Like the non-predicated vrev64q patterns, mve_vrev64q_m_<supf><mode> and mve_vrev64q_m_f<mode> need an early clobber constraint, otherwise we can generate an unpredictable instruction:
Warning: 64-bit element size and same destination and source operands makes instruction UNPREDICTABLE when calling vrevq64_m* with the same first and second arguments. Regression-tested on arm-none-eabi, bootstap in progress on armv8l-unknown-linux-gnueabihf. OK for trunk? Thanks, Christophe gcc/ChangeLog: * config/arm/mve.md: (mve_vrev64q_m_<supf><mode>): Add early clobber. (mve_vrev64q_m_f<mode>): Likewise. gcc/testsuite/ChangeLog: * gcc.target/arm/mve/intrinsics/vrev64q_m_s16-clobber.c: New test. --- gcc/config/arm/mve.md | 4 ++-- .../arm/mve/intrinsics/vrev64q_m_s16-clobber.c | 17 +++++++++++++++++ 2 files changed, 19 insertions(+), 2 deletions(-) create mode 100644 gcc/testsuite/gcc.target/arm/mve/intrinsics/vrev64q_m_s16-clobber.c diff --git a/gcc/config/arm/mve.md b/gcc/config/arm/mve.md index 714178609f7..62186f124da 100644 --- a/gcc/config/arm/mve.md +++ b/gcc/config/arm/mve.md @@ -3503,7 +3503,7 @@ (define_insn "mve_vqshlq_m_r_<supf><mode>" ;; (define_insn "mve_vrev64q_m_<supf><mode>" [ - (set (match_operand:MVE_2 0 "s_register_operand" "=w") + (set (match_operand:MVE_2 0 "s_register_operand" "=&w") (unspec:MVE_2 [(match_operand:MVE_2 1 "s_register_operand" "0") (match_operand:MVE_2 2 "s_register_operand" "w") (match_operand:<MVE_VPRED> 3 "vpr_register_operand" "Up")] @@ -4598,7 +4598,7 @@ (define_insn "mve_vrev32q_m_<supf><mode>" ;; (define_insn "mve_vrev64q_m_f<mode>" [ - (set (match_operand:MVE_0 0 "s_register_operand" "=w") + (set (match_operand:MVE_0 0 "s_register_operand" "=&w") (unspec:MVE_0 [(match_operand:MVE_0 1 "s_register_operand" "0") (match_operand:MVE_0 2 "s_register_operand" "w") (match_operand:<MVE_VPRED> 3 "vpr_register_operand" "Up")] diff --git a/gcc/testsuite/gcc.target/arm/mve/intrinsics/vrev64q_m_s16-clobber.c b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vrev64q_m_s16-clobber.c new file mode 100644 index 00000000000..6464c96181d --- /dev/null +++ b/gcc/testsuite/gcc.target/arm/mve/intrinsics/vrev64q_m_s16-clobber.c @@ -0,0 +1,17 @@ +/* { dg-require-effective-target arm_v8_1m_mve_fp_ok } */ +/* { dg-add-options arm_v8_1m_mve_fp } */ +/* { dg-additional-options "-O2" } */ + +#include "arm_mve.h" + +int16x8_t +foo (int16x8_t a, mve_pred16_t p) +{ + return vrev64q_m_s16 (a, a, p); +} + +float16x8_t +foo2 (float16x8_t a, mve_pred16_t p) +{ + return vrev64q_m_f16 (a, a, p); +} -- 2.34.1