X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=riscv%2Fdecode.h;h=596a2ad9bfea1bfe24b38ed1b733e5a8c91bb688;hb=4299874ad4b07ef457776513a64e5b2397a6a75e;hp=fd3b2923a7b57430bcf3ee7f4e47ea3c89f9727f;hpb=614902fd205a1640ef9aeb7db3b75be3622d485b;p=riscv-isa-sim.git

diff --git a/riscv/decode.h b/riscv/decode.h
index fd3b292..596a2ad 100644
--- a/riscv/decode.h
+++ b/riscv/decode.h
@@ -1,38 +1,35 @@
+// See LICENSE for license details.
+
 #ifndef _RISCV_DECODE_H
 #define _RISCV_DECODE_H
 
-#define __STDC_LIMIT_MACROS
-#include <stdint.h>
+#if (-1 != ~0) || ((-1 >> 1) != -1)
+# error spike requires a two''s-complement c++ implementation
+#endif
+
+#ifdef WORDS_BIGENDIAN
+# error spike requires a little-endian host
+#endif
+
+#include <cstdint>
 #include <string.h>
-#include "pcr.h"
+#include <strings.h>
+#include "encoding.h"
 #include "config.h"
-
-typedef int int128_t __attribute__((mode(TI)));
-typedef unsigned int uint128_t __attribute__((mode(TI)));
+#include "common.h"
+#include "softfloat_types.h"
+#include "specialize.h"
+#include <cinttypes>
 
 typedef int64_t sreg_t;
 typedef uint64_t reg_t;
-typedef uint64_t freg_t;
 
-const int OPCODE_BITS = 7;
+const int NXPR = 32;
+const int NFPR = 32;
+const int NCSR = 4096;
 
-const int XPRID_BITS = 5;
-const int NXPR = 1 << XPRID_BITS;
-
-const int FPR_BITS = 64;
-const int FPRID_BITS = 5;
-const int NFPR = 1 << FPRID_BITS;
-
-const int IMM_BITS = 12;
-const int IMMLO_BITS = 7;
-const int TARGET_BITS = 25;
-const int FUNCT_BITS = 3;
-const int FUNCTR_BITS = 7;
-const int FFUNCT_BITS = 2;
-const int RM_BITS = 3;
-const int BIGIMM_BITS = 20;
-const int BRANCH_ALIGN_BITS = 1;
-const int JUMP_ALIGN_BITS = 1;
+#define X_RA 1
+#define X_SP 2
 
 #define FP_RD_NE  0
 #define FP_RD_0   1
@@ -57,238 +54,221 @@ const int JUMP_ALIGN_BITS = 1;
 #define FSR_NXA  (FPEXC_NX << FSR_AEXC_SHIFT)
 #define FSR_AEXC (FSR_NVA | FSR_OFA | FSR_UFA | FSR_DZA | FSR_NXA)
 
-#define FSR_ZERO ~(FSR_RD | FSR_AEXC)
-
-// note: bit fields are in little-endian order
-struct itype_t
-{
-  unsigned opcode : OPCODE_BITS;
-  unsigned funct : FUNCT_BITS;
-  signed imm12 : IMM_BITS;
-  unsigned rs1 : XPRID_BITS;
-  unsigned rd : XPRID_BITS;
-};
-
-struct btype_t
-{
-  unsigned opcode : OPCODE_BITS;
-  unsigned funct : FUNCT_BITS;
-  unsigned immlo : IMMLO_BITS;
-  unsigned rs2 : XPRID_BITS;
-  unsigned rs1 : XPRID_BITS;
-  signed immhi : IMM_BITS-IMMLO_BITS;
-};
-
-struct jtype_t
-{
-  unsigned jump_opcode : OPCODE_BITS;
-  signed target : TARGET_BITS;
-};
-
-struct rtype_t
-{
-  unsigned opcode : OPCODE_BITS;
-  unsigned funct : FUNCT_BITS;
-  unsigned functr : FUNCTR_BITS;
-  unsigned rs2 : XPRID_BITS;
-  unsigned rs1 : XPRID_BITS;
-  unsigned rd : XPRID_BITS;
-};
-
-struct ltype_t
-{
-  unsigned opcode : OPCODE_BITS;
-  unsigned bigimm : BIGIMM_BITS;
-  unsigned rd : XPRID_BITS;
-};
-
-struct ftype_t
-{
-  unsigned opcode : OPCODE_BITS;
-  unsigned ffunct : FFUNCT_BITS;
-  unsigned rm : RM_BITS;
-  unsigned rs3 : FPRID_BITS;
-  unsigned rs2 : FPRID_BITS;
-  unsigned rs1 : FPRID_BITS;
-  unsigned rd  : FPRID_BITS;
-};
-
-union insn_t
-{
-  itype_t itype;
-  jtype_t jtype;
-  rtype_t rtype;
-  btype_t btype;
-  ltype_t ltype;
-  ftype_t ftype;
-  uint32_t bits;
-};
+#define insn_length(x) \
+  (((x) & 0x03) < 0x03 ? 2 : \
+   ((x) & 0x1f) < 0x1f ? 4 : \
+   ((x) & 0x3f) < 0x3f ? 6 : \
+   8)
+#define MAX_INSN_LENGTH 8
+#define PC_ALIGN 2
 
