summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
Diffstat (limited to 'src')
-rw-r--r--src/common/CMakeLists.txt6
-rw-r--r--src/common/memory_hook.cpp (renamed from src/core/memory_hook.cpp)6
-rw-r--r--src/common/memory_hook.h (renamed from src/core/memory_hook.h)4
-rw-r--r--src/common/page_table.cpp29
-rw-r--r--src/common/page_table.h80
-rw-r--r--src/common/uint128.cpp41
-rw-r--r--src/common/uint128.h14
-rw-r--r--src/core/CMakeLists.txt2
-rw-r--r--src/core/arm/dynarmic/arm_dynarmic.cpp5
-rw-r--r--src/core/arm/dynarmic/arm_dynarmic.h4
-rw-r--r--src/core/core_timing_util.cpp6
-rw-r--r--src/core/core_timing_util.h3
-rw-r--r--src/core/hle/ipc_helpers.h30
-rw-r--r--src/core/hle/kernel/process.cpp2
-rw-r--r--src/core/hle/kernel/scheduler.cpp2
-rw-r--r--src/core/hle/kernel/thread.cpp58
-rw-r--r--src/core/hle/kernel/thread.h10
-rw-r--r--src/core/hle/kernel/vm_manager.cpp6
-rw-r--r--src/core/hle/kernel/vm_manager.h9
-rw-r--r--src/core/memory.cpp215
-rw-r--r--src/core/memory.h80
-rw-r--r--src/core/memory_setup.h19
-rw-r--r--src/tests/core/arm/arm_test_common.cpp3
-rw-r--r--src/tests/core/arm/arm_test_common.h8
-rw-r--r--src/video_core/dma_pusher.cpp7
-rw-r--r--src/video_core/engines/kepler_memory.cpp15
-rw-r--r--src/video_core/engines/maxwell_3d.cpp41
-rw-r--r--src/video_core/engines/maxwell_dma.cpp28
-rw-r--r--src/video_core/gpu.cpp13
-rw-r--r--src/video_core/memory_manager.cpp55
-rw-r--r--src/video_core/memory_manager.h17
-rw-r--r--src/video_core/morton.cpp15
-rw-r--r--src/video_core/morton.h2
-rw-r--r--src/video_core/renderer_opengl/gl_buffer_cache.cpp8
-rw-r--r--src/video_core/renderer_opengl/gl_global_cache.cpp33
-rw-r--r--src/video_core/renderer_opengl/gl_global_cache.h6
-rw-r--r--src/video_core/renderer_opengl/gl_primitive_assembler.cpp7
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp6
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer_cache.cpp39
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer_cache.h4
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.cpp37
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.h8
-rw-r--r--src/video_core/renderer_opengl/renderer_opengl.cpp3
-rw-r--r--src/video_core/textures/decoders.cpp32
-rw-r--r--src/video_core/textures/decoders.h13
-rw-r--r--src/yuzu/debugger/graphics/graphics_surface.cpp5
46 files changed, 579 insertions, 457 deletions
diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt
index 3d30f0e3e..43ae8a9e7 100644
--- a/src/common/CMakeLists.txt
+++ b/src/common/CMakeLists.txt
@@ -92,10 +92,14 @@ add_library(common STATIC
logging/text_formatter.cpp
logging/text_formatter.h
math_util.h
+ memory_hook.cpp
+ memory_hook.h
microprofile.cpp
microprofile.h
microprofileui.h
misc.cpp
+ page_table.cpp
+ page_table.h
param_package.cpp
param_package.h
quaternion.h
@@ -114,6 +118,8 @@ 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/core/memory_hook.cpp b/src/common/memory_hook.cpp
index c61c6c1fb..3986986d6 100644
--- a/src/core/memory_hook.cpp
+++ b/src/common/memory_hook.cpp
@@ -2,10 +2,10 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
-#include "core/memory_hook.h"
+#include "common/memory_hook.h"
-namespace Memory {
+namespace Common {
MemoryHook::~MemoryHook() = default;
-} // namespace Memory
+} // namespace Common
diff --git a/src/core/memory_hook.h b/src/common/memory_hook.h
index 940777107..adaa4c2c5 100644
--- a/src/core/memory_hook.h
+++ b/src/common/memory_hook.h
@@ -9,7 +9,7 @@
#include "common/common_types.h"
-namespace Memory {
+namespace Common {
/**
* Memory hooks have two purposes:
@@ -44,4 +44,4 @@ public:
};
using MemoryHookPointer = std::shared_ptr<MemoryHook>;
-} // namespace Memory
+} // namespace Common
diff --git a/src/common/page_table.cpp b/src/common/page_table.cpp
new file mode 100644
index 000000000..8eba1c3f1
--- /dev/null
+++ b/src/common/page_table.cpp
@@ -0,0 +1,29 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/page_table.h"
+
+namespace Common {
+
+PageTable::PageTable(std::size_t page_size_in_bits) : page_size_in_bits{page_size_in_bits} {}
+
+PageTable::~PageTable() = default;
+
+void PageTable::Resize(std::size_t address_space_width_in_bits) {
+ const std::size_t num_page_table_entries = 1ULL
+ << (address_space_width_in_bits - page_size_in_bits);
+
+ pointers.resize(num_page_table_entries);
+ attributes.resize(num_page_table_entries);
+
+ // The default is a 39-bit address space, which causes an initial 1GB allocation size. If the
+ // vector size is subsequently decreased (via resize), the vector might not automatically
+ // actually reallocate/resize its underlying allocation, which wastes up to ~800 MB for
+ // 36-bit titles. Call shrink_to_fit to reduce capacity to what's actually in use.
