diff options
Diffstat (limited to 'indra/llmath')
| -rw-r--r-- | indra/llmath/CMakeLists.txt | 12 | ||||
| -rw-r--r-- | indra/llmath/llcalcparser.h | 6 | ||||
| -rw-r--r-- | indra/llmath/llquaternion2.inl | 5 | ||||
| -rw-r--r-- | indra/llmath/llsimdmath.h | 10 | ||||
| -rw-r--r-- | indra/llmath/llvector4a.cpp | 10 | ||||
| -rw-r--r-- | indra/llmath/llvector4a.inl | 15 | ||||
| -rw-r--r-- | indra/llmath/llvolume.cpp | 2 |
7 files changed, 49 insertions, 11 deletions
diff --git a/indra/llmath/CMakeLists.txt b/indra/llmath/CMakeLists.txt index bd860adc31..e21b6fdb84 100644 --- a/indra/llmath/CMakeLists.txt +++ b/indra/llmath/CMakeLists.txt @@ -6,7 +6,6 @@ include(00-Common) include(LLCommon) include(bugsplat) include(Boost) -include(SSE2NEON) set(llmath_SOURCE_FILES llbbox.cpp @@ -99,11 +98,20 @@ set(llmath_HEADER_FILES list(APPEND llmath_SOURCE_FILES ${llmath_HEADER_FILES}) +include_directories(${LIBS_PREBUILT_DIR}/include) + add_library (llmath ${llmath_SOURCE_FILES}) -target_link_libraries(llmath llcommon llmeshoptimizer ll::sse2neon) +target_link_libraries(llmath llcommon llmeshoptimizer) target_include_directories( llmath INTERFACE ${CMAKE_CURRENT_SOURCE_DIR}) +if (CMAKE_CXX_COMPILER_ID MATCHES GNU) + set_source_files_properties(llcalc.cpp PROPERTIES + COMPILE_FLAGS -Wno-dangling-pointer) +endif () + +include(LibraryInstall) + # Add tests if (LL_TESTS) include(LLAddBuildTest) diff --git a/indra/llmath/llcalcparser.h b/indra/llmath/llcalcparser.h index ea71752ebc..ec7f6a2cb6 100644 --- a/indra/llmath/llcalcparser.h +++ b/indra/llmath/llcalcparser.h @@ -131,14 +131,14 @@ struct LLCalcParser : grammar<LLCalcParser> power = unary_expr[power.value = arg1] >> - *('^' >> assert_syntax(unary_expr[power.value = phoenix::bind(&powf)(power.value, arg1)])) + *('^' >> assert_syntax(unary_expr[power.value = phoenix::bind(&LLCalcParser::_pow)(self, power.value, arg1)])) ; term = power[term.value = arg1] >> *(('*' >> assert_syntax(power[term.value *= arg1])) | ('/' >> assert_syntax(power[term.value /= arg1])) | - ('%' >> assert_syntax(power[term.value = phoenix::bind(&fmodf)(term.value, arg1)])) + ('%' >> assert_syntax(power[term.value = phoenix::bind(&LLCalcParser::_fmod)(self, term.value, arg1)])) ) ; @@ -177,6 +177,8 @@ private: F32 _floor(const F32& a) const { return (F32)llfloor(a); } F32 _ceil(const F32& a) const { return (F32)llceil(a); } F32 _atan2(const F32& a,const F32& b) const { return atan2(a,b); } + F32 _pow(const F32& a, const F32& b) const { return powf(a, b); } + F32 _fmod(const F32&a, const F32& b) const { return fmodf(a, b); } LLCalc::calc_map_t* mConstants; LLCalc::calc_map_t* mVariables; diff --git a/indra/llmath/llquaternion2.inl b/indra/llmath/llquaternion2.inl index ce5ed73926..b431d5766c 100644 --- a/indra/llmath/llquaternion2.inl +++ b/indra/llmath/llquaternion2.inl @@ -26,8 +26,13 @@ #include "llquaternion2.h" +#if _M_ARM64 +static const LLQuad LL_V4A_PLUS_ONE = {.n128_f32 = {1.f, 1.f, 1.f, 1.f}}; +static const LLQuad LL_V4A_MINUS_ONE = {.n128_f32 = {-1.f, -1.f, -1.f, -1.f}}; +#else static const LLQuad LL_V4A_PLUS_ONE = {1.f, 1.f, 1.f, 1.f}; static const LLQuad LL_V4A_MINUS_ONE = {-1.f, -1.f, -1.f, -1.f}; +#endif // Ctor from LLQuaternion inline LLQuaternion2::LLQuaternion2( const LLQuaternion& quat ) diff --git a/indra/llmath/llsimdmath.h b/indra/llmath/llsimdmath.h index b27b034cf3..440561e589 100644 --- a/indra/llmath/llsimdmath.h +++ b/indra/llmath/llsimdmath.h @@ -31,10 +31,8 @@ #error "Please include llmath.h before this file." #endif -// the check for this error case must be split into multiple parts -// because some versions of VS complain about '__SSE2__' #if ( ( LL_DARWIN || LL_LINUX ) ) - #if !(__SSE2__) && !(__arm64__) && !(__aarch64__) + #if !