|
Gromacs
2019
|
|
Gromacs
2019
|
Collaboration diagram for SIMD intrinsics interface (simd):Provides an architecture-independent way of doing SIMD coding.
Overview of the SIMD implementation is provided in Single-instruction Multiple-data (SIMD) coding. The details are documented in gromacs/simd/simd.h and the reference implementation impl_reference.h.
Namespaces | |
| gmx | |
| Generic GROMACS namespace. | |
SIMD implementation capability definitions | |
| #define | GMX_SIMD 1 |
| 1 if any SIMD support is present, otherwise 0. | |
| #define | GMX_SIMD_HAVE_FLOAT 1 |
| 1 when SIMD float support is present, otherwise 0 More... | |
| #define | GMX_SIMD_HAVE_DOUBLE 1 |
| 1 if SIMD double support is present, otherwise 0 | |
| #define | GMX_SIMD_HAVE_LOADU 1 |
| 1 if the SIMD implementation supports unaligned loads, otherwise 0 | |
| #define | GMX_SIMD_HAVE_STOREU 1 |
| 1 if the SIMD implementation supports unaligned stores, otherwise 0 | |
| #define | GMX_SIMD_HAVE_FMA 0 |
| 1 if the SIMD implementation has fused-multiply add hardware More... | |
| #define | GMX_SIMD_HAVE_LOGICAL 1 |
| 1 if SIMD impl has logical operations on floating-point data, otherwise 0 | |
| #define | GMX_SIMD_HAVE_FINT32_EXTRACT 1 |
| Support for extracting integers from gmx::SimdFInt32 (1/0 for present/absent) | |
| #define | GMX_SIMD_HAVE_FINT32_LOGICAL 1 |
| 1 if SIMD logical ops are supported for gmx::SimdFInt32, otherwise 0 | |
| #define | GMX_SIMD_HAVE_FINT32_ARITHMETICS 1 |
| 1 if SIMD arithmetic ops are supported for gmx::SimdFInt32, otherwise 0 | |
| #define | GMX_SIMD_HAVE_DINT32_EXTRACT 1 |
| Support for extracting integer from gmx::SimdDInt32 (1/0 for present/absent) | |
| #define | GMX_SIMD_HAVE_DINT32_LOGICAL 1 |
| 1 if logical operations are supported for gmx::SimdDInt32, otherwise 0 | |
| #define | GMX_SIMD_HAVE_DINT32_ARITHMETICS 1 |
| 1 if SIMD arithmetic ops are supported for gmx::SimdDInt32, otherwise 0 | |
| #define | GMX_SIMD_HAVE_NATIVE_COPYSIGN_FLOAT 0 |
| 1 if implementation provides single precision copysign() More... | |
| #define | GMX_SIMD_HAVE_NATIVE_RSQRT_ITER_FLOAT 0 |
| 1 if implementation provides single precision 1/sqrt(x) N-R iterations faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_RCP_ITER_FLOAT 0 |
| 1 if implementation provides single precision 1/x N-R iterations faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_LOG_FLOAT 0 |
| 1 if implementation provides single precision log() faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_EXP2_FLOAT 0 |
| 1 if implementation provides single precision exp2() faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_EXP_FLOAT 0 |
| 1 if implementation provides single precision exp() faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_COPYSIGN_DOUBLE 0 |
| 1 if implementation provides double precision copysign() More... | |
| #define | GMX_SIMD_HAVE_NATIVE_RSQRT_ITER_DOUBLE 0 |
| 1 if implementation provides double precision 1/sqrt(x) N-R iterations faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_RCP_ITER_DOUBLE 0 |
| 1 if implementation provides double precision 1/x N-R iterations faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_LOG_DOUBLE 0 |
| 1 if implementation provides double precision log() faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_EXP2_DOUBLE 0 |
| 1 if implementation provides double precision exp2() faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_NATIVE_EXP_DOUBLE 0 |
| 1 if implementation provides double precision exp() faster than simd_math.h More... | |
| #define | GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_FLOAT 1 |
| 1 if gmx::gatherLoadUBySimdIntTranspose is present, otherwise 0 | |
| #define | GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_DOUBLE 1 |
| 1 if gmx::gatherLoadUBySimdIntTranspose is present, otherwise 0 | |
| #define | GMX_SIMD_HAVE_HSIMD_UTIL_FLOAT 1 |
| 1 if float half-register load/store/reduce utils present, otherwise 0 | |
| #define | GMX_SIMD_HAVE_HSIMD_UTIL_DOUBLE 1 |
| 1 if double half-register load/store/reduce utils present, otherwise 0 | |
| #define | GMX_SIMD_FLOAT_WIDTH 4 |
| Width of the gmx::SimdFloat datatype. | |
| #define | GMX_SIMD_DOUBLE_WIDTH 4 |
| Width of the gmx::SimdDouble datatype. | |
| #define | GMX_SIMD_HAVE_4NSIMD_UTIL_FLOAT 1 |
| 1 if float 4xN load utils present, otherwise 0 | |
| #define | GMX_SIMD_HAVE_4NSIMD_UTIL_DOUBLE 1 |
| 1 if double 4xN load utils present, otherwise 0 | |
| #define | GMX_SIMD4_HAVE_FLOAT 1 |
| 1 if implementation provides gmx::Simd4Float, otherwise 0. | |
| #define | GMX_SIMD4_HAVE_DOUBLE 1 |
| 1 if the implementation provides gmx::Simd4Double, otherwise 0. | |
| #define | GMX_SIMD_FINT32_WIDTH GMX_SIMD_FLOAT_WIDTH |
| Width of the gmx::SimdFInt32 datatype. | |
| #define | GMX_SIMD_DINT32_WIDTH GMX_SIMD_DOUBLE_WIDTH |
| Width of the gmx::SimdDInt32 datatype. | |
| #define | GMX_SIMD4_WIDTH 4 |
| The SIMD4 type is always four units wide, but this makes code more explicit. | |
| #define | GMX_SIMD_ALIGNMENT 8 |
| Required alignment in bytes for aligned load/store (always defined, even without SIMD) | |
| #define | GMX_SIMD_RSQRT_BITS 23 |
| Accuracy of SIMD 1/sqrt(x) lookup. Used to determine number of iterations. | |
| #define | GMX_SIMD_RCP_BITS 23 |
| Accuracy of SIMD 1/x lookup. Used to determine number of iterations. | |
Constant width-4 double precision SIMD types and instructions | |
| static Simd4Double gmx_simdcall | gmx::load4 (const double *m) |
| Load 4 double values from aligned memory into SIMD4 variable. More... | |
| static void gmx_simdcall | gmx::store4 (double *m, Simd4Double a) |
| Store the contents of SIMD4 double to aligned memory m. More... | |
| static Simd4Double gmx_simdcall | gmx::load4U (const double *m) |
| Load SIMD4 double from unaligned memory. More... | |
| static void gmx_simdcall | gmx::store4U (double *m, Simd4Double a) |
| Store SIMD4 double to unaligned memory. More... | |
| static Simd4Double gmx_simdcall | gmx::simd4SetZeroD () |
| Set all SIMD4 double elements to 0. More... | |
| static Simd4Double gmx_simdcall | gmx::operator& (Simd4Double a, Simd4Double b) |
| Bitwise and for two SIMD4 double variables. More... | |
| static Simd4Double gmx_simdcall | gmx::andNot (Simd4Double a, Simd4Double b) |
| Bitwise andnot for two SIMD4 double variables. c=(~a) & b. More... | |
| static Simd4Double gmx_simdcall | gmx::operator| (Simd4Double a, Simd4Double b) |
| Bitwise or for two SIMD4 doubles. More... | |
| static Simd4Double gmx_simdcall | gmx::operator^ (Simd4Double a, Simd4Double b) |
| Bitwise xor for two SIMD4 double variables. More... | |
| static Simd4Double gmx_simdcall | gmx::operator+ (Simd4Double a, Simd4Double b) |
| Add two double SIMD4 variables. More... | |
| static Simd4Double gmx_simdcall | gmx::operator- (Simd4Double a, Simd4Double b) |
| Subtract two SIMD4 variables. More... | |
| static Simd4Double gmx_simdcall | gmx::operator- (Simd4Double a) |
| SIMD4 floating-point negate. More... | |
| static Simd4Double gmx_simdcall | gmx::operator* (Simd4Double a, Simd4Double b) |
| Multiply two SIMD4 variables. More... | |
| static Simd4Double gmx_simdcall | gmx::fma (Simd4Double a, Simd4Double b, Simd4Double c) |
| SIMD4 Fused-multiply-add. Result is a*b+c. More... | |
| static Simd4Double gmx_simdcall | gmx::fms (Simd4Double a, Simd4Double b, Simd4Double c) |
| SIMD4 Fused-multiply-subtract. Result is a*b-c. More... | |
| static Simd4Double gmx_simdcall | gmx::fnma (Simd4Double a, Simd4Double b, Simd4Double c) |
| SIMD4 Fused-negated-multiply-add. Result is -a*b+c. More... | |
| static Simd4Double gmx_simdcall | gmx::fnms (Simd4Double a, Simd4Double b, Simd4Double c) |
| SIMD4 Fused-negated-multiply-subtract. Result is -a*b-c. More... | |
| static Simd4Double gmx_simdcall | gmx::rsqrt (Simd4Double x) |
| SIMD4 1.0/sqrt(x) lookup. More... | |
| static Simd4Double gmx_simdcall | gmx::abs (Simd4Double a) |
| SIMD4 Floating-point abs(). More... | |
| static Simd4Double gmx_simdcall | gmx::max (Simd4Double a, Simd4Double b) |
| Set each SIMD4 element to the largest from two variables. More... | |
| static Simd4Double gmx_simdcall | gmx::min (Simd4Double a, Simd4Double b) |
| Set each SIMD4 element to the largest from two variables. More... | |
| static Simd4Double gmx_simdcall | gmx::round (Simd4Double a) |
| SIMD4 Round to nearest integer value (in floating-point format). More... | |
| static Simd4Double gmx_simdcall | gmx::trunc (Simd4Double a) |
| Truncate SIMD4, i.e. round towards zero - common hardware instruction. More... | |
| static double gmx_simdcall | gmx::dotProduct (Simd4Double a, Simd4Double b) |
| Return dot product of two double precision SIMD4 variables. More... | |
| static void gmx_simdcall | gmx::transpose (Simd4Double *v0, Simd4Double *v1, Simd4Double *v2, Simd4Double *v3) |
| SIMD4 double transpose. More... | |
| static Simd4DBool gmx_simdcall | gmx::operator== (Simd4Double a, Simd4Double b) |
| a==b for SIMD4 double More... | |
| static Simd4DBool gmx_simdcall | gmx::operator!= (Simd4Double a, Simd4Double b) |
| a!=b for SIMD4 double More... | |
| static Simd4DBool gmx_simdcall | gmx::operator< (Simd4Double a, Simd4Double b) |
| a<b for SIMD4 double More... | |
| static Simd4DBool gmx_simdcall | gmx::operator<= (Simd4Double a, Simd4Double b) |
| a<=b for SIMD4 double. More... | |
| static Simd4DBool gmx_simdcall | gmx::operator&& (Simd4DBool a, Simd4DBool b) |
| Logical and on single precision SIMD4 booleans. More... | |
| static Simd4DBool gmx_simdcall | gmx::operator|| (Simd4DBool a, Simd4DBool b) |
| Logical or on single precision SIMD4 booleans. More... | |
| static bool gmx_simdcall | gmx::anyTrue (Simd4DBool a) |
| Returns non-zero if any of the boolean in SIMD4 a is True, otherwise 0. More... | |
| static Simd4Double gmx_simdcall | gmx::selectByMask (Simd4Double a, Simd4DBool mask) |
| Select from single precision SIMD4 variable where boolean is true. More... | |
| static Simd4Double gmx_simdcall | gmx::selectByNotMask (Simd4Double a, Simd4DBool mask) |
| Select from single precision SIMD4 variable where boolean is false. More... | |
| static Simd4Double gmx_simdcall | gmx::blend (Simd4Double a, Simd4Double b, Simd4DBool sel) |
| Vector-blend SIMD4 selection. More... | |
| static double gmx_simdcall | gmx::reduce (Simd4Double a) |
| Return sum of all elements in SIMD4 double variable. More... | |
Constant width-4 single precision SIMD types and instructions | |
| static Simd4Float gmx_simdcall | gmx::load4 (const float *m) |
| Load 4 float values from aligned memory into SIMD4 variable. More... | |
| static void gmx_simdcall | gmx::store4 (float *m, Simd4Float a) |
| Store the contents of SIMD4 float to aligned memory m. More... | |
| static Simd4Float gmx_simdcall | gmx::load4U (const float *m) |
| Load SIMD4 float from unaligned memory. More... | |
| static void gmx_simdcall | gmx::store4U (float *m, Simd4Float a) |
| Store SIMD4 float to unaligned memory. More... | |
| static Simd4Float gmx_simdcall | gmx::simd4SetZeroF () |
| Set all SIMD4 float elements to 0. More... | |
| static Simd4Float gmx_simdcall | gmx::operator& (Simd4Float a, Simd4Float b) |
| Bitwise and for two SIMD4 float variables. More... | |
| static Simd4Float gmx_simdcall | gmx::andNot (Simd4Float a, Simd4Float b) |
