PIPS
Macros | Typedefs
neon.h File Reference
#include <arm_neon.h>
+ Include dependency graph for neon.h:

Go to the source code of this file.

Macros

#define SIMD_LOAD_V4SF(vec, arr)   vec=vld1q_f32(arr)
 float More...
 
#define SIMD_LOADA_V4SF(vec, arr)   vec=vld1q_f32(arr)
 
#define SIMD_MULPS(vec1, vec2, vec3)   vec1=vmulq_f32(vec2,vec3)
 
#define SIMD_DIVPS(vec1, vec2, vec3)
 
#define SIMD_ADDPS(vec1, vec2, vec3)   vec1=vaddq_f32(vec2,vec3)
 
#define SIMD_SUBPS(vec1, vec2, vec3)   vec1=vsubq_f32(vec2, vec3)
 
#define SIMD_MULADDPS(vec1, vec2, vec3, vec4)   vec1=vmlaq_f32(vec2,vec3,vec4)
 
#define SIMD_UMINPS(vec1, vec2)   vec1=vnegq_f32(vec2)
 
#define SIMD_STORE_V4SF(vec, arr)   vst1q_f32(arr,vec)
 
#define SIMD_STOREA_V4SF(vec, arr)   vst1q_f32(arr,vec)
 
#define SIMD_STORE_GENERIC_V4SF(vec, v0, v1, v2, v3)
 
#define SIMD_ZERO_V4SF(vec)   vec = vsubq_f32(vec,vec)
 
#define SIMD_LOAD_GENERIC_V4SF(vec, v0, v1, v2, v3)
 
#define SIMD_STORE_MASKED_V4SF(vec, arr)
 handle padded value, this is a very bad implementation ... More...
 
#define SIMD_LOAD_V4SI_TO_V4SF(v, f)
 
#define SIMD_LOAD_V2DI(vec, arr)   vec=vld1q_s64(arr)
 int64_t More...
 
#define SIMD_STORE_V2DI(vec, arr)   vst1q_s64(arr,vec)
 
#define SIMD_ZERO_V2DI(vec)   vec = veorq_s64(vec,vec)
 
#define SIMD_ADDDI(v1, v2, v3)   v1=vaddq_s64(v2,v3)
 
#define SIMD_SUBDI(v1, v2, v3)   v1=vsubq_s64(v2,v3)
 
#define SIMD_DIVDI(vec1, vec2, vec3)
 
#define SIMD_MULDI(v1, v2, v3)   v1=vmulq_s64(v2,v3)
 
#define SIMD_MULADDDI(vec1, vec2, vec3, vec4)   vec1=vmlaq_s64(vec2,vec3,vec4)
 
#define SIMD_LOAD_V4SI(vec, arr)   vec=vld1q_s32(arr)
 int32_t More...
 
#define SIMD_STORE_V4SI(vec, arr)   vst1q_s32(arr,vec)
 
#define SIMD_ZERO_V4SI(vec)   vec = veorq_s32(vec,vec)
 
#define SIMD_ADDD(v1, v2, v3)   v1=vaddq_s32(v2,v3)
 
#define SIMD_SUBD(v1, v2, v3)   v1=vsubq_s32(v2,v3)
 
#define SIMD_DIVD(vec1, vec2, vec3)
 
#define SIMD_MULD(v1, v2, v3)   v1=vmulq_s32(v2,v3)
 
#define SIMD_MULADDD(vec1, vec2, vec3, vec4)   vec1=vmlaq_s32(vec2,vec3,vec4)
 
#define SIMD_LOAD_V8HI(vec, arr)   vec=vld1q_s16(arr)
 int16_t More...
 
