early-access version 3332

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 3330.
+This is the source code for early-access 3332.
 
 ## Legal Notice
 

src/core/hle/kernel/k_condition_variable.cpp

@@ -171,7 +171,7 @@ Result KConditionVariable::WaitForAddress(Handle handle, VAddr addr, u32 value)
         R_UNLESS(owner_thread != nullptr, ResultInvalidHandle);
 
         // Update the lock.
-        cur_thread->SetAddressKey(addr, value);
+        cur_thread->SetUserAddressKey(addr, value);
         owner_thread->AddWaiter(cur_thread);
 
         // Begin waiting.

src/core/hle/kernel/k_light_lock.cpp

@@ -68,7 +68,7 @@ bool KLightLock::LockSlowPath(uintptr_t _owner, uintptr_t _cur_thread) {
 
         // Add the current thread as a waiter on the owner.
         KThread* owner_thread = reinterpret_cast<KThread*>(_owner & ~1ULL);
-        cur_thread->SetAddressKey(reinterpret_cast<uintptr_t>(std::addressof(tag)));
+        cur_thread->SetKernelAddressKey(reinterpret_cast<uintptr_t>(std::addressof(tag)));
         owner_thread->AddWaiter(cur_thread);
 
         // Begin waiting to hold the lock.

src/core/hle/kernel/k_memory_layout.h

@@ -67,9 +67,9 @@ constexpr size_t KernelPageBufferAdditionalSize = 0x33C000;
 constexpr std::size_t KernelResourceSize = KernelPageTableHeapSize + KernelInitialPageHeapSize +
                                            KernelSlabHeapSize + KernelPageBufferHeapSize;
 
-constexpr bool IsKernelAddressKey(VAddr key) {
+//! NB: Use KThread::GetAddressKeyIsKernel().
-    return KernelVirtualAddressSpaceBase <= key && key <= KernelVirtualAddressSpaceLast;
+//! See explanation for deviation of GetAddressKey.
-}
+bool IsKernelAddressKey(VAddr key) = delete;
 
 constexpr bool IsKernelAddress(VAddr address) {
     return KernelVirtualAddressSpaceBase <= address && address < KernelVirtualAddressSpaceEnd;

src/core/hle/kernel/k_thread.cpp

@@ -330,7 +330,7 @@ void KThread::Finalize() {
             KThread* const waiter = std::addressof(*it);
 
             // The thread shouldn't be a kernel waiter.
-            ASSERT(!IsKernelAddressKey(waiter->GetAddressKey()));
+            ASSERT(!waiter->GetAddressKeyIsKernel());
 
             // Clear the lock owner.
             waiter->SetLockOwner(nullptr);
@@ -763,19 +763,6 @@ void KThread::Continue() {
     KScheduler::OnThreadStateChanged(kernel, this, old_state);
 }
 
-void KThread::WaitUntilSuspended() {
-    // Make sure we have a suspend requested.
-    ASSERT(IsSuspendRequested());
-
-    // Loop until the thread is not executing on any core.
-    for (std::size_t i = 0; i < static_cast<std::size_t>(Core::Hardware::NUM_CPU_CORES); ++i) {
-        KThread* core_thread{};
-        do {
-            core_thread = kernel.Scheduler(i).GetSchedulerCurrentThread();
-        } while (core_thread == this);
-    }
-}
-
 Result KThread::SetActivity(Svc::ThreadActivity activity) {
     // Lock ourselves.
     KScopedLightLock lk(activity_pause_lock);
@@ -897,7 +884,7 @@ void KThread::AddWaiterImpl(KThread* thread) {
     }
 
     // Keep track of how many kernel waiters we have.
-    if (IsKernelAddressKey(thread->GetAddressKey())) {
+    if (thread->GetAddressKeyIsKernel()) {
         ASSERT((num_kernel_waiters++) >= 0);
         KScheduler::SetSchedulerUpdateNeeded(kernel);
     }
@@ -911,7 +898,7 @@ void KThread::RemoveWaiterImpl(KThread* thread) {
     ASSERT(kernel.GlobalSchedulerContext().IsLocked());
 
