https://gcc.gnu.org/g:81584f5a499f20283f151c66679b93f7cabea40d
commit r17-6136-g81584f5a499f20283f151c66679b93f7cabea40d Author: Pan Li <[email protected]> Date: Fri Nov 6 10:39:51 2026 +0800 RISC-V: Add test cases for vwaddu.wv vs1 reg overlap Add test cases to ensure the vs1 of the vwaddu.wv doesn't overlap the register group of the vs2. gcc/testsuite/ChangeLog: * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m1.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m2.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m4.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf2.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf4.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf8.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m1.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m2.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m4.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf2.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf4.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m1.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m2.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m4.c: New test. * gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-mf2.c: New test. Signed-off-by: Pan Li <[email protected]> Diff: --- .../autovec/group_overlap/vwaddu_wv_vs1-u16-m1.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u16-m2.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u16-m4.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u16-mf2.c | 30 +++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u16-mf4.c | 30 +++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u32-m1.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u32-m2.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u32-m4.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u32-mf2.c | 30 +++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u8-m1.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u8-m2.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u8-m4.c | 38 ++++++++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u8-mf2.c | 30 +++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u8-mf4.c | 30 +++++++++++++++++ .../autovec/group_overlap/vwaddu_wv_vs1-u8-mf8.c | 30 +++++++++++++++++ 15 files changed, 522 insertions(+) diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m1.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m1.c new file mode 100644 index 000000000000..890fab803d98 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m1.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint16m1_t +pick1_u16m2_u16m1 (vuint32m2_t v) +{ + vuint16m2_t n = __riscv_vreinterpret_v_u32m2_u16m2 (v); + + return __riscv_vget_v_u16m2_u16m1 (n, 1); +} + +static vuint16m1_t +pick0_u16m2_u16m1 (vuint32m2_t v) +{ + vuint16m2_t n = __riscv_vreinterpret_v_u32m2_u16m2 (v); + + return __riscv_vget_v_u16m2_u16m1 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e16m1, + vuint16m1_t, + vuint32m2_t, + __riscv_vle32_v_u32m2, + pick1_u16m2_u16m1, + pick0_u16m2_u16m1, + __riscv_vwaddu_wv_u32m2, + __riscv_vse32_v_u32m2, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v10,v0,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v8,v6,v5([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1)|v2,v(2|3)|v4,v(4|5)|v6,v(6|7)|v8,v(8|9)|v10,v(10|11)|v12,v(12|13)|v14,v(14|15)|v16,v(16|17)|v18,v(18|19)|v20,v(20|21)|v22,v(22|23)|v24,v(24|25)|v26,v(26|27)|v28,v(28|29)|v30,v(30|31))([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m2.c new file mode 100644 index 000000000000..d664ffa94dc9 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m2.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint16m2_t +pick1_u16m4_u16m2 (vuint32m4_t v) +{ + vuint16m4_t n = __riscv_vreinterpret_v_u32m4_u16m4 (v); + + return __riscv_vget_v_u16m4_u16m2 (n, 1); +} + +static vuint16m2_t +pick0_u16m4_u16m2 (vuint32m4_t v) +{ + vuint16m4_t n = __riscv_vreinterpret_v_u32m4_u16m4 (v); + + return __riscv_vget_v_u16m4_u16m2 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e16m2, + vuint16m2_t, + vuint32m4_t, + __riscv_vle32_v_u32m4, + pick1_u16m4_u16m2, + pick0_u16m4_u16m2, + __riscv_vwaddu_wv_u32m4, + __riscv_vse32_v_u32m4, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v16,v12,v10([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v20,v0,v4([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1|2|3)|v4,v(4|5|6|7)|v8,v(8|9|10|11)|v12,v(12|13|14|15)|v16,v(16|17|18|19)|v20,v(20|21|22|23)|v24,v(24|25|26|27)|v28,v(28|29|30|31))([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m4.