+
+ pointers.shrink_to_fit();
+ attributes.shrink_to_fit();
+}
+
+} // namespace Common
diff --git a/src/common/page_table.h b/src/common/page_table.h
new file mode 100644
index 000000000..8339f2890
--- /dev/null
+++ b/src/common/page_table.h
@@ -0,0 +1,80 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+#include <boost/icl/interval_map.hpp>
+#include "common/common_types.h"
+#include "common/memory_hook.h"
+
+namespace Common {
+
+enum class PageType : u8 {
+ /// Page is unmapped and should cause an access error.
+ Unmapped,
+ /// Page is mapped to regular memory. This is the only type you can get pointers to.
+ Memory,
+ /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
+ /// invalidation
+ RasterizerCachedMemory,
+ /// Page is mapped to a I/O region. Writing and reading to this page is handled by functions.
+ Special,
+};
+
+struct SpecialRegion {
+ enum class Type {
+ DebugHook,
+ IODevice,
+ } type;
+
+ MemoryHookPointer handler;
+
+ bool operator<(const SpecialRegion& other) const {
+ return std::tie(type, handler) < std::tie(other.type, other.handler);
+ }
+
+ bool operator==(const SpecialRegion& other) const {
+ return std::tie(type, handler) == std::tie(other.type, other.handler);
+ }
+};
+
+/**
+ * A (reasonably) fast way of allowing switchable and remappable process address spaces. It loosely
+ * mimics the way a real CPU page table works.
+ */
+struct PageTable {
+ explicit PageTable(std::size_t page_size_in_bits);
+ ~PageTable();
+
+ /**
+ * Resizes the page table to be able to accomodate enough pages within
+ * a given address space.
+ *
+ * @param address_space_width_in_bits The address size width in bits.
+ */
+ void Resize(std::size_t address_space_width_in_bits);
+
+ /**
+ * Vector of memory pointers backing each page. An entry can only be non-null if the
+ * corresponding entry in the `attributes` vector is of type `Memory`.
+ */
+ std::vector<u8*> pointers;
+
+ /**
+ * Contains MMIO handlers that back memory regions whose entries in the `attribute` vector is
+ * of type `Special`.
+ */
+ boost::icl::interval_map<VAddr, std::set<SpecialRegion>> special_regions;
+
+ /**
+ * Vector of fine grained page attributes. If it is set to any value other than `Memory`, then
+ * the corresponding entry in `pointers` MUST be set to null.
+ */
+ std::vector<PageType> attributes;
+
+ const std::size_t page_size_in_bits{};
+};
+
+} // namespace Common
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/CMakeLists.txt b/src/core/CMakeLists.txt
index 8ccb2d5f0..aee8bc27d 100644
--- a/src/core/CMakeLists.txt
+++ b/src/core/CMakeLists.txt
@@ -437,8 +437,6 @@ add_library(core STATIC
loader/xci.h
memory.cpp
memory.h
- memory_hook.cpp
- memory_hook.h
memory_setup.h
perf_stats.cpp
perf_stats.h
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/arm/dynarmic/arm_dynarmic.h b/src/core/arm/dynarmic/arm_dynarmic.h
index 6cc458296..aada1e862 100644
--- a/src/core/arm/dynarmic/arm_dynarmic.h
+++ b/src/core/arm/dynarmic/arm_dynarmic.h
@@ -12,7 +12,7 @@
#include "core/arm/exclusive_monitor.h"
#include "core/arm/unicorn/arm_unicorn.h"
-namespace Memory {
+namespace Common {
struct PageTable;
}
@@ -70,7 +70,7 @@ private:
Timing::CoreTiming& core_timing;
DynarmicExclusiveMonitor& exclusive_monitor;
- Memory::PageTable* current_page_table = nullptr;
+ Common::PageTable* current_page_table = nullptr;
};
class DynarmicExclusiveMonitor final : public ExclusiveMonitor {
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/ipc_helpers.h b/src/core/hle/ipc_helpers.h
index a1e4be070..68406eb63 100644
--- a/src/core/hle/ipc_helpers.h
+++ b/src/core/hle/ipc_helpers.h
@@ -275,6 +275,20 @@ inline void ResponseBuilder::Push(u64 value) {
}
template <>
+inline void ResponseBuilder::Push(float value) {
+ u32 integral;
+ std::memcpy(&integral, &value, sizeof(u32));
+ Push(integral);
+}
+
+template <>
+inline void ResponseBuilder::Push(double value) {
+ u64 integral;
+ std::memcpy(&integral, &value, sizeof(u64));
+ Push(integral);
+}
+
+template <>
inline void ResponseBuilder::Push(bool value) {
Push(static_cast<u8>(value));
}
@@ -416,6 +430,22 @@ inline s64 RequestParser::Pop() {
}
template <>
+inline float RequestParser::Pop() {
+ const u32 value = Pop<u32>();
+ float real;
+ std::memcpy(&real, &value, sizeof(real));
+ return real;
+}
+
+template <>
+inline double RequestParser::Pop() {
+ const u64 value = Pop<u64>();
+ float real;
+ std::memcpy(&real, &value, sizeof(real));
+ return real;
+}
+
+template <>
