gcc/testsuite/gcc.target/aarch64/acle/fpmr-2.c
raw| 1 | /* { dg-do compile } */ |
| 2 | /* { dg-options "-O1 -march=armv8-a+fp8fma -moverride=tune=cheap_fpmr_write" } */ |
| 3 | |
| 4 | #include <arm_neon.h> |
| 5 | |
| 6 | void foo(float16_t ap[20000], mfloat8x16_t b, mfloat8x16_t c) |
| 7 | { |
| 8 | for (int i = 0; i < 103; i++) |
| 9 | { |
| 10 | float16x8_t a = vld1q_f16(ap + 8*i); |
| 11 | a = vmlalbq_f16_mf8_fpm(a, b, c, 13); |
| 12 | vst1q_f16(ap + 8*i, a); |
| 13 | } |
| 14 | } |
| 15 | /* { dg-final { scan-assembler "msr\tfpmr.*\n\.L2" } } */ |