vector: Check if there is any vector extension before using vector CSRs

riscv/insns/vnclip_wi.h

@@ -1,9 +1,9 @@
 // vnclip: vd[i] = clip(round(vs2[i] + rnd) >> simm)
-VRM xrm = P.VU.get_vround_mode();
-int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
-int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
 VI_VI_LOOP_NARROW
 ({
+  VRM xrm = P.VU.get_vround_mode();
+  int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
+  int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
   int128_t result = vs2;
   unsigned shift = zimm5 & ((sew * 2) - 1);
 

riscv/insns/vnclip_wv.h

@@ -1,9 +1,9 @@
 // vnclip: vd[i] = clip(round(vs2[i] + rnd) >> vs1[i])
-VRM xrm = P.VU.get_vround_mode();
-int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
-int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
 VI_VV_LOOP_NARROW
 ({
+  VRM xrm = P.VU.get_vround_mode();
+  int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
+  int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
   int128_t result = vs2;
   unsigned shift = vs1 & ((sew * 2) - 1);
 

riscv/insns/vnclipu_wi.h

@@ -1,9 +1,9 @@
 // vnclipu: vd[i] = clip(round(vs2[i] + rnd) >> simm)
-VRM xrm = P.VU.get_vround_mode();
-uint64_t uint_max = UINT64_MAX >> (64 - P.VU.vsew);
-uint64_t sign_mask = UINT64_MAX << P.VU.vsew;
 VI_VI_LOOP_NARROW
 ({
+  VRM xrm = P.VU.get_vround_mode();
+  uint64_t uint_max = UINT64_MAX >> (64 - P.VU.vsew);
+  uint64_t sign_mask = UINT64_MAX << P.VU.vsew;
   uint128_t result = vs2_u;
   unsigned shift = zimm5 & ((sew * 2) - 1);
 

riscv/insns/vnclipu_wv.h

@@ -1,9 +1,9 @@
 // vnclipu: vd[i] = clip(round(vs2[i] + rnd) >> vs1[i])
-VRM xrm = P.VU.get_vround_mode();
-uint64_t uint_max = UINT64_MAX >> (64 - P.VU.vsew);
-uint64_t sign_mask = UINT64_MAX << P.VU.vsew;
 VI_VV_LOOP_NARROW
 ({
+  VRM xrm = P.VU.get_vround_mode();
+  uint64_t uint_max = UINT64_MAX >> (64 - P.VU.vsew);
+  uint64_t sign_mask = UINT64_MAX << P.VU.vsew;
   uint128_t result = vs2_u;
   unsigned shift = vs1 & ((sew * 2) - 1);
 

riscv/insns/vnclipu_wx.h

@@ -1,9 +1,9 @@
 // vnclipu: vd[i] = clip(round(vs2[i] + rnd) >> rs1[i])
-VRM xrm = P.VU.get_vround_mode();
-uint64_t uint_max = UINT64_MAX >> (64 - P.VU.vsew);
-uint64_t sign_mask = UINT64_MAX << P.VU.vsew;
 VI_VX_LOOP_NARROW
 ({
+  VRM xrm = P.VU.get_vround_mode();
+  uint64_t uint_max = UINT64_MAX >> (64 - P.VU.vsew);
+  uint64_t sign_mask = UINT64_MAX << P.VU.vsew;
   uint128_t result = vs2_u;
   unsigned shift = rs1 & ((sew * 2) - 1);
 

riscv/insns/vsmul_vv.h

@@ -1,10 +1,10 @@
 // vsmul.vv vd, vs2, vs1
-VRM xrm = P.VU.get_vround_mode();
-int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
-int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
-
 VI_VV_LOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
+  int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
+  int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
+
   bool overflow = vs1 == vs2 && vs1 == int_min;
   int128_t result = (int128_t)vs1 * (int128_t)vs2;
 

