T0b1/fadec
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

decode: Merge zeroreg and vexreg

Browse Source 
There is no instruction that uses an implicit register and an
VEX-encoded register at the same time. Thus, we can merge vexreg and
zeroreg in the instruction descriptor; the zeroreg value will be added
to the vex-operand (which is zero unless set by a VEX prefix).

This also frees 4 descriptor bits for use with AVX-512 (which will
probably need 1-2 additional unused bits, probably from the type).
This commit is contained in:
Alexis Engelke
2022-11-20 15:25:37 +01:00
parent 19b9d82630
commit 1c04ded9b7
2 changed files with 23 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
decode.c
View File

@@ -78,11 +78,9 @@ struct InstrDesc
#define DESC_MODREG_IDX(desc) ((((desc)->operand_indices >> 2) & 3) ^ 3) #define DESC_MODREG_IDX(desc) ((((desc)->operand_indices >> 2) & 3) ^ 3)
#define DESC_HAS_VEXREG(desc) (((desc)->operand_indices & (3 << 4)) != 0) #define DESC_HAS_VEXREG(desc) (((desc)->operand_indices & (3 << 4)) != 0)
#define DESC_VEXREG_IDX(desc) ((((desc)->operand_indices >> 4) & 3) ^ 3) #define DESC_VEXREG_IDX(desc) ((((desc)->operand_indices >> 4) & 3) ^ 3)
#define DESC_HAS_IMPLICIT(desc) (((desc)->operand_indices & (3 << 6)) != 0)
#define DESC_IMPLICIT_IDX(desc) ((((desc)->operand_indices >> 6) & 3) ^ 3)
#define DESC_IMM_CONTROL(desc) (((desc)->operand_indices >> 12) & 0x7) #define DESC_IMM_CONTROL(desc) (((desc)->operand_indices >> 12) & 0x7)
#define DESC_IMM_IDX(desc) ((((desc)->operand_indices >> 8) & 3) ^ 3) #define DESC_IMM_IDX(desc) ((((desc)->operand_indices >> 6) & 3) ^ 3)
#define DESC_IMPLICIT_VAL(desc) (((desc)->operand_indices >> 10) & 1) #define DESC_ZEROREG_VAL(desc) (((desc)->operand_indices >> 10) & 1)
#define DESC_LOCK(desc) (((desc)->operand_indices >> 11) & 1) #define DESC_LOCK(desc) (((desc)->operand_indices >> 11) & 1)
#define DESC_VSIB(desc) (((desc)->operand_indices >> 15) & 1) #define DESC_VSIB(desc) (((desc)->operand_indices >> 15) & 1)
#define DESC_OPSIZE(desc) (((desc)->operand_sizes >> 8) & 3) #define DESC_OPSIZE(desc) (((desc)->operand_sizes >> 8) & 3)
@@ -94,7 +92,6 @@ struct InstrDesc
#define DESC_REGTY_MODRM(desc) (((desc)->reg_types >> 0) & 7) #define DESC_REGTY_MODRM(desc) (((desc)->reg_types >> 0) & 7)
#define DESC_REGTY_MODREG(desc) (((desc)->reg_types >> 3) & 7) #define DESC_REGTY_MODREG(desc) (((desc)->reg_types >> 3) & 7)
#define DESC_REGTY_VEXREG(desc) (((desc)->reg_types >> 6) & 3) #define DESC_REGTY_VEXREG(desc) (((desc)->reg_types >> 6) & 3)
#define DESC_REGTY_ZEROREG(desc) (((desc)->reg_types >> 8) & 3)
int int
fd_decode(const uint8_t* buffer, size_t len_sz, int mode_int, uintptr_t address, fd_decode(const uint8_t* buffer, size_t len_sz, int mode_int, uintptr_t address,
@@ -333,15 +330,6 @@ prefix_end:
goto skip_modrm; goto skip_modrm;
} }
if (UNLIKELY(DESC_HAS_IMPLICIT(desc)))
{
FdOp* operand = &instr->operands[DESC_IMPLICIT_IDX(desc)];
operand->type = FD_OT_REG;
operand->reg = DESC_IMPLICIT_VAL(desc);
unsigned reg_ty = DESC_REGTY_ZEROREG(desc); // GPL VEC FPU
