From 3240e0498a03e16aae28933cbe16a7abdcfc633b Mon Sep 17 00:00:00 2001 From: pineappleEA Date: Thu, 10 Mar 2022 07:08:15 +0100 Subject: [PATCH] early-access version 2537 --- README.md | 2 +- src/common/bit_util.h | 7 ++ src/common/x64/cpu_detect.cpp | 102 +++++++++---------- src/common/x64/cpu_detect.h | 64 +++++++----- src/video_core/texture_cache/texture_cache.h | 18 +--- 5 files changed, 97 insertions(+), 96 deletions(-) diff --git a/README.md b/README.md index d36e28d0f..3377f8fe7 100755 --- a/README.md +++ b/README.md @@ -1,7 +1,7 @@ yuzu emulator early access ============= -This is the source code for early-access 2535. +This is the source code for early-access 2537. ## Legal Notice diff --git a/src/common/bit_util.h b/src/common/bit_util.h index f50d3308a..f37538e06 100755 --- a/src/common/bit_util.h +++ b/src/common/bit_util.h @@ -57,4 +57,11 @@ requires std::is_integral_v return static_cast(1ULL << ((8U * sizeof(T)) - std::countl_zero(value - 1U))); } +template +requires std::is_integral_v +[[nodiscard]] constexpr bool Bit(const T value) { + static_assert(bit_index < BitSize(), "bit_index must be smaller than size of T"); + return ((value >> bit_index) & T(1)) == T(1); +} + } // namespace Common diff --git a/src/common/x64/cpu_detect.cpp b/src/common/x64/cpu_detect.cpp index fbeacc7e2..99d87f586 100755 --- a/src/common/x64/cpu_detect.cpp +++ b/src/common/x64/cpu_detect.cpp @@ -1,8 +1,12 @@ -// Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project / 2022 Yuzu Emulator +// Project Licensed under GPLv2 or any later version Refer to the license.txt file included. +#include #include +#include +#include +#include +#include "common/bit_util.h" #include "common/common_types.h" #include "common/x64/cpu_detect.h" @@ -17,7 +21,7 @@ // clang-format on #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__) // 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); @@ -30,7 +34,7 @@ static inline void __cpuidex(int info[4], int function_id, int subfunction_id) { #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); } @@ -45,6 +49,17 @@ static inline u64 _xgetbv(u32 index) { 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 static CPUCaps Detect() { CPUCaps caps = {}; @@ -53,57 +68,44 @@ static CPUCaps Detect() { // yuzu at all anyway 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); - 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::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; + std::memset(caps.brand_string, 0, std::size(caps.brand_string)); + std::memcpy(&caps.brand_string[0], &cpu_id[1], sizeof(u32)); + std::memcpy(&caps.brand_string[4], &cpu_id[3], sizeof(u32)); + std::memcpy(&caps.brand_string[8], &cpu_id[2], sizeof(u32)); + + caps.manufacturer = CPUCaps::ParseManufacturer(caps.brand_string); + + // Set reasonable default cpu string even if brand string not available + std::strncpy(caps.cpu_string, caps.brand_string, std::size(caps.brand_string)); __cpuid(cpu_id, 0x80000000); - 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); + const u32 max_ex_fn = cpu_id[0]; // Detect family and other miscellaneous features if (max_std_fn >= 1) { __cpuid(cpu_id, 0x00000001); - if ((cpu_id[3] >> 25) & 1) - caps.sse = true; - if ((cpu_id[3] >> 26) & 1) - caps.sse2 = true; - if ((cpu_id[2]) & 1) - caps.sse3 = true; - if ((cpu_id[2] >> 9) & 1) - 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; + caps.sse = Common::Bit<25>(cpu_id[3]); + caps.sse2 = Common::Bit<26>(cpu_id[3]); + caps.sse3 = Common::Bit<0>(cpu_id[2]); + caps.ssse3 = Common::Bit<9>(cpu_id[2]); + caps.sse4_1 = Common::Bit<19>(cpu_id[2]); + caps.sse4_2 = Common::Bit<20>(cpu_id[2]); + caps.aes = Common::Bit<25>(cpu_id[2]); // AVX support requires 3 separate checks: // - Is the AVX bit set in CPUID? // - Is the