| 1 | /* |
| 2 | * SME outer product, [ 1 2 3 4 ] squared |
| 3 | * SPDX-License-Identifier: GPL-2.0-or-later |
| 4 | */ |
| 5 | |
| 6 | #include <stdio.h> |
| 7 | #include <stdint.h> |
| 8 | #include <string.h> |
| 9 | #include <math.h> |
| 10 | |
| 11 | static const float i_1234[4] = { |
| 12 | 1.0f, 2.0f, 3.0f, 4.0f |
| 13 | }; |
| 14 | |
| 15 | static const float expected[4] = { |
| 16 | 4.515625f, 5.750000f, 6.984375f, 8.218750f |
| 17 | }; |
| 18 | |
| 19 | static void test_fmopa(float *result) |
| 20 | { |
| 21 | asm(".arch_extension sme\n\t" |
| 22 | "smstart\n\t" /* ZArray cleared */ |
| 23 | "ptrue p2.b, vl16\n\t" /* Limit vector length to 16 */ |
| 24 | "ld1w {z0.s}, p2/z, [%1]\n\t" |
| 25 | "mov w15, #0\n\t" |
| 26 | "mov za3h.s[w15, 0], p2/m, z0.s\n\t" |
| 27 | "mov za3h.s[w15, 1], p2/m, z0.s\n\t" |
| 28 | "mov w15, #2\n\t" |
| 29 | "mov za3h.s[w15, 0], p2/m, z0.s\n\t" |
| 30 | "mov za3h.s[w15, 1], p2/m, z0.s\n\t" |
| 31 | "msr fpcr, xzr\n\t" |
| 32 | "fmopa za3.s, p2/m, p2/m, z0.h, z0.h\n\t" |
| 33 | "mov w15, #0\n\t" |
| 34 | "st1w {za3h.s[w15, 0]}, p2, [%0]\n" |
| 35 | "add %0, %0, #16\n\t" |
| 36 | "st1w {za3h.s[w15, 1]}, p2, [%0]\n\t" |
| 37 | "mov w15, #2\n\t" |
| 38 | "add %0, %0, #16\n\t" |
| 39 | "st1w {za3h.s[w15, 0]}, p2, [%0]\n\t" |
| 40 | "add %0, %0, #16\n\t" |
| 41 | "st1w {za3h.s[w15, 1]}, p2, [%0]\n\t" |
| 42 | "smstop" |
| 43 | : "+r"(result) : "r"(i_1234) |
| 44 | : "x15", "x16", "p2", "d0", "memory"); |
| 45 | } |
| 46 | |
| 47 | int main(void) |
| 48 | { |
| 49 | float result[4 * 4] = { }; |
| 50 | int ret = 0; |
| 51 | |
| 52 | test_fmopa(result); |
| 53 | |
| 54 | for (int i = 0; i < 4; i++) { |
| 55 | float actual = result[i]; |
| 56 | if (fabsf(actual - expected[i]) > 0.001f) { |
| 57 | printf("Test failed at element %d: Expected %f, got %f\n", |
| 58 | i, expected[i], actual); |
| 59 | ret = 1; |
| 60 | } |
| 61 | } |
| 62 | return ret; |
| 63 | } |