operand->misc = (0461 >> (3 * reg_ty)) & 0x7;
}
if (DESC_HAS_MODREG(desc)) if (DESC_HAS_MODREG(desc))
{ {
FdOp* op_modreg = &instr->operands[DESC_MODREG_IDX(desc)]; FdOp* op_modreg = &instr->operands[DESC_MODREG_IDX(desc)];
@@ -431,14 +419,15 @@ skip_modrm:
if (UNLIKELY(DESC_HAS_VEXREG(desc))) if (UNLIKELY(DESC_HAS_VEXREG(desc)))
{ {
// Without VEX prefix, this encodes an implicit register
FdOp* operand = &instr->operands[DESC_VEXREG_IDX(desc)]; FdOp* operand = &instr->operands[DESC_VEXREG_IDX(desc)];
operand->type = FD_OT_REG; operand->type = FD_OT_REG;
if (mode == DECODE_32) if (mode == DECODE_32)
vex_operand &= 0x7; vex_operand &= 0x7;
operand->reg = vex_operand; operand->reg = vex_operand | DESC_ZEROREG_VAL(desc);
unsigned reg_ty = DESC_REGTY_VEXREG(desc); // GPL VEC MSK unsigned reg_ty = DESC_REGTY_VEXREG(desc); // GPL VEC MSK FPU
operand->misc = (0761 >> (3 * reg_ty)) & 0x7; operand->misc = (04761 >> (3 * reg_ty)) & 0x7;
} }
else if (vex_operand != 0) else if (vex_operand != 0)
{ {

39
parseinstrs.py
View File

@@ -11,9 +11,9 @@ from typing import NamedTuple, FrozenSet, List, Tuple, Union, Optional, ByteStri
INSTR_FLAGS_FIELDS, INSTR_FLAGS_SIZES = zip(*[ INSTR_FLAGS_FIELDS, INSTR_FLAGS_SIZES = zip(*[
("modrm_idx", 2), ("modrm_idx", 2),
("modreg_idx", 2), ("modreg_idx", 2),
("vexreg_idx", 2), ("vexreg_idx", 2), # note: vexreg w/o vex prefix is zeroreg_val
("zeroreg_idx", 2),
("imm_idx", 2), ("imm_idx", 2),
("unused1", 2),
("zeroreg_val", 1), ("zeroreg_val", 1),
("lock", 1), ("lock", 1),
("imm_control", 3), ("imm_control", 3),
@@ -29,8 +29,7 @@ INSTR_FLAGS_FIELDS, INSTR_FLAGS_SIZES = zip(*[
("modrm_ty", 3), ("modrm_ty", 3),
("modreg_ty", 3), ("modreg_ty", 3),
("vexreg_ty", 2), ("vexreg_ty", 2),
("zeroreg_ty", 2), ("unused", 6),
("unused", 4),
("modrm", 1), ("modrm", 1),
("ign66", 1), ("ign66", 1),
][::-1]) ][::-1])
@@ -49,25 +48,25 @@ ENCODINGS = {
"M": InstrFlags(modrm=1, modrm_idx=0^3), "M": InstrFlags(modrm=1, modrm_idx=0^3),
"M1": InstrFlags(modrm=1, modrm_idx=0^3, imm_idx=1^3, imm_control=1), "M1": InstrFlags(modrm=1, modrm_idx=0^3, imm_idx=1^3, imm_control=1),
"MI": InstrFlags(modrm=1, modrm_idx=0^3, imm_idx=1^3, imm_control=4), "MI": InstrFlags(modrm=1, modrm_idx=0^3, imm_idx=1^3, imm_control=4),
"MC": InstrFlags(modrm=1, modrm_idx=0^3, zeroreg_idx=1^3, zeroreg_val=1), "MC": InstrFlags(modrm=1, modrm_idx=0^3, vexreg_idx=1^3, zeroreg_val=1),
"MR": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=1^3), "MR": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=1^3),
"RM": InstrFlags(modrm=1, modrm_idx=1^3, modreg_idx=0^3), "RM": InstrFlags(modrm=1, modrm_idx=1^3, modreg_idx=0^3),
"RMA": InstrFlags(modrm=1, modrm_idx=1^3, modreg_idx=0^3, zeroreg_idx=2^3), "RMA": InstrFlags(modrm=1, modrm_idx=1^3, modreg_idx=0^3, vexreg_idx=2^3),
"MRI": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=1^3, imm_idx=2^3, imm_control=4), "MRI": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=1^3, imm_idx=2^3, imm_control=4),
"RMI": InstrFlags(modrm=1, modrm_idx=1^3, modreg_idx=0^3, imm_idx=2^3, imm_control=4), "RMI": InstrFlags(modrm=1, modrm_idx=1^3, modreg_idx=0^3, imm_idx=2^3, imm_control=4),