inline bool RequestParser::Pop() {
return Pop<u8>() != 0;
}
diff --git a/src/core/hle/kernel/process.cpp b/src/core/hle/kernel/process.cpp
index 49fced7b1..65c51003d 100644
--- a/src/core/hle/kernel/process.cpp
+++ b/src/core/hle/kernel/process.cpp
@@ -31,7 +31,7 @@ namespace {
*/
void SetupMainThread(Process& owner_process, KernelCore& kernel, VAddr entry_point, u32 priority) {
// Setup page table so we can write to memory
- SetCurrentPageTable(&owner_process.VMManager().page_table);
+ Memory::SetCurrentPageTable(&owner_process.VMManager().page_table);
// Initialize new "main" thread
const VAddr stack_top = owner_process.VMManager().GetTLSIORegionEndAddress();
diff --git a/src/core/hle/kernel/scheduler.cpp b/src/core/hle/kernel/scheduler.cpp
index e524509df..cc189cc64 100644
--- a/src/core/hle/kernel/scheduler.cpp
+++ b/src/core/hle/kernel/scheduler.cpp
@@ -96,7 +96,7 @@ void Scheduler::SwitchContext(Thread* new_thread) {
auto* const thread_owner_process = current_thread->GetOwnerProcess();
if (previous_process != thread_owner_process) {
system.Kernel().MakeCurrentProcess(thread_owner_process);
- SetCurrentPageTable(&thread_owner_process->VMManager().page_table);
+ Memory::SetCurrentPageTable(&thread_owner_process->VMManager().page_table);
}
cpu_core.LoadContext(new_thread->GetContext());
diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp
index 2e712c9cb..d9ffebc3f 100644
--- a/src/core/hle/kernel/thread.cpp
+++ b/src/core/hle/kernel/thread.cpp
@@ -7,8 +7,6 @@
#include <optional>
#include <vector>
-#include <boost/range/algorithm_ext/erase.hpp>
-
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
@@ -68,12 +66,6 @@ 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)
@@ -264,8 +256,8 @@ void Thread::AddMutexWaiter(SharedPtr<Thread> thread) {
if (thread->lock_owner == this) {
// If the thread is already waiting for this thread to release the mutex, ensure that the
// waiters list is consistent and return without doing anything.
- auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
- ASSERT(itr != wait_mutex_threads.end());
+ const auto iter = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+ ASSERT(iter != wait_mutex_threads.end());
return;
}
@@ -273,11 +265,16 @@ void Thread::AddMutexWaiter(SharedPtr<Thread> thread) {
ASSERT(thread->lock_owner == nullptr);
// Ensure that the thread is not already in the list of mutex waiters
- auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
- ASSERT(itr == wait_mutex_threads.end());
-
+ const auto iter = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+ ASSERT(iter == wait_mutex_threads.end());
+
+ // Keep the list in an ordered fashion
+ const auto insertion_point = std::find_if(
+ wait_mutex_threads.begin(), wait_mutex_threads.end(),
+ [&thread](const auto& entry) { return entry->GetPriority() > thread->GetPriority(); });
+ wait_mutex_threads.insert(insertion_point, thread);
thread->lock_owner = this;
- wait_mutex_threads.emplace_back(std::move(thread));
+
UpdatePriority();
}
@@ -285,32 +282,43 @@ void Thread::RemoveMutexWaiter(SharedPtr<Thread> thread) {
ASSERT(thread->lock_owner == this);
// Ensure that the thread is in the list of mutex waiters
- auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
- ASSERT(itr != wait_mutex_threads.end());
+ const auto iter = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+ ASSERT(iter != wait_mutex_threads.end());
+
+ wait_mutex_threads.erase(iter);
- boost::remove_erase(wait_mutex_threads, thread);
thread->lock_owner = nullptr;
UpdatePriority();
}
void Thread::UpdatePriority() {
- // Find the highest priority among all the threads that are waiting for this thread's lock
+ // If any of the threads waiting on the mutex have a higher priority
+ // (taking into account priority inheritance), then this thread inherits
+ // that thread's priority.
u32 new_priority = nominal_priority;
- for (const auto& thread : wait_mutex_threads) {
- if (thread->nominal_priority < new_priority)
- new_priority = thread->nominal_priority;
+ if (!wait_mutex_threads.empty()) {
+ if (wait_mutex_threads.front()->current_priority < new_priority) {
+ new_priority = wait_mutex_threads.front()->current_priority;
+ }
}
- if (new_priority == current_priority)
+ if (new_priority == current_priority) {
return;
+ }
scheduler->SetThreadPriority(this, new_priority);
-
current_priority = new_priority;
+ if (!lock_owner) {
+ return;
+ }
+
+ // Ensure that the thread is within the correct location in the waiting list.
