diff options
Diffstat (limited to 'src')
30 files changed, 287 insertions, 265 deletions
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt index 3d30f0e3e..c538c6415 100644 --- a/src/common/CMakeLists.txt +++ b/src/common/CMakeLists.txt @@ -114,6 +114,8 @@ add_library(common STATIC threadsafe_queue.h timer.cpp timer.h + uint128.cpp + uint128.h vector_math.h web_result.h ) diff --git a/src/common/uint128.cpp b/src/common/uint128.cpp new file mode 100644 index 000000000..2238a52c5 --- /dev/null +++ b/src/common/uint128.cpp @@ -0,0 +1,41 @@ +#ifdef _MSC_VER +#include <intrin.h> + +#pragma intrinsic(_umul128) +#endif +#include <cstring> +#include "common/uint128.h" + +namespace Common { + +u128 Multiply64Into128(u64 a, u64 b) { + u128 result; +#ifdef _MSC_VER + result[0] = _umul128(a, b, &result[1]); +#else + unsigned __int128 tmp = a; + tmp *= b; + std::memcpy(&result, &tmp, sizeof(u128)); +#endif + return result; +} + +std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor) { + u64 remainder = dividend[0] % divisor; + u64 accum = dividend[0] / divisor; + if (dividend[1] == 0) + return {accum, remainder}; + // We ignore dividend[1] / divisor as that overflows + const u64 first_segment = (dividend[1] % divisor) << 32; + accum += (first_segment / divisor) << 32; + const u64 second_segment = (first_segment % divisor) << 32; + accum += (second_segment / divisor); + remainder += second_segment % divisor; + if (remainder >= divisor) { + accum++; + remainder -= divisor; + } + return {accum, remainder}; +} + +} // namespace Common diff --git a/src/common/uint128.h b/src/common/uint128.h new file mode 100644 index 000000000..52e6b46eb --- /dev/null +++ b/src/common/uint128.h @@ -0,0 +1,14 @@ + +#include <utility> +#include "common/common_types.h" + +namespace Common { + +// This function multiplies 2 u64 values and produces a u128 value; +u128 Multiply64Into128(u64 a, u64 b); + +// This function divides a u128 by a u32 value and produces two u64 values: +// the result of division and the remainder +std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor); + +} // namespace Common diff --git a/src/core/arm/dynarmic/arm_dynarmic.cpp b/src/core/arm/dynarmic/arm_dynarmic.cpp index 9b7ca4030..4fdc12f11 100644 --- a/src/core/arm/dynarmic/arm_dynarmic.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic.cpp @@ -12,6 +12,7 @@ #include "core/core.h" #include "core/core_cpu.h" #include "core/core_timing.h" +#include "core/core_timing_util.h" #include "core/gdbstub/gdbstub.h" #include "core/hle/kernel/process.h" #include "core/hle/kernel/svc.h" @@ -119,7 +120,7 @@ public: return std::max(parent.core_timing.GetDowncount(), 0); } u64 GetCNTPCT() override { - return parent.core_timing.GetTicks(); + return Timing::CpuCyclesToClockCycles(parent.core_timing.GetTicks()); } ARM_Dynarmic& parent; @@ -151,7 +152,7 @@ std::unique_ptr<Dynarmic::A64::Jit> ARM_Dynarmic::MakeJit() const { config.tpidr_el0 = &cb->tpidr_el0; config.dczid_el0 = 4; config.ctr_el0 = 0x8444c004; - config.cntfrq_el0 = 19200000; // Value from fusee. + config.cntfrq_el0 = Timing::CNTFREQ; // Unpredictable instructions config.define_unpredictable_behaviour = true; diff --git a/src/core/core_timing_util.cpp b/src/core/core_timing_util.cpp index 88ff70233..7942f30d6 100644 --- a/src/core/core_timing_util.cpp +++ b/src/core/core_timing_util.cpp @@ -7,6 +7,7 @@ #include <cinttypes> #include <limits> #include "common/logging/log.h" +#include "common/uint128.h" namespace Core::Timing { @@ -60,4 +61,9 @@ s64 nsToCycles(u64 ns) { return (BASE_CLOCK_RATE * static_cast<s64>(ns)) / 1000000000; } +u64 CpuCyclesToClockCycles(u64 ticks) { + const u128 temporal = Common::Multiply64Into128(ticks, CNTFREQ); + return Common::Divide128On32(temporal, static_cast<u32>(BASE_CLOCK_RATE)).first; +} + } // namespace Core::Timing diff --git a/src/core/core_timing_util.h b/src/core/core_timing_util.h index 513cfac1b..679aa3123 100644 --- a/src/core/core_timing_util.h +++ b/src/core/core_timing_util.h @@ -11,6 +11,7 @@ namespace Core::Timing { // The below clock rate is based on Switch's clockspeed being widely known as 1.020GHz // The exact value used is of course unverified. constexpr u64 BASE_CLOCK_RATE = 1019215872; // Switch clock speed is 1020MHz un/docked +constexpr u64 CNTFREQ = 19200000; // Value from fusee. inline s64 msToCycles(int ms) { // since ms is int there is no way to overflow @@ -61,4 +62,6 @@ inline u64 cyclesToMs(s64 cycles) { return cycles * 1000 / BASE_CLOCK_RATE; } +u64 CpuCyclesToClockCycles(u64 ticks); + } // namespace Core::Timing diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index 2e712c9cb..89f180bd9 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -68,12 +68,6 @@ void Thread::Stop() { owner_process->FreeTLSSlot(tls_address); } -void ExitCurrentThread() { - Thread* thread = GetCurrentThread(); - thread->Stop(); - Core::System::GetInstance().CurrentScheduler().RemoveThread(thread); -} - void Thread::WakeAfterDelay(s64 nanoseconds) { // Don't schedule a wakeup if the thread wants to wait forever if (nanoseconds == -1) diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 4fde53033..e0cc5175f 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -18,6 +18,7 @@ #include "core/hle/lock.h" #include "core/memory.h" #include "core/memory_setup.h" +#include "video_core/gpu.h" #include "video_core/renderer_base.h" namespace Memory { @@ -69,8 +70,8 @@ static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, Pa // During boot, current_page_table might not be set yet, in which case we need not flush if (current_page_table) { - RasterizerFlushVirtualRegion(base << PAGE_BITS, size * PAGE_SIZE, - FlushMode::FlushAndInvalidate); + Core::System::GetInstance().GPU().FlushAndInvalidateRegion(base << PAGE_BITS, + size * PAGE_SIZE); } VAddr end = base + size; @@ -183,10 +184,10 @@ T Read(const VAddr vaddr) { ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr); break; case PageType::RasterizerCachedMemory: { - RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Flush); - + auto host_ptr{GetPointerFromVMA(vaddr)}; + Core::System::GetInstance().GPU().FlushRegion(ToCacheAddr(host_ptr), sizeof(T)); T value; - std::memcpy(&value, GetPointerFromVMA(vaddr), sizeof(T)); + std::memcpy(&value, host_ptr, sizeof(T)); return value; } default: @@ -214,8 +215,9 @@ void Write(const VAddr vaddr, const T data) { ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr); break; case PageType::RasterizerCachedMemory: { - RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Invalidate); - std::memcpy(GetPointerFromVMA(vaddr), &data, sizeof(T)); + auto host_ptr{GetPointerFromVMA(vaddr)}; + Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(T)); + std::memcpy(host_ptr, &data, sizeof(T)); break; } default: @@ -338,47 +340,6 @@ void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) { } } -void RasterizerFlushVirtualRegion(VAddr