early-access version 2537

This commit is contained in:
pineappleEA 2022-03-10 07:08:15 +01:00
parent 870c5bbb4e
commit 3240e0498a
5 changed files with 97 additions and 96 deletions

View file

@ -1,7 +1,7 @@
yuzu emulator early access yuzu emulator early access
============= =============
This is the source code for early-access 2535. This is the source code for early-access 2537.
## Legal Notice ## Legal Notice

View file

@ -57,4 +57,11 @@ requires std::is_integral_v<T>
return static_cast<T>(1ULL << ((8U * sizeof(T)) - std::countl_zero(value - 1U))); return static_cast<T>(1ULL << ((8U * sizeof(T)) - std::countl_zero(value - 1U)));
} }
template <size_t bit_index, typename T>
requires std::is_integral_v<T>
[[nodiscard]] constexpr bool Bit(const T value) {
static_assert(bit_index < BitSize<T>(), "bit_index must be smaller than size of T");
return ((value >> bit_index) & T(1)) == T(1);
}
} // namespace Common } // namespace Common

View file

@ -1,8 +1,12 @@
// Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project // Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project / 2022 Yuzu Emulator
// Licensed under GPLv2 or any later version // Project Licensed under GPLv2 or any later version Refer to the license.txt file included.
// Refer to the license.txt file included.
#include <array>
#include <cstring> #include <cstring>
#include <iterator>
#include <span>
#include <string_view>
#include "common/bit_util.h"
#include "common/common_types.h" #include "common/common_types.h"
#include "common/x64/cpu_detect.h" #include "common/x64/cpu_detect.h"
@ -17,7 +21,7 @@
// clang-format on // clang-format on
#endif #endif
static inline void __cpuidex(int info[4], int function_id, int subfunction_id) { static inline void __cpuidex(int info[4], u32 function_id, u32 subfunction_id) {
#if defined(__DragonFly__) || defined(__FreeBSD__) #if defined(__DragonFly__) || defined(__FreeBSD__)
// Despite the name, this is just do_cpuid() with ECX as second input. // Despite the name, this is just do_cpuid() with ECX as second input.
cpuid_count((u_int)function_id, (u_int)subfunction_id, (u_int*)info); cpuid_count((u_int)function_id, (u_int)subfunction_id, (u_int*)info);
@ -30,7 +34,7 @@ static inline void __cpuidex(int info[4], int function_id, int subfunction_id) {
#endif #endif
} }
static inline void __cpuid(int info[4], int function_id) { static inline void __cpuid(int info[4], u32 function_id) {
return __cpuidex(info, function_id, 0); return __cpuidex(info, function_id, 0);
} }
@ -45,6 +49,17 @@ static inline u64 _xgetbv(u32 index) {
namespace Common { namespace Common {
CPUCaps::Manufacturer CPUCaps::ParseManufacturer(std::string_view brand_string) {
if (brand_string == "GenuineIntel") {
return Manufacturer::Intel;
} else if (brand_string == "AuthenticAMD") {
return Manufacturer::AMD;
} else if (brand_string == "HygonGenuine") {
return Manufacturer::Hygon;
}
return Manufacturer::Unknown;
}
// Detects the various CPU features // Detects the various CPU features
static CPUCaps Detect() { static CPUCaps Detect() {
CPUCaps caps = {}; CPUCaps caps = {};
@ -53,57 +68,44 @@ static CPUCaps Detect() {
// yuzu at all anyway // yuzu at all anyway
int cpu_id[4]; int cpu_id[4];
memset(caps.brand_string, 0, sizeof(caps.brand_string));
// Detect CPU's CPUID capabilities and grab CPU string // Detect CPU's CPUID capabilities and grab manufacturer string
__cpuid(cpu_id, 0x00000000); __cpuid(cpu_id, 0x00000000);
u32 max_std_fn = cpu_id[0]; // EAX const u32 max_std_fn = cpu_id[0]; // EAX
std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(int)); std::memset(caps.brand_string, 0, std::size(caps.brand_string));
std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(int)); std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(u32));
std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(int)); std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(u32));