+ lock_owner->RemoveMutexWaiter(this);
+ lock_owner->AddMutexWaiter(this);
+
// Recursively update the priority of the thread that depends on the priority of this one.
- if (lock_owner)
- lock_owner->UpdatePriority();
+ lock_owner->UpdatePriority();
}
void Thread::ChangeCore(u32 core, u64 mask) {
diff --git a/src/core/hle/kernel/thread.h b/src/core/hle/kernel/thread.h
index ccdefeecc..faad5f391 100644
--- a/src/core/hle/kernel/thread.h
+++ b/src/core/hle/kernel/thread.h
@@ -401,8 +401,14 @@ private:
VAddr entry_point = 0;
VAddr stack_top = 0;
- u32 nominal_priority = 0; ///< Nominal thread priority, as set by the emulated application
- u32 current_priority = 0; ///< Current thread priority, can be temporarily changed
+ /// Nominal thread priority, as set by the emulated application.
+ /// The nominal priority is the thread priority without priority
+ /// inheritance taken into account.
+ u32 nominal_priority = 0;
+
+ /// Current thread priority. This may change over the course of the
+ /// thread's lifetime in order to facilitate priority inheritance.
+ u32 current_priority = 0;
u64 total_cpu_time_ticks = 0; ///< Total CPU running ticks.
u64 last_running_ticks = 0; ///< CPU tick when thread was last running
diff --git a/src/core/hle/kernel/vm_manager.cpp b/src/core/hle/kernel/vm_manager.cpp
index 05c59af34..3def3e52c 100644
--- a/src/core/hle/kernel/vm_manager.cpp
+++ b/src/core/hle/kernel/vm_manager.cpp
@@ -7,13 +7,13 @@
#include <utility>
#include "common/assert.h"
#include "common/logging/log.h"
+#include "common/memory_hook.h"
#include "core/arm/arm_interface.h"
#include "core/core.h"
#include "core/file_sys/program_metadata.h"
#include "core/hle/kernel/errors.h"
#include "core/hle/kernel/vm_manager.h"
#include "core/memory.h"
-#include "core/memory_hook.h"
#include "core/memory_setup.h"
namespace Kernel {
@@ -177,7 +177,7 @@ ResultVal<VAddr> VMManager::FindFreeRegion(u64 size) const {
ResultVal<VMManager::VMAHandle> VMManager::MapMMIO(VAddr target, PAddr paddr, u64 size,
MemoryState state,
- Memory::MemoryHookPointer mmio_handler) {
+ Common::MemoryHookPointer mmio_handler) {
// This is the appropriately sized VMA that will turn into our allocation.
CASCADE_RESULT(VMAIter vma_handle, CarveVMA(target, size));
VirtualMemoryArea& final_vma = vma_handle->second;
@@ -624,7 +624,7 @@ void VMManager::ClearPageTable() {
std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
page_table.special_regions.clear();
std::fill(page_table.attributes.begin(), page_table.attributes.end(),
- Memory::PageType::Unmapped);
+ Common::PageType::Unmapped);
}
VMManager::CheckResults VMManager::CheckRangeState(VAddr address, u64 size, MemoryState state_mask,
diff --git a/src/core/hle/kernel/vm_manager.h b/src/core/hle/kernel/vm_manager.h
index 88e0b3c02..b96980f8f 100644
--- a/src/core/hle/kernel/vm_manager.h
+++ b/src/core/hle/kernel/vm_manager.h
@@ -9,9 +9,10 @@
#include <tuple>
#include <vector>
#include "common/common_types.h"
+#include "common/memory_hook.h"
+#include "common/page_table.h"
#include "core/hle/result.h"
#include "core/memory.h"
-#include "core/memory_hook.h"
namespace FileSys {
enum class ProgramAddressSpaceType : u8;
@@ -290,7 +291,7 @@ struct VirtualMemoryArea {
// Settings for type = MMIO
/// Physical address of the register area this VMA maps to.
PAddr paddr = 0;
- Memory::MemoryHookPointer mmio_handler = nullptr;
+ Common::MemoryHookPointer mmio_handler = nullptr;
/// Tests if this area can be merged to the right with `next`.
bool CanBeMergedWith(const VirtualMemoryArea& next) const;
@@ -368,7 +369,7 @@ public:
* @param mmio_handler The handler that will implement read and write for this MMIO region.
*/
ResultVal<VMAHandle> MapMMIO(VAddr target, PAddr paddr, u64 size, MemoryState state,
- Memory::MemoryHookPointer mmio_handler);
+ Common::MemoryHookPointer mmio_handler);
/// Unmaps a range of addresses, splitting VMAs as necessary.
ResultCode UnmapRange(VAddr target, u64 size);
@@ -509,7 +510,7 @@ public:
/// Each VMManager has its own page table, which is set as the main one when the owning process
/// is scheduled.
