Texture Cahe/Shader decompiler: Resize PointSize on rescaling, refactor and make reaper more agressive on 4Gb GPUs.

This commit is contained in:
FernandoS27 2021-10-29 17:02:57 +02:00 committed by Fernando Sahmkow
parent 917b2466ad
commit de1c8c5c2c
5 changed files with 29 additions and 22 deletions

View File

@ -75,6 +75,14 @@ void PatchFragCoord(IR::Block& block, IR::Inst& inst) {
inst.ReplaceUsesWith(downscaled_frag_coord); inst.ReplaceUsesWith(downscaled_frag_coord);
} }
void PatchPointSize(IR::Block& block, IR::Inst& inst) {
IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
const IR::F32 point_value{inst.Arg(1)};
const IR::F32 up_factor{ir.FPRecip(ir.ResolutionDownFactor())};
const IR::F32 upscaled_point_value{ir.FPMul(point_value, up_factor)};
inst.SetArg(1, upscaled_point_value);
}
[[nodiscard]] IR::U32 Scale(IR::IREmitter& ir, const IR::U1& is_scaled, const IR::U32& value) { [[nodiscard]] IR::U32 Scale(IR::IREmitter& ir, const IR::U1& is_scaled, const IR::U32& value) {
IR::U32 scaled_value{value}; IR::U32 scaled_value{value};
if (const u32 up_scale = Settings::values.resolution_info.up_scale; up_scale != 1) { if (const u32 up_scale = Settings::values.resolution_info.up_scale; up_scale != 1) {
@ -253,6 +261,19 @@ void Visit(const IR::Program& program, IR::Block& block, IR::Inst& inst) {
} }
break; break;
} }
case IR::Opcode::SetAttribute: {
const IR::Attribute attr{inst.Arg(0).Attribute()};
switch (attr) {
case IR::Attribute::PointSize:
if (inst.Flags<u32>() != 0xDEADBEEF) {
PatchPointSize(block, inst);
}
break;
default:
break;
}
break;
}
case IR::Opcode::ImageQueryDimensions: case IR::Opcode::ImageQueryDimensions:
PatchImageQueryDimensions(block, inst); PatchImageQueryDimensions(block, inst);
break; break;

View File

@ -976,8 +976,9 @@ void RasterizerOpenGL::SyncPointState() {
oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable); oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable);
oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d.regs.vp_point_size.enable); oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d.regs.vp_point_size.enable);
const bool is_rescaling{texture_cache.IsRescaling()};
glPointSize(std::max(1.0f, maxwell3d.regs.point_size)); const float scale = is_rescaling ? Settings::values.resolution_info.up_factor : 1.0f;
glPointSize(std::max(1.0f, maxwell3d.regs.point_size * scale));
} }
void RasterizerOpenGL::SyncLineState() { void RasterizerOpenGL::SyncLineState() {

View File

@ -38,7 +38,6 @@ enum class ImageFlagBits : u32 {
Rescaled = 1 << 12, Rescaled = 1 << 12,
CheckingRescalable = 1 << 13, CheckingRescalable = 1 << 13,
IsRescalable = 1 << 14, IsRescalable = 1 << 14,
Blacklisted = 1 << 15,
}; };
DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits) DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits)

View File

@ -135,7 +135,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index)
type = ImageType::e3D; type = ImageType::e3D;
size.depth = rt.depth; size.depth = rt.depth;
} else { } else {
rescaleable = block.depth == 0 && size.height > 256; rescaleable = block.depth == 0;
downscaleable = size.height > 512; downscaleable = size.height > 512;
type = ImageType::e2D; type = ImageType::e2D;
resources.layers = rt.depth; resources.layers = rt.depth;
@ -165,7 +165,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept {
type = ImageType::e3D; type = ImageType::e3D;
size.depth = regs.zeta_depth; size.depth = regs.zeta_depth;
} else { } else {
rescaleable = block.depth == 0 && size.height > 256; rescaleable = block.depth == 0;
downscaleable = size.height > 512; downscaleable = size.height > 512;
type = ImageType::e2D; type = ImageType::e2D;
resources.layers = regs.zeta_depth; resources.layers = regs.zeta_depth;
@ -199,7 +199,7 @@ ImageInfo::ImageInfo(const Tegra::Engines::Fermi2D::Surface& config) noexcept {
.height = config.height, .height = config.height,
.depth = 1, .depth = 1,
}; };
rescaleable = block.depth == 0 && size.height > 256; rescaleable = block.depth == 0;
downscaleable = size.height > 512; downscaleable = size.height > 512;
} }
} }