| Bitwise andnot for two SIMD4 float variables. c=(~a) & b. More... | |
| static Simd4Float gmx_simdcall | gmx::operator| (Simd4Float a, Simd4Float b) |
| Bitwise or for two SIMD4 floats. More... | |
| static Simd4Float gmx_simdcall | gmx::operator^ (Simd4Float a, Simd4Float b) |
| Bitwise xor for two SIMD4 float variables. More... | |
| static Simd4Float gmx_simdcall | gmx::operator+ (Simd4Float a, Simd4Float b) |
| Add two float SIMD4 variables. More... | |
| static Simd4Float gmx_simdcall | gmx::operator- (Simd4Float a, Simd4Float b) |
| Subtract two SIMD4 variables. More... | |
| static Simd4Float gmx_simdcall | gmx::operator- (Simd4Float a) |
| SIMD4 floating-point negate. More... | |
| static Simd4Float gmx_simdcall | gmx::operator* (Simd4Float a, Simd4Float b) |
| Multiply two SIMD4 variables. More... | |
| static Simd4Float gmx_simdcall | gmx::fma (Simd4Float a, Simd4Float b, Simd4Float c) |
| SIMD4 Fused-multiply-add. Result is a*b+c. More... | |
| static Simd4Float gmx_simdcall | gmx::fms (Simd4Float a, Simd4Float b, Simd4Float c) |
| SIMD4 Fused-multiply-subtract. Result is a*b-c. More... | |
| static Simd4Float gmx_simdcall | gmx::fnma (Simd4Float a, Simd4Float b, Simd4Float c) |
| SIMD4 Fused-negated-multiply-add. Result is -a*b+c. More... | |
| static Simd4Float gmx_simdcall | gmx::fnms (Simd4Float a, Simd4Float b, Simd4Float c) |
| SIMD4 Fused-negated-multiply-subtract. Result is -a*b-c. More... | |
| static Simd4Float gmx_simdcall | gmx::rsqrt (Simd4Float x) |
| SIMD4 1.0/sqrt(x) lookup. More... | |
| static Simd4Float gmx_simdcall | gmx::abs (Simd4Float a) |
| SIMD4 Floating-point fabs(). More... | |
| static Simd4Float gmx_simdcall | gmx::max (Simd4Float a, Simd4Float b) |
| Set each SIMD4 element to the largest from two variables. More... | |
| static Simd4Float gmx_simdcall | gmx::min (Simd4Float a, Simd4Float b) |
| Set each SIMD4 element to the largest from two variables. More... | |
| static Simd4Float gmx_simdcall | gmx::round (Simd4Float a) |
| SIMD4 Round to nearest integer value (in floating-point format). More... | |
| static Simd4Float gmx_simdcall | gmx::trunc (Simd4Float a) |
| Truncate SIMD4, i.e. round towards zero - common hardware instruction. More... | |
| static float gmx_simdcall | gmx::dotProduct (Simd4Float a, Simd4Float b) |
| Return dot product of two single precision SIMD4 variables. More... | |
| static void gmx_simdcall | gmx::transpose (Simd4Float *v0, Simd4Float *v1, Simd4Float *v2, Simd4Float *v3) |
| SIMD4 float transpose. More... | |
| static Simd4FBool gmx_simdcall | gmx::operator== (Simd4Float a, Simd4Float b) |
| a==b for SIMD4 float More... | |
| static Simd4FBool gmx_simdcall | gmx::operator!= (Simd4Float a, Simd4Float b) |
| a!=b for SIMD4 float More... | |
| static Simd4FBool gmx_simdcall | gmx::operator< (Simd4Float a, Simd4Float b) |
| a<b for SIMD4 float More... | |
| static Simd4FBool gmx_simdcall | gmx::operator<= (Simd4Float a, Simd4Float b) |
| a<=b for SIMD4 float. More... | |
| static Simd4FBool gmx_simdcall | gmx::operator&& (Simd4FBool a, Simd4FBool b) |
| Logical and on single precision SIMD4 booleans. More... | |
| static Simd4FBool gmx_simdcall | gmx::operator|| (Simd4FBool a, Simd4FBool b) |
| Logical or on single precision SIMD4 booleans. More... | |
| static bool gmx_simdcall | gmx::anyTrue (Simd4FBool a) |
| Returns non-zero if any of the boolean in SIMD4 a is True, otherwise 0. More... | |
| static Simd4Float gmx_simdcall | gmx::selectByMask (Simd4Float a, Simd4FBool mask) |
| Select from single precision SIMD4 variable where boolean is true. More... | |
| static Simd4Float gmx_simdcall | gmx::selectByNotMask (Simd4Float a, Simd4FBool mask) |
| Select from single precision SIMD4 variable where boolean is false. More... | |
| static Simd4Float gmx_simdcall | gmx::blend (Simd4Float a, Simd4Float b, Simd4FBool sel) |
| Vector-blend SIMD4 selection. More... | |
| static float gmx_simdcall | gmx::reduce (Simd4Float a) |
| Return sum of all elements in SIMD4 float variable. More... | |
SIMD predefined macros to describe high-level capabilities | |
These macros are used to describe the features available in default Gromacs real precision. They are set from the lower-level implementation files that have macros describing single and double precision individually, as well as the implementation details. | |
| #define | GMX_SIMD_HAVE_REAL GMX_SIMD_HAVE_FLOAT |
| 1 if SimdReal is available, otherwise 0. More... | |
| #define | GMX_SIMD_REAL_WIDTH GMX_SIMD_FLOAT_WIDTH |
| Width of SimdReal. More... | |
| #define | GMX_SIMD_HAVE_INT32_EXTRACT GMX_SIMD_HAVE_FINT32_EXTRACT |
| 1 if support is available for extracting elements from SimdInt32, otherwise 0 More... | |
| #define | GMX_SIMD_HAVE_INT32_LOGICAL GMX_SIMD_HAVE_FINT32_LOGICAL |
| 1 if logical ops are supported on SimdInt32, otherwise 0. More... | |
| #define | GMX_SIMD_HAVE_INT32_ARITHMETICS GMX_SIMD_HAVE_FINT32_ARITHMETICS |
| 1 if arithmetic ops are supported on SimdInt32, otherwise 0. More... | |
| #define | GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_REAL GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_FLOAT |
| 1 if gmx::simdGatherLoadUBySimdIntTranspose is present, otherwise 0 More... | |
| #define | GMX_SIMD_HAVE_HSIMD_UTIL_REAL GMX_SIMD_HAVE_HSIMD_UTIL_FLOAT |
| 1 if real half-register load/store/reduce utils present, otherwise 0 More... | |
| #define | GMX_SIMD4_HAVE_REAL GMX_SIMD4_HAVE_FLOAT |
| 1 if Simd4Real is available, otherwise 0. More... | |
Single precision SIMD math functions | |
| |
| static SimdFloat gmx_simdcall | gmx::copysign (SimdFloat x, SimdFloat y) |
| Composes floating point value with the magnitude of x and the sign of y. More... | |
| static SimdFloat gmx_simdcall | gmx::rsqrtIter (SimdFloat lu, SimdFloat x) |
| Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD float. More... | |
| static SimdFloat gmx_simdcall | gmx::invsqrt (SimdFloat x) |
| Calculate 1/sqrt(x) for SIMD float. More... | |
| static void gmx_simdcall | gmx::invsqrtPair (SimdFloat x0, SimdFloat x1, SimdFloat *out0, SimdFloat *out1) |
| Calculate 1/sqrt(x) for two SIMD floats. More... | |
| static SimdFloat gmx_simdcall | gmx::rcpIter (SimdFloat lu, SimdFloat x) |
| Perform one Newton-Raphson iteration to improve 1/x for SIMD float. More... | |
| static SimdFloat gmx_simdcall | gmx::inv (SimdFloat x) |
| Calculate 1/x for SIMD float. More... | |
| static SimdFloat gmx_simdcall | gmx::operator/ (SimdFloat nom, SimdFloat denom) |
| Division for SIMD floats. More... | |
| static SimdFloat | gmx::maskzInvsqrt (SimdFloat x, SimdFBool m) |
| Calculate 1/sqrt(x) for masked entries of SIMD float. More... | |
| static SimdFloat gmx_simdcall | gmx::maskzInv (SimdFloat x, SimdFBool m) |
| Calculate 1/x for SIMD float, masked version. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdFloat gmx_simdcall | gmx::sqrt (SimdFloat x) |
| Calculate sqrt(x) for SIMD floats. More... | |
| static SimdFloat gmx_simdcall | gmx::log (SimdFloat x) |
| SIMD float log(x). This is the natural logarithm. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdFloat gmx_simdcall | gmx::exp2 (SimdFloat x) |
| SIMD float 2^x. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdFloat gmx_simdcall | gmx::exp (SimdFloat x) |
| SIMD float exp(x). More... | |
| static SimdFloat gmx_simdcall | gmx::erf (SimdFloat x) |
| SIMD float erf(x). More... | |
| static SimdFloat gmx_simdcall | gmx::erfc (SimdFloat x) |
| SIMD float erfc(x). More... | |
| static void gmx_simdcall | gmx::sincos (SimdFloat x, SimdFloat *sinval, SimdFloat *cosval) |
| SIMD float sin & cos. More... | |
| static SimdFloat gmx_simdcall | gmx::sin (SimdFloat x) |
| SIMD float sin(x). More... | |
| static SimdFloat gmx_simdcall | gmx::cos (SimdFloat x) |
| SIMD float cos(x). More... | |
| static SimdFloat gmx_simdcall | gmx::tan (SimdFloat x) |
| SIMD float tan(x). More... | |
| static SimdFloat gmx_simdcall | gmx::asin (SimdFloat x) |
| SIMD float asin(x). More... | |
| static SimdFloat gmx_simdcall | gmx::acos (SimdFloat x) |
| SIMD float acos(x). More... | |
| static SimdFloat gmx_simdcall | gmx::atan (SimdFloat x) |
| SIMD float asin(x). More... | |
| static SimdFloat gmx_simdcall | gmx::atan2 (SimdFloat y, SimdFloat x) |
| SIMD float atan2(y,x). More... | |
| static SimdFloat gmx_simdcall | gmx::pmeForceCorrection (SimdFloat z2) |
| Calculate the force correction due to PME analytically in SIMD float. More... | |
| static SimdFloat gmx_simdcall | gmx::pmePotentialCorrection (SimdFloat z2) |
| Calculate the potential correction due to PME analytically in SIMD float. More... | |
Double precision SIMD math functions | |
| |
| static SimdDouble gmx_simdcall | gmx::copysign (SimdDouble x, SimdDouble y) |
| Composes floating point value with the magnitude of x and the sign of y. More... | |
| static SimdDouble gmx_simdcall | gmx::rsqrtIter (SimdDouble lu, SimdDouble x) |
| Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD double. More... | |
| static SimdDouble gmx_simdcall | gmx::invsqrt (SimdDouble x) |
| Calculate 1/sqrt(x) for SIMD double. More... | |
| static void gmx_simdcall | gmx::invsqrtPair (SimdDouble x0, SimdDouble x1, SimdDouble *out0, SimdDouble *out1) |
| Calculate 1/sqrt(x) for two SIMD doubles. More... | |
| static SimdDouble gmx_simdcall | gmx::rcpIter (SimdDouble lu, SimdDouble x) |
| Perform one Newton-Raphson iteration to improve 1/x for SIMD double. More... | |
| static SimdDouble gmx_simdcall | gmx::inv (SimdDouble x) |
| Calculate 1/x for SIMD double. More... | |
| static SimdDouble gmx_simdcall | gmx::operator/ (SimdDouble nom, SimdDouble denom) |
| Division for SIMD doubles. More... | |
| static SimdDouble | gmx::maskzInvsqrt (SimdDouble x, SimdDBool m) |
| Calculate 1/sqrt(x) for masked entries of SIMD double. More... | |
| static SimdDouble gmx_simdcall | gmx::maskzInv (SimdDouble x, SimdDBool m) |
| Calculate 1/x for SIMD double, masked version. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdDouble gmx_simdcall | gmx::sqrt (SimdDouble x) |
| Calculate sqrt(x) for SIMD doubles. More... | |
| static SimdDouble gmx_simdcall | gmx::log (SimdDouble x) |
| SIMD double log(x). This is the natural logarithm. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdDouble gmx_simdcall | gmx::exp2 (SimdDouble x) |
| SIMD double 2^x. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdDouble gmx_simdcall | gmx::exp (SimdDouble x) |
| SIMD double exp(x). More... | |
| static SimdDouble gmx_simdcall | gmx::erf (SimdDouble x) |
| SIMD double erf(x). More... | |
| static SimdDouble gmx_simdcall | gmx::erfc (SimdDouble x) |
| SIMD double erfc(x). More... | |
| static void gmx_simdcall | gmx::sincos (SimdDouble x, SimdDouble *sinval, SimdDouble *cosval) |
| SIMD double sin & cos. More... | |
| static SimdDouble gmx_simdcall | gmx::sin (SimdDouble x) |
| SIMD double sin(x). More... | |
| static SimdDouble gmx_simdcall | gmx::cos (SimdDouble x) |
| SIMD double cos(x). More... | |
| static SimdDouble gmx_simdcall | gmx::tan (SimdDouble x) |
| SIMD double tan(x). More... | |
| static SimdDouble gmx_simdcall | gmx::asin (SimdDouble x) |
| SIMD double asin(x). More... | |
| static SimdDouble gmx_simdcall | gmx::acos (SimdDouble x) |
| SIMD double acos(x). More... | |