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m4.c new file mode 100644 index 000000000000..5d26a32fe05c --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-m4.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint16m4_t +pick1_u16m8_u16m4 (vuint32m8_t v) +{ + vuint16m8_t n = __riscv_vreinterpret_v_u32m8_u16m8 (v); + + return __riscv_vget_v_u16m8_u16m4 (n, 1); +} + +static vuint16m4_t +pick0_u16m8_u16m4 (vuint32m8_t v) +{ + vuint16m8_t n = __riscv_vreinterpret_v_u32m8_u16m8 (v); + + return __riscv_vget_v_u16m8_u16m4 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e16m4, + vuint16m4_t, + vuint32m8_t, + __riscv_vle32_v_u32m8, + pick1_u16m8_u16m4, + pick0_u16m8_u16m4, + __riscv_vwaddu_wv_u32m8, + __riscv_vse32_v_u32m8, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v0,v24,v20([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v24,v24,v8([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1|2|3|4|5|6|7)|v8,v(8|9|10|11|12|13|14|15)|v16,v(16|17|18|19|20|21|22|23)|v24,v(24|25|26|27|28|29|30|31))([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf2.c new file mode 100644 index 000000000000..702ef3f61ee0 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf2.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint16mf2_t +pick_u16m1_u16mf2 (vuint32m1_t v) +{ + vuint16m1_t n = __riscv_vreinterpret_v_u32m1_u16m1 (v); + + return __riscv_vlmul_trunc_v_u16m1_u16mf2 (n); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e16mf2, + vuint16mf2_t, + vuint32m1_t, + __riscv_vle32_v_u32m1, + pick_u16m1_u16mf2, + pick_u16m1_u16mf2, + __riscv_vwaddu_wv_u32m1, + __riscv_vse32_v_u32m1, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v4,v3,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v5,v0,v1([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v[0-9]+),\1([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf4.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf4.c new file mode 100644 index 000000000000..18de399909a8 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u16-mf4.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint16mf4_t +pick_u16mf2_u16mf4 (vuint32mf2_t v) +{ + vuint16mf2_t n = __riscv_vreinterpret_v_u32mf2_u16mf2 (v); + + return __riscv_vlmul_trunc_v_u16mf2_u16mf4 (n); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e16mf4, + vuint16mf4_t, + vuint32mf2_t, + __riscv_vle32_v_u32mf2, + pick_u16mf2_u16mf4, + pick_u16mf2_u16mf4, + __riscv_vwaddu_wv_u32mf2, + __riscv_vse32_v_u32mf2, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v4,v3,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v5,v0,v1([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v[0-9]+),\1([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m1.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m1.c new file mode 100644 index 000000000000..be09de5becb0 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m1.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint32m1_t +pick1_u32m2_u32m1 (vuint64m2_t v) +{ + vuint32m2_t n = __riscv_vreinterpret_v_u64m2_u32m2 (v); + + return __riscv_vget_v_u32m2_u32m1 (n, 1); +} + +static vuint32m1_t +pick0_u32m2_u32m1 (vuint64m2_t v) +{ + vuint32m2_t n = __riscv_vreinterpret_v_u64m2_u32m2 (v); + + return __riscv_vget_v_u32m2_u32m1 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e32m1, + vuint32m1_t, + vuint64m2_t, + __riscv_vle64_v_u64m2, + pick1_u32m2_u32m1, + pick0_u32m2_u32m1, + __riscv_vwaddu_wv_u64m2, + __riscv_vse64_v_u64m2, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v10,v0,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v8,v6,v5([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1)|v2,v(2|3)|v4,v(4|5)|v6,v(6|7)|v8,v(8|9)|v10,v(10|11)|v12,v(12|13)|v14,v(14|15)|v16,v(16|17)|v18,v(18|19)|v20,v(20|21)|v22,v(22|23)|v24,v(24|25)|v26,v(26|27)|v28,v(28|29)|v30,v(30|31))([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m2.c new file mode 100644 index 000000000000..626c3506a9f6 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m2.