_with_mipmaps;
const auto& dest_rect{new_params.GetRect()};
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, copy_pbo.handle);
if (dest_format.compressed) {
LOG_CRITICAL(HW_GPU, "Compressed copy is unimplemented!");
UNREACHABLE();
} else {
switch (new_params.target) {
case SurfaceParams::SurfaceTarget::Texture1D:
glTextureSubImage1D(new_surface->Texture().handle, 0, 0,
static_cast<GLsizei>(dest_rect.GetWidth()), dest_format.format,
dest_format.type, nullptr);
break;
case SurfaceParams::SurfaceTarget::Texture2D:
glTextureSubImage2D(new_surface->Texture().handle, 0, 0, 0,
static_cast<GLsizei>(dest_rect.GetWidth()),
static_cast<GLsizei>(dest_rect.GetHeight()), dest_format.format,
dest_format.type, nullptr);
break;
case SurfaceParams::SurfaceTarget::Texture3D:
case SurfaceParams::SurfaceTarget::Texture2DArray:
glTextureSubImage3D(new_surface->Texture().handle, 0, 0, 0, 0,
static_cast<GLsizei>(dest_rect.GetWidth()),
static_cast<GLsizei>(dest_rect.GetHeight()),
static_cast<GLsizei>(new_params.depth), dest_format.format,
dest_format.type, nullptr);
break;
case SurfaceParams::SurfaceTarget::TextureCubemap:
for (std::size_t face = 0; face < new_params.depth; ++face) {
glTextureSubImage3D(
new_surface->Texture().handle, 0, 0, 0, static_cast<GLint>(face),
static_cast<GLsizei>(dest_rect.GetWidth()),
static_cast<GLsizei>(dest_rect.GetHeight()), static_cast<GLsizei>(1),
dest_format.format, dest_format.type, nullptr);
} }
break; break;
}
default: default:
LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}", LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}",
static_cast<u32>(new_params.target)); static_cast<u32>(new_params.target));
UNREACHABLE(); UNREACHABLE();
} }
}
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
}
return new_surface; return new_surface;
} }

View file

@ -9,6 +9,7 @@
#include <memory> #include <memory>
#include <vector> #include <vector>
#include "common/alignment.h"
#include "common/common_types.h" #include "common/common_types.h"
#include "common/hash.h" #include "common/hash.h"
#include "common/math_util.h" #include "common/math_util.h"
@ -706,6 +707,29 @@ struct SurfaceParams {
return SizeInBytes2D() * depth; return SizeInBytes2D() * depth;
} }
/**
* Returns the size in bytes of the 2D surface with mipmaps. Each mipmap level proceeds the
* previous with half the width and half the height. Once the size of the next mip reaches 0, we
* are done.
*/
std::size_t SizeInBytes2DWithMipmap() const {
std::size_t size_in_bytes{};
auto mip_params{*this};
for (std::size_t level = 0; level < max_mip_level; level++) {
size_in_bytes += mip_params.SizeInBytes2D();
mip_params.width /= 2;
mip_params.height /= 2;
if (!mip_params.width || !mip_params.height) {
break;
}
}
// TODO(bunnei): This alignup is unverified, but necessary in games tested (e.g. in SMO)
return Common::AlignUp(size_in_bytes, 0x1000);
}
/// Creates SurfaceParams from a texture configuration /// Creates SurfaceParams from a texture configuration
static SurfaceParams CreateForTexture(const Tegra::Texture::FullTextureInfo& config, static SurfaceParams CreateForTexture(const Tegra::Texture::FullTextureInfo& config,
const GLShader::SamplerEntry& entry); const GLShader::SamplerEntry& entry);
@ -738,6 +762,15 @@ struct SurfaceParams {
std::size_t size_in_bytes_total; std::size_t size_in_bytes_total;
std::size_t size_in_bytes_2d; std::size_t size_in_bytes_2d;
SurfaceTarget target; SurfaceTarget target;
u32 max_mip_level;
// Render target specific parameters, not used in caching
struct {
u32 index;
u32 array_mode;
u32 layer_stride;
u32 base_layer;
} rt;
}; };
}; // namespace OpenGL }; // namespace OpenGL
@ -747,6 +780,7 @@ struct SurfaceReserveKey : Common::HashableStruct<OpenGL::SurfaceParams> {
static SurfaceReserveKey Create(const OpenGL::SurfaceParams& params) { static SurfaceReserveKey Create(const OpenGL::SurfaceParams& params) {
SurfaceReserveKey res; SurfaceReserveKey res;
res.state = params; res.state = params;
res.state.rt = {}; // Ignore rt config in caching
return res; return res;
} }
}; };

View file

@ -165,6 +165,8 @@ struct TICEntry {
// High 16 bits of the pitch value // High 16 bits of the pitch value
BitField<0, 16, u32> pitch_high; BitField<0, 16, u32> pitch_high;
BitField<28, 4, u32> max_mip_level;
}; };
union { union {
BitField<0, 16, u32> width_minus_1; BitField<0, 16, u32> width_minus_1;