hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 2156

Browse Source
This commit is contained in:
pineappleEA
2021-10-24 06:39:01 +02:00
parent 238cebb24e
commit f85f34d123
119 changed files with 8931 additions and 610 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/video_core/texture_cache/image_base.cpp
View File

@@ -60,15 +60,17 @@ namespace {
ImageBase::ImageBase(const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_)
: info{info_}, guest_size_bytes{CalculateGuestSizeInBytes(info)},
unswizzled_size_bytes{CalculateUnswizzledSizeBytes(info)},
converted_size_bytes{CalculateConvertedSizeBytes(info)}, gpu_addr{gpu_addr_},
cpu_addr{cpu_addr_}, cpu_addr_end{cpu_addr + guest_size_bytes},
mip_level_offsets{CalculateMipLevelOffsets(info)} {
converted_size_bytes{CalculateConvertedSizeBytes(info)}, scale_rating{}, scale_tick{},
has_scaled{}, gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_},
cpu_addr_end{cpu_addr + guest_size_bytes}, mip_level_offsets{CalculateMipLevelOffsets(info)} {
if (info.type == ImageType::e3D) {
slice_offsets = CalculateSliceOffsets(info);
slice_subresources = CalculateSliceSubresources(info);
}
}
ImageBase::ImageBase(const NullImageParams&) {}
ImageMapView::ImageMapView(GPUVAddr gpu_addr_, VAddr cpu_addr_, size_t size_, ImageId image_id_)
: gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_}, size{size_}, image_id{image_id_} {}
@@ -254,6 +256,8 @@ void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_i
}
lhs.aliased_images.push_back(std::move(lhs_alias));
rhs.aliased_images.push_back(std::move(rhs_alias));
lhs.flags &= ~ImageFlagBits::IsRescalable;
rhs.flags &= ~ImageFlagBits::IsRescalable;
}
} // namespace VideoCommon

16
src/video_core/texture_cache/image_base.h
View File

@@ -33,6 +33,12 @@ enum class ImageFlagBits : u32 {
///< garbage collection priority
Alias = 1 << 11, ///< This image has aliases and has priority on garbage
///< collection
// Rescaler
Rescaled = 1 << 12,
CheckingRescalable = 1 << 13,
IsRescalable = 1 << 14,
Blacklisted = 1 << 15,
};
DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits)
@@ -43,8 +49,11 @@ struct AliasedImage {
ImageId id;
};
struct NullImageParams {};
struct ImageBase {
explicit ImageBase(const ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr);
explicit ImageBase(const NullImageParams&);
[[nodiscard]] std::optional<SubresourceBase> TryFindBase(GPUVAddr other_addr) const noexcept;
@@ -68,11 +77,18 @@ struct ImageBase {
void CheckBadOverlapState();
void CheckAliasState();
bool HasScaled() {
return has_scaled;
}
ImageInfo info;
u32 guest_size_bytes = 0;
u32 unswizzled_size_bytes = 0;
u32 converted_size_bytes = 0;
u32 scale_rating = 0;
u64 scale_tick = 0;
bool has_scaled = false;
ImageFlagBits flags = ImageFlagBits::CpuModified;
GPUVAddr gpu_addr = 0;

18
src/video_core/texture_cache/image_info.cpp
View File

@@ -31,6 +31,7 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept {
.depth = config.block_depth,
};
}
rescaleable = false;
tile_width_spacing = config.tile_width_spacing;
if (config.texture_type != TextureType::Texture2D &&
config.texture_type != TextureType::Texture2DNoMipmap) {
@@ -41,6 +42,7 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept {
ASSERT(config.BaseLayer() == 0);
type = ImageType::e1D;
size.width = config.Width();
resources.layers = 1;
break;
case TextureType::Texture1DArray:
UNIMPLEMENTED_IF(config.BaseLayer() != 0);
@@ -52,12 +54,14 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept {
case TextureType::Texture2DNoMipmap:
ASSERT(config.Depth() == 1);
type = config.IsPitchLinear() ? ImageType::Linear : ImageType::e2D;
rescaleable = !config.IsPitchLinear();
size.width = config.Width();
size.height = config.Height();
resources.layers = config.BaseLayer() + 1;
break;
case TextureType::Texture2DArray:
type = ImageType::e2D;
rescaleable = true;
size.width = config.Width();
size.height = config.Height();
resources.layers = config.BaseLayer() + config.Depth();
@@ -82,10 +86,12 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept {
size.width = config.Width();
size.height = config.Height();
size.depth = config.Depth();
resources.layers = 1;
break;
case TextureType::Texture1DBuffer:
type = ImageType::Buffer;
size.width = config.Width();
resources.layers = 1;
break;
default:
UNREACHABLE_MSG("Invalid texture_type={}", static_cast<int>(config.texture_type.Value()));
@@ -95,12 +101,15 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept {
// FIXME: Call this without passing *this
layer_stride = CalculateLayerStride(*this);
maybe_unaligned_layer_stride = CalculateLayerSize(*this);
rescaleable &= (block.depth == 0) && resources.levels == 1;
downscaleable = size.height > 512;
}
}
ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index) noexcept {
const auto& rt = regs.rt[index];
format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(rt.format);
rescaleable = false;
if (rt.tile_mode.is_pitch_linear) {
ASSERT(rt.tile_mode.is_3d == 0);
type = ImageType::Linear;
@@ -126,6 +135,8 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index)
type = ImageType::e3D;
size.depth = rt.depth;
} else {
rescaleable = block.depth == 0 && size.height > 256;
downscaleable = size.height > 512;
type = ImageType::e2D;
resources.layers = rt.depth;
}
@@ -135,6 +146,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept {
format = VideoCore::Surface::PixelFormatFromDepthFormat(regs.zeta.format);
size.width = regs.zeta_width;
size.height = regs.zeta_height;
rescaleable = false;
resources.levels = 1;
layer_stride = regs.zeta.layer_stride * 4;
maybe_unaligned_layer_stride = layer_stride;
@@ -153,6 +165,8 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept {
type = ImageType::e3D;
size.depth = regs.zeta_depth;
} else {
rescaleable = block.depth == 0 && size.height > 256;
downscaleable = size.height > 512;
type = ImageType::e2D;
resources.layers = regs.zeta_depth;
}
@@ -161,6 +175,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept {
ImageInfo::ImageInfo(const Tegra::Engines::Fermi2D::Surface& config) noexcept {
UNIMPLEMENTED_IF_MSG(config.layer != 0, "Surface layer is not zero");
format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(config.format);
rescaleable = false;
if (config.linear == Tegra::Engines::Fermi2D::MemoryLayout::Pitch) {
type = ImageType::Linear;
size = Extent3D{
@@ -171,6 +186,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Fermi2D::Surface& config) noexcept {
pitch = config.pitch;
} else {
type = config.block_depth > 0 ? ImageType::e3D : ImageType::e2D;
block = Extent3D{
.width = config.block_width,
.height = config.block_height,
@@ -183,6 +199,8 @@ ImageInfo::ImageInfo(const Tegra::Engines::Fermi2D::Surface& config) noexcept {
.height = config.height,
.depth = 1,
};
rescaleable = block.depth == 0 && size.height > 256;
downscaleable = size.height > 512;
}
}