if (cpu_id[1] == 0x756e6547 && cpu_id[2] == 0x6c65746e && cpu_id[3] == 0x49656e69) std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(u32));
caps.manufacturer = Manufacturer::Intel;
else if (cpu_id[1] == 0x68747541 && cpu_id[2] == 0x444d4163 && cpu_id[3] == 0x69746e65) caps.manufacturer = CPUCaps::ParseManufacturer(caps.brand_string);
caps.manufacturer = Manufacturer::AMD;
else if (cpu_id[1] == 0x6f677948 && cpu_id[2] == 0x656e6975 && cpu_id[3] == 0x6e65476e) // Set reasonable default cpu string even if brand string not available
caps.manufacturer = Manufacturer::Hygon; std::strncpy(caps.cpu_string, caps.brand_string, std::size(caps.brand_string));
else
caps.manufacturer = Manufacturer::Unknown;
__cpuid(cpu_id, 0x80000000); __cpuid(cpu_id, 0x80000000);
u32 max_ex_fn = cpu_id[0]; const u32 max_ex_fn = cpu_id[0];
// Set reasonable default brand string even if brand string not available
strcpy(caps.cpu_string, caps.brand_string);
// Detect family and other miscellaneous features // Detect family and other miscellaneous features
if (max_std_fn >= 1) { if (max_std_fn >= 1) {
__cpuid(cpu_id, 0x00000001); __cpuid(cpu_id, 0x00000001);
if ((cpu_id[3] >> 25) & 1) caps.sse = Common::Bit<25>(cpu_id[3]);
caps.sse = true; caps.sse2 = Common::Bit<26>(cpu_id[3]);
if ((cpu_id[3] >> 26) & 1) caps.sse3 = Common::Bit<0>(cpu_id[2]);
caps.sse2 = true; caps.ssse3 = Common::Bit<9>(cpu_id[2]);
if ((cpu_id[2]) & 1) caps.sse4_1 = Common::Bit<19>(cpu_id[2]);
caps.sse3 = true; caps.sse4_2 = Common::Bit<20>(cpu_id[2]);
if ((cpu_id[2] >> 9) & 1) caps.aes = Common::Bit<25>(cpu_id[2]);
caps.ssse3 = true;
if ((cpu_id[2] >> 19) & 1)
caps.sse4_1 = true;
if ((cpu_id[2] >> 20) & 1)
caps.sse4_2 = true;
if ((cpu_id[2] >> 25) & 1)
caps.aes = true;
// AVX support requires 3 separate checks: // AVX support requires 3 separate checks:
// - Is the AVX bit set in CPUID? // - Is the AVX bit set in CPUID?
// - Is the XSAVE bit set in CPUID? // - Is the XSAVE bit set in CPUID?
// - XGETBV result has the XCR bit set. // - XGETBV result has the XCR bit set.
if (((cpu_id[2] >> 28) & 1) && ((cpu_id[2] >> 27) & 1)) { if (Common::Bit<28>(cpu_id[2]) && Common::Bit<27>(cpu_id[2])) {
if ((_xgetbv(_XCR_XFEATURE_ENABLED_MASK) & 0x6) == 0x6) { if ((_xgetbv(_XCR_XFEATURE_ENABLED_MASK) & 0x6) == 0x6) {
caps.avx = true; caps.avx = true;
if ((cpu_id[2] >> 12) & 1) if (Common::Bit<12>(cpu_id[2]))
caps.fma = true; caps.fma = true;
} }
} }
@ -111,15 +113,13 @@ static CPUCaps Detect() {
if (max_std_fn >= 7) { if (max_std_fn >= 7) {
__cpuidex(cpu_id, 0x00000007, 0x00000000); __cpuidex(cpu_id, 0x00000007, 0x00000000);
// Can't enable AVX2 unless the XSAVE/XGETBV checks above passed // Can't enable AVX2 unless the XSAVE/XGETBV checks above passed
if ((cpu_id[1] >> 5) & 1) caps.avx2 = caps.avx && Common::Bit<5>(cpu_id[1]);
caps.avx2 = caps.avx; caps.bmi1 = Common::Bit<3>(cpu_id[1]);
if ((cpu_id[1] >> 3) & 1) caps.bmi2 = Common::Bit<8>(cpu_id[1]);
caps.bmi1 = true;
if ((cpu_id[1] >> 8) & 1)
caps.bmi2 = true;
// Checks for AVX512F, AVX512CD, AVX512VL, AVX512DQ, AVX512BW (Intel Skylake-X/SP) // 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 && if (Common::Bit<16>(cpu_id[1]) && Common::Bit<28>(cpu_id[1]) &&
(cpu_id[1] >> 17) & 1 && (cpu_id[1] >> 30) & 1) { Common::Bit<31>(cpu_id[1]) && Common::Bit<17>(cpu_id[1]) &&
Common::Bit<30>(cpu_id[1])) {
caps.avx512 = caps.avx2; caps.avx512 = caps.avx2;
} }
} }
@ -138,15 +138,13 @@ static CPUCaps Detect() {
if (max_ex_fn >= 0x80000001) { if (max_ex_fn >= 0x80000001) {
// Check for more features // Check for more features
__cpuid(cpu_id, 0x80000001); __cpuid(cpu_id, 0x80000001);
if ((cpu_id[2] >> 16) & 1) caps.lzcnt = Common::Bit<5>(cpu_id[2]);
caps.fma4 = true; caps.fma4 = Common::Bit<16>(cpu_id[2]);
} }
if (max_ex_fn >= 0x80000007) { if (max_ex_fn >= 0x80000007) {
__cpuid(cpu_id, 0x80000007); __cpuid(cpu_id, 0x80000007);
if (cpu_id[3] & (1 << 8)) { caps.invariant_tsc = Common::Bit<8>(cpu_id[3]);
caps.invariant_tsc = true;
}
} }
if (max_std_fn >= 0x16) { if (max_std_fn >= 0x16) {

