| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */ |
| 3 |
|
| 4 |
#include "riscv_vector.h" |
| 5 |
|
| 6 |
double __attribute__ ((noinline)) |
| 7 |
sumation (double sum0, double sum1, double sum2, double sum3) |
| 8 |
{ |
| 9 |
return sum0 + sum1 + sum2 + sum3; |
| 10 |
} |
| 11 |
|
| 12 |
double |
| 13 |
foo (char const *buf, size_t len) |
| 14 |
{ |
| 15 |
double sum = 0; |
| 16 |
size_t vl = __riscv_vsetvlmax_e8m8 (); |
| 17 |
size_t step = vl * 4; |
| 18 |
const char *it = buf, *end = buf + len; |
| 19 |
for (; it + step <= end;) |
| 20 |
{ |
| 21 |
vfloat32m1_t v0 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 22 |
it += vl; |
| 23 |
vfloat32m1_t v1 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 24 |
it += vl; |
| 25 |
vfloat32m1_t v2 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 26 |
it += vl; |
| 27 |
vfloat32m1_t v3 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 28 |
it += vl; |
| 29 |
|
| 30 |
asm volatile("nop" ::: "memory"); |
| 31 |
vint64m2_t vw0 = __riscv_vfwcvt_rtz_x_f_v_i64m2 (v0, vl); |
| 32 |
vint64m2_t vw1 = __riscv_vfwcvt_rtz_x_f_v_i64m2 (v1, vl); |
| 33 |
vint64m2_t vw2 = __riscv_vfwcvt_rtz_x_f_v_i64m2 (v2, vl); |
| 34 |
vint64m2_t vw3 = __riscv_vfwcvt_rtz_x_f_v_i64m2 (v3, vl); |
| 35 |
|
| 36 |
asm volatile("nop" ::: "memory"); |
| 37 |
double sum0 = __riscv_vmv_x_s_i64m2_i64 (vw0); |
| 38 |
double sum1 = __riscv_vmv_x_s_i64m2_i64 (vw1); |
| 39 |
double sum2 = __riscv_vmv_x_s_i64m2_i64 (vw2); |
| 40 |
double sum3 = __riscv_vmv_x_s_i64m2_i64 (vw3); |
| 41 |
|
| 42 |
sum += sumation (sum0, sum1, sum2, sum3); |
| 43 |
} |
| 44 |
return sum; |
| 45 |
} |
| 46 |
|
| 47 |
double |
| 48 |
foo2 (char const *buf, size_t len) |
| 49 |
{ |
| 50 |
double sum = 0; |
| 51 |
size_t vl = __riscv_vsetvlmax_e8m8 (); |
| 52 |
size_t step = vl * 4; |
| 53 |
const char *it = buf, *end = buf + len; |
| 54 |
for (; it + step <= end;) |
| 55 |
{ |
| 56 |
vfloat32m1_t v0 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 57 |
it += vl; |
| 58 |
vfloat32m1_t v1 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 59 |
it += vl; |
| 60 |
vfloat32m1_t v2 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 61 |
it += vl; |
| 62 |
vfloat32m1_t v3 = __riscv_vle32_v_f32m1 ((void *) it, vl); |
| 63 |
it += vl; |
| 64 |
|
| 65 |
asm volatile("nop" ::: "memory"); |
| 66 |
vint64m2_t vw0 = __riscv_vfwcvt_x_f_v_i64m2 (v0, vl); |
| 67 |
vint64m2_t vw1 = __riscv_vfwcvt_x_f_v_i64m2 (v1, vl); |
| 68 |
vint64m2_t vw2 = __riscv_vfwcvt_x_f_v_i64m2 (v2, vl); |
| 69 |
vint64m2_t vw3 = __riscv_vfwcvt_x_f_v_i64m2 (v3, vl); |
| 70 |
|
| 71 |
asm volatile("nop" ::: "memory"); |
| 72 |
double sum0 = __riscv_vmv_x_s_i64m2_i64 (vw0); |
| 73 |
double sum1 = __riscv_vmv_x_s_i64m2_i64 (vw1); |
| 74 |
double sum2 = __riscv_vmv_x_s_i64m2_i64 (vw2); |
| 75 |
double sum3 = __riscv_vmv_x_s_i64m2_i64 (vw3); |
| 76 |
|
| 77 |
sum += sumation (sum0, sum1, sum2, sum3); |
| 78 |
} |
| 79 |
return sum; |
| 80 |
} |
| 81 |
|
| 82 |
/* { dg-final { scan-assembler-not {vmv1r} } } */ |
| 83 |
/* { dg-final { scan-assembler-not {vmv2r} } } */ |
| 84 |
/* { dg-final { scan-assembler-not {vmv4r} } } */ |
| 85 |
/* { dg-final { scan-assembler-not {vmv8r} } } */ |
| 86 |
/* { dg-final { scan-assembler-not {csrr} } } */ |