- Add MASM assembler files for MSVC x64 projects to support vm_x86 in x64 mode

- Clean up ftol()/snapvector() mess
- Make use of SSE instructions for ftol()/snapvector() if available
- move ftol/snapvector pure assembler to inline assembler, this will add x86_64 and improve support for different calling conventions
- Set FPU control word at program startup to get consistent behaviour on all platforms
This commit is contained in:
Thilo Schulz 2011-06-13 09:56:39 +00:00
parent 471182cba0
commit 03201aff22
22 changed files with 540 additions and 460 deletions

90
code/asm/ftola.asm Normal file
View file

@ -0,0 +1,90 @@
; ===========================================================================
; Copyright (C) 2011 Thilo Schulz <thilo@tjps.eu>
;
; This file is part of Quake III Arena source code.
;
; Quake III Arena source code is free software; you can redistribute it
; and/or modify it under the terms of the GNU General Public License as
; published by the Free Software Foundation; either version 2 of the License,
; or (at your option) any later version.
;
; Quake III Arena source code is distributed in the hope that it will be
; useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
; GNU General Public License for more details.
;
; You should have received a copy of the GNU General Public License
; along with Quake III Arena source code; if not, write to the Free Software
; Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
; ===========================================================================
; MASM ftol conversion functions using SSE or FPU
; assume __cdecl calling convention is being used for x86, __fastcall for x64
IFNDEF idx64
.model flat, c
ENDIF
; .data
; ifndef idx64
; fpucw WORD 0F7Fh
; endif
.code
IFDEF idx64
; qftol using SSE
qftolsse PROC
cvttss2si eax, xmm0
ret
qftolsse ENDP
qvmftolsse PROC
movss xmm0, dword ptr [rdi + rbx * 4]
cvttss2si eax, xmm0
ret
qvmftolsse ENDP
ELSE
; qftol using FPU
qftolx87m macro src
; not necessary, fpucw is set with _controlfp at startup
; sub esp, 2
; fnstcw word ptr [esp]
; fldcw fpucw
fld dword ptr src
fistp dword ptr src
; fldcw [esp]
mov eax, src
; add esp, 2
ret
endm
qftolx87 PROC
; need this line when storing FPU control word on stack
; qftolx87m [esp + 6]
qftolx87m [esp + 4]
qftolx87 ENDP
qvmftolx87 PROC
qftolx87m [edi + ebx * 4]
qvmftolx87 ENDP
; qftol using SSE
qftolsse PROC
movss xmm0, dword ptr [esp + 4]
cvttss2si eax, xmm0
ret
qftolsse ENDP
qvmftolsse PROC
movss xmm0, dword ptr [edi + ebx * 4]
cvttss2si eax, xmm0
ret
qvmftolsse ENDP
ENDIF
end

View file

@ -1,157 +0,0 @@
/*
===========================================================================
Copyright (C) 1999-2005 Id Software, Inc.
This file is part of Quake III Arena source code.
Quake III Arena source code is free software; you can redistribute it
and/or modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the License,
or (at your option) any later version.
Quake III Arena source code is distributed in the hope that it will be
useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with Quake III Arena source code; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
===========================================================================
*/
//
// qftol -- fast floating point to long conversion.
//
// 23/09/05 Ported to gas by intel2gas, best supporting actor Tim Angus
// <tim@ngus.net>
#include "qasm.h"
.data
temp: .single 0.0
fpucw: .long 0
// Precision Control Field , 2 bits / 0x0300
// PC24 0x0000 Single precision (24 bits).
// PC53 0x0200 Double precision (53 bits).
// PC64 0x0300 Extended precision (64 bits).
// Rounding Control Field, 2 bits / 0x0C00
// RCN 0x0000 Rounding to nearest (even).
// RCD 0x0400 Rounding down (directed, minus).
// RCU 0x0800 Rounding up (directed plus).
// RC0 0x0C00 Rounding towards zero (chop mode).
// rounding towards nearest (even)
cw027F: .long 0x027F
cw037F: .long 0x037F
// rounding towards zero (chop mode)
cw0E7F: .long 0x0E7F
cw0F7F: .long 0x0F7F
.text
//
// int qftol( void ) - default control word
//
.globl C(qftol)
C(qftol):
fistpl temp
movl temp,%eax
ret
//
// int qftol027F( void ) - DirectX FPU
//
.globl C(qftol027F)
C(qftol027F):
fnstcw fpucw
fldcw cw027F
fistpl temp
fldcw fpucw
movl temp,%eax
ret
//
// int qftol037F( void ) - Linux FPU
//
.globl C(qftol037F)
C(qftol037F):
fnstcw fpucw
fldcw cw037F
fistpl temp
fldcw fpucw
movl temp,%eax
ret
//
// int qftol0F7F( void ) - ANSI
//
.globl C(qftol0F7F)
C(qftol0F7F):
fnstcw fpucw
fldcw cw0F7F
fistpl temp
fldcw fpucw
movl temp,%eax
ret
//
// int qftol0E7F( void )
//
.globl C(qftol0E7F)
C(qftol0E7F):
fnstcw fpucw
fldcw cw0E7F
fistpl temp
fldcw fpucw
movl temp,%eax
ret
//
// long Q_ftol( float q )
//
.globl C(Q_ftol)
C(Q_ftol):
flds 4(%esp)
fistpl temp
movl temp,%eax
ret
//
// long qftol0F7F( float q ) - Linux FPU
//
.globl C(Q_ftol0F7F)
C(Q_ftol0F7F):
fnstcw fpucw
flds 4(%esp)
fldcw cw0F7F
fistpl temp
fldcw fpucw
movl temp,%eax
ret

