Merge pull request #2207 from riscv-software-src/fix-2206HEADmaster

Only set mstatus.VS for legal vector instructions

32 files changed, 83 insertions, 76 deletions

diff --git a/riscv/csrs.cc b/riscv/csrs.cc
index cc12e1d..27c26d3 100644
--- a/riscv/csrs.cc
+++ b/riscv/csrs.cc
@@ -1598,7 +1598,7 @@ void vector_csr_t::write_raw(const reg_t val) noexcept {
 
 bool vector_csr_t::unlogged_write(const reg_t val) noexcept {
   if (mask == 0) return false;
-  dirty_vs_state;
+  STATE.sstatus->dirty(SSTATUS_VS);
   return basic_csr_t::unlogged_write(val & mask);
 }
 
@@ -1612,7 +1612,7 @@ void vxsat_csr_t::verify_permissions(insn_t insn, bool write) const {
 }
 
 bool vxsat_csr_t::unlogged_write(const reg_t val) noexcept {
-  dirty_vs_state;
+  STATE.sstatus->dirty(SSTATUS_VS);
   return masked_csr_t::unlogged_write(val);
 }
 
diff --git a/riscv/decode_macros.h b/riscv/decode_macros.h
index 6f24799..b778668 100644
--- a/riscv/decode_macros.h
+++ b/riscv/decode_macros.h
@@ -110,7 +110,6 @@
 #define FRS3_D READ_FREG_D(insn.rs3())
 #define dirty_fp_state  STATE.sstatus->dirty(SSTATUS_FS)
 #define dirty_ext_state STATE.sstatus->dirty(SSTATUS_XS)
-#define dirty_vs_state  STATE.sstatus->dirty(SSTATUS_VS)
 #define DO_WRITE_FREG(reg, value) (STATE.FPR.write(reg, value), dirty_fp_state)
 #define WRITE_FRD(value) WRITE_FREG(insn.rd(), value)
 #define WRITE_FRD_H(value) \
@@ -173,14 +172,12 @@ static inline bool is_aligned(const unsigned val, const unsigned pos)
     if (alu && !P.VU.vstart_alu) \
       require(P.VU.vstart->read() == 0); \
     WRITE_VSTATUS; \
-    dirty_vs_state; \
   } while (0);
 #define require_vector_novtype(is_log) \
   do { \
     require_vector_vs; \
     if (is_log) \
       WRITE_VSTATUS; \
-    dirty_vs_state; \
   } while (0);
 #define require_align(val, pos) require(is_aligned(val, pos))
 #define require_noover(astart, asize, bstart, bsize) \
diff --git a/riscv/insns/vcompress_vm.h b/riscv/insns/vcompress_vm.h
index 6624d8b..8754eb0 100644
--- a/riscv/insns/vcompress_vm.h
+++ b/riscv/insns/vcompress_vm.h
@@ -27,4 +27,5 @@ VI_GENERAL_LOOP_BASE
 
     ++pos;
   }
-VI_LOOP_END_BASE;
+
+VI_LOOP_END;
diff --git a/riscv/insns/vcpop_m.h b/riscv/insns/vcpop_m.h
index 26a1276..260f45d 100644
--- a/riscv/insns/vcpop_m.h
+++ b/riscv/insns/vcpop_m.h
@@ -10,3 +10,5 @@ for (reg_t i=P.VU.vstart->read(); i<vl; ++i) {
   popcount += vs2_bit && (insn.v_vm() || P.VU.mask_elt(0, i));
 }
 WRITE_RD(popcount);
+
+VECTOR_END;
diff --git a/riscv/insns/vfbdot_vv.h b/riscv/insns/vfbdot_vv.h
index 8d4c792..b70e309 100644
--- a/riscv/insns/vfbdot_vv.h
+++ b/riscv/insns/vfbdot_vv.h
@@ -14,3 +14,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vfirst_m.h b/riscv/insns/vfirst_m.h
index e3f5263..261ea5a 100644
--- a/riscv/insns/vfirst_m.h
+++ b/riscv/insns/vfirst_m.h
@@ -14,3 +14,5 @@ for (reg_t i=P.VU.vstart->read(); i < vl; ++i) {
   }
 }
 WRITE_RD(pos);
+
+VECTOR_END;
diff --git a/riscv/insns/vfmv_f_s.h b/riscv/insns/vfmv_f_s.h
index 65a3cff..f4e984e 100644
--- a/riscv/insns/vfmv_f_s.h
+++ b/riscv/insns/vfmv_f_s.h
@@ -29,4 +29,4 @@ if (FLEN == 64) {
   WRITE_FRD(f32(vs2_0));
 }
 
