5 files changed, 216 insertions, 66 deletions

diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp
index c9349a6b4..fccd2eee5 100644
--- a/src/common/x64/cpu_detect.cpp
+++ b/src/common/x64/cpu_detect.cpp
@@ -62,6 +62,17 @@ static CPUCaps Detect() {
     std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(int));
     std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(int));
     std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(int));
+    if (cpu_id[1] == 0x756e6547 && cpu_id[2] == 0x6c65746e && cpu_id[3] == 0x49656e69)
+        caps.manufacturer = Manufacturer::Intel;
+    else if (cpu_id[1] == 0x68747541 && cpu_id[2] == 0x444d4163 && cpu_id[3] == 0x69746e65)
+        caps.manufacturer = Manufacturer::AMD;
+    else if (cpu_id[1] == 0x6f677948 && cpu_id[2] == 0x656e6975 && cpu_id[3] == 0x6e65476e)
+        caps.manufacturer = Manufacturer::Hygon;
+    else
+        caps.manufacturer = Manufacturer::Unknown;
+
+    u32 family = {};
+    u32 model = {};
 
     __cpuid(cpu_id, 0x80000000);
 
@@ -73,6 +84,14 @@ static CPUCaps Detect() {
     // Detect family and other miscellaneous features
     if (max_std_fn >= 1) {
         __cpuid(cpu_id, 0x00000001);
+        family = (cpu_id[0] >> 8) & 0xf;
+        model = (cpu_id[0] >> 4) & 0xf;
+        if (family == 0xf) {
+            family += (cpu_id[0] >> 20) & 0xff;
+        }
+        if (family >= 6) {
+            model += ((cpu_id[0] >> 16) & 0xf) << 4;
+        }
 
         if ((cpu_id[3] >> 25) & 1)
             caps.sse = true;
@@ -110,6 +129,11 @@ static CPUCaps Detect() {
                 caps.bmi1 = true;
             if ((cpu_id[1] >> 8) & 1)
                 caps.bmi2 = true;
+            // Checks for AVX512F, AVX512CD, AVX512VL, AVX512DQ, AVX512BW (Intel Skylake-X/SP)
+            if ((cpu_id[1] >> 16) & 1 && (cpu_id[1] >> 28) & 1 && (cpu_id[1] >> 31) & 1 &&
+                (cpu_id[1] >> 17) & 1 && (cpu_id[1] >> 30) & 1) {
+                caps.avx512 = caps.avx2;
+            }
         }
     }
 
@@ -130,6 +154,20 @@ static CPUCaps Detect() {
             caps.fma4 = true;
     }
 
+    if (max_ex_fn >= 0x80000007) {
+        __cpuid(cpu_id, 0x80000007);
+        if (cpu_id[3] & (1 << 8)) {
+            caps.invariant_tsc = true;
+        }
+    }
+
+    if (max_std_fn >= 0x16) {
+        __cpuid(cpu_id, 0x16);
+        caps.base_frequency = cpu_id[0];
+        caps.max_frequency = cpu_id[1];
+        caps.bus_frequency = cpu_id[2];
+    }
+
     return caps;
 }
 
diff --git a/src/common/x64/cpu_detect.h b/src/common/x64/cpu_detect.h
index 20f2ba234..e3b63302e 100644
--- a/src/common/x64/cpu_detect.h
+++ b/src/common/x64/cpu_detect.h
@@ -6,8 +6,16 @@
 
 namespace Common {
 
+enum class Manufacturer : u32 {
+    Intel = 0,
+    AMD = 1,
+    Hygon = 2,
+    Unknown = 3,
+};
+
 /// x86/x64 CPU capabilities that may be detected by this module
 struct CPUCaps {
+    Manufacturer manufacturer;
     char cpu_string[0x21];
     char brand_string[0x41];
     bool sse;
@@ -19,11 +27,16 @@ struct CPUCaps {
     bool lzcnt;
     bool avx;
     bool avx2;
+    bool avx512;
     bool bmi1;
     bool bmi2;
     bool fma;
     bool fma4;
     bool aes;
+    bool invariant_tsc;
+    u32 base_frequency;
+    u32 max_frequency;
+    u32 bus_frequency;
 };
 
