Merge pull request #6324 from ps54all/fix/agc-alias-copy-store-destination

fix(agc): store an image-copied buffer copy's destination before later labels

core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp

@@ -2626,12 +2626,10 @@ VulkanDevice::CopyOutcome VulkanDevice::CopyBuffer(std::uint64_t destination, st
         }
         outcome.syncMs = std::chrono::duration<double, std::milli>(std::chrono::steady_clock::now() - syncStart).count();
     }
-    // A whole surface copied into another: a device copy between their images, whose results reach
-    // the destination's memory through the image's deferred write-back (nothing crosses to host
-    // memory for the copy itself).
     if (AliasCopy(context, destination, source, bytes)) {
         noteWriter({}, 0);
         recorder.NotePendingWrite(destination, bytes);
+        Graphics::StorageTexture::FlushPending(destination, bytes, nullptr, "buffer copy destination", Graphics::PublishScope::Whole);
         outcome.path = 3;
         return outcome;
     }

core/libs/prx/libSceAgcDriver/tests/execution/AliasCopyLabel.cpp

@@ -0,0 +1,148 @@
+#include "prx/libSceAgcDriver/Execution/include/GuestMemory.hpp"
+#include "prx/libc/include/GuestAllocations.hpp"
+#include "prx/libSceAgcDriver/Execution/include/VulkanDevice.hpp"
+#include "prx/libSceAgcDriver/Graphics/include/GuestTextureResource.hpp"
+#include "prx/libSceAgcDriver/Graphics/include/ShaderResources.hpp"
+#include "prx/libSceAgcDriver/Graphics/include/Texture.hpp"
+#include "prx/libSceAgcDriver/Graphics/include/TextureTiling.hpp"
+#include "Recompiler.hpp"
+#include "VulkanTestDevice.hpp"
+#include <algorithm>
+#include <array>
+#include <chrono>
+#include <cstdint>
+#include <cstdio>
+#include <cstring>
+#include <iostream>
+#include <limits>
+#include <mutex>
+#include <span>
+#include <string>
+#include <thread>
+#include <vector>
+
+namespace {
+
+using AgcDriver::Graphics::Require;
+using AgcDriver::Graphics::StorageTexture;
+using ShaderRecompiler::ShaderStage;
+
+constexpr std::uint32_t Threads = 8;
+constexpr std::uint32_t Side = 256;
+constexpr std::uint32_t Format32UInt = 20;
+constexpr std::uint32_t TileR64KBX = 0x1b;
+constexpr std::uint32_t Type2D = 9;
+constexpr std::size_t SurfaceBytes = 0x40000;
+constexpr std::size_t LabelOffset = 2 * SurfaceBytes;
+constexpr std::size_t BlockBytes = LabelOffset + 0x10000;
+constexpr std::uint32_t LabelValue = 0x600dcafeu;
+
+alignas(256) constexpr std::array<std::uint32_t, 7> StoreCode{
+    0x7e040300, 0x7e060280, 0x7e0202ff, 0x12345678u, 0xf0200108, 0x00020102, 0xbf810000,
+};
+
+std::uint64_t AddressOf(const void* data) {
+    return static_cast<std::uint64_t>(reinterpret_cast<std::uintptr_t>(data));
+}
+
+std::array<std::uint32_t, 8> TextureDescriptor(const std::uint8_t* texels) {
+    const auto address = AddressOf(texels);
+    return {
+        static_cast<std::uint32_t>(address >> 8u),
+        static_cast<std::uint32_t>((address >> 40u) & 0xffu) | (Format32UInt << 20u) | (((Side - 1u) & 3u) << 30u),
+        ((Side - 1u) >> 2u) | ((Side - 1u) << 14u),
+        0xfacu | (TileR64KBX << 20u) | (Type2D << 28u),
+        0u,
+        0u,
+        0u,
+        0u,
+    };
+}
+
+void Store(AgcDriver::VulkanDevice& device, const std::uint8_t* texels) {
+    std::vector<std::uint32_t> userData(16, 0u);
+    const auto texture = TextureDescriptor(texels);
+    std::copy(texture.begin(), texture.end(), userData.begin() + 8);
+    const std::span<const std::uint32_t> code(StoreCode);
+    const std::array<ShaderRecompiler::MemoryRegion, 1> memory{{{reinterpret_cast<std::uintptr_t>(code.data()), std::as_bytes(code)}}};
+    const ShaderRecompiler::ShaderComputeStageInfo compute{{Threads, 1, 1}, 0u, {false, false, false}, false, 1};
+    ShaderRecompiler::RecompileRequest request{
+        {ShaderStage::Compute, reinterpret_cast<std::uintptr_t>(code.data()), code, 0, {}},