- Memory::PageTable page_table;
+ Common::PageTable page_table{Memory::PAGE_BITS};
private:
using VMAIter = VMAMap::iterator;
diff --git a/src/core/memory.cpp b/src/core/memory.cpp
index 4fde53033..365ac82b4 100644
--- a/src/core/memory.cpp
+++ b/src/core/memory.cpp
@@ -10,6 +10,7 @@
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
+#include "common/page_table.h"
#include "common/swap.h"
#include "core/arm/arm_interface.h"
#include "core/core.h"
@@ -18,13 +19,14 @@
#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 {
-static PageTable* current_page_table = nullptr;
+static Common::PageTable* current_page_table = nullptr;
-void SetCurrentPageTable(PageTable* page_table) {
+void SetCurrentPageTable(Common::PageTable* page_table) {
current_page_table = page_table;
auto& system = Core::System::GetInstance();
@@ -36,41 +38,19 @@ void SetCurrentPageTable(PageTable* page_table) {
}
}
-PageTable* GetCurrentPageTable() {
+Common::PageTable* GetCurrentPageTable() {
return current_page_table;
}
-PageTable::PageTable() = default;
-
-PageTable::PageTable(std::size_t address_space_width_in_bits) {
- Resize(address_space_width_in_bits);
-}
-
-PageTable::~PageTable() = default;
-
-void PageTable::Resize(std::size_t address_space_width_in_bits) {
- const std::size_t num_page_table_entries = 1ULL << (address_space_width_in_bits - PAGE_BITS);
-
- pointers.resize(num_page_table_entries);
- attributes.resize(num_page_table_entries);
-
- // The default is a 39-bit address space, which causes an initial 1GB allocation size. If the
- // vector size is subsequently decreased (via resize), the vector might not automatically
- // actually reallocate/resize its underlying allocation, which wastes up to ~800 MB for
- // 36-bit titles. Call shrink_to_fit to reduce capacity to what's actually in use.
-
- pointers.shrink_to_fit();
- attributes.shrink_to_fit();
-}
-
-static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, PageType type) {
+static void MapPages(Common::PageTable& page_table, VAddr base, u64 size, u8* memory,
+ Common::PageType type) {
LOG_DEBUG(HW_Memory, "Mapping {} onto {:016X}-{:016X}", fmt::ptr(memory), base * PAGE_SIZE,
(base + size) * PAGE_SIZE);
// 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;
@@ -91,41 +71,47 @@ static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, Pa
}
}
-void MapMemoryRegion(PageTable& page_table, VAddr base, u64 size, u8* target) {
+void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, u8* target) {
ASSERT_MSG((size & PAGE_MASK) == 0, "non-page aligned size: {:016X}", size);
ASSERT_MSG((base & PAGE_MASK) == 0, "non-page aligned base: {:016X}", base);
- MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, target, PageType::Memory);
+ MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, target, Common::PageType::Memory);
}
-void MapIoRegion(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer mmio_handler) {
+void MapIoRegion(Common::PageTable& page_table, VAddr base, u64 size,
+ Common::MemoryHookPointer mmio_handler) {
ASSERT_MSG((size & PAGE_MASK) == 0, "non-page aligned size: {:016X}", size);
ASSERT_MSG((base & PAGE_MASK) == 0, "non-page aligned base: {:016X}", base);
- MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, PageType::Special);
+ MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, Common::PageType::Special);
auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
- SpecialRegion region{SpecialRegion::Type::IODevice, std::move(mmio_handler)};
- page_table.special_regions.add(std::make_pair(interval, std::set<SpecialRegion>{region}));
+ Common::SpecialRegion region{Common::SpecialRegion::Type::IODevice, std::move(mmio_handler)};
+ page_table.special_regions.add(
+ std::make_pair(interval, std::set<Common::SpecialRegion>{region}));
}
-void UnmapRegion(PageTable& page_table, VAddr base, u64 size) {
+void UnmapRegion(Common::PageTable& page_table, VAddr base, u64 size) {
ASSERT_MSG((size & PAGE_MASK) == 0, "non-page aligned size: {:016X}", size);
ASSERT_MSG((base & PAGE_MASK) == 0, "non-page aligned base: {:016X}", base);
- MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, PageType::Unmapped);
+ MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, Common::PageType::Unmapped);
auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
page_table.special_regions.erase(interval);
}
-void AddDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook) {
+void AddDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+ Common::MemoryHookPointer hook) {
auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
- SpecialRegion region{SpecialRegion::Type::DebugHook, std::move(hook)};
- page_table.special_regions.add(std::make_pair(interval, std::set<SpecialRegion>{region}));
+ Common::SpecialRegion region{Common::SpecialRegion::Type::DebugHook, std::move(hook)};
+ page_table.special_regions.add(
+ std::make_pair(interval, std::set<Common::SpecialRegion>{region}));