4
src/video_core/texture_cache/image_info.h
View File

@@ -15,7 +15,7 @@ using Tegra::Texture::TICEntry;
using VideoCore::Surface::PixelFormat;
struct ImageInfo {
explicit ImageInfo() = default;
ImageInfo() = default;
explicit ImageInfo(const TICEntry& config) noexcept;
explicit ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index) noexcept;
explicit ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept;
@@ -33,6 +33,8 @@ struct ImageInfo {
u32 maybe_unaligned_layer_stride = 0;
u32 num_samples = 1;
u32 tile_width_spacing = 0;
bool rescaleable = false;
bool downscaleable = false;
};
} // namespace VideoCommon

13
src/video_core/texture_cache/image_view_base.cpp
View File

@@ -37,14 +37,15 @@ ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_i
}
ImageViewBase::ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info)
: format{info.format}, type{ImageViewType::Buffer}, size{
.width = info.size.width,
.height = 1,
.depth = 1,
} {
: image_id{NULL_IMAGE_ID}, format{info.format}, type{ImageViewType::Buffer},
size{
.width = info.size.width,
.height = 1,
.depth = 1,
} {
ASSERT_MSG(view_info.type == ImageViewType::Buffer, "Expected texture buffer");
}
ImageViewBase::ImageViewBase(const NullImageParams&) {}
ImageViewBase::ImageViewBase(const NullImageViewParams&) : image_id{NULL_IMAGE_ID} {}
} // namespace VideoCommon

4
src/video_core/texture_cache/image_view_base.h
View File

@@ -15,7 +15,7 @@ using VideoCore::Surface::PixelFormat;
struct ImageViewInfo;
struct ImageInfo;
struct NullImageParams {};
struct NullImageViewParams {};
enum class ImageViewFlagBits : u16 {
PreemtiveDownload = 1 << 0,
@@ -28,7 +28,7 @@ struct ImageViewBase {
explicit ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_info,
ImageId image_id);
explicit ImageViewBase(const ImageInfo& info, const ImageViewInfo& view_info);
explicit ImageViewBase(const NullImageParams&);
explicit ImageViewBase(const NullImageViewParams&);
[[nodiscard]] bool IsBuffer() const noexcept {
return type == ImageViewType::Buffer;

497
src/video_core/texture_cache/texture_cache.h
View File

@@ -7,6 +7,7 @@
#include <unordered_set>
#include "common/alignment.h"
#include "common/settings.h"
#include "video_core/dirty_flags.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/texture_cache/image_view_base.h"
@@ -44,21 +45,22 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
// Make sure the first index is reserved for the null resources
// This way the null resource becomes a compile time constant
void(slot_image_views.insert(runtime, NullImageParams{}));
void(slot_images.insert(NullImageParams{}));
void(slot_image_views.insert(runtime, NullImageViewParams{}));
void(slot_samplers.insert(runtime, sampler_descriptor));
if constexpr (HAS_DEVICE_MEMORY_INFO) {
const auto device_memory = runtime.GetDeviceLocalMemory();
const u64 possible_expected_memory = (device_memory * 3) / 10;
const u64 possible_critical_memory = (device_memory * 6) / 10;
const u64 possible_expected_memory = (device_memory * 4) / 10;
const u64 possible_critical_memory = (device_memory * 7) / 10;
expected_memory = std::max(possible_expected_memory, DEFAULT_EXPECTED_MEMORY);
critical_memory = std::max(possible_critical_memory, DEFAULT_CRITICAL_MEMORY);
minimum_memory = 0;
} else {
// on OGL we can be more conservatives as the driver takes care.
// On OpenGL we can be more conservatives as the driver takes care.
expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB;
critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB;
minimum_memory = expected_memory;
minimum_memory = 0;
}
}
@@ -67,7 +69,7 @@ void TextureCache<P>::RunGarbageCollector() {
const bool high_priority_mode = total_used_memory >= expected_memory;
const bool aggressive_mode = total_used_memory >= critical_memory;
const u64 ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 100ULL;
size_t num_iterations = aggressive_mode ? 10000 : (high_priority_mode ? 100 : 5);
size_t num_iterations = aggressive_mode ? 300 : (high_priority_mode ? 50 : 10);
const auto clean_up = [this, &num_iterations, high_priority_mode](ImageId image_id) {
if (num_iterations == 0) {
return true;
@@ -89,7 +91,7 @@ void TextureCache<P>::RunGarbageCollector() {
UntrackImage(image, image_id);
}
UnregisterImage(image_id);
DeleteImage(image_id);
DeleteImage(image_id, image.scale_tick > frame_tick + 5);
return false;
};
lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up);
@@ -103,6 +105,7 @@ void TextureCache<P>::TickFrame() {
sentenced_images.Tick();
sentenced_framebuffers.Tick();
sentenced_image_view.Tick();
runtime.TickFrame();
++frame_tick;
}
@@ -122,15 +125,14 @@ void TextureCache<P>::MarkModification(ImageId id) noexcept {
}
template <class P>
void TextureCache<P>::FillGraphicsImageViews(std::span<const u32> indices,
std::span<ImageViewId> image_view_ids) {
FillImageViews(graphics_image_table, graphics_image_view_ids, indices, image_view_ids);
template <bool has_blacklists>
void TextureCache<P>::FillGraphicsImageViews(std::span<ImageViewInOut> views) {
FillImageViews<has_blacklists>(graphics_image_table, graphics_image_view_ids, views);
}
template <class P>
void TextureCache<P>::FillComputeImageViews(std::span<const u32> indices,
std::span<ImageViewId> image_view_ids) {
FillImageViews(compute_image_table, compute_image_view_ids, indices, image_view_ids);
void TextureCache<P>::FillComputeImageViews(std::span<ImageViewInOut> views) {
FillImageViews<true>(compute_image_table, compute_image_view_ids, views);
}
template <class P>
@@ -202,24 +204,109 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) {
PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id));
return;
}
flags[Dirty::RenderTargets] = false;
// Render target control is used on all render targets, so force look ups when this one is up
const bool force = flags[Dirty::RenderTargetControl];
flags[Dirty::RenderTargetControl] = false;
u32 scale_rating = 0;
bool rescaled = false;
std::array<ImageId, NUM_RT> tmp_color_images{};
ImageId tmp_depth_image{};
do {
flags[Dirty::RenderTargets] = false;
has_deleted_images = false;
// Render target control is used on all render targets, so force look ups when this one is
// up
const bool force = flags[Dirty::RenderTargetControl];
flags[Dirty::RenderTargetControl] = false;
scale_rating = 0;
bool any_rescaled = false;
bool can_rescale = true;
const auto check_rescale = [&](ImageViewId view_id, ImageId& id_save) {
if (view_id != NULL_IMAGE_VIEW_ID && view_id != ImageViewId{}) {
const auto& view = slot_image_views[view_id];
const auto image_id = view.image_id;
id_save = image_id;
auto& image = slot_images[image_id];
can_rescale &= ImageCanRescale(image);
any_rescaled |= True(image.flags & ImageFlagBits::Rescaled) ||
GetFormatType(image.info.format) != SurfaceType::ColorTexture;
scale_rating = std::max<u32>(scale_rating, image.scale_tick <= frame_tick
? image.scale_rating + 1U
: image.scale_rating);
} else {
id_save = CORRUPT_ID;
}
};
for (size_t index = 0; index < NUM_RT; ++index) {
ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index];
if (flags[Dirty::ColorBuffer0 + index] || force) {
flags[Dirty::ColorBuffer0 + index] = false;
BindRenderTarget(&color_buffer_id, FindColorBuffer(index, is_clear));
}
check_rescale(color_buffer_id, tmp_color_images[index]);
}
if (flags[Dirty::ZetaBuffer] || force) {
flags[Dirty::ZetaBuffer] = false;
BindRenderTarget(&render_targets.depth_buffer_id, FindDepthBuffer(is_clear));
