8 files changed, 74 insertions, 41 deletions

diff --git a/src/common/input.h b/src/common/input.h
index ea30770ae..2c4ccea22 100644
--- a/src/common/input.h
+++ b/src/common/input.h
@@ -75,8 +75,10 @@ enum class DriverResult {
     ErrorWritingData,
     NoDeviceDetected,
     InvalidHandle,
+    InvalidParameters,
     NotSupported,
     Disabled,
+    Delayed,
     Unknown,
 };
 
diff --git a/src/common/ring_buffer.h b/src/common/ring_buffer.h
index 416680d44..5c961b202 100644
--- a/src/common/ring_buffer.h
+++ b/src/common/ring_buffer.h
@@ -54,7 +54,7 @@ public:
         return push_count;
     }
 
-    std::size_t Push(const std::span<T> input) {
+    std::size_t Push(std::span<const T> input) {
         return Push(input.data(), input.size());
     }
 
diff --git a/src/common/scratch_buffer.h b/src/common/scratch_buffer.h
index 6fe907953..d5961b020 100644
--- a/src/common/scratch_buffer.h
+++ b/src/common/scratch_buffer.h
@@ -5,7 +5,6 @@
 
 #include <iterator>
 
-#include "common/concepts.h"
 #include "common/make_unique_for_overwrite.h"
 
 namespace Common {
@@ -19,15 +18,22 @@ namespace Common {
 template <typename T>
 class ScratchBuffer {
 public:
-    using iterator = T*;
-    using const_iterator = const T*;
-    using value_type = T;
     using element_type = T;
-    using iterator_category = std::contiguous_iterator_tag;
+    using value_type = T;
+    using size_type = size_t;
+    using difference_type = std::ptrdiff_t;
+    using pointer = T*;
+    using const_pointer = const T*;
+    using reference = T&;
+    using const_reference = const T&;
+    using iterator = pointer;
+    using const_iterator = const_pointer;
+    using iterator_category = std::random_access_iterator_tag;
+    using iterator_concept = std::contiguous_iterator_tag;
 
     ScratchBuffer() = default;
 
-    explicit ScratchBuffer(size_t initial_capacity)
+    explicit ScratchBuffer(size_type initial_capacity)
         : last_requested_size{initial_capacity}, buffer_capacity{initial_capacity},
           buffer{Common::make_unique_for_overwrite<T[]>(initial_capacity)} {}
 
@@ -39,7 +45,7 @@ public:
 
     /// This will only grow the buffer's capacity if size is greater than the current capacity.
     /// The previously held data will remain intact.
-    void resize(size_t size) {
+    void resize(size_type size) {
         if (size > buffer_capacity) {
             auto new_buffer = Common::make_unique_for_overwrite<T[]>(size);
             std::move(buffer.get(), buffer.get() + buffer_capacity, new_buffer.get());
@@ -51,7 +57,7 @@ public:
 
     /// This will only grow the buffer's capacity if size is greater than the current capacity.
     /// The previously held data will be destroyed if a reallocation occurs.
-    void resize_destructive(size_t size) {
+    void resize_destructive(size_type size) {
         if (size > buffer_capacity) {
             buffer_capacity = size;
             buffer = Common::make_unique_for_overwrite<T[]>(buffer_capacity);
@@ -59,43 +65,43 @@ public:
         last_requested_size = size;
     }
 
-    [[nodiscard]] T* data() noexcept {
+    [[nodiscard]] pointer data() noexcept {
         return buffer.get();
     }
 
-    [[nodiscard]] const T* data() const noexcept {
+    [[nodiscard]] const_pointer data() const noexcept {
         return buffer.get();
     }
 
-    [[nodiscard]] T* begin() noexcept {
+    [[nodiscard]] iterator begin() noexcept {
         return data();
     }
 
-    [[nodiscard]] const T* begin() const noexcept {
+    [[nodiscard]] const_iterator begin() const noexcept {
         return data();
     }
 
-    [[nodiscard]] T* end() noexcept {
+    [[nodiscard]] iterator end() noexcept {
         return data() + last_requested_size;
     }
 