-#include <stdio.h>
-template <class T>
-class write_port_t
+typedef uint64_t insn_bits_t;
+class insn_t
 {
 public:
-  write_port_t(T& _t) : t(_t) {}
-  T& operator = (const T& rhs)
-  {
-    return t = rhs;
-  }
-  operator T()
-  {
-    return t;
-  }
+  insn_t() = default;
+  insn_t(insn_bits_t bits) : b(bits) {}
+  insn_bits_t bits() { return b; }
+  int length() { return insn_length(b); }
+  int64_t i_imm() { return int64_t(b) >> 20; }
+  int64_t s_imm() { return x(7, 5) + (xs(25, 7) << 5); }
+  int64_t sb_imm() { return (x(8, 4) << 1) + (x(25,6) << 5) + (x(7,1) << 11) + (imm_sign() << 12); }
+  int64_t u_imm() { return int64_t(b) >> 12 << 12; }
+  int64_t uj_imm() { return (x(21, 10) << 1) + (x(20, 1) << 11) + (x(12, 8) << 12) + (imm_sign() << 20); }
+  uint64_t rd() { return x(7, 5); }
+  uint64_t rs1() { return x(15, 5); }
+  uint64_t rs2() { return x(20, 5); }
+  uint64_t rs3() { return x(27, 5); }
+  uint64_t rm() { return x(12, 3); }
+  uint64_t csr() { return x(20, 12); }
+
+  int64_t rvc_imm() { return x(2, 5) + (xs(12, 1) << 5); }
+  int64_t rvc_zimm() { return x(2, 5) + (x(12, 1) << 5); }
+  int64_t rvc_addi4spn_imm() { return (x(6, 1) << 2) + (x(5, 1) << 3) + (x(11, 2) << 4) + (x(7, 4) << 6); }
+  int64_t rvc_addi16sp_imm() { return (x(6, 1) << 4) + (x(2, 1) << 5) + (x(5, 1) << 6) + (x(3, 2) << 7) + (xs(12, 1) << 9); }
+  int64_t rvc_lwsp_imm() { return (x(4, 3) << 2) + (x(12, 1) << 5) + (x(2, 2) << 6); }
+  int64_t rvc_ldsp_imm() { return (x(5, 2) << 3) + (x(12, 1) << 5) + (x(2, 3) << 6); }
+  int64_t rvc_swsp_imm() { return (x(9, 4) << 2) + (x(7, 2) << 6); }
+  int64_t rvc_sdsp_imm() { return (x(10, 3) << 3) + (x(7, 3) << 6); }
+  int64_t rvc_lw_imm() { return (x(6, 1) << 2) + (x(10, 3) << 3) + (x(5, 1) << 6); }
+  int64_t rvc_ld_imm() { return (x(10, 3) << 3) + (x(5, 2) << 6); }
+  int64_t rvc_j_imm() { return (x(3, 3) << 1) + (x(11, 1) << 4) + (x(2, 1) << 5) + (x(7, 1) << 6) + (x(6, 1) << 7) + (x(9, 2) << 8) + (x(8, 1) << 10) + (xs(12, 1) << 11); }
+  int64_t rvc_b_imm() { return (x(3, 2) << 1) + (x(10, 2) << 3) + (x(2, 1) << 5) + (x(5, 2) << 6) + (xs(12, 1) << 8); }
+  int64_t rvc_simm3() { return x(10, 3); }
+  uint64_t rvc_rd() { return rd(); }
+  uint64_t rvc_rs1() { return rd(); }
+  uint64_t rvc_rs2() { return x(2, 5); }
+  uint64_t rvc_rs1s() { return 8 + x(7, 3); }
+  uint64_t rvc_rs2s() { return 8 + x(2, 3); }
 private:
-  T& t;
+  insn_bits_t b;
+  uint64_t x(int lo, int len) { return (b >> lo) & ((insn_bits_t(1) << len)-1); }
+  uint64_t xs(int lo, int len) { return int64_t(b) << (64-lo-len) >> (64-len); }
+  uint64_t imm_sign() { return xs(63, 1); }
 };
+
 template <class T, size_t N, bool zero_reg>
 class regfile_t
 {
 public:
-  void reset()
-  {
-    memset(data, 0, sizeof(data));
-  }
-  write_port_t<T> write_port(size_t i)
+  void write(size_t i, T value)
   {
-    return write_port_t<T>(data[i]);
+    if (!zero_reg || i != 0)
+      data[i] = value;
   }
   const T& operator [] (size_t i) const
   {
-    if (zero_reg)
-      const_cast<T&>(data[0]) = 0;
     return data[i];
   }
 private:
   T data[N];
 };
 