+        {32, 0, userData, compute, std::nullopt, std::nullopt, memory},
+        device.Target(),
+        {0, 0, 0, 128}
+    };
+    request.useCache = false;
+    const auto result = ShaderRecompiler::Recompile(request);
+    device.Dispatch(result, 1, 1, 1, {}, reinterpret_cast<std::uintptr_t>(code.data()));
+    device.WaitIdle();
+}
+
+}
+
+int main() {
+    try {
+        const auto device = OpenVulkanTestDevice();
+        if (!device) return VulkanTestSkipped;
+        std::vector<std::uint8_t> storage(BlockBytes + 0x10000u);
+        auto* block = reinterpret_cast<std::uint8_t*>((reinterpret_cast<std::uintptr_t>(storage.data()) + 0xffffu) & ~std::uintptr_t{0xffffu});
+        auto* source = block;
+        auto* destination = block + SurfaceBytes;
+        auto* label = reinterpret_cast<volatile std::uint32_t*>(block + LabelOffset);
+        for (std::size_t at = 0; at < SurfaceBytes; ++at) source[at] = static_cast<std::uint8_t>(0x31u + at % 67u);
+        std::memset(destination, 0xa5, SurfaceBytes);
+        *label = 0;
+        {
+            GuestAllocations::Mutation mutation;
+            mutation.Add(block, BlockBytes, true, true, true);
+        }
+        const auto surfaceBytes = AgcDriver::Graphics::DescribeSurface(AgcDriver::Graphics::DecodeTextureResource(TextureDescriptor(source))).guestBytes;
+        Require(surfaceBytes == SurfaceBytes, "the 256x256 32-bit surface must cover exactly its 256 KiB block");
+
+        Store(*device, source);
+        Require(StorageTexture::FindPending(AddressOf(source), SurfaceBytes) != nullptr, "the compute store left no pending image over the source");
+
+        {
+            std::lock_guard lock(AgcDriver::GuestMemory::GpuMutex());
+            const auto outcome = device->CopyBuffer(AddressOf(destination), AddressOf(source), SurfaceBytes, 0, std::numeric_limits<std::size_t>::max(), 0, 0, 0, [](std::span<const std::byte>, std::uint64_t) {});
+            if (outcome.path != 3) {
+                std::printf("skipped, the copy took path %d instead of the image copy\n", outcome.path);
+                return VulkanTestSkipped;
+            }
+            const std::uint32_t value = LabelValue;
+            const int reason = device->WriteLabelOnGpu(AddressOf(const_cast<std::uint32_t*>(label)), std::as_bytes(std::span(&value, 1)), 1, 0);
+            if (reason != 0) {
+                std::printf("skipped, the label was not recorded on the GPU (reason %d)\n", reason);
+                return VulkanTestSkipped;
+            }
+            device->SubmitRecorded(true);
+        }
+        const auto deadline = std::chrono::steady_clock::now() + std::chrono::seconds(10);
+        while (*label != LabelValue) {
+            Require(std::chrono::steady_clock::now() < deadline, "the label after the image copy never landed");
+            std::this_thread::yield();
+        }
+        std::vector<std::uint8_t> seen(destination, destination + SurfaceBytes);
+
+        std::vector<std::byte> expected(SurfaceBytes);
+        AgcDriver::GuestMemory::Read(AddressOf(source), expected);
+        std::size_t differences = 0;
+        for (std::size_t at = 0; at < SurfaceBytes; ++at) differences += static_cast<std::byte>(seen[at]) != expected[at];
+        Require(differences == 0, "a CPU read after the label that followed the image copy found " + std::to_string(differences) + " of " + std::to_string(SurfaceBytes) + " destination bytes still unwritten");
+
+        {
+            std::lock_guard lock(AgcDriver::GuestMemory::GpuMutex());
+            AgcDriver::Graphics::FlushCachedTextures(device->Device());
+        }
+        device->WaitIdle();
+        {
+            GuestAllocations::Mutation mutation;
+            mutation.Remove(block);
+        }
+        std::puts("alias copy label tests passed");
+        return 0;
+    } catch (const std::exception& error) {
+        std::cerr << error.what() << '\n';
+        return 1;
+    }
+}