diff --git a/metamod/src/callback_jit.cpp b/metamod/src/callback_jit.cpp index 8c3f292..3445301 100644 --- a/metamod/src/callback_jit.cpp +++ b/metamod/src/callback_jit.cpp @@ -66,22 +66,27 @@ void CForwardCallbackJIT::naked_main() enum // stack map { - orig_ret = 0, - over_ret = sizeof(int) + /* META GLOBALS BACKUP */ + /* STRING BUFFER */ + over_ret = sizeof(int), + orig_ret = 0 }; auto globals = ebx; - auto mg_backup = m_jitdata->has_ret ? sizeof(int) * 2 /* orig + over */ : 0; - auto framesize = align(mg_backup + sizeof(meta_globals_t) + /* for align */m_jitdata->args_count * sizeof(int), xmmreg_size) - m_jitdata->args_count * sizeof(int); + auto locals_size = m_jitdata->has_ret ? sizeof(int) * 2 /* orig + over */ : 0; + auto framesize = align(locals_size + sizeof(meta_globals_t) + /* for align */m_jitdata->args_count * sizeof(int), xmmreg_size) - m_jitdata->args_count * sizeof(int); if (m_jitdata->has_varargs) { - size_t buf_offset = framesize; + size_t strbuf_offset = locals_size; sub(esp, framesize += align(MAX_STRBUF_LEN, xmmreg_size)); // format varargs lea(edx, dword_ptr[ebp + first_arg_offset + m_jitdata->args_count * sizeof(int)]); // varargs ptr - lea(eax, dword_ptr[esp + buf_offset]); // buf ptr + if (strbuf_offset) + lea(eax, dword_ptr[esp + strbuf_offset]); // buf ptr + else + mov(eax, esp); mov(ecx, size_t(vsnprintf)); push(edx); @@ -94,12 +99,14 @@ void CForwardCallbackJIT::naked_main() else sub(esp, framesize); + size_t mg_backup = framesize - xmmreg_size - sizeof(int); + // setup globals ptr and backup old data mov(globals, size_t(&g_metaGlobals)); - movups(xmm0, xmmword_ptr[globals]); + movaps(xmm0, xmmword_ptr[globals]); mov(eax, dword_ptr[globals + xmmreg_size]); - movups(xmmword_ptr[esp + mg_backup], xmm0); - mov(dword_ptr[esp + mg_backup + xmmreg_size], eax); + movaps(xmmword_ptr[esp + mg_backup + sizeof(int)], xmm0); + mov(dword_ptr[esp + mg_backup], eax); // call metamod's pre hook if present if (m_jitdata->mm_hook && m_jitdata->mm_hook_time == P_PRE) { @@ -255,9 +262,9 @@ void CForwardCallbackJIT::naked_main() call_func(ecx); } - movups(xmm0, xmmword_ptr[esp + mg_backup]); - mov(eax, dword_ptr[esp + mg_backup + xmmreg_size]); - movups(xmmword_ptr[globals], xmm0); + movaps(xmm0, xmmword_ptr[esp + mg_backup + sizeof(int)]); + mov(eax, dword_ptr[esp + mg_backup]); + movaps(xmmword_ptr[globals], xmm0); mov(dword_ptr[globals + xmmreg_size], eax); if (m_jitdata->has_ret) { @@ -276,12 +283,16 @@ void CForwardCallbackJIT::naked_main() void CForwardCallbackJIT::call_func(jitasm::Frontend::Reg32 addr) { const size_t fixed_args_count = m_jitdata->args_count - (m_jitdata->has_varargs ? 1u /* excluding format string */ : 0u); - const size_t strbuf_offset = align((m_jitdata->has_ret ? sizeof(int) * 2 : 0) + sizeof(meta_globals_t) + /* for align */m_jitdata->args_count * sizeof(int), xmmreg_size) - m_jitdata->args_count * sizeof(int); + const size_t strbuf_offset = m_jitdata->has_ret ? sizeof(int) * 2 /* orig + over */ : 0; // push formatted buf instead of format string if (m_jitdata->has_varargs) { - lea(eax, dword_ptr[esp + strbuf_offset]); - push(eax); + if (strbuf_offset) { + lea(eax, dword_ptr[esp + strbuf_offset]); + push(eax); + } + else + push(esp); } // push normal args diff --git a/metamod/src/jitasm.h b/metamod/src/jitasm.h index f74840d..006a225 100644 --- a/metamod/src/jitasm.h +++ b/metamod/src/jitasm.h @@ -321,7 +321,7 @@ namespace detail bool operator==(const Opd& rhs) const { - if ((opdtype_ & O_TYPE_TYPE_MASK) != (rhs.opdtype_ & O_TYPE_TYPE_MASK) || rhs.opdsize_ != opdsize_) {return false;} + if ((opdtype_ & O_TYPE_TYPE_MASK) != (rhs.opdtype_ & O_TYPE_TYPE_MASK) || opdsize_ != opdsize_) {return false;} if (IsReg()) {return reg_ == rhs.reg_ && reg_assignable_ == rhs.reg_assignable_;} if (IsMem()) {return base_ == rhs.base_ && index_ == rhs.index_ && scale_ == rhs.scale_ && disp_ == rhs.disp_ && addrsize_ == rhs.addrsize_;} if (IsImm()) {return imm_ == rhs.imm_;} @@ -1924,7 +1924,6 @@ struct Frontend #ifndef JITASM64 void call(const Reg16& dst) {AppendInstr(I_CALL, 0xFF, E_OPERAND_SIZE_PREFIX, Imm8(2), R(dst));} void call(const Reg32& dst) {AppendInstr(I_CALL, 0xFF, 0, Imm8(2), R(dst));} - void call(const Mem32& dst) {AppendInstr(I_CALL, 0xFF, 0, Imm8(2), R(dst));} // Imm8(2) = register/opcode #else void call(const Reg64& dst) {AppendInstr(I_CALL, 0xFF, 0, Imm8(2), R(dst));} #endif @@ -2267,7 +2266,6 @@ struct Frontend #ifdef JITASM64 void iretq() {AppendInstr(I_IRETQ, 0xCF, E_REXW_PREFIX);} #endif - void jmp(const Mem32& dst) {AppendInstr(I_JMP, 0xFF, 0, Imm8(4), R(dst));} void jmp(const std::string& label_name) {AppendJmp(GetLabelID(label_name));} void ja(const std::string& label_name) {AppendJcc(JCC_A, GetLabelID(label_name));} void jae(const std::string& label_name) {AppendJcc(JCC_AE, GetLabelID(label_name));} @@ -3996,7 +3994,7 @@ struct Frontend void vblendvps(const YmmReg& dst, const YmmReg& src1, const Mem256& src2, const YmmReg& mask) {AppendInstr(I_BLENDVPS, 0x4A, E_VEX_256 | E_VEX_66_0F3A, W(dst), R(src2), R(src1), R(mask));} void vbroadcastss(const XmmReg& dst, const Mem32& src) {AppendInstr(I_VBROADCASTSS, 0x18, E_VEX_128_66_0F38_W0, W(dst), R(src));} void vbroadcastss(const YmmReg& dst, const Mem32& src) {AppendInstr(I_VBROADCASTSS, 0x18, E_VEX_256_66_0F38_W0, W(dst), R(src));} - void vbroadcastsd(const YmmReg& dst, const Mem64& src) {AppendInstr(I_VBROADCASTSD, 0x19, E_VEX_256_66_0F38_W0, W(dst), R(src));} + void vbroadcastsd(const YmmReg& dst, const Mem64 src) {AppendInstr(I_VBROADCASTSD, 0x19, E_VEX_256_66_0F38_W0, W(dst), R(src));} void vbroadcastf128(const YmmReg& dst, const Mem128& src) {AppendInstr(I_VBROADCASTF128, 0x1A, E_VEX_256_66_0F38_W0, W(dst), R(src));} void vcmppd(const XmmReg& dst, const XmmReg& src1, const XmmReg& src2, const Imm8& imm) {AppendInstr(I_CMPPD, 0xC2, E_VEX_128_66_0F_WIG, W(dst), R(src2), R(src1), imm);} void vcmppd(const XmmReg& dst, const XmmReg& src1, const Mem128& src2, const Imm8& imm) {AppendInstr(I_CMPPD, 0xC2, E_VEX_128_66_0F_WIG, W(dst), R(src2), R(src1), imm);} @@ -4038,8 +4036,8 @@ struct Frontend void vcvtps2pd(const XmmReg& dst, const Mem64& src) {AppendInstr(I_CVTPS2PD, 0x5A, E_VEX_128_0F_WIG, W(dst), R(src));} void vcvtps2pd(const YmmReg& dst, const XmmReg& src) {AppendInstr(I_CVTPS2PD, 0x5A, E_VEX_256_0F_WIG, W(dst), R(src));} void vcvtps2pd(const YmmReg& dst, const Mem128& src) {AppendInstr(I_CVTPS2PD, 0x5A, E_VEX_256_0F_WIG, W(dst), R(src));} - void vcvtsd2si(const Reg32& dst, const XmmReg& src) {AppendInstr(I_CVTSD2SI, 0x2D, E_VEX_128 | E_VEX_F2_0F | E_VEX_W0, W(dst), R(src));} - void vcvtsd2si(const Reg32& dst, const Mem64& src) {AppendInstr(I_CVTSD2SI, 0x2D, E_VEX_128 | E_VEX_F2_0F | E_VEX_W0, W(dst), R(src));} + void vcvtsd2si(const Reg32 dst, const XmmReg& src) {AppendInstr(I_CVTSD2SI, 0x2D, E_VEX_128 | E_VEX_F2_0F | E_VEX_W0, W(dst), R(src));} + void vcvtsd2si(const Reg32 dst, const Mem64& src) {AppendInstr(I_CVTSD2SI, 0x2D, E_VEX_128 | E_VEX_F2_0F | E_VEX_W0, W(dst), R(src));} #ifdef JITASM64 void vcvtsd2si(const Reg64 dst, const XmmReg& src) {AppendInstr(I_CVTSD2SI, 0x2D, E_VEX_128 | E_VEX_F2_0F | E_VEX_W1, W(dst), R(src));} void vcvtsd2si(const Reg64 dst, const Mem64& src) {AppendInstr(I_CVTSD2SI, 0x2D, E_VEX_128 | E_VEX_F2_0F | E_VEX_W1, W(dst), R(src));} @@ -6245,6 +6243,7 @@ namespace compiler InstrID instr_id = f.instrs_[instr_idx].GetID(); if (Frontend::IsJump(instr_id) || instr_id == I_RET || instr_id == I_IRET) { // Jump instruction always terminate basic block + BlockList::iterator next_block; if (instr_idx + 1 < cur_block->instr_end) { // Split basic block split(blocks_.begin() + block_idx, instr_idx + 1); diff --git a/metamod/src/metamod.cpp b/metamod/src/metamod.cpp index 12793a5..83fdd7a 100644 --- a/metamod/src/metamod.cpp +++ b/metamod/src/metamod.cpp @@ -16,6 +16,7 @@ option_t g_global_options[] = gamedll_t g_GameDLL; +ALIGN16 meta_globals_t g_metaGlobals; meta_enginefuncs_t g_plugin_engfuncs;