| static SimdDouble gmx_simdcall | gmx::atan (SimdDouble x) |
| SIMD double asin(x). More... | |
| static SimdDouble gmx_simdcall | gmx::atan2 (SimdDouble y, SimdDouble x) |
| SIMD double atan2(y,x). More... | |
| static SimdDouble gmx_simdcall | gmx::pmeForceCorrection (SimdDouble z2) |
| Calculate the force correction due to PME analytically in SIMD double. More... | |
| static SimdDouble gmx_simdcall | gmx::pmePotentialCorrection (SimdDouble z2) |
| Calculate the potential correction due to PME analytically in SIMD double. More... | |
SIMD math functions for double prec. data, single prec. accuracy | |
| |
| static SimdDouble gmx_simdcall | gmx::invsqrtSingleAccuracy (SimdDouble x) |
| Calculate 1/sqrt(x) for SIMD double, but in single accuracy. More... | |
| static SimdDouble | gmx::maskzInvsqrtSingleAccuracy (SimdDouble x, SimdDBool m) |
| 1/sqrt(x) for masked-in entries of SIMD double, but in single accuracy. More... | |
| static void gmx_simdcall | gmx::invsqrtPairSingleAccuracy (SimdDouble x0, SimdDouble x1, SimdDouble *out0, SimdDouble *out1) |
| Calculate 1/sqrt(x) for two SIMD doubles, but single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::invSingleAccuracy (SimdDouble x) |
| Calculate 1/x for SIMD double, but in single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::maskzInvSingleAccuracy (SimdDouble x, SimdDBool m) |
| 1/x for masked entries of SIMD double, single accuracy. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdDouble gmx_simdcall | gmx::sqrtSingleAccuracy (SimdDouble x) |
| Calculate sqrt(x) (correct for 0.0) for SIMD double, with single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::logSingleAccuracy (SimdDouble x) |
| SIMD log(x). Double precision SIMD data, single accuracy. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdDouble gmx_simdcall | gmx::exp2SingleAccuracy (SimdDouble x) |
| SIMD 2^x. Double precision SIMD, single accuracy. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdDouble gmx_simdcall | gmx::expSingleAccuracy (SimdDouble x) |
| SIMD exp(x). Double precision SIMD, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::erfSingleAccuracy (SimdDouble x) |
| SIMD erf(x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::erfcSingleAccuracy (SimdDouble x) |
| SIMD erfc(x). Double precision SIMD data, single accuracy. More... | |
| static void gmx_simdcall | gmx::sinCosSingleAccuracy (SimdDouble x, SimdDouble *sinval, SimdDouble *cosval) |
| SIMD sin & cos. Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::sinSingleAccuracy (SimdDouble x) |
| SIMD sin(x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::cosSingleAccuracy (SimdDouble x) |
| SIMD cos(x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::tanSingleAccuracy (SimdDouble x) |
| SIMD tan(x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::asinSingleAccuracy (SimdDouble x) |
| SIMD asin(x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::acosSingleAccuracy (SimdDouble x) |
| SIMD acos(x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::atanSingleAccuracy (SimdDouble x) |
| SIMD asin(x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::atan2SingleAccuracy (SimdDouble y, SimdDouble x) |
| SIMD atan2(y,x). Double precision SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::pmeForceCorrectionSingleAccuracy (SimdDouble z2) |
| Analytical PME force correction, double SIMD data, single accuracy. More... | |
| static SimdDouble gmx_simdcall | gmx::pmePotentialCorrectionSingleAccuracy (SimdDouble z2) |
| Analytical PME potential correction, double SIMD data, single accuracy. More... | |
SIMD4 math functions | |
| |
| static Simd4Float gmx_simdcall | gmx::rsqrtIter (Simd4Float lu, Simd4Float x) |
| Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD4 float. More... | |
| static Simd4Float gmx_simdcall | gmx::invsqrt (Simd4Float x) |
| Calculate 1/sqrt(x) for SIMD4 float. More... | |
| static Simd4Double gmx_simdcall | gmx::rsqrtIter (Simd4Double lu, Simd4Double x) |
| Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD4 double. More... | |
| static Simd4Double gmx_simdcall | gmx::invsqrt (Simd4Double x) |
| Calculate 1/sqrt(x) for SIMD4 double. More... | |
| static Simd4Double gmx_simdcall | gmx::invsqrtSingleAccuracy (Simd4Double x) |
| Calculate 1/sqrt(x) for SIMD4 double, but in single accuracy. More... | |
Classes | |
| class | gmx::Simd4Double |
| SIMD4 double type. More... | |
| class | gmx::Simd4DBool |
| SIMD4 variable type to use for logical comparisons on doubles. More... | |
| class | gmx::Simd4Float |
| SIMD4 float type. More... | |
| class | gmx::Simd4FBool |
| SIMD4 variable type to use for logical comparisons on floats. More... | |
| class | gmx::SimdDouble |
| Double SIMD variable. Available if GMX_SIMD_HAVE_DOUBLE is 1. More... | |
| class | gmx::SimdDInt32 |
| Integer SIMD variable type to use for conversions to/from double. More... | |
| class | gmx::SimdDBool |
| Boolean type for double SIMD data. More... | |
| class | gmx::SimdDIBool |
| Boolean type for integer datatypes corresponding to double SIMD. More... | |
| class | gmx::SimdFloat |
| Float SIMD variable. Available if GMX_SIMD_HAVE_FLOAT is 1. More... | |
| class | gmx::SimdFInt32 |
| Integer SIMD variable type to use for conversions to/from float. More... | |
| class | gmx::SimdFBool |
| Boolean type for float SIMD data. More... | |
| class | gmx::SimdFIBool |
| Boolean type for integer datatypes corresponding to float SIMD. More... | |
Functions | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadTranspose (const double *base, const std::int32_t offset[], SimdDouble *v0, SimdDouble *v1, SimdDouble *v2, SimdDouble *v3) |
| Load 4 consecutive double from each of GMX_SIMD_DOUBLE_WIDTH offsets, and transpose into 4 SIMD double variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadTranspose (const double *base, const std::int32_t offset[], SimdDouble *v0, SimdDouble *v1) |
| Load 2 consecutive double from each of GMX_SIMD_DOUBLE_WIDTH offsets, and transpose into 2 SIMD double variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadUTranspose (const double *base, const std::int32_t offset[], SimdDouble *v0, SimdDouble *v1, SimdDouble *v2) |
| Load 3 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets, and transpose into 3 SIMD double variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::transposeScatterStoreU (double *base, const std::int32_t offset[], SimdDouble v0, SimdDouble v1, SimdDouble v2) |
| Transpose and store 3 SIMD doubles to 3 consecutive addresses at GMX_SIMD_DOUBLE_WIDTH offsets. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::transposeScatterIncrU (double *base, const std::int32_t offset[], SimdDouble v0, SimdDouble v1, SimdDouble v2) |
| Transpose and add 3 SIMD doubles to 3 consecutive addresses at GMX_SIMD_DOUBLE_WIDTH offsets. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::transposeScatterDecrU (double *base, const std::int32_t offset[], SimdDouble v0, SimdDouble v1, SimdDouble v2) |
| Transpose and subtract 3 SIMD doubles to 3 consecutive addresses at GMX_SIMD_DOUBLE_WIDTH offsets. More... | |
| static void gmx_simdcall | gmx::expandScalarsToTriplets (SimdDouble scalar, SimdDouble *triplets0, SimdDouble *triplets1, SimdDouble *triplets2) |
| Expand each element of double SIMD variable into three identical consecutive elements in three SIMD outputs. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadBySimdIntTranspose (const double *base, SimdDInt32 offset, SimdDouble *v0, SimdDouble *v1, SimdDouble *v2, SimdDouble *v3) |
| Load 4 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets specified by a SIMD integer, transpose into 4 SIMD double variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadUBySimdIntTranspose (const double *base, SimdDInt32 offset, SimdDouble *v0, SimdDouble *v1) |
| Load 2 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets (unaligned) specified by SIMD integer, transpose into 2 SIMD doubles. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadBySimdIntTranspose (const double *base, SimdDInt32 offset, SimdDouble *v0, SimdDouble *v1) |
| Load 2 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets specified by a SIMD integer, transpose into 2 SIMD double variables. More... | |
| static double gmx_simdcall | gmx::reduceIncr4ReturnSum (double *m, SimdDouble v0, SimdDouble v1, SimdDouble v2, SimdDouble v3) |
| Reduce each of four SIMD doubles, add those values to four consecutive doubles in memory, return sum. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadTranspose (const float *base, const std::int32_t offset[], SimdFloat *v0, SimdFloat *v1, SimdFloat *v2, SimdFloat *v3) |
| Load 4 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets, and transpose into 4 SIMD float variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadTranspose (const float *base, const std::int32_t offset[], SimdFloat *v0, SimdFloat *v1) |
| Load 2 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets, and transpose into 2 SIMD float variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadUTranspose (const float *base, const std::int32_t offset[], SimdFloat *v0, SimdFloat *v1, SimdFloat *v2) |
| Load 3 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets, and transpose into 3 SIMD float variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::transposeScatterStoreU (float *base, const std::int32_t offset[], SimdFloat v0, SimdFloat v1, SimdFloat v2) |
| Transpose and store 3 SIMD floats to 3 consecutive addresses at GMX_SIMD_FLOAT_WIDTH offsets. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::transposeScatterIncrU (float *base, const std::int32_t offset[], SimdFloat v0, SimdFloat v1, SimdFloat v2) |
| Transpose and add 3 SIMD floats to 3 consecutive addresses at GMX_SIMD_FLOAT_WIDTH offsets. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::transposeScatterDecrU (float *base, const std::int32_t offset[], SimdFloat v0, SimdFloat v1, SimdFloat v2) |
| Transpose and subtract 3 SIMD floats to 3 consecutive addresses at GMX_SIMD_FLOAT_WIDTH offsets. More... | |
| static void gmx_simdcall | gmx::expandScalarsToTriplets (SimdFloat scalar, SimdFloat *triplets0, SimdFloat *triplets1, SimdFloat *triplets2) |
| Expand each element of float SIMD variable into three identical consecutive elements in three SIMD outputs. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadBySimdIntTranspose (const float *base, SimdFInt32 offset, SimdFloat *v0, SimdFloat *v1, SimdFloat *v2, SimdFloat *v3) |
| Load 4 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets specified by a SIMD integer, transpose into 4 SIMD float variables. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadUBySimdIntTranspose (const float *base, SimdFInt32 offset, SimdFloat *v0, SimdFloat *v1) |
| Load 2 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets (unaligned) specified by SIMD integer, transpose into 2 SIMD floats. More... | |
| template<int align> | |
| static void gmx_simdcall | gmx::gatherLoadBySimdIntTranspose (const float *base, SimdFInt32 offset, SimdFloat *v0, SimdFloat *v1) |
| Load 2 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets specified by a SIMD integer, transpose into 2 SIMD float variables. More... | |
