gl_rasterizer_cache: Use GPUVAddr as cache key, not parameter set.
This commit is contained in:
parent
cd47391c2d
commit
4301f0b539
3 changed files with 46 additions and 57 deletions
|
@ -413,11 +413,13 @@ void RasterizerOpenGL::Clear() {
|
|||
glClear(clear_mask);
|
||||
|
||||
// Mark framebuffer surfaces as dirty
|
||||
if (dirty_color_surface != nullptr) {
|
||||
res_cache.MarkSurfaceAsDirty(dirty_color_surface);
|
||||
}
|
||||
if (dirty_depth_surface != nullptr) {
|
||||
res_cache.MarkSurfaceAsDirty(dirty_depth_surface);
|
||||
if (Settings::values.use_accurate_framebuffers) {
|
||||
if (dirty_color_surface != nullptr) {
|
||||
res_cache.FlushSurface(dirty_color_surface);
|
||||
}
|
||||
if (dirty_depth_surface != nullptr) {
|
||||
res_cache.FlushSurface(dirty_depth_surface);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
@ -520,11 +522,13 @@ void RasterizerOpenGL::DrawArrays() {
|
|||
state.Apply();
|
||||
|
||||
// Mark framebuffer surfaces as dirty
|
||||
if (dirty_color_surface != nullptr) {
|
||||
res_cache.MarkSurfaceAsDirty(dirty_color_surface);
|
||||
}
|
||||
if (dirty_depth_surface != nullptr) {
|
||||
res_cache.MarkSurfaceAsDirty(dirty_depth_surface);
|
||||
if (Settings::values.use_accurate_framebuffers) {
|
||||
if (dirty_color_surface != nullptr) {
|
||||
res_cache.FlushSurface(dirty_color_surface);
|
||||
}
|
||||
if (dirty_depth_surface != nullptr) {
|
||||
res_cache.FlushSurface(dirty_depth_surface);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
@ -565,17 +565,9 @@ void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {
|
|||
surface->UploadGLTexture(read_framebuffer.handle, draw_framebuffer.handle);
|
||||
}
|
||||
|
||||
void RasterizerCacheOpenGL::MarkSurfaceAsDirty(const Surface& surface) {
|
||||
if (Settings::values.use_accurate_framebuffers) {
|
||||
// If enabled, always flush dirty surfaces
|
||||
surface->DownloadGLTexture(read_framebuffer.handle, draw_framebuffer.handle);
|
||||
surface->FlushGLBuffer();
|
||||
} else {
|
||||
// Otherwise, don't mark surfaces that we write to as cached, because the resulting loads
|
||||
// and flushes are very slow and do not seem to improve accuracy
|
||||
const auto& params{surface->GetSurfaceParams()};
|
||||
Memory::RasterizerMarkRegionCached(params.addr, params.size_in_bytes, false);
|
||||
}
|
||||
void RasterizerCacheOpenGL::FlushSurface(const Surface& surface) {
|
||||
surface->DownloadGLTexture(read_framebuffer.handle, draw_framebuffer.handle);
|
||||
surface->FlushGLBuffer();
|
||||
}
|
||||
|
||||
Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params) {
|
||||
|
@ -589,21 +581,22 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params) {
|
|||
return {};
|
||||
|
||||
// Check for an exact match in existing surfaces
|
||||
const auto& surface_key{SurfaceKey::Create(params)};
|
||||
const auto& search{surface_cache.find(surface_key)};
|
||||
const auto& search{surface_cache.find(params.addr)};
|
||||
Surface surface;
|
||||
if (search != surface_cache.end()) {
|
||||
surface = search->second;
|
||||
if (Settings::values.use_accurate_framebuffers) {
|
||||
// Reload the surface from Switch memory
|
||||
LoadSurface(surface);
|
||||
if (surface->GetSurfaceParams() != params || Settings::values.use_accurate_framebuffers) {
|
||||
FlushSurface(surface);
|
||||
UnregisterSurface(surface);
|
||||
} else {
|
||||
return surface;
|
||||
}
|
||||
} else {
|
||||
surface = std::make_shared<CachedSurface>(params);
|
||||
RegisterSurface(surface);
|
||||
LoadSurface(surface);
|
||||
}
|
||||
|
||||
surface = std::make_shared<CachedSurface>(params);
|
||||
RegisterSurface(surface);
|
||||
LoadSurface(surface);
|
||||
|
||||
return surface;
|
||||
}
|
||||
|
||||
|
@ -652,22 +645,20 @@ void RasterizerCacheOpenGL::InvalidateRegion(Tegra::GPUVAddr addr, size_t size)
|
|||
|
||||
void RasterizerCacheOpenGL::RegisterSurface(const Surface& surface) {
|
||||
const auto& params{surface->GetSurfaceParams()};
