Menu
▾
▴
[OMPT-svn-commit] OpenMPT r25340 - trunk/OpenMPT/src/openmpt/soundbase
|
From: <sv...@op...> - 2026-05-20 16:42:27
|
Author: manx Date: Wed May 20 18:42:18 2026 New Revision: 25340 URL: https://source.openmpt.org/browse/openmpt/?op=revision&rev=25340 Log: [Ref] openmpt/soundbase/SampleConvert.hpp: Use mpt::lround and mpt::llround when appropriate, instead of using mpt::round and casting the result to integer. This gives the compiler more freedom. It also removes 1 additional case where fastround() is no longer faster. Modified: trunk/OpenMPT/src/openmpt/soundbase/SampleConvert.hpp Modified: trunk/OpenMPT/src/openmpt/soundbase/SampleConvert.hpp ============================================================================== --- trunk/OpenMPT/src/openmpt/soundbase/SampleConvert.hpp Wed May 20 18:40:16 2026 (r25339) +++ trunk/OpenMPT/src/openmpt/soundbase/SampleConvert.hpp Wed May 20 18:42:18 2026 (r25340) @@ -49,11 +49,23 @@ // amd64 SSE4.2 precise inline+fixup inline + // amd64 AVX2 precise inline+fixup inline + // -// See <https://godbolt.org/z/dYYGenjKT> for a microbenchmark. +// std::round: <https://godbolt.org/z/dYYGenjKT> +// +// arch SSE /fp: std::lround fastlround better // -// TODO: Real-world benchmark of amd64 SSE4.2 and amd64 AVX2. -// The inlined additional code might fold away, or nearly away, -// in which case the canonical semantically correct std::round should be preferred. +// i386 x87 fast 2*call fixup+2*call - +// i386 SSE2 fast inline inline+fixup - +// amd64 SSE2 fast call inline + +// amd64 SSE4.2 fast call inline + +// amd64 AVX2 fast call inline + +// +// i386 x87 precise call fixup+2*call - +// i386 SSE2 precise call fixup+call - +// amd64 SSE2 precise call fixup+call - +// amd64 SSE4.2 precise call inline + +// amd64 AVX2 precise call inline + +// +// std::lround: <https://godbolt.org/z/rcbcbK7b7> #if MPT_COMPILER_MSVC #if defined(_M_FP_FAST) && (_M_FP_FAST == 1) @@ -62,7 +74,7 @@ #elif MPT_ARCH_X86 && (defined(MPT_ARCH_X86_FPU) && !defined(MPT_ARCH_X86_SSE2)) #define MPT_SOUNDBASE_SAMPLECONVERT_HPP_FASTROUND 0 #elif MPT_ARCH_X86 && (defined(MPT_ARCH_X86_SSE2)) -#define MPT_SOUNDBASE_SAMPLECONVERT_HPP_FASTROUND 1 +#define MPT_SOUNDBASE_SAMPLECONVERT_HPP_FASTROUND 0 #elif MPT_ARCH_AMD64 && (defined(MPT_ARCH_X86_SSE2) && !defined(MPT_ARCH_X86_SSE4_2)) #define MPT_SOUNDBASE_SAMPLECONVERT_HPP_FASTROUND 1 #elif MPT_ARCH_AMD64 && (defined(MPT_ARCH_X86_SSE4_2) && !defined(MPT_ARCH_X86_AVX2)) @@ -98,6 +110,18 @@ static_assert(std::is_floating_point<Tfloat>::value); return std::floor(x + static_cast<Tfloat>(0.5)); } +template <typename Tfloat> +MPT_ATTR_ALWAYSINLINE MPT_INLINE_FORCE long fastlround(Tfloat x) +{ + static_assert(std::is_floating_point<Tfloat>::value); + return static_cast<long>(std::floor(x + static_cast<Tfloat>(0.5))); +} +template <typename Tfloat> +MPT_ATTR_ALWAYSINLINE MPT_INLINE_FORCE long long fastllround(Tfloat x) +{ + static_assert(std::is_floating_point<Tfloat>::value); + return static_cast<long long>(std::floor(x + static_cast<Tfloat>(0.5))); +} #else template <typename Tfloat> MPT_ATTR_ALWAYSINLINE MPT_INLINE_FORCE Tfloat fastround(Tfloat x) @@ -105,6 +129,18 @@ static_assert(std::is_floating_point<Tfloat>::value); return mpt::round(x); } +template <typename Tfloat> +MPT_ATTR_ALWAYSINLINE MPT_INLINE_FORCE long fastlround(Tfloat x) +{ + static_assert(std::is_floating_point<Tfloat>::value); + return mpt::lround(x); +} +template <typename Tfloat> +MPT_ATTR_ALWAYSINLINE MPT_INLINE_FORCE long long fastllround(Tfloat x) +{ + static_assert(std::is_floating_point<Tfloat>::value); + return mpt::llround(x); +} #endif @@ -221,7 +257,7 @@ { val = mpt::safe_clamp(val, -1.0f, 1.0f); val *= 128.0f; - return static_cast<uint8>(mpt::saturate_cast<int8>(static_cast<int>(SC::fastround(val))) + 0x80); + return static_cast<uint8>(mpt::saturate_cast<int8>(SC::fastlround(val)) + 0x80); } }; @@ -234,7 +270,7 @@ { val = mpt::safe_clamp(val, -1.0, 1.0); val *= 128.0; - return static_cast<uint8>(mpt::saturate_cast<int8>(static_cast<int>(SC::fastround(val))) + 0x80); + return static_cast<uint8>(mpt::saturate_cast<int8>(SC::fastlround(val)) + 0x80); } }; @@ -302,7 +338,7 @@ { val = mpt::safe_clamp(val, -1.0f, 1.0f); val *= 128.0f; - return mpt::saturate_cast<int8>(static_cast<int>(SC::fastround(val))); + return mpt::saturate_cast<int8>(SC::fastlround(val)); } }; @@ -315,7 +351,7 @@ { val = mpt::safe_clamp(val, -1.0, 1.0); val *= 128.0; - return mpt::saturate_cast<int8>(static_cast<int>(SC::fastround(val))); + return mpt::saturate_cast<int8>(SC::fastlround(val)); } }; @@ -383,7 +419,7 @@ { val = mpt::safe_clamp(val, -1.0f, 1.0f); val *= 32768.0f; - return mpt::saturate_cast<int16>(static_cast<int>(SC::fastround(val))); + return mpt::saturate_cast<int16>(SC::fastlround(val)); } }; @@ -396,7 +432,7 @@ { val = mpt::safe_clamp(val, -1.0, 1.0); val *= 32768.0; - return mpt::saturate_cast<int16>(static_cast<int>(SC::fastround(val))); + return mpt::saturate_cast<int16>(SC::fastlround(val)); } }; @@ -464,7 +500,7 @@ { val = mpt::safe_clamp(val, -1.0f, 1.0f); val *= 2147483648.0f; - return static_cast<int24>(mpt::rshift_signed(mpt::saturate_cast<int32>(static_cast<int64>(SC::fastround(val))), 8)); + return static_cast<int24>(mpt::rshift_signed(mpt::saturate_cast<int32>(SC::fastllround(val)), 8)); } }; @@ -477,7 +513,7 @@ { val = mpt::safe_clamp(val, -1.0, 1.0); val *= 2147483648.0; - return static_cast<int24>(mpt::rshift_signed(mpt::saturate_cast<int32>(static_cast<int64>(SC::fastround(val))), 8)); + return static_cast<int24>(mpt::rshift_signed(mpt::saturate_cast<int32>(SC::fastllround(val)), 8)); } }; @@ -545,7 +581,7 @@ { val = mpt::safe_clamp(val, -1.0f, 1.0f); val *= 2147483648.0f; - return mpt::saturate_cast<int32>(static_cast<int64>(SC::fastround(val))); + return mpt::saturate_cast<int32>(SC::fastllround(val)); } }; @@ -558,7 +594,7 @@ { val = mpt::safe_clamp(val, -1.0, 1.0); val *= 2147483648.0; - return mpt::saturate_cast<int32>(static_cast<int64>(SC::fastround(val))); + return mpt::saturate_cast<int32>(SC::fastllround(val)); } }; |