}
check_rescale(render_targets.depth_buffer_id, tmp_depth_image);
if (can_rescale) {
rescaled = any_rescaled || scale_rating >= 2;
const auto scale_up = [this](ImageId image_id) {
if (image_id != CORRUPT_ID) {
Image& image = slot_images[image_id];
ScaleUp(image);
}
};
if (rescaled) {
for (size_t index = 0; index < NUM_RT; ++index) {
scale_up(tmp_color_images[index]);
}
scale_up(tmp_depth_image);
scale_rating = 2;
}
} else {
rescaled = false;
const auto scale_down = [this](ImageId image_id) {
if (image_id != CORRUPT_ID) {
Image& image = slot_images[image_id];
ScaleDown(image);
}
};
for (size_t index = 0; index < NUM_RT; ++index) {
scale_down(tmp_color_images[index]);
}
scale_down(tmp_depth_image);
scale_rating = 1;
}
} while (has_deleted_images);
// Rescale End
const auto set_rating = [this, scale_rating](ImageId image_id) {
if (image_id != CORRUPT_ID) {
Image& image = slot_images[image_id];
image.scale_rating = scale_rating;
if (image.scale_tick <= frame_tick) {
image.scale_tick = frame_tick + 1;
}
}
};
for (size_t index = 0; index < NUM_RT; ++index) {
set_rating(tmp_color_images[index]);
}
set_rating(tmp_depth_image);
if (is_rescaling != rescaled) {
flags[Dirty::RescaleViewports] = true;
flags[Dirty::RescaleScissors] = true;
is_rescaling = rescaled;
}
for (size_t index = 0; index < NUM_RT; ++index) {
ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index];
if (flags[Dirty::ColorBuffer0 + index] || force) {
flags[Dirty::ColorBuffer0 + index] = false;
BindRenderTarget(&color_buffer_id, FindColorBuffer(index, is_clear));
}
PrepareImageView(color_buffer_id, true, is_clear && IsFullClear(color_buffer_id));
}
if (flags[Dirty::ZetaBuffer] || force) {
flags[Dirty::ZetaBuffer] = false;
BindRenderTarget(&render_targets.depth_buffer_id, FindDepthBuffer(is_clear));
}
const ImageViewId depth_buffer_id = render_targets.depth_buffer_id;
PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id));
@@ -227,9 +314,15 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) {
for (size_t index = 0; index < NUM_RT; ++index) {
render_targets.draw_buffers[index] = static_cast<u8>(maxwell3d.regs.rt_control.Map(index));
}
u32 up_scale = 1;
u32 down_shift = 0;
if (is_rescaling) {
up_scale = Settings::values.resolution_info.up_scale;
down_shift = Settings::values.resolution_info.down_shift;
}
render_targets.size = Extent2D{
maxwell3d.regs.render_area.width,
maxwell3d.regs.render_area.height,
(maxwell3d.regs.render_area.width * up_scale) >> down_shift,
(maxwell3d.regs.render_area.height * up_scale) >> down_shift,
};
flags[Dirty::DepthBiasGlobal] = true;
@@ -241,17 +334,29 @@ typename P::Framebuffer* TextureCache<P>::GetFramebuffer() {
}
template <class P>
template <bool has_blacklists>
void TextureCache<P>::FillImageViews(DescriptorTable<TICEntry>& table,
std::span<ImageViewId> cached_image_view_ids,
std::span<const u32> indices,
std::span<ImageViewId> image_view_ids) {
ASSERT(indices.size() <= image_view_ids.size());
std::span<ImageViewInOut> views) {
bool has_blacklisted;
do {
has_deleted_images = false;
std::ranges::transform(indices, image_view_ids.begin(), [&](u32 index) {
return VisitImageView(table, cached_image_view_ids, index);
});
} while (has_deleted_images);
if constexpr (has_blacklists) {
has_blacklisted = false;
}
for (ImageViewInOut& view : views) {
view.id = VisitImageView(table, cached_image_view_ids, view.index);
if constexpr (has_blacklists) {
if (view.blacklist && view.id != NULL_IMAGE_VIEW_ID) {
const ImageViewBase& image_view{slot_image_views[view.id]};
auto& image = slot_images[image_view.image_id];
image.flags |= ImageFlagBits::Blacklisted;
has_blacklisted |= ScaleDown(image);
image.scale_rating = 0;
}
}
}
} while (has_deleted_images || (has_blacklists && has_blacklisted));
}
template <class P>
@@ -369,8 +474,43 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
PrepareImage(src_id, false, false);
PrepareImage(dst_id, true, false);
ImageBase& dst_image = slot_images[dst_id];
const ImageBase& src_image = slot_images[src_id];
Image& dst_image = slot_images[dst_id];
Image& src_image = slot_images[src_id];
bool is_src_rescaled = True(src_image.flags & ImageFlagBits::Rescaled);
bool is_dst_rescaled = True(dst_image.flags & ImageFlagBits::Rescaled);
const bool is_resolve = src_image.info.num_samples != 1 && dst_image.info.num_samples == 1;
if (is_src_rescaled != is_dst_rescaled) {
if (ImageCanRescale(src_image)) {
ScaleUp(src_image);
is_src_rescaled = True(src_image.flags & ImageFlagBits::Rescaled);
if (is_resolve) {
dst_image.info.rescaleable = true;
for (const auto& alias : dst_image.aliased_images) {
Image& other_image = slot_images[alias.id];
other_image.info.rescaleable = true;
}
}
}
if (ImageCanRescale(dst_image)) {
ScaleUp(dst_image);
is_dst_rescaled = True(dst_image.flags & ImageFlagBits::Rescaled);
}
}
if (is_resolve && (is_src_rescaled != is_dst_rescaled)) {
// A resolve requires both images to be the same dimensions. Resize down if needed.
ScaleDown(src_image);
ScaleDown(dst_image);
is_src_rescaled = True(src_image.flags & ImageFlagBits::Rescaled);
is_dst_rescaled = True(dst_image.flags & ImageFlagBits::Rescaled);
}
const auto& resolution = Settings::values.resolution_info;
const auto scale_region = [&](Region2D& region) {
region.start.x = resolution.ScaleUp(region.start.x);
region.start.y = resolution.ScaleUp(region.start.y);
region.end.x = resolution.ScaleUp(region.end.x);