#define SIMD_STORE_V8HI(vec, arr)   vst1q_s16(arr,vec)
 
#define SIMD_ZERO_V8HI(vec)   vec = veorq_s16(vec,vec)
 
#define SIMD_ADDHI(v1, v2, v3)   v1=vaddq_s16(v2,v3)
 
#define SIMD_SUBHI(v1, v2, v3)   v1=vsubq_s16(v2,v3)
 
#define SIMD_DIVHI(vec1, vec2, vec3)
 
#define SIMD_MULHI(v1, v2, v3)   v1=vmulq_s16(v2,v3)
 
#define SIMD_STORE_V8HI_TO_V8SI(vec, arr)    SIMD_STORE_V8HI(vec,arr)
 
#define SIMD_LOAD_V8SI_TO_V8HI(vec, arr)    SIMD_LOAD_V8HI(vec,arr)
 
#define SIMD_MULADDHI(vec1, vec2, vec3, vec4)   vec1=vmlaq_s16(vec2,vec3,vec4)
 
#define SIMD_LOAD_V16QI(vec, arr)   vec=vld1q_s8(arr)
 int8_t More...
 
#define SIMD_STORE_V16QI(vec, arr)   vst1q_s8(arr,vec)
 
#define SIMD_ZERO_V16QI(vec)   vec = veorq_s8(vec,vec)
 
#define SIMD_ADDQI(v1, v2, v3)   v1=vaddq_s8(v2,v3)
 
#define SIMD_SUBQI(v1, v2, v3)   v1=vsubq_s8(v2,v3)
 
#define SIMD_DIVQI(vec1, vec2, vec3)
 
#define SIMD_MULQI(v1, v2, v3)   v1=vmulq_s8(v2,v3)
 
#define SIMD_MULADDQI(vec1, vec2, vec3, vec4)   vec1=vmlaq_s8(vec2,vec3,vec4)
 

Typedefs

typedef float32_t a4sf[8] __attribute__((aligned(32)))
 Uses 128-bits NEON instructions. More...
 
typedef float32x4_t v4sf
 
typedef int64x2_t v2di
 
typedef int32x4_t v4si
 
typedef int16x8_t v8hi
 
typedef int8x16_t v16qi
 

Macro Definition Documentation

◆ SIMD_ADDD

#define SIMD_ADDD (   v1,
  v2,
  v3 
)    v1=vaddq_s32(v2,v3)

Definition at line 116 of file neon.h.

◆ SIMD_ADDDI

#define SIMD_ADDDI (   v1,
  v2,
  v3 
)    v1=vaddq_s64(v2,v3)

Definition at line 99 of file neon.h.

◆ SIMD_ADDHI

#define SIMD_ADDHI (   v1,
  v2,
  v3 
)    v1=vaddq_s16(v2,v3)

Definition at line 133 of file neon.h.

◆ SIMD_ADDPS

#define SIMD_ADDPS (   vec1,
  vec2,
  vec3 
)    vec1=vaddq_f32(vec2,vec3)

Definition at line 45 of file neon.h.

◆ SIMD_ADDQI

#define SIMD_ADDQI (   v1,
  v2,
  v3 
)    v1=vaddq_s8(v2,v3)

Definition at line 156 of file neon.h.

◆ SIMD_DIVD

#define SIMD_DIVD (   vec1,
  vec2,
  vec3 
)
Value:
do {\
vec3=vrecpeq_s32(vec3);\
vec1=vmulq_s32(vec2,vec3);\
}\
while (0)

Definition at line 118 of file neon.h.

◆ SIMD_DIVDI

#define SIMD_DIVDI (   vec1,
  vec2,
  vec3 
)
Value:
do {\
vec3=vrecpeq_s64(vec3);\
vec1=vmulq_s64(vec2,vec3);\
}\
while (0)

Definition at line 101 of file neon.h.

◆ SIMD_DIVHI

#define SIMD_DIVHI (   vec1,
  vec2,
  vec3 
)
Value:
do {\
vec3=vrecpeq_s16(vec3);\
vec1=vmulq_s16(vec2,vec3);\
}\
while (0)

Definition at line 135 of file neon.h.

