Buffer cache: Fixes, Clang and Feedback.

This commit is contained in:
Fernando Sahmkow 2021-07-14 19:04:45 +02:00
parent 1a95a7cdd9
commit 1ae4b684ff
3 changed files with 10 additions and 11 deletions

View file

@ -172,7 +172,7 @@ public:
[[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
/// Return true when a region is registered on the cache /// Return true when a region is registered on the cache
[[nodiscard]] bool IsRegionRegistered(VAddr addr, size_t size) const; [[nodiscard]] bool IsRegionRegistered(VAddr addr, size_t size);
/// Return true when a CPU region is modified from the CPU /// Return true when a CPU region is modified from the CPU
[[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size); [[nodiscard]] bool IsRegionCpuModified(VAddr addr, size_t size);
@ -503,10 +503,6 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
auto& src_buffer = slot_buffers[buffer_a]; auto& src_buffer = slot_buffers[buffer_a];
auto& dest_buffer = slot_buffers[buffer_b]; auto& dest_buffer = slot_buffers[buffer_b];
SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast<u32>(amount)); SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast<u32>(amount));
const VAddr aligned_dst = Common::AlignUp(*cpu_dest_address, 64);
const u64 diff = aligned_dst - *cpu_dest_address;
const u64 new_amount = diff > amount ? 0 : amount - diff;
dest_buffer.UnmarkRegionAsCpuModified(aligned_dst, Common::AlignDown(new_amount, 64));
SynchronizeBuffer(dest_buffer, *cpu_dest_address, static_cast<u32>(amount)); SynchronizeBuffer(dest_buffer, *cpu_dest_address, static_cast<u32>(amount));
std::array copies{BufferCopy{ std::array copies{BufferCopy{
.src_offset = src_buffer.Offset(*cpu_src_address), .src_offset = src_buffer.Offset(*cpu_src_address),
@ -552,21 +548,19 @@ bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) {
return false; return false;
} }
const IntervalType subtract_interval{*cpu_dst_address, *cpu_dst_address + amount * sizeof(u32)}; const size_t size = amount * sizeof(u32);
const IntervalType subtract_interval{*cpu_dst_address, *cpu_dst_address + size};
ClearDownload(subtract_interval); ClearDownload(subtract_interval);
common_ranges.subtract(subtract_interval); common_ranges.subtract(subtract_interval);
const size_t size = amount * sizeof(u32);
BufferId buffer; BufferId buffer;
do { do {
has_deleted_buffers = false; has_deleted_buffers = false;
buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size)); buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
} while (has_deleted_buffers); } while (has_deleted_buffers);
auto& dest_buffer = slot_buffers[buffer]; auto& dest_buffer = slot_buffers[buffer];
const u32 offset = static_cast<u32>(*cpu_dst_address - dest_buffer.CpuAddr()); const u32 offset = static_cast<u32>(*cpu_dst_address - dest_buffer.CpuAddr());
runtime.ClearBuffer(dest_buffer, offset, size, value); runtime.ClearBuffer(dest_buffer, offset, size, value);
dest_buffer.UnmarkRegionAsCpuModified(*cpu_dst_address, size);
return true; return true;
} }
@ -828,7 +822,7 @@ bool BufferCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
} }
template <class P> template <class P>
bool BufferCache<P>::IsRegionRegistered(VAddr addr, size_t size) const { bool BufferCache<P>::IsRegionRegistered(VAddr addr, size_t size) {
const VAddr end_addr = addr + size; const VAddr end_addr = addr + size;
const u64 page_end = Common::DivCeil(end_addr, PAGE_SIZE); const u64 page_end = Common::DivCeil(end_addr, PAGE_SIZE);
for (u64 page = addr >> PAGE_BITS; page < page_end;) { for (u64 page = addr >> PAGE_BITS; page < page_end;) {

View file

@ -4,6 +4,7 @@
#include "common/assert.h" #include "common/assert.h"
#include "common/logging/log.h" #include "common/logging/log.h"
#include "common/microprofile.h"
#include "common/settings.h" #include "common/settings.h"
#include "core/core.h" #include "core/core.h"
#include "video_core/engines/maxwell_3d.h" #include "video_core/engines/maxwell_3d.h"
@ -12,6 +13,9 @@
#include "video_core/renderer_base.h" #include "video_core/renderer_base.h"
#include "video_core/textures/decoders.h" #include "video_core/textures/decoders.h"
MICROPROFILE_DECLARE(GPU_DMAEngine);
MICROPROFILE_DEFINE(GPU_DMAEngine, "GPU", "DMA Engine", MP_RGB(224, 224, 128));
namespace Tegra::Engines { namespace Tegra::Engines {
using namespace Texture; using namespace Texture;
@ -43,6 +47,7 @@ void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
} }
void MaxwellDMA::Launch() { void MaxwellDMA::Launch() {
MICROPROFILE_SCOPE(GPU_DMAEngine);
LOG_TRACE(Render_OpenGL, "DMA copy 0x{:x} -> 0x{:x}", static_cast<GPUVAddr>(regs.offset_in), LOG_TRACE(Render_OpenGL, "DMA copy 0x{:x} -> 0x{:x}", static_cast<GPUVAddr>(regs.offset_in),
static_cast<GPUVAddr>(regs.offset_out)); static_cast<GPUVAddr>(regs.offset_out));

View file

@ -327,7 +327,7 @@ void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, s
// Invalidate must happen on the rasterizer interface, such that memory is always // Invalidate must happen on the rasterizer interface, such that memory is always
// synchronous when it is written (even when in asynchronous GPU mode). // synchronous when it is written (even when in asynchronous GPU mode).
rasterizer->UnmapMemory(dest_addr, copy_amount); rasterizer->InvalidateRegion(dest_addr, copy_amount);
system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount); system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount);
} }