     // Keep track of how many kernel waiters we have.
-    if (IsKernelAddressKey(thread->GetAddressKey())) {
+    if (thread->GetAddressKeyIsKernel()) {
         ASSERT((num_kernel_waiters--) > 0);
         KScheduler::SetSchedulerUpdateNeeded(kernel);
     }
@@ -987,7 +974,7 @@ KThread* KThread::RemoveWaiterByKey(s32* out_num_waiters, VAddr key) {
             KThread* thread = std::addressof(*it);
 
             // Keep track of how many kernel waiters we have.
-            if (IsKernelAddressKey(thread->GetAddressKey())) {
+            if (thread->GetAddressKeyIsKernel()) {
                 ASSERT((num_kernel_waiters--) > 0);
                 KScheduler::SetSchedulerUpdateNeeded(kernel);
             }

src/core/hle/kernel/k_thread.h

@@ -214,8 +214,6 @@ public:
 
     void Continue();
 
-    void WaitUntilSuspended();
-
     constexpr void SetSyncedIndex(s32 index) {
         synced_index = index;
     }
@@ -607,13 +605,30 @@ public:
         return address_key_value;
     }
 
-    void SetAddressKey(VAddr key) {
+    [[nodiscard]] bool GetAddressKeyIsKernel() const {
-        address_key = key;
+        return address_key_is_kernel;
     }
 
-    void SetAddressKey(VAddr key, u32 val) {
+    //! NB: intentional deviation from official kernel.
+    //
+    // Separate SetAddressKey into user and kernel versions
+    // to cope with arbitrary host pointers making their way
+    // into things.
+
+    void SetUserAddressKey(VAddr key) {
+        address_key = key;
+        address_key_is_kernel = false;
+    }
+
+    void SetUserAddressKey(VAddr key, u32 val) {
         address_key = key;
         address_key_value = val;
+        address_key_is_kernel = false;
+    }
+
+    void SetKernelAddressKey(VAddr key) {
+        address_key = key;
+        address_key_is_kernel = true;
     }
 
     void ClearWaitQueue() {
@@ -772,6 +787,7 @@ private:
     bool debug_attached{};
     s8 priority_inheritance_count{};
     bool resource_limit_release_hint{};
+    bool address_key_is_kernel{};
     StackParameters stack_parameters{};
     Common::SpinLock context_guard{};
 

src/core/hle/kernel/kernel.cpp

@@ -1198,27 +1198,34 @@ void KernelCore::Suspend(bool suspended) {
     const bool should_suspend{exception_exited || suspended};
     const auto activity = should_suspend ? ProcessActivity::Paused : ProcessActivity::Runnable;
 
-    std::vector<KScopedAutoObject<KThread>> process_threads;
+    //! This refers to the application process, not the current process.
-    {
+    KScopedAutoObject<KProcess> process = CurrentProcess();
-        KScopedSchedulerLock sl{*this};
+    if (process.IsNull()) {
-
-        if (auto* process = CurrentProcess(); process != nullptr) {
-            process->SetActivity(activity);
-
-            if (!should_suspend) {
-                // Runnable now; no need to wait.
         return;
     }
 
-            for (auto* thread : process->GetThreadList()) {
+    // Set the new activity.
-                process_threads.emplace_back(thread);
+    process->SetActivity(activity);
-            }
-        }
-    }
 
-    // Wait for execution to stop.
+    // Wait for process execution to stop.
-    for (auto& thread : process_threads) {
+    bool must_wait{should_suspend};
-        thread->WaitUntilSuspended();
+
+    // KernelCore::Suspend must be called from locked context, or we
+    // could race another call to SetActivity, interfering with waiting.
+    while (must_wait) {
+        KScopedSchedulerLock sl{*this};
+
+        // Assume that all threads have finished running.
+        must_wait = false;
+
+        for (auto i = 0; i < static_cast<s32>(Core::Hardware::NUM_CPU_CORES); ++i) {
+            if (Scheduler(i).GetSchedulerCurrentThread()->GetOwnerProcess() ==
+                process.GetPointerUnsafe()) {
+                // A thread has not finished running yet.
+                // Continue waiting.
+                must_wait = true;
+            }
+        }
     }
 }
 

src/shader_recompiler/backend/glasm/emit_glasm_bitwise_conversion.cpp

@@ -43,10 +43,6 @@ void EmitBitCastU64F64(EmitContext&, IR::Inst& inst, const IR::Value& value) {
     Alias(inst, value);
 }
 