-#define throw_illegal_instruction \
-  ({ if (utmode) throw trap_vector_illegal_instruction; \
-     else throw trap_illegal_instruction; })
-
 // helpful macros, etc
-#define RS1 XPR[insn.rtype.rs1]
-#define RS2 XPR[insn.rtype.rs2]
-#define RD XPR.write_port(insn.rtype.rd)
-#define RA XPR.write_port(1)
-#define FRS1 FPR[insn.ftype.rs1]
-#define FRS2 FPR[insn.ftype.rs2]
-#define FRS3 FPR[insn.ftype.rs3]
-#define FRD FPR.write_port(insn.ftype.rd)
-#define BIGIMM insn.ltype.bigimm
-#define SIMM insn.itype.imm12
-#define BIMM ((signed)insn.btype.immlo | (insn.btype.immhi << IMMLO_BITS))
-#define SHAMT (insn.itype.imm12 & 0x3F)
-#define SHAMTW (insn.itype.imm12 & 0x1F)
-#define TARGET insn.jtype.target
-#define BRANCH_TARGET (pc + (BIMM << BRANCH_ALIGN_BITS))
-#define JUMP_TARGET (pc + (TARGET << JUMP_ALIGN_BITS))
-#define RM ({ int rm = insn.ftype.rm; \
-              if(rm == 7) rm = (fsr & FSR_RD) >> FSR_RD_SHIFT; \
-              if(rm > 4) throw_illegal_instruction; \
-              rm; })
-
-#define xpr64 (xprlen == 64)
-
-#define require_supervisor if(unlikely(!(sr & SR_S))) throw trap_privileged_instruction
-#define require_xpr64 if(unlikely(!xpr64)) throw_illegal_instruction
-#define require_xpr32 if(unlikely(xpr64)) throw_illegal_instruction
-#ifndef RISCV_ENABLE_FPU
-# define require_fp throw trap_illegal_instruction
-#else
-# define require_fp if(unlikely(!(sr & SR_EF))) throw trap_fp_disabled
-#endif
-#ifndef RISCV_ENABLE_VEC
-# define require_vector throw trap_illegal_instruction
+#define MMU (*p->get_mmu())
+#define STATE (*p->get_state())
+#define READ_REG(reg) STATE.XPR[reg]
+#define READ_FREG(reg) STATE.FPR[reg]
+#define RS1 READ_REG(insn.rs1())
+#define RS2 READ_REG(insn.rs2())
+#define WRITE_RD(value) WRITE_REG(insn.rd(), value)
+
+#ifndef RISCV_ENABLE_COMMITLOG
+# define WRITE_REG(reg, value) STATE.XPR.write(reg, value)
+# define WRITE_FREG(reg, value) DO_WRITE_FREG(reg, freg(value))
 #else
-# define require_vector \
-  ({ if(!(sr & SR_EV)) throw trap_vector_disabled; \
-    else if (!utmode && (vecbanks_count < 3)) throw trap_vector_bank; \
+# define WRITE_REG(reg, value) ({ \
+    reg_t wdata = (value); /* value may have side effects */ \
+    STATE.log_reg_write = (commit_log_reg_t){(reg) << 1, {wdata, 0}}; \
+    STATE.XPR.write(reg, wdata); \
+  })
+# define WRITE_FREG(reg, value) ({ \
+    freg_t wdata = freg(value); /* value may have side effects */ \
+    STATE.log_reg_write = (commit_log_reg_t){((reg) << 1) | 1, wdata}; \
+    DO_WRITE_FREG(reg, wdata); \
   })
 #endif
 