}
-void RemoveDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook) {
+void RemoveDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+ Common::MemoryHookPointer hook) {
auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
- SpecialRegion region{SpecialRegion::Type::DebugHook, std::move(hook)};
- page_table.special_regions.subtract(std::make_pair(interval, std::set<SpecialRegion>{region}));
+ Common::SpecialRegion region{Common::SpecialRegion::Type::DebugHook, std::move(hook)};
+ page_table.special_regions.subtract(
+ std::make_pair(interval, std::set<Common::SpecialRegion>{region}));
}
/**
@@ -174,19 +160,19 @@ T Read(const VAddr vaddr) {
return value;
}
- PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
+ Common::PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
switch (type) {
- case PageType::Unmapped:
+ case Common::PageType::Unmapped:
LOG_ERROR(HW_Memory, "Unmapped Read{} @ 0x{:08X}", sizeof(T) * 8, vaddr);
return 0;
- case PageType::Memory:
+ case Common::PageType::Memory:
ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
break;
- case PageType::RasterizerCachedMemory: {
- RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Flush);
-
+ case Common::PageType::RasterizerCachedMemory: {
+ 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:
@@ -204,18 +190,19 @@ void Write(const VAddr vaddr, const T data) {
return;
}
- PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
+ Common::PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
switch (type) {
- case PageType::Unmapped:
+ case Common::PageType::Unmapped:
LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
static_cast<u32>(data), vaddr);
return;
- case PageType::Memory:
+ case Common::PageType::Memory:
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));
+ case Common::PageType::RasterizerCachedMemory: {
+ auto host_ptr{GetPointerFromVMA(vaddr)};
+ Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(T));
+ std::memcpy(host_ptr, &data, sizeof(T));
break;
}
default:
@@ -230,10 +217,10 @@ bool IsValidVirtualAddress(const Kernel::Process& process, const VAddr vaddr) {
if (page_pointer)
return true;
- if (page_table.attributes[vaddr >> PAGE_BITS] == PageType::RasterizerCachedMemory)
+ if (page_table.attributes[vaddr >> PAGE_BITS] == Common::PageType::RasterizerCachedMemory)
return true;
- if (page_table.attributes[vaddr >> PAGE_BITS] != PageType::Special)
+ if (page_table.attributes[vaddr >> PAGE_BITS] != Common::PageType::Special)
return false;
return false;
@@ -253,7 +240,8 @@ u8* GetPointer(const VAddr vaddr) {
return page_pointer + (vaddr & PAGE_MASK);
}
- if (current_page_table->attributes[vaddr >> PAGE_BITS] == PageType::RasterizerCachedMemory) {
+ if (current_page_table->attributes[vaddr >> PAGE_BITS] ==
+ Common::PageType::RasterizerCachedMemory) {
return GetPointerFromVMA(vaddr);
}
@@ -287,20 +275,20 @@ void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
u64 num_pages = ((vaddr + size - 1) >> PAGE_BITS) - (vaddr >> PAGE_BITS) + 1;
for (unsigned i = 0; i < num_pages; ++i, vaddr += PAGE_SIZE) {
- PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
+ Common::PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
if (cached) {
// Switch page type to cached if now cached
switch (page_type) {
- case PageType::Unmapped:
+ case Common::PageType::Unmapped:
// It is not necessary for a process to have this region mapped into its address
// space, for example, a system module need not have a VRAM mapping.
break;
- case PageType::Memory:
- page_type = PageType::RasterizerCachedMemory;
+ case Common::PageType::Memory:
+ page_type = Common::PageType::RasterizerCachedMemory;
current_page_table->pointers[vaddr >> PAGE_BITS] = nullptr;
break;
- case PageType::RasterizerCachedMemory:
+ case Common::PageType::RasterizerCachedMemory:
// There can be more than one GPU region mapped per CPU region, so it's common that
// this area is already marked as cached.
break;
@@ -310,23 +298,23 @@ void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
} else {
// Switch page type to uncached if now uncached
switch (page_type) {
- case PageType::Unmapped:
+ case Common::PageType::Unmapped:
// It is not necessary for a process to have this region mapped into its address
// space, for example, a system module need not have a VRAM mapping.
break;
- case PageType::Memory:
+ case Common::PageType::Memory:
// There can be more than one GPU region mapped per CPU region, so it's common that
// this area is already unmarked as cached.
break;
- case PageType::RasterizerCachedMemory: {
+ case Common::PageType::RasterizerCachedMemory: {
u8* pointer = GetPointerFromVMA(vaddr & ~PAGE_MASK);
if (pointer == nullptr) {
// It's possible that this function has been called while updating the pagetable
// after unmapping a VMA. In that case the underlying VMA will no longer exist,
// and we should just leave the pagetable entry blank.