◆ SIMD_DIVPS

#define SIMD_DIVPS (   vec1,
  vec2,
  vec3 
)
Value:
do {\
vec3=vrecpeq_f32(vec3);\
vec1=vmulq_f32(vec2,vec3);\
}\
while (0)

Definition at line 38 of file neon.h.

◆ SIMD_DIVQI

#define SIMD_DIVQI (   vec1,
  vec2,
  vec3 
)
Value:
do {\
vec3=vrecpeq_s8(vec3);\
vec1=vmulq_s8(vec2,vec3);\
}\
while (0)

Definition at line 158 of file neon.h.

◆ SIMD_LOAD_GENERIC_V4SF

#define SIMD_LOAD_GENERIC_V4SF (   vec,
  v0,
  v1,
  v2,
  v3 
)
Value:
do { \
float __pips_v[4] __attribute ((aligned (16)));\
__pips_v[0]=v0;\
__pips_v[1]=v1;\
__pips_v[2]=v2;\
__pips_v[3]=v3;\
SIMD_LOADA_V4SF(vec,&__pips_v[0]); \
} while(0)

Definition at line 63 of file neon.h.

◆ SIMD_LOAD_V16QI

#define SIMD_LOAD_V16QI (   vec,
  arr 
)    vec=vld1q_s8(arr)

int8_t

Definition at line 151 of file neon.h.

◆ SIMD_LOAD_V2DI

#define SIMD_LOAD_V2DI (   vec,
  arr 
)    vec=vld1q_s64(arr)

int64_t

Definition at line 94 of file neon.h.

◆ SIMD_LOAD_V4SF

#define SIMD_LOAD_V4SF (   vec,
  arr 
)    vec=vld1q_f32(arr)

float

Definition at line 35 of file neon.h.

◆ SIMD_LOAD_V4SI

#define SIMD_LOAD_V4SI (   vec,
  arr 
)    vec=vld1q_s32(arr)

int32_t

Definition at line 111 of file neon.h.

◆ SIMD_LOAD_V4SI_TO_V4SF

#define SIMD_LOAD_V4SI_TO_V4SF (   v,
  f 
)
Value:
do { \
float __pips_tmp[4]; \
__pips_tmp[0] = (f)[0]; \
__pips_tmp[1] = (f)[1]; \
__pips_tmp[2] = (f)[2]; \
__pips_tmp[3] = (f)[3]; \
SIMD_LOAD_V4SF(v, __pips_tmp); \
} while(0)
f
int f(int off1, int off2, int n, float r[n], float a[n], float b[n])
Definition: offsets.c:15

Definition at line 83 of file neon.h.

◆ SIMD_LOAD_V8HI

#define SIMD_LOAD_V8HI (   vec,
  arr 
)    vec=vld1q_s16(arr)

int16_t

Definition at line 128 of file neon.h.

◆ SIMD_LOAD_V8SI_TO_V8HI

#define SIMD_LOAD_V8SI_TO_V8HI (   vec,
  arr 
)     SIMD_LOAD_V8HI(vec,arr)

Definition at line 145 of file neon.h.

◆ SIMD_LOADA_V4SF

#define SIMD_LOADA_V4SF (   vec,
  arr 
)    vec=vld1q_f32(arr)

Definition at line 36 of file neon.h.

◆ SIMD_MULADDD

#define SIMD_MULADDD (   vec1,
  vec2,
  vec3,
  vec4 
)    vec1=vmlaq_s32(vec2,vec3,vec4)

Definition at line 125 of file neon.h.

◆ SIMD_MULADDDI

#define SIMD_MULADDDI (   vec1,
  vec2,
  vec3,
  vec4 
)    vec1=vmlaq_s64(vec2,vec3,vec4)

Definition at line 108 of file neon.h.

◆ SIMD_MULADDHI

#define SIMD_MULADDHI (   vec1,
  vec2,
  vec3,
  vec4 
)    vec1=vmlaq_s16(vec2,vec3,vec4)

Definition at line 148 of file neon.h.