|
||||
const auto& surface_key{SurfaceKey::Create(params)};
|
||||
const auto& search{surface_cache.find(surface_key)};
|
||||
const auto& search{surface_cache.find(params.addr)};
|
||||
|
||||
if (search != surface_cache.end()) {
|
||||
// Registered already
|
||||
return;
|
||||
}
|
||||
|
||||
surface_cache[surface_key] = surface;
|
||||
surface_cache[params.addr] = surface;
|
||||
UpdatePagesCachedCount(params.addr, params.size_in_bytes, 1);
|
||||
}
|
||||
|
||||
void RasterizerCacheOpenGL::UnregisterSurface(const Surface& surface) {
|
||||
const auto& params{surface->GetSurfaceParams()};
|
||||
const auto& surface_key{SurfaceKey::Create(params)};
|
||||
const auto& search{surface_cache.find(surface_key)};
|
||||
const auto& search{surface_cache.find(params.addr)};
|
||||
|
||||
if (search == surface_cache.end()) {
|
||||
// Unregistered already
|
||||
|
|
|
@ -10,7 +10,6 @@
|
|||
#include <vector>
|
||||
#include <boost/icl/interval_map.hpp>
|
||||
#include "common/common_types.h"
|
||||
#include "common/hash.h"
|
||||
#include "common/math_util.h"
|
||||
#include "video_core/engines/maxwell_3d.h"
|
||||
#include "video_core/renderer_opengl/gl_resource_manager.h"
|
||||
|
@ -137,6 +136,7 @@ struct SurfaceParams {
|
|||
ASSERT(static_cast<size_t>(format) < bpp_table.size());
|
||||
return bpp_table[static_cast<size_t>(format)];
|
||||
}
|
||||
|
||||
u32 GetFormatBpp() const {
|
||||
return GetFormatBpp(pixel_format);
|
||||
}
|
||||
|
@ -369,6 +369,18 @@ struct SurfaceParams {
|
|||
Tegra::GPUVAddr zeta_address,
|
||||
Tegra::DepthFormat format);
|
||||
|
||||
bool operator==(const SurfaceParams& other) const {
|
||||
return std::tie(addr, is_tiled, block_height, pixel_format, component_type, type, width,
|
||||
height, unaligned_height, size_in_bytes) ==
|
||||
std::tie(other.addr, other.is_tiled, other.block_height, other.pixel_format,
|
||||
other.component_type, other.type, other.width, other.height,
|
||||
other.unaligned_height, other.size_in_bytes);
|
||||
}
|
||||
|
||||
bool operator!=(const SurfaceParams& other) const {
|
||||
return !operator==(other);
|
||||
}
|
||||
|
||||
Tegra::GPUVAddr addr;
|
||||
bool is_tiled;
|
||||
u32 block_height;
|
||||
|
@ -381,24 +393,6 @@ struct SurfaceParams {
|
|||
size_t size_in_bytes;
|
||||
};
|
||||
|
||||
/// Hashable variation of SurfaceParams, used for a key in the surface cache
|
||||
struct SurfaceKey : Common::HashableStruct<SurfaceParams> {
|
||||
static SurfaceKey Create(const SurfaceParams& params) {
|
||||
SurfaceKey res;
|
||||
res.state = params;
|
||||
return res;
|
||||
}
|
||||
};
|
||||
|
||||
namespace std {
|
||||
template <>
|
||||
struct hash<SurfaceKey> {
|
||||
size_t operator()(const SurfaceKey& k) const {
|
||||
return k.Hash();
|
||||
}
|
||||
};
|
||||
} // namespace std
|
||||
|
||||
class CachedSurface final {
|
||||
public:
|
||||
CachedSurface(const SurfaceParams& params);
|
||||
|
@ -444,8 +438,8 @@ public:
|
|||
SurfaceSurfaceRect_Tuple GetFramebufferSurfaces(bool using_color_fb, bool using_depth_fb,
|
||||
const MathUtil::Rectangle<s32>& viewport);
|
||||
|
||||
/// Marks the specified surface as "dirty", in that it is out of sync with Switch memory
|
||||
void MarkSurfaceAsDirty(const Surface& surface);
|
||||
/// Flushes the surface to Switch memory
|
||||
void FlushSurface(const Surface& surface);
|
||||
|
||||
/// Tries to find a framebuffer GPU address based on the provided CPU address
|
||||
Surface TryFindFramebufferSurface(VAddr cpu_addr) const;
|
||||
|
@ -469,7 +463,7 @@ private:
|
|||
/// Increase/decrease the number of surface in pages touching the specified region
|
||||
void UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta);
|
||||
|
||||
std::unordered_map<SurfaceKey, Surface> surface_cache;
|
||||
std::unordered_map<Tegra::GPUVAddr, Surface> surface_cache;
|
||||
PageMap cached_pages;
|
||||
|
||||
OGLFramebuffer read_framebuffer;
|
||||
|
|
Loading…
Reference in a new issue