- page_type = PageType::Unmapped;
+ page_type = Common::PageType::Unmapped;
} else {
- page_type = PageType::Memory;
+ page_type = Common::PageType::Memory;
current_page_table->pointers[vaddr >> PAGE_BITS] = pointer;
}
break;
@@ -338,47 +326,6 @@ 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);
}
@@ -409,24 +356,24 @@ void ReadBlock(const Kernel::Process& process, const VAddr src_addr, void* dest_
const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
switch (page_table.attributes[page_index]) {
- case PageType::Unmapped: {
+ case Common::PageType::Unmapped: {
LOG_ERROR(HW_Memory,
"Unmapped ReadBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
current_vaddr, src_addr, size);
std::memset(dest_buffer, 0, copy_amount);
break;
}
- case PageType::Memory: {
+ case Common::PageType::Memory: {
DEBUG_ASSERT(page_table.pointers[page_index]);
const u8* src_ptr = page_table.pointers[page_index] + page_offset;
std::memcpy(dest_buffer, src_ptr, copy_amount);
break;
}
- case PageType::RasterizerCachedMemory: {
- RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
- FlushMode::Flush);
- std::memcpy(dest_buffer, GetPointerFromVMA(process, current_vaddr), copy_amount);
+ case Common::PageType::RasterizerCachedMemory: {
+ 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:
@@ -473,23 +420,23 @@ void WriteBlock(const Kernel::Process& process, const VAddr dest_addr, const voi
const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
switch (page_table.attributes[page_index]) {
- case PageType::Unmapped: {
+ case Common::PageType::Unmapped: {
LOG_ERROR(HW_Memory,
"Unmapped WriteBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
current_vaddr, dest_addr, size);
break;
}
- case PageType::Memory: {
+ case Common::PageType::Memory: {
DEBUG_ASSERT(page_table.pointers[page_index]);
u8* dest_ptr = page_table.pointers[page_index] + page_offset;
std::memcpy(dest_ptr, src_buffer, copy_amount);
break;
}
- case PageType::RasterizerCachedMemory: {
- RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
- FlushMode::Invalidate);
- std::memcpy(GetPointerFromVMA(process, current_vaddr), src_buffer, copy_amount);
+ case Common::PageType::RasterizerCachedMemory: {
+ 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:
@@ -519,23 +466,23 @@ void ZeroBlock(const Kernel::Process& process, const VAddr dest_addr, const std:
const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
switch (page_table.attributes[page_index]) {
- case PageType::Unmapped: {
+ case Common::PageType::Unmapped: {
LOG_ERROR(HW_Memory,
"Unmapped ZeroBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
current_vaddr, dest_addr, size);
break;
}
- case PageType::Memory: {
+ case Common::PageType::Memory: {
DEBUG_ASSERT(page_table.pointers[page_index]);
u8* dest_ptr = page_table.pointers[page_index] + page_offset;
std::memset(dest_ptr, 0, copy_amount);
break;
}
- case PageType::RasterizerCachedMemory: {
- RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
- FlushMode::Invalidate);
- std::memset(GetPointerFromVMA(process, current_vaddr), 0, copy_amount);
+ case Common::PageType::RasterizerCachedMemory: {
+ 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:
@@ -561,23 +508,23 @@ void CopyBlock(const Kernel::Process& process, VAddr dest_addr, VAddr src_addr,
const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
switch (page_table.attributes[page_index]) {
- case PageType::Unmapped: {
+ case Common::PageType::Unmapped: {
LOG_ERROR(HW_Memory,
"Unmapped CopyBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
current_vaddr, src_addr, size);
ZeroBlock(process, dest_addr, copy_amount);
break;
}
- case PageType::Memory: {
+ case Common::PageType::Memory: {
DEBUG_ASSERT(page_table.pointers[page_index]);
const u8* src_ptr = page_table.pointers[page_index] + page_offset;
WriteBlock(process, dest_addr, src_ptr, copy_amount);
break;
}
- case PageType::RasterizerCachedMemory: {
- RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
- FlushMode::Flush);
- WriteBlock(process, dest_addr, GetPointerFromVMA(process, current_vaddr), copy_amount);
+ case Common::PageType::RasterizerCachedMemory: {
+ 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..3f60d868c 100644
--- a/src/core/memory.h
+++ b/src/core/memory.h
@@ -10,7 +10,10 @@
#include <vector>
#include <boost/icl/interval_map.hpp>
#include "common/common_types.h"
-#include "core/memory_hook.h"
+
+namespace Common {
+struct PageTable;
+}
namespace Kernel {
class Process;
@@ -26,71 +29,6 @@ constexpr std::size_t PAGE_BITS = 12;
constexpr u64 PAGE_SIZE = 1ULL << PAGE_BITS;
constexpr u64 PAGE_MASK = PAGE_SIZE - 1;
-enum class PageType : u8 {
- /// Page is unmapped and should cause an access error.
- Unmapped,
- /// Page is mapped to regular memory. This is the only type you can get pointers to.
- Memory,
- /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
- /// invalidation
- RasterizerCachedMemory,
- /// Page is mapped to a I/O region. Writing and reading to this page is handled by functions.
- Special,
-};
-
-struct SpecialRegion {
- enum class Type {
- DebugHook,
- IODevice,
- } type;
-
- MemoryHookPointer handler;
-
- bool operator<(const SpecialRegion& other) const {
- return std::tie(type, handler) < std::tie(other.type, other.handler);
- }
-
- bool operator==(const SpecialRegion& other) const {
- return std::tie(type, handler) == std::tie(other.type, other.handler);
- }
-};
-
-/**
- * A (reasonably) fast way of allowing switchable and remappable process address spaces. It loosely
- * mimics the way a real CPU page table works.
- */
-struct PageTable {
- explicit PageTable();
- explicit PageTable(std::size_t address_space_width_in_bits);
- ~PageTable();
-
- /**
- * Resizes the page table to be able to accomodate enough pages within
- * a given address space.
- *
- * @param address_space_width_in_bits The address size width in bits.
- */
- void Resize(std::size_t address_space_width_in_bits);
-
- /**
- * Vector of memory pointers backing each page. An entry can only be non-null if the
- * corresponding entry in the `attributes` vector is of type `Memory`.
- */
- std::vector<u8*> pointers;
-
- /**
- * Contains MMIO handlers that back memory regions whose entries in the `attribute` vector is
- * of type `Special`.