◆ SIMD_MULADDPS

#define SIMD_MULADDPS (   vec1,
  vec2,
  vec3,
  vec4 
)    vec1=vmlaq_f32(vec2,vec3,vec4)

Definition at line 47 of file neon.h.

◆ SIMD_MULADDQI

#define SIMD_MULADDQI (   vec1,
  vec2,
  vec3,
  vec4 
)    vec1=vmlaq_s8(vec2,vec3,vec4)

Definition at line 166 of file neon.h.

◆ SIMD_MULD

#define SIMD_MULD (   v1,
  v2,
  v3 
)    v1=vmulq_s32(v2,v3)

Definition at line 124 of file neon.h.

◆ SIMD_MULDI

#define SIMD_MULDI (   v1,
  v2,
  v3 
)    v1=vmulq_s64(v2,v3)

Definition at line 107 of file neon.h.

◆ SIMD_MULHI

#define SIMD_MULHI (   v1,
  v2,
  v3 
)    v1=vmulq_s16(v2,v3)

Definition at line 141 of file neon.h.

◆ SIMD_MULPS

#define SIMD_MULPS (   vec1,
  vec2,
  vec3 
)    vec1=vmulq_f32(vec2,vec3)

Definition at line 37 of file neon.h.

◆ SIMD_MULQI

#define SIMD_MULQI (   v1,
  v2,
  v3 
)    v1=vmulq_s8(v2,v3)

Definition at line 164 of file neon.h.

◆ SIMD_STORE_GENERIC_V4SF

#define SIMD_STORE_GENERIC_V4SF (   vec,
  v0,
  v1,
  v2,
  v3 
)
Value:
do { \
float __pips_tmp[4] __attribute__ ((aligned (16))); \
SIMD_STOREA_V4SF(vec,&__pips_tmp[0]); \
*(v0)=__pips_tmp[0]; \
*(v1)=__pips_tmp[1]; \
*(v2)=__pips_tmp[2]; \
*(v3)=__pips_tmp[3]; \
} while (0)
__attribute__
float32_t a4sf[8] __attribute__((aligned(32)))
Uses 128-bits NEON instructions.
Definition: neon.h:22

Definition at line 51 of file neon.h.

◆ SIMD_STORE_MASKED_V4SF

#define SIMD_STORE_MASKED_V4SF (   vec,
  arr 
)
Value:
do { \
float __pips_tmp[4] __attribute__ ((aligned (16))); \
SIMD_STOREA_V4SF(vec,&__pips_tmp[0]); \
(arr)[0] = __pips_tmp[0]; \
(arr)[1] = __pips_tmp[1]; \
(arr)[2] = __pips_tmp[2]; \
} while(0)

handle padded value, this is a very bad implementation ...

Definition at line 74 of file neon.h.

◆ SIMD_STORE_V16QI

#define SIMD_STORE_V16QI (   vec,
  arr 
)    vst1q_s8(arr,vec)

Definition at line 152 of file neon.h.

◆ SIMD_STORE_V2DI

#define SIMD_STORE_V2DI (   vec,
  arr 
)    vst1q_s64(arr,vec)

Definition at line 95 of file neon.h.

◆ SIMD_STORE_V4SF

#define SIMD_STORE_V4SF (   vec,
  arr 
)    vst1q_f32(arr,vec)

Definition at line 49 of file neon.h.

◆ SIMD_STORE_V4SI

#define SIMD_STORE_V4SI (   vec,
  arr 
)    vst1q_s32(arr,vec)

Definition at line 112 of file neon.h.

◆ SIMD_STORE_V8HI

#define SIMD_STORE_V8HI (   vec,
  arr 
)    vst1q_s16(arr,vec)

Definition at line 129 of file neon.h.

◆ SIMD_STORE_V8HI_TO_V8SI

#define SIMD_STORE_V8HI_TO_V8SI (   vec,
  arr 
)     SIMD_STORE_V8HI(vec,arr)

Definition at line 143 of file neon.h.