start, u64 size, FlushMode mode) { - auto& system_instance = Core::System::GetInstance(); - - // Since pages are unmapped on shutdown after video core is shutdown, the renderer may be - // null here - if (!system_instance.IsPoweredOn()) { - return; - } - - const VAddr end = start + size; - - const auto CheckRegion = [&](VAddr region_start, VAddr region_end) { - if (start >= region_end || end <= region_start) { - // No overlap with region - return; - } - - const VAddr overlap_start = std::max(start, region_start); - const VAddr overlap_end = std::min(end, region_end); - const VAddr overlap_size = overlap_end - overlap_start; - - auto& gpu = system_instance.GPU(); - switch (mode) { - case FlushMode::Flush: - gpu.FlushRegion(ToCacheAddr(GetPointer(overlap_start)), overlap_size); - break; - case FlushMode::Invalidate: - gpu.InvalidateRegion(ToCacheAddr(GetPointer(overlap_start)), overlap_size); - break; - case FlushMode::FlushAndInvalidate: - gpu.FlushAndInvalidateRegion(ToCacheAddr(GetPointer(overlap_start)), overlap_size); - break; - } - }; - - const auto& vm_manager = Core::CurrentProcess()->VMManager(); - - CheckRegion(vm_manager.GetCodeRegionBaseAddress(), vm_manager.GetCodeRegionEndAddress()); - CheckRegion(vm_manager.GetHeapRegionBaseAddress(), vm_manager.GetHeapRegionEndAddress()); -} - u8 Read8(const VAddr addr) { return Read<u8>(addr); } @@ -424,9 +385,9 @@ void ReadBlock(const Kernel::Process& process, const VAddr src_addr, void* dest_ break; } case PageType::RasterizerCachedMemory: { - RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount), - FlushMode::Flush); - std::memcpy(dest_buffer, GetPointerFromVMA(process, current_vaddr), copy_amount); + const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)}; + Core::System::GetInstance().GPU().FlushRegion(ToCacheAddr(host_ptr), copy_amount); + std::memcpy(dest_buffer, host_ptr, copy_amount); break; } default: @@ -487,9 +448,9 @@ void WriteBlock(const Kernel::Process& process, const VAddr dest_addr, const voi break; } case PageType::RasterizerCachedMemory: { - RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount), - FlushMode::Invalidate); - std::memcpy(GetPointerFromVMA(process, current_vaddr), src_buffer, copy_amount); + const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)}; + Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), copy_amount); + std::memcpy(host_ptr, src_buffer, copy_amount); break; } default: @@ -533,9 +494,9 @@ void ZeroBlock(const Kernel::Process& process, const VAddr dest_addr, const std: break; } case PageType::RasterizerCachedMemory: { - RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount), - FlushMode::Invalidate); - std::memset(GetPointerFromVMA(process, current_vaddr), 0, copy_amount); + const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)}; + Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), copy_amount); + std::memset(host_ptr, 0, copy_amount); break; } default: @@ -575,9 +536,9 @@ void CopyBlock(const Kernel::Process& process, VAddr dest_addr, VAddr src_addr, break; } case PageType::RasterizerCachedMemory: { - RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount), - FlushMode::Flush); - WriteBlock(process, dest_addr, GetPointerFromVMA(process, current_vaddr), copy_amount); + const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)}; + Core::System::GetInstance().GPU().FlushRegion(ToCacheAddr(host_ptr), copy_amount); + WriteBlock(process, dest_addr, host_ptr, copy_amount); break; } default: diff --git a/src/core/memory.h b/src/core/memory.h index 1acf5ce8c..c2c6643ee 100644 --- a/src/core/memory.h +++ b/src/core/memory.h @@ -161,10 +161,4 @@ enum class FlushMode { */ void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached); -/** - * Flushes and invalidates any externally cached rasterizer resources touching the given virtual - * address region. - */ -void RasterizerFlushVirtualRegion(VAddr start, u64 size, FlushMode mode); - } // namespace Memory diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index bff1a37ff..8b1bea1ae 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -55,12 +55,9 @@ bool DmaPusher::Step() { } // Push buffer non-empty, read a word - const auto address = gpu.MemoryManager().GpuToCpuAddress(dma_get); - ASSERT_MSG(address, "Invalid GPU address"); - command_headers.resize(command_list_header.size); - - Memory::ReadBlock(*address, command_headers.data(), command_list_header.size * sizeof(u32)); + gpu.MemoryManager().ReadBlock(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); for (const CommandHeader& command_header : command_headers) { diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp index daefa43a6..0931b9626 100644 --- a/src/video_core/engines/kepler_memory.cpp +++ b/src/video_core/engines/kepler_memory.cpp @@ -41,18 +41,13 @@ void KeplerMemory::ProcessData(u32 data) { ASSERT_MSG(regs.exec.linear, "Non-linear uploads are not supported"); ASSERT(regs.dest.x == 0 && regs.dest.y == 0 && regs.dest.z == 0); - const GPUVAddr address = regs.dest.Address(); - const auto dest_address = - memory_manager.GpuToCpuAddress(address + state.write_offset * sizeof(u32)); - ASSERT_MSG(dest_address, "Invalid GPU address"); - // We have to invalidate the destination region to evict any outdated surfaces from the cache. - // We do this before actually writing the new data because the destination address might contain - // a dirty surface that will have to be written back to memory. - system.Renderer().Rasterizer().InvalidateRegion(ToCacheAddr(Memory::GetPointer(*dest_address)), - sizeof(u32)); + // We do this before actually writing the new data because the destination address might + // contain a dirty surface that will have to be written back to memory. + const GPUVAddr address{regs.dest.Address() + state.write_offset * sizeof(u32)}; + rasterizer.InvalidateRegion(ToCacheAddr(memory_manager.GetPointer(address)), sizeof(u32)); + memory_manager.Write32(address, data); - Memory::Write32(*dest_address, data); system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite(); state.write_offset++; diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 49979694e..c5d5be4ef 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -270,11 +270,9 @@ void Maxwell3D::ProcessMacroBind(u32 data) { } void Maxwell3D::ProcessQueryGet() { - GPUVAddr sequence_address = regs.query.QueryAddress(); + const GPUVAddr sequence_address{regs.query.QueryAddress()}; // Since the sequence address is given as a GPU VAddr, we have to convert it to an application // VAddr before writing. - const auto address = memory_manager.GpuToCpuAddress(sequence_address); - ASSERT_MSG(address, "Invalid GPU address"); // TODO(Subv): Support the other query units. ASSERT_MSG(regs.query.query_get.unit == Regs::QueryUnit::Crop, @@ -309,7 +307,7 @@ void Maxwell3D::ProcessQueryGet() { // Write the current query sequence to the sequence address. // TODO(Subv): Find out what happens if you use a long query type but mark it as a short // query. - Memory::Write32(*address, sequence); + memory_manager.Write32(sequence_address, sequence); } else { // Write the 128-bit result structure in long mode. Note: We emulate an infinitely fast // GPU, this command may actually take a while to complete in real hardware due to GPU @@ -318,7 +316,7 @@ void Maxwell3D::ProcessQueryGet() { query_result.value = result; // TODO(Subv): Generate a real GPU timestamp and write it here instead of CoreTiming query_result.timestamp = system.CoreTiming().GetTicks(); - Memory::WriteBlock(*address, &query_result, sizeof(query_result)); + memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result)); } dirty_flags.OnMemoryWrite(); break; @@ -393,12 +391,11 @@ void Maxwell3D::ProcessCBData(u32 value) { // Don't allow writing past the end of the buffer. ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size); - const auto address = memory_manager.GpuToCpuAddress(buffer_address + regs.const_buffer.cb_pos); - ASSERT_MSG(address, "Invalid GPU address"); + const GPUVAddr address{buffer_address + regs.const_buffer.cb_pos}; - u8* ptr{Memory::GetPointer(*address)}; + u8* ptr{memory_manager.GetPointer(address)}; rasterizer.InvalidateRegion(ToCacheAddr(ptr), sizeof(u32)); - std::memcpy(ptr, &value, sizeof(u32)); + memory_manager.Write32(address, value); dirty_flags.OnMemoryWrite(); @@ -407,14 +404,10 @@ void Maxwell3D::ProcessCBData(u32 value) { } Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const { - const GPUVAddr tic_base_address = regs.tic.TICAddress(); - - const GPUVAddr tic_address_gpu = tic_base_address + tic_index * sizeof(Texture::TICEntry); - const auto tic_address_cpu = memory_manager.GpuToCpuAddress(tic_address_gpu); - ASSERT_MSG(tic_address_cpu, "Invalid GPU address"); + const GPUVAddr tic_address_gpu{regs.tic.TICAddress() + tic_index * sizeof(Texture::TICEntry)}; Texture::TICEntry tic_entry; - Memory::ReadBlock(*tic_address_cpu, &tic_entry, sizeof(Texture::TICEntry)); + memory_manager.ReadBlock(tic_address_gpu, &tic_entry, sizeof(Texture::TICEntry)); ASSERT_MSG(tic_entry.header_version == Texture::TICHeaderVersion::BlockLinear || tic_entry.header_version == Texture::TICHeaderVersion::Pitch, @@ -432,14 +425,10 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const { } Texture::TSCEntry Maxwell3D::GetTSCEntry(u32 tsc_index) const { - const GPUVAddr tsc_base_address = regs.tsc.TSCAddress(); - - const GPUVAddr tsc_address_gpu = tsc_base_address + tsc_index * sizeof(Texture::TSCEntry); - const auto tsc_address_cpu = memory_manager.GpuToCpuAddress(tsc_address_gpu); - ASSERT_MSG(tsc_address_cpu, "Invalid GPU address"); + const GPUVAddr tsc_address_gpu{regs.tsc.TSCAddress() + tsc_index * sizeof(Texture::TSCEntry)}; Texture::TSCEntry tsc_entry; - Memory::ReadBlock(*tsc_address_cpu, &tsc_entry, sizeof(Texture::TSCEntry)); + memory_manager.ReadBlock(tsc_address_gpu, &tsc_entry, sizeof(Texture::TSCEntry)); return tsc_entry; } @@ -458,10 +447,7 @@ std::vector<Texture::FullTextureInfo> Maxwell3D::GetStageTextures(Regs::ShaderSt for (GPUVAddr current_texture = tex_info_buffer.address + TextureInfoOffset; current_texture < tex_info_buffer_end; current_texture += sizeof(Texture::TextureHandle)) { - const auto address = memory_manager.GpuToCpuAddress(current_texture); - ASSERT_MSG(address, "Invalid GPU address"); - - const Texture::TextureHandle tex_handle{Memory::Read32(*address)}; + const Texture::TextureHandle tex_handle{memory_manager.Read32(current_texture)}; Texture::FullTextureInfo tex_info{}; // TODO(Subv): Use the shader to determine which textures are actually accessed. @@ -496,10 +482,7 @@ Texture::FullTextureInfo Maxwell3D::GetStageTexture(Regs::ShaderStage stage, ASSERT(tex_info_address < tex_info_buffer.address + tex_info_buffer.size); - const auto tex_address_cpu = memory_manager.GpuToCpuAddress(tex_info_address); - ASSERT_MSG(tex_address_cpu, "Invalid GPU address"); - - const Texture::TextureHandle tex_handle{Memory::Read32(*tex_address_cpu)}; + const Texture::TextureHandle tex_handle{memory_manager.Read32(tex_info_address)}; Texture::FullTextureInfo tex_info{}; tex_info.index = static_cast<u32>(offset); diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 415a6319a..a0ded4c25 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -43,11 +43,6 @@ void MaxwellDMA::HandleCopy() { const GPUVAddr source = regs.src_address.Address(); const GPUVAddr dest = regs.dst_address.Address(); - const auto source_cpu = memory_manager.GpuToCpuAddress(source); - const auto dest_cpu = memory_manager.GpuToCpuAddress(dest); - ASSERT_MSG(source_cpu, "Invalid source GPU address"); - ASSERT_MSG(dest_cpu, "Invalid destination GPU address"); - // TODO(Subv): Perform more research and implement all features of this engine. ASSERT(regs.exec.enable_swizzle == 0); ASSERT(regs.exec.query_mode == Regs::QueryMode::None); @@ -70,7 +65,7 @@ void MaxwellDMA::HandleCopy() { // buffer of length `x_count`, otherwise we copy a 2D image of dimensions (x_count, // y_count). if (!regs.exec.enable_2d) { - Memory::CopyBlock(*dest_cpu, *source_cpu, regs.x_count); + memory_manager.CopyBlock(dest, source, regs.x_count); return; } @@ -79,9 +74,9 @@ void MaxwellDMA::HandleCopy() { // rectangle. There is no need to manually flush/invalidate the regions because // CopyBlock does that for us. for (u32 line = 0; line < regs.y_count; ++line) { - const VAddr source_line = *source_cpu + line * regs.src_pitch; - const VAddr dest_line = *dest_cpu + line * regs.dst_pitch; - Memory::CopyBlock(dest_line, source_line, regs.x_count); + const GPUVAddr source_line = source + line * regs.src_pitch; + const GPUVAddr dest_line = dest + line * regs.dst_pitch; + memory_manager.CopyBlock(dest_line, source_line, regs.x_count); } return; } @@ -90,17 +85,18 @@ void MaxwellDMA::HandleCopy() { const std::size_t copy_size = regs.x_count * regs.y_count; + auto source_ptr{memory_manager.GetPointer(source)}; + auto dst_ptr{memory_manager.GetPointer(dest)}; + const auto FlushAndInvalidate = [&](u32 src_size, u64 dst_size) { // TODO(Subv): For now, manually flush the regions until we implement GPU-accelerated // copying. - Core::System::GetInstance().Renderer().Rasterizer().FlushRegion( - ToCacheAddr(Memory::GetPointer(*source_cpu)), src_size); + rasterizer.FlushRegion(ToCacheAddr(source_ptr), src_size); // We have to invalidate the destination region to evict any outdated surfaces from the // cache. We do this before actually writing the new data because the destination address // might contain a dirty surface that will have to be written back to memory. - Core::System::GetInstance().Renderer().Rasterizer().InvalidateRegion( - ToCacheAddr(Memory::GetPointer(*dest_cpu)), dst_size); + rasterizer.InvalidateRegion(ToCacheAddr(dst_ptr), dst_size); }; if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) { @@ -113,8 +109,8 @@ void MaxwellDMA::HandleCopy() { copy_size * src_bytes_per_pixel); Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch, - regs.src_params.size_x, src_bytes_per_pixel, *source_cpu, - *dest_cpu, regs.src_params.BlockHeight(), regs.src_params.pos_x, + regs.src_params.size_x, src_bytes_per_pixel, source_ptr, dst_ptr, + regs.src_params.BlockHeight(), regs.src_params.pos_x, regs.src_params.pos_y); } else { ASSERT(regs.dst_params.size_z == 1); @@ -127,7 +123,7 @@ void MaxwellDMA::HandleCopy() { // If the input is linear and the output is tiled, swizzle the input and copy it over. Texture::SwizzleSubrect(regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x, - src_bpp, *dest_cpu, *source_cpu, regs.dst_params.BlockHeight()); + src_bpp, dst_ptr, source_ptr, regs.dst_params.BlockHeight()); } } diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 08abf8ac9..66c690494 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -274,7 +274,6 @@ void GPU::ProcessSemaphoreTriggerMethod() { const auto op = static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask); if (op == GpuSemaphoreOperation::WriteLong) { - auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress()); struct Block { u32 sequence; u32 zeros = 0; @@ -286,11 +285,9 @@ void GPU::ProcessSemaphoreTriggerMethod() { // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of // CoreTiming block.timestamp = Core::System::GetInstance().CoreTiming().GetTicks(); - Memory::WriteBlock(*address, &block, sizeof(block)); + memory_manager->WriteBlock(regs.smaphore_address.SmaphoreAddress(), &block, sizeof(block)); } else { - const auto address = - memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress()); - const u32 word = Memory::Read32(*address); + const u32 word{memory_manager->Read32(regs.smaphore_address.SmaphoreAddress())}; if ((op == GpuSemaphoreOperation::AcquireEqual && word == regs.semaphore_sequence) || (op == GpuSemaphoreOperation::AcquireGequal && static_cast<s32>(word - regs.semaphore_sequence) > 0) || @@ -317,13 +314,11 @@ void GPU::ProcessSemaphoreTriggerMethod() { } void GPU::ProcessSemaphoreRelease() { - const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress()); - Memory::Write32(*address, regs.semaphore_release); + memory_manager->Write32(regs.smaphore_address.SmaphoreAddress(), regs.semaphore_release); } void GPU::ProcessSemaphoreAcquire() { - const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress()); - const u32 word = Memory::Read32(*address); + const u32 word = memory_manager->Read32(regs.smaphore_address.SmaphoreAddress()); const auto value = regs.semaphore_acquire; if (word != value) { regs.acquire_active = true; diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index 54abe5298..8e8f36f28 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -5,6 +5,7 @@ #include "common/alignment.h" #include "common/assert.h" #include "common/logging/log.h" +#include "core/memory.h" #include "video_core/memory_manager.h" namespace Tegra { @@ -162,15 +163,51 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) { return base_addr + (gpu_addr & PAGE_MASK); } -std::vector<GPUVAddr> MemoryManager::CpuToGpuAddress(VAddr cpu_addr) const { - std::vector<GPUVAddr> results; - for (const auto& region : mapped_regions) { - if (cpu_addr >= region.cpu_addr && cpu_addr < (region.cpu_addr + region.size)) { - const u64 offset{cpu_addr - region.cpu_addr}; - results.push_back(region.gpu_addr + offset); - } - } - return results; +u8 MemoryManager::Read8(GPUVAddr addr) { + return Memory::Read8(*GpuToCpuAddress(addr)); +} + +u16 MemoryManager::Read16(GPUVAddr addr) { + return Memory::Read16(*GpuToCpuAddress(addr)); +} + +u32 MemoryManager::Read32(GPUVAddr addr) { + return Memory::Read32(*GpuToCpuAddress(addr)); +} + +u64 MemoryManager::Read64(GPUVAddr addr) { + return Memory::Read64(*GpuToCpuAddress(addr)); +} + +void MemoryManager::Write8(GPUVAddr addr, u8 data) { + Memory::Write8(*GpuToCpuAddress(addr), data); +} + +void MemoryManager::Write16(GPUVAddr addr, u16 data) { + Memory::Write16(*GpuToCpuAddress(addr), data); +} + +void MemoryManager::Write32(GPUVAddr addr, u32 data) { + Memory::Write32(*GpuToCpuAddress(addr), data); +} + +void MemoryManager::Write64(GPUVAddr addr, u64 data) { + Memory::Write64(*GpuToCpuAddress(addr), data); +} + +u8* MemoryManager::GetPointer(GPUVAddr addr) { + return Memory::GetPointer(*GpuToCpuAddress(addr)); +} + +void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) { + std::memcpy(dest_buffer, GetPointer(src_addr), size); +} +void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size) { + std::memcpy(GetPointer(dest_addr), src_buffer, size); +} + +void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size) { + std::memcpy(GetPointer(dest_addr), GetPointer(src_addr), size); } VAddr& MemoryManager::PageSlot(GPUVAddr gpu_addr) { diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index fb03497ca..425e2f31c 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -27,12 +27,27 @@ public: GPUVAddr UnmapBuffer(GPUVAddr gpu_addr, u64 size); GPUVAddr GetRegionEnd(GPUVAddr region_start) const; std::optional<VAddr> GpuToCpuAddress(GPUVAddr gpu_addr); - std::vector<GPUVAddr> CpuToGpuAddress(VAddr cpu_addr) const; static constexpr u64 PAGE_BITS = 16; static constexpr u64 PAGE_SIZE = 1 << PAGE_BITS; static constexpr u64 PAGE_MASK = PAGE_SIZE - 1; + u8 Read8(GPUVAddr addr); + u16 Read16(GPUVAddr addr); + u32 Read32(GPUVAddr addr); + u64 Read64(GPUVAddr addr); + + void Write8(GPUVAddr addr, u8 data); + void Write16(GPUVAddr addr, u16 data); + void Write32(GPUVAddr addr, u32 data); + void Write64(GPUVAddr addr, u64 data); + + u8* GetPointer(GPUVAddr vaddr); + + void ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size); + void WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size); + void CopyBlock(VAddr dest_addr, VAddr src_addr, std::size_t size); + private: enum class PageStatus : u64 { Unmapped = 0xFFFFFFFFFFFFFFFFULL, diff --git a/src/video_core/morton.cpp b/src/video_core/morton.cpp index 9692ce143..3e91cbc83 100644 --- a/src/video_core/morton.cpp +++ b/src/video_core/morton.cpp @@ -6,7 +6,6 @@ #include <cstring> #include "common/assert.h" #include "common/common_types.h" -#include "core/memory.h" #include "video_core/morton.h" #include "video_core/surface.h" #include "video_core/textures/decoders.h" @@ -16,12 +15,12 @@ namespace VideoCore { using Surface::GetBytesPerPixel; using Surface::PixelFormat; -using MortonCopyFn = void (*)(u32, u32, u32, u32, u32, u32, u8*, VAddr); +using MortonCopyFn = void (*)(u32, u32, u32, u32, u32, u32, u8*, u8*); using ConversionArray = std::array<MortonCopyFn, Surface::MaxPixelFormat>; template <bool morton_to_linear, PixelFormat format> static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, - u32 tile_width_spacing, u8* buffer, VAddr addr) { + u32 