riscv/insns/vsmul_vx.h

@@ -1,10 +1,10 @@
 // vsmul.vx vd, vs2, rs1
-VRM xrm = P.VU.get_vround_mode();
-int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
-int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
-
 VI_VX_LOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
+  int64_t int_max = INT64_MAX >> (64 - P.VU.vsew);
+  int64_t int_min = INT64_MIN >> (64 - P.VU.vsew);
+
   bool overflow = rs1 == vs2 && rs1 == int_min;
   int128_t result = (int128_t)rs1 * (int128_t)vs2;
 

riscv/insns/vssra_vi.h

@@ -1,7 +1,7 @@
 // vssra.vi vd, vs2, simm5
-VRM xrm = P.VU.get_vround_mode();
 VI_VI_LOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
   int sh = simm5 & (sew - 1) & 0x1f;
   int128_t val = vs2;
 

riscv/insns/vssra_vv.h

@@ -1,7 +1,7 @@
 // vssra.vv vd, vs2, vs1
-VRM xrm = P.VU.get_vround_mode();
 VI_VV_LOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
   int sh = vs1 & (sew - 1);
   int128_t val = vs2;
 

riscv/insns/vssra_vx.h

@@ -1,7 +1,7 @@
 // vssra.vx vd, vs2, rs1
-VRM xrm = P.VU.get_vround_mode();
 VI_VX_LOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
   int sh = rs1 & (sew - 1);
   int128_t val = vs2;
 

riscv/insns/vssrl_vi.h

@@ -1,7 +1,7 @@
 // vssra.vi vd, vs2, simm5
-VRM xrm = P.VU.get_vround_mode();
 VI_VI_ULOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
   int sh = zimm5 & (sew - 1) & 0x1f;
   uint128_t val = vs2;
 

riscv/insns/vssrl_vv.h

@@ -1,7 +1,7 @@
 // vssrl.vv vd, vs2, vs1
-VRM xrm = P.VU.get_vround_mode();
 VI_VV_ULOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
   int sh = vs1 & (sew - 1);
   uint128_t val = vs2;
 

riscv/insns/vssrl_vx.h

@@ -1,7 +1,7 @@
 // vssrl.vx vd, vs2, rs1
-VRM xrm = P.VU.get_vround_mode();
 VI_VX_ULOOP
 ({
+  VRM xrm = P.VU.get_vround_mode();
   int sh = rs1 & (sew - 1);
   uint128_t val = vs2;
 

riscv/v_ext_macros.h

@@ -1144,32 +1144,32 @@ static inline bool is_overlapped_widen(const int astart, int asize,
 
 // average loop
 #define VI_VV_LOOP_AVG(op) \
-VRM xrm = p->VU.get_vround_mode(); \
 VI_VV_LOOP({ \
+  VRM xrm = p->VU.get_vround_mode(); \
   uint128_t res = ((uint128_t)vs2) op vs1; \
   INT_ROUNDING(res, xrm, 1); \
   vd = res >> 1; \
 })
 
 #define VI_VX_LOOP_AVG(op) \
-VRM xrm = p->VU.get_vround_mode(); \
 VI_VX_LOOP({ \
+  VRM xrm = p->VU.get_vround_mode(); \
   uint128_t res = ((uint128_t)vs2) op rs1; \
   INT_ROUNDING(res, xrm, 1); \
   vd = res >> 1; \
 })
 
 #define VI_VV_ULOOP_AVG(op) \
-VRM xrm = p->VU.get_vround_mode(); \
 VI_VV_ULOOP({ \
+  VRM xrm = p->VU.get_vround_mode(); \
   uint128_t res = ((uint128_t)vs2) op vs1; \
   INT_ROUNDING(res, xrm, 1); \
   vd = res >> 1; \
 })
 
 #define VI_VX_ULOOP_AVG(op) \
-VRM xrm = p->VU.get_vround_mode(); \
 VI_VX_ULOOP({ \
+  VRM xrm = p->VU.get_vround_mode(); \
   uint128_t res = ((uint128_t)vs2) op rs1; \
   INT_ROUNDING(res, xrm, 1); \
   vd = res >> 1; \