claunia/86Box
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Added the IBM 5161 ISA expansion for PC and XT;

Browse Source 
Cleaned up the parallel port emulation, added IRQ support, and made enabling/disabling per port;
Added the Award 430NX and the Intel Classic/PCI (Alfredo, 420TX);
Finished the 586MC1;
Added 8087 emulation;
Moved Cyrix 6x86'es to the Dev branch;
Sanitized/cleaned up memregs.c/h and intel.c/h;
Split the chipsets from machines and sanitized Port 92 emulation;
Added support for the 15bpp mode to the Compaq ATI 28800;
Moved the MR 386DX and 486 machines to the Dev branch;
Ported the new dynamic recompiler from PCem, but it remains in Dev branch until after v2.00;
Ported the new timer code from PCem;
Cleaned up the CPU table of unused stuff and better optimized its structure;
Ported the Open-XT and Open-AT from VARCem, the Open-AT is in the Dev branch;
Ported the XT MFM controller rewrite and adding of more controllers (incl. two RLL ones), from VARCem;
Added the AHA-1540A and the BusTek BT-542B;
Moved the Sumo SCSI-AT to the Dev branch;
Minor IDE, FDC, and floppy drive code clean-ups;
Made NCR 5380/53C400-based cards' BIOS address configurable;
Got rid of the legacy romset variable;
Unified (video) buffer and buffer32 into one and make the unified buffer 32-bit;
Added the Amstead PPC512 per PCem patch by John Elliott;
Switched memory mapping granularity from 16k to 4k (less than 1k not possible due to internal pages);
Rewrote the CL-GD 54xx blitter, fixes Win-OS/2 on the 54x6 among other thing;
Added the Image Manager 1024 and Professional Graphics Controller per PCem patch by John Elliott and work done on VARCem;
Added Headland HT-216, GC-205 and Video 7 VGA 1024i emulation based on PCem commit;
Implemented the fuction keys for the Toshiba T1000/T1200/T3100 enhancement;
Amstrad MegaPC does now works correctly with non-internal graphics card;
The SLiRP code no longer casts a packed struct type to a non-packed struct type;
The Xi8088 and PB410a no longer hang on 86Box when PS/2 mouse is not present;
The S3 Virge on BeOS is no longer broken (was broken by build #1591);
OS/2 2.0 build 6.167 now sees key presses again;
Xi8088 now work on CGA again;
86F images converted from either the old or new variants of the HxC MFM format now work correctly;
Hardware interrupts with a vector of 0xFF are now handled correctly;
OPTi 495SX boards no longer incorrectly have 64 MB maximum RAM when 32 MB is correct;
Fixed VNC keyboard input bugs;
Fixed AT RTC periodic interrupt - Chicago 58s / 73f / 73g  / 81 MIDI play no longer hangs with the build's own VTD driver;
Fixed mouse polling with internal mice - Amstrad and Olivetti mice now work correctly;
Triones ATAPI DMA driver now correctly reads a file at the end of a CD image with a sectors number not divisible by 4;
Compaq Portable now works with all graphics cards;
Fixed various MDSI Genius bugs;
Added segment limit checks and improved page fault checks for several CPU instructions - Memphis 15xx WINSETUP and Chicago 58s WINDISK.CPL no longer issue a GPF, and some S3 drivers that used to have glitches, now work correctly;
Further improved the 808x emulation, also fixes the noticably choppy sound when using 808x CPU's, also fixes #355;
OS/2 installer no logner locks up on splash screen on PS/2 Model 70 and 80, fixes #400.
Fixed several Amstead bugs, GEM no longer crashes on the Amstrad 1640, fixes #391.
Ported John Elliott's Amstrad fixes and improvement from PCem, and fixed the default language so it's correctly Engliish, fixes #278, fixes #389.
Fixed a minor IDE timing bug, fixes #388.
Fixed Toshiba T1000 RAM issues, fixes #379.
Fixed EGA/(S)VGA overscan border handling, fixes #378;
Got rid of the now long useless IDE channel 2 auto-removal, fixes #370;
Fixed the BIOS files used by the AMSTRAD PC1512, fixes #366;
Ported the Unicode CD image file name fix from VARCem, fixes #365;
Fixed high density floppy disks on the Xi8088, fixes #359;
Fixed some bugs in the Hercules emulation, fixes #346, fixes #358;
Fixed the SCSI hard disk mode sense pages, fixes #356;
Removed the AMI Unknown 386SX because of impossibility to identify the chipset, closes #349;
Fixed bugs in the serial mouse emulation, fixes #344;
Compiled 86Box binaries now include all the required .DLL's, fixes #341;
Made some combo boxes in the Settings dialog slightly wider, fixes #276.
This commit is contained in:
OBattler
2019-09-20 14:02:30 +02:00
parent b06296bbf6
commit 552a87ea3d
524 changed files with 129555 additions and 21862 deletions
Download Patch File Download Diff File Expand all files Collapse all files

