1 // Auto-generated file. Do not edit! 2 // Template: src/f32-relu/scalar.c.in 3 // Generator: tools/xngen 4 // 5 // Copyright 2020 Google LLC 6 // 7 // This source code is licensed under the BSD-style license found in the 8 // LICENSE file in the root directory of this source tree. 9 10 #include <assert.h> 11 12 #include <xnnpack/vunary.h> 13 #include <xnnpack/common.h> 14 #include <xnnpack/math.h> 15 xnn_f32_relu_ukernel__scalar_x4(size_t n,const float * x_ptr,float * y_ptr,const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS (1)])16void xnn_f32_relu_ukernel__scalar_x4( 17 size_t n, 18 const float* x_ptr, 19 float* y_ptr, 20 const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS(1)]) 21 { 22 assert(n != 0); 23 assert(n % sizeof(float) == 0); 24 assert(x_ptr != NULL); 25 assert(y_ptr != NULL); 26 27 const uint32_t* x = (const uint32_t*)x_ptr; 28 uint32_t* y = (uint32_t*)y_ptr; 29 30 for (; n >= 4 * sizeof(uint32_t); n -= 4 * sizeof(uint32_t)) { 31 uint32_t vacc0 = x[0]; 32 uint32_t vacc1 = x[1]; 33 uint32_t vacc2 = x[2]; 34 uint32_t vacc3 = x[3]; 35 x += 4; 36 37 vacc0 = ((vacc0 >> 31) - 1) & vacc0; 38 vacc1 = ((vacc1 >> 31) - 1) & vacc1; 39 vacc2 = ((vacc2 >> 31) - 1) & vacc2; 40 vacc3 = ((vacc3 >> 31) - 1) & vacc3; 41 42 y[0] = vacc0; 43 y[1] = vacc1; 44 y[2] = vacc2; 45 y[3] = vacc3; 46 y += 4; 47 } 48 if XNN_UNLIKELY(n != 0) { 49 do { 50 uint32_t vacc = *x++; 51 vacc = ((vacc >> 31) - 1) & vacc; 52 *y++ = vacc; 53 n -= sizeof(uint32_t); 54 } while (n != 0); 55 } 56 } 57