Merge pull request #2253 from 5265325/feat/zvzip

Add Zvzip extension.

README.md

@@ -77,6 +77,7 @@ Spike supports the following RISC-V ISA features:
   - Zvkt  extension, v1.0
   - Zvkn, Zvknc, Zvkng extension, v1.0
   - Zvks, Zvksc, Zvksg extension, v1.0 
+  - Zvzip extension, v0.1
   - Zicond extension, v1.0
   - Zilsd extension, v1.0
   - Zclsd extension, v1.0

disasm/disasm.cc

@@ -2183,6 +2183,14 @@ void disassembler_t::add_instructions(const isa_parser_t* isa, bool strict)
     DEFINE_VECTOR_MULTIPLYADD_VV(vwabdau_vv);
   }
 
+  if (ext_enabled(EXT_ZVZIP)) {
+    DEFINE_VECTOR_VV(vzip_vv);
+    DEFINE_VECTOR_V(vunzipe_v);
+    DEFINE_VECTOR_V(vunzipo_v);
+    DEFINE_VECTOR_VV(vpaire_vv);
+    DEFINE_VECTOR_VV(vpairo_vv);
+  }
+
   if (ext_enabled(EXT_ZVBB)) {
 #define DEFINE_VECTOR_VIU_ZIMM6(code) \
   add_vector_viu_z6_insn(this, #code, match_##code, mask_##code)

disasm/isa_parser.cc

@@ -321,6 +321,8 @@ void isa_parser_t::add_extension(const std::string& ext_str, const char* str)
   } else if (ext_str == "zvfwldot16bf") {
     extension_table[EXT_ZVFWLDOT16BF] = true;
   } else if (ext_str == "zvkt") {
+  } else if (ext_str == "zvzip") {
+    extension_table[EXT_ZVZIP] = true;
   } else if (ext_str == "sstc") {
       extension_table[EXT_SSTC] = true;
   } else if (ext_str == "smcsrind") {

riscv/insns/vpaire_vv.h

@@ -0,0 +1,7 @@
+// vunzipe.vv vd, vs1, vs2, vm
+
+#include "zvzip_ext_macros.h"
+
+require_zvzip;
+
+VI_VPAIR_VV_LOOP(i - 1, i)

riscv/insns/vpairo_vv.h

@@ -0,0 +1,7 @@
+// vunzipo.vv vd, vs1, vs2, vm
+
+#include "zvzip_ext_macros.h"
+
+require_zvzip;
+
+VI_VPAIR_VV_LOOP(i, i + 1)

riscv/insns/vunzipe_v.h

@@ -0,0 +1,7 @@
+// vunzipe.v vd, vs2, vm
+
+#include "zvzip_ext_macros.h"
+
+require_zvzip;
+
+VI_VUNZIP_V_LOOP(i << 1)

riscv/insns/vunzipo_v.h

@@ -0,0 +1,7 @@
+// vunzipo.v vd, vs2, vm
+
+#include "zvzip_ext_macros.h"
+
+require_zvzip;
+
+VI_VUNZIP_V_LOOP((i << 1) + 1)

riscv/insns/vzip_vv.h

@@ -0,0 +1,7 @@
+// vzip.vv vd, vs1, vs2, vm
+
+#include "zvzip_ext_macros.h"
+
+require_zvzip;
+
+VI_VZIP_VV_LOOP(i >> 1, i >> 1)

riscv/isa_parser.h

@@ -88,6 +88,7 @@ typedef enum {
   EXT_ZVQLDOT16I,
   EXT_ZVFQLDOT8F,
   EXT_ZVFWLDOT16BF,
+  EXT_ZVZIP,
   EXT_SSTC,
   EXT_ZAAMO,
   EXT_ZALRSC,

riscv/riscv.mk.in

@@ -1143,6 +1143,13 @@ riscv_insn_ext_zvabd = \
 	vwabda_vv \
 	vwabdau_vv \
 
+riscv_insn_ext_zvzip = \
+	vzip_vv \
+	vunzipe_v \
+	vunzipo_v \
+	vpaire_vv \
+	vpairo_vv \
+
 riscv_insn_list = \
 	$(riscv_insn_ext_i) \
 	$(riscv_insn_ext_c) \
@@ -1174,6 +1181,7 @@ riscv_insn_list = \
 	$(riscv_insn_ext_zvbdot) \
 	$(riscv_insn_ext_zvldot) \
 	$(riscv_insn_ext_zvabd) \
+	$(riscv_insn_ext_zvzip) \
 	$(riscv_insn_priv) \
 	$(riscv_insn_smrnmi) \
 	$(riscv_insn_svinval) \

