Merge pull request #868 from eopXD/simplify-narrowing-inst

Simplify narrowing instruction

13 files changed, 72 insertions, 65 deletions

diff --git a/riscv/decode.h b/riscv/decode.h
index 4d37530..313d9f0 100644
--- a/riscv/decode.h
+++ b/riscv/decode.h
@@ -794,19 +794,19 @@ static inline bool is_aligned(const unsigned val, const unsigned pos)
   auto &vd = P.VU.elt<type_sew_t<x>::type>(rd_num, i, true); \
   auto vs2 = P.VU.elt<type_sew_t<x>::type>(rs2_num, i - offset);
 
-#define VI_NSHIFT_PARAMS(sew1, sew2) \
+#define VI_NARROW_PARAMS(sew1, sew2) \
   auto &vd = P.VU.elt<type_usew_t<sew1>::type>(rd_num, i, true); \
   auto vs2_u = P.VU.elt<type_usew_t<sew2>::type>(rs2_num, i); \
   auto vs2 = P.VU.elt<type_sew_t<sew2>::type>(rs2_num, i); \
   auto zimm5 = (type_usew_t<sew1>::type)insn.v_zimm5();
 
-#define VX_NSHIFT_PARAMS(sew1, sew2) \
+#define VX_NARROW_PARAMS(sew1, sew2) \
   auto &vd = P.VU.elt<type_usew_t<sew1>::type>(rd_num, i, true); \
   auto vs2_u = P.VU.elt<type_usew_t<sew2>::type>(rs2_num, i); \
   auto vs2 = P.VU.elt<type_sew_t<sew2>::type>(rs2_num, i); \
   auto rs1 = (type_sew_t<sew1>::type)RS1;
 
-#define VV_NSHIFT_PARAMS(sew1, sew2) \
+#define VV_NARROW_PARAMS(sew1, sew2) \
   auto &vd = P.VU.elt<type_usew_t<sew1>::type>(rd_num, i, true); \
   auto vs2_u = P.VU.elt<type_usew_t<sew2>::type>(rs2_num, i); \
   auto vs2 = P.VU.elt<type_sew_t<sew2>::type>(rs2_num, i); \
@@ -1174,84 +1174,91 @@ static inline bool is_aligned(const unsigned val, const unsigned pos)
 
 // narrow operation loop
 #define VI_VV_LOOP_NARROW(BODY) \
-VI_NARROW_CHECK_COMMON; \
-VI_LOOP_BASE \
-if (sew == e8){ \
-  VI_NARROW_SHIFT(e8, e16) \
-  BODY; \
-}else if(sew == e16){ \
-  VI_NARROW_SHIFT(e16, e32) \
-  BODY; \
-}else if(sew == e32){ \
-  VI_NARROW_SHIFT(e32, e64) \
-  BODY; \
-} \
-VI_LOOP_END 
-
-#define VI_NARROW_SHIFT(sew1, sew2) \
-  type_usew_t<sew1>::type &vd = P.VU.elt<type_usew_t<sew1>::type>(rd_num, i, true); \
-  type_usew_t<sew2>::type vs2_u = P.VU.elt<type_usew_t<sew2>::type>(rs2_num, i); \
-  type_usew_t<sew1>::type zimm5 = (type_usew_t<sew1>::type)insn.v_zimm5(); \
-  type_sew_t<sew2>::type vs2 = P.VU.elt<type_sew_t<sew2>::type>(rs2_num, i); \
-  type_sew_t<sew1>::type vs1 = P.VU.elt<type_sew_t<sew1>::type>(rs1_num, i); \
-  type_sew_t<sew1>::type rs1 = (type_sew_t<sew1>::type)RS1; 
-
-#define VI_VVXI_LOOP_NARROW(BODY, is_vs1) \
-  VI_CHECK_SDS(is_vs1); \
+  VI_CHECK_SDS(true); \
   VI_LOOP_BASE \
   if (sew == e8){ \
-    VI_NARROW_SHIFT(e8, e16) \
+    VV_NARROW_PARAMS(e8, e16) \
     BODY; \
-  } else if (sew == e16) { \
-    VI_NARROW_SHIFT(e16, e32) \
+  }else if(sew == e16){ \
+    VV_NARROW_PARAMS(e16, e32) \
     BODY; \
-  } else if (sew == e32) { \
-    VI_NARROW_SHIFT(e32, e64) \
+  }else if(sew == e32){ \
+    VV_NARROW_PARAMS(e32, e64) \
+    BODY; \
+  } \
+  VI_LOOP_END
+
+#define VI_VX_LOOP_NARROW(BODY) \
+  VI_CHECK_SDS(false); \
+  VI_LOOP_BASE \
+  if (sew == e8){ \
+    VX_NARROW_PARAMS(e8, e16) \
+    BODY; \
+  }else if(sew == e16){ \
+    VX_NARROW_PARAMS(e16, e32) \
+    BODY; \
+  }else if(sew == e32){ \
+    VX_NARROW_PARAMS(e32, e64) \
+    BODY; \
+  } \
+  VI_LOOP_END
+
+#define VI_VI_LOOP_NARROW(BODY) \
+  VI_CHECK_SDS(false); \
+  VI_LOOP_BASE \
+  if (sew == e8){ \
+    VI_NARROW_PARAMS(e8, e16) \
+    BODY; \
+  }else if(sew == e16){ \
+    VI_NARROW_PARAMS(e16, e32) \
+    BODY; \
+  }else if(sew == e32){ \
+    VI_NARROW_PARAMS(e32, e64) \
     BODY; \
   } \
   VI_LOOP_END
 