tile_width_spacing, u8* buffer, u8* addr) { constexpr u32 bytes_per_pixel = GetBytesPerPixel(format); // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual @@ -34,10 +33,10 @@ static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth stride, height, depth, block_height, block_depth, tile_width_spacing); } else { - Tegra::Texture::CopySwizzledData( - (stride + tile_size_x - 1) / tile_size_x, (height + tile_size_y - 1) / tile_size_y, - depth, bytes_per_pixel, bytes_per_pixel, Memory::GetPointer(addr), buffer, false, - block_height, block_depth, tile_width_spacing); + Tegra::Texture::CopySwizzledData((stride + tile_size_x - 1) / tile_size_x, + (height + tile_size_y - 1) / tile_size_y, depth, + bytes_per_pixel, bytes_per_pixel, addr, buffer, false, + block_height, block_depth, tile_width_spacing); } } @@ -282,7 +281,7 @@ static u32 GetMortonOffset128(u32 x, u32 y, u32 bytes_per_pixel) { void MortonSwizzle(MortonSwizzleMode mode, Surface::PixelFormat format, u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing, - u8* buffer, VAddr addr) { + u8* buffer, u8* addr) { GetSwizzleFunction(mode, format)(stride, block_height, height, block_depth, depth, tile_width_spacing, buffer, addr); } diff --git a/src/video_core/morton.h b/src/video_core/morton.h index b565204b5..ee5b45555 100644 --- a/src/video_core/morton.h +++ b/src/video_core/morton.h @@ -13,7 +13,7 @@ enum class MortonSwizzleMode { MortonToLinear, LinearToMorton }; void MortonSwizzle(MortonSwizzleMode mode, VideoCore::Surface::PixelFormat format, u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing, - u8* buffer, VAddr addr); + u8* buffer, u8* addr); void MortonCopyPixels128(MortonSwizzleMode mode, u32 width, u32 height, u32 bytes_per_pixel, u32 linear_bytes_per_pixel, u8* morton_data, u8* linear_data); diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index a4eea61a6..5048ed6ce 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -24,14 +24,12 @@ OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size) GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, std::size_t alignment, bool cache) { auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager(); - const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)}; - ASSERT_MSG(cpu_addr, "Invalid GPU address"); // Cache management is a big overhead, so only cache entries with a given size. // TODO: Figure out which size is the best for given games. cache &= size >= 2048; - const auto& host_ptr{Memory::GetPointer(*cpu_addr)}; + const auto& host_ptr{memory_manager.GetPointer(gpu_addr)}; if (cache) { auto entry = TryGet(host_ptr); if (entry) { @@ -54,8 +52,8 @@ GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size buffer_offset += size; if (cache) { - auto entry = std::make_shared<CachedBufferEntry>(*cpu_addr, size, uploaded_offset, - alignment, host_ptr); + auto entry = std::make_shared<CachedBufferEntry>( + *memory_manager.GpuToCpuAddress(gpu_addr), size, uploaded_offset, alignment, host_ptr); Register(entry); } diff --git a/src/video_core/renderer_opengl/gl_global_cache.cpp b/src/video_core/renderer_opengl/gl_global_cache.cpp index a2c509c24..c8dbcacbd 100644 --- a/src/video_core/renderer_opengl/gl_global_cache.cpp +++ b/src/video_core/renderer_opengl/gl_global_cache.cpp @@ -7,7 +7,6 @@ #include "common/assert.h" #include "common/logging/log.h" #include "core/core.h" -#include "core/memory.h" #include "video_core/renderer_opengl/gl_global_cache.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" @@ -39,7 +38,7 @@ void CachedGlobalRegion::Reload(u32 size_) { glBufferData(GL_SHADER_STORAGE_BUFFER, size, GetHostPtr(), GL_DYNAMIC_DRAW); } -GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(VAddr addr, u32 size) const { +GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const { const auto search{reserve.find(addr)}; if (search == reserve.end()) { return {}; @@ -47,11 +46,14 @@ GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(VAddr addr, u32 return search->second; } -GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 size, u8* host_ptr) { - GlobalRegion region{TryGetReservedGlobalRegion(addr, size)}; +GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size, + u8* host_ptr) { + GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)}; if (!region) { // No reserved surface available, create a new one and reserve it - region = std::make_shared<CachedGlobalRegion>(addr, size, host_ptr); + auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()}; + const auto cpu_addr = *memory_manager.GpuToCpuAddress(addr); + region = std::make_shared<CachedGlobalRegion>(cpu_addr, size, host_ptr); ReserveGlobalRegion(region); } region->Reload(size); @@ -59,7 +61,7 @@ GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 si } void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) { - reserve.insert_or_assign(region->GetCpuAddr(), std::move(region)); + reserve.insert_or_assign(region->GetCacheAddr(), std::move(region)); } GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer) @@ -70,23 +72,20 @@ GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion( Tegra::Engines::Maxwell3D::Regs::ShaderStage stage) { auto& gpu{Core::System::GetInstance().GPU()}; - const auto cbufs = gpu.Maxwell3D().state.shader_stages[static_cast<u64>(stage)]; - const auto cbuf_addr = gpu.MemoryManager().GpuToCpuAddress( - cbufs.const_buffers[global_region.GetCbufIndex()].address + global_region.GetCbufOffset()); - ASSERT(cbuf_addr); - - const auto actual_addr_gpu = Memory::Read64(*cbuf_addr); - const auto size = Memory::Read32(*cbuf_addr + 8); - const auto actual_addr = gpu.MemoryManager().GpuToCpuAddress(actual_addr_gpu); - ASSERT(actual_addr); + auto& memory_manager{gpu.MemoryManager()}; + const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<u64>(stage)]}; + const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address + + global_region.GetCbufOffset()}; + const auto actual_addr{memory_manager.Read64(addr)}; + const auto size{memory_manager.Read32(addr + 8)}; // Look up global region in the cache based on address - const auto& host_ptr{Memory::GetPointer(*actual_addr)}; + const auto& host_ptr{memory_manager.GetPointer(actual_addr)}; GlobalRegion region{TryGet(host_ptr)}; if (!region) { // No global region found - create a new one - region = GetUncachedGlobalRegion(*actual_addr, size, host_ptr); + region = GetUncachedGlobalRegion(actual_addr, size, host_ptr); Register(region); } diff --git a/src/video_core/renderer_opengl/gl_global_cache.h b/src/video_core/renderer_opengl/gl_global_cache.h index e497a0619..a840491f7 100644 --- a/src/video_core/renderer_opengl/gl_global_cache.h +++ b/src/video_core/renderer_opengl/gl_global_cache.h @@ -65,11 +65,11 @@ public: Tegra::Engines::Maxwell3D::Regs::ShaderStage stage); private: - GlobalRegion TryGetReservedGlobalRegion(VAddr addr, u32 size) const; - GlobalRegion GetUncachedGlobalRegion(VAddr