| static float gmx_simdcall | gmx::reduceIncr4ReturnSum (float *m, SimdFloat v0, SimdFloat v1, SimdFloat v2, SimdFloat v3) |
| Reduce each of four SIMD floats, add those values to four consecutive floats in memory, return sum. More... | |
| static SimdFloat gmx_simdcall | gmx::invsqrtSingleAccuracy (SimdFloat x) |
| Calculate 1/sqrt(x) for SIMD float, only targeting single accuracy. More... | |
| static SimdFloat | gmx::maskzInvsqrtSingleAccuracy (SimdFloat x, SimdFBool m) |
| Calculate 1/sqrt(x) for masked SIMD floats, only targeting single accuracy. More... | |
| static void gmx_simdcall | gmx::invsqrtPairSingleAccuracy (SimdFloat x0, SimdFloat x1, SimdFloat *out0, SimdFloat *out1) |
| Calculate 1/sqrt(x) for two SIMD floats, only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::invSingleAccuracy (SimdFloat x) |
| Calculate 1/x for SIMD float, only targeting single accuracy. More... | |
| static SimdFloat | gmx::maskzInvSingleAccuracy (SimdFloat x, SimdFBool m) |
| Calculate 1/x for masked SIMD floats, only targeting single accuracy. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdFloat gmx_simdcall | gmx::sqrtSingleAccuracy (SimdFloat x) |
| Calculate sqrt(x) for SIMD float, always targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::logSingleAccuracy (SimdFloat x) |
| SIMD float log(x), only targeting single accuracy. This is the natural logarithm. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdFloat gmx_simdcall | gmx::exp2SingleAccuracy (SimdFloat x) |
| SIMD float 2^x, only targeting single accuracy. More... | |
| template<MathOptimization opt = MathOptimization::Safe> | |
| static SimdFloat gmx_simdcall | gmx::expSingleAccuracy (SimdFloat x) |
| SIMD float e^x, only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::erfSingleAccuracy (SimdFloat x) |
| SIMD float erf(x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::erfcSingleAccuracy (SimdFloat x) |
| SIMD float erfc(x), only targeting single accuracy. More... | |
| static void gmx_simdcall | gmx::sinCosSingleAccuracy (SimdFloat x, SimdFloat *sinval, SimdFloat *cosval) |
| SIMD float sin & cos, only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::sinSingleAccuracy (SimdFloat x) |
| SIMD float sin(x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::cosSingleAccuracy (SimdFloat x) |
| SIMD float cos(x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::tanSingleAccuracy (SimdFloat x) |
| SIMD float tan(x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::asinSingleAccuracy (SimdFloat x) |
| SIMD float asin(x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::acosSingleAccuracy (SimdFloat x) |
| SIMD float acos(x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::atanSingleAccuracy (SimdFloat x) |
| SIMD float atan(x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::atan2SingleAccuracy (SimdFloat y, SimdFloat x) |
| SIMD float atan2(y,x), only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::pmeForceCorrectionSingleAccuracy (SimdFloat z2) |
| SIMD Analytic PME force correction, only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::pmePotentialCorrectionSingleAccuracy (SimdFloat z2) |
| SIMD Analytic PME potential correction, only targeting single accuracy. More... | |
| static Simd4Float gmx_simdcall | gmx::invsqrtSingleAccuracy (Simd4Float x) |
| Calculate 1/sqrt(x) for SIMD4 float, only targeting single accuracy. More... | |
| static SimdFloat gmx_simdcall | gmx::iprod (SimdFloat ax, SimdFloat ay, SimdFloat az, SimdFloat bx, SimdFloat by, SimdFloat bz) |
| SIMD float inner product of multiple float vectors. More... | |
| static SimdFloat gmx_simdcall | gmx::norm2 (SimdFloat ax, SimdFloat ay, SimdFloat az) |
| SIMD float norm squared of multiple vectors. More... | |
| static void gmx_simdcall | gmx::cprod (SimdFloat ax, SimdFloat ay, SimdFloat az, SimdFloat bx, SimdFloat by, SimdFloat bz, SimdFloat *cx, SimdFloat *cy, SimdFloat *cz) |
| SIMD float cross-product of multiple vectors. More... | |
| static SimdDouble gmx_simdcall | gmx::iprod (SimdDouble ax, SimdDouble ay, SimdDouble az, SimdDouble bx, SimdDouble by, SimdDouble bz) |
| SIMD double inner product of multiple double vectors. More... | |
| static SimdDouble gmx_simdcall | gmx::norm2 (SimdDouble ax, SimdDouble ay, SimdDouble az) |
| SIMD double norm squared of multiple vectors. More... | |
| static void gmx_simdcall | gmx::cprod (SimdDouble ax, SimdDouble ay, SimdDouble az, SimdDouble bx, SimdDouble by, SimdDouble bz, SimdDouble *cx, SimdDouble *cy, SimdDouble *cz) |
| SIMD double cross-product of multiple vectors. More... | |
| static Simd4Float gmx_simdcall | gmx::norm2 (Simd4Float ax, Simd4Float ay, Simd4Float az) |
| SIMD4 float norm squared of multiple vectors. More... | |
| static Simd4Double gmx_simdcall | gmx::norm2 (Simd4Double ax, Simd4Double ay, Simd4Double az) |
| SIMD4 double norm squared of multiple vectors. More... | |
Variables | |
| static const int | gmx::c_simdBestPairAlignmentDouble = 2 |
| Best alignment to use for aligned pairs of double data. More... | |
| static const int | gmx::c_simdBestPairAlignmentFloat = 2 |
| Best alignment to use for aligned pairs of float data. More... | |
Directories | |
| directory | simd |
| SIMD intrinsics interface (simd) | |
| directory | tests |
| Unit tests for SIMD intrinsics interface (simd). | |
Files | |
| file | impl_reference.h |
| Reference SIMD implementation, including SIMD documentation. | |
| file | impl_reference_definitions.h |
| Reference SIMD implementation, including SIMD documentation. | |
| file | impl_reference_general.h |
| Reference SIMD implementation, general utility functions. | |
| file | impl_reference_simd4_double.h |
| Reference implementation, SIMD4 single precision. | |
| file | impl_reference_simd4_float.h |
| Reference implementation, SIMD4 single precision. | |
| file | impl_reference_simd_double.h |
| Reference implementation, SIMD double precision. | |
| file | impl_reference_simd_float.h |
| Reference implementation, SIMD single precision. | |
| file | impl_reference_util_double.h |
| Reference impl., higher-level double prec. SIMD utility functions. | |
| file | impl_reference_util_float.h |
| Reference impl., higher-level single prec. SIMD utility functions. | |
| file | scalar.h |
| Scalar float functions corresponding to GROMACS SIMD functions. | |
| file | scalar_math.h |
| Scalar math functions mimicking GROMACS SIMD math functions. | |
| file | scalar_util.h |
| Scalar utility functions mimicking GROMACS SIMD utility functions. | |
| file | simd.h |
| Definitions, capabilities, and wrappers for SIMD module. | |
| file | simd_math.h |
| Math functions for SIMD datatypes. | |
| file | simd_memory.h |
| Declares SimdArrayRef. | |
| file | support.h |
| Functions to query compiled and supported SIMD architectures. | |
| file | vector_operations.h |
| SIMD operations corresponding to Gromacs rvec datatypes. | |
| #define GMX_SIMD4_HAVE_REAL GMX_SIMD4_HAVE_FLOAT |
1 if Simd4Real is available, otherwise 0.
GMX_SIMD4_HAVE_DOUBLE if GMX_DOUBLE is 1, otherwise GMX_SIMD4_HAVE_FLOAT.
| #define GMX_SIMD_HAVE_FLOAT 1 |
1 when SIMD float support is present, otherwise 0
You should only use this to specifically check for single precision SIMD, support, even when the rest of Gromacs uses double precision.
| #define GMX_SIMD_HAVE_FMA 0 |
1 if the SIMD implementation has fused-multiply add hardware
| #define GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_REAL GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_FLOAT |
1 if gmx::simdGatherLoadUBySimdIntTranspose is present, otherwise 0
GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_DOUBLE if GMX_DOUBLE is 1, otherwise GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE_FLOAT.
| #define GMX_SIMD_HAVE_HSIMD_UTIL_REAL GMX_SIMD_HAVE_HSIMD_UTIL_FLOAT |
1 if real half-register load/store/reduce utils present, otherwise 0
GMX_SIMD_HAVE_HSIMD_UTIL_DOUBLE if GMX_DOUBLE is 1, otherwise GMX_SIMD_HAVE_HSIMD_UTIL_FLOAT.
| #define GMX_SIMD_HAVE_INT32_ARITHMETICS GMX_SIMD_HAVE_FINT32_ARITHMETICS |
1 if arithmetic ops are supported on SimdInt32, otherwise 0.
GMX_SIMD_HAVE_DINT32_ARITHMETICS if GMX_DOUBLE is 1, otherwise GMX_SIMD_HAVE_FINT32_ARITHMETICS.
| #define GMX_SIMD_HAVE_INT32_EXTRACT GMX_SIMD_HAVE_FINT32_EXTRACT |
1 if support is available for extracting elements from SimdInt32, otherwise 0
GMX_SIMD_HAVE_DINT32_EXTRACT if GMX_DOUBLE is 1, otherwise GMX_SIMD_HAVE_FINT32_EXTRACT.
| #define GMX_SIMD_HAVE_INT32_LOGICAL GMX_SIMD_HAVE_FINT32_LOGICAL |
1 if logical ops are supported on SimdInt32, otherwise 0.
GMX_SIMD_HAVE_DINT32_LOGICAL if GMX_DOUBLE is 1, otherwise GMX_SIMD_HAVE_FINT32_LOGICAL.
| #define GMX_SIMD_HAVE_NATIVE_COPYSIGN_DOUBLE 0 |
1 if implementation provides double precision copysign()
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_COPYSIGN_FLOAT 0 |
1 if implementation provides single precision copysign()
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_EXP2_DOUBLE 0 |
1 if implementation provides double precision exp2() faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_EXP2_FLOAT 0 |
1 if implementation provides single precision exp2() faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_EXP_DOUBLE 0 |
1 if implementation provides double precision exp() faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_EXP_FLOAT 0 |
1 if implementation provides single precision exp() faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_LOG_DOUBLE 0 |
1 if implementation provides double precision log() faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_LOG_FLOAT 0 |
1 if implementation provides single precision log() faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_RCP_ITER_DOUBLE 0 |
1 if implementation provides double precision 1/x N-R iterations faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_RCP_ITER_FLOAT 0 |
1 if implementation provides single precision 1/x N-R iterations faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_RSQRT_ITER_DOUBLE 0 |
1 if implementation provides double precision 1/sqrt(x) N-R iterations faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_NATIVE_RSQRT_ITER_FLOAT 0 |
1 if implementation provides single precision 1/sqrt(x) N-R iterations faster than simd_math.h
Only used in simd_math.h to selectively override the generic implementation.
| #define GMX_SIMD_HAVE_REAL GMX_SIMD_HAVE_FLOAT |
1 if SimdReal is available, otherwise 0.
GMX_SIMD_HAVE_DOUBLE if GMX_DOUBLE is 1, otherwise GMX_SIMD_HAVE_FLOAT.
| #define GMX_SIMD_REAL_WIDTH GMX_SIMD_FLOAT_WIDTH |
Width of SimdReal.
GMX_SIMD_DOUBLE_WIDTH if GMX_DOUBLE is 1, otherwise GMX_SIMD_FLOAT_WIDTH.
|
inlinestatic |
SIMD4 Floating-point fabs().
| a | any floating point values |
|
inlinestatic |
|
inlinestatic |
SIMD float acos(x).
| x | The argument to evaluate acos for |
|
inlinestatic |
SIMD double acos(x).
| x | The argument to evaluate acos for |
|
inlinestatic |
SIMD acos(x). Double precision SIMD data, single accuracy.
| x | The argument to evaluate acos for |
|
inlinestatic |
SIMD float acos(x), only targeting single accuracy.
| x | The argument to evaluate acos for |
|
inlinestatic |
Bitwise andnot for two SIMD4 double variables. c=(~a) & b.
Available if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Bitwise andnot for two SIMD4 float variables. c=(~a) & b.