riscv/zvzip_ext_macros.h

@@ -0,0 +1,110 @@
+#ifndef RISCV_ZVZIP_MACROS_H_
+#define RISCV_ZVZIP_MACROS_H_
+
+#define require_zvzip \
+  do { \
+    require_vector(true); \
+    require_extension(EXT_ZVZIP); \
+  } while (0)
+
+#define ZVZIP_EXTRACT_V(x, VS2_IDX) \
+  type_sew_t<x>::type UNUSED &vd = P.VU.elt<type_sew_t<x>::type>(rd_num, i, true); \
+  vd = P.VU.elt<type_sew_t<x>::type>(rs2_num, VS2_IDX);
+
+#define ZVZIP_EXTRACT_VV(x, VS1_IDX, VS2_IDX) \
+  type_sew_t<x>::type UNUSED &vd = P.VU.elt<type_sew_t<x>::type>(rd_num, i, true); \
+  vd = (i & 1) \
+    ? ((VS1_IDX) >= P.VU.vlmax ? 0 : P.VU.elt<type_sew_t<x>::type>(rs1_num, VS1_IDX)) \
+    : ((VS2_IDX) >= P.VU.vlmax ? 0 : P.VU.elt<type_sew_t<x>::type>(rs2_num, VS2_IDX));
+
+#define VI_VZIP_VV_CHECK \
+  require_vector(true); \
+  require(P.VU.vflmul <= 4); \
+  require_align(insn.rd(), P.VU.vflmul * 2); \
+  require_align(insn.rs2(), P.VU.vflmul); \
+  require_align(insn.rs1(), P.VU.vflmul); \
+  require_vm; \
+  if (P.VU.vflmul < 1) { \
+    require_noover(insn.rd(), P.VU.vflmul * 2, insn.rs2(), P.VU.vflmul); \
+    require_noover(insn.rd(), P.VU.vflmul * 2, insn.rs1(), P.VU.vflmul); \
+  } else { \
+    require_noover_widen(insn.rd(), P.VU.vflmul * 2, insn.rs2(), P.VU.vflmul); \
+    require_noover_widen(insn.rd(), P.VU.vflmul * 2, insn.rs1(), P.VU.vflmul); \
+  }
+
+#define VI_VZIP_VV_LOOP_BASE \
+  require(P.VU.vsew >= e8 && P.VU.vsew <= e64); \
+  reg_t vl = P.VU.vl->read() * 2; \
+  reg_t UNUSED sew = P.VU.vsew; \
+  reg_t UNUSED rd_num = insn.rd(); \
+  reg_t UNUSED rs1_num = insn.rs1(); \
+  reg_t rs2_num = insn.rs2(); \
+  for (reg_t i = P.VU.vstart->read(); i < vl; ++i) { \
+    VI_LOOP_ELEMENT_SKIP();
+
+#define VI_VZIP_VV_LOOP_END \
+  } \
+  P.VU.vstart->write(0);
+
+#define VI_VZIP_VV_LOOP(VS1_IDX, VS2_IDX) \
+  VI_VZIP_VV_CHECK \
+  VI_VZIP_VV_LOOP_BASE \
+  if (sew == e8) { \
+    ZVZIP_EXTRACT_VV(e8, VS1_IDX, VS2_IDX); \
+  } else if (sew == e16) { \
+    ZVZIP_EXTRACT_VV(e16, VS1_IDX, VS2_IDX); \
+  } else if (sew == e32) { \
+    ZVZIP_EXTRACT_VV(e32, VS1_IDX, VS2_IDX); \
+  } else if (sew == e64) { \
+    ZVZIP_EXTRACT_VV(e64, VS1_IDX, VS2_IDX); \
+  } \
+  VI_VZIP_VV_LOOP_END
+
+#define VI_VUNZIP_V_CHECK \
+  require_vector(true); \
+  require(P.VU.vflmul <= 4); \
+  require_align(insn.rs2(), P.VU.vflmul * 2); \
+  require_align(insn.rd(), P.VU.vflmul); \
+  require_vm; \
+  if (insn.rd() != insn.rs2()) { \
+    require_noover(insn.rd(), P.VU.vflmul, insn.rs2(), P.VU.vflmul * 2); \
+  }
+
+#define VI_VUNZIP_V_LOOP(VS2_IDX) \
+  VI_VUNZIP_V_CHECK \
+  VI_LOOP_BASE \
+  if (sew == e8) { \
+    ZVZIP_EXTRACT_V(e8, VS2_IDX); \
+  } else if (sew == e16) { \
+    ZVZIP_EXTRACT_V(e16, VS2_IDX); \
+  } else if (sew == e32) { \
+    ZVZIP_EXTRACT_V(e32, VS2_IDX); \
+  } else if (sew == e64) { \
+    ZVZIP_EXTRACT_V(e64, VS2_IDX); \
+  } \
+  VI_LOOP_END
+
+#define VI_VPAIR_VV_CHECK \
+  require_vector(true); \
+  require_align(insn.rd(), P.VU.vflmul); \
+  require_align(insn.rs2(), P.VU.vflmul); \
+  require_align(insn.rs1(), P.VU.vflmul); \
+  require_vm; \
+  require_noover(insn.rd(), P.VU.vflmul, insn.rs2(), P.VU.vflmul); \
+  require_noover(insn.rd(), P.VU.vflmul, insn.rs1(), P.VU.vflmul);
+
+#define VI_VPAIR_VV_LOOP(VS1_IDX, VS2_IDX) \
+  VI_VPAIR_VV_CHECK \
+  VI_LOOP_BASE \
+  if (sew == e8) { \
+    ZVZIP_EXTRACT_VV(e8, VS1_IDX, VS2_IDX); \
+  } else if (sew == e16) { \
+    ZVZIP_EXTRACT_VV(e16, VS1_IDX, VS2_IDX); \
+  } else if (sew == e32) { \
+    ZVZIP_EXTRACT_VV(e32, VS1_IDX, VS2_IDX); \
+  } else if (sew == e64) { \
+    ZVZIP_EXTRACT_VV(e64, VS1_IDX, VS2_IDX); \
+  } \
+  VI_LOOP_END
+
+#endif