-    [[nodiscard]] const T* end() const noexcept {
+    [[nodiscard]] const_iterator end() const noexcept {
         return data() + last_requested_size;
     }
 
-    [[nodiscard]] T& operator[](size_t i) {
+    [[nodiscard]] reference operator[](size_type i) {
         return buffer[i];
     }
 
-    [[nodiscard]] const T& operator[](size_t i) const {
+    [[nodiscard]] const_reference operator[](size_type i) const {
         return buffer[i];
     }
 
-    [[nodiscard]] size_t size() const noexcept {
+    [[nodiscard]] size_type size() const noexcept {
         return last_requested_size;
     }
 
-    [[nodiscard]] size_t capacity() const noexcept {
+    [[nodiscard]] size_type capacity() const noexcept {
         return buffer_capacity;
     }
 
@@ -106,8 +112,8 @@ public:
     }
 
 private:
-    size_t last_requested_size{};
-    size_t buffer_capacity{};
+    size_type last_requested_size{};
+    size_type buffer_capacity{};
     std::unique_ptr<T[]> buffer{};
 };
 
diff --git a/src/common/settings.h b/src/common/settings.h
index ae5ed93d8..59e96e74f 100644
--- a/src/common/settings.h
+++ b/src/common/settings.h
@@ -527,12 +527,10 @@ struct Values {
     Setting<bool> mouse_panning{false, "mouse_panning"};
     Setting<u8, true> mouse_panning_x_sensitivity{50, 1, 100, "mouse_panning_x_sensitivity"};
     Setting<u8, true> mouse_panning_y_sensitivity{50, 1, 100, "mouse_panning_y_sensitivity"};
-    Setting<u8, true> mouse_panning_deadzone_x_counterweight{
-        0, 0, 100, "mouse_panning_deadzone_x_counterweight"};
-    Setting<u8, true> mouse_panning_deadzone_y_counterweight{
-        0, 0, 100, "mouse_panning_deadzone_y_counterweight"};
-    Setting<u8, true> mouse_panning_decay_strength{22, 0, 100, "mouse_panning_decay_strength"};
-    Setting<u8, true> mouse_panning_min_decay{5, 0, 100, "mouse_panning_min_decay"};
+    Setting<u8, true> mouse_panning_deadzone_counterweight{20, 0, 100,
+                                                           "mouse_panning_deadzone_counterweight"};
+    Setting<u8, true> mouse_panning_decay_strength{18, 0, 100, "mouse_panning_decay_strength"};
+    Setting<u8, true> mouse_panning_min_decay{6, 0, 100, "mouse_panning_min_decay"};
 
     Setting<bool> mouse_enabled{false, "mouse_enabled"};
     Setting<bool> emulate_analog_keyboard{false, "emulate_analog_keyboard"};
diff --git a/src/common/telemetry.cpp b/src/common/telemetry.cpp
index 91352912d..929ed67e4 100644
--- a/src/common/telemetry.cpp
+++ b/src/common/telemetry.cpp
@@ -93,6 +93,7 @@ void AppendCPUInfo(FieldCollection& fc) {
     add_field("CPU_Extension_x64_GFNI", caps.gfni);
     add_field("CPU_Extension_x64_INVARIANT_TSC", caps.invariant_tsc);
     add_field("CPU_Extension_x64_LZCNT", caps.lzcnt);
+    add_field("CPU_Extension_x64_MONITORX", caps.monitorx);
     add_field("CPU_Extension_x64_MOVBE", caps.movbe);
     add_field("CPU_Extension_x64_PCLMULQDQ", caps.pclmulqdq);
     add_field("CPU_Extension_x64_POPCNT", caps.popcnt);
diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp
index c998b1197..780120a5b 100644
--- a/src/common/x64/cpu_detect.cpp
+++ b/src/common/x64/cpu_detect.cpp
@@ -168,6 +168,7 @@ static CPUCaps Detect() {
         __cpuid(cpu_id, 0x80000001);
         caps.lzcnt = Common::Bit<5>(cpu_id[2]);
         caps.fma4 = Common::Bit<16>(cpu_id[2]);
+        caps.monitorx = Common::Bit<29>(cpu_id[2]);
     }
 