107
code/asm/snapvector.asm Normal file
View file

@ -0,0 +1,107 @@
; ===========================================================================
; Copyright (C) 2011 Thilo Schulz <thilo@tjps.eu>
;
; This file is part of Quake III Arena source code.
;
; Quake III Arena source code is free software; you can redistribute it
; and/or modify it under the terms of the GNU General Public License as
; published by the Free Software Foundation; either version 2 of the License,
; or (at your option) any later version.
;
; Quake III Arena source code is distributed in the hope that it will be
; useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
; GNU General Public License for more details.
;
; You should have received a copy of the GNU General Public License
; along with Quake III Arena source code; if not, write to the Free Software
; Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
; ===========================================================================
; MASM version of snapvector conversion function using SSE or FPU
; assume __cdecl calling convention is being used for x86, __fastcall for x64
;
; function prototype:
; void qsnapvector(vec3_t vec)
IFNDEF idx64
.model flat, c
ENDIF
.data
ALIGN 16
ssemask DWORD 0FFFFFFFFh, 0FFFFFFFFh, 0FFFFFFFFh, 00000000h
ssecw DWORD 00001F80h
IFNDEF idx64
fpucw WORD 037Fh
ENDIF
.code
IFDEF idx64
; qsnapvector using SSE
qsnapvectorsse PROC
sub rsp, 4
stmxcsr [rsp] ; save SSE control word
ldmxcsr ssecw ; set to round nearest
push rdi
mov rdi, rcx ; maskmovdqu uses rdi as implicit memory operand
movaps xmm1, ssemask ; initialize the mask register for maskmovdqu
movups xmm0, [rdi] ; here is stored our vector. Read 4 values in one go
cvtps2dq xmm0, xmm0 ; convert 4 single fp to int
cvtdq2ps xmm0, xmm0 ; convert 4 int to single fp
maskmovdqu xmm0, xmm1 ; write 3 values back to memory
pop rdi
ldmxcsr [rsp] ; restore sse control word to old value
add rsp, 4
ret
qsnapvectorsse ENDP
ELSE
qsnapvectorsse PROC
sub esp, 4
stmxcsr [esp] ; save SSE control word
ldmxcsr ssecw ; set to round nearest
push edi
mov edi, dword ptr 12[esp] ; maskmovdqu uses edi as implicit memory operand
movaps xmm1, ssemask ; initialize the mask register for maskmovdqu
movups xmm0, [edi] ; here is stored our vector. Read 4 values in one go
cvtps2dq xmm0, xmm0 ; convert 4 single fp to int
cvtdq2ps xmm0, xmm0 ; convert 4 int to single fp
maskmovdqu xmm0, xmm1 ; write 3 values back to memory
pop edi
ldmxcsr [esp] ; restore sse control word to old value
add esp, 4
ret
qsnapvectorsse ENDP
qroundx87 macro src
fld dword ptr src
fistp dword ptr src
fild dword ptr src
fstp dword ptr src
endm
qsnapvectorx87 PROC
mov eax, dword ptr 4[esp]
sub esp, 2
fnstcw word ptr [esp]
fldcw fpucw
qroundx87 [eax]
qroundx87 4[eax]
qroundx87 8[eax]
fldcw [esp]
add esp, 2
qsnapvectorx87 ENDP
ENDIF
end

