Merge pull request #8896 from teepean/perf/recent-host-imports

perf(agc): find a host import among the last ones found first

core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp

@@ -297,6 +297,10 @@ struct HostImports {
     std::uint8_t reportNoted = 0;
     std::map<std::uint64_t, std::uint64_t> spaceUses;
     std::uint64_t budgetRefusals = 0;
+    std::array<const HostImport*, 4> recentImports{};
+    std::uint64_t recentEpoch = std::numeric_limits<std::uint64_t>::max();
+    std::size_t recentSize = 0;
+    std::size_t recentNext = 0;
 };
 
 HostImports& Imports() {
@@ -304,6 +308,31 @@ HostImports& Imports() {
     return imports;
 }
 
+bool recentImportsEnabled() {
+    static const bool enabled = std::getenv("APS5_NO_RECENT_IMPORTS") == nullptr;
+    return enabled;
+}
+
+const HostImport* recentImport(HostImports& state, std::uint64_t begin, std::uint64_t end) {
+    if (state.recentEpoch != state.epoch || state.recentSize != state.imports.size()) return nullptr;
+    for (const auto* entry : state.recentImports) {
+        if (entry != nullptr && begin >= entry->base && end <= entry->base + entry->bytes) {
+            entry->lastUse = ++state.useCounter;
+            return entry;
+        }
+    }
+    return nullptr;
+}
+
+void rememberImport(HostImports& state, const HostImport* entry) {
+    if (state.recentEpoch != state.epoch || state.recentSize != state.imports.size()) {
+        state.recentImports = {};
+        state.recentEpoch = state.epoch;
+        state.recentSize = state.imports.size();
+    }
+    state.recentImports[state.recentNext++ % state.recentImports.size()] = entry;
+}
+
 #ifdef _WIN32
 void waitSettleBatch(const ImportSettle& settle, VkDevice device, const SettleBatch& batch) {
     if (batch.busy && settle.waitForFences != nullptr) Check(settle.waitForFences(device, 1, &batch.fence, VK_TRUE, 60'000'000'000ull), "vkWaitForFences (settle batch)");
@@ -4023,8 +4052,15 @@ const HostImport* HostImportFor(const Context& context, std::uint64_t address, s
 #endif
     // A hit is only valid while the registry has not changed since the imports were reconciled.
     if (!importsStale(context, state)) {
+        if (recentImportsEnabled()) {
+            if (const auto* entry = recentImport(state, address, address + bytes)) {
+                entry->lastUse = ++state.useCounter;
+                return entry;
+            }
+        }
         if (const auto* entry = findImport(state, address, address + bytes)) {
             entry->lastUse = ++state.useCounter;
+            if (recentImportsEnabled()) rememberImport(state, entry);
             return entry;
         }
     }

core/libs/prx/libSceAgcDriver/tests/HostImportRetireHold.cpp

@@ -360,6 +360,23 @@ void BusyRetireOutlivesALaterBatch() {
     Expect(!Live(retired), "the dropped import outlived every batch of its hold");
 }
 
+void RecentImportFollowsReplacement() {
+    Heap heap;
+    ActiveRecorder active;
+    const auto context = mockContext();
+    std::lock_guard gpu(AgcDriver::GuestMemory::GpuMutex());
+    const auto* first = HostImportFor(context, heap.Address(), 4096);
+    Expect(first != nullptr, "the first range was not imported");
+    if (first == nullptr) return;
+    Expect(HostImportFor(context, heap.Address(), 4096) == first, "a repeated lookup did not find the same import");
+    const auto retired = first->buffer;
+    heap.Grow();
+    const auto* grown = HostImportFor(context, heap.Address(RangeBytes), 4096);
+    Expect(grown != nullptr && grown->buffer != retired, "the grown run was not imported as a new import");
+    if (grown == nullptr) return;
+    Expect(HostImportFor(context, heap.Address(), 4096) == grown, "a lookup after the replacement served the retired import instead of the one that replaced it");
+}
+
 }
 
 int main() {
@@ -374,6 +391,7 @@ int main() {
         IdleRetireOutlivesTheHold();
         BusyRetireOutlivesALaterBatch();
         AgcDriver::Tests::RunHostImportFailureTests(mockContext(), false);
+        RecentImportFollowsReplacement();
     } catch (const std::exception& error) {
         std::fprintf(stderr, "%s\n", error.what());
         return 1;

docs/dev/ENVIRONMENT.md

@@ -305,6 +305,7 @@ The code reads environment names through `getenv`, `IsEnvironmentSet`, `kibSetti
 | `APS5_NO_PROC_TABLE` | presence | `core/libs/prx/libSceAgcDriver/Execution/src/VulkanDevice.cpp` |  | resolves the device's Vulkan entry points per call instead of filling the function table once |
 | `APS5_NO_QUEUED_DRAW_READ_ORDER` | presence | `core/libs/prx/libSceAgcDriver/Execution/src/Driver/Draw/DrawPipeline.cpp` |  | stops queued draws and DMA copies from holding back write-backs over the guest memory they read at commit |
 | `APS5_NO_QUEUED_LABELS` | presence | `core/libs/prx/libSceAgcDriver/Execution/src/Driver/Synchronization/DeferredLabels.cpp` |  | disables the queued-label table |
+| `APS5_NO_RECENT_IMPORTS` | presence | `core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp` |  | looks every host import up in the import map instead of first checking the last four found while the map is unchanged |
 | `APS5_NO_RECORDED_CLEAR` | presence | `core/libs/prx/libSceAgcDriver/Graphics/src/Texture.cpp` |  | waits for a storage image's fast clear in a batch of its own instead of recording it |
 | `APS5_NO_RECORDER_SYNC_DRAWS` | presence | `core/libs/prx/libSceAgcDriver/Graphics/src/Draw.cpp` |  | gives the draws the recorder waits for their own batch instead of recording them with the wait |
 | `APS5_NO_RECORD_COPIED_DRAWS` | presence | `core/libs/prx/libSceAgcDriver/Graphics/src/Draw.cpp` |  | waits for copied-write draws at once instead of listing their write-back for later |