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint32m2_t +pick1_u32m4_u32m2 (vuint64m4_t v) +{ + vuint32m4_t n = __riscv_vreinterpret_v_u64m4_u32m4 (v); + + return __riscv_vget_v_u32m4_u32m2 (n, 1); +} + +static vuint32m2_t +pick0_u32m4_u32m2 (vuint64m4_t v) +{ + vuint32m4_t n = __riscv_vreinterpret_v_u64m4_u32m4 (v); + + return __riscv_vget_v_u32m4_u32m2 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e32m2, + vuint32m2_t, + vuint64m4_t, + __riscv_vle64_v_u64m4, + pick1_u32m4_u32m2, + pick0_u32m4_u32m2, + __riscv_vwaddu_wv_u64m4, + __riscv_vse64_v_u64m4, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v16,v12,v10([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v20,v0,v4([^0-9]|$)} 1 } } */ +/* { 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]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m4.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m4.c new file mode 100644 index 000000000000..929f43729085 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-m4.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint32m4_t +pick1_u32m8_u32m4 (vuint64m8_t v) +{ + vuint32m8_t n = __riscv_vreinterpret_v_u64m8_u32m8 (v); + + return __riscv_vget_v_u32m8_u32m4 (n, 1); +} + +static vuint32m4_t +pick0_u32m8_u32m4 (vuint64m8_t v) +{ + vuint32m8_t n = __riscv_vreinterpret_v_u64m8_u32m8 (v); + + return __riscv_vget_v_u32m8_u32m4 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e32m4, + vuint32m4_t, + vuint64m8_t, + __riscv_vle64_v_u64m8, + pick1_u32m8_u32m4, + pick0_u32m8_u32m4, + __riscv_vwaddu_wv_u64m8, + __riscv_vse64_v_u64m8, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v0,v24,v20([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v24,v24,v8([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1|2|3|4|5|6|7)|v8,v(8|9|10|11|12|13|14|15)|v16,v(16|17|18|19|20|21|22|23)|v24,v(24|25|26|27|28|29|30|31))([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-mf2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-mf2.c new file mode 100644 index 000000000000..e23ba9a0dd5c --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u32-mf2.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint32mf2_t +pick_u32m1_u32mf2 (vuint64m1_t v) +{ + vuint32m1_t n = __riscv_vreinterpret_v_u64m1_u32m1 (v); + + return __riscv_vlmul_trunc_v_u32m1_u32mf2 (n); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e32mf2, + vuint32mf2_t, + vuint64m1_t, + __riscv_vle64_v_u64m1, + pick_u32m1_u32mf2, + pick_u32m1_u32mf2, + __riscv_vwaddu_wv_u64m1, + __riscv_vse64_v_u64m1, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v4,v3,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v5,v0,v1([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v[0-9]+),\1([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m1.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m1.c new file mode 100644 index 000000000000..2e6027ad9950 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m1.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint8m1_t +pick1_u8m2_u8m1 (vuint16m2_t v) +{ + vuint8m2_t n = __riscv_vreinterpret_v_u16m2_u8m2 (v); + + return __riscv_vget_v_u8m2_u8m1 (n, 1); +} + +static vuint8m1_t +pick0_u8m2_u8m1 (vuint16m2_t v) +{ + vuint8m2_t n = __riscv_vreinterpret_v_u16m2_u8m2 (v); + + return __riscv_vget_v_u8m2_u8m1 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e8m1, + vuint8m1_t, + vuint16m2_t, + __riscv_vle16_v_u16m2, + pick1_u8m2_u8m1, + pick0_u8m2_u8m1, + __riscv_vwaddu_wv_u16m2, + __riscv_vse16_v_u16m2, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v10,v0,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v8,v6,v5([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1)|v2,v(2|3)|v4,v(4|5)|v6,v(6|7)|v8,v(8|9)|v10,v(10|11)|v12,v(12|13)|v14,v(14|15)|v16,v(16|17)|v18,v(18|19)|v20,v(20|21)|v22,v(22|23)|v24,v(24|25)|v26,v(26|27)|v28,v(28|29)|v30,v(30|31))([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m2.c new file mode 100644 index 000000000000..a7510dc2b8b8 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m2.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint8m2_t +pick1_u8m4_u8m2 (vuint16m4_t v) +{ + vuint8m4_t n = __riscv_vreinterpret_v_u16m4_u8m4 (v); + + return __riscv_vget_v_u8m4_u8m2 (n, 1); +} + +static vuint8m2_t +pick0_u8m4_u8m2 (vuint16m4_t v) +{ + vuint8m4_t n = __riscv_vreinterpret_v_u16m4_u8m4 (v); + + return __riscv_vget_v_u8m4_u8m2 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e8m2, + vuint8m2_t, + vuint16m4_t, + __riscv_vle16_v_u16m4, + pick1_u8m4_u8m2, + pick0_u8m4_u8m2, + __riscv_vwaddu_wv_u16m4, + __riscv_vse16_v_u16m4, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v16,v12,v10([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v20,v0,v4([^0-9]|$)} 1 } } */ +/* { 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]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m4.