"MRC": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=1^3, zeroreg_idx=2^3, zeroreg_val=1), "MRC": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=1^3, vexreg_idx=2^3, zeroreg_val=1),
"AM": InstrFlags(modrm=1, modrm_idx=1^3, zeroreg_idx=0^3), "AM": InstrFlags(modrm=1, modrm_idx=1^3, vexreg_idx=0^3),
"MA": InstrFlags(modrm=1, modrm_idx=0^3, zeroreg_idx=1^3), "MA": InstrFlags(modrm=1, modrm_idx=0^3, vexreg_idx=1^3),
"I": InstrFlags(imm_idx=0^3, imm_control=4), "I": InstrFlags(imm_idx=0^3, imm_control=4),
"IA": InstrFlags(zeroreg_idx=0^3, imm_idx=1^3, imm_control=4), "IA": InstrFlags(vexreg_idx=0^3, imm_idx=1^3, imm_control=4),
"O": InstrFlags(modrm_idx=0^3), "O": InstrFlags(modrm_idx=0^3),
"OI": InstrFlags(modrm_idx=0^3, imm_idx=1^3, imm_control=4), "OI": InstrFlags(modrm_idx=0^3, imm_idx=1^3, imm_control=4),
"OA": InstrFlags(modrm_idx=0^3, zeroreg_idx=1^3), "OA": InstrFlags(modrm_idx=0^3, vexreg_idx=1^3),
"S": InstrFlags(modreg_idx=0^3), # segment register in bits 3,4,5 "S": InstrFlags(modreg_idx=0^3), # segment register in bits 3,4,5
"A": InstrFlags(zeroreg_idx=0^3), "A": InstrFlags(vexreg_idx=0^3),
"D": InstrFlags(imm_idx=0^3, imm_control=6), "D": InstrFlags(imm_idx=0^3, imm_control=6),
"FD": InstrFlags(zeroreg_idx=0^3, imm_idx=1^3, imm_control=2), "FD": InstrFlags(vexreg_idx=0^3, imm_idx=1^3, imm_control=2),
"TD": InstrFlags(zeroreg_idx=1^3, imm_idx=0^3, imm_control=2), "TD": InstrFlags(vexreg_idx=1^3, imm_idx=0^3, imm_control=2),
"RVM": InstrFlags(modrm=1, modrm_idx=2^3, modreg_idx=0^3, vexreg_idx=1^3), "RVM": InstrFlags(modrm=1, modrm_idx=2^3, modreg_idx=0^3, vexreg_idx=1^3),
"RVMI": InstrFlags(modrm=1, modrm_idx=2^3, modreg_idx=0^3, vexreg_idx=1^3, imm_idx=3^3, imm_control=4), "RVMI": InstrFlags(modrm=1, modrm_idx=2^3, modreg_idx=0^3, vexreg_idx=1^3, imm_idx=3^3, imm_control=4),
@@ -77,7 +76,7 @@ ENCODINGS = {
"VMI": InstrFlags(modrm=1, modrm_idx=1^3, vexreg_idx=0^3, imm_idx=2^3, imm_control=4), "VMI": InstrFlags(modrm=1, modrm_idx=1^3, vexreg_idx=0^3, imm_idx=2^3, imm_control=4),
"MVR": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=2^3, vexreg_idx=1^3), "MVR": InstrFlags(modrm=1, modrm_idx=0^3, modreg_idx=2^3, vexreg_idx=1^3),
} }
ENCODING_OPTYS = ["modrm", "modreg", "vexreg", "zeroreg", "imm"] ENCODING_OPTYS = ["modrm", "modreg", "vexreg", "imm"]
ENCODING_OPORDER = { enc: sorted(ENCODING_OPTYS, key=lambda ty: getattr(ENCODINGS[enc], ty+"_idx")^3) for enc in ENCODINGS} ENCODING_OPORDER = { enc: sorted(ENCODING_OPTYS, key=lambda ty: getattr(ENCODINGS[enc], ty+"_idx")^3) for enc in ENCODINGS}
OPKIND_CANONICALIZE = { OPKIND_CANONICALIZE = {
@@ -164,8 +163,7 @@ class InstrDesc(NamedTuple):
OPKIND_REGTYS_MODRM = { "GP": 0, "XMM": 1, "MMX": 4, "FPU": 5, "MASK": 6, } OPKIND_REGTYS_MODRM = { "GP": 0, "XMM": 1, "MMX": 4, "FPU": 5, "MASK": 6, }
OPKIND_REGTYS_MODREG = { "GP": 0, "XMM": 1, "MASK": 2, "MMX": 4, "SEG": 5, OPKIND_REGTYS_MODREG = { "GP": 0, "XMM": 1, "MASK": 2, "MMX": 4, "SEG": 5,
"CR": 0, "DR": 0 } # CR/DR handled in code. "CR": 0, "DR": 0 } # CR/DR handled in code.