- */
- boost::icl::interval_map<VAddr, std::set<SpecialRegion>> special_regions;
-
- /**
- * Vector of fine grained page attributes. If it is set to any value other than `Memory`, then
- * the corresponding entry in `pointers` MUST be set to null.
- */
- std::vector<PageType> attributes;
-};
-
/// Virtual user-space memory regions
enum : VAddr {
/// Read-only page containing kernel and system configuration values.
@@ -116,8 +54,8 @@ enum : VAddr {
};
/// Currently active page table
-void SetCurrentPageTable(PageTable* page_table);
-PageTable* GetCurrentPageTable();
+void SetCurrentPageTable(Common::PageTable* page_table);
+Common::PageTable* GetCurrentPageTable();
/// Determines if the given VAddr is valid for the specified process.
bool IsValidVirtualAddress(const Kernel::Process& process, VAddr vaddr);
@@ -161,10 +99,4 @@ 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/core/memory_setup.h b/src/core/memory_setup.h
index 9a1a4f4be..5225ee8e2 100644
--- a/src/core/memory_setup.h
+++ b/src/core/memory_setup.h
@@ -5,7 +5,11 @@
#pragma once
#include "common/common_types.h"
-#include "core/memory_hook.h"
+#include "common/memory_hook.h"
+
+namespace Common {
+struct PageTable;
+}
namespace Memory {
@@ -17,7 +21,7 @@ namespace Memory {
* @param size The amount of bytes to map. Must be page-aligned.
* @param target Buffer with the memory backing the mapping. Must be of length at least `size`.
*/
-void MapMemoryRegion(PageTable& page_table, VAddr base, u64 size, u8* target);
+void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, u8* target);
/**
* Maps a region of the emulated process address space as a IO region.
@@ -26,11 +30,14 @@ void MapMemoryRegion(PageTable& page_table, VAddr base, u64 size, u8* target);
* @param size The amount of bytes to map. Must be page-aligned.
* @param mmio_handler The handler that backs the mapping.
*/
-void MapIoRegion(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer mmio_handler);
+void MapIoRegion(Common::PageTable& page_table, VAddr base, u64 size,
+ Common::MemoryHookPointer mmio_handler);
-void UnmapRegion(PageTable& page_table, VAddr base, u64 size);
+void UnmapRegion(Common::PageTable& page_table, VAddr base, u64 size);
-void AddDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook);
-void RemoveDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook);
+void AddDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+ Common::MemoryHookPointer hook);
+void RemoveDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+ Common::MemoryHookPointer hook);
} // namespace Memory
diff --git a/src/tests/core/arm/arm_test_common.cpp b/src/tests/core/arm/arm_test_common.cpp
index 6fe56833d..3e1a735c3 100644
--- a/src/tests/core/arm/arm_test_common.cpp
+++ b/src/tests/core/arm/arm_test_common.cpp
@@ -4,6 +4,7 @@
#include <algorithm>
+#include "common/page_table.h"
#include "core/core.h"
#include "core/hle/kernel/process.h"
#include "core/memory.h"
@@ -22,7 +23,7 @@ TestEnvironment::TestEnvironment(bool mutable_memory_)
std::fill(page_table->pointers.begin(), page_table->pointers.end(), nullptr);
page_table->special_regions.clear();
std::fill(page_table->attributes.begin(), page_table->attributes.end(),
- Memory::PageType::Unmapped);
+ Common::PageType::Unmapped);
Memory::MapIoRegion(*page_table, 0x00000000, 0x80000000, test_memory);
Memory::MapIoRegion(*page_table, 0x80000000, 0x80000000, test_memory);
diff --git a/src/tests/core/arm/arm_test_common.h b/src/tests/core/arm/arm_test_common.h
index 0b7539601..d145dbfcc 100644
--- a/src/tests/core/arm/arm_test_common.h
+++ b/src/tests/core/arm/arm_test_common.h
@@ -9,10 +9,10 @@
#include <vector>
#include "common/common_types.h"
+#include "common/memory_hook.h"
#include "core/hle/kernel/kernel.h"
-#include "core/memory_hook.h"
-namespace Memory {
+namespace Common {
struct PageTable;
}
@@ -58,7 +58,7 @@ public:
private:
friend struct TestMemory;
- struct TestMemory final : Memory::MemoryHook {
+ struct TestMemory final : Common::MemoryHook {
explicit TestMemory(TestEnvironment* env_) : env(env_) {}
TestEnvironment* env;
@@ -86,7 +86,7 @@ private:
bool mutable_memory;
std::shared_ptr<TestMemory> test_memory;
std::vector<WriteRecord> write_records;
- Memory::PageTable* page_table = nullptr;
+ Common::PageTable* page_table = nullptr;
Kernel::KernelCore kernel;
};
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.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index bb6de5477..198c54872 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -749,11 +749,17 @@ void RasterizerOpenGL::FlushAll() {}
void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+ if (!addr || !size) {
+ return;
+ }
res_cache.FlushRegion(addr, size);
}
void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+ if (!addr || !size) {
+ return;
+ }
res_cache.InvalidateRegion(addr, size);
shader_cache.InvalidateRegion(addr, size);
global_cache.InvalidateRegion(addr, size);
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);