     if (max_ex_fn >= 0x80000007) {
diff --git a/src/common/x64/cpu_detect.h b/src/common/x64/cpu_detect.h
index 8253944d6..756459417 100644
--- a/src/common/x64/cpu_detect.h
+++ b/src/common/x64/cpu_detect.h
@@ -63,6 +63,7 @@ struct CPUCaps {
     bool gfni : 1;
     bool invariant_tsc : 1;
     bool lzcnt : 1;
+    bool monitorx : 1;
     bool movbe : 1;
     bool pclmulqdq : 1;
     bool popcnt : 1;
diff --git a/src/common/x64/cpu_wait.cpp b/src/common/x64/cpu_wait.cpp
index c53dd4945..41d385f59 100644
--- a/src/common/x64/cpu_wait.cpp
+++ b/src/common/x64/cpu_wait.cpp
@@ -13,36 +13,60 @@
 
 namespace Common::X64 {
 
+namespace {
+
+// 100,000 cycles is a reasonable amount of time to wait to save on CPU resources.
+// For reference:
+// At 1 GHz, 100K cycles is 100us
+// At 2 GHz, 100K cycles is 50us
+// At 4 GHz, 100K cycles is 25us
+constexpr auto PauseCycles = 100'000U;
+
+} // Anonymous namespace
+
 #ifdef _MSC_VER
 __forceinline static void TPAUSE() {
-    // 100,000 cycles is a reasonable amount of time to wait to save on CPU resources.
-    // For reference:
-    // At 1 GHz, 100K cycles is 100us
-    // At 2 GHz, 100K cycles is 50us
-    // At 4 GHz, 100K cycles is 25us
-    static constexpr auto PauseCycles = 100'000;
-    _tpause(0, FencedRDTSC() + PauseCycles);
+    static constexpr auto RequestC02State = 0U;
+    _tpause(RequestC02State, FencedRDTSC() + PauseCycles);
+}
+
+__forceinline static void MWAITX() {
+    static constexpr auto EnableWaitTimeFlag = 1U << 1;
+    static constexpr auto RequestC1State = 0U;
+
+    // monitor_var should be aligned to a cache line.
+    alignas(64) u64 monitor_var{};
+    _mm_monitorx(&monitor_var, 0, 0);
+    _mm_mwaitx(EnableWaitTimeFlag, RequestC1State, PauseCycles);
 }
 #else
 static void TPAUSE() {
-    // 100,000 cycles is a reasonable amount of time to wait to save on CPU resources.
-    // For reference:
-    // At 1 GHz, 100K cycles is 100us
-    // At 2 GHz, 100K cycles is 50us
-    // At 4 GHz, 100K cycles is 25us
-    static constexpr auto PauseCycles = 100'000;
+    static constexpr auto RequestC02State = 0U;
     const auto tsc = FencedRDTSC() + PauseCycles;
     const auto eax = static_cast<u32>(tsc & 0xFFFFFFFF);
     const auto edx = static_cast<u32>(tsc >> 32);
-    asm volatile("tpause %0" : : "r"(0), "d"(edx), "a"(eax));
+    asm volatile("tpause %0" : : "r"(RequestC02State), "d"(edx), "a"(eax));
+}
+
+static void MWAITX() {
+    static constexpr auto EnableWaitTimeFlag = 1U << 1;
+    static constexpr auto RequestC1State = 0U;
+
+    // monitor_var should be aligned to a cache line.
+    alignas(64) u64 monitor_var{};
+    asm volatile("monitorx" : : "a"(&monitor_var), "c"(0), "d"(0));
+    asm volatile("mwaitx" : : "a"(RequestC1State), "b"(PauseCycles), "c"(EnableWaitTimeFlag));
 }
 #endif
 
 void MicroSleep() {
     static const bool has_waitpkg = GetCPUCaps().waitpkg;
+    static const bool has_monitorx = GetCPUCaps().monitorx;
 
     if (has_waitpkg) {
         TPAUSE();
+    } else if (has_monitorx) {
+        MWAITX();
     } else {
         std::this_thread::yield();
     }