-#define VI_VI_LOOP_NSHIFT(BODY, is_vs1) \
-  VI_CHECK_SDS(is_vs1); \
+#define VI_VI_LOOP_NSHIFT(BODY) \
+  VI_CHECK_SDS(false); \
   VI_LOOP_NSHIFT_BASE \
   if (sew == e8){ \
-    VI_NSHIFT_PARAMS(e8, e16) \
+    VI_NARROW_PARAMS(e8, e16) \
     BODY; \
   } else if (sew == e16) { \
-    VI_NSHIFT_PARAMS(e16, e32) \
+    VI_NARROW_PARAMS(e16, e32) \
     BODY; \
   } else if (sew == e32) { \
-    VI_NSHIFT_PARAMS(e32, e64) \
+    VI_NARROW_PARAMS(e32, e64) \
     BODY; \
   } \
   VI_LOOP_END
 
-#define VI_VX_LOOP_NSHIFT(BODY, is_vs1) \
-  VI_CHECK_SDS(is_vs1); \
+#define VI_VX_LOOP_NSHIFT(BODY) \
+  VI_CHECK_SDS(false); \
   VI_LOOP_NSHIFT_BASE \
   if (sew == e8){ \
-    VX_NSHIFT_PARAMS(e8, e16) \
+    VX_NARROW_PARAMS(e8, e16) \
     BODY; \
   } else if (sew == e16) { \
-    VX_NSHIFT_PARAMS(e16, e32) \
+    VX_NARROW_PARAMS(e16, e32) \
     BODY; \
   } else if (sew == e32) { \
-    VX_NSHIFT_PARAMS(e32, e64) \
+    VX_NARROW_PARAMS(e32, e64) \
     BODY; \
   } \
   VI_LOOP_END
 
-#define VI_VV_LOOP_NSHIFT(BODY, is_vs1) \
-  VI_CHECK_SDS(is_vs1); \
+#define VI_VV_LOOP_NSHIFT(BODY) \
+  VI_CHECK_SDS(true); \
   VI_LOOP_NSHIFT_BASE \
   if (sew == e8){ \
-    VV_NSHIFT_PARAMS(e8, e16) \
+    VV_NARROW_PARAMS(e8, e16) \
     BODY; \
   } else if (sew == e16) { \
-    VV_NSHIFT_PARAMS(e16, e32) \
+    VV_NARROW_PARAMS(e16, e32) \
     BODY; \
   } else if (sew == e32) { \
-    VV_NSHIFT_PARAMS(e32, e64) \
+    VV_NARROW_PARAMS(e32, e64) \
     BODY; \
   } \
   VI_LOOP_END
diff --git a/riscv/insns/vnclip_wi.h b/riscv/insns/vnclip_wi.h
index 6b68e1d..ea6898c 100644
--- a/riscv/insns/vnclip_wi.h
+++ b/riscv/insns/vnclip_wi.h
@@ -2,7 +2,7 @@
 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_VVXI_LOOP_NARROW
+VI_VI_LOOP_NARROW
 ({
   int128_t result = vs2;
   unsigned shift = zimm5 & ((sew * 2) - 1);
@@ -22,4 +22,4 @@ VI_VVXI_LOOP_NARROW
   }
 
   vd = result;
-}, false)
+})
diff --git a/riscv/insns/vnclip_wv.h b/riscv/insns/vnclip_wv.h
index 5f87697..63b84c6 100644
--- a/riscv/insns/vnclip_wv.h
+++ b/riscv/insns/vnclip_wv.h
@@ -2,7 +2,7 @@
 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_VVXI_LOOP_NARROW
+VI_VV_LOOP_NARROW
 ({
   int128_t result = vs2;
   unsigned shift = vs1 & ((sew * 2) - 1);
@@ -22,4 +22,4 @@ VI_VVXI_LOOP_NARROW
   }
 
   vd = result;
-}, true)
+})
diff --git a/riscv/insns/vnclip_wx.h b/riscv/insns/vnclip_wx.h
index 5436936..482eace 100644
--- a/riscv/insns/vnclip_wx.h
+++ b/riscv/insns/vnclip_wx.h
@@ -2,7 +2,7 @@
 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_VVXI_LOOP_NARROW
+VI_VX_LOOP_NARROW
 ({
   int128_t result = vs2;
   unsigned shift = rs1 & ((sew * 2) - 1);
@@ -22,4 +22,4 @@ VI_VVXI_LOOP_NARROW
   }
 