-P.VU.vstart->write(0);
+VECTOR_END;
diff --git a/riscv/insns/vfmv_s_f.h b/riscv/insns/vfmv_s_f.h
index 917948d..11975f2 100644
--- a/riscv/insns/vfmv_s_f.h
+++ b/riscv/insns/vfmv_s_f.h
@@ -21,4 +21,4 @@ if (vl > 0 && P.VU.vstart->read() < vl) {
       break;
   }
 }
-P.VU.vstart->write(0);
+VECTOR_END;
diff --git a/riscv/insns/vfqbdot_alt_vv.h b/riscv/insns/vfqbdot_alt_vv.h
index f1df781..dcdcb11 100644
--- a/riscv/insns/vfqbdot_alt_vv.h
+++ b/riscv/insns/vfqbdot_alt_vv.h
@@ -15,3 +15,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vfqbdot_vv.h b/riscv/insns/vfqbdot_vv.h
index fe3e652..b4fdffe 100644
--- a/riscv/insns/vfqbdot_vv.h
+++ b/riscv/insns/vfqbdot_vv.h
@@ -15,3 +15,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vfqldot_alt_vv.h b/riscv/insns/vfqldot_alt_vv.h
index ea18828..a2b7fd2 100644
--- a/riscv/insns/vfqldot_alt_vv.h
+++ b/riscv/insns/vfqldot_alt_vv.h
@@ -15,3 +15,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vfqldot_vv.h b/riscv/insns/vfqldot_vv.h
index b03ec29..ca46d26 100644
--- a/riscv/insns/vfqldot_vv.h
+++ b/riscv/insns/vfqldot_vv.h
@@ -15,3 +15,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vfwbdot_vv.h b/riscv/insns/vfwbdot_vv.h
index b8d35a7..47f7c9b 100644
--- a/riscv/insns/vfwbdot_vv.h
+++ b/riscv/insns/vfwbdot_vv.h
@@ -13,3 +13,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vfwldot_vv.h b/riscv/insns/vfwldot_vv.h
index 63a4e47..fe1cf09 100644
--- a/riscv/insns/vfwldot_vv.h
+++ b/riscv/insns/vfwldot_vv.h
@@ -13,3 +13,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vid_v.h b/riscv/insns/vid_v.h
index 510132d..ad7da21 100644
--- a/riscv/insns/vid_v.h
+++ b/riscv/insns/vid_v.h
@@ -25,4 +25,4 @@ for (reg_t i = P.VU.vstart->read() ; i < P.VU.vl->read(); ++i) {
   }
 }
 
-P.VU.vstart->write(0);
+VECTOR_END;
diff --git a/riscv/insns/viota_m.h b/riscv/insns/viota_m.h
index 00155db..4a67769 100644
--- a/riscv/insns/viota_m.h
+++ b/riscv/insns/viota_m.h
@@ -45,3 +45,4 @@ for (reg_t i = 0; i < vl; ++i) {
   }
 }
 
+VECTOR_END;
diff --git a/riscv/insns/vmsbf_m.h b/riscv/insns/vmsbf_m.h
index 3f736e0..7c907a6 100644
--- a/riscv/insns/vmsbf_m.h
+++ b/riscv/insns/vmsbf_m.h
@@ -25,3 +25,5 @@ for (reg_t i = P.VU.vstart->read(); i < vl; ++i) {
     P.VU.set_mask_elt(rd_num, i, res);
   }
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vmsif_m.h b/riscv/insns/vmsif_m.h
index b029327..7d98655 100644
--- a/riscv/insns/vmsif_m.h
+++ b/riscv/insns/vmsif_m.h
@@ -26,3 +26,5 @@ for (reg_t i = P.VU.vstart->read(); i < vl; ++i) {
     P.VU.set_mask_elt(rd_num, i, res);
   }
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vmsof_m.h b/riscv/insns/vmsof_m.h
index 5753dbf..caaff04 100644
--- a/riscv/insns/vmsof_m.h
+++ b/riscv/insns/vmsof_m.h
@@ -24,3 +24,5 @@ for (reg_t i = P.VU.vstart->read() ; i < vl; ++i) {
     P.VU.set_mask_elt(rd_num, i, res);
   }
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vmv_s_x.h b/riscv/insns/vmv_s_x.h
index 23a6b56..bd848a8 100644
--- a/riscv/insns/vmv_s_x.h
+++ b/riscv/insns/vmv_s_x.h
@@ -26,4 +26,4 @@ if (vl > 0 && P.VU.vstart->read() < vl) {
   vl = 0;
 }
 
