Debian Codegraph @

gcc/testsuite/gcc.target/i386/avx512fp16-vaddsh-1b.c

raw
1 /* { dg-do run { target avx512fp16 } } */
2 /* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
3
4
5 #define AVX512FP16
6 #include "avx512fp16-helper.h"
7
8 #define N_ELEMS 8
9
10 void NOINLINE
11 emulate_add_sh(V512 * dest, V512 op1, V512 op2,
12 __mmask8 k, int zero_mask)
13 {
14 V512 v1, v2, v3, v4, v5, v6, v7, v8;
15 int i;
16
17 unpack_ph_2twops(op1, &v1, &v2);
18 unpack_ph_2twops(op2, &v3, &v4);
19 unpack_ph_2twops(*dest, &v7, &v8);
20
21 if ((k&1) || !k)
22 v5.f32[0] = v1.f32[0] + v3.f32[0];
23 else if (zero_mask)
24 v5.f32[0] = 0;
25 else
26 v5.f32[0] = v7.f32[0];
27
28 for (i = 1; i < 8; i++)
29 v5.f32[i] = v1.f32[i];
30
31 *dest = pack_twops_2ph(v5, v6);
32 }
33
34 void
35 test_512 (void)
36 {
37 V512 res;
38 V512 exp;
39
40 init_src();
41
42 emulate_add_sh(&exp, src1, src2, 0x1, 0);
43 res.xmmh[0] = _mm_add_sh(src1.xmmh[0], src2.xmmh[0]);
44 check_results(&res, &exp, N_ELEMS, "_mm_add_sh");
45
46 //DEST.fp16[0] := SRC1.fp16[0] + SRC2.fp16[0]
47 emulate_add_sh(&exp, src1, src2, 0x1, 0);
48 res.xmmh[0] = _mm_mask_add_sh(res.xmmh[0], 0x1,
49 src1.xmmh[0], src2.xmmh[0]);
50 check_results(&res, &exp, N_ELEMS, "_mm_mask_add_sh");
51
52 //dest.fp16[0] remains unchanged
53 init_dest(&res, &exp);
54 emulate_add_sh(&exp, src1, src2, 0x2, 0);
55 res.xmmh[0] = _mm_mask_add_sh(res.xmmh[0], 0x2,
56 src1.xmmh[0], src2.xmmh[0]);
57 check_results(&res, &exp, N_ELEMS, "_mm_mask_add_sh");
58
59 //dest.fp16[0] = 0
60 emulate_add_sh(&exp, src1, src2, 0x2, 1);
61 res.xmmh[0] = _mm_maskz_add_sh(0x2, src1.xmmh[0], src2.xmmh[0]);
62 check_results(&res, &exp, N_ELEMS, "_mm_maskz_add_sh");
63
64 //DEST.fp16[0] := SRC1.fp16[0] + SRC2.fp16[0]
65 emulate_add_sh(&exp, src1, src2, 0x3, 1);
66 res.xmmh[0] = _mm_maskz_add_sh(0x3, src1.xmmh[0], src2.xmmh[0]);
67 check_results(&res, &exp, N_ELEMS, "_mm_maskz_add_sh");
68
69 //DEST.fp16[0] := SRC1.fp16[0] + SRC2.fp16[0]
70 emulate_add_sh(&exp, src1, src2, 0x1, 0);
71 res.xmmh[0] = _mm_add_round_sh(src1.xmmh[0],
72 src2.xmmh[0], _ROUND_NINT);
73 check_results(&res, &exp, N_ELEMS, "_mm_add_round_sh");
74
75 //DEST.fp16[0] := SRC1.fp16[0] + SRC2.fp16[0]
76 emulate_add_sh(&exp, src1, src2, 0x1, 0);
77 res.xmmh[0] = _mm_mask_add_round_sh(res.xmmh[0], 0x1, src1.xmmh[0],
78 src2.xmmh[0], _ROUND_NINT);
79 check_results(&res, &exp, N_ELEMS, "_mm_mask_add_round_sh");
80
81 //dest.fp16[0] remains unchanged
82 init_dest(&res, &exp);
83 emulate_add_sh(&exp, src1, src2, 0x2, 0);
84 res.xmmh[0] = _mm_mask_add_round_sh(res.xmmh[0], 0x2, src1.xmmh[0],
85 src2.xmmh[0], _ROUND_NINT);
86 check_results(&res, &exp, N_ELEMS, "_mm_mask_add_round_sh");
87
88 //dest.fp16[0] = 0
89 emulate_add_sh(&exp, src1, src2, 0x2, 1);
90 res.xmmh[0] = _mm_maskz_add_round_sh(0x2, src1.xmmh[0],
91 src2.xmmh[0], _ROUND_NINT);
92 check_results(&res, &exp, N_ELEMS, "_mm_maskz_add_round_sh");
93
94 //DEST.fp16[0] := SRC1.fp16[0] + SRC2.fp16[0]
95 emulate_add_sh(&exp, src1, src2, 0x3, 1);
96 res.xmmh[0] = _mm_maskz_add_round_sh(0x3, src1.xmmh[0],
97 src2.xmmh[0], _ROUND_NINT);
98 check_results(&res, &exp, N_ELEMS, "_mm_maskz_add_round_sh");
99
100 if (n_errs != 0) {
101 abort ();
102 }
103 }
104