Mercurial > libpostproc.hg

comparison postprocess_altivec_template.c @ 100:b944f0b99b23 libpostproc

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Remove declarations after statements from vertClassify_altivec
author lu_zero
date Sun, 23 Mar 2008 15:33:24 +0000
parents fa0ecbc87f51
children db57626d7d76
comparison
equal deleted inserted replaced
99:fa0ecbc87f51 100:b944f0b99b23
82 const vector signed short mask = vec_splat_s16(1); 82 const vector signed short mask = vec_splat_s16(1);
83 vector signed int v_numEq = vec_splat_s32(0); 83 vector signed int v_numEq = vec_splat_s32(0);
84 vector signed short v_data = vec_ld(0, data); 84 vector signed short v_data = vec_ld(0, data);
85 vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3, 85 vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3,
86 v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7; 86 v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7;
87 //FIXME avoid this mess if possible
88 register int j0 = 0,
89 j1 = stride,
90 j2 = 2 * stride,
91 j3 = 3 * stride,
92 j4 = 4 * stride,
93 j5 = 5 * stride,
94 j6 = 6 * stride,
95 j7 = 7 * stride;
96 vector unsigned char v_srcA0, v_srcA1, v_srcA2, v_srcA3,
97 v_srcA4, v_srcA5, v_srcA6, v_srcA7;
87 98
88 v_dcOffset = vec_splat(v_data, 0); 99 v_dcOffset = vec_splat(v_data, 0);
89 v_dcThreshold = (vector unsigned short)vec_splat(v_data, 1); 100 v_dcThreshold = (vector unsigned short)vec_splat(v_data, 1);
90 v2QP = vec_splat(v_data, 2); 101 v2QP = vec_splat(v_data, 2);
91 v4QP = (vector unsigned short)vec_splat(v_data, 3); 102 v4QP = (vector unsigned short)vec_splat(v_data, 3);
92 103
93 src2 += stride * 4; 104 src2 += stride * 4;
94 105
95 106
96 #define LOAD_LINE(i) \ 107 #define LOAD_LINE(i) \
97 register int j##i = i * stride; \ 108 { \
98 vector unsigned char perm##i = vec_lvsl(j##i, src2); \ 109 vector unsigned char perm##i = vec_lvsl(j##i, src2); \
99 const vector unsigned char v_srcA1##i = vec_ld(j##i, src2); \
100 vector unsigned char v_srcA2##i; \ 110 vector unsigned char v_srcA2##i; \
111 vector unsigned char v_srcA1##i = vec_ld(j##i, src2); \
101 if (two_vectors) \ 112 if (two_vectors) \
102 v_srcA2##i = vec_ld(j##i + 16, src2); \ 113 v_srcA2##i = vec_ld(j##i + 16, src2); \
103 const vector unsigned char v_srcA##i = \ 114 v_srcA##i = \
104 vec_perm(v_srcA1##i, v_srcA2##i, perm##i); \ 115 vec_perm(v_srcA1##i, v_srcA2##i, perm##i); \
105 v_srcAss##i = \ 116 v_srcAss##i = \
106 (vector signed short)vec_mergeh((vector signed char)zero, \ 117 (vector signed short)vec_mergeh((vector signed char)zero, \
107 (vector signed char)v_srcA##i) 118 (vector signed char)v_srcA##i); }
108 119
109 #define LOAD_LINE_ALIGNED(i) \ 120 #define LOAD_LINE_ALIGNED(i) \
110 register int j##i = i * stride; \ 121 v_srcA##i = vec_ld(j##i, src2); \
111 const vector unsigned char v_srcA##i = vec_ld(j##i, src2); \
112 v_srcAss##i = \ 122 v_srcAss##i = \
113 (vector signed short)vec_mergeh((vector signed char)zero, \ 123 (vector signed short)vec_mergeh((vector signed char)zero, \
114 (vector signed char)v_srcA##i) 124 (vector signed char)v_srcA##i)
115 125
116 /* Special-casing the aligned case is worthwhile, as all calls from 126 /* Special-casing the aligned case is worthwhile, as all calls from
144 const vector signed short v_sum##i = \ 154 const vector signed short v_sum##i = \
145 vec_add(v_diff##i, v_dcOffset); \ 155 vec_add(v_diff##i, v_dcOffset); \
146 const vector signed short v_comp##i = \ 156 const vector signed short v_comp##i = \
147 (vector signed short)vec_cmplt((vector unsigned short)v_sum##i, \ 157 (vector signed short)vec_cmplt((vector unsigned short)v_sum##i, \
148 v_dcThreshold); \ 158 v_dcThreshold); \
149 const vector signed short v_part##i = vec_and(mask, v_comp##i); \ 159 const vector signed short v_part##i = vec_and(mask, v_comp##i);
150 v_numEq = vec_sum4s(v_part##i, v_numEq); 160
151 161 {
152 ITER(0, 1); 162 ITER(0, 1)
153 ITER(1, 2); 163 ITER(1, 2)
154 ITER(2, 3); 164 ITER(2, 3)
155 ITER(3, 4); 165 ITER(3, 4)
156 ITER(4, 5); 166 ITER(4, 5)
157 ITER(5, 6); 167 ITER(5, 6)
158 ITER(6, 7); 168 ITER(6, 7)
169
170 v_numEq = vec_sum4s(v_part0, v_numEq);
171 v_numEq = vec_sum4s(v_part1, v_numEq);
172 v_numEq = vec_sum4s(v_part2, v_numEq);
173 v_numEq = vec_sum4s(v_part3, v_numEq);
174 v_numEq = vec_sum4s(v_part4, v_numEq);
175 v_numEq = vec_sum4s(v_part5, v_numEq);
176 v_numEq = vec_sum4s(v_part6, v_numEq);
177 }
178
159 #undef ITER 179 #undef ITER
160 180
161 v_numEq = vec_sums(v_numEq, zero); 181 v_numEq = vec_sums(v_numEq, zero);
162 182
163 v_numEq = vec_splat(v_numEq, 3); 183 v_numEq = vec_splat(v_numEq, 3);