OPKIND_REGTYS_VEXREG = { "GP": 0, "XMM": 1, "MASK": 2 } OPKIND_REGTYS_VEXREG = { "GP": 0, "XMM": 1, "MASK": 2, "FPU": 3 } # also zeroreg
OPKIND_REGTYS_ZEROREG = { "GP": 0, "XMM": 1, "FPU": 2 }
OPKIND_REGTYS_ENC = {"SEG": 3, "FPU": 4, "MMX": 5, "XMM": 6, "BND": 8, OPKIND_REGTYS_ENC = {"SEG": 3, "FPU": 4, "MMX": 5, "XMM": 6, "BND": 8,
"CR": 9, "DR": 10} "CR": 9, "DR": 10}
OPKIND_SIZES = { OPKIND_SIZES = {
@@ -203,7 +201,6 @@ class InstrDesc(NamedTuple):
if flags.modrm_idx: optypes[flags.modrm_idx^3] = "rM"[flags.modrm] if flags.modrm_idx: optypes[flags.modrm_idx^3] = "rM"[flags.modrm]
if flags.modreg_idx: optypes[flags.modreg_idx^3] = "r" if flags.modreg_idx: optypes[flags.modreg_idx^3] = "r"
if flags.vexreg_idx: optypes[flags.vexreg_idx^3] = "r" if flags.vexreg_idx: optypes[flags.vexreg_idx^3] = "r"
if flags.zeroreg_idx: optypes[flags.zeroreg_idx^3] = "r"
if flags.imm_control: optypes[flags.imm_idx^3] = " iariioo"[flags.imm_control] if flags.imm_control: optypes[flags.imm_idx^3] = " iariioo"[flags.imm_control]
return "".join(optypes) return "".join(optypes)
@@ -252,8 +249,6 @@ class InstrDesc(NamedTuple):
extraflags["modreg_ty"] = self.OPKIND_REGTYS_MODREG[opkind.kind] extraflags["modreg_ty"] = self.OPKIND_REGTYS_MODREG[opkind.kind]
elif opname == "vexreg": elif opname == "vexreg":
extraflags["vexreg_ty"] = self.OPKIND_REGTYS_VEXREG[opkind.kind] extraflags["vexreg_ty"] = self.OPKIND_REGTYS_VEXREG[opkind.kind]
elif opname == "zeroreg":
extraflags["zeroreg_ty"] = self.OPKIND_REGTYS_ZEROREG[opkind.kind]
else: else:
if opkind.kind not in ("IMM", "MEM", "XMM"): if opkind.kind not in ("IMM", "MEM", "XMM"):
raise Exception("invalid regty for op 3, must be VEC") raise Exception("invalid regty for op 3, must be VEC")
@@ -769,8 +764,8 @@ def encode2_table(entries, args):
# XCHG eax, eax must not be encoded as 90 -- that'd be NOP. # XCHG eax, eax must not be encoded as 90 -- that'd be NOP.
code += f" if (op_reg_idx(op0)==0&&op_reg_idx(op1)==0) goto next{i};\n" code += f" if (op_reg_idx(op0)==0&&op_reg_idx(op1)==0) goto next{i};\n"
neednext = True neednext = True
if flags.zeroreg_idx: if flags.vexreg_idx and not opcode.vex: # vexreg w/o vex is zeroreg
code += f" if (op_reg_idx(op{flags.zeroreg_idx^3})!={flags.zeroreg_val}) goto next{i};\n" code += f" if (op_reg_idx(op{flags.vexreg_idx^3})!={flags.zeroreg_val}) goto next{i};\n"
neednext = True neednext = True
if flags.imm_control: if flags.imm_control:
if flags.imm_control != 3: if flags.imm_control != 3: