File jm_simd.h¶

FileList > doppler > jm_simd.h

Public Types¶

Type	Name
typedef float	JM_VEC_F32
typedef double	JM_VEC_F64

Public Static Functions¶

Type	Name
float	jm_dot_f32 (const float *JM_RESTRICT a, const float *JM_RESTRICT b, int n)
double	jm_dot_f64 (const double *JM_RESTRICT a, const double *JM_RESTRICT b, int n)

Macros¶

Type	Name
define	JM_ADD_F32 (a, b) `((a) + (b))`
define	JM_ADD_F64 (a, b) `((a) + (b))`
define	JM_FMA_F32 (acc, a, b) `((acc) += (a) \* (b))`
define	JM_FMA_F64 (acc, a, b) `((acc) += (a) \* (b))`
define	JM_HSUM_F32 (v) `(v)`
define	JM_HSUM_F64 (v) `(v)`
define	JM_LOAD_F32 (p) `(\*(p))`
define	JM_LOAD_F64 (p) `(\*(p))`
define	JM_MAC_F32 (acc, ptr, s) `((acc) += (\(ptr)) \ (s))`
define	JM_MAC_F64 (acc, ptr, s) `((acc) += (\(ptr)) \ (s))`
define	JM_MUL_F32 (a, b) `((a) \* (b))`
define	JM_MUL_F64 (a, b) `((a) \* (b))`
define	JM_RESTRICT `restrict`
define	JM_SIMD_WIDTH `1`
define	JM_SIMD_WIDTH_F32 `1`
define	JM_SIMD_WIDTH_F64 `1`
define	JM_SPLAT_F32 (x) `(x)`
define	JM_SPLAT_F64 (x) `(x)`
define	JM_STORE_F32 (p, v) `(\*(p) = (v))`
define	JM_STORE_F64 (p, v) `(\*(p) = (v))`
define	JM_ZERO_F32 () `0.0f`
define	JM_ZERO_F64 () `0.0`

Public Types Documentation¶

typedef JM_VEC_F32¶

typedef float JM_VEC_F32;

typedef JM_VEC_F64¶

typedef double JM_VEC_F64;

Public Static Functions Documentation¶

function jm_dot_f32¶

static inline float jm_dot_f32 (
    const float * JM_RESTRICT a,
    const float * JM_RESTRICT b,
    int n
)

function jm_dot_f64¶

static inline double jm_dot_f64 (
    const double * JM_RESTRICT a,
    const double * JM_RESTRICT b,
    int n
)

Macro Definition Documentation¶

define JM_ADD_F32¶

#define JM_ADD_F32 (
    a,
    b
) `((a) + (b))`

define JM_ADD_F64¶

#define JM_ADD_F64 (
    a,
    b
) `((a) + (b))`

define JM_FMA_F32¶

#define JM_FMA_F32 (
    acc,
    a,
    b
) `((acc) += (a) * (b))`

define JM_FMA_F64¶

#define JM_FMA_F64 (
    acc,
    a,
    b
) `((acc) += (a) * (b))`

define JM_HSUM_F32¶

#define JM_HSUM_F32 (
    v
) `(v)`

define JM_HSUM_F64¶

#define JM_HSUM_F64 (
    v
) `(v)`

define JM_LOAD_F32¶

#define JM_LOAD_F32 (
    p
) `(*(p))`

define JM_LOAD_F64¶

#define JM_LOAD_F64 (
    p
) `(*(p))`

define JM_MAC_F32¶

#define JM_MAC_F32 (
    acc,
    ptr,
    s
) `((acc) += (*(ptr)) * (s))`

define JM_MAC_F64¶

#define JM_MAC_F64 (
    acc,
    ptr,
    s
) `((acc) += (*(ptr)) * (s))`

define JM_MUL_F32¶

#define JM_MUL_F32 (
    a,
    b
) `((a) * (b))`

define JM_MUL_F64¶

#define JM_MUL_F64 (
    a,
    b
) `((a) * (b))`

define JM_RESTRICT¶

#define JM_RESTRICT `restrict`

jm_simd.h — width-portable SIMD operation macros.

Selects the widest available instruction set at compile time: AVX-512F -> 16 float / 8 double lanes (JM_SIMD_WIDTH_F32 = 16) AVX2+FMA -> 8 float / 4 double lanes (JM_SIMD_WIDTH_F32 = 8) NEON -> 4 float / 2 double lanes (JM_SIMD_WIDTH_F32 = 4) (aarch64) Scalar -> 1 lane (auto-vectorisation still applies)

Typical usage (FIR inner loop):

JM_VEC_F32 acc = JM_ZERO_F32();
for (int k = 0; k < N_TAPS; k++)
    JM_MAC_F32(acc, window + k, coeffs[k]);
*out = JM_HSUM_F32(acc);

JM_SIMD_WIDTH_F32 tells you how many floats the loop above advances per iteration — stride your outer loop accordingly.

Can be included standalone; reuses JM_RESTRICT from jm_perf.h if already defined, otherwise provides its own fallback.

define JM_SIMD_WIDTH¶

#define JM_SIMD_WIDTH `1`

define JM_SIMD_WIDTH_F32¶

#define JM_SIMD_WIDTH_F32 `1`

define JM_SIMD_WIDTH_F64¶

#define JM_SIMD_WIDTH_F64 `1`

define JM_SPLAT_F32¶

#define JM_SPLAT_F32 (
    x
) `(x)`

define JM_SPLAT_F64¶

#define JM_SPLAT_F64 (
    x
) `(x)`

define JM_STORE_F32¶

#define JM_STORE_F32 (
    p,
    v
) `(*(p) = (v))`

define JM_STORE_F64¶

#define JM_STORE_F64 (
    p,
    v
) `(*(p) = (v))`

define JM_ZERO_F32¶

#define JM_ZERO_F32 (

) `0.0f`

define JM_ZERO_F64¶

#define JM_ZERO_F64 (

) `0.0`

The documentation for this class was generated from the following file native/inc/doppler/jm_simd.h