| 1 | /* |
| 2 | * SME outer product, FZ vs FZ16 |
| 3 | * SPDX-License-Identifier: GPL-2.0-or-later |
| 4 | */ |
| 5 | |
| 6 | #include <stdint.h> |
| 7 | #include <stdio.h> |
| 8 | |
| 9 | static void test_fmopa(uint32_t *result) |
| 10 | { |
| 11 | asm(".arch_extension sme\n\t" |
| 12 | "smstart\n\t" /* Z*, P* and ZArray cleared */ |
| 13 | "ptrue p2.b, vl16\n\t" /* Limit vector length to 16 */ |
| 14 | "ptrue p5.b, vl16\n\t" |
| 15 | "movi d0, #0x00ff\n\t" /* fp16 denormal */ |
| 16 | "movi d16, #0x00ff\n\t" |
| 17 | "mov w15, #0x0001000000\n\t" /* FZ=1, FZ16=0 */ |
| 18 | "msr fpcr, x15\n\t" |
| 19 | "fmopa za3.s, p2/m, p5/m, z16.h, z0.h\n\t" |
| 20 | "mov w15, #0\n\t" |
| 21 | "st1w {za3h.s[w15, 0]}, p2, [%0]\n\t" |
| 22 | "add %0, %0, #16\n\t" |
| 23 | "st1w {za3h.s[w15, 1]}, p2, [%0]\n\t" |
| 24 | "mov w15, #2\n\t" |
| 25 | "add %0, %0, #16\n\t" |
| 26 | "st1w {za3h.s[w15, 0]}, p2, [%0]\n\t" |
| 27 | "add %0, %0, #16\n\t" |
| 28 | "st1w {za3h.s[w15, 1]}, p2, [%0]\n\t" |
| 29 | "smstop" |
| 30 | : "+r"(result) : |
| 31 | : "x15", "x16", "p2", "p5", "d0", "d16", "memory"); |
| 32 | } |
| 33 | |
| 34 | int main(void) |
| 35 | { |
| 36 | uint32_t result[4 * 4] = { }; |
| 37 | |
| 38 | test_fmopa(result); |
| 39 | |
| 40 | if (result[0] != 0x2f7e0100) { |
| 41 | printf("Test failed: Incorrect output in first 4 bytes\n" |
| 42 | "Expected: %08x\n" |
| 43 | "Got: %08x\n", |
| 44 | 0x2f7e0100, result[0]); |
| 45 | return 1; |
| 46 | } |
| 47 | |
| 48 | for (int i = 1; i < 16; ++i) { |
| 49 | if (result[i] != 0) { |
| 50 | printf("Test failed: Non-zero word at position %d\n", i); |
| 51 | return 1; |
| 52 | } |
| 53 | } |
| 54 | |
| 55 | return 0; |
| 56 | } |