Available if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Returns non-zero if any of the boolean in SIMD4 a is True, otherwise 0.
| a | Logical variable. |
The actual return value for truth will depend on the architecture, so any non-zero value is considered truth.
|
inlinestatic |
Returns non-zero if any of the boolean in SIMD4 a is True, otherwise 0.
| a | Logical variable. |
The actual return value for truth will depend on the architecture, so any non-zero value is considered truth.
|
inlinestatic |
SIMD float asin(x).
| x | The argument to evaluate asin for |
|
inlinestatic |
SIMD double asin(x).
| x | The argument to evaluate asin for |
|
inlinestatic |
SIMD asin(x). Double precision SIMD data, single accuracy.
| x | The argument to evaluate asin for |
|
inlinestatic |
SIMD float asin(x), only targeting single accuracy.
| x | The argument to evaluate asin for |
|
inlinestatic |
SIMD float asin(x).
| x | The argument to evaluate atan for |
|
inlinestatic |
SIMD double asin(x).
| x | The argument to evaluate atan for |
|
inlinestatic |
SIMD float atan2(y,x).
| y | Y component of vector, any quartile |
| x | X component of vector, any quartile |
|
inlinestatic |
SIMD double atan2(y,x).
| y | Y component of vector, any quartile |
| x | X component of vector, any quartile |
|
inlinestatic |
SIMD atan2(y,x). Double precision SIMD data, single accuracy.
| y | Y component of vector, any quartile |
| x | X component of vector, any quartile |
|
inlinestatic |
SIMD float atan2(y,x), only targeting single accuracy.
| y | Y component of vector, any quartile |
| x | X component of vector, any quartile |
|
inlinestatic |
SIMD asin(x). Double precision SIMD data, single accuracy.
| x | The argument to evaluate atan for |
|
inlinestatic |
SIMD float atan(x), only targeting single accuracy.
| x | The argument to evaluate atan for |
|
inlinestatic |
Vector-blend SIMD4 selection.
| a | First source |
| b | Second source |
| sel | Boolean selector |
|
inlinestatic |
Vector-blend SIMD4 selection.
| a | First source |
| b | Second source |
| sel | Boolean selector |
|
inlinestatic |
Composes floating point value with the magnitude of x and the sign of y.
| x | Values to set sign for |
| y | Values used to set sign |
|
inlinestatic |
Composes floating point value with the magnitude of x and the sign of y.
| x | Values to set sign for |
| y | Values used to set sign |
|
inlinestatic |
SIMD float cos(x).
| x | The argument to evaluate cos for |
|
inlinestatic |
SIMD double cos(x).
| x | The argument to evaluate cos for |
|
inlinestatic |
SIMD cos(x). Double precision SIMD data, single accuracy.
| x | The argument to evaluate cos for |
|
inlinestatic |
SIMD float cos(x), only targeting single accuracy.
| x | The argument to evaluate cos for |
|
inlinestatic |
SIMD float cross-product of multiple vectors.
| ax | X components of first vectors | |
| ay | Y components of first vectors | |
| az | Z components of first vectors | |
| bx | X components of second vectors | |
| by | Y components of second vectors | |
| bz | Z components of second vectors | |
| [out] | cx | X components of cross product vectors |
| [out] | cy | Y components of cross product vectors |
| [out] | cz | Z components of cross product vectors |
This calculates C = A x B, where the cross denotes the cross product. The arguments x/y/z denotes the different components, and each element corresponds to a separate vector.
|
inlinestatic |
SIMD double cross-product of multiple vectors.
| ax | X components of first vectors | |
| ay | Y components of first vectors | |
| az | Z components of first vectors | |
| bx | X components of second vectors | |
| by | Y components of second vectors | |
| bz | Z components of second vectors | |
| [out] | cx | X components of cross product vectors |
| [out] | cy | Y components of cross product vectors |
| [out] | cz | Z components of cross product vectors |
This calculates C = A x B, where the cross denotes the cross product. The arguments x/y/z denotes the different components, and each element corresponds to a separate vector.
|
inlinestatic |
Return dot product of two single precision SIMD4 variables.
The dot product is calculated between the first three elements in the two vectors, while the fourth is ignored. The result is returned as a scalar.
| a | vector1 |
| b | vector2 |
|
inlinestatic |
Return dot product of two double precision SIMD4 variables.
The dot product is calculated between the first three elements in the two vectors, while the fourth is ignored. The result is returned as a scalar.
| a | vector1 |
| b | vector2 |
|
inlinestatic |
SIMD float erf(x).
| x | The value to calculate erf(x) for. |
This routine achieves very close to full precision, but we do not care about the last bit or the subnormal result range.
|
inlinestatic |
SIMD double erf(x).
| x | The value to calculate erf(x) for. |
This routine achieves very close to full precision, but we do not care about the last bit or the subnormal result range.
|
inlinestatic |
SIMD float erfc(x).
| x | The value to calculate erfc(x) for. |
This routine achieves full precision (bar the last bit) over most of the input range, but for large arguments where the result is getting close to the minimum representable numbers we accept slightly larger errors (think results that are in the ballpark of 10^-30 for single precision) since that is not relevant for MD.
|
inlinestatic |
SIMD double erfc(x).
| x | The value to calculate erfc(x) for. |
This routine achieves full precision (bar the last bit) over most of the input range, but for large arguments where the result is getting close to the minimum representable numbers we accept slightly larger errors (think results that are in the ballpark of 10^-200 for double) since that is not relevant for MD.
|
inlinestatic |
SIMD erfc(x). Double precision SIMD data, single accuracy.
| x | The value to calculate erfc(x) for. |
This routine achieves singleprecision (bar the last bit) over most of the input range, but for large arguments where the result is getting close to the minimum representable numbers we accept slightly larger errors (think results that are in the ballpark of 10^-30) since that is not relevant for MD.
|
inlinestatic |
SIMD float erfc(x), only targeting single accuracy.
| x | The value to calculate erfc(x) for. |
This routine achieves singleprecision (bar the last bit) over most of the input range, but for large arguments where the result is getting close to the minimum representable numbers we accept slightly larger errors (think results that are in the ballpark of 10^-30) since that is not relevant for MD.
|
inlinestatic |
SIMD erf(x). Double precision SIMD data, single accuracy.
| x | The value to calculate erf(x) for. |
This routine achieves very close to single precision, but we do not care about the last bit or the subnormal result range.
|
inlinestatic |
SIMD float erf(x), only targeting single accuracy.
| x | The value to calculate erf(x) for. |
This routine achieves very close to single precision, but we do not care about the last bit or the subnormal result range.
|
inlinestatic |
SIMD float exp(x).
In addition to scaling the argument for 2^x this routine correctly does extended precision arithmetics to improve accuracy.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to input arguments reaching -126*ln(2)=-87.3 in single, or -1022*ln(2)=-708.4 (double). Similarly, it might overflow for arguments reaching 127*ln(2)=88.0 (single) or 1023*ln(2)=709.1 (double). If the unsafe math optimizations are enabled, small input values that would result in zero-clamped output are not allowed. |
|
inlinestatic |
SIMD double exp(x).
In addition to scaling the argument for 2^x this routine correctly does extended precision arithmetics to improve accuracy.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to input arguments reaching -126*ln(2)=-87.3 in single, or -1022*ln(2)=-708.4 (double). Similarly, it might overflow for arguments reaching 127*ln(2)=88.0 (single) or 1023*ln(2)=709.1 (double). If the unsafe math optimizations are enabled, small input values that would result in zero-clamped output are not allowed. |
|
inlinestatic |
SIMD float 2^x.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to inputs below -126 in single or -1022 in double, and it might overflow for arguments reaching 127 (single) or 1023 (double). If you enable the unsafe math optimization, very small arguments will not necessarily be zero-clamped, but can produce undefined results. |
|
inlinestatic |
SIMD double 2^x.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to inputs below -126 in single or -1022 in double, and it might overflow for arguments reaching 127 (single) or 1023 (double). If you enable the unsafe math optimization, very small arguments will not necessarily be zero-clamped, but can produce undefined results. |
|
inlinestatic |
SIMD 2^x. Double precision SIMD, single accuracy.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to inputs below -126 in single or -1022 in double, and it might overflow for arguments reaching 127 (single) or 1023 (double). If you enable the unsafe math optimization, very small arguments will not necessarily be zero-clamped, but can produce undefined results. |
|
inlinestatic |
SIMD float 2^x, only targeting single accuracy.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to inputs below -126 in single or -1022 in double, and it might overflow for arguments reaching 127 (single) or 1023 (double). If you enable the unsafe math optimization, very small arguments will not necessarily be zero-clamped, but can produce undefined results. |
|
inlinestatic |
Expand each element of double SIMD variable into three identical consecutive elements in three SIMD outputs.
| scalar | Floating-point input, e.g. [s0 s1 s2 s3] if width=4. | |
| [out] | triplets0 | First output, e.g. [s0 s0 s0 s1] if width=4. |
| [out] | triplets1 | Second output, e.g. [s1 s1 s2 s2] if width=4. |
| [out] | triplets2 | Third output, e.g. [s2 s3 s3 s3] if width=4. |
This routine is meant to use for things like scalar-vector multiplication, where the vectors are stored in a merged format like [x0 y0 z0 x1 y1 z1 ...], while the scalars are stored as [s0 s1 s2...], and the data cannot easily be changed to SIMD-friendly layout.
In this case, load 3 full-width SIMD variables from the vector array (This will always correspond to GMX_SIMD_DOUBLE_WIDTH triplets), load a single full-width variable from the scalar array, and call this routine to expand the data. You can then simply multiply the first, second and third pair of SIMD variables, and store the three results back into a suitable vector-format array.
|
inlinestatic |
Expand each element of float SIMD variable into three identical consecutive elements in three SIMD outputs.
| scalar | Floating-point input, e.g. [s0 s1 s2 s3] if width=4. | |
| [out] | triplets0 | First output, e.g. [s0 s0 s0 s1] if width=4. |
| [out] | triplets1 | Second output, e.g. [s1 s1 s2 s2] if width=4. |
| [out] | triplets2 | Third output, e.g. [s2 s3 s3 s3] if width=4. |
This routine is meant to use for things like scalar-vector multiplication, where the vectors are stored in a merged format like [x0 y0 z0 x1 y1 z1 ...], while the scalars are stored as [s0 s1 s2...], and the data cannot easily be changed to SIMD-friendly layout.
In this case, load 3 full-width SIMD variables from the vector array (This will always correspond to GMX_SIMD_FLOAT_WIDTH triplets), load a single full-width variable from the scalar array, and call this routine to expand the data. You can then simply multiply the first, second and third pair of SIMD variables, and store the three results back into a suitable vector-format array.
|
inlinestatic |
SIMD exp(x). Double precision SIMD, single accuracy.
In addition to scaling the argument for 2^x this routine correctly does extended precision arithmetics to improve accuracy.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to input arguments reaching -126*ln(2)=-87.3 in single, or -1022*ln(2)=-708.4 (double). Similarly, it might overflow for arguments reaching 127*ln(2)=88.0 (single) or 1023*ln(2)=709.1 (double). If the unsafe math optimizations are enabled, small input values that would result in zero-clamped output are not allowed. |
|
inlinestatic |
SIMD float e^x, only targeting single accuracy.
In addition to scaling the argument for 2^x this routine correctly does extended precision arithmetics to improve accuracy.
| opt | If this is changed from the default (safe) into the unsafe option, input values that would otherwise lead to zero-clamped results are not allowed and will lead to undefined results. |
| x | Argument. For the default (safe) function version this can be arbitrarily small value, but the routine might clamp the result to zero for arguments that would produce subnormal IEEE754-2008 results. This corresponds to input arguments reaching -126*ln(2)=-87.3 in single, or -1022*ln(2)=-708.4 (double). Similarly, it might overflow for arguments reaching 127*ln(2)=88.0 (single) or 1023*ln(2)=709.1 (double). If the unsafe math optimizations are enabled, small input values that would result in zero-clamped output are not allowed. |
|
inlinestatic |
SIMD4 Fused-multiply-add. Result is a*b+c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
SIMD4 Fused-multiply-add. Result is a*b+c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
SIMD4 Fused-multiply-subtract. Result is a*b-c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
SIMD4 Fused-multiply-subtract. Result is a*b-c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
SIMD4 Fused-negated-multiply-add. Result is -a*b+c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
SIMD4 Fused-negated-multiply-add. Result is -a*b+c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
SIMD4 Fused-negated-multiply-subtract. Result is -a*b-c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
SIMD4 Fused-negated-multiply-subtract. Result is -a*b-c.
| a | factor1 |
| b | factor2 |
| c | term |
|
inlinestatic |
Load 4 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets specified by a SIMD integer, transpose into 4 SIMD double variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 4 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Aligned pointer to the start of the memory. | |
| offset | SIMD integer type with offsets to the start of each triplet. | |
| [out] | v0 | First component, base[align*offset[i]] for each i. |
| [out] | v1 | Second component, base[align*offset[i] + 1] for each i. |
| [out] | v2 | Third component, base[align*offset[i] + 2] for each i. |
| [out] | v3 | Fourth component, base[align*offset[i] + 3] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of four elements and the floating-point SIMD width.
|
inlinestatic |
Load 4 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets specified by a SIMD integer, transpose into 4 SIMD float variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 4 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Aligned pointer to the start of the memory. | |
| offset | SIMD integer type with offsets to the start of each triplet. | |
| [out] | v0 | First component, base[align*offset[i]] for each i. |
| [out] | v1 | Second component, base[align*offset[i] + 1] for each i. |
| [out] | v2 | Third component, base[align*offset[i] + 2] for each i. |
| [out] | v3 | Fourth component, base[align*offset[i] + 3] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of four elements and the floating-point SIMD width.
|
inlinestatic |
Load 2 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets specified by a SIMD integer, transpose into 2 SIMD double variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 2 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Aligned pointer to the start of the memory. | |
| offset | SIMD integer type with offsets to the start of each triplet. | |
| [out] | v0 | First component, base[align*offset[i]] for each i. |
| [out] | v1 | Second component, base[align*offset[i] + 1] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of two elements and the floating-point SIMD width.
|
inlinestatic |
Load 2 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets specified by a SIMD integer, transpose into 2 SIMD float variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 2 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Aligned pointer to the start of the memory. | |
| offset | SIMD integer type with offsets to the start of each triplet. | |
| [out] | v0 | First component, base[align*offset[i]] for each i. |
| [out] | v1 | Second component, base[align*offset[i] + 1] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of two elements and the floating-point SIMD width.
|
inlinestatic |
Load 4 consecutive double from each of GMX_SIMD_DOUBLE_WIDTH offsets, and transpose into 4 SIMD double variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 4 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory area | |
| offset | Array with offsets to the start of each data point. | |
| [out] | v0 | 1st component of data, base[align*offset[i]] for each i. |
| [out] | v1 | 2nd component of data, base[align*offset[i] + 1] for each i. |
| [out] | v2 | 3rd component of data, base[align*offset[i] + 2] for each i. |
| [out] | v3 | 4th component of data, base[align*offset[i] + 3] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of four elements and the floating-point SIMD width.