(__SSE2__ || __ARM_NEON) #error SSE2 not enabled. LLVector4a and related class will not compile. #endif #elif ( LL_WINDOWS && ( _M_IX86_FP < 2 && ADDRESS_SIZE == 32 ) ) @@ -45,11 +43,11 @@ #include <stdint.h> #endif -#if defined(__arm64__) || defined(__aarch64__) -#include "sse2neon.h" -#else +#if defined(__i386__) || defined(__x86_64__) || _M_X64 #include <xmmintrin.h> #include <emmintrin.h> +#else +#include <sse2neon.h> #endif #include "llmemory.h" diff --git a/indra/llmath/llvector4a.cpp b/indra/llmath/llvector4a.cpp index b81d50f0f9..df20585d16 100644 --- a/indra/llmath/llvector4a.cpp +++ b/indra/llmath/llvector4a.cpp @@ -30,6 +30,15 @@ #include "llmath.h" #include "llquantize.h" +#if _M_ARM64 +extern const LLQuad F_ZERO_4A = {.n128_f32 = {0, 0, 0, 0}}; +extern const LLQuad F_APPROXIMATELY_ZERO_4A = {.n128_f32 = { + F_APPROXIMATELY_ZERO, + F_APPROXIMATELY_ZERO, + F_APPROXIMATELY_ZERO, + F_APPROXIMATELY_ZERO +}}; +#else extern const LLQuad F_ZERO_4A = { 0, 0, 0, 0 }; extern const LLQuad F_APPROXIMATELY_ZERO_4A = { F_APPROXIMATELY_ZERO, @@ -37,6 +46,7 @@ extern const LLQuad F_APPROXIMATELY_ZERO_4A = { F_APPROXIMATELY_ZERO, F_APPROXIMATELY_ZERO }; +#endif extern const LLVector4a LL_V4A_ZERO = reinterpret_cast<const LLVector4a&> ( F_ZERO_4A ); extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast<const LLVector4a&> ( F_APPROXIMATELY_ZERO_4A ); diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl index 443a46c317..0f7c4123ac 100644 --- a/indra/llmath/llvector4a.inl +++ b/indra/llmath/llvector4a.inl @@ -351,8 +351,13 @@ inline void LLVector4a::normalize3() LLVector4a lenSqrd; lenSqrd.setAllDot3( *this, *this ); // rsqrt = approximate reciprocal square (i.e., { ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2 } const LLQuad rsqrt = _mm_rsqrt_ps(lenSqrd.mQ); +#if _M_ARM64 + static const LLQuad half = {.n128_f32 = {0.5f, 0.5f, 0.5f, 0.5f}}; + static const LLQuad three = {.n128_f32 = {3.f, 3.f, 3.f, 3.f }}; +#else static const LLQuad half = { 0.5f, 0.5f, 0.5f, 0.5f }; static const LLQuad three = {3.f, 3.f, 3.f, 3.f }; +#endif // Now we do one round of Newton-Raphson approximation to get full accuracy // According to the Newton-Raphson method, given a first 'w' for the root of f(x) = 1/x^2 - a (i.e., x = 1/sqrt(a)) // the next better approximation w[i+1] = w - f(w)/f'(w) = w - (1/w^2 - a)/(-2*w^(-3)) @@ -375,8 +380,13 @@ inline void LLVector4a::normalize4() LLVector4a lenSqrd; lenSqrd.setAllDot4( *this, *this ); // rsqrt = approximate reciprocal square (i.e., { ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2 } const LLQuad rsqrt = _mm_rsqrt_ps(lenSqrd.mQ); +#if _M_ARM64 + static const LLQuad half = {.n128_f32 = {0.5f, 0.5f, 0.5f, 0.5f}}; + static const LLQuad three = {.n128_f32 = {3.f, 3.f, 3.f, 3.f}}; +#else static const LLQuad half = { 0.5f, 0.5f, 0.5f, 0.5f }; static const LLQuad three = {3.f, 3.f, 3.f, 3.f }; +#endif // Now we do one round of Newton-Raphson approximation to get full accuracy // According to the Newton-Raphson method, given a first 'w' for the root of f(x) = 1/x^2 - a (i.e., x = 1/sqrt(a)) // the next better approximation w[i+1] = w - f(w)/f'(w) = w - (1/w^2 - a)/(-2*w^(-3)) @@ -399,8 +409,13 @@ inline LLSimdScalar LLVector4a::normalize3withLength() LLVector4a lenSqrd; lenSqrd.setAllDot3( *this, *this ); // rsqrt = approximate reciprocal square (i.e., { ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2 } const LLQuad rsqrt = _mm_rsqrt_ps(lenSqrd.mQ); +#if _M_ARM64 + static const LLQuad half = {.n128_f32 = {0.5f, 0.5f, 0.5f, 0.5f}}; + static const LLQuad three = {.n128_f32 = {3.f, 3.f, 3.f, 3.f}}; +#else static const LLQuad half = { 0.5f, 0.5f, 0.5f, 0.5f }; static const LLQuad three = {3.f, 3.f, 3.f, 3.f }; +#endif // Now we do one round of Newton-Raphson approximation to get full accuracy // According to the Newton-Raphson method, given a first 'w' for the root of f(x) = 1/x^2 - a (i.e., x = 1/sqrt(a)) // the next better approximation w[i+1] = w - f(w)/f'(w) = w - (1/w^2 - a)/(-2*w^(-3)) diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index b3cb278d59..0cf5ecd3d0 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -55,7 +55,7 @@ #include "mikktspace/mikktspace.hh" -#include "meshoptimizer/meshoptimizer.h" +#include <meshoptimizer.h> #define DEBUG_SILHOUETTE_BINORMALS 0 #define DEBUG_SILHOUETTE_NORMALS 0 // TomY: Use this to display normals using the silhouette |