addr, u32 size, u8* host_ptr); + GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const; + GlobalRegion GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size, u8* host_ptr); void ReserveGlobalRegion(GlobalRegion region); - std::unordered_map<VAddr, GlobalRegion> reserve; + std::unordered_map<CacheAddr, GlobalRegion> reserve; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_primitive_assembler.cpp b/src/video_core/renderer_opengl/gl_primitive_assembler.cpp index 77d5cedd2..75d816795 100644 --- a/src/video_core/renderer_opengl/gl_primitive_assembler.cpp +++ b/src/video_core/renderer_opengl/gl_primitive_assembler.cpp @@ -46,10 +46,7 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size auto [dst_pointer, index_offset] = buffer_cache.ReserveMemory(map_size); auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager(); - const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)}; - ASSERT_MSG(cpu_addr, "Invalid GPU address"); - - const u8* source{Memory::GetPointer(*cpu_addr)}; + const u8* source{memory_manager.GetPointer(gpu_addr)}; for (u32 primitive = 0; primitive < count / 4; ++primitive) { for (std::size_t i = 0; i < TRIANGLES_PER_QUAD; ++i) { @@ -64,4 +61,4 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size return index_offset; } -} // namespace OpenGL
\ No newline at end of file +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 451de00e8..57329cd61 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -57,11 +57,9 @@ static void ApplyTextureDefaults(GLuint texture, u32 max_mip_level) { void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) { auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()}; - const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr_)}; - addr = cpu_addr ? *cpu_addr : 0; gpu_addr = gpu_addr_; - host_ptr = Memory::GetPointer(addr); + host_ptr = memory_manager.GetPointer(gpu_addr_); size_in_bytes = SizeInBytesRaw(); if (IsPixelFormatASTC(pixel_format)) { @@ -447,7 +445,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params, MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level), params.MipBlockHeight(mip_level), params.MipHeight(mip_level), params.MipBlockDepth(mip_level), 1, params.tile_width_spacing, - gl_buffer.data() + offset_gl, params.addr + offset); + gl_buffer.data() + offset_gl, params.host_ptr + offset); offset += layer_size; offset_gl += gl_size; } @@ -456,7 +454,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params, MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level), params.MipBlockHeight(mip_level), params.MipHeight(mip_level), params.MipBlockDepth(mip_level), depth, params.tile_width_spacing, - gl_buffer.data(), params.addr + offset); + gl_buffer.data(), params.host_ptr + offset); } } @@ -514,9 +512,9 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac "reinterpretation but the texture is tiled."); } const std::size_t remaining_size = dst_params.size_in_bytes - src_params.size_in_bytes; - + auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()}; glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes, remaining_size, - Memory::GetPointer(dst_params.addr + src_params.size_in_bytes)); + memory_manager.GetPointer(dst_params.gpu_addr + src_params.size_in_bytes)); } glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); @@ -604,7 +602,7 @@ CachedSurface::CachedSurface(const SurfaceParams& params) ApplyTextureDefaults(texture.handle, params.max_mip_level); - OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.addr, params.IdentityString()); + OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.gpu_addr, params.IdentityString()); // Clamp size to mapped GPU memory region // TODO(bunnei): Super Mario Odyssey maps a 0x40000 byte region and then uses it for a 0x80000 @@ -617,6 +615,8 @@ CachedSurface::CachedSurface(const SurfaceParams& params) LOG_ERROR(HW_GPU, "Surface size {} exceeds region size {}", params.size_in_bytes, max_size); cached_size_in_bytes = max_size; } + + cpu_addr = *memory_manager.GpuToCpuAddress(params.gpu_addr); } MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 192, 64)); @@ -925,7 +925,7 @@ void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) { } Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool preserve_contents) { - if (params.addr == 0 || params.height * params.width == 0) { + if (params.gpu_addr == 0 || params.height * params.width == 0) { return {}; } @@ -979,14 +979,16 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface, const Surface& dst_surface) { const auto& init_params{src_surface->GetSurfaceParams()}; const auto& dst_params{dst_surface->GetSurfaceParams()}; - VAddr address = init_params.addr; - const std::size_t layer_size = dst_params.LayerMemorySize(); + auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()}; + Tegra::GPUVAddr address{init_params.gpu_addr}; + const std::size_t layer_size{dst_params.LayerMemorySize()}; for (u32 layer = 0; layer < dst_params.depth; layer++) { for (u32 mipmap = 0; mipmap < dst_params.max_mip_level; mipmap++) { - const VAddr sub_address = address + dst_params.GetMipmapLevelOffset(mipmap); - const Surface& copy = TryGet(Memory::GetPointer(sub_address)); - if (!copy) + const Tegra::GPUVAddr sub_address{address + dst_params.GetMipmapLevelOffset(mipmap)}; + const Surface& copy{TryGet(memory_manager.GetPointer(sub_address))}; + if (!copy) { continue; + } const auto& src_params{copy->GetSurfaceParams()}; const u32 width{std::min(src_params.width, dst_params.MipWidth(mipmap))}; const u32 height{std::min(src_params.height, dst_params.MipHeight(mipmap))}; @@ -1242,9 +1244,10 @@ static std::optional<u32> TryFindBestMipMap(std::size_t memory, const SurfacePar return {}; } -static std::optional<u32> TryFindBestLayer(VAddr addr, const SurfaceParams params, u32 mipmap) { - const std::size_t size = params.LayerMemorySize(); - VAddr start = params.addr + params.GetMipmapLevelOffset(mipmap); +static std::optional<u32> TryFindBestLayer(Tegra::GPUVAddr addr, const SurfaceParams params, + u32 mipmap) { + const std::size_t size{params.LayerMemorySize()}; + Tegra::GPUVAddr start{params.gpu_addr + params.GetMipmapLevelOffset(mipmap)}; for (u32 i = 0; i < params.depth; i++) { if (start == addr) { return {i}; @@ -1266,7 +1269,7 @@ static bool LayerFitReinterpretSurface(RasterizerCacheOpenGL& cache, const Surfa src_params.height == dst_params.MipHeight(*level) && src_params.block_height >= dst_params.MipBlockHeight(*level)) { const std::optional<u32> slot = - TryFindBestLayer(render_surface->GetCpuAddr(), dst_params, *level); + TryFindBestLayer(render_surface->GetSurfaceParams().gpu_addr, dst_params, *level); if (slot.has_value()) { glCopyImageSubData(render_surface->Texture().handle, SurfaceTargetToGL(src_params.target), 0, 0, 0, 0, diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index b3afad139..9366f47f2 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -296,7 +296,6 @@ struct SurfaceParams { bool is_array; bool srgb_conversion; // Parameters used for caching - VAddr addr; u8* host_ptr; Tegra::GPUVAddr gpu_addr; std::size_t size_in_bytes; @@ -349,7 +348,7 @@ public: explicit CachedSurface(const SurfaceParams& params); VAddr GetCpuAddr() const override { - return params.addr; + return cpu_addr; } std::size_t GetSizeInBytes() const override { @@ -433,6 +432,7 @@ private: std::size_t memory_size; bool reinterpreted = false; bool must_reload = false; + VAddr cpu_addr{}; }; class RasterizerCacheOpenGL final : public RasterizerCache<Surface> { diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 60a04e146..1ed740877 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -32,13 +32,10 @@ struct UnspecializedShader { namespace { /// Gets the address for the specified shader stage program -VAddr GetShaderAddress(Maxwell::ShaderProgram program) { - const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D(); - const auto& shader_config = gpu.regs.shader_config[static_cast<std::size_t>(program)]; - const auto address = gpu.memory_manager.GpuToCpuAddress(gpu.regs.code_address.CodeAddress() + - shader_config.offset); - ASSERT_MSG(address, "Invalid GPU address"); - return *address; +Tegra::GPUVAddr GetShaderAddress(Maxwell::ShaderProgram program) { + const auto& gpu{Core::System::GetInstance().GPU().Maxwell3D()}; + const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]}; + return gpu.regs.code_address.CodeAddress() + shader_config.offset; } /// Gets the shader program code from memory for the specified address @@ -214,11 +211,11 @@ std::set<GLenum> GetSupportedFormats() { } // namespace -CachedShader::CachedShader(VAddr guest_addr, u64 unique_identifier, +CachedShader::CachedShader(VAddr cpu_addr, u64 unique_identifier, Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache, const PrecompiledPrograms& precompiled_programs, ProgramCode&& program_code, ProgramCode&& program_code_b, u8* host_ptr) - : host_ptr{host_ptr}, guest_addr{guest_addr}, unique_identifier{unique_identifier}, + : host_ptr{host_ptr}, cpu_addr{cpu_addr}, unique_identifier{unique_identifier}, program_type{program_type}, disk_cache{disk_cache}, precompiled_programs{precompiled_programs}, RasterizerCacheObject{host_ptr} { @@ -244,11 +241,11 @@ CachedShader::CachedShader(VAddr guest_addr, u64 unique_identifier, disk_cache.SaveRaw(raw); } -CachedShader::CachedShader(VAddr guest_addr, u64 unique_identifier, +CachedShader::CachedShader(VAddr cpu_addr, u64 unique_identifier, Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache, const PrecompiledPrograms& precompiled_programs, GLShader::ProgramResult result, u8* host_ptr) - : guest_addr{guest_addr}, unique_identifier{unique_identifier}, program_type{program_type}, + : cpu_addr{cpu_addr}, unique_identifier{unique_identifier}, program_type{program_type}, disk_cache{disk_cache}, precompiled_programs{precompiled_programs}, RasterizerCacheObject{ host_ptr} { @@ -273,7 +270,7 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings)); } - LabelGLObject(GL_PROGRAM, program->handle, guest_addr); + LabelGLObject(GL_PROGRAM, program->handle, cpu_addr); } handle = program->handle; @@ -325,7 +322,7 @@ GLuint CachedShader::LazyGeometryProgram(CachedProgram& target_program, BaseBind disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings)); } - LabelGLObject(GL_PROGRAM, target_program->handle, guest_addr, debug_name); + LabelGLObject(GL_PROGRAM, target_program->handle, cpu_addr, debug_name); return target_program->handle; }; @@ -488,31 +485,31 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) { return last_shaders[static_cast<u32>(program)]; } - const VAddr program_addr{GetShaderAddress(program)}; + auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()}; + const Tegra::GPUVAddr program_addr{GetShaderAddress(program)}; // Look up shader in the cache based on address - const auto& host_ptr{Memory::GetPointer(program_addr)}; + const auto& host_ptr{memory_manager.GetPointer(program_addr)}; Shader shader{TryGet(host_ptr)}; if (!shader) { // No shader found - create a new one - const auto& host_ptr{Memory::GetPointer(program_addr)}; ProgramCode program_code{GetShaderCode(host_ptr)}; ProgramCode program_code_b; if (program == Maxwell::ShaderProgram::VertexA) { program_code_b = GetShaderCode( - Memory::GetPointer(GetShaderAddress(Maxwell::ShaderProgram::VertexB))); + memory_manager.GetPointer(GetShaderAddress(Maxwell::ShaderProgram::VertexB))); } const u64 unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b); - + const VAddr cpu_addr{*memory_manager.GpuToCpuAddress(program_addr)}; const auto found = precompiled_shaders.find(unique_identifier); if (found != precompiled_shaders.end()) { shader = - std::make_shared<CachedShader>(program_addr, unique_identifier, program, disk_cache, + std::make_shared<CachedShader>(cpu_addr, unique_identifier, program, disk_cache, precompiled_programs, found->second, host_ptr); } else { shader = std::make_shared<CachedShader>( - program_addr, unique_identifier, program, disk_cache, precompiled_programs, + cpu_addr, unique_identifier, program, disk_cache, precompiled_programs, std::move(program_code), std::move(program_code_b), host_ptr); } Register(shader); diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index 81fe716b4..fd1c85115 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -39,18 +39,18 @@ using PrecompiledShaders = std::unordered_map<u64, GLShader::ProgramResult>; class CachedShader final : public RasterizerCacheObject { public: - explicit CachedShader(VAddr guest_addr, u64 unique_identifier, + explicit CachedShader(VAddr cpu_addr, u64 unique_identifier, Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache, const PrecompiledPrograms& precompiled_programs, ProgramCode&& program_code, ProgramCode&& program_code_b, u8* host_ptr); - explicit CachedShader(VAddr guest_addr, u64 unique_identifier, + explicit CachedShader(VAddr cpu_addr, u64 unique_identifier, Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache, const PrecompiledPrograms& precompiled_programs, GLShader::ProgramResult result, u8* host_ptr); VAddr GetCpuAddr() const override { - return guest_addr; + return cpu_addr; } std::size_t GetSizeInBytes() const override { @@ -92,7 +92,7 @@ private: ShaderDiskCacheUsage GetUsage(GLenum primitive_mode, BaseBindings base_bindings) const; u8* host_ptr{}; - VAddr guest_addr{}; + VAddr cpu_addr{}; u64 unique_identifier{}; Maxwell::ShaderProgram program_type{}; ShaderDiskCacheOpenGL& disk_cache; diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index b97576309..5e3d862c6 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -164,8 +164,7 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf // Reset the screen info's display texture to its own permanent texture screen_info.display_texture = screen_info.texture.resource.handle; - Memory::RasterizerFlushVirtualRegion(framebuffer_addr, size_in_bytes, - Memory::FlushMode::Flush); + rasterizer->FlushRegion(ToCacheAddr(Memory::GetPointer(framebuffer_addr)), size_in_bytes); constexpr u32 linear_bpp = 4; VideoCore::MortonCopyPixels128(VideoCore::MortonSwizzleMode::MortonToLinear, diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index cad7340f5..995d0e068 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -6,7 +6,6 @@ #include <cstring> #include "common/alignment.h" #include "common/assert.h" -#include "core/memory.h" #include "video_core/gpu.h" #include "video_core/textures/decoders.h" #include "video_core/textures/texture.h" @@ -230,18 +229,18 @@ u32 BytesPerPixel(TextureFormat format) { } } -void UnswizzleTexture(u8* const unswizzled_data, VAddr address, u32 tile_size_x, u32 tile_size_y, +void UnswizzleTexture(u8* const unswizzled_data, u8* address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth, u32 width_spacing) { CopySwizzledData((width + tile_size_x - 1) / tile_size_x, (height + tile_size_y - 1) / tile_size_y, depth, bytes_per_pixel, - bytes_per_pixel, Memory::GetPointer(address), unswizzled_data, true, - block_height, block_depth, width_spacing); + bytes_per_pixel, address, unswizzled_data, true, block_height, block_depth, + width_spacing); } -std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y, - u32 bytes_per_pixel, u32 width, u32 height, u32 depth, - u32 block_height, u32 block_depth, u32 width_spacing) { +std::vector<u8> UnswizzleTexture(u8* address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, + u32 width, u32 height, u32 depth, u32 block_height, + u32 block_depth, u32 width_spacing) { std::vector<u8> unswizzled_data(width * height * depth * bytes_per_pixel); UnswizzleTexture(unswizzled_data.data(), address, tile_size_x, tile_size_y, bytes_per_pixel, width, height, depth, block_height, block_depth, width_spacing); @@ -249,8 +248,7 @@ std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y } void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data, - u32 block_height) { + u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height) { const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + (gob_size_x - 1)) / gob_size_x}; for (u32 line = 0; line < subrect_height; ++line) { @@ -262,17 +260,17 @@ void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 const u32 gob_address = gob_address_y + (x * bytes_per_pixel / gob_size_x) * gob_size * block_height; const u32 swizzled_offset = gob_address + table[(x * bytes_per_pixel) % gob_size_x]; - const VAddr source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel; - const VAddr dest_addr = swizzled_data + swizzled_offset; + u8* source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel; + u8* dest_addr = swizzled_data + swizzled_offset; - Memory::CopyBlock(dest_addr, source_line, bytes_per_pixel); + std::memcpy(dest_addr, source_line, bytes_per_pixel); } } } void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width, - u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data, - u32 block_height, u32 offset_x, u32 offset_y) { + u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height, + u32 offset_x, u32 offset_y) { for (u32 line = 0; line < subrect_height; ++line) { const u32 y2 = line + offset_y; const u32 gob_address_y = (y2 / (gob_size_y * block_height)) * gob_size * block_height + @@ -282,10 +280,10 @@ void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 const u32 x2 = (x + offset_x) * bytes_per_pixel; const u32 gob_address = gob_address_y + (x2 / gob_size_x) * gob_size * block_height; const u32 swizzled_offset = gob_address + table[x2 % gob_size_x]; - const VAddr dest_line = unswizzled_data + line * dest_pitch + x * bytes_per_pixel; - const VAddr source_addr = swizzled_data + swizzled_offset; + u8* dest_line = unswizzled_data + line * dest_pitch + x * bytes_per_pixel; + u8* source_addr = swizzled_data + swizzled_offset; - Memory::CopyBlock(dest_line, source_addr, bytes_per_pixel); + std::memcpy(dest_line, source_addr, bytes_per_pixel); } } } diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index 65df86890..e078fa274 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -17,14 +17,14 @@ inline std::size_t GetGOBSize() { } /// Unswizzles a swizzled texture without changing its format. -void UnswizzleTexture(u8* unswizzled_data, VAddr address, u32 tile_size_x, u32 tile_size_y, +void UnswizzleTexture(u8* unswizzled_data, u8* address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height = TICEntry::DefaultBlockHeight, u32 block_depth = TICEntry::DefaultBlockHeight, u32 width_spacing = 0); /// Unswizzles a swizzled texture without changing its format. -std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y, - u32 bytes_per_pixel, u32 width, u32 height, u32 depth, +std::vector<u8> UnswizzleTexture(u8* address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel, + u32 width, u32 height, u32 depth, u32 block_height = TICEntry::DefaultBlockHeight, u32 block_depth = TICEntry::DefaultBlockHeight, u32 width_spacing = 0); @@ -44,12 +44,11 @@ std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height /// Copies an untiled subrectangle into a tiled surface. void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width, - u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data, - u32 block_height); + u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height); /// Copies a tiled subrectangle into a linear surface. void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width, - u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data, - u32 block_height, u32 offset_x, u32 offset_y); + u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height, + u32 offset_x, u32 offset_y); } // namespace Tegra::Texture diff --git a/src/yuzu/debugger/graphics/graphics_surface.cpp b/src/yuzu/debugger/graphics/graphics_surface.cpp index 71683da8e..29f01dfb2 100644 --- a/src/yuzu/debugger/graphics/graphics_surface.cpp +++ b/src/yuzu/debugger/graphics/graphics_surface.cpp @@ -383,13 +383,12 @@ void GraphicsSurfaceWidget::OnUpdate() { // TODO: Implement a good way to visualize alpha components! QImage decoded_image(surface_width, surface_height, QImage::Format_ARGB32); - std::optional<VAddr> address = gpu.MemoryManager().GpuToCpuAddress(surface_address); // TODO(bunnei): Will not work with BCn formats that swizzle 4x4 tiles. // Needs to be fixed if we plan to use this feature more, otherwise we may remove it. auto unswizzled_data = Tegra::Texture::UnswizzleTexture( - *address, 1, 1, Tegra::Texture::BytesPerPixel(surface_format), surface_width, - surface_height, 1U); + gpu.MemoryManager().GetPointer(surface_address), 1, 1, + Tegra::Texture::BytesPerPixel(surface_format), surface_width, surface_height, 1U); auto texture_data = Tegra::Texture::DecodeTexture(unswizzled_data, surface_format, surface_width, surface_height); |