1
0
Fork 0
forked from suyu/suyu

Reaper: Address Feedback.

This commit is contained in:
Fernando Sahmkow 2021-06-14 13:42:22 +02:00
parent 954ad2a61e
commit 0dd98842bf
6 changed files with 43 additions and 20 deletions

View file

@ -24,6 +24,7 @@ enum : u64 {
Size_128_MB = 128ULL * Size_1_MB, Size_128_MB = 128ULL * Size_1_MB,
Size_448_MB = 448ULL * Size_1_MB, Size_448_MB = 448ULL * Size_1_MB,
Size_507_MB = 507ULL * Size_1_MB, Size_507_MB = 507ULL * Size_1_MB,
Size_512_MB = 512ULL * Size_1_MB,
Size_562_MB = 562ULL * Size_1_MB, Size_562_MB = 562ULL * Size_1_MB,
Size_1554_MB = 1554ULL * Size_1_MB, Size_1554_MB = 1554ULL * Size_1_MB,
Size_2048_MB = 2048ULL * Size_1_MB, Size_2048_MB = 2048ULL * Size_1_MB,

View file

@ -16,6 +16,7 @@
#include <boost/container/small_vector.hpp> #include <boost/container/small_vector.hpp>
#include "common/common_sizes.h"
#include "common/common_types.h" #include "common/common_types.h"
#include "common/div_ceil.h" #include "common/div_ceil.h"
#include "common/microprofile.h" #include "common/microprofile.h"
@ -65,8 +66,8 @@ class BufferCache {
static constexpr BufferId NULL_BUFFER_ID{0}; static constexpr BufferId NULL_BUFFER_ID{0};
static constexpr u64 expected_memory = 512ULL * 1024ULL * 1024ULL; static constexpr u64 EXPECTED_MEMORY = Common::Size_512_MB;
static constexpr u64 critical_memory = 1024ULL * 1024ULL * 1024ULL; static constexpr u64 CRITICAL_MEMORY = Common::Size_1_GB;
using Maxwell = Tegra::Engines::Maxwell3D::Regs; using Maxwell = Tegra::Engines::Maxwell3D::Regs;
@ -368,13 +369,13 @@ void BufferCache<P>::TickFrame() {
const bool skip_preferred = hits * 256 < shots * 251; const bool skip_preferred = hits * 256 < shots * 251;
uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0; uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0;
const bool activate_gc = enabled_gc && total_used_memory >= expected_memory; const bool activate_gc = enabled_gc && total_used_memory >= EXPECTED_MEMORY;
if (!activate_gc) { if (!activate_gc) {
return; return;
} }
const bool agressive_gc = total_used_memory >= critical_memory; const bool aggressive_gc = total_used_memory >= CRITICAL_MEMORY;
const u64 ticks_to_destroy = agressive_gc ? 60 : 120; const u64 ticks_to_destroy = aggressive_gc ? 60 : 120;
int num_iterations = agressive_gc ? 64 : 32; int num_iterations = aggressive_gc ? 64 : 32;
for (; num_iterations > 0; --num_iterations) { for (; num_iterations > 0; --num_iterations) {
if (deletion_iterator == slot_buffers.end()) { if (deletion_iterator == slot_buffers.end()) {
deletion_iterator = slot_buffers.begin(); deletion_iterator = slot_buffers.begin();

View file

@ -283,4 +283,11 @@ std::pair<u32, u32> GetASTCBlockSize(PixelFormat format) {
return {DefaultBlockWidth(format), DefaultBlockHeight(format)}; return {DefaultBlockWidth(format), DefaultBlockHeight(format)};
} }
u64 EstimatedDecompressedSize(u64 base_size, PixelFormat format) {
constexpr u64 RGBA8_PIXEL_SIZE = 4;
const u64 base_block_size = static_cast<u64>(DefaultBlockWidth(format)) *
static_cast<u64>(DefaultBlockHeight(format)) * RGBA8_PIXEL_SIZE;
return (base_size * base_block_size) / BytesPerBlock(format);
}
} // namespace VideoCore::Surface } // namespace VideoCore::Surface

View file

@ -462,4 +462,6 @@ bool IsPixelFormatSRGB(PixelFormat format);
std::pair<u32, u32> GetASTCBlockSize(PixelFormat format); std::pair<u32, u32> GetASTCBlockSize(PixelFormat format);
u64 EstimatedDecompressedSize(u64 base_size, PixelFormat format);
} // namespace VideoCore::Surface } // namespace VideoCore::Surface