   vd = result;
-}, false)
+})
diff --git a/riscv/insns/vnclipu_wi.h b/riscv/insns/vnclipu_wi.h
index ba39905..441a3a7 100644
--- a/riscv/insns/vnclipu_wi.h
+++ b/riscv/insns/vnclipu_wi.h
@@ -2,7 +2,7 @@
 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_VVXI_LOOP_NARROW
+VI_VI_LOOP_NARROW
 ({
   uint128_t result = vs2_u;
   unsigned shift = zimm5 & ((sew * 2) - 1);
@@ -20,4 +20,4 @@ VI_VVXI_LOOP_NARROW
   }
 
   vd = result;
-}, false)
+})
diff --git a/riscv/insns/vnclipu_wv.h b/riscv/insns/vnclipu_wv.h
index d6647c6..8072489 100644
--- a/riscv/insns/vnclipu_wv.h
+++ b/riscv/insns/vnclipu_wv.h
@@ -2,7 +2,7 @@
 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_VVXI_LOOP_NARROW
+VI_VV_LOOP_NARROW
 ({
   uint128_t result = vs2_u;
   unsigned shift = vs1 & ((sew * 2) - 1);
@@ -19,4 +19,4 @@ VI_VVXI_LOOP_NARROW
   }
 
   vd = result;
-}, true)
+})
diff --git a/riscv/insns/vnclipu_wx.h b/riscv/insns/vnclipu_wx.h
index bf44233..b2d91c3 100644
--- a/riscv/insns/vnclipu_wx.h
+++ b/riscv/insns/vnclipu_wx.h
@@ -2,7 +2,7 @@
 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_VVXI_LOOP_NARROW
+VI_VX_LOOP_NARROW
 ({
   uint128_t result = vs2_u;
   unsigned shift = rs1 & ((sew * 2) - 1);
@@ -19,4 +19,4 @@ VI_VVXI_LOOP_NARROW
   }
 
   vd = result;
-}, false)
+})
diff --git a/riscv/insns/vnsra_wi.h b/riscv/insns/vnsra_wi.h
index f41979e..0502ff1 100644
--- a/riscv/insns/vnsra_wi.h
+++ b/riscv/insns/vnsra_wi.h
@@ -2,4 +2,4 @@
 VI_VI_LOOP_NSHIFT
 ({
   vd = vs2 >> (zimm5 & (sew * 2 - 1) & 0x1f);
-}, false)
+})
diff --git a/riscv/insns/vnsra_wv.h b/riscv/insns/vnsra_wv.h
index 59f255e..555ce3f 100644
--- a/riscv/insns/vnsra_wv.h
+++ b/riscv/insns/vnsra_wv.h
@@ -2,4 +2,4 @@
 VI_VV_LOOP_NSHIFT
 ({
   vd = vs2 >> (vs1 & (sew * 2 - 1));
-}, true)
+})
diff --git a/riscv/insns/vnsra_wx.h b/riscv/insns/vnsra_wx.h
index adaa24c..05a55e3 100644
--- a/riscv/insns/vnsra_wx.h
+++ b/riscv/insns/vnsra_wx.h
@@ -2,4 +2,4 @@
 VI_VX_LOOP_NSHIFT
 ({
   vd = vs2 >> (rs1 & (sew * 2 - 1));
-}, false)
+})
diff --git a/riscv/insns/vnsrl_wi.h b/riscv/insns/vnsrl_wi.h
index 91402c0..d4dfcf0 100644
--- a/riscv/insns/vnsrl_wi.h
+++ b/riscv/insns/vnsrl_wi.h
@@ -2,4 +2,4 @@
 VI_VI_LOOP_NSHIFT
 ({
   vd = vs2_u >> (zimm5 & (sew * 2 - 1));
-}, false)
+})
diff --git a/riscv/insns/vnsrl_wv.h b/riscv/insns/vnsrl_wv.h
index 609299f..ab72b84 100644
--- a/riscv/insns/vnsrl_wv.h
+++ b/riscv/insns/vnsrl_wv.h
@@ -2,4 +2,4 @@
 VI_VV_LOOP_NSHIFT
 ({
   vd = vs2_u >> (vs1 & (sew * 2 - 1));
-}, true)
+})
diff --git a/riscv/insns/vnsrl_wx.h b/riscv/insns/vnsrl_wx.h
index 8356a2b..e149b38 100644
--- a/riscv/insns/vnsrl_wx.h
+++ b/riscv/insns/vnsrl_wx.h
@@ -2,4 +2,4 @@
 VI_VX_LOOP_NSHIFT
 ({
   vd = vs2_u >> (rs1 & (sew * 2 - 1));
-}, false)
+})