View file

@ -1,42 +1,50 @@
// Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project // Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project / 2022 Yuzu Emulator
// Licensed under GPLv2 or any later version // Project Project Licensed under GPLv2 or any later version Refer to the license.txt file included.
// Refer to the license.txt file included.
#pragma once #pragma once
namespace Common { #include <string_view>
#include "common/common_types.h"
enum class Manufacturer : u32 { namespace Common {
Intel = 0,
AMD = 1,
Hygon = 2,
Unknown = 3,
};
/// x86/x64 CPU capabilities that may be detected by this module /// x86/x64 CPU capabilities that may be detected by this module
struct CPUCaps { struct CPUCaps {
enum class Manufacturer : u8 {
Unknown = 0,
Intel = 1,
AMD = 2,
Hygon = 3,
};
static Manufacturer ParseManufacturer(std::string_view brand_string);
Manufacturer manufacturer; Manufacturer manufacturer;
char cpu_string[0x21]; char brand_string[13];
char brand_string[0x41];
bool sse; char cpu_string[48];
bool sse2;
bool sse3;
bool ssse3;
bool sse4_1;
bool sse4_2;
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 base_frequency;
u32 max_frequency; u32 max_frequency;
u32 bus_frequency; u32 bus_frequency;
bool sse : 1;
bool sse2 : 1;
bool sse3 : 1;
bool ssse3 : 1;
bool sse4_1 : 1;
bool sse4_2 : 1;
bool lzcnt : 1;
bool avx : 1;
bool avx2 : 1;
bool avx512 : 1;
bool bmi1 : 1;
bool bmi2 : 1;
bool fma : 1;
bool fma4 : 1;
bool aes : 1;
bool invariant_tsc : 1;
}; };
/** /**

View file

@ -369,7 +369,7 @@ template <bool has_blacklists>
void TextureCache<P>::FillImageViews(DescriptorTable<TICEntry>& table, void TextureCache<P>::FillImageViews(DescriptorTable<TICEntry>& table,
std::span<ImageViewId> cached_image_view_ids, std::span<ImageViewId> cached_image_view_ids,
std::span<ImageViewInOut> views) { std::span<ImageViewInOut> views) {
bool has_blacklisted; bool has_blacklisted = false;
do { do {
has_deleted_images = false; has_deleted_images = false;
if constexpr (has_blacklists) { if constexpr (has_blacklists) {
@ -1763,7 +1763,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) {
}); });
const auto& resolution = Settings::values.resolution_info; const auto& resolution = Settings::values.resolution_info;
for (const AliasedImage* const aliased : aliased_images) { for (const AliasedImage* const aliased : aliased_images) {
if (!resolution.active | !any_rescaled) { if (!resolution.active || !any_rescaled) {
CopyImage(image_id, aliased->id, aliased->copies); CopyImage(image_id, aliased->id, aliased->copies);
continue; continue;
} }
@ -1774,19 +1774,7 @@ void TextureCache<P>::SynchronizeAliases(ImageId image_id) {
continue; continue;
} }
ScaleUp(aliased_image); ScaleUp(aliased_image);
CopyImage(image_id, aliased->id, aliased->copies);
const bool both_2d{image.info.type == ImageType::e2D &&
aliased_image.info.type == ImageType::e2D};
auto copies = aliased->copies;
for (auto copy : copies) {
copy.extent.width = std::max<u32>(
(copy.extent.width * resolution.up_scale) >> resolution.down_shift, 1);
if (both_2d) {
copy.extent.height = std::max<u32>(
(copy.extent.height * resolution.up_scale) >> resolution.down_shift, 1);
}
}
CopyImage(image_id, aliased->id, copies);
} }
} }