texture_cache: Handle out of bound texture blits

Some games interleave a texture blit using regions which are out-of-bounds. This addresses the interleaving to avoid oob reads from the src texture.
This commit is contained in:
ameerj 2021-05-07 22:14:21 -04:00
parent 260b841dc3
commit 3671fd0a97
8 changed files with 99 additions and 61 deletions

View file

@ -148,7 +148,9 @@ public:
/// Blit an image with the given parameters
void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const Tegra::Engines::Fermi2D::Surface& src,
const Tegra::Engines::Fermi2D::Config& copy);
const Tegra::Engines::Fermi2D::Config& copy,
std::optional<Region2D> src_region_override = {},
std::optional<Region2D> dst_region_override = {});
/// Invalidate the contents of the color buffer index
/// These contents become unspecified, the cache can assume aggressive optimizations.
@ -615,7 +617,9 @@ void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) {
template <class P>
void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const Tegra::Engines::Fermi2D::Surface& src,
const Tegra::Engines::Fermi2D::Config& copy) {
const Tegra::Engines::Fermi2D::Config& copy,
std::optional<Region2D> src_override,
std::optional<Region2D> dst_override) {
const BlitImages images = GetBlitImages(dst, src);
const ImageId dst_id = images.dst_id;
const ImageId src_id = images.src_id;
@ -631,20 +635,42 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
const ImageViewInfo dst_view_info(ImageViewType::e2D, images.dst_format, dst_range);
const auto [dst_framebuffer_id, dst_view_id] = RenderTargetFromImage(dst_id, dst_view_info);
const auto [src_samples_x, src_samples_y] = SamplesLog2(src_image.info.num_samples);
const std::array src_region{
Offset2D{.x = copy.src_x0 >> src_samples_x, .y = copy.src_y0 >> src_samples_y},
Offset2D{.x = copy.src_x1 >> src_samples_x, .y = copy.src_y1 >> src_samples_y},
// out of bounds texture blit checking
const bool use_override = src_override.has_value();
const s32 src_x0 = copy.src_x0 >> src_samples_x;
s32 src_x1 = use_override ? src_override->end.x : copy.src_x1 >> src_samples_x;
const s32 src_y0 = copy.src_y0 >> src_samples_y;
const s32 src_y1 = copy.src_y1 >> src_samples_y;
const auto src_width = static_cast<s32>(src_image.info.size.width);
const bool width_oob = src_x1 > src_width;
const auto width_diff = width_oob ? src_x1 - src_width : 0;
if (width_oob) {
src_x1 = src_width;
}
const Region2D src_dimensions{
Offset2D{.x = src_x0, .y = src_y0},
Offset2D{.x = src_x1, .y = src_y1},
};
const auto src_region = use_override ? *src_override : src_dimensions;
const std::optional src_base = src_image.TryFindBase(src.Address());
const SubresourceRange src_range{.base = src_base.value(), .extent = {1, 1}};
const ImageViewInfo src_view_info(ImageViewType::e2D, images.src_format, src_range);
const auto [src_framebuffer_id, src_view_id] = RenderTargetFromImage(src_id, src_view_info);
const auto [dst_samples_x, dst_samples_y] = SamplesLog2(dst_image.info.num_samples);
const std::array dst_region{
Offset2D{.x = copy.dst_x0 >> dst_samples_x, .y = copy.dst_y0 >> dst_samples_y},
Offset2D{.x = copy.dst_x1 >> dst_samples_x, .y = copy.dst_y1 >> dst_samples_y},
const s32 dst_x0 = copy.dst_x0 >> dst_samples_x;
const s32 dst_x1 = copy.dst_x1 >> dst_samples_x;
const s32 dst_y0 = copy.dst_y0 >> dst_samples_y;
const s32 dst_y1 = copy.dst_y1 >> dst_samples_y;
const Region2D dst_dimensions{
Offset2D{.x = dst_x0, .y = dst_y0},
Offset2D{.x = dst_x1 - width_diff, .y = dst_y1},
};
const auto dst_region = use_override ? *dst_override : dst_dimensions;
// Always call this after src_framebuffer_id was queried, as the address might be invalidated.
Framebuffer* const dst_framebuffer = &slot_framebuffers[dst_framebuffer_id];
@ -661,6 +687,21 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
runtime.BlitImage(dst_framebuffer, dst_view, src_view, dst_region, src_region, copy.filter,
copy.operation);
}
if (width_oob) {
// Continue copy of the oob region of the texture on the next row
auto oob_src = src;
oob_src.height++;
const Region2D src_region_override{
Offset2D{.x = 0, .y = src_y0 + 1},
Offset2D{.x = width_diff, .y = src_y1 + 1},
};
const Region2D dst_region_override{
Offset2D{.x = dst_x1 - width_diff, .y = dst_y0},
Offset2D{.x = dst_x1, .y = dst_y1},
};
BlitImage(dst, oob_src, copy, src_region_override, dst_region_override);
}
}
template <class P>

View file

@ -64,6 +64,13 @@ struct Offset3D {
s32 z;
};
struct Region2D {
constexpr auto operator<=>(const Region2D&) const noexcept = default;
Offset2D start;
Offset2D end;
};
struct Extent2D {
constexpr auto operator<=>(const Extent2D&) const noexcept = default;