The offset memory must be aligned to GMX_SIMD_DINT32_WIDTH.
|
inlinestatic |
Load 4 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets, and transpose into 4 SIMD float variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 4 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory area | |
| offset | Array with offsets to the start of each data point. | |
| [out] | v0 | 1st component of data, base[align*offset[i]] for each i. |
| [out] | v1 | 2nd component of data, base[align*offset[i] + 1] for each i. |
| [out] | v2 | 3rd component of data, base[align*offset[i] + 2] for each i. |
| [out] | v3 | 4th component of data, base[align*offset[i] + 3] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of four elements and the floating-point SIMD width.
The offset memory must be aligned to GMX_SIMD_DINT32_WIDTH.
|
inlinestatic |
Load 2 consecutive double from each of GMX_SIMD_DOUBLE_WIDTH offsets, and transpose into 2 SIMD double variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 2 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory area | |
| offset | Array with offsets to the start of each data point. | |
| [out] | v0 | 1st component of data, base[align*offset[i]] for each i. |
| [out] | v1 | 2nd component of data, base[align*offset[i] + 1] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of two elements and the floating-point SIMD width.
The offset memory must be aligned to GMX_SIMD_DINT32_WIDTH.
|
inlinestatic |
Load 2 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets, and transpose into 2 SIMD float variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 2 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory area | |
| offset | Array with offsets to the start of each data point. | |
| [out] | v0 | 1st component of data, base[align*offset[i]] for each i. |
| [out] | v1 | 2nd component of data, base[align*offset[i] + 1] for each i. |
The floating-point memory locations must be aligned, but only to the smaller of two elements and the floating-point SIMD width.
The offset memory must be aligned to GMX_SIMD_FINT32_WIDTH.
To achieve the best possible performance, you should store your data with alignment c_simdBestPairAlignmentFloat in single, or c_simdBestPairAlignmentDouble in double.
|
inlinestatic |
Load 2 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets (unaligned) specified by SIMD integer, transpose into 2 SIMD doubles.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 2 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory. | |
| offset | SIMD integer type with offsets to the start of each triplet. | |
| [out] | v0 | First component, base[align*offset[i]] for each i. |
| [out] | v1 | Second component, base[align*offset[i] + 1] for each i. |
Since some SIMD architectures cannot handle any unaligned loads, this routine is only available if GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE is 1.
|
inlinestatic |
Load 2 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets (unaligned) specified by SIMD integer, transpose into 2 SIMD floats.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 2 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory. | |
| offset | SIMD integer type with offsets to the start of each triplet. | |
| [out] | v0 | First component, base[align*offset[i]] for each i. |
| [out] | v1 | Second component, base[align*offset[i] + 1] for each i. |
Since some SIMD architectures cannot handle any unaligned loads, this routine is only available if GMX_SIMD_HAVE_GATHER_LOADU_BYSIMDINT_TRANSPOSE is 1.
|
inlinestatic |
Load 3 consecutive doubles from each of GMX_SIMD_DOUBLE_WIDTH offsets, and transpose into 3 SIMD double variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory area | |
| offset | Array with offsets to the start of each data point. | |
| [out] | v0 | 1st component of data, base[align*offset[i]] for each i. |
| [out] | v1 | 2nd component of data, base[align*offset[i] + 1] for each i. |
| [out] | v2 | 3rd component of data, base[align*offset[i] + 2] for each i. |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory load operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Load 3 consecutive floats from each of GMX_SIMD_FLOAT_WIDTH offsets, and transpose into 3 SIMD float variables.
| align | Alignment of the memory from which we read, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the input data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are loaded. |
| base | Pointer to the start of the memory area | |
| offset | Array with offsets to the start of each data point. | |
| [out] | v0 | 1st component of data, base[align*offset[i]] for each i. |
| [out] | v1 | 2nd component of data, base[align*offset[i] + 1] for each i. |
| [out] | v2 | 3rd component of data, base[align*offset[i] + 2] for each i. |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory load operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Calculate 1/x for SIMD float.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/x for SIMD double.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/x for SIMD double, but in single accuracy.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/x for SIMD float, only targeting single accuracy.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD float.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD double.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD4 float.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD4 double.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for two SIMD floats.
| x0 | First set of arguments, x0 must be in single range (see below). | |
| x1 | Second set of arguments, x1 must be in single range (see below). | |
| [out] | out0 | Result 1/sqrt(x0) |
| [out] | out1 | Result 1/sqrt(x1) |
In particular for double precision we can sometimes calculate square root pairs slightly faster by using single precision until the very last step.
|
inlinestatic |
Calculate 1/sqrt(x) for two SIMD doubles.
| x0 | First set of arguments, x0 must be in single range (see below). | |
| x1 | Second set of arguments, x1 must be in single range (see below). | |
| [out] | out0 | Result 1/sqrt(x0) |
| [out] | out1 | Result 1/sqrt(x1) |
In particular for double precision we can sometimes calculate square root pairs slightly faster by using single precision until the very last step.
|
inlinestatic |
Calculate 1/sqrt(x) for two SIMD doubles, but single accuracy.
| x0 | First set of arguments, x0 must be in single range (see below). | |
| x1 | Second set of arguments, x1 must be in single range (see below). | |
| [out] | out0 | Result 1/sqrt(x0) |
| [out] | out1 | Result 1/sqrt(x1) |
In particular for double precision we can sometimes calculate square root pairs slightly faster by using single precision until the very last step.
|
inlinestatic |
Calculate 1/sqrt(x) for two SIMD floats, only targeting single accuracy.
| x0 | First set of arguments, x0 must be in single range (see below). | |
| x1 | Second set of arguments, x1 must be in single range (see below). | |
| [out] | out0 | Result 1/sqrt(x0) |
| [out] | out1 | Result 1/sqrt(x1) |
In particular for double precision we can sometimes calculate square root pairs slightly faster by using single precision until the very last step.
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD double, but in single accuracy.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD4 double, but in single accuracy.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD float, only targeting single accuracy.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Calculate 1/sqrt(x) for SIMD4 float, only targeting single accuracy.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
SIMD float inner product of multiple float vectors.
| ax | X components of first vectors |
| ay | Y components of first vectors |
| az | Z components of first vectors |
| bx | X components of second vectors |
| by | Y components of second vectors |
| bz | Z components of second vectors |
|
inlinestatic |
SIMD double inner product of multiple double vectors.
| ax | X components of first vectors |
| ay | Y components of first vectors |
| az | Z components of first vectors |
| bx | X components of second vectors |
| by | Y components of second vectors |
| bz | Z components of second vectors |
|
inlinestatic |
Load 4 float values from aligned memory into SIMD4 variable.
| m | Pointer to memory aligned to 4 elements. |
|
inlinestatic |
Load 4 double values from aligned memory into SIMD4 variable.
| m | Pointer to memory aligned to 4 elements. |
|
inlinestatic |
Load SIMD4 float from unaligned memory.
Available if GMX_SIMD_HAVE_LOADU is 1.
| m | Pointer to memory, no alignment requirement. |
|
inlinestatic |
Load SIMD4 double from unaligned memory.
Available if GMX_SIMD_HAVE_LOADU is 1.
| m | Pointer to memory, no alignment requirement. |
|
inlinestatic |
SIMD float log(x). This is the natural logarithm.
| x | Argument, should be >0. |
|
inlinestatic |
SIMD double log(x). This is the natural logarithm.
| x | Argument, should be >0. |
|
inlinestatic |
SIMD log(x). Double precision SIMD data, single accuracy.
| x | Argument, should be >0. |
|
inlinestatic |
SIMD float log(x), only targeting single accuracy. This is the natural logarithm.
| x | Argument, should be >0. |
|
inlinestatic |
Calculate 1/x for SIMD float, masked version.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX for masked-in entries. See invsqrt for the discussion about argument restrictions. |
| m | Mask |
|
inlinestatic |
Calculate 1/x for SIMD double, masked version.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX for masked-in entries. See invsqrt for the discussion about argument restrictions. |
| m | Mask |
|
inlinestatic |
1/x for masked entries of SIMD double, single accuracy.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
| m | Mask |
|
inlinestatic |
Calculate 1/x for masked SIMD floats, only targeting single accuracy.
| x | Argument with magnitude larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
| m | Mask |
|
inlinestatic |
Calculate 1/sqrt(x) for masked entries of SIMD float.
This routine only evaluates 1/sqrt(x) for elements for which mask is true. Illegal values in the masked-out elements will not lead to floating-point exceptions.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX for masked-in entries. See invsqrt for the discussion about argument restrictions. |
| m | Mask |
|
inlinestatic |
Calculate 1/sqrt(x) for masked entries of SIMD double.
This routine only evaluates 1/sqrt(x) for elements for which mask is true. Illegal values in the masked-out elements will not lead to floating-point exceptions.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX for masked-in entries. See invsqrt for the discussion about argument restrictions. |
| m | Mask |
|
inlinestatic |
1/sqrt(x) for masked-in entries of SIMD double, but in single accuracy.
This routine only evaluates 1/sqrt(x) for elements for which mask is true. Illegal values in the masked-out elements will not lead to floating-point exceptions.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
| m | Mask |
|
inlinestatic |
Calculate 1/sqrt(x) for masked SIMD floats, only targeting single accuracy.
This routine only evaluates 1/sqrt(x) for elements for which mask is true. Illegal values in the masked-out elements will not lead to floating-point exceptions.
| x | Argument that must be larger than GMX_FLOAT_MIN and smaller than GMX_FLOAT_MAX, i.e. within the range of single precision. For the single precision implementation this is obviously always true for positive values, but for double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
| m | Mask |
|
inlinestatic |
Set each SIMD4 element to the largest from two variables.
| a | Any floating-point value |
| b | Any floating-point value |
|
inlinestatic |
Set each SIMD4 element to the largest from two variables.
| a | Any floating-point value |
| b | Any floating-point value |
|
inlinestatic |
Set each SIMD4 element to the largest from two variables.
| a | Any floating-point value |
| b | Any floating-point value |
|
inlinestatic |
Set each SIMD4 element to the largest from two variables.
| a | Any floating-point value |
| b | Any floating-point value |
|
inlinestatic |
SIMD float norm squared of multiple vectors.
| ax | X components of vectors |
| ay | Y components of vectors |
| az | Z components of vectors |
|
inlinestatic |
SIMD double norm squared of multiple vectors.
| ax | X components of vectors |
| ay | Y components of vectors |
| az | Z components of vectors |
|
inlinestatic |
SIMD4 float norm squared of multiple vectors.
| ax | X components of vectors |
| ay | Y components of vectors |
| az | Z components of vectors |
|
inlinestatic |
SIMD4 double norm squared of multiple vectors.
| ax | X components of vectors |
| ay | Y components of vectors |
| az | Z components of vectors |
|
inlinestatic |
a!=b for SIMD4 float
| a | value1 |
| b | value2 |
|
inlinestatic |
a!=b for SIMD4 double
| a | value1 |
| b | value2 |
|
inlinestatic |
Bitwise and for two SIMD4 float variables.