View file

@ -1,103 +0,0 @@
/*
===========================================================================
Copyright (C) 1999-2005 Id Software, Inc.
This file is part of Quake III Arena source code.
Quake III Arena source code is free software; you can redistribute it
and/or modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the License,
or (at your option) any later version.
Quake III Arena source code is distributed in the hope that it will be
useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with Quake III Arena source code; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
===========================================================================
*/
//
// Sys_SnapVector NASM code (Andrew Henderson)
// See win32/win_shared.c for the Win32 equivalent
// This code is provided to ensure that the
// rounding behavior (and, if necessary, the
// precision) of DLL and QVM code are identical
// e.g. for network-visible operations.
// See ftol.nasm for operations on a single float,
// as used in compiled VM and DLL code that does
// not use this system trap.
//
// 23/09/05 Ported to gas by intel2gas, best supporting actor Tim Angus
// <tim@ngus.net>
#include "qasm.h"
#if id386
.data
fpucw: .long 0
cw037F: .long 0x037F
.text
// void Sys_SnapVector( float *v )
.globl C(Sys_SnapVector)
C(Sys_SnapVector):
pushl %eax
pushl %ebp
movl %esp,%ebp
fnstcw fpucw
movl 12(%ebp),%eax
fldcw cw037F
flds (%eax)
fistpl (%eax)
fildl (%eax)
fstps (%eax)
flds 4(%eax)
fistpl 4(%eax)
fildl 4(%eax)
fstps 4(%eax)
flds 8(%eax)
fistpl 8(%eax)
fildl 8(%eax)
fstps 8(%eax)
fldcw fpucw
popl %ebp
popl %eax
ret
// void Sys_SnapVectorCW( float *v, unsigned short int cw )
.globl C(Sys_SnapVectorCW)
C(Sys_SnapVectorCW):
pushl %eax
pushl %ebp
movl %esp,%ebp
fnstcw fpucw
movl 12(%ebp),%eax
fldcw 16(%ebp)
flds (%eax)
fistpl (%eax)
fildl (%eax)
fstps (%eax)
flds 4(%eax)
fistpl 4(%eax)
fildl 4(%eax)
fstps 4(%eax)
flds 8(%eax)
fistpl 8(%eax)
fildl 8(%eax)
fstps 8(%eax)
fldcw fpucw
popl %ebp
popl %eax
ret
#endif

76
code/asm/vm_x86_64.asm Normal file
View file

@ -0,0 +1,76 @@
; ===========================================================================
; Copyright (C) 2011 Thilo Schulz <thilo@tjps.eu>
;
; This file is part of Quake III Arena source code.
;
; Quake III Arena source code is free software; you can redistribute it
; and/or modify it under the terms of the GNU General Public License as
; published by the Free Software Foundation; either version 2 of the License,
; or (at your option) any later version.
;
; Quake III Arena source code is distributed in the hope that it will be
; useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
; GNU General Public License for more details.
;
; You should have received a copy of the GNU General Public License
; along with Quake III Arena source code; if not, write to the Free Software
; Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
; ===========================================================================
; Call wrapper for vm_x86 when built with MSVC in 64 bit mode,
; since MSVC does not support inline x64 assembler code anymore.
;
; assumes __fastcall calling convention
DoSyscall PROTO
.code
; Call to static void DoSyscall(int syscallNum, int programStack, int *opStackBase, uint8_t opStackOfs, intptr_t arg)
qsyscall64 PROC
sub rsp, 28h ; after this esp will be aligned to 16 byte boundary
mov qword ptr [rsp + 20h], rcx ; 5th parameter "arg" is passed on stack
mov r9b, bl ; opStackOfs
mov r8, rdi ; opStackBase
mov edx, esi ; programStack
mov ecx, eax ; syscallNum
mov rax, DoSyscall ; store call address of DoSyscall in rax
call rax
add rsp, 28h
ret
qsyscall64 ENDP
; Call to compiled code after setting up the register environment for the VM
; prototype:
; uint8_t qvmcall64(int *programStack, int *opStack, intptr_t *instructionPointers, byte *dataBase);
qvmcall64 PROC
push rsi ; push non-volatile registers to stack
push rdi
push rbx
; need to save pointer in rcx so we can write back the programData value to caller
push rcx
; registers r8 and r9 have correct value already thanx to __fastcall
xor rbx, rbx ; opStackOfs starts out being 0
mov rdi, rdx ; opStack
mov esi, dword ptr [rcx] ; programStack
call qword ptr [r8] ; instructionPointers[0] is also the entry point
pop rcx
mov dword ptr [rcx], esi ; write back the programStack value
mov al, bl ; return opStack offset
pop rbx
pop rdi
pop rsi
ret
qvmcall64 ENDP
end