The spec for vmv<nf>r.v says: 'the instructions operate as if EEW=SEW, EMUL = NREG, effective length evl= EMUL * VLEN/SEW.'
So the start byte for vstart != 0 should take sew into account Signed-off-by: Weiwei Li <liwei...@iscas.ac.cn> Signed-off-by: Junqiang Wang <wangjunqi...@iscas.ac.cn> --- target/riscv/vector_helper.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/target/riscv/vector_helper.c b/target/riscv/vector_helper.c index 1d4982ef7f..a17ae9dfd1 100644 --- a/target/riscv/vector_helper.c +++ b/target/riscv/vector_helper.c @@ -4890,13 +4890,15 @@ GEN_VEXT_VCOMPRESS_VM(vcompress_vm_d, uint64_t, H8) /* Vector Whole Register Move */ void HELPER(vmvr_v)(void *vd, void *vs2, CPURISCVState *env, uint32_t desc) { - /* EEW = 8 */ + /* EEW = SEW */ uint32_t maxsz = simd_maxsz(desc); - uint32_t i = env->vstart; + uint32_t sewb = 1 << FIELD_EX64(env->vtype, VTYPE, VSEW) + uint32_t startb = env->vstart * sewb; + uint32_t i = startb; memcpy((uint8_t *)vd + H1(i), (uint8_t *)vs2 + H1(i), - maxsz - env->vstart); + maxsz - startb); env->vstart = 0; } -- 2.17.1