-void EmitBitCastS32F32(EmitContext&, IR::Inst& inst, const IR::Value& value) {
-    Alias(inst, value);
-}
-
 void EmitBitCastF16U16(EmitContext&, IR::Inst& inst, const IR::Value& value) {
     Alias(inst, value);
 }

src/shader_recompiler/backend/glasm/emit_glasm_instructions.h

@@ -197,7 +197,6 @@ void EmitSelectF64(EmitContext& ctx, ScalarS32 cond, Register true_value, Regist
 void EmitBitCastU16F16(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
 void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
 void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
-void EmitBitCastS32F32(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
 void EmitBitCastF16U16(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
 void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);
 void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, const IR::Value& value);

src/shader_recompiler/backend/glsl/emit_glsl_bitwise_conversion.cpp

@@ -48,10 +48,6 @@ void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value)
     ctx.AddU64("{}=doubleBitsToUint64({});", inst, value);
 }
 
-void EmitBitCastS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) {
-    ctx.AddF32("{}=ftoi({});", inst, value);
-}
-
 void EmitBitCastF16U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst) {
     NotImplemented();
 }

src/shader_recompiler/backend/glsl/emit_glsl_instructions.h

@@ -231,7 +231,6 @@ void EmitSelectF64(EmitContext& ctx, IR::Inst& inst, std::string_view cond,
 void EmitBitCastU16F16(EmitContext& ctx, IR::Inst& inst);
 void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
 void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value);
-void EmitBitCastS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
 void EmitBitCastF16U16(EmitContext& ctx, IR::Inst& inst);
 void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value);
 void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value);

src/shader_recompiler/backend/spirv/emit_spirv_bitwise_conversion.cpp

@@ -18,10 +18,6 @@ void EmitBitCastU64F64(EmitContext&) {
     throw NotImplementedException("SPIR-V Instruction");
 }
 
-void EmitBitCastS32F32(EmitContext&) {
-    throw NotImplementedException("SPIR-V Instruction");
-}
-
 void EmitBitCastF16U16(EmitContext&) {
     throw NotImplementedException("SPIR-V Instruction");
 }

src/shader_recompiler/backend/spirv/emit_spirv_instructions.h

@@ -179,7 +179,6 @@ Id EmitSelectF64(EmitContext& ctx, Id cond, Id true_value, Id false_value);
 void EmitBitCastU16F16(EmitContext& ctx);
 Id EmitBitCastU32F32(EmitContext& ctx, Id value);
 void EmitBitCastU64F64(EmitContext& ctx);
-void EmitBitCastS32F32(EmitContext& ctx);
 void EmitBitCastF16U16(EmitContext&);
 Id EmitBitCastF32U32(EmitContext& ctx, Id value);
 void EmitBitCastF64U64(EmitContext& ctx);

src/shader_recompiler/frontend/ir/ir_emitter.cpp

@@ -703,11 +703,6 @@ IR::U32 IREmitter::BitCast<IR::U32, IR::F32>(const IR::F32& value) {
     return Inst<IR::U32>(Opcode::BitCastU32F32, value);
 }
 
-template <>
-IR::S32 IREmitter::BitCast<IR::S32, IR::F32>(const IR::F32& value) {
-    return Inst<IR::S32>(Opcode::BitCastS32F32, value);
-}
-
 template <>
 IR::F32 IREmitter::BitCast<IR::F32, IR::U32>(const IR::U32& value) {
     return Inst<IR::F32>(Opcode::BitCastF32U32, value);

src/shader_recompiler/frontend/ir/opcodes.h

@@ -38,7 +38,6 @@ constexpr Type U8{Type::U8};
 constexpr Type U16{Type::U16};
 constexpr Type U32{Type::U32};
 constexpr Type U64{Type::U64};
-constexpr Type S32{Type::S32};
 constexpr Type F16{Type::F16};
 constexpr Type F32{Type::F32};
 constexpr Type F64{Type::F64};