Supported if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Bitwise and for two SIMD4 double variables.
Supported if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Logical and on single precision SIMD4 booleans.
| a | logical vars 1 |
| b | logical vars 2 |
|
inlinestatic |
Logical and on single precision SIMD4 booleans.
| a | logical vars 1 |
| b | logical vars 2 |
|
inlinestatic |
Multiply two SIMD4 variables.
| a | factor1 |
| b | factor2 |
|
inlinestatic |
Multiply two SIMD4 variables.
| a | factor1 |
| b | factor2 |
|
inlinestatic |
Add two double SIMD4 variables.
| a | term1 |
| b | term2 |
|
inlinestatic |
Add two float SIMD4 variables.
| a | term1 |
| b | term2 |
|
inlinestatic |
Subtract two SIMD4 variables.
| a | term1 |
| b | term2 |
|
inlinestatic |
Subtract two SIMD4 variables.
| a | term1 |
| b | term2 |
|
inlinestatic |
SIMD4 floating-point negate.
| a | SIMD4 floating-point value |
|
inlinestatic |
SIMD4 floating-point negate.
| a | SIMD4 floating-point value |
|
inlinestatic |
Division for SIMD floats.
| nom | Nominator |
| denom | Denominator, with magnitude in range (GMX_FLOAT_MIN,GMX_FLOAT_MAX). For single precision this is equivalent to a nonzero argument, but in double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
Division for SIMD doubles.
| nom | Nominator |
| denom | Denominator, with magnitude in range (GMX_FLOAT_MIN,GMX_FLOAT_MAX). For single precision this is equivalent to a nonzero argument, but in double precision it adds an extra restriction since the first lookup step might have to be performed in single precision on some architectures. Note that the responsibility for checking falls on you - this routine does not check arguments. |
|
inlinestatic |
a<b for SIMD4 float
| a | value1 |
| b | value2 |
|
inlinestatic |
a<b for SIMD4 double
| a | value1 |
| b | value2 |
|
inlinestatic |
a<=b for SIMD4 float.
| a | value1 |
| b | value2 |
|
inlinestatic |
a<=b for SIMD4 double.
| a | value1 |
| b | value2 |
|
inlinestatic |
a==b for SIMD4 float
| a | value1 |
| b | value2 |
|
inlinestatic |
a==b for SIMD4 double
| a | value1 |
| b | value2 |
|
inlinestatic |
Bitwise xor for two SIMD4 float variables.
Available if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Bitwise xor for two SIMD4 double variables.
Available if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Bitwise or for two SIMD4 floats.
Available if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Bitwise or for two SIMD4 doubles.
Available if GMX_SIMD_HAVE_LOGICAL is 1.
| a | data1 |
| b | data2 |
|
inlinestatic |
Logical or on single precision SIMD4 booleans.
| a | logical vars 1 |
| b | logical vars 2 |
Note that this is not necessarily a bitwise operation - the storage format of booleans is implementation-dependent.
|
inlinestatic |
Logical or on single precision SIMD4 booleans.
| a | logical vars 1 |
| b | logical vars 2 |
Note that this is not necessarily a bitwise operation - the storage format of booleans is implementation-dependent.
|
inlinestatic |
Calculate the force correction due to PME analytically in SIMD float.
| z2 |  - see below for details. |
This routine is meant to enable analytical evaluation of the direct-space PME electrostatic force to avoid tables.
The direct-space potential should be 
, but there are some problems evaluating that:
First, the error function is difficult (read: expensive) to approxmiate accurately for intermediate to large arguments, and this happens already in ranges of 
that occur in simulations. Second, we now try to avoid calculating potentials in Gromacs but use forces directly.
We can simply things slight by noting that the PME part is really a correction to the normal Coulomb force since 
, i.e.
![\[ V = \frac{1}{r} - \frac{\mbox{erf}(\beta r)}{r} \]](form_18.png)
The first term we already have from the inverse square root, so that we can leave out of this routine.
For pme tolerances of 1e-3 to 1e-8 and cutoffs of 0.5nm to 1.8nm, the argument 
will be in the range 0.15 to ~4, which is the range used for the minimax fit. Use your favorite plotting program to realize how well-behaved 
is in this range!
We approximate 
with a rational minimax polynomial. However, it turns out it is more efficient to approximate 
and then only use even powers. This is another minor optimization, since we actually want , because it is going to be multiplied by the vector between the two atoms to get the vectorial force. The fastest flops are the ones we can avoid calculating!
So, here's how it should be used:
.
, so you get 
.
as the argument.The return value is the expression:
![\[ \frac{2 \exp{-z^2}}{\sqrt{\pi} z^2}-\frac{\mbox{erf}(z)}{z^3} \]](form_27.png)
Multiply the entire expression by 
. This will get you
![\[ \frac{2 \beta^3 \exp(-z^2)}{\sqrt{\pi} z^2} - \frac{\beta^3 \mbox{erf}(z)}{z^3} \]](form_29.png)
or, switching back to 
(since 
):
![\[ \frac{2 \beta \exp(-r^2 \beta^2)}{\sqrt{\pi} r^2} - \frac{\mbox{erf}(r \beta)}{r^3} \]](form_32.png)
With a bit of math exercise you should be able to confirm that this is exactly
![\[ \frac{\frac{d}{dr}\left( \frac{\mbox{erf}(\beta r)}{r} \right)}{r} \]](form_33.png)
, multiply by the product of the charges, and you have your force (divided by ). A final multiplication with the vector connecting the two particles and you have your vectorial force to add to the particles.This approximation achieves an error slightly lower than 1e-6 in single precision and 1e-11 in double precision for arguments smaller than 16 ( 
); when added to 
the error will be insignificant. For 
the return value can be inf or NaN.
|
inlinestatic |
Calculate the force correction due to PME analytically in SIMD double.
| z2 | This should be the value , where r is your interaction distance and beta the ewald splitting parameters. |
This routine is meant to enable analytical evaluation of the direct-space PME electrostatic force to avoid tables. For details, see the single precision function.
|
inlinestatic |
Analytical PME force correction, double SIMD data, single accuracy.
| z2 | - see below for details. |
This routine is meant to enable analytical evaluation of the direct-space PME electrostatic force to avoid tables.
The direct-space potential should be , but there are some problems evaluating that:
First, the error function is difficult (read: expensive) to approxmiate accurately for intermediate to large arguments, and this happens already in ranges of that occur in simulations. Second, we now try to avoid calculating potentials in Gromacs but use forces directly.
We can simply things slight by noting that the PME part is really a correction to the normal Coulomb force since , i.e.
The first term we already have from the inverse square root, so that we can leave out of this routine.
For pme tolerances of 1e-3 to 1e-8 and cutoffs of 0.5nm to 1.8nm, the argument will be in the range 0.15 to ~4. Use your favorite plotting program to realize how well-behaved is in this range!
We approximate with a rational minimax polynomial. However, it turns out it is more efficient to approximate and then only use even powers. This is another minor optimization, since we actually want , because it is going to be multiplied by the vector between the two atoms to get the vectorial force. The fastest flops are the ones we can avoid calculating!
So, here's how it should be used:
., so you get . as the argument.The return value is the expression:
Multiply the entire expression by . This will get you
or, switching back to (since ):
With a bit of math exercise you should be able to confirm that this is exactly
, multiply by the product of the charges, and you have your force (divided by ). A final multiplication with the vector connecting the two particles and you have your vectorial force to add to the particles.This approximation achieves an accuracy slightly lower than 1e-6; when added to the error will be insignificant.
|
inlinestatic |
SIMD Analytic PME force correction, only targeting single accuracy.
| z2 | - see default single precision version for details. |
|
inlinestatic |
Calculate the potential correction due to PME analytically in SIMD float.
| z2 | - see below for details. |
See pmeForceCorrection for details about the approximation.
This routine calculates 
, although you should provide as the input argument.
Here's how it should be used:
., so you get 
.The return value is the expression:
![\[ \frac{\mbox{erf}(z)}{z} \]](form_40.png)
Multiply the entire expression by beta and switching back to (since ):
![\[ \frac{\mbox{erf}(r \beta)}{r} \]](form_41.png)
, multiply by the product of the charges, and you have your potential.This approximation achieves an error slightly lower than 1e-6 in single precision and 4e-11 in double precision for arguments smaller than 16 ( 
); for 
the error can be twice as high; when added to the error will be insignificant. For 
the return value can be inf or NaN.
|
inlinestatic |
Calculate the potential correction due to PME analytically in SIMD double.
| z2 | This should be the value , where r is your interaction distance and beta the ewald splitting parameters. |
This routine is meant to enable analytical evaluation of the direct-space PME electrostatic potential to avoid tables. For details, see the single precision function.
|
inlinestatic |
Analytical PME potential correction, double SIMD data, single accuracy.
| z2 | - see below for details. |
This routine calculates , although you should provide as the input argument.
Here's how it should be used:
., so you get .The return value is the expression:
Multiply the entire expression by beta and switching back to (since ):
, multiply by the product of the charges, and you have your potential.This approximation achieves an accuracy slightly lower than 1e-6; when added to the error will be insignificant.
|
inlinestatic |
SIMD Analytic PME potential correction, only targeting single accuracy.
| z2 | - see default single precision version for details. |
|
inlinestatic |
Perform one Newton-Raphson iteration to improve 1/x for SIMD float.
This is a low-level routine that should only be used by SIMD math routine that evaluates the reciprocal.
| lu | Approximation of 1/x, typically obtained from lookup. |
| x | The reference (starting) value x for which we want 1/x. |
|
inlinestatic |
Perform one Newton-Raphson iteration to improve 1/x for SIMD double.
This is a low-level routine that should only be used by SIMD math routine that evaluates the reciprocal.
| lu | Approximation of 1/x, typically obtained from lookup. |
| x | The reference (starting) value x for which we want 1/x. |
|
inlinestatic |
Return sum of all elements in SIMD4 float variable.
| a | SIMD4 variable to reduce/sum. |
|
inlinestatic |
Return sum of all elements in SIMD4 double variable.
| a | SIMD4 variable to reduce/sum. |
|
inlinestatic |
Reduce each of four SIMD doubles, add those values to four consecutive doubles in memory, return sum.
| m | Pointer to memory where four doubles should be incremented |
| v0 | SIMD variable whose sum should be added to m[0] |
| v1 | SIMD variable whose sum should be added to m[1] |
| v2 | SIMD variable whose sum should be added to m[2] |
| v3 | SIMD variable whose sum should be added to m[3] |
The pointer m must be aligned to the smaller of four elements and the floating-point SIMD width.
|
inlinestatic |
Reduce each of four SIMD floats, add those values to four consecutive floats in memory, return sum.
| m | Pointer to memory where four floats should be incremented |
| v0 | SIMD variable whose sum should be added to m[0] |
| v1 | SIMD variable whose sum should be added to m[1] |
| v2 | SIMD variable whose sum should be added to m[2] |
| v3 | SIMD variable whose sum should be added to m[3] |
The pointer m must be aligned to the smaller of four elements and the floating-point SIMD width.
|
inlinestatic |
SIMD4 Round to nearest integer value (in floating-point format).
| a | Any floating-point value |
|
inlinestatic |
SIMD4 Round to nearest integer value (in floating-point format).
| a | Any floating-point value |
|
inlinestatic |
SIMD4 1.0/sqrt(x) lookup.
This is a low-level instruction that should only be called from routines implementing the inverse square root in simd_math.h.
| x | Argument, x>0 |
|
inlinestatic |
SIMD4 1.0/sqrt(x) lookup.
This is a low-level instruction that should only be called from routines implementing the inverse square root in simd_math.h.
| x | Argument, x>0 |
|
inlinestatic |
Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD float.
This is a low-level routine that should only be used by SIMD math routine that evaluates the inverse square root.
| lu | Approximation of 1/sqrt(x), typically obtained from lookup. |
| x | The reference (starting) value x for which we want 1/sqrt(x). |
|
inlinestatic |
Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD double.
This is a low-level routine that should only be used by SIMD math routine that evaluates the inverse square root.
| lu | Approximation of 1/sqrt(x), typically obtained from lookup. |
| x | The reference (starting) value x for which we want 1/sqrt(x). |
|
inlinestatic |
Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD4 float.
This is a low-level routine that should only be used by SIMD math routine that evaluates the inverse square root.
| lu | Approximation of 1/sqrt(x), typically obtained from lookup. |
| x | The reference (starting) value x for which we want 1/sqrt(x). |
|
inlinestatic |
Perform one Newton-Raphson iteration to improve 1/sqrt(x) for SIMD4 double.