-#define cmp_trunc(reg) (reg_t(reg) << (64-xprlen))
-#define set_fp_exceptions ({ set_fsr(fsr | \
-                               (softfloat_exceptionFlags << FSR_AEXC_SHIFT)); \
+// RVC macros
+#define WRITE_RVC_RS1S(value) WRITE_REG(insn.rvc_rs1s(), value)
+#define WRITE_RVC_RS2S(value) WRITE_REG(insn.rvc_rs2s(), value)
+#define WRITE_RVC_FRS2S(value) WRITE_FREG(insn.rvc_rs2s(), value)
+#define RVC_RS1 READ_REG(insn.rvc_rs1())
+#define RVC_RS2 READ_REG(insn.rvc_rs2())
+#define RVC_RS1S READ_REG(insn.rvc_rs1s())
+#define RVC_RS2S READ_REG(insn.rvc_rs2s())
+#define RVC_FRS2 READ_FREG(insn.rvc_rs2())
+#define RVC_FRS2S READ_FREG(insn.rvc_rs2s())
+#define RVC_SP READ_REG(X_SP)
+
+// FPU macros
+#define FRS1 READ_FREG(insn.rs1())
+#define FRS2 READ_FREG(insn.rs2())
+#define FRS3 READ_FREG(insn.rs3())
+#define dirty_fp_state (STATE.mstatus |= MSTATUS_FS | (xlen == 64 ? MSTATUS64_SD : MSTATUS32_SD))
+#define dirty_ext_state (STATE.mstatus |= MSTATUS_XS | (xlen == 64 ? MSTATUS64_SD : MSTATUS32_SD))
+#define DO_WRITE_FREG(reg, value) (STATE.FPR.write(reg, value), dirty_fp_state)
+#define WRITE_FRD(value) WRITE_FREG(insn.rd(), value)
+ 
+#define SHAMT (insn.i_imm() & 0x3F)
+#define BRANCH_TARGET (pc + insn.sb_imm())
+#define JUMP_TARGET (pc + insn.uj_imm())
+#define RM ({ int rm = insn.rm(); \
+              if(rm == 7) rm = STATE.frm; \
+              if(rm > 4) throw trap_illegal_instruction(0); \
+              rm; })
+
+#define get_field(reg, mask) (((reg) & (decltype(reg))(mask)) / ((mask) & ~((mask) << 1)))
+#define set_field(reg, mask, val) (((reg) & ~(decltype(reg))(mask)) | (((decltype(reg))(val) * ((mask) & ~((mask) << 1))) & (decltype(reg))(mask)))
+
+#define require(x) if (unlikely(!(x))) throw trap_illegal_instruction(0)
+#define require_privilege(p) require(STATE.prv >= (p))
+#define require_rv64 require(xlen == 64)
+#define require_rv32 require(xlen == 32)
+#define require_extension(s) require(p->supports_extension(s))
+#define require_fp require((STATE.mstatus & MSTATUS_FS) != 0)
+#define require_accelerator require((STATE.mstatus & MSTATUS_XS) != 0)
+
+#define set_fp_exceptions ({ if (softfloat_exceptionFlags) { \
+                               dirty_fp_state; \
+                               STATE.fflags |= softfloat_exceptionFlags; \
+                             } \
                              softfloat_exceptionFlags = 0; })
 
