void test_vceqQf32 (void)
{
uint32x4_t out_uint32x4_t;
float32x4_t arg0_float32x4_t;
float32x4_t arg1_float32x4_t;
out_uint32x4_t = vceqq_f32 (arg0_float32x4_t, arg1_float32x4_t);
}
void dumb() {
V4SF x = {{ 0.0903333798051, 0.0903333798051, 0.0903333798051, 0.0903333798051 }};
V4SF w; w.v = log_ps(x.v);
float z = cephes_logf(x.f[0]);
printf("log_ps returned "); print4(w.v);
printf("\ncephes returned: %14.12g and logf(%g)=%14.12g\n", z, x.f[0], logf(x.f[0]));
print4i(vceqq_f32(x.v, x.v)); printf("\n");
exit(1);
}
static inline float32x4_t floor_neon(float32x4_t a)
{
#if __ARM_ARCH >= 8
return vrndqm_f32(a);
#else
const float32x4_t round32 = vdupq_n_f32(12582912.0f);
const float32x4_t vhalf = vdupq_n_f32(0.5f);
float32x4_t rounded = vsubq_f32(vaddq_f32(a, round32), round32);
uint32x4_t mask = vceqq_f32(a, rounded);
float32x4_t floored = vsubq_f32(vaddq_f32(vsubq_f32(a, vhalf), round32), round32);
return vreinterpretq_f32_u32(vorrq_u32(vandq_u32(vreinterpretq_u32_f32(a), mask),
vbicq_u32(vreinterpretq_u32_f32(floored), mask)));
#endif
}
//Kernel function: saxpy
void saxpy_vector(KernelArgs* args) {
//Setup
const float32x4_t MASK_FALSE = vdupq_n_f32(0.f);
const float32x4_t MASK_TRUE = vcvtq_f32_u32(vceqq_f32(MASK_FALSE, MASK_FALSE));
//Uniforms
//Fuses
//Literals
//Stack variables
float32x4_t scale, x, y, result, var060, var061;
//Loop over input
uint64_t index;
for(index = 0; index < args->N; index += 4) {
//Inputs
scale = vld1q_f32(&args->scale[index]);
x = vld1q_f32(&args->x[index]);
y = vld1q_f32(&args->y[index]);
//Begin kernel logic
{
//>>> result = scale * x + y
var061 = vmulq_f32(scale, x);
var060 = vaddq_f32(var061, y);
result = vbslq_f32(vcvtq_u32_f32(MASK_TRUE), var060, result);
}
//End kernel logic
//Outputs
vst1q_f32(&args->result[index], result);
}
}
inline uint32x4_t vceqq(const float32x4_t & v0, const float32x4_t & v1) { return vceqq_f32(v0, v1); }
