1 /* { dg-do compile } */
2 /* { dg-options "-O2 -msse2" } */
3 /* { dg-require-effective-target sse2 } */
5 typedef float __v4sf __attribute__ ((__vector_size__ (16)));
6 typedef float __m128 __attribute__ ((__vector_size__ (16)));
7 typedef long long __v2di __attribute__ ((__vector_size__ (16)));
10 _mm_cmpeq_ps (__m128 __A, __m128 __B)
12 return (__m128) __builtin_ia32_cmpeqps ((__v4sf)__A, (__v4sf)__B);
15 static __inline __m128
16 _mm_setr_ps (float __Z, float __Y, float __X, float __W)
18 return __extension__ (__m128)(__v4sf){__Z, __Y, __X, __W };
21 static __inline __m128
22 _mm_and_si128 (__m128 __A, __m128 __B)
24 return (__m128)__builtin_ia32_pand128 ((__v2di)__A, (__v2di)__B);
27 static __inline __m128
28 _mm_or_si128 (__m128 __A, __m128 __B)
30 return (__m128)__builtin_ia32_por128 ((__v2di)__A, (__v2di)__B);
38 __attribute__ ((aligned (16))) um128;
43 sse_max_abs_indexf (float *v, int step, int n)
49 mm = __builtin_ia32_andps ((__m128) (__v4sf)
50 { 0.0, v[step], v[step2], v[step3] }
55 for (i = 0; i < n4; ++i);
56 msk = (__m128) _mm_cmpeq_ps (m1, mm);
57 mim = _mm_or_si128 (_mm_and_si128 (msk, mi), mim);
59 ui.xmmi = (__m128) mim;
64 sse_swap_rowf (float *r1, float *r2, int n)
67 float *r14end = r1 + n4;
76 ludcompf (float *m, int nw, int *prow, int n)
80 for (i = 0, pm = m; i < n - 1; ++i, pm += nw)
82 int vi = sse_max_abs_indexf (pm + i, nw, n - i);
87 sse_swap_rowf (pm, pm + vi * nw, nw);
88 swap_index (prow, i, i + vi);
90 for (j = i + 1, pt = pm + nw; j < n; ++j, pt += nw)
91 sse_add_rowf (pt + i + 1, pm + i + 1, -1.0, n - i - 1);