 /**
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
new file mode 100644
index 000000000..26d4d0ba6
--- /dev/null
+++ b/src/common/x64/native_clock.cpp
@@ -0,0 +1,95 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <chrono>
+#include <thread>
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#else
+#include <x86intrin.h>
+#endif
+
+#include "common/uint128.h"
+#include "common/x64/native_clock.h"
+
+namespace Common {
+
+u64 EstimateRDTSCFrequency() {
+    const auto milli_10 = std::chrono::milliseconds{10};
+    // get current time
+    _mm_mfence();
+    const u64 tscStart = __rdtsc();
+    const auto startTime = std::chrono::high_resolution_clock::now();
+    // wait roughly 3 seconds
+    while (true) {
+        auto milli = std::chrono::duration_cast<std::chrono::milliseconds>(
+            std::chrono::high_resolution_clock::now() - startTime);
+        if (milli.count() >= 3000)
+            break;
+        std::this_thread::sleep_for(milli_10);
+    }
+    const auto endTime = std::chrono::high_resolution_clock::now();
+    _mm_mfence();
+    const u64 tscEnd = __rdtsc();
+    // calculate difference
+    const u64 timer_diff =
+        std::chrono::duration_cast<std::chrono::nanoseconds>(endTime - startTime).count();
+    const u64 tsc_diff = tscEnd - tscStart;
+    const u64 tsc_freq = MultiplyAndDivide64(tsc_diff, 1000000000ULL, timer_diff);
+    return tsc_freq;
+}
+
+namespace X64 {
+NativeClock::NativeClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency,
+                         u64 rtsc_frequency)
+    : WallClock(emulated_cpu_frequency, emulated_clock_frequency, true), rtsc_frequency{
+                                                                             rtsc_frequency} {
+    _mm_mfence();
+    last_measure = __rdtsc();
+    accumulated_ticks = 0U;
+}
+
+u64 NativeClock::GetRTSC() {
+    rtsc_serialize.lock();
+    _mm_mfence();
+    const u64 current_measure = __rdtsc();
+    u64 diff = current_measure - last_measure;
+    diff = diff & ~static_cast<u64>(static_cast<s64>(diff) >> 63); // max(diff, 0)
+    if (current_measure > last_measure) {
+        last_measure = current_measure;
+    }
+    accumulated_ticks += diff;
+    rtsc_serialize.unlock();
+    return accumulated_ticks;
+}
+
+std::chrono::nanoseconds NativeClock::GetTimeNS() {
+    const u64 rtsc_value = GetRTSC();
+    return std::chrono::nanoseconds{MultiplyAndDivide64(rtsc_value, 1000000000, rtsc_frequency)};
+}
+
+std::chrono::microseconds NativeClock::GetTimeUS() {
+    const u64 rtsc_value = GetRTSC();
+    return std::chrono::microseconds{MultiplyAndDivide64(rtsc_value, 1000000, rtsc_frequency)};
+}
+
+std::chrono::milliseconds NativeClock::GetTimeMS() {
+    const u64 rtsc_value = GetRTSC();
+    return std::chrono::milliseconds{MultiplyAndDivide64(rtsc_value, 1000, rtsc_frequency)};
+}
+
+u64 NativeClock::GetClockCycles() {
+    const u64 rtsc_value = GetRTSC();
+    return MultiplyAndDivide64(rtsc_value, emulated_clock_frequency, rtsc_frequency);
+}
+
+u64 NativeClock::GetCPUCycles() {
+    const u64 rtsc_value = GetRTSC();
+    return MultiplyAndDivide64(rtsc_value, emulated_cpu_frequency, rtsc_frequency);
+}
+
+} // namespace X64
+
+} // namespace Common
diff --git a/src/common/x64/native_clock.h b/src/common/x64/native_clock.h
new file mode 100644
index 000000000..b58cf9f5a
--- /dev/null
+++ b/src/common/x64/native_clock.h
@@ -0,0 +1,41 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <optional>
+
+#include "common/spin_lock.h"
+#include "common/wall_clock.h"
+
+namespace Common {
+
+namespace X64 {
+class NativeClock : public WallClock {
+public:
+    NativeClock(u64 emulated_cpu_frequency, u64 emulated_clock_frequency, u64 rtsc_frequency);
+
+    std::chrono::nanoseconds GetTimeNS() override;
+
+    std::chrono::microseconds GetTimeUS() override;
+
+    std::chrono::milliseconds GetTimeMS() override;
+
+    u64 GetClockCycles() override;
+
+    u64 GetCPUCycles() override;
+
+private:
+    u64 GetRTSC();
+
+    SpinLock rtsc_serialize{};
+    u64 last_measure{};
+    u64 accumulated_ticks{};
+    u64 rtsc_frequency;
+};
+} // namespace X64
+
+u64 EstimateRDTSCFrequency();
+
+} // namespace Common
diff --git a/src/common/x64/xbyak_abi.h b/src/common/x64/xbyak_abi.h
index 794da8a52..a5f5d4fc1 100644
--- a/src/common/x64/xbyak_abi.h
+++ b/src/common/x64/xbyak_abi.h
@@ -11,7 +11,7 @@
 