-P.VU.vstart->write(0);
+VECTOR_END;
diff --git a/riscv/insns/vmv_x_s.h b/riscv/insns/vmv_x_s.h
index 57a9e1a..c28dbf4 100644
--- a/riscv/insns/vmv_x_s.h
+++ b/riscv/insns/vmv_x_s.h
@@ -24,4 +24,4 @@ default:
 
 WRITE_RD(sext_xlen(res));
 
-P.VU.vstart->write(0);
+VECTOR_END;
diff --git a/riscv/insns/vmvnfr_v.h b/riscv/insns/vmvnfr_v.h
index 9c52810..fa47bb6 100644
--- a/riscv/insns/vmvnfr_v.h
+++ b/riscv/insns/vmvnfr_v.h
@@ -24,4 +24,4 @@ if (vd != vs2 && start < size) {
   }
 }
 
-P.VU.vstart->write(0);
+VECTOR_END;
diff --git a/riscv/insns/vqbdots_vv.h b/riscv/insns/vqbdots_vv.h
index 55c3dd2..de34835 100644
--- a/riscv/insns/vqbdots_vv.h
+++ b/riscv/insns/vqbdots_vv.h
@@ -21,3 +21,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vqbdotu_vv.h b/riscv/insns/vqbdotu_vv.h
index a73d568..8eee3a9 100644
--- a/riscv/insns/vqbdotu_vv.h
+++ b/riscv/insns/vqbdotu_vv.h
@@ -21,3 +21,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vqldots_vv.h b/riscv/insns/vqldots_vv.h
index ce6376a..bde2296 100644
--- a/riscv/insns/vqldots_vv.h
+++ b/riscv/insns/vqldots_vv.h
@@ -21,3 +21,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vqldotu_vv.h b/riscv/insns/vqldotu_vv.h
index 2b674b1..cd3691a 100644
--- a/riscv/insns/vqldotu_vv.h
+++ b/riscv/insns/vqldotu_vv.h
@@ -21,3 +21,5 @@ switch (P.VU.vsew) {
   }
   default: require(false);
 }
+
+VECTOR_END;
diff --git a/riscv/insns/vsetivli.h b/riscv/insns/vsetivli.h
index f880e96..f30564f 100644
--- a/riscv/insns/vsetivli.h
+++ b/riscv/insns/vsetivli.h
@@ -1,2 +1,3 @@
 require_vector_novtype(false);
 WRITE_RD(P.VU.set_vl(insn.rd(), -1, insn.rs1(), insn.v_zimm10()));
+VECTOR_END;
diff --git a/riscv/insns/vsetvl.h b/riscv/insns/vsetvl.h
index 4d03542..f18d5be 100644
--- a/riscv/insns/vsetvl.h
+++ b/riscv/insns/vsetvl.h
@@ -1,2 +1,3 @@
 require_vector_novtype(false);
 WRITE_RD(P.VU.set_vl(insn.rd(), insn.rs1(), RS1, RS2));
+VECTOR_END;
diff --git a/riscv/insns/vsetvli.h b/riscv/insns/vsetvli.h
index d1f43b5..140cc30 100644
--- a/riscv/insns/vsetvli.h
+++ b/riscv/insns/vsetvli.h
@@ -1,2 +1,3 @@
 require_vector_novtype(false);
 WRITE_RD(P.VU.set_vl(insn.rd(), insn.rs1(), RS1, insn.v_zimm11()));
+VECTOR_END;
diff --git a/riscv/v_ext_macros.h b/riscv/v_ext_macros.h
index 7f5256c..e96fc12 100644
--- a/riscv/v_ext_macros.h
+++ b/riscv/v_ext_macros.h
@@ -233,16 +233,19 @@ static inline bool is_overlapped_widen(const int astart, int asize,
 #define VI_LOOP_END_BASE \
  }
 
