https://gcc.gnu.org/g:a2a11abfad353f3e39c3a46abf278ad6258e5547
commit r17-6137-ga2a11abfad353f3e39c3a46abf278ad6258e5547 Author: Pan Li <[email protected]> Date: Fri Nov 6 10:39:51 2026 +0800 RISC-V: Add more test cases for vwadd[u].wv For vs1 not overlap the vs2 register group asm check when the vs1 comes from the vs2 register group. gcc/testsuite/ChangeLog: * gcc.target/riscv/rvv/base/pr112431-65.c: New test. * gcc.target/riscv/rvv/base/pr112431-66.c: New test. Signed-off-by: Pan Li <[email protected]> Diff: --- .../gcc.target/riscv/rvv/base/pr112431-65.c | 62 ++++++++++++++++++++++ .../gcc.target/riscv/rvv/base/pr112431-66.c | 62 ++++++++++++++++++++++ 2 files changed, 124 insertions(+) diff --git a/gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-65.c b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-65.c new file mode 100644 index 000000000000..bd019786975f --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-65.c @@ -0,0 +1,62 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */ + +#include "riscv_vector.h" + +void +foo (void *in, void *out, int n) +{ + for (int i = 0; i < n; i++) + { + asm volatile("nop" ::: "memory"); + vint16m4_t v0 = __riscv_vle16_v_i16m4 (in, 4);in+=100; + vint8m2_t v0_n = __riscv_vreinterpret_v_i16m2_i8m2 (__riscv_vget_v_i16m4_i16m2 (v0, 1)); + v0 = __riscv_vwadd_wv_i16m4_tu (v0, v0, v0_n, 4); + asm volatile("nop" ::: "memory"); + vint16m4_t v1 = __riscv_vle16_v_i16m4 (in, 4);in+=100; + vint8m2_t v1_n = __riscv_vreinterpret_v_i16m2_i8m2 (__riscv_vget_v_i16m4_i16m2 (v1, 1)); + v1 = __riscv_vwadd_wv_i16m4_tu (v1, v1, v1_n, 4); + asm volatile("nop" ::: "memory"); + vint16m4_t v2 = __riscv_vle16_v_i16m4 (in, 4);in+=100; + vint8m2_t v2_n = __riscv_vreinterpret_v_i16m2_i8m2 (__riscv_vget_v_i16m4_i16m2 (v2, 1)); + v2 = __riscv_vwadd_wv_i16m4_tu (v2, v2, v2_n, 4); + asm volatile("nop" ::: "memory"); + vint16m4_t v3 = __riscv_vle16_v_i16m4 (in, 4);in+=100; + vint8m2_t v3_n = __riscv_vreinterpret_v_i16m2_i8m2 (__riscv_vget_v_i16m4_i16m2 (v3, 1)); + v3 = __riscv_vwadd_wv_i16m4_tu (v3, v3, v3_n, 4); + asm volatile("nop" ::: "memory"); + vint16m4_t v4 = __riscv_vle16_v_i16m4 (in, 4);in+=100; + vint8m2_t v4_n = __riscv_vreinterpret_v_i16m2_i8m2 (__riscv_vget_v_i16m4_i16m2 (v4, 1)); + v4 = __riscv_vwadd_wv_i16m4_tu (v4, v4, v4_n, 4); + asm volatile("nop" ::: "memory"); + vint16m4_t v5 = __riscv_vle16_v_i16m4 (in, 4);in+=100; + vint8m2_t v5_n = __riscv_vreinterpret_v_i16m2_i8m2 (__riscv_vget_v_i16m4_i16m2 (v5, 1)); + v5 = __riscv_vwadd_wv_i16m4_tu (v5, v5, v5_n, 4); + asm volatile("nop" ::: "memory"); + vint16m4_t v6 = __riscv_vle16_v_i16m4 (in, 4);in+=100; + vint8m2_t v6_n = __riscv_vreinterpret_v_i16m2_i8m2 (__riscv_vget_v_i16m4_i16m2 (v6, 1)); + v6 = __riscv_vwadd_wv_i16m4_tu (v6, v6, v6_n, 4); + asm volatile("nop" ::: "memory"); + vint8m2_t v7_n = __riscv_vle8_v_i8m2 (in, 4);in+=100; + vint16m4_t v7 = __riscv_vwcvt_x_x_v_i16m4 (v7_n, 4); + + asm volatile("nop" ::: "memory"); + __riscv_vsse16_v_i16m4 (out, 4, v0, 4);out+=100; + __riscv_vsse16_v_i16m4 (out, 4, v1, 4);out+=100; + __riscv_vsse16_v_i16m4 (out, 4, v2, 4);out+=100; + __riscv_vsse16_v_i16m4 (out, 4, v3, 4);out+=100; + __riscv_vsse16_v_i16m4 (out, 4, v4, 4);out+=100; + __riscv_vsse16_v_i16m4 (out, 4, v5, 4);out+=100; + __riscv_vsse16_v_i16m4 (out, 4, v6, 4);out+=100; + __riscv_vsse16_v_i16m4 (out, 4, v7, 4);out+=100; + } +} + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-not {vwadd\.wv\s+v[0-9]+,(v0,v(0|1|2|3)|v4,v(4|5|6|7)|v8,v(8|9|10|11)|v12,v(12|13|14|15)|v16,v(16|17|18|19)|v20,v(20|21|22|23)|v24,v(24|25|26|27)|v28,v(28|29|30|31))([^0-9]|$)} } } */ +/* { dg-final { scan-assembler-not {vmv1r} } } */ +/* { dg-final { scan-assembler-not {vmv2r} } } */ +/* { dg-final { scan-assembler-not {vmv4r} { xfail riscv*-*-* } } } */ +/* { dg-final { scan-assembler-not {vmv8r} } } */ +/* { dg-final { scan-assembler-not {csrr} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-66.c b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-66.c new file mode 100644 index 000000000000..2576cfbf6d76 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112431-66.c @@ -0,0 +1,62 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */ + +#include "riscv_vector.h" + +void +foo (void *in, void *out, int n) +{ + for (int i = 0; i < n; i++) + { + asm volatile("nop" ::: "memory"); + vuint16m4_t v0 = __riscv_vle16_v_u16m4 (in, 4);in+=100; + vuint8m2_t v0_n = __riscv_vreinterpret_v_u16m2_u8m2 (__riscv_vget_v_u16m4_u16m2 (v0, 1)); + v0 = __riscv_vwaddu_wv_u16m4_tu (v0, v0, v0_n, 4); + asm volatile("nop" ::: "memory"); + vuint16m4_t v1 = __riscv_vle16_v_u16m4 (in, 4);in+=100; + vuint8m2_t v1_n = __riscv_vreinterpret_v_u16m2_u8m2 (__riscv_vget_v_u16m4_u16m2 (v1, 1)); + v1 = __riscv_vwaddu_wv_u16m4_tu (v1, v1, v1_n, 4); + asm volatile("nop" ::: "memory"); + vuint16m4_t v2 = __riscv_vle16_v_u16m4 (in, 4);in+=100; + vuint8m2_t v2_n = __riscv_vreinterpret_v_u16m2_u8m2 (__riscv_vget_v_u16m4_u16m2 (v2, 1)); + v2 = __riscv_vwaddu_wv_u16m4_tu (v2, v2, v2_n, 4); + asm volatile("nop" ::: "memory"); + vuint16m4_t v3 = __riscv_vle16_v_u16m4 (in, 4);in+=100; + vuint8m2_t v3_n = __riscv_vreinterpret_v_u16m2_u8m2 (__riscv_vget_v_u16m4_u16m2 (v3, 1)); + v3 = __riscv_vwaddu_wv_u16m4_tu (v3, v3, v3_n, 4); + asm volatile("nop" ::: "memory"); + vuint16m4_t v4 = __riscv_vle16_v_u16m4 (in, 4);in+=100; + vuint8m2_t v4_n = __riscv_vreinterpret_v_u16m2_u8m2 (__riscv_vget_v_u16m4_u16m2 (v4, 1)); + v4 = __riscv_vwaddu_wv_u16m4_tu (v4, v4, v4_n, 4); + asm volatile("nop" ::: "memory"); + vuint16m4_t v5 = __riscv_vle16_v_u16m4 (in, 4);in+=100; + vuint8m2_t v5_n = __riscv_vreinterpret_v_u16m2_u8m2 (__riscv_vget_v_u16m4_u16m2 (v5, 1)); + v5 = __riscv_vwaddu_wv_u16m4_tu (v5, v5, v5_n, 4); + asm volatile("nop" ::: "memory"); + vuint16m4_t v6 = __riscv_vle16_v_u16m4 (in, 4);in+=100; + vuint8m2_t v6_n = __riscv_vreinterpret_v_u16m2_u8m2 (__riscv_vget_v_u16m4_u16m2 (v6, 1)); + v6 = __riscv_vwaddu_wv_u16m4_tu (v6, v6, v6_n, 4); + asm volatile("nop" ::: "memory"); + vuint8m2_t v7_n = __riscv_vle8_v_u8m2 (in, 4);in+=100; + vuint16m4_t v7 = __riscv_vwcvtu_x_x_v_u16m4 (v7_n, 4); + + asm volatile("nop" ::: "memory"); + __riscv_vsse16_v_u16m4 (out, 4, v0, 4);out+=100; + __riscv_vsse16_v_u16m4 (out, 4, v1, 4);out+=100; + __riscv_vsse16_v_u16m4 (out, 4, v2, 4);out+=100; + __riscv_vsse16_v_u16m4 (out, 4, v3, 4);out+=100; + __riscv_vsse16_v_u16m4 (out, 4, v4, 4);out+=100; + __riscv_vsse16_v_u16m4 (out, 4, v5, 4);out+=100; + __riscv_vsse16_v_u16m4 (out, 4, v6, 4);out+=100; + __riscv_vsse16_v_u16m4 (out, 4, v7, 4);out+=100; + } +} + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1|2|3)|v4,v(4|5|6|7)|v8,v(8|9|10|11)|v12,v(12|13|14|15)|v16,v(16|17|18|19)|v20,v(20|21|22|23)|v24,v(24|25|26|27)|v28,v(28|29|30|31))([^0-9]|$)} } } */ +/* { dg-final { scan-assembler-not {vmv1r} } } */ +/* { dg-final { scan-assembler-not {vmv2r} } } */ +/* { dg-final { scan-assembler-not {vmv4r} { xfail riscv*-*-* } } } */ +/* { dg-final { scan-assembler-not {vmv8r} } } */ +/* { dg-final { scan-assembler-not {csrr} } } */
