Merge pull request #150 from shadps4-emu/stabilization_one

video_core: various fixes

src/core/libraries/libc/libc_math.cpp

@@ -2,6 +2,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cmath>
+#include "common/assert.h"
 #include "core/libraries/libc/libc_math.h"
 
 namespace Libraries::LibC {
@@ -38,8 +39,13 @@ double PS4_SYSV_ABI ps4__Sin(double x) {
     return sin(x);
 }
 
-float PS4_SYSV_ABI ps4__Fsin(float arg) {
+float PS4_SYSV_ABI ps4__Fsin(float arg, unsigned int m, int n) {
+    ASSERT(n == 0);
+    if (m != 0) {
+        return cosf(arg);
+    } else {
         return sinf(arg);
+    }
 }
 
 double PS4_SYSV_ABI ps4_exp2(double arg) {

src/core/libraries/libc/libc_math.h

@@ -13,7 +13,7 @@ float PS4_SYSV_ABI ps4_tanf(float num);
 float PS4_SYSV_ABI ps4_asinf(float num);
 double PS4_SYSV_ABI ps4_pow(double base, double exponent);
 double PS4_SYSV_ABI ps4__Sin(double x);
-float PS4_SYSV_ABI ps4__Fsin(float arg);
+float PS4_SYSV_ABI ps4__Fsin(float arg, unsigned int, int);
 double PS4_SYSV_ABI ps4_exp2(double arg);
 float PS4_SYSV_ABI ps4_powf(float x, float y);
 float PS4_SYSV_ABI ps4_roundf(float arg);

src/core/libraries/videoout/driver.h

@@ -18,7 +18,7 @@ struct VideoOutPort {
     bool is_open = false;
     SceVideoOutResolutionStatus resolution;
     std::array<VideoOutBuffer, MaxDisplayBuffers> buffer_slots;
-    std::array<uintptr_t, MaxDisplayBuffers> buffer_labels; // should be contiguous in memory
+    std::array<u64, MaxDisplayBuffers> buffer_labels; // should be contiguous in memory
     static_assert(sizeof(buffer_labels[0]) == 8u);
     std::array<BufferAttributeGroup, MaxDisplayBufferGroups> groups;
     FlipStatus flip_status;

src/core/libraries/videoout/video_out.cpp

@@ -251,6 +251,7 @@ s32 sceVideoOutSubmitEopFlip(s32 handle, u32 buf_id, u32 mode, u32 arg, void** u
     Platform::IrqC::Instance()->RegisterOnce(
         Platform::InterruptId::GfxFlip, [=](Platform::InterruptId irq) {
             ASSERT_MSG(irq == Platform::InterruptId::GfxFlip, "An unexpected IRQ occured");
+            ASSERT_MSG(port->buffer_labels[buf_id] == 1, "Out of order flip IRQ");
             const auto result = driver->SubmitFlip(port, buf_id, arg, true);
             ASSERT_MSG(result, "EOP flip submission failed");
         });

src/core/platform.h

@@ -66,7 +66,7 @@ struct IrqController {
             h(irq);
         }
 
-        while (!ctx.one_time_subscribers.empty()) {
+        if (!ctx.one_time_subscribers.empty()) {
             const auto& h = ctx.one_time_subscribers.front();
             h(irq);
 

src/video_core/amdgpu/liverpool.cpp

@@ -18,17 +18,16 @@ Liverpool::Liverpool() {
 
 Liverpool::~Liverpool() {
     process_thread.request_stop();
-    cv_submit.notify_one();
+    num_submits = -1;
+    num_submits.notify_one();
+    process_thread.join();
 }
 
 void Liverpool::Process(std::stop_token stoken) {
     Common::SetCurrentThreadName("GPU_CommandProcessor");
 
     while (!stoken.stop_requested()) {
-        {
+        num_submits.wait(0);
-            std::unique_lock lock{m_submit};
-            cv_submit.wait(lock, stoken, [this]() { return num_submits != 0; });
-        }
 
         if (stoken.stop_requested()) {
             break;
@@ -62,13 +61,14 @@ void Liverpool::Process(std::stop_token stoken) {
                 --num_submits;
             }
         }
-        cv_complete.notify_all(); // Notify GPU idle
+        num_submits.notify_all();
     }
 }
 
 void Liverpool::WaitGpuIdle() {
-    std::unique_lock lock{m_submit};
+    while (const auto old = num_submits.load()) {
-    cv_complete.wait(lock, [this]() { return num_submits == 0; });
+        num_submits.wait(old);
+    }
 }
 
 Liverpool::Task Liverpool::ProcessCeUpdate(std::span<const u32> ccb) {
@@ -308,11 +308,8 @@ void Liverpool::SubmitGfx(std::span<const u32> dcb, std::span<const u32> ccb) {
         queue.submits.emplace(task.handle);
     }
 
-    {
-        std::unique_lock lock{m_submit};
     ++num_submits;
-    }
+    num_submits.notify_one();
-    cv_submit.notify_one();
 }
 
 void Liverpool::SubmitAsc(u32 vqid, std::span<const u32> acb) {
@@ -325,11 +322,8 @@ void Liverpool::SubmitAsc(u32 vqid, std::span<const u32> acb) {
         queue.submits.emplace(task.handle);
     }
 
-    {
-        std::unique_lock lock{m_submit};
     ++num_submits;
-    }
+    num_submits.notify_one();
-    cv_submit.notify_one();
 }
 
 } // namespace AmdGpu

src/video_core/amdgpu/liverpool.h

@@ -771,9 +771,6 @@ private:
 
     Vulkan::Rasterizer* rasterizer{};
     std::jthread process_thread{};
-    std::condition_variable_any cv_submit{};
-    std::condition_variable cv_complete{};
-    std::mutex m_submit{};
     std::atomic<u32> num_submits{};
 };
 

src/video_core/renderer_vulkan/vk_instance.cpp

@@ -206,6 +206,7 @@ bool Instance::CreateDevice() {
         },
         vk::PhysicalDeviceVulkan13Features{
             .dynamicRendering = true,
+            .maintenance4 = true,
         },
         vk::PhysicalDeviceCustomBorderColorFeaturesEXT{
             .customBorderColors = true,
@@ -214,9 +215,6 @@ bool Instance::CreateDevice() {
         vk::PhysicalDeviceIndexTypeUint8FeaturesEXT{
             .indexTypeUint8 = true,
         },
-        vk::PhysicalDeviceMaintenance4Features{
-            .maintenance4 = true,
-        },
     };
 
     if (!index_type_uint8) {

src/video_core/texture_cache/image.h

@@ -53,9 +53,14 @@ struct UniqueImage {
     UniqueImage(const UniqueImage&) = delete;
     UniqueImage& operator=(const UniqueImage&) = delete;
 
-    UniqueImage(UniqueImage&& other) : image{std::exchange(other.image, VK_NULL_HANDLE)} {}
+    UniqueImage(UniqueImage&& other)
+        : image{std::exchange(other.image, VK_NULL_HANDLE)},
+          allocator{std::exchange(other.allocator, VK_NULL_HANDLE)},
+          allocation{std::exchange(other.allocation, VK_NULL_HANDLE)} {}
     UniqueImage& operator=(UniqueImage&& other) {
         image = std::exchange(other.image, VK_NULL_HANDLE);
+        allocator = std::exchange(other.allocator, VK_NULL_HANDLE);
+        allocation = std::exchange(other.allocation, VK_NULL_HANDLE);
         return *this;
     }