 #define sext32(x) ((sreg_t)(int32_t)(x))
 #define zext32(x) ((reg_t)(uint32_t)(x))
-#define sext_xprlen(x) (((sreg_t)(x) << (64-xprlen)) >> (64-xprlen))
-#define zext_xprlen(x) (((reg_t)(x) << (64-xprlen)) >> (64-xprlen))
-
-// RVC stuff
-
-#define INSN_IS_RVC(x) (((x) & 0x3) < 0x3)
-#define insn_length(x) (INSN_IS_RVC(x) ? 2 : 4)
-#define require_rvc if(!(sr & SR_EC)) throw_illegal_instruction
-
-#define CRD_REGNUM ((insn.bits >> 5) & 0x1f)
-#define CRD XPR.write_port(CRD_REGNUM)
-#define CRS1 XPR[(insn.bits >> 10) & 0x1f]
-#define CRS2 XPR[(insn.bits >> 5) & 0x1f]
-#define CIMM6 ((int32_t)((insn.bits >> 10) & 0x3f) << 26 >> 26)
-#define CIMM5U ((insn.bits >> 5) & 0x1f)
-#define CIMM5 ((int32_t)CIMM5U << 27 >> 27)
-#define CIMM10 ((int32_t)((insn.bits >> 5) & 0x3ff) << 22 >> 22)
-#define CBRANCH_TARGET (pc + (CIMM5 << BRANCH_ALIGN_BITS))
-#define CJUMP_TARGET (pc + (CIMM10 << JUMP_ALIGN_BITS))
-
-static const int rvc_rs1_regmap[8] = { 20, 21, 2, 3, 4, 5, 6, 7 };
-#define rvc_rd_regmap rvc_rs1_regmap
-#define rvc_rs2b_regmap rvc_rs1_regmap
-static const int rvc_rs2_regmap[8] = { 20, 21, 2, 3, 4, 5, 6, 0 };
-#define CRDS XPR.write_port(rvc_rd_regmap[(insn.bits >> 13) & 0x7])
-#define FCRDS FPR.write_port(rvc_rd_regmap[(insn.bits >> 13) & 0x7])
-#define CRS1S XPR[rvc_rs1_regmap[(insn.bits >> 10) & 0x7]]
-#define CRS2S XPR[rvc_rs2_regmap[(insn.bits >> 13) & 0x7]]
-#define CRS2BS XPR[rvc_rs2b_regmap[(insn.bits >> 5) & 0x7]]
-#define FCRS2S FPR[rvc_rs2_regmap[(insn.bits >> 13) & 0x7]]
-
-// vector stuff
-#define VL vl
-
-#define UT_RS1(idx) uts[idx]->XPR[insn.rtype.rs1]
-#define UT_RS2(idx) uts[idx]->XPR[insn.rtype.rs2]
-#define UT_RD(idx) uts[idx]->XPR.write_port(insn.rtype.rd)
-#define UT_RA(idx) uts[idx]->XPR.write_port(1)
-#define UT_FRS1(idx) uts[idx]->FPR[insn.ftype.rs1]
-#define UT_FRS2(idx) uts[idx]->FPR[insn.ftype.rs2]
-#define UT_FRS3(idx) uts[idx]->FPR[insn.ftype.rs3]
-#define UT_FRD(idx) uts[idx]->FPR.write_port(insn.ftype.rd)
-#define UT_RM(idx) ((insn.ftype.rm != 7) ? insn.ftype.rm : \
-              ((uts[idx]->fsr & FSR_RD) >> FSR_RD_SHIFT))
-
-#define UT_LOOP_START for (int i=0;i<VL; i++) {
-#define UT_LOOP_END }
-#define UT_LOOP_RS1 UT_RS1(i)
-#define UT_LOOP_RS2 UT_RS2(i)
-#define UT_LOOP_RD UT_RD(i)
-#define UT_LOOP_RA UT_RA(i)
-#define UT_LOOP_FRS1 UT_FRS1(i)
-#define UT_LOOP_FRS2 UT_FRS2(i)
-#define UT_LOOP_FRS3 UT_FRS3(i)
-#define UT_LOOP_FRD UT_FRD(i)
-#define UT_LOOP_RM UT_RM(i)
-
-#define VEC_LOAD(dst, func, inc) \
-  reg_t addr = RS1; \
-  UT_LOOP_START \
-    UT_LOOP_##dst = mmu.func(addr); \
-    addr += inc; \
-  UT_LOOP_END
-
-#define VEC_STORE(src, func, inc) \
-  reg_t addr = RS1; \
-  UT_LOOP_START \
-    mmu.func(addr, UT_LOOP_##src); \
-    addr += inc; \