View file

@ -20,6 +20,7 @@
#include "common/alignment.h" #include "common/alignment.h"
#include "common/common_funcs.h" #include "common/common_funcs.h"
#include "common/common_sizes.h"
#include "common/common_types.h" #include "common/common_types.h"
#include "common/logging/log.h" #include "common/logging/log.h"
#include "common/settings.h" #include "common/settings.h"
@ -76,8 +77,8 @@ class TextureCache {
/// Sampler ID for bugged sampler ids /// Sampler ID for bugged sampler ids
static constexpr SamplerId NULL_SAMPLER_ID{0}; static constexpr SamplerId NULL_SAMPLER_ID{0};
static constexpr u64 expected_memory = 1024ULL * 1024ULL * 1024ULL; static constexpr u64 EXPECTED_MEMORY = Common::Size_1_GB;
static constexpr u64 critical_memory = 2 * 1024ULL * 1024ULL * 1024ULL; static constexpr u64 CRITICAL_MEMORY = Common::Size_2_GB;
using Runtime = typename P::Runtime; using Runtime = typename P::Runtime;
using Image = typename P::Image; using Image = typename P::Image;
@ -394,8 +395,8 @@ void TextureCache<P>::TickFrame() {
++frame_tick; ++frame_tick;
return; return;
} }
const bool high_priority_mode = total_used_memory >= expected_memory; const bool high_priority_mode = total_used_memory >= EXPECTED_MEMORY;
const bool aggressive_mode = total_used_memory >= critical_memory; const bool aggressive_mode = total_used_memory >= CRITICAL_MEMORY;
const u64 ticks_to_destroy = high_priority_mode ? 60 : 100; const u64 ticks_to_destroy = high_priority_mode ? 60 : 100;
int num_iterations = aggressive_mode ? 256 : (high_priority_mode ? 128 : 64); int num_iterations = aggressive_mode ? 256 : (high_priority_mode ? 128 : 64);
for (; num_iterations > 0; --num_iterations) { for (; num_iterations > 0; --num_iterations) {
@ -405,7 +406,8 @@ void TextureCache<P>::TickFrame() {
break; break;
} }
} }
const auto [image_id, image] = *deletion_iterator; auto [image_id, image_tmp] = *deletion_iterator;
Image* image = image_tmp; // fix clang error.
const bool is_alias = True(image->flags & ImageFlagBits::Alias); const bool is_alias = True(image->flags & ImageFlagBits::Alias);
const bool is_bad_overlap = True(image->flags & ImageFlagBits::BadOverlap); const bool is_bad_overlap = True(image->flags & ImageFlagBits::BadOverlap);
const bool must_download = image->IsSafeDownload(); const bool must_download = image->IsSafeDownload();
@ -417,8 +419,8 @@ void TextureCache<P>::TickFrame() {
should_care |= aggressive_mode; should_care |= aggressive_mode;
if (should_care && image->frame_tick + ticks_needed < frame_tick) { if (should_care && image->frame_tick + ticks_needed < frame_tick) {
if (is_bad_overlap) { if (is_bad_overlap) {
const bool overlap_check = const bool overlap_check = std::ranges::all_of(
std::ranges::all_of(image->overlapping_images, [&](const ImageId& overlap_id) { image->overlapping_images, [&, image](const ImageId& overlap_id) {
auto& overlap = slot_images[overlap_id]; auto& overlap = slot_images[overlap_id];
return overlap.frame_tick >= image->frame_tick; return overlap.frame_tick >= image->frame_tick;
}); });
@ -428,8 +430,8 @@ void TextureCache<P>::TickFrame() {
} }
} }
if (!is_bad_overlap && must_download) { if (!is_bad_overlap && must_download) {
const bool alias_check = const bool alias_check = std::ranges::none_of(
std::ranges::none_of(image->aliased_images, [&](const AliasedImage& alias) { image->aliased_images, [&, image](const AliasedImage& alias) {
auto& alias_image = slot_images[alias.id]; auto& alias_image = slot_images[alias.id];
return (alias_image.frame_tick < image->frame_tick) || return (alias_image.frame_tick < image->frame_tick) ||
(alias_image.modification_tick < image->modification_tick); (alias_image.modification_tick < image->modification_tick);
@ -1275,8 +1277,13 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
image.flags |= ImageFlagBits::Registered; image.flags |= ImageFlagBits::Registered;
ForEachPage(image.cpu_addr, image.guest_size_bytes, ForEachPage(image.cpu_addr, image.guest_size_bytes,
[this, image_id](u64 page) { page_table[page].push_back(image_id); }); [this, image_id](u64 page) { page_table[page].push_back(image_id); });
total_used_memory += u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
Common::AlignUp(std::max(image.guest_size_bytes, image.unswizzled_size_bytes), 1024); 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);
} }
template <class P> template <class P>
@ -1286,8 +1293,13 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
"Trying to unregister an already registered image"); "Trying to unregister an already registered image");
image.flags &= ~ImageFlagBits::Registered; image.flags &= ~ImageFlagBits::Registered;
image.flags &= ~ImageFlagBits::BadOverlap; image.flags &= ~ImageFlagBits::BadOverlap;
total_used_memory -= u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
Common::AlignUp(std::max(image.guest_size_bytes, image.unswizzled_size_bytes), 1024); 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);
ForEachPage(image.cpu_addr, image.guest_size_bytes, [this, image_id](u64 page) { ForEachPage(image.cpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
const auto page_it = page_table.find(page); const auto page_it = page_table.find(page);
if (page_it == page_table.end()) { if (page_it == page_table.end()) {

View file

@ -106,7 +106,7 @@
<item> <item>
<widget class="QCheckBox" name="use_caches_gc"> <widget class="QCheckBox" name="use_caches_gc">
<property name="toolTip"> <property name="toolTip">
<string>Enables garbage collection for the GPU caches, this will try to keep VRAM within 3-4Gb and flush least used textures/buffers. This option may be unsafe on a few games</string> <string>Enables garbage collection for the GPU caches, this will try to keep VRAM within 3-4 GB by flushing the least used textures/buffers. May cause issues in a few games.</string>
</property> </property>
<property name="text"> <property name="text">
<string>Enable GPU caches garbage collection (unsafe)</string> <string>Enable GPU caches garbage collection (unsafe)</string>