region.end.y = resolution.ScaleUp(region.end.y);
};
// TODO: Deduplicate
const std::optional src_base = src_image.TryFindBase(src.Address());
@@ -378,20 +518,26 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
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 [src_samples_x, src_samples_y] = SamplesLog2(src_image.info.num_samples);
const Region2D src_region{
Region2D 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},
};
if (is_src_rescaled) {
scale_region(src_region);
}
const std::optional dst_base = dst_image.TryFindBase(dst.Address());
const SubresourceRange dst_range{.base = dst_base.value(), .extent = {1, 1}};
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 [dst_samples_x, dst_samples_y] = SamplesLog2(dst_image.info.num_samples);
const Region2D dst_region{
Region2D 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},
};
if (is_dst_rescaled) {
scale_region(dst_region);
}
// Always call this after src_framebuffer_id was queried, as the address might be invalidated.
Framebuffer* const dst_framebuffer = &slot_framebuffers[dst_framebuffer_id];
@@ -486,6 +632,20 @@ void TextureCache<P>::PopAsyncFlushes() {
committed_downloads.pop();
}
template <class P>
bool TextureCache<P>::IsRescaling() const noexcept {
return is_rescaling;
}
template <class P>
bool TextureCache<P>::IsRescaling(const ImageViewBase& image_view) const noexcept {
if (image_view.type == ImageViewType::Buffer) {
return false;
}
const ImageBase& image = slot_images[image_view.image_id];
return True(image.flags & ImageFlagBits::Rescaled);
}
template <class P>
bool TextureCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
bool is_modified = false;
@@ -623,6 +783,105 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
return image_id;
}
template <class P>
bool TextureCache<P>::ImageCanRescale(ImageBase& image) {
if (!image.info.rescaleable) {
return false;
}
if (Settings::values.resolution_info.downscale && !image.info.downscaleable) {
return false;
}
if (True(image.flags & (ImageFlagBits::Rescaled | ImageFlagBits::CheckingRescalable))) {
return true;
}
if (True(image.flags & ImageFlagBits::IsRescalable)) {
return true;
}
image.flags |= ImageFlagBits::CheckingRescalable;
for (const auto& alias : image.aliased_images) {
Image& other_image = slot_images[alias.id];
if (!ImageCanRescale(other_image)) {
image.flags &= ~ImageFlagBits::CheckingRescalable;
return false;
}
}
image.flags &= ~ImageFlagBits::CheckingRescalable;
image.flags |= ImageFlagBits::IsRescalable;
return true;
}
template <class P>
void TextureCache<P>::InvalidateScale(Image& image) {
if (image.scale_tick <= frame_tick) {
image.scale_tick = frame_tick + 1;
}
const std::span<const ImageViewId> image_view_ids = image.image_view_ids;
auto& dirty = maxwell3d.dirty.flags;
dirty[Dirty::RenderTargets] = true;
dirty[Dirty::ZetaBuffer] = true;
for (size_t rt = 0; rt < NUM_RT; ++rt) {
dirty[Dirty::ColorBuffer0 + rt] = true;
}
for (const ImageViewId image_view_id : image_view_ids) {
std::ranges::replace(render_targets.color_buffer_ids, image_view_id, ImageViewId{});
if (render_targets.depth_buffer_id == image_view_id) {
render_targets.depth_buffer_id = ImageViewId{};
}
}
RemoveImageViewReferences(image_view_ids);
RemoveFramebuffers(image_view_ids);
for (const ImageViewId image_view_id : image_view_ids) {
sentenced_image_view.Push(std::move(slot_image_views[image_view_id]));
slot_image_views.erase(image_view_id);
}
image.image_view_ids.clear();
image.image_view_infos.clear();
if constexpr (ENABLE_VALIDATION) {
std::ranges::fill(graphics_image_view_ids, CORRUPT_ID);
std::ranges::fill(compute_image_view_ids, CORRUPT_ID);
}
graphics_image_table.Invalidate();
compute_image_table.Invalidate();
has_deleted_images = true;
}
template <class P>
u64 TextureCache<P>::GetScaledImageSizeBytes(ImageBase& image) {
const u64 scale_up = static_cast<u64>(Settings::values.resolution_info.up_scale *
Settings::values.resolution_info.up_scale);
const u64 down_shift = static_cast<u64>(Settings::values.resolution_info.down_shift +
Settings::values.resolution_info.down_shift);
const u64 image_size_bytes =
static_cast<u64>(std::max(image.guest_size_bytes, image.unswizzled_size_bytes));
const u64 tentative_size = (image_size_bytes * scale_up) >> down_shift;
const u64 fitted_size = Common::AlignUp(tentative_size, 1024);
return fitted_size;
}
template <class P>
bool TextureCache<P>::ScaleUp(Image& image) {
const bool has_copy = image.HasScaled();
const bool rescaled = image.ScaleUp();
if (!rescaled) {
return false;
}
if (!has_copy) {
total_used_memory += GetScaledImageSizeBytes(image);
}
InvalidateScale(image);
return true;
}
template <class P>
bool TextureCache<P>::ScaleDown(Image& image) {
const bool rescaled = image.ScaleDown();
if (!rescaled) {
return false;
}
InvalidateScale(image);
return true;
}
template <class P>
ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
RelaxedOptions options) {
@@ -660,12 +919,18 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
std::vector<ImageId> right_aliased_ids;
std::unordered_set<ImageId> ignore_textures;
std::vector<ImageId> bad_overlap_ids;
std::vector<ImageId> all_siblings;
const bool this_is_linear = info.type == ImageType::Linear;
const auto region_check = [&](ImageId overlap_id, ImageBase& overlap) {
if (True(overlap.flags & ImageFlagBits::Remapped)) {
ignore_textures.insert(overlap_id);
return;
}
if (info.type == ImageType::Linear) {