◆ SIMD_STOREA_V4SF

#define SIMD_STOREA_V4SF (   vec,
  arr 
)    vst1q_f32(arr,vec)

Definition at line 50 of file neon.h.

◆ SIMD_SUBD

#define SIMD_SUBD (   v1,
  v2,
  v3 
)    v1=vsubq_s32(v2,v3)

Definition at line 117 of file neon.h.

◆ SIMD_SUBDI

#define SIMD_SUBDI (   v1,
  v2,
  v3 
)    v1=vsubq_s64(v2,v3)

Definition at line 100 of file neon.h.

◆ SIMD_SUBHI

#define SIMD_SUBHI (   v1,
  v2,
  v3 
)    v1=vsubq_s16(v2,v3)

Definition at line 134 of file neon.h.

◆ SIMD_SUBPS

#define SIMD_SUBPS (   vec1,
  vec2,
  vec3 
)    vec1=vsubq_f32(vec2, vec3)

Definition at line 46 of file neon.h.

◆ SIMD_SUBQI

#define SIMD_SUBQI (   v1,
  v2,
  v3 
)    v1=vsubq_s8(v2,v3)

Definition at line 157 of file neon.h.

◆ SIMD_UMINPS

#define SIMD_UMINPS (   vec1,
  vec2 
)    vec1=vnegq_f32(vec2)

Definition at line 48 of file neon.h.

◆ SIMD_ZERO_V16QI

#define SIMD_ZERO_V16QI (   vec)    vec = veorq_s8(vec,vec)

Definition at line 154 of file neon.h.

◆ SIMD_ZERO_V2DI

#define SIMD_ZERO_V2DI (   vec)    vec = veorq_s64(vec,vec)

Definition at line 97 of file neon.h.

◆ SIMD_ZERO_V4SF

#define SIMD_ZERO_V4SF (   vec)    vec = vsubq_f32(vec,vec)

Definition at line 61 of file neon.h.

◆ SIMD_ZERO_V4SI

#define SIMD_ZERO_V4SI (   vec)    vec = veorq_s32(vec,vec)

Definition at line 114 of file neon.h.

◆ SIMD_ZERO_V8HI

#define SIMD_ZERO_V8HI (   vec)    vec = veorq_s16(vec,vec)

Definition at line 131 of file neon.h.

Typedef Documentation

◆ __attribute__

typedef int8_t a16qi [32] __attribute__((aligned(32)))

Uses 128-bits NEON instructions.

Notes : NEON can also operate on 64-bits vectors. NEON does not operate on double-precision float. However, VFP can work on double-precision 64-bits vectors, but VFP is not a simd unit : it processes vectors scalar by scalar. reminder : NEON data types : signed/unsigned 8-bit, 16-bit, 32-bit, 64-bit, single precision floating point TODO: alignement: [1] says that each instruction has an alignement offset argument., but I can't find it in the intrinsics... An other option is to used an isntruction that set the alignement offset before each call, but it sounds like wasted cycles... TODO: a feature of NEON is to load/store up to 4 vectors with just one instruction (see vst{1,2,3,4}/vld{1,2,3,4} variants and [2] for examples). This needs (I think) a modification in SAC. [2] is a nice summary of the intrinsics used here

[1] : http://infocenter.arm.com/help/topic/com.arm.doc.dui0489b/CIHGIAEH.html [2] : http://gcc.gnu.org/onlinedocs/gcc/ARM-NEON-Intrinsics.html

Definition at line 22 of file neon.h.

◆ v16qi

typedef int8x16_t v16qi

Definition at line 32 of file neon.h.

◆ v2di

typedef int64x2_t v2di

Definition at line 29 of file neon.h.

◆ v4sf

typedef float32x4_t v4sf

Definition at line 28 of file neon.h.

◆ v4si

typedef int32x4_t v4si

Definition at line 30 of file neon.h.

◆ v8hi

typedef int16x8_t v8hi

Definition at line 31 of file neon.h.