+#define VECTOR_END \
+  P.VU.vstart->write(0)
+
 #define VI_LOOP_END \
   VI_LOOP_END_BASE \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_LOOP_REDUCTION_END(x) \
   } \
   if (vl > 0) { \
     vd_0_des = vd_0_res; \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_LOOP_CARRY_BASE \
   VI_GENERAL_LOOP_BASE \
@@ -253,7 +256,7 @@ static inline bool is_overlapped_widen(const int astart, int asize,
 #define VI_LOOP_CARRY_END \
     P.VU.set_mask_elt(insn.rd(), i, res); \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 #define VI_LOOP_WITH_CARRY_BASE \
   VI_GENERAL_LOOP_BASE \
   const uint128_t op_mask = (UINT64_MAX >> (64 - sew)); \
@@ -274,7 +277,7 @@ static inline bool is_overlapped_widen(const int astart, int asize,
 #define VI_LOOP_CMP_END \
     P.VU.set_mask_elt(insn.rd(), i, res); \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_LOOP_MASK(op) \
   require(P.VU.vsew <= e64); \
@@ -285,7 +288,7 @@ static inline bool is_overlapped_widen(const int astart, int asize,
     bool vs1 = P.VU.mask_elt(insn.rs1(), i); \
     P.VU.set_mask_elt(insn.rd(), i, (op)); \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_LOOP_NSHIFT_BASE \
   VI_GENERAL_LOOP_BASE; \
@@ -1199,7 +1202,7 @@ VI_VX_ULOOP({ \
       P.VU.elt<elt_width##_t>(vd + fn * emul, vreg_inx, true) = val; \
     } \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_LDST_GET_INDEX(elt_width) \
   reg_t index; \
@@ -1252,7 +1255,7 @@ VI_VX_ULOOP({ \
       } \
     } \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_ST(stride, offset, elt_width, is_mask_ldst) \
   const reg_t nf = insn.v_nf() + 1; \
@@ -1270,7 +1273,7 @@ VI_VX_ULOOP({ \
         baseAddr + (stride) + (offset) * sizeof(elt_width##_t), val); \
     } \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_ST_INDEX(elt_width, is_seg) \
   const reg_t nf = insn.v_nf() + 1; \
@@ -1306,7 +1309,7 @@ VI_VX_ULOOP({ \
       } \
     } \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_LDST_FF(elt_width) \
   const reg_t nf = insn.v_nf() + 1; \
@@ -1325,8 +1328,10 @@ VI_VX_ULOOP({ \
         val = MMU.load<elt_width##_t>( \
           baseAddr + (i * nf + fn) * sizeof(elt_width##_t)); \
       } catch (trap_t& t) { \
-        if (i == 0) \
+        if (i == 0) { \
+          P.VU.vstart->write(0); /* dirty VS */ \
           throw; /* Only take exception on zeroth element */ \
+        } \
         /* Reduce VL if an exception occurs on a later element */ \
         early_stop = true; \
         P.VU.vl->write_raw(i); \
@@ -1339,7 +1344,7 @@ VI_VX_ULOOP({ \
       break; \
     } \
   } \
-  p->VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_LD_WHOLE(elt_width) \
   require_vector_novtype(true); \
@@ -1350,28 +1355,12 @@ VI_VX_ULOOP({ \
   require_align(vd, len); \
   const reg_t elt_per_reg = P.VU.vlenb / sizeof(elt_width ## _t); \
   const reg_t size = len * elt_per_reg; \
-  if (P.VU.vstart->read() < size) { \
-    reg_t i = P.VU.vstart->read() / elt_per_reg; \
-    reg_t off = P.VU.vstart->read() % elt_per_reg; \
-    if (off) { \
-      for (reg_t pos = off; pos < elt_per_reg; ++pos) { \
-        auto val = MMU.load<elt_width##_t>(baseAddr + \
-          P.VU.vstart->read() * sizeof(elt_width ## _t)); \
-        P.VU.elt<elt_width ## _t>(vd + i, pos, true) = val; \
-        P.VU.vstart->write(P.VU.vstart->read() + 1); \
-      } \
-      ++i; \
-    } \
-    for (; i < len; ++i) { \
-      for (reg_t pos = 0; pos < elt_per_reg; ++pos) { \
-        auto val = MMU.load<elt_width##_t>(baseAddr + \
-          P.VU.vstart->read() * sizeof(elt_width ## _t)); \
-        P.VU.elt<elt_width ## _t>(vd + i, pos, true) = val; \
-        P.VU.vstart->write(P.VU.vstart->read() + 1); \
-      } \
-    } \
+  for (reg_t i = P.VU.vstart->read(); i < size; i++) { \
+    P.VU.vstart->write(i); \
+    auto val = MMU.load<elt_width##_t>(baseAddr + i * sizeof(elt_width ## _t)); \
+    P.VU.elt<elt_width ## _t>(vd, i, true) = val; \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_ST_WHOLE \
   require_vector_novtype(true); \
@@ -1380,27 +1369,12 @@ VI_VX_ULOOP({ \
   const reg_t len = insn.v_nf() + 1; \
   require_align(vs3, len); \
   const reg_t size = len * P.VU.vlenb; \
-  \
-  if (P.VU.vstart->read() < size) { \
-    reg_t i = P.VU.vstart->read() / P.VU.vlenb; \
-    reg_t off = P.VU.vstart->read() % P.VU.vlenb; \
-    if (off) { \
-      for (reg_t pos = off; pos < P.VU.vlenb; ++pos) { \
-        auto val = P.VU.elt<uint8_t>(vs3 + i, pos); \
-        MMU.store<uint8_t>(baseAddr + P.VU.vstart->read(), val); \
-        P.VU.vstart->write(P.VU.vstart->read() + 1); \
-      } \
-      i++; \
-    } \
-    for (; i < len; ++i) { \
-      for (reg_t pos = 0; pos < P.VU.vlenb; ++pos) { \
-        auto val = P.VU.elt<uint8_t>(vs3 + i, pos); \
-        MMU.store<uint8_t>(baseAddr + P.VU.vstart->read(), val); \
-        P.VU.vstart->write(P.VU.vstart->read() + 1); \
-      } \
-    } \
+  for (reg_t i = P.VU.vstart->read(); i < size; i++) { \
+    P.VU.vstart->write(i); \
+    auto val = P.VU.elt<uint8_t>(vs3, i); \
+    MMU.store<uint8_t>(baseAddr + i, val); \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_EXT_CHECK(div) \
   require(insn.rd() != insn.rs2()); \
@@ -1509,11 +1483,11 @@ VI_VX_ULOOP({ \
 