-  UT_LOOP_END
-
-enum vt_command_t
+#define sext_xlen(x) (((sreg_t)(x) << (64-xlen)) >> (64-xlen))
+#define zext_xlen(x) (((reg_t)(x) << (64-xlen)) >> (64-xlen))
+
+#define set_pc(x) \
+  do { p->check_pc_alignment(x); \
+       npc = sext_xlen(x); \
+     } while(0)
+
+#define set_pc_and_serialize(x) \
+  do { reg_t __npc = (x); \
+       npc = PC_SERIALIZE_AFTER; \
+       STATE.pc = __npc; \
+     } while(0)
+
+#define serialize() set_pc_and_serialize(npc)
+
+/* Sentinel PC values to serialize simulator pipeline */
+#define PC_SERIALIZE_BEFORE 3
+#define PC_SERIALIZE_AFTER 5
+#define invalid_pc(pc) ((pc) & 1)
+
+/* Convenience wrappers to simplify softfloat code sequences */
+#define isBoxedF32(r) (isBoxedF64(r) && ((uint32_t)((r.v[0] >> 32) + 1) == 0))
+#define unboxF32(r) (isBoxedF32(r) ? (uint32_t)r.v[0] : defaultNaNF32UI)
+#define isBoxedF64(r) ((r.v[1] + 1) == 0)
+#define unboxF64(r) (isBoxedF64(r) ? r.v[0] : defaultNaNF64UI)
+typedef float128_t freg_t;
+inline float32_t f32(uint32_t v) { return { v }; }
+inline float64_t f64(uint64_t v) { return { v }; }
+inline float32_t f32(freg_t r) { return f32(unboxF32(r)); }
+inline float64_t f64(freg_t r) { return f64(unboxF64(r)); }
+inline float128_t f128(freg_t r) { return r; }
+inline freg_t freg(float32_t f) { return { ((uint64_t)-1 << 32) | f.v, (uint64_t)-1 }; }
+inline freg_t freg(float64_t f) { return { f.v, (uint64_t)-1 }; }
+inline freg_t freg(float128_t f) { return f; }
+#define F32_SIGN ((uint32_t)1 << 31)
+#define F64_SIGN ((uint64_t)1 << 63)
+#define fsgnj32(a, b, n, x) \
+  f32((f32(a).v & ~F32_SIGN) | ((((x) ? f32(a).v : (n) ? F32_SIGN : 0) ^ f32(b).v) & F32_SIGN))
+#define fsgnj64(a, b, n, x) \
+  f64((f64(a).v & ~F64_SIGN) | ((((x) ? f64(a).v : (n) ? F64_SIGN : 0) ^ f64(b).v) & F64_SIGN))
+
+#define isNaNF128(x) isNaNF128UI(x.v[1], x.v[0])
+inline float128_t defaultNaNF128()
 {
-  vt_command_stop,
-};
+  float128_t nan;
+  nan.v[1] = defaultNaNF128UI64;
+  nan.v[0] = defaultNaNF128UI0;
+  return nan;
+}
+inline freg_t fsgnj128(freg_t a, freg_t b, bool n, bool x)
+{
+  a.v[1] = (a.v[1] & ~F64_SIGN) | (((x ? a.v[1] : n ? F64_SIGN : 0) ^ b.v[1]) & F64_SIGN);
+  return a;
+}
+inline freg_t f128_negate(freg_t a)
+{
+  a.v[1] ^= F64_SIGN;
+  return a;
+}
+
+#define validate_csr(which, write) ({ \
+  if (!STATE.serialized) return PC_SERIALIZE_BEFORE; \
+  STATE.serialized = false; \
+  unsigned csr_priv = get_field((which), 0x300); \
+  unsigned csr_read_only = get_field((which), 0xC00) == 3; \
+  if (((write) && csr_read_only) || STATE.prv < csr_priv) \
+    throw trap_illegal_instruction(0); \
+  (which); })
+
+// Seems that 0x0 doesn't work.
+#define DEBUG_START             0x100
+#define DEBUG_END                 (0x1000 - 1)
 
 #endif