const bool overlap_is_linear = overlap.info.type == ImageType::Linear;
if (this_is_linear != overlap_is_linear) {
return;
}
if (this_is_linear && overlap_is_linear) {
if (info.pitch == overlap.info.pitch && gpu_addr == overlap.gpu_addr) {
// Alias linear images with the same pitch
left_aliased_ids.push_back(overlap_id);
@@ -681,6 +946,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
cpu_addr = solution->cpu_addr;
new_info.resources = solution->resources;
overlap_ids.push_back(overlap_id);
all_siblings.push_back(overlap_id);
return;
}
static constexpr auto options = RelaxedOptions::Size | RelaxedOptions::Format;
@@ -688,10 +954,12 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
if (IsSubresource(new_info, overlap, gpu_addr, options, broken_views, native_bgr)) {
left_aliased_ids.push_back(overlap_id);
overlap.flags |= ImageFlagBits::Alias;
all_siblings.push_back(overlap_id);
} else if (IsSubresource(overlap.info, new_image_base, overlap.gpu_addr, options,
broken_views, native_bgr)) {
right_aliased_ids.push_back(overlap_id);
overlap.flags |= ImageFlagBits::Alias;
all_siblings.push_back(overlap_id);
} else {
bad_overlap_ids.push_back(overlap_id);
overlap.flags |= ImageFlagBits::BadOverlap;
@@ -709,6 +977,37 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
}
};
ForEachSparseImageInRegion(gpu_addr, size_bytes, region_check_gpu);
bool can_rescale = info.rescaleable;
bool any_rescaled = false;
bool any_blacklisted = false;
for (const ImageId sibling_id : all_siblings) {
if (!can_rescale) {
break;
}
Image& sibling = slot_images[sibling_id];
can_rescale &= ImageCanRescale(sibling);
any_rescaled |= True(sibling.flags & ImageFlagBits::Rescaled);
any_blacklisted |= True(sibling.flags & ImageFlagBits::Blacklisted);
}
can_rescale &= any_rescaled;
if (can_rescale) {
for (const ImageId sibling_id : all_siblings) {
Image& sibling = slot_images[sibling_id];
ScaleUp(sibling);
}
} else {
for (const ImageId sibling_id : all_siblings) {
Image& sibling = slot_images[sibling_id];
ScaleDown(sibling);
if (any_blacklisted) {
sibling.flags |= ImageFlagBits::Blacklisted;
}
}
}
const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr);
Image& new_image = slot_images[new_image_id];
@@ -731,14 +1030,23 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
// TODO: Only upload what we need
RefreshContents(new_image, new_image_id);
if (can_rescale) {
ScaleUp(new_image);
} else {
ScaleDown(new_image);
}
for (const ImageId overlap_id : overlap_ids) {
Image& overlap = slot_images[overlap_id];
if (overlap.info.num_samples != new_image.info.num_samples) {
LOG_WARNING(HW_GPU, "Copying between images with different samples is not implemented");
} else {
const auto& resolution = Settings::values.resolution_info;
const SubresourceBase base = new_image.TryFindBase(overlap.gpu_addr).value();
const auto copies = MakeShrinkImageCopies(new_info, overlap.info, base);
runtime.CopyImage(new_image, overlap, copies);
const u32 up_scale = can_rescale ? resolution.up_scale : 1;
const u32 down_shift = can_rescale ? resolution.down_shift : 0;
auto copies = MakeShrinkImageCopies(new_info, overlap.info, base, up_scale, down_shift);
runtime.CopyImage(new_image, overlap, std::move(copies));
}
if (True(overlap.flags & ImageFlagBits::Tracked)) {
UntrackImage(overlap, overlap_id);
@@ -1083,13 +1391,6 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
"Trying to unregister an already registered image");
image.flags &= ~ImageFlagBits::Registered;
image.flags &= ~ImageFlagBits::BadOverlap;
u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
if ((IsPixelFormatASTC(image.info.format) &&
True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
True(image.flags & ImageFlagBits::Converted)) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
}
total_used_memory -= Common::AlignUp(tentative_size, 1024);
lru_cache.Free(image.lru_index);
const auto& clear_page_table =
[this, image_id](
@@ -1213,8 +1514,18 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
}
template <class P>
void TextureCache<P>::DeleteImage(ImageId image_id) {
void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
ImageBase& image = slot_images[image_id];
if (image.HasScaled()) {
total_used_memory -= GetScaledImageSizeBytes(image);
}
u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
if ((IsPixelFormatASTC(image.info.format) &&
True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
True(image.flags & ImageFlagBits::Converted)) {
tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
}
total_used_memory -= Common::AlignUp(tentative_size, 1024);
const GPUVAddr gpu_addr = image.gpu_addr;
const auto alloc_it = image_allocs_table.find(gpu_addr);
if (alloc_it == image_allocs_table.end()) {
@@ -1269,10 +1580,14 @@ void TextureCache<P>::DeleteImage(ImageId image_id) {
num_removed_overlaps);
}
for (const ImageViewId image_view_id : image_view_ids) {
sentenced_image_view.Push(std::move(slot_image_views[image_view_id]));
if (!immediate_delete) {
sentenced_image_view.Push(std::move(slot_image_views[image_view_id]));
}
slot_image_views.erase(image_view_id);
}
sentenced_images.Push(std::move(slot_images[image_id]));
if (!immediate_delete) {
sentenced_images.Push(std::move(slot_images[image_id]));
}
slot_images.erase(image_id);
alloc_images.erase(alloc_image_it);
@@ -1322,26 +1637,68 @@ void TextureCache<P>::MarkModification(ImageBase& image) noexcept {
template <class P>