 namespace Common::X64 {
 
-inline int RegToIndex(const Xbyak::Reg& reg) {
+inline std::size_t RegToIndex(const Xbyak::Reg& reg) {
     using Kind = Xbyak::Reg::Kind;
     ASSERT_MSG((reg.getKind() & (Kind::REG | Kind::XMM)) != 0,
                "RegSet only support GPRs and XMM registers.");
@@ -19,17 +19,17 @@ inline int RegToIndex(const Xbyak::Reg& reg) {
     return reg.getIdx() + (reg.getKind() == Kind::REG ? 0 : 16);
 }
 
-inline Xbyak::Reg64 IndexToReg64(int reg_index) {
+inline Xbyak::Reg64 IndexToReg64(std::size_t reg_index) {
     ASSERT(reg_index < 16);
-    return Xbyak::Reg64(reg_index);
+    return Xbyak::Reg64(static_cast<int>(reg_index));
 }
 
-inline Xbyak::Xmm IndexToXmm(int reg_index) {
+inline Xbyak::Xmm IndexToXmm(std::size_t reg_index) {
     ASSERT(reg_index >= 16 && reg_index < 32);
-    return Xbyak::Xmm(reg_index - 16);
+    return Xbyak::Xmm(static_cast<int>(reg_index - 16));
 }
 
-inline Xbyak::Reg IndexToReg(int reg_index) {
+inline Xbyak::Reg IndexToReg(std::size_t reg_index) {
     if (reg_index < 16) {
         return IndexToReg64(reg_index);
     } else {
@@ -151,9 +151,13 @@ constexpr size_t ABI_SHADOW_SPACE = 0;
 
 #endif
 
-inline void ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
-                                   size_t needed_frame_size, s32* out_subtraction,
-                                   s32* out_xmm_offset) {
+struct ABIFrameInfo {
+    s32 subtraction;
+    s32 xmm_offset;
+};
+
+inline ABIFrameInfo ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
+                                           size_t needed_frame_size) {
     const auto count = (regs & ABI_ALL_GPRS).count();
     rsp_alignment -= count * 8;
     size_t subtraction = 0;
@@ -170,33 +174,28 @@ inline void ABI_CalculateFrameSize(std::bitset<32> regs, size_t rsp_alignment,
     rsp_alignment -= subtraction;
     subtraction += rsp_alignment & 0xF;
 
