2 /* { dg-require-effective-target fma4 } */
3 /* { dg-options "-O0 -mfma4" } */
5 #include "fma4-check.h"
18 } dst, res, src1, src2, src3;
21 /* Note that in macc*,msub*,mnmacc* and mnsub* instructions, the intermdediate
22 product is not rounded, only the addition is rounded. */
28 for (i = 0; i < NUM * 4; i++)
40 for (i = 0; i < NUM * 4; i++)
51 int i, j, check_fails = 0;
52 for (i = 0; i < NUM * 4; i = i + 4)
53 for (j = 0; j < 4; j++)
55 res.f[i + j] = (src1.f[i + j] * src2.f[i + j]) + src3.f[i + j];
56 if (dst.f[i + j] != res.f[i + j])
65 int i, j, check_fails = 0;
66 for (i = 0; i < NUM * 2; i = i + 2)
67 for (j = 0; j < 2; j++)
69 res.d[i + j] = (src1.d[i + j] * src2.d[i + j]) + src3.d[i + j];
70 if (dst.d[i + j] != res.d[i + j])
80 int i, j, check_fails = 0;
81 for (i = 0; i < NUM * 4; i= i + 4)
83 res.f[i] = (src1.f[i] * src2.f[i]) + src3.f[i];
84 if (dst.f[i] != res.f[i])
93 int i, j, check_fails = 0;
94 for (i = 0; i < NUM * 2; i = i + 2)
96 res.d[i] = (src1.d[i] * src2.d[i]) + src3.d[i];
97 if (dst.d[i] != res.d[i])
100 return check_fails++;
110 for (i = 0; i < NUM; i++)
111 dst.x[i] = _mm_macc_ps (src1.x[i], src2.x[i], src3.x[i]);
116 for (i = 0; i < NUM; i++)
117 dst.x[i] = _mm_macc_ss (src1.x[i], src2.x[i], src3.x[i]);
124 for (i = 0; i < NUM; i++)
125 dst.y[i] = _mm_macc_pd (src1.y[i], src2.y[i], src3.y[i]);
130 for (i = 0; i < NUM; i++)
131 dst.y[i] = _mm_macc_sd (src1.y[i], src2.y[i], src3.y[i]);