void TextureCache<P>::SynchronizeAliases(ImageId image_id) {
boost::container::small_vector<const AliasedImage*, 1> aliased_images;
ImageBase& image = slot_images[image_id];
Image& image = slot_images[image_id];
bool any_rescaled = True(image.flags & ImageFlagBits::Rescaled);
bool any_blacklisted = True(image.flags & ImageFlagBits::Blacklisted);
u64 most_recent_tick = image.modification_tick;
for (const AliasedImage& aliased : image.aliased_images) {
ImageBase& aliased_image = slot_images[aliased.id];
if (image.modification_tick < aliased_image.modification_tick) {
most_recent_tick = std::max(most_recent_tick, aliased_image.modification_tick);
aliased_images.push_back(&aliased);
any_rescaled |= True(aliased_image.flags & ImageFlagBits::Rescaled);
any_blacklisted |= True(aliased_image.flags & ImageFlagBits::Blacklisted);
}
}
if (aliased_images.empty()) {
return;
}
const bool can_rescale = ImageCanRescale(image);
if (any_rescaled) {
if (can_rescale) {
ScaleUp(image);
} else {
ScaleDown(image);
if (any_blacklisted) {
image.flags |= ImageFlagBits::Blacklisted;
}
}
}
image.modification_tick = most_recent_tick;
std::ranges::sort(aliased_images, [this](const AliasedImage* lhs, const AliasedImage* rhs) {
const ImageBase& lhs_image = slot_images[lhs->id];
const ImageBase& rhs_image = slot_images[rhs->id];
return lhs_image.modification_tick < rhs_image.modification_tick;
});
const auto& resolution = Settings::values.resolution_info;
for (const AliasedImage* const aliased : aliased_images) {
CopyImage(image_id, aliased->id, aliased->copies);
if (!resolution.active | !any_rescaled) {
CopyImage(image_id, aliased->id, aliased->copies);
continue;
}
Image& aliased_image = slot_images[aliased->id];
if (!can_rescale) {
ScaleDown(aliased_image);
if (any_blacklisted) {
aliased_image.flags |= ImageFlagBits::Blacklisted;
}
CopyImage(image_id, aliased->id, aliased->copies);
continue;
}
ScaleUp(aliased_image);
const bool both_2d{image.info.type == ImageType::e2D &&
aliased_image.info.type == ImageType::e2D};
auto copies = aliased->copies;
for (auto copy : copies) {
copy.extent.width = std::max<u32>(
(copy.extent.width * resolution.up_scale) >> resolution.down_shift, 1);
if (both_2d) {
copy.extent.height = std::max<u32>(
(copy.extent.height * resolution.up_scale) >> resolution.down_shift, 1);
}
}
CopyImage(image_id, aliased->id, copies);
}
}
@@ -1377,9 +1734,25 @@ void TextureCache<P>::PrepareImageView(ImageViewId image_view_id, bool is_modifi
}
template <class P>
void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::span<const ImageCopy> copies) {
void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::vector<ImageCopy> copies) {
Image& dst = slot_images[dst_id];
Image& src = slot_images[src_id];
const bool is_rescaled = True(src.flags & ImageFlagBits::Rescaled);
if (is_rescaled) {
ASSERT(True(dst.flags & ImageFlagBits::Rescaled));
const bool both_2d{src.info.type == ImageType::e2D && dst.info.type == ImageType::e2D};
const auto& resolution = Settings::values.resolution_info;
for (auto& copy : copies) {
copy.src_offset.x = resolution.ScaleUp(copy.src_offset.x);
copy.dst_offset.x = resolution.ScaleUp(copy.dst_offset.x);
copy.extent.width = resolution.ScaleUp(copy.extent.width);
if (both_2d) {
copy.src_offset.y = resolution.ScaleUp(copy.src_offset.y);
copy.dst_offset.y = resolution.ScaleUp(copy.dst_offset.y);
copy.extent.height = resolution.ScaleUp(copy.extent.height);
}
}
}
const auto dst_format_type = GetFormatType(dst.info.format);
const auto src_format_type = GetFormatType(src.info.format);
if (src_format_type == dst_format_type) {
@@ -1424,7 +1797,7 @@ void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::span<const
};
UNIMPLEMENTED_IF(copy.extent != expected_size);
runtime.ConvertImage(dst_framebuffer, dst_view, src_view);
runtime.ConvertImage(dst_framebuffer, dst_view, src_view, is_rescaled);
}
}
@@ -1433,8 +1806,8 @@ void TextureCache<P>::BindRenderTarget(ImageViewId* old_id, ImageViewId new_id)
if (*old_id == new_id) {
return;
}
if (*old_id) {
const ImageViewBase& old_view = slot_image_views[*old_id];
if (new_id) {
const ImageViewBase& old_view = slot_image_views[new_id];
if (True(old_view.flags & ImageViewFlagBits::PreemtiveDownload)) {
uncommitted_downloads.push_back(old_view.image_id);
}
@@ -1447,10 +1820,18 @@ std::pair<FramebufferId, ImageViewId> TextureCache<P>::RenderTargetFromImage(
ImageId image_id, const ImageViewInfo& view_info) {
const ImageViewId view_id = FindOrEmplaceImageView(image_id, view_info);
const ImageBase& image = slot_images[image_id];
const bool is_rescaled = True(image.flags & ImageFlagBits::Rescaled);
const bool is_color = GetFormatType(image.info.format) == SurfaceType::ColorTexture;
const ImageViewId color_view_id = is_color ? view_id : ImageViewId{};
const ImageViewId depth_view_id = is_color ? ImageViewId{} : view_id;
const Extent3D extent = MipSize(image.info.size, view_info.range.base.level);
Extent3D extent = MipSize(image.info.size, view_info.range.base.level);
if (is_rescaled) {
const auto& resolution = Settings::values.resolution_info;
extent.width = resolution.ScaleUp(extent.width);
if (image.info.type == ImageType::e2D) {
extent.height = resolution.ScaleUp(extent.height);
}
}
const u32 num_samples = image.info.num_samples;
const auto [samples_x, samples_y] = SamplesLog2(num_samples);
const FramebufferId framebuffer_id = GetFramebufferId(RenderTargets{

