60 unsigned int num_points)
62 const unsigned int num_bytes = num_points * 2;
66 int bound = (num_bytes >> 4);
67 int bound_copy = bound;
68 int leftovers = (num_bytes >> 1) & 7;
70 __m128i *p_target, *p_src0, *p_src1, *p_src2, *p_src3;
71 p_target = (__m128i*)target;
72 p_src0 = (__m128i*)src0;
73 p_src1 = (__m128i*)src1;
74 p_src2 = (__m128i*)src2;
75 p_src3 = (__m128i*)src3;
77 __m128i xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8;
79 while (bound_copy > 0) {
80 xmm1 = _mm_load_si128(p_src0);
81 xmm2 = _mm_load_si128(p_src1);
82 xmm3 = _mm_load_si128(p_src2);
83 xmm4 = _mm_load_si128(p_src3);
85 xmm5 = _mm_setzero_si128();
86 xmm6 = _mm_setzero_si128();
90 xmm1 = _mm_sub_epi16(xmm2, xmm1);
92 xmm3 = _mm_sub_epi16(xmm4, xmm3);
94 xmm5 = _mm_cmpgt_epi16(xmm1, xmm5);
95 xmm6 = _mm_cmpgt_epi16(xmm3, xmm6);
97 xmm2 = _mm_and_si128(xmm5, xmm2);
98 xmm4 = _mm_and_si128(xmm6, xmm4);
99 xmm5 = _mm_andnot_si128(xmm5, xmm7);
100 xmm6 = _mm_andnot_si128(xmm6, xmm8);
102 xmm5 = _mm_add_epi16(xmm2, xmm5);
103 xmm6 = _mm_add_epi16(xmm4, xmm6);
105 xmm1 = _mm_xor_si128(xmm1, xmm1);
107 xmm5 = _mm_sub_epi16(xmm6, xmm5);
111 xmm1 = _mm_cmpgt_epi16(xmm5, xmm1);
114 xmm6 = _mm_and_si128(xmm1, xmm6);
116 xmm1 = _mm_andnot_si128(xmm1, xmm2);
119 xmm1 = _mm_add_epi16(xmm6, xmm1);
122 _mm_store_si128(p_target, xmm1);
128 for (
i = bound * 8;
i < (bound * 8) + leftovers; ++
i) {
129 temp0 = ((short)(src0[
i] - src1[
i]) > 0) ? src0[
i] : src1[
i];
130 temp1 = ((short)(src2[
i] - src3[
i]) > 0) ? src2[
i] : src3[
i];
131 target[
i] = ((short)(temp0 - temp1) > 0) ? temp0 : temp1;
147 unsigned int num_points)
149 const unsigned int eighth_points = num_points / 8;
152 int16x8_t src0_vec, src1_vec, src2_vec, src3_vec;
153 int16x8_t diff12, diff34;
154 int16x8_t comp0, comp1, comp2, comp3;
155 int16x8_t result1_vec, result2_vec;
157 zeros = vdupq_n_s16(0);
158 for (
i = 0;
i < eighth_points; ++
i) {
159 src0_vec = vld1q_s16(src0);
160 src1_vec = vld1q_s16(src1);
161 src2_vec = vld1q_s16(src2);
162 src3_vec = vld1q_s16(src3);
163 diff12 = vsubq_s16(src0_vec, src1_vec);
164 diff34 = vsubq_s16(src2_vec, src3_vec);
165 comp0 = (int16x8_t)vcgeq_s16(diff12, zeros);
166 comp1 = (int16x8_t)vcltq_s16(diff12, zeros);
167 comp2 = (int16x8_t)vcgeq_s16(diff34, zeros);
168 comp3 = (int16x8_t)vcltq_s16(diff34, zeros);
169 comp0 = vandq_s16(src0_vec, comp0);
170 comp1 = vandq_s16(src1_vec, comp1);
171 comp2 = vandq_s16(src2_vec, comp2);
172 comp3 = vandq_s16(src3_vec, comp3);
174 result1_vec = vaddq_s16(comp0, comp1);
175 result2_vec = vaddq_s16(comp2, comp3);
177 diff12 = vsubq_s16(result1_vec, result2_vec);
178 comp0 = (int16x8_t)vcgeq_s16(diff12, zeros);
179 comp1 = (int16x8_t)vcltq_s16(diff12, zeros);
180 comp0 = vandq_s16(result1_vec, comp0);
181 comp1 = vandq_s16(result2_vec, comp1);
182 result1_vec = vaddq_s16(comp0, comp1);
183 vst1q_s16(target, result1_vec);
193 for (
i = eighth_points * 8;
i < num_points; ++
i) {
194 temp0 = ((short)(*src0 - *src1) > 0) ? *src0 : *src1;
195 temp1 = ((short)(*src2 - *src3) > 0) ? *src2 : *src3;
196 *target++ = ((short)(temp0 - temp1) > 0) ? temp0 : temp1;
static void volk_16i_x4_quad_max_star_16i_generic(short *target, short *src0, short *src1, short *src2, short *src3, unsigned int num_points)
Definition volk_16i_x4_quad_max_star_16i.h:207
static void volk_16i_x4_quad_max_star_16i_neon(short *target, short *src0, short *src1, short *src2, short *src3, unsigned int num_points)
Definition volk_16i_x4_quad_max_star_16i.h:142
static void volk_16i_x4_quad_max_star_16i_a_sse2(short *target, short *src0, short *src1, short *src2, short *src3, unsigned int num_points)
Definition volk_16i_x4_quad_max_star_16i.h:55