 #define VI_VFP_LOOP_END \
   } \
-  P.VU.vstart->write(0); \
+  VECTOR_END; \
 
 #define VI_VFP_LOOP_REDUCTION_END(x) \
   } \
-  P.VU.vstart->write(0); \
+  VECTOR_END; \
   if (vl > 0) { \
     if (is_propagate && !is_active) { \
       switch (x) { \
@@ -1575,7 +1549,7 @@ VI_VX_ULOOP({ \
       break; \
     }; \
   } \
-  P.VU.vstart->write(0);
+  VECTOR_END;
 
 #define VI_VFP_VV_LOOP(BODY16, BODY32, BODY64) \
   VI_CHECK_SSS(true); \
diff --git a/riscv/vector_unit.cc b/riscv/vector_unit.cc
index a7ba018..6eadf59 100644
--- a/riscv/vector_unit.cc
+++ b/riscv/vector_unit.cc
@@ -89,7 +89,6 @@ reg_t vectorUnit_t::vectorUnit_t::set_vl(int rd, int rs1, reg_t reqVL, reg_t new
     vl->write_raw(std::min(reqVL, vlmax));
   }
 
-  vstart->write_raw(0);
   return vl->read();
 }
 
diff --git a/riscv/zvk_ext_macros.h b/riscv/zvk_ext_macros.h
index e96e0a8..1ab4331 100644
--- a/riscv/zvk_ext_macros.h
+++ b/riscv/zvk_ext_macros.h
@@ -322,7 +322,7 @@
       VV_VD_VS1_VS2_EGU32x4_PARAMS(vd_num, vs1_num, vs2_num, idx_eg); \
       EG_BODY \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)
 
 // Processes all 32b*8 element groups available in the vector register
@@ -377,7 +377,7 @@
       VV_VD_VS1_VS2_EGU32x8_PARAMS(vd_num, vs1_num, vs2_num, idx_eg); \
       EG_BODY \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)
 
 // Processes all 32b*4 element groups available in the vector register
@@ -445,7 +445,7 @@
         EG_BODY \
       } \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)
 
 // Processes all 32b*4 element groups available in the vector register
@@ -513,7 +513,7 @@
         EG_BODY \
       } \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)
 
 // Processes all 32b*4 element groups available in the vector registers
@@ -560,7 +560,7 @@
       VV_VD_VS2_EGU32x4_PARAMS(vd_num, vs2_num, idx_eg); \
       EG_BODY \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)
 
 // Processes all 32b*4 element groups available in the vector registers
@@ -616,7 +616,7 @@
         EG_BODY \
       } \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)
 
 // Processes all 32b*8 element groups available in the vector registers
@@ -672,7 +672,7 @@
         EG_BODY \
       } \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)
 
 // Processes all 64b*4 element groups available in the vector registers
@@ -726,7 +726,7 @@
       VV_VD_VS1_VS2_EGU64x4_PARAMS(vd_num, vs1_num, vs2_num, idx_eg); \
       EG_BODY \
     } \
-    P.VU.vstart->write(0); \
+    VECTOR_END; \
   } while (0)