src/shader_recompiler/frontend/ir/opcodes.inc

@@ -175,7 +175,6 @@ OPCODE(SelectF64,                                           F64,            U1,
 OPCODE(BitCastU16F16,                                       U16,            F16,                                                                            )
 OPCODE(BitCastU32F32,                                       U32,            F32,                                                                            )
 OPCODE(BitCastU64F64,                                       U64,            F64,                                                                            )
-OPCODE(BitCastS32F32,                                       S32,            F32,                                                                            )
 OPCODE(BitCastF16U16,                                       F16,            U16,                                                                            )
 OPCODE(BitCastF32U32,                                       F32,            U32,                                                                            )
 OPCODE(BitCastF64U64,                                       F64,            U64,                                                                            )

src/shader_recompiler/frontend/ir/type.h

@@ -24,22 +24,21 @@ enum class Type {
     U16 = 1 << 7,
     U32 = 1 << 8,
     U64 = 1 << 9,
-    S32 = 1 << 10,
+    F16 = 1 << 10,
-    F16 = 1 << 11,
+    F32 = 1 << 11,
-    F32 = 1 << 12,
+    F64 = 1 << 12,
-    F64 = 1 << 13,
+    U32x2 = 1 << 13,
-    U32x2 = 1 << 14,
+    U32x3 = 1 << 14,
-    U32x3 = 1 << 15,
+    U32x4 = 1 << 15,
-    U32x4 = 1 << 16,
+    F16x2 = 1 << 16,
-    F16x2 = 1 << 17,
+    F16x3 = 1 << 17,
-    F16x3 = 1 << 18,
+    F16x4 = 1 << 18,
-    F16x4 = 1 << 19,
+    F32x2 = 1 << 19,
-    F32x2 = 1 << 20,
+    F32x3 = 1 << 20,
-    F32x3 = 1 << 21,
+    F32x4 = 1 << 21,
-    F32x4 = 1 << 22,
+    F64x2 = 1 << 22,
-    F64x2 = 1 << 23,
+    F64x3 = 1 << 23,
-    F64x3 = 1 << 24,
+    F64x4 = 1 << 24,
-    F64x4 = 1 << 25,
 };
 DECLARE_ENUM_FLAG_OPERATORS(Type)
 

src/shader_recompiler/frontend/ir/value.cpp

@@ -23,8 +23,6 @@ Value::Value(u16 value) noexcept : type{Type::U16}, imm_u16{value} {}
 
 Value::Value(u32 value) noexcept : type{Type::U32}, imm_u32{value} {}
 
-Value::Value(s32 value) noexcept : type{Type::S32}, imm_s32{value} {}
-
 Value::Value(f32 value) noexcept : type{Type::F32}, imm_f32{value} {}
 
 Value::Value(u64 value) noexcept : type{Type::U64}, imm_u64{value} {}
@@ -71,7 +69,6 @@ bool Value::operator==(const Value& other) const {
         return imm_u16 == other.imm_u16;
     case Type::U32:
     case Type::F32:
-    case Type::S32:
         return imm_u32 == other.imm_u32;
     case Type::U64:
     case Type::F64:

src/shader_recompiler/frontend/ir/value.h

@@ -268,7 +268,6 @@ using U8 = TypedValue<Type::U8>;
 using U16 = TypedValue<Type::U16>;
 using U32 = TypedValue<Type::U32>;
 using U64 = TypedValue<Type::U64>;
-using S32 = TypedValue<Type::S32>;
 using F16 = TypedValue<Type::F16>;
 using F32 = TypedValue<Type::F32>;
 using F64 = TypedValue<Type::F64>;

src/shader_recompiler/ir_opt/texture_pass.cpp

@@ -486,10 +486,10 @@ void PatchTexelFetch(IR::Block& block, IR::Inst& inst, TexturePixelFormat pixel_
     const IR::F32 w(ir.CompositeExtract(new_inst, 3));
     const IR::F16F32F64 max_value(ir.Imm32(get_max_value()));
     const IR::Value converted =
-        ir.CompositeConstruct(ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(x)), max_value),
+        ir.CompositeConstruct(ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(x)), max_value),
-                              ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(y)), max_value),
+                              ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(y)), max_value),
-                              ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(z)), max_value),
+                              ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(z)), max_value),
-                              ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(w)), max_value));
+                              ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(w)), max_value));
     inst.ReplaceUsesWith(converted);
 }
 } // Anonymous namespace