3 files changed, 165 insertions, 7 deletions

diff --git a/src/core/core_timing.cpp b/src/core/core_timing.cpp
index 6da2dcfb4..0ed6f9b19 100644
--- a/src/core/core_timing.cpp
+++ b/src/core/core_timing.cpp
@@ -13,6 +13,8 @@
 #include "common/thread.h"
 #include "core/core_timing_util.h"
 
+#pragma optoimize("", off)
+
 namespace Core::Timing {
 
 constexpr int MAX_SLICE_LENGTH = 10000;
@@ -114,7 +116,7 @@ void CoreTiming::UnscheduleEvent(const EventType* event_type, u64 userdata) {
 u64 CoreTiming::GetTicks() const {
     u64 ticks = static_cast<u64>(global_timer);
     if (!is_global_timer_sane) {
-        ticks += time_slice[current_context] - downcounts[current_context];
+        ticks += accumulated_ticks;
     }
     return ticks;
 }
@@ -124,6 +126,7 @@ u64 CoreTiming::GetIdleTicks() const {
 }
 
 void CoreTiming::AddTicks(u64 ticks) {
+    accumulated_ticks += ticks;
     downcounts[current_context] -= static_cast<s64>(ticks);
 }
 
@@ -151,7 +154,6 @@ void CoreTiming::ForceExceptionCheck(s64 cycles) {
 
     // downcount is always (much) smaller than MAX_INT so we can safely cast cycles to an int
     // here. Account for cycles already executed by adjusting the g.slice_length
-    slice_length -= downcounts[current_context] - static_cast<int>(cycles);
     downcounts[current_context] = static_cast<int>(cycles);
 }
 
@@ -172,8 +174,8 @@ std::optional<u64> CoreTiming::NextAvailableCore(const s64 needed_ticks) const {
 void CoreTiming::Advance() {
     std::unique_lock<std::mutex> guard(inner_mutex);
 
-    const int cycles_executed = time_slice[current_context] - downcounts[current_context];
-    time_slice[current_context] = std::max<s64>(0, downcounts[current_context]);
+    const int cycles_executed = accumulated_ticks;
+    time_slice[current_context] = std::max<s64>(0, time_slice[current_context] - accumulated_ticks);
     global_timer += cycles_executed;
 
     is_global_timer_sane = true;
@@ -198,6 +200,8 @@ void CoreTiming::Advance() {
         }
     }
 
+    accumulated_ticks = 0;
+
     downcounts[current_context] = time_slice[current_context];
 }
 
@@ -212,6 +216,9 @@ void CoreTiming::ResetRun() {
         s64 needed_ticks = std::min<s64>(event_queue.front().time - global_timer, MAX_SLICE_LENGTH);
         downcounts[current_context] = needed_ticks;
     }
+
+    is_global_timer_sane = false;
+    accumulated_ticks = 0;
 }
 
 void CoreTiming::Idle() {
diff --git a/src/core/core_timing.h b/src/core/core_timing.h
index ec0a6d2c0..8bba45beb 100644
--- a/src/core/core_timing.h
+++ b/src/core/core_timing.h
@@ -130,6 +130,7 @@ private:
     s64 global_timer = 0;
     s64 idled_cycles = 0;
     s64 slice_length = 0;
+    u64 accumulated_ticks = 0;
     std::array<s64, num_cpu_cores> downcounts{};
     // Slice of time assigned to each core per run.
     std::array<s64, num_cpu_cores> time_slice{};
diff --git a/src/tests/core/core_timing.cpp b/src/tests/core/core_timing.cpp
index 596a2f4aa..467eb4736 100644
--- a/src/tests/core/core_timing.cpp
+++ b/src/tests/core/core_timing.cpp
@@ -6,6 +6,7 @@
 
 #include <array>
 #include <bitset>
+#include <cstdlib>
 #include <string>
 #include "common/file_util.h"
 #include "core/core.h"
@@ -13,7 +14,7 @@
 