This is a low-level routine that should only be used by SIMD math routine that evaluates the inverse square root.
| lu | Approximation of 1/sqrt(x), typically obtained from lookup. |
| x | The reference (starting) value x for which we want 1/sqrt(x). |
|
inlinestatic |
Select from single precision SIMD4 variable where boolean is true.
| a | Floating-point variable to select from |
| mask | Boolean selector |
|
inlinestatic |
Select from single precision SIMD4 variable where boolean is true.
| a | Floating-point variable to select from |
| mask | Boolean selector |
|
inlinestatic |
Select from single precision SIMD4 variable where boolean is false.
| a | Floating-point variable to select from |
| mask | Boolean selector |
|
inlinestatic |
Select from single precision SIMD4 variable where boolean is false.
| a | Floating-point variable to select from |
| mask | Boolean selector |
|
inlinestatic |
Set all SIMD4 double elements to 0.
You should typically just call gmx::setZero(), which uses proxy objects internally to handle all types rather than adding the suffix used here.
|
inlinestatic |
Set all SIMD4 float elements to 0.
You should typically just call gmx::setZero(), which uses proxy objects internally to handle all types rather than adding the suffix used here.
|
inlinestatic |
SIMD float sin(x).
| x | The argument to evaluate sin for |
|
inlinestatic |
SIMD double sin(x).
| x | The argument to evaluate sin for |
|
inlinestatic |
SIMD float sin & cos.
| x | The argument to evaluate sin/cos for | |
| [out] | sinval | Sin(x) |
| [out] | cosval | Cos(x) |
This version achieves close to machine precision, but for very large magnitudes of the argument we inherently begin to lose accuracy due to the argument reduction, despite using extended precision arithmetics internally.
|
inlinestatic |
SIMD double sin & cos.
| x | The argument to evaluate sin/cos for | |
| [out] | sinval | Sin(x) |
| [out] | cosval | Cos(x) |
This version achieves close to machine precision, but for very large magnitudes of the argument we inherently begin to lose accuracy due to the argument reduction, despite using extended precision arithmetics internally.
|
inlinestatic |
SIMD sin & cos. Double precision SIMD data, single accuracy.
| x | The argument to evaluate sin/cos for | |
| [out] | sinval | Sin(x) |
| [out] | cosval | Cos(x) |
|
inlinestatic |
SIMD float sin & cos, only targeting single accuracy.
| x | The argument to evaluate sin/cos for | |
| [out] | sinval | Sin(x) |
| [out] | cosval | Cos(x) |
|
inlinestatic |
SIMD sin(x). Double precision SIMD data, single accuracy.
| x | The argument to evaluate sin for |
|
inlinestatic |
SIMD float sin(x), only targeting single accuracy.
| x | The argument to evaluate sin for |
|
inlinestatic |
Calculate sqrt(x) for SIMD floats.
| opt | By default, this function checks if the input value is 0.0 and masks this to return the correct result. If you are certain your argument will never be zero, and you know you need to save every single cycle you can, you can alternatively call the function as sqrt<MathOptimization::Unsafe>(x). |
| x | Argument that must be in range 0 <=x <= GMX_FLOAT_MAX, since the lookup step often has to be implemented in single precision. Arguments smaller than GMX_FLOAT_MIN will always lead to a zero result, even in double precision. If you are using the unsafe math optimization parameter, the argument must be in the range GMX_FLOAT_MIN <= x <= GMX_FLOAT_MAX. |
|
inlinestatic |
Calculate sqrt(x) for SIMD doubles.
| opt | By default, this function checks if the input value is 0.0 and masks this to return the correct result. If you are certain your argument will never be zero, and you know you need to save every single cycle you can, you can alternatively call the function as sqrt<MathOptimization::Unsafe>(x). |
| x | Argument that must be in range 0 <=x <= GMX_FLOAT_MAX, since the lookup step often has to be implemented in single precision. Arguments smaller than GMX_FLOAT_MIN will always lead to a zero result, even in double precision. If you are using the unsafe math optimization parameter, the argument must be in the range GMX_FLOAT_MIN <= x <= GMX_FLOAT_MAX. |
|
inlinestatic |
Calculate sqrt(x) (correct for 0.0) for SIMD double, with single accuracy.
| opt | By default, this function checks if the input value is 0.0 and masks this to return the correct result. If you are certain your argument will never be zero, and you know you need to save every single cycle you can, you can alternatively call the function as sqrt<MathOptimization::Unsafe>(x). |
| x | Argument that must be in range 0 <=x <= GMX_FLOAT_MAX, since the lookup step often has to be implemented in single precision. Arguments smaller than GMX_FLOAT_MIN will always lead to a zero result, even in double precision. If you are using the unsafe math optimization parameter, the argument must be in the range GMX_FLOAT_MIN <= x <= GMX_FLOAT_MAX. |
|
inlinestatic |
Calculate sqrt(x) for SIMD float, always targeting single accuracy.
| opt | By default, this function checks if the input value is 0.0 and masks this to return the correct result. If you are certain your argument will never be zero, and you know you need to save every single cycle you can, you can alternatively call the function as sqrt<MathOptimization::Unsafe>(x). |
| x | Argument that must be in range 0 <=x <= GMX_FLOAT_MAX, since the lookup step often has to be implemented in single precision. Arguments smaller than GMX_FLOAT_MIN will always lead to a zero result, even in double precision. If you are using the unsafe math optimization parameter, the argument must be in the range GMX_FLOAT_MIN <= x <= GMX_FLOAT_MAX. |
|
inlinestatic |
Store the contents of SIMD4 double to aligned memory m.
| [out] | m | Pointer to memory, aligned to 4 elements. |
| a | SIMD4 variable to store |
|
inlinestatic |
Store the contents of SIMD4 float to aligned memory m.
| [out] | m | Pointer to memory, aligned to 4 elements. |
| a | SIMD4 variable to store |
|
inlinestatic |
Store SIMD4 float to unaligned memory.
Available if GMX_SIMD_HAVE_STOREU is 1.
| [out] | m | Pointer to memory, no alignment requirement. |
| a | SIMD4 variable to store. |
|
inlinestatic |
Store SIMD4 double to unaligned memory.
Available if GMX_SIMD_HAVE_STOREU is 1.
| [out] | m | Pointer to memory, no alignment requirement. |
| a | SIMD4 variable to store. |
|
inlinestatic |
SIMD float tan(x).
| x | The argument to evaluate tan for |
|
inlinestatic |
SIMD double tan(x).
| x | The argument to evaluate tan for |
|
inlinestatic |
SIMD tan(x). Double precision SIMD data, single accuracy.
| x | The argument to evaluate tan for |
|
inlinestatic |
SIMD float tan(x), only targeting single accuracy.
| x | The argument to evaluate tan for |
|
inlinestatic |
SIMD4 float transpose.
| [in,out] | v0 | Row 0 on input, column 0 on output |
| [in,out] | v1 | Row 1 on input, column 1 on output |
| [in,out] | v2 | Row 2 on input, column 2 on output |
| [in,out] | v3 | Row 3 on input, column 3 on output |
|
inlinestatic |
SIMD4 double transpose.
| [in,out] | v0 | Row 0 on input, column 0 on output |
| [in,out] | v1 | Row 1 on input, column 1 on output |
| [in,out] | v2 | Row 2 on input, column 2 on output |
| [in,out] | v3 | Row 3 on input, column 3 on output |
|
inlinestatic |
Transpose and subtract 3 SIMD doubles to 3 consecutive addresses at GMX_SIMD_DOUBLE_WIDTH offsets.
| align | Alignment of the memory to which we write, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the output data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are decremented. |
| [out] | base | Pointer to start of memory. |
| offset | Aligned array with offsets to the start of each triplet. | |
| v0 | 1st component, subtracted from base[align*offset[i]] | |
| v1 | 2nd component, subtracted from base[align*offset[i]+1] | |
| v2 | 3rd component, subtracted from base[align*offset[i]+2] |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory load/store operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Transpose and subtract 3 SIMD floats to 3 consecutive addresses at GMX_SIMD_FLOAT_WIDTH offsets.
| align | Alignment of the memory to which we write, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the output data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are decremented. |
| [out] | base | Pointer to start of memory. |
| offset | Aligned array with offsets to the start of each triplet. | |
| v0 | 1st component, subtracted from base[align*offset[i]] | |
| v1 | 2nd component, subtracted from base[align*offset[i]+1] | |
| v2 | 3rd component, subtracted from base[align*offset[i]+2] |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory load/store operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Transpose and add 3 SIMD doubles to 3 consecutive addresses at GMX_SIMD_DOUBLE_WIDTH offsets.
| align | Alignment of the memory to which we write, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the output data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are incremented. |
| [out] | base | Pointer to the start of the memory area |
| offset | Aligned array with offsets to the start of each triplet. | |
| v0 | 1st component of triplets, added to base[align*offset[i]]. | |
| v1 | 2nd component of triplets, added to base[align*offset[i] + 1]. | |
| v2 | 3rd component of triplets, added to base[align*offset[i] + 2]. |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory load/store operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Transpose and add 3 SIMD floats to 3 consecutive addresses at GMX_SIMD_FLOAT_WIDTH offsets.
| align | Alignment of the memory to which we write, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the output data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are incremented. |
| [out] | base | Pointer to the start of the memory area |
| offset | Aligned array with offsets to the start of each triplet. | |
| v0 | 1st component of triplets, added to base[align*offset[i]]. | |
| v1 | 2nd component of triplets, added to base[align*offset[i] + 1]. | |
| v2 | 3rd component of triplets, added to base[align*offset[i] + 2]. |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory load/store operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Transpose and store 3 SIMD doubles to 3 consecutive addresses at GMX_SIMD_DOUBLE_WIDTH offsets.
| align | Alignment of the memory to which we write, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the output data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are written. |
| [out] | base | Pointer to the start of the memory area |
| offset | Aligned array with offsets to the start of each triplet. | |
| v0 | 1st component of triplets, written to base[align*offset[i]]. | |
| v1 | 2nd component of triplets, written to base[align*offset[i] + 1]. | |
| v2 | 3rd component of triplets, written to base[align*offset[i] + 2]. |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory store operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Transpose and store 3 SIMD floats to 3 consecutive addresses at GMX_SIMD_FLOAT_WIDTH offsets.
| align | Alignment of the memory to which we write, i.e. distance (measured in elements, not bytes) between index points. When this is identical to the number of SIMD variables (i.e., 3 for this routine) the output data is packed without padding in memory. See the SIMD parameters for exactly what memory positions are written. |
| [out] | base | Pointer to the start of the memory area |
| offset | Aligned array with offsets to the start of each triplet. | |
| v0 | 1st component of triplets, written to base[align*offset[i]]. | |
| v1 | 2nd component of triplets, written to base[align*offset[i] + 1]. | |
| v2 | 3rd component of triplets, written to base[align*offset[i] + 2]. |
This function can work with both aligned (better performance) and unaligned memory. When the align parameter is not a power-of-two (align==3 would be normal for packed atomic coordinates) the memory obviously cannot be aligned, and we account for this. However, in the case where align is a power-of-two, we assume the base pointer also has the same alignment, which will enable many platforms to use faster aligned memory store operations. An easy way to think of this is that each triplet of data in memory must be aligned to the align parameter you specify when it's a power-of-two.
The offset memory must always be aligned to GMX_SIMD_FINT32_WIDTH, since this enables us to use SIMD loads and gather operations on platforms that support it.
|
inlinestatic |
Truncate SIMD4, i.e. round towards zero - common hardware instruction.
| a | Any floating-point value |
|
inlinestatic |
Truncate SIMD4, i.e. round towards zero - common hardware instruction.
| a | Any floating-point value |
|
static |
Best alignment to use for aligned pairs of double data.
The routines to load and transpose data will work with a wide range of alignments, but some might be faster than others, depending on the load instructions available in the hardware. This specifies the best alignment for each implementation when working with pairs of data.
To allow each architecture to use the most optimal form, we use a constant that code outside the SIMD module should use to store things properly. It must be at least 2. For example, a value of 2 means the two parameters A & B are stored as [A0 B0 A1 B1] while align-4 means [A0 B0 - - A1 B1 - -].
This alignment depends on the efficiency of partial-register load/store operations, and will depend on the architecture.
|
static |
Best alignment to use for aligned pairs of float data.
The routines to load and transpose data will work with a wide range of alignments, but some might be faster than others, depending on the load instructions available in the hardware. This specifies the best alignment for each implementation when working with pairs of data.
To allow each architecture to use the most optimal form, we use a constant that code outside the SIMD module should use to store things properly. It must be at least 2. For example, a value of 2 means the two parameters A & B are stored as [A0 B0 A1 B1] while align-4 means [A0 B0 - - A1 B1 - -].
This alignment depends on the efficiency of partial-register load/store operations, and will depend on the architecture.
1.8.5