38
src/video_core/texture_cache/texture_cache_base.h
View File

@@ -21,6 +21,7 @@
#include "video_core/texture_cache/descriptor_table.h"
#include "video_core/texture_cache/image_base.h"
#include "video_core/texture_cache/image_info.h"
#include "video_core/texture_cache/image_view_base.h"
#include "video_core/texture_cache/image_view_info.h"
#include "video_core/texture_cache/render_targets.h"
#include "video_core/texture_cache/slot_vector.h"
@@ -39,6 +40,12 @@ using VideoCore::Surface::PixelFormatFromDepthFormat;
using VideoCore::Surface::PixelFormatFromRenderTargetFormat;
using namespace Common::Literals;
struct ImageViewInOut {
u32 index{};
bool blacklist{};
ImageViewId id{};
};
template <class P>
class TextureCache {
/// Address shift for caching images into a hash table
@@ -53,11 +60,6 @@ class TextureCache {
/// True when the API can provide info about the memory of the device.
static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO;
/// Image view ID for null descriptors
static constexpr ImageViewId NULL_IMAGE_VIEW_ID{0};
/// Sampler ID for bugged sampler ids
static constexpr SamplerId NULL_SAMPLER_ID{0};
static constexpr u64 DEFAULT_EXPECTED_MEMORY = 1_GiB;
static constexpr u64 DEFAULT_CRITICAL_MEMORY = 2_GiB;
@@ -99,11 +101,11 @@ public:
void MarkModification(ImageId id) noexcept;
/// Fill image_view_ids with the graphics images in indices
void FillGraphicsImageViews(std::span<const u32> indices,
std::span<ImageViewId> image_view_ids);
template <bool has_blacklists>
void FillGraphicsImageViews(std::span<ImageViewInOut> views);
/// Fill image_view_ids with the compute images in indices
void FillComputeImageViews(std::span<const u32> indices, std::span<ImageViewId> image_view_ids);
void FillComputeImageViews(std::span<ImageViewInOut> views);
/// Get the sampler from the graphics descriptor table in the specified index
Sampler* GetGraphicsSampler(u32 index);
@@ -160,6 +162,10 @@ public:
/// Return true when a CPU region is modified from the GPU
[[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
[[nodiscard]] bool IsRescaling() const noexcept;
[[nodiscard]] bool IsRescaling(const ImageViewBase& image_view) const noexcept;
std::mutex mutex;
private:
@@ -198,9 +204,10 @@ private:
void RunGarbageCollector();
/// Fills image_view_ids in the image views in indices
template <bool has_blacklists>
void FillImageViews(DescriptorTable<TICEntry>& table,
std::span<ImageViewId> cached_image_view_ids, std::span<const u32> indices,
std::span<ImageViewId> image_view_ids);
std::span<ImageViewId> cached_image_view_ids,
std::span<ImageViewInOut> views);
/// Find or create an image view in the guest descriptor table
ImageViewId VisitImageView(DescriptorTable<TICEntry>& table,
@@ -285,7 +292,7 @@ private:
void UntrackImage(ImageBase& image, ImageId image_id);
/// Delete image from the cache
void DeleteImage(ImageId image);
void DeleteImage(ImageId image, bool immediate_delete = false);
/// Remove image views references from the cache
void RemoveImageViewReferences(std::span<const ImageViewId> removed_views);
@@ -306,7 +313,7 @@ private:
void PrepareImageView(ImageViewId image_view_id, bool is_modification, bool invalidate);
/// Execute copies from one image to the other, even if they are incompatible
void CopyImage(ImageId dst_id, ImageId src_id, std::span<const ImageCopy> copies);
void CopyImage(ImageId dst_id, ImageId src_id, std::vector<ImageCopy> copies);
/// Bind an image view as render target, downloading resources preemtively if needed
void BindRenderTarget(ImageViewId* old_id, ImageViewId new_id);
@@ -318,6 +325,12 @@ private:
/// Returns true if the current clear parameters clear the whole image of a given image view
[[nodiscard]] bool IsFullClear(ImageViewId id);
bool ImageCanRescale(ImageBase& image);
void InvalidateScale(Image& image);
bool ScaleUp(Image& image);
bool ScaleDown(Image& image);
u64 GetScaledImageSizeBytes(ImageBase& image);
Runtime& runtime;
VideoCore::RasterizerInterface& rasterizer;
Tegra::Engines::Maxwell3D& maxwell3d;
@@ -349,6 +362,7 @@ private:
VAddr virtual_invalid_space{};
bool has_deleted_images = false;
bool is_rescaling = false;
u64 total_used_memory = 0;
u64 minimum_memory;
u64 expected_memory;

7
src/video_core/texture_cache/types.h
View File

@@ -22,6 +22,13 @@ using ImageAllocId = SlotId;
using SamplerId = SlotId;
using FramebufferId = SlotId;
/// Fake image ID for null image views
constexpr ImageId NULL_IMAGE_ID{0};
/// Image view ID for null descriptors
constexpr ImageViewId NULL_IMAGE_VIEW_ID{0};
/// Sampler ID for bugged sampler ids
constexpr SamplerId NULL_SAMPLER_ID{0};
enum class ImageType : u32 {
e1D,
e2D,

12
src/video_core/texture_cache/util.cpp
View File

@@ -723,7 +723,7 @@ ImageViewType RenderTargetImageViewType(const ImageInfo& info) noexcept {
}
std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageInfo& src,
SubresourceBase base) {
SubresourceBase base, u32 up_scale, u32 down_shift) {
ASSERT(dst.resources.levels >= src.resources.levels);
ASSERT(dst.num_samples == src.num_samples);
@@ -732,7 +732,7 @@ std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageIn
ASSERT(src.type == ImageType::e3D);
ASSERT(src.resources.levels == 1);
}
const bool both_2d{src.type == ImageType::e2D && dst.type == ImageType::e2D};
std::vector<ImageCopy> copies;
copies.reserve(src.resources.levels);
for (s32 level = 0; level < src.resources.levels; ++level) {
@@ -762,6 +762,10 @@ std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageIn
if (is_dst_3d) {
copy.extent.depth = src.size.depth;
}
copy.extent.width = std::max<u32>((copy.extent.width * up_scale) >> down_shift, 1);
if (both_2d) {
copy.extent.height = std::max<u32>((copy.extent.height * up_scale) >> down_shift, 1);
}
}
return copies;
}
@@ -1153,10 +1157,10 @@ void DeduceBlitImages(ImageInfo& dst_info, ImageInfo& src_info, const ImageBase*
if (dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) {
dst_info.format = dst->info.format;
}
if (!dst && src && GetFormatType(src->info.format) != SurfaceType::ColorTexture) {
if (src && GetFormatType(src->info.format) != SurfaceType::ColorTexture) {
dst_info.format = src->info.format;
}
if (!src && dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) {
if (dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) {
src_info.format = dst->info.format;
}
}

3
src/video_core/texture_cache/util.h
View File

@@ -55,7 +55,8 @@ struct OverlapResult {
[[nodiscard]] std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst,
const ImageInfo& src,
SubresourceBase base);
SubresourceBase base, u32 up_scale = 1,
u32 down_shift = 0);
[[nodiscard]] bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config);