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m4.c new file mode 100644 index 000000000000..3009756ce8ea --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-m4.c @@ -0,0 +1,38 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint8m4_t +pick1_u8m8_u8m4 (vuint16m8_t v) +{ + vuint8m8_t n = __riscv_vreinterpret_v_u16m8_u8m8 (v); + + return __riscv_vget_v_u8m8_u8m4 (n, 1); +} + +static vuint8m4_t +pick0_u8m8_u8m4 (vuint16m8_t v) +{ + vuint8m8_t n = __riscv_vreinterpret_v_u16m8_u8m8 (v); + + return __riscv_vget_v_u8m8_u8m4 (n, 0); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e8m4, + vuint8m4_t, + vuint16m8_t, + __riscv_vle16_v_u16m8, + pick1_u8m8_u8m4, + pick0_u8m8_u8m4, + __riscv_vwaddu_wv_u16m8, + __riscv_vse16_v_u16m8, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v0,v24,v20([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v24,v24,v8([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v0,v(0|1|2|3|4|5|6|7)|v8,v(8|9|10|11|12|13|14|15)|v16,v(16|17|18|19|20|21|22|23)|v24,v(24|25|26|27|28|29|30|31))([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf2.c new file mode 100644 index 000000000000..ccd8a7a74daf --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf2.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint8mf2_t +pick_u8m1_u8mf2 (vuint16m1_t v) +{ + vuint8m1_t n = __riscv_vreinterpret_v_u16m1_u8m1 (v); + + return __riscv_vlmul_trunc_v_u8m1_u8mf2 (n); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e8mf2, + vuint8mf2_t, + vuint16m1_t, + __riscv_vle16_v_u16m1, + pick_u8m1_u8mf2, + pick_u8m1_u8mf2, + __riscv_vwaddu_wv_u16m1, + __riscv_vse16_v_u16m1, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v4,v3,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v5,v0,v1([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v[0-9]+),\1([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf4.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf4.c new file mode 100644 index 000000000000..93fac9aea5ad --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf4.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint8mf4_t +pick_u8mf2_u8mf4 (vuint16mf2_t v) +{ + vuint8mf2_t n = __riscv_vreinterpret_v_u16mf2_u8mf2 (v); + + return __riscv_vlmul_trunc_v_u8mf2_u8mf4 (n); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e8mf4, + vuint8mf4_t, + vuint16mf2_t, + __riscv_vle16_v_u16mf2, + pick_u8mf2_u8mf4, + pick_u8mf2_u8mf4, + __riscv_vwaddu_wv_u16mf2, + __riscv_vse16_v_u16mf2, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* The vs1 is read with EEW = SEW while the vs2 is read with EEW = 2 * SEW, + thus vs1 must not fall inside the vs2 register group. */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v4,v3,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v5,v0,v1([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v[0-9]+),\1([^0-9]|$)} } } */ diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf8.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf8.c new file mode 100644 index 000000000000..1b8d0899e524 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/group_overlap/vwaddu_wv_vs1-u8-mf8.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gcv -mabi=lp64d" } */ + +#include "group_overlap.h" + +static vuint8mf8_t +pick_u8mf4_u8mf8 (vuint16mf4_t v) +{ + vuint8mf4_t n = __riscv_vreinterpret_v_u16mf4_u8mf4 (v); + + return __riscv_vlmul_trunc_v_u8mf4_u8mf8 (n); +} + +DEF_GROUP_OVERLAP_BINARY_4( + __riscv_vsetvlmax_e8mf8, + vuint8mf8_t, + vuint16mf4_t, + __riscv_vle16_v_u16mf4, + pick_u8mf4_u8mf8, + pick_u8mf4_u8mf8, + __riscv_vwaddu_wv_u16mf4, + __riscv_vse16_v_u16mf4, + vwaddu_wv_vs1, + LOOP_WIDEN_BINARY_WV_BODY_OVERLAP_X2) + +/* 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-times {vwaddu\.wv\s+v4,v3,v2([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-times {vwaddu\.wv\s+v5,v0,v1([^0-9]|$)} 1 } } */ +/* { dg-final { scan-assembler-not {vwaddu\.wv\s+v[0-9]+,(v[0-9]+),\1([^0-9]|$)} } } */
