Debian Codegraph @

gcc/testsuite/gcc.target/aarch64/sve/fmla_1.c

raw
1 /* { dg-do assemble { target aarch64_asm_sve-b16b16_ok } } */
2 /* { dg-do compile { target { ! aarch64_asm_sve-b16b16_ok } } } */
3 /* { dg-options "-O3 -msve-vector-bits=256 --save-temps" } */
4
5 typedef __bf16 vnx8bf __attribute__((vector_size(32)));
6 typedef _Float16 vnx8hf __attribute__((vector_size(32)));
7 typedef float vnx4sf __attribute__((vector_size(32)));
8 typedef double vnx2df __attribute__((vector_size(32)));
9
10 #define DO_OP(TYPE) \
11 void vmad##TYPE (TYPE *x, TYPE y, TYPE z) \
12 { \
13 register TYPE dst asm("z0"); \
14 register TYPE src1 asm("z2"); \
15 register TYPE src2 asm("z4"); \
16 dst = *x; \
17 src1 = y; \
18 src2 = z; \
19 asm volatile ("" :: "w" (dst), "w" (src1), "w" (src2)); \
20 dst = (src1 * src2) + dst; \
21 asm volatile ("" :: "w" (dst)); \
22 *x = dst; \
23 }
24
25 DO_OP (vnx8hf)
26 DO_OP (vnx4sf)
27 DO_OP (vnx2df)
28
29 #pragma GCC target "+sve2+sve-b16b16"
30
31 DO_OP (vnx8bf)
32
33 /* { dg-final { scan-assembler-times {\tbfmla\tz0\.h, p[0-7]/m, z2\.h, z4\.h\n} 1 } } */
34 /* { dg-final { scan-assembler-times {\tfmla\tz0\.h, p[0-7]/m, z2\.h, z4\.h\n} 1 } } */
35 /* { dg-final { scan-assembler-times {\tfmla\tz0\.s, p[0-7]/m, z2\.s, z4\.s\n} 1 } } */
36 /* { dg-final { scan-assembler-times {\tfmla\tz0\.d, p[0-7]/m, z2\.d, z4\.d\n} 1 } } */