 // Numbers are chosen randomly to make sure the correct one is given.
 static constexpr std::array<u64, 5> CB_IDS{{42, 144, 93, 1026, UINT64_C(0xFFFF7FFFF7FFFF)}};
-static constexpr int MAX_SLICE_LENGTH = 20000; // Copied from CoreTiming internals
+static constexpr int MAX_SLICE_LENGTH = 10000; // Copied from CoreTiming internals
 
 static std::bitset<CB_IDS.size()> callbacks_ran_flags;
 static u64 expected_callback = 0;
@@ -28,6 +29,12 @@ void CallbackTemplate(u64 userdata, s64 cycles_late) {
     REQUIRE(lateness == cycles_late);
 }
 
+static u64 callbacks_done = 0;
+
+void EmptyCallback(u64 userdata, s64 cycles_late) {
+    ++callbacks_done;
+}
+
 struct ScopeInit final {
     ScopeInit() {
         core_timing.Initialize();
@@ -39,16 +46,159 @@ struct ScopeInit final {
     Core::Timing::CoreTiming core_timing;
 };
 
-static void AdvanceAndCheck(Core::Timing::CoreTiming& core_timing, u32 idx, int downcount,
+static void AdvanceAndCheck(Core::Timing::CoreTiming& core_timing, u32 idx, u32 context = 0,
                             int expected_lateness = 0, int cpu_downcount = 0) {
     callbacks_ran_flags = 0;
     expected_callback = CB_IDS[idx];
     lateness = expected_lateness;
 
     // Pretend we executed X cycles of instructions.
+    core_timing.SwitchContext(context);
     core_timing.AddTicks(core_timing.GetDowncount() - cpu_downcount);
     core_timing.Advance();
+    core_timing.SwitchContext((context + 1) % 4);
 
