tree-optimization/110897 - Fix missed vectorization of shift on both RISC-V and aarch64
Checks
Commit Message
Consider this following case:
#include <stdint.h>
#define TEST2_TYPE(TYPE) \
__attribute__((noipa)) \
void vshiftr_##TYPE (TYPE *__restrict dst, TYPE *__restrict a, TYPE *__restrict b, int n) \
{ \
for (int i = 0; i < n; i++) \
dst[i] = (a[i]) >> b[i]; \
}
#define TEST_ALL() \
TEST2_TYPE(uint8_t) \
TEST2_TYPE(uint16_t) \
TEST2_TYPE(uint32_t) \
TEST2_TYPE(uint64_t) \
TEST_ALL()
Both RISC-V and aarch64 of trunk GCC failed vectorize uint8_t/uint16_t with following missed report:
<source>:17:1: missed: couldn't vectorize loop
<source>:17:1: missed: not vectorized: relevant stmt not supported: patt_46 = MIN_EXPR <_6, 7>;
<source>:17:1: missed: couldn't vectorize loop
<source>:17:1: missed: not vectorized: relevant stmt not supported: patt_47 = MIN_EXPR <_7, 15>;
Compiler returned: 0
Both GCC 13.1 can vectorize, see:
https://godbolt.org/z/6vaMK5M1o
Bootstrap and regression on X86 passed.
Ok for trunk ?
gcc/ChangeLog:
* tree-vect-patterns.cc (vect_recog_over_widening_pattern): Add op vectype.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/binop/narrow-1.c: Adapt testcase.
---
gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/narrow-1.c | 4 ++--
gcc/tree-vect-patterns.cc | 3 ++-
2 files changed, 4 insertions(+), 3 deletions(-)
Comments
On Mon, 7 Aug 2023, Juzhe-Zhong wrote:
> Consider this following case:
>
> #include <stdint.h>
>
> #define TEST2_TYPE(TYPE) \
> __attribute__((noipa)) \
> void vshiftr_##TYPE (TYPE *__restrict dst, TYPE *__restrict a, TYPE *__restrict b, int n) \
> { \
> for (int i = 0; i < n; i++) \
> dst[i] = (a[i]) >> b[i]; \
> }
>
> #define TEST_ALL() \
> TEST2_TYPE(uint8_t) \
> TEST2_TYPE(uint16_t) \
> TEST2_TYPE(uint32_t) \
> TEST2_TYPE(uint64_t) \
>
> TEST_ALL()
>
> Both RISC-V and aarch64 of trunk GCC failed vectorize uint8_t/uint16_t with following missed report:
>
> <source>:17:1: missed: couldn't vectorize loop
> <source>:17:1: missed: not vectorized: relevant stmt not supported: patt_46 = MIN_EXPR <_6, 7>;
> <source>:17:1: missed: couldn't vectorize loop
> <source>:17:1: missed: not vectorized: relevant stmt not supported: patt_47 = MIN_EXPR <_7, 15>;
> Compiler returned: 0
>
> Both GCC 13.1 can vectorize, see:
>
> https://godbolt.org/z/6vaMK5M1o
>
> Bootstrap and regression on X86 passed.
>
> Ok for trunk ?
OK.
> gcc/ChangeLog:
>
> * tree-vect-patterns.cc (vect_recog_over_widening_pattern): Add op vectype.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/riscv/rvv/autovec/binop/narrow-1.c: Adapt testcase.
>
> ---
> gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/narrow-1.c | 4 ++--
> gcc/tree-vect-patterns.cc | 3 ++-
> 2 files changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/narrow-1.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/narrow-1.c
> index 3de8d85b52d..b12cb6355c8 100644
> --- a/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/narrow-1.c
> +++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/binop/narrow-1.c
> @@ -27,5 +27,5 @@
>
> TEST_ALL ()
>
> -/* { dg-final { scan-assembler-times {\tvnsra\.wv} 6 } } */
> -/* { dg-final { scan-assembler-times {\tvnsrl\.wv} 5 } } */
> +/* { dg-final { scan-assembler-times {\tvnsra\.wv} 4 } } */
> +/* { dg-final { scan-assembler-times {\tvnsrl\.wv} 4 } } */
> diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
> index 2cedf238450..cda27fed95b 100644
> --- a/gcc/tree-vect-patterns.cc
> +++ b/gcc/tree-vect-patterns.cc
> @@ -3133,7 +3133,8 @@ vect_recog_over_widening_pattern (vec_info *vinfo,
> return NULL;
> }
> else
> - append_pattern_def_seq (vinfo, last_stmt_info, pattern_stmt);
> + append_pattern_def_seq (vinfo, last_stmt_info, pattern_stmt,
> + op_vectype);
> ops[1] = new_var;
> }
> }
>
Committed to trunk, thanks Richard and Juzhe.
@@ -27,5 +27,5 @@
TEST_ALL ()
-/* { dg-final { scan-assembler-times {\tvnsra\.wv} 6 } } */
-/* { dg-final { scan-assembler-times {\tvnsrl\.wv} 5 } } */
+/* { dg-final { scan-assembler-times {\tvnsra\.wv} 4 } } */
+/* { dg-final { scan-assembler-times {\tvnsrl\.wv} 4 } } */
@@ -3133,7 +3133,8 @@ vect_recog_over_widening_pattern (vec_info *vinfo,
return NULL;
}
else
- append_pattern_def_seq (vinfo, last_stmt_info, pattern_stmt);
+ append_pattern_def_seq (vinfo, last_stmt_info, pattern_stmt,
+ op_vectype);
ops[1] = new_var;
}
}