-    *out_subtraction = (s32)subtraction;
-    *out_xmm_offset = (s32)(subtraction - xmm_base_subtraction);
+    return ABIFrameInfo{static_cast<s32>(subtraction),
+                        static_cast<s32>(subtraction - xmm_base_subtraction)};
 }
 
 inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs,
                                               size_t rsp_alignment, size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
+    auto frame_info = ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size);
+
     for (std::size_t i = 0; i < regs.size(); ++i) {
         if (regs[i] && ABI_ALL_GPRS[i]) {
-            code.push(IndexToReg64(static_cast<int>(i)));
+            code.push(IndexToReg64(i));
         }
     }
-    if (subtraction != 0) {
-        code.sub(code.rsp, subtraction);
-    }
 
-    for (int i = 0; i < regs.count(); i++) {
-        if (regs.test(i) & ABI_ALL_GPRS.test(i)) {
-            code.push(IndexToReg64(i));
-        }
+    if (frame_info.subtraction != 0) {
+        code.sub(code.rsp, frame_info.subtraction);
     }
 
     for (std::size_t i = 0; i < regs.size(); ++i) {
         if (regs[i] && ABI_ALL_XMMS[i]) {
-            code.movaps(code.xword[code.rsp + xmm_offset], IndexToXmm(static_cast<int>(i)));
-            xmm_offset += 0x10;
+            code.movaps(code.xword[code.rsp + frame_info.xmm_offset], IndexToXmm(i));
+            frame_info.xmm_offset += 0x10;
         }
     }
 
@@ -205,59 +204,23 @@ inline size_t ABI_PushRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::b
 
 inline void ABI_PopRegistersAndAdjustStack(Xbyak::CodeGenerator& code, std::bitset<32> regs,
                                            size_t rsp_alignment, size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
+    auto frame_info = ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size);
 
     for (std::size_t i = 0; i < regs.size(); ++i) {
         if (regs[i] && ABI_ALL_XMMS[i]) {
-            code.movaps(IndexToXmm(static_cast<int>(i)), code.xword[code.rsp + xmm_offset]);
-            xmm_offset += 0x10;
+            code.movaps(IndexToXmm(i), code.xword[code.rsp + frame_info.xmm_offset]);
+            frame_info.xmm_offset += 0x10;
         }
     }
 
-    if (subtraction != 0) {
-        code.add(code.rsp, subtraction);
+    if (frame_info.subtraction != 0) {
+        code.add(code.rsp, frame_info.subtraction);
     }
 
     // GPRs need to be popped in reverse order
-    for (int i = 15; i >= 0; i--) {
-        if (regs[i]) {
-            code.pop(IndexToReg64(i));
-        }
-    }
-}
-
-inline size_t ABI_PushRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs,
-                                                 size_t rsp_alignment,
-                                                 size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
-
-    for (std::size_t i = 0; i < regs.size(); ++i) {
+    for (std::size_t j = 0; j < regs.size(); ++j) {
+        const std::size_t i = regs.size() - j - 1;
         if (regs[i] && ABI_ALL_GPRS[i]) {
-            code.push(IndexToReg64(static_cast<int>(i)));
-        }
-    }
-
-    if (subtraction != 0) {
-        code.sub(code.rsp, subtraction);
-    }
-
-    return ABI_SHADOW_SPACE;
-}
-
-inline void ABI_PopRegistersAndAdjustStackGPS(Xbyak::CodeGenerator& code, std::bitset<32> regs,
-                                              size_t rsp_alignment, size_t needed_frame_size = 0) {
-    s32 subtraction, xmm_offset;
-    ABI_CalculateFrameSize(regs, rsp_alignment, needed_frame_size, &subtraction, &xmm_offset);
-
-    if (subtraction != 0) {
-        code.add(code.rsp, subtraction);
-    }
-
-    // GPRs need to be popped in reverse order
-    for (int i = 15; i >= 0; i--) {
-        if (regs[i]) {
             code.pop(IndexToReg64(i));
         }
     }