     REQUIRE(decltype(callbacks_ran_flags)().set(idx) == callbacks_ran_flags);
-    REQUIRE(downcount == core_timing.GetDowncount());
+}
+
+TEST_CASE("CoreTiming[BasicOrder]", "[core]") {
+    ScopeInit guard;
+    auto& core_timing = guard.core_timing;
+
+    Core::Timing::EventType* cb_a = core_timing.RegisterEvent("callbackA", CallbackTemplate<0>);
+    Core::Timing::EventType* cb_b = core_timing.RegisterEvent("callbackB", CallbackTemplate<1>);
+    Core::Timing::EventType* cb_c = core_timing.RegisterEvent("callbackC", CallbackTemplate<2>);
+    Core::Timing::EventType* cb_d = core_timing.RegisterEvent("callbackD", CallbackTemplate<3>);
+    Core::Timing::EventType* cb_e = core_timing.RegisterEvent("callbackE", CallbackTemplate<4>);
+
+    // Enter slice 0
+    core_timing.ResetRun();
+
+    // D -> B -> C -> A -> E
+    core_timing.SwitchContext(0);
+    core_timing.ScheduleEvent(1000, cb_a, CB_IDS[0]);
+    REQUIRE(1000 == core_timing.GetDowncount());
+    core_timing.ScheduleEvent(500, cb_b, CB_IDS[1]);
+    REQUIRE(500 == core_timing.GetDowncount());
+    core_timing.ScheduleEvent(800, cb_c, CB_IDS[2]);
+    REQUIRE(500 == core_timing.GetDowncount());
+    core_timing.ScheduleEvent(100, cb_d, CB_IDS[3]);
+    REQUIRE(100 == core_timing.GetDowncount());
+    core_timing.ScheduleEvent(1200, cb_e, CB_IDS[4]);
+    REQUIRE(100 == core_timing.GetDowncount());
+
+    AdvanceAndCheck(core_timing, 3, 0);
+    AdvanceAndCheck(core_timing, 1, 1);
+    AdvanceAndCheck(core_timing, 2, 2);
+    AdvanceAndCheck(core_timing, 0, 3);
+    AdvanceAndCheck(core_timing, 4, 0);
+}
+
+TEST_CASE("CoreTiming[FairSharing]", "[core]") {
+
+    ScopeInit guard;
+    auto& core_timing = guard.core_timing;
+
+    Core::Timing::EventType* empty_callback =
+        core_timing.RegisterEvent("empty_callback", EmptyCallback);
+
+    callbacks_done = 0;
+    u64 MAX_CALLBACKS = 10;
+    for (std::size_t i = 0; i < 10; i++) {
+        core_timing.ScheduleEvent(i * 3333U, empty_callback, 0);
+    }
+
+    const s64 advances = MAX_SLICE_LENGTH / 10;
+    core_timing.ResetRun();
+    u64 current_time = core_timing.GetTicks();
+    bool keep_running{};
+    do {
+        keep_running = false;
+        for (u32 active_core = 0; active_core < 4; ++active_core) {
+            core_timing.SwitchContext(active_core);
+            if (core_timing.CurrentContextCanRun()) {
+                core_timing.AddTicks(std::min<s64>(advances, core_timing.GetDowncount()));
+                core_timing.Advance();
+            }
+            keep_running |= core_timing.CurrentContextCanRun();
+        }
+    } while (keep_running);
+    u64 current_time_2 = core_timing.GetTicks();
+
+    REQUIRE(MAX_CALLBACKS == callbacks_done);
+    REQUIRE(current_time_2 == current_time + MAX_SLICE_LENGTH * 4);
+}
+
+TEST_CASE("Core::Timing[PredictableLateness]", "[core]") {
+    ScopeInit guard;
+    auto& core_timing = guard.core_timing;
+
+    Core::Timing::EventType* cb_a = core_timing.RegisterEvent("callbackA", CallbackTemplate<0>);
+    Core::Timing::EventType* cb_b = core_timing.RegisterEvent("callbackB", CallbackTemplate<1>);
+
+    // Enter slice 0
+    core_timing.ResetRun();
+
+    core_timing.ScheduleEvent(100, cb_a, CB_IDS[0]);
+    core_timing.ScheduleEvent(200, cb_b, CB_IDS[1]);
+
+    AdvanceAndCheck(core_timing, 0, 0, 10, -10); // (100 - 10)
+    AdvanceAndCheck(core_timing, 1, 1, 50, -50);
+}
+
+namespace ChainSchedulingTest {
+static int reschedules = 0;
+
+static void RescheduleCallback(Core::Timing::CoreTiming& core_timing, u64 userdata,
+                               s64 cycles_late) {
+    --reschedules;
+    REQUIRE(reschedules >= 0);
+    REQUIRE(lateness == cycles_late);
+
+    if (reschedules > 0) {
+        core_timing.ScheduleEvent(1000, reinterpret_cast<Core::Timing::EventType*>(userdata),
+                                  userdata);
+    }
+}
+} // namespace ChainSchedulingTest
+
+TEST_CASE("CoreTiming[ChainScheduling]", "[core]") {
+    using namespace ChainSchedulingTest;
+
+    ScopeInit guard;
+    auto& core_timing = guard.core_timing;
+
+    Core::Timing::EventType* cb_a = core_timing.RegisterEvent("callbackA", CallbackTemplate<0>);
+    Core::Timing::EventType* cb_b = core_timing.RegisterEvent("callbackB", CallbackTemplate<1>);
+    Core::Timing::EventType* cb_c = core_timing.RegisterEvent("callbackC", CallbackTemplate<2>);
+    Core::Timing::EventType* cb_rs = core_timing.RegisterEvent(
+        "callbackReschedule", [&core_timing](u64 userdata, s64 cycles_late) {
+            RescheduleCallback(core_timing, userdata, cycles_late);
+        });
+
+    // Enter slice 0
+    core_timing.ResetRun();
+
+    core_timing.ScheduleEvent(800, cb_a, CB_IDS[0]);
+    core_timing.ScheduleEvent(1000, cb_b, CB_IDS[1]);
+    core_timing.ScheduleEvent(2200, cb_c, CB_IDS[2]);
+    core_timing.ScheduleEvent(1000, cb_rs, reinterpret_cast<u64>(cb_rs));
+    REQUIRE(800 == core_timing.GetDowncount());
+
+    reschedules = 3;
+    AdvanceAndCheck(core_timing, 0, 0); // cb_a
+    AdvanceAndCheck(core_timing, 1, 1); // cb_b, cb_rs
+    REQUIRE(2 == reschedules);
+
+    core_timing.AddTicks(core_timing.GetDowncount());
+    core_timing.Advance(); // cb_rs
+    core_timing.SwitchContext(3);
+    REQUIRE(1 == reschedules);
+    REQUIRE(200 == core_timing.GetDowncount());
+
+    AdvanceAndCheck(core_timing, 2, 3); // cb_c
+
+    core_timing.AddTicks(core_timing.GetDowncount());
+    core_timing.Advance(); // cb_rs
+    REQUIRE(0 == reschedules);
 }