Migrating a Short Vector Math Library (SVML)

Replaceability

Intel Short Vector Math Library (Intel SVML) is an automatic vectorization interface provided by Intel for the compiler. It converts one pair of scalar operations into parallel processing of multiple pairs of vector operations, which is an optimization of scalar operations by the compiler. The Intel SVML interface is not provided for users. KML_SVML provides 30 interfaces for the compiler. For details, see Table 1.

**Table 1** Compiler vectorization interfaces
SVML Interface	MATH Interface	Description
__svml_sin? __svml_sinf?	_ZGVnN2v_sin _ZGVnN4v_sinf	Sine function
__svml_cos? __svml_cosf?	_ZGVnN2v_cos _ZGVnN4v_cosf	Cosine function
__svml_tan? __svml_tanf?	_ZGVnN2v_tan _ZGVnN4v_tanf	Tangent function
__svml_atan? __svml_atanf?	_ZGVnN2v_atan _ZGVnN4v_atanf	Arctangent function
__svml_atan2? __svml_atan2f?	_ZGVnN2vv_atan2 _ZGVnN4vv_atan2f	Arctangent value of src1 or src2
__svml_sinh? __svml_sinhf?	_ZGVnN2v_sinh _ZGVnN4v_sinhf	Hyperbolic sine function
__svml_cosh? __svml_coshf?	_ZGVnN2v_cosh _ZGVnN4v_coshf	Hyperbolic cosine function
__svml_tanh? __svml_tanhf?	_ZGVnN2v_tanh _ZGVnN4v_tanhf	Hyperbolic tangent function
__svml_exp? __svml_expf?	_ZGVnN2v_exp _ZGVnN4v_expf	Exponential function (base e)
__svml_exp2? __svml_exp2f?	_ZGVnN2v_exp2 _ZGVnN4v_exp2f	Exponential function (base 2)
__svml_expm1? __svml_expm1f?	_ZGVnN2v_expm1 _ZGVnN4v_expm1f	Exponential minus one (base e)
__svml_log? __svml_logf?	_ZGVnN2v_log _ZGVnN4v_logf	Logarithmic function (base e)
__svml_log10? __svml_log10f?	_ZGVnN2v_log10 _ZGVnN4v_log10f	Logarithm function (base 10)
__svml_log1p? __svml_log1pf?	_ZGVnN2v_log1p _ZGVnN4v_log1pf	Logarithmic function (base e) of (1 + x)
__svml_pow? __svml_powf?	_ZGVnN2vv_pow _ZGVnN4vv_powf	Power function

Using GCC for Compilation

For details about how to enable GCC automatic vectorization, see "Kunpeng Math Library Developer Guide" > "Basic Math Libraries" > "KML_SVML Library Functions" > "Compiler Automatic Vectorization" > "GCC Vectorization" in Kunpeng HPCKit 26.1.RC1 Developer Guide.

Using the BiSheng Compiler for Compilation

For details about how to enable automatic vectorization of the BiSheng Compiler, see "Kunpeng Math Library Developer Guide" > "Basic Math Libraries" > "KML_SVML Library Functions" > "Compiler Automatic Vectorization" > "BiSheng Compiler Vectorization" in Kunpeng HPCKit 26.1.RC1 Developer Guide.