View File

@ -53,8 +53,8 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
const auto device_memory = runtime.GetDeviceLocalMemory(); const auto device_memory = runtime.GetDeviceLocalMemory();
const u64 possible_expected_memory = (device_memory * 4) / 10; const u64 possible_expected_memory = (device_memory * 4) / 10;
const u64 possible_critical_memory = (device_memory * 7) / 10; const u64 possible_critical_memory = (device_memory * 7) / 10;
expected_memory = std::max(possible_expected_memory, DEFAULT_EXPECTED_MEMORY); expected_memory = std::max(possible_expected_memory, DEFAULT_EXPECTED_MEMORY - 256_MiB);
critical_memory = std::max(possible_critical_memory, DEFAULT_CRITICAL_MEMORY); critical_memory = std::max(possible_critical_memory, DEFAULT_CRITICAL_MEMORY - 512_MiB);
minimum_memory = 0; minimum_memory = 0;
} else { } else {
// On OpenGL we can be more conservatives as the driver takes care. // On OpenGL we can be more conservatives as the driver takes care.
@ -355,7 +355,6 @@ void TextureCache<P>::FillImageViews(DescriptorTable<TICEntry>& table,
if (view.blacklist && view.id != NULL_IMAGE_VIEW_ID) { if (view.blacklist && view.id != NULL_IMAGE_VIEW_ID) {
const ImageViewBase& image_view{slot_image_views[view.id]}; const ImageViewBase& image_view{slot_image_views[view.id]};
auto& image = slot_images[image_view.image_id]; auto& image = slot_images[image_view.image_id];
image.flags |= ImageFlagBits::Blacklisted;
has_blacklisted |= ScaleDown(image); has_blacklisted |= ScaleDown(image);
image.scale_rating = 0; image.scale_rating = 0;
} }
@ -985,7 +984,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
bool can_rescale = info.rescaleable; bool can_rescale = info.rescaleable;
bool any_rescaled = false; bool any_rescaled = false;
bool any_blacklisted = false;
for (const ImageId sibling_id : all_siblings) { for (const ImageId sibling_id : all_siblings) {
if (!can_rescale) { if (!can_rescale) {
break; break;
@ -993,7 +991,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
Image& sibling = slot_images[sibling_id]; Image& sibling = slot_images[sibling_id];
can_rescale &= ImageCanRescale(sibling); can_rescale &= ImageCanRescale(sibling);
any_rescaled |= True(sibling.flags & ImageFlagBits::Rescaled); any_rescaled |= True(sibling.flags & ImageFlagBits::Rescaled);
any_blacklisted |= True(sibling.flags & ImageFlagBits::Blacklisted);
} }
can_rescale &= any_rescaled; can_rescale &= any_rescaled;
@ -1007,9 +1004,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
for (const ImageId sibling_id : all_siblings) { for (const ImageId sibling_id : all_siblings) {
Image& sibling = slot_images[sibling_id]; Image& sibling = slot_images[sibling_id];
ScaleDown(sibling); ScaleDown(sibling);
if (any_blacklisted) {
sibling.flags |= ImageFlagBits::Blacklisted;
}
} }
} }
@ -1644,7 +1638,6 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) {
boost::container::small_vector<const AliasedImage*, 1> aliased_images; boost::container::small_vector<const AliasedImage*, 1> aliased_images;
Image& image = slot_images[image_id]; Image& image = slot_images[image_id];
bool any_rescaled = True(image.flags & ImageFlagBits::Rescaled); bool any_rescaled = True(image.flags & ImageFlagBits::Rescaled);
bool any_blacklisted = True(image.flags & ImageFlagBits::Blacklisted);
u64 most_recent_tick = image.modification_tick; u64 most_recent_tick = image.modification_tick;
for (const AliasedImage& aliased : image.aliased_images) { for (const AliasedImage& aliased : image.aliased_images) {
ImageBase& aliased_image = slot_images[aliased.id]; ImageBase& aliased_image = slot_images[aliased.id];
@ -1652,7 +1645,6 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) {
most_recent_tick = std::max(most_recent_tick, aliased_image.modification_tick); most_recent_tick = std::max(most_recent_tick, aliased_image.modification_tick);
aliased_images.push_back(&aliased); aliased_images.push_back(&aliased);
any_rescaled |= True(aliased_image.flags & ImageFlagBits::Rescaled); any_rescaled |= True(aliased_image.flags & ImageFlagBits::Rescaled);
any_blacklisted |= True(aliased_image.flags & ImageFlagBits::Blacklisted);
} }
} }
if (aliased_images.empty()) { if (aliased_images.empty()) {
@ -1664,9 +1656,6 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) {
ScaleUp(image); ScaleUp(image);
} else { } else {
ScaleDown(image); ScaleDown(image);
if (any_blacklisted) {
image.flags |= ImageFlagBits::Blacklisted;
}
} }
} }
image.modification_tick = most_recent_tick; image.modification_tick = most_recent_tick;
@ -1684,9 +1673,6 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) {
Image& aliased_image = slot_images[aliased->id]; Image& aliased_image = slot_images[aliased->id];
if (!can_rescale) { if (!can_rescale) {
ScaleDown(aliased_image); ScaleDown(aliased_image);
if (any_blacklisted) {
aliased_image.flags |= ImageFlagBits::Blacklisted;
}
CopyImage(image_id, aliased->id, aliased->copies); CopyImage(image_id, aliased->id, aliased->copies);
continue; continue;
} }