XSAVE bit set in CPUID? // - 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) { caps.avx = true; - if ((cpu_id[2] >> 12) & 1) + if (Common::Bit<12>(cpu_id[2])) caps.fma = true; } } @@ -111,15 +113,13 @@ static CPUCaps Detect() { if (max_std_fn >= 7) { __cpuidex(cpu_id, 0x00000007, 0x00000000); // Can't enable AVX2 unless the XSAVE/XGETBV checks above passed - if ((cpu_id[1] >> 5) & 1) - caps.avx2 = caps.avx; - if ((cpu_id[1] >> 3) & 1) - caps.bmi1 = true; - if ((cpu_id[1] >> 8) & 1) - caps.bmi2 = true; + caps.avx2 = caps.avx && Common::Bit<5>(cpu_id[1]); + caps.bmi1 = Common::Bit<3>(cpu_id[1]); + caps.bmi2 = Common::Bit<8>(cpu_id[1]); // 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) { + if (Common::Bit<16>(cpu_id[1]) && Common::Bit<28>(cpu_id[1]) && + Common::Bit<31>(cpu_id[1]) && Common::Bit<17>(cpu_id[1]) && + Common::Bit<30>(cpu_id[1])) { caps.avx512 = caps.avx2; } } @@ -138,15 +138,13 @@ static CPUCaps Detect() { if (max_ex_fn >= 0x80000001) { // Check for more features __cpuid(cpu_id, 0x80000001); - if ((cpu_id[2] >> 16) & 1) - caps.fma4 = true; + caps.lzcnt = Common::Bit<5>(cpu_id[2]); + caps.fma4 = Common::Bit<16>(cpu_id[2]); } if (max_ex_fn >= 0x80000007) { __cpuid(cpu_id, 0x80000007); - if (cpu_id[3] & (1 << 8)) { - caps.invariant_tsc = true; - } + caps.invariant_tsc = Common::Bit<8>(cpu_id[3]); } if (max_std_fn >= 0x16) { diff --git a/src/common/x64/cpu_detect.h b/src/common/x64/cpu_detect.h index e3b63302e..3e6d808f3 100755 --- a/src/common/x64/cpu_detect.h +++ b/src/common/x64/cpu_detect.h @@ -1,42 +1,50 @@ -// Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// Copyright 2013 Dolphin Emulator Project / 2015 Citra Emulator Project / 2022 Yuzu Emulator +// Project Project Licensed under GPLv2 or any later version Refer to the license.txt file included. #pragma once -namespace Common { +#include +#include "common/common_types.h" -enum class Manufacturer : u32 { - Intel = 0, - AMD = 1, - Hygon = 2, - Unknown = 3, -}; +namespace Common { /// x86/x64 CPU capabilities that may be detected by this module struct CPUCaps { + + enum class Manufacturer : u8 { + Unknown = 0, + Intel = 1, + AMD = 2, + Hygon = 3, + }; + + static Manufacturer ParseManufacturer(std::string_view brand_string); + Manufacturer manufacturer; - char cpu_string[0x21]; - char brand_string[0x41]; - bool sse; - 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; + char brand_string[13]; + + char cpu_string[48]; + u32 base_frequency; u32 max_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; }; /** diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 00dcd66b7..efc1c4525 100755 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -369,7 +369,7 @@ template void TextureCache

::FillImageViews(DescriptorTable& table, std::span cached_image_view_ids, std::span views) { - bool has_blacklisted; + bool has_blacklisted = false; do { has_deleted_images = false; if constexpr (has_blacklists) { @@ -1763,7 +1763,7 @@ void TextureCache

::SynchronizeAliases(ImageId image_id) { }); const auto& resolution = Settings::values.resolution_info; for (const AliasedImage* const aliased : aliased_images) { - if (!resolution.active | !any_rescaled) { + if (!resolution.active || !any_rescaled) { CopyImage(image_id, aliased->id, aliased->copies); continue; } @@ -1774,19 +1774,7 @@ void TextureCache

::SynchronizeAliases(ImageId image_id) { continue; } ScaleUp(aliased_image); - - 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( - (copy.extent.width * resolution.up_scale) >> resolution.down_shift, 1); - if (both_2d) { - copy.extent.height = std::max( - (copy.extent.height * resolution.up_scale) >> resolution.down_shift, 1); - } - } - CopyImage(image_id, aliased->id, copies); + CopyImage(image_id, aliased->id, aliased->copies); } }