391
src/cpu_new/codegen_backend_x86-64.c Normal file
View File

@@ -0,0 +1,391 @@
#ifdef __amd64__
#include <stdint.h>
#include "../86box.h"
#include "cpu.h"
#include "../mem.h"
#include "codegen.h"
#include "codegen_allocator.h"
#include "codegen_backend.h"
#include "codegen_backend_x86-64_defs.h"
#include "codegen_backend_x86-64_ops.h"
#include "codegen_backend_x86-64_ops_sse.h"
#include "codegen_reg.h"
#include "x86.h"
#if defined(__linux__) || defined(__APPLE__)
#include <sys/mman.h>
#include <unistd.h>
#endif
#if defined WIN32 || defined _WIN32 || defined _WIN32
#include <windows.h>
#endif
void *codegen_mem_load_byte;
void *codegen_mem_load_word;
void *codegen_mem_load_long;
void *codegen_mem_load_quad;
void *codegen_mem_load_single;
void *codegen_mem_load_double;
void *codegen_mem_store_byte;
void *codegen_mem_store_word;
void *codegen_mem_store_long;
void *codegen_mem_store_quad;
void *codegen_mem_store_single;
void *codegen_mem_store_double;
void *codegen_gpf_rout;
void *codegen_exit_rout;
host_reg_def_t codegen_host_reg_list[CODEGEN_HOST_REGS] =
{
/*Note: while EAX and EDX are normally volatile registers under x86
calling conventions, the recompiler will explicitly save and restore
them across funcion calls*/
{REG_EAX, 0},
{REG_EBX, 0},
{REG_EDX, 0}
};
host_reg_def_t codegen_host_fp_reg_list[CODEGEN_HOST_FP_REGS] =
{
#if WIN64
/*Windows x86-64 calling convention preserves XMM6-XMM15*/
{REG_XMM6, 0},
{REG_XMM7, 0},
#else
/*System V AMD64 calling convention does not preserve any XMM registers*/
{REG_XMM6, HOST_REG_FLAG_VOLATILE},
{REG_XMM7, HOST_REG_FLAG_VOLATILE},
#endif
{REG_XMM1, HOST_REG_FLAG_VOLATILE},
{REG_XMM2, HOST_REG_FLAG_VOLATILE},
{REG_XMM3, HOST_REG_FLAG_VOLATILE},
{REG_XMM4, HOST_REG_FLAG_VOLATILE},
{REG_XMM5, HOST_REG_FLAG_VOLATILE}
};
static void build_load_routine(codeblock_t *block, int size, int is_float)
{
uint8_t *branch_offset;
uint8_t *misaligned_offset;
/*In - ESI = address
Out - ECX = data, ESI = abrt*/
/*MOV ECX, ESI
SHR ESI, 12
MOV RSI, [readlookup2+ESI*4]
CMP ESI, -1
JNZ +
MOVZX ECX, B[RSI+RCX]
XOR ESI,ESI
RET
* PUSH EAX
PUSH EDX
PUSH ECX
CALL readmembl
POP ECX
POP EDX
POP EAX
MOVZX ECX, AL
RET
*/
host_x86_MOV32_REG_REG(block, REG_ECX, REG_ESI);
host_x86_SHR32_IMM(block, REG_ESI, 12);
host_x86_MOV64_REG_IMM(block, REG_RDI, (uint64_t)(uintptr_t)readlookup2);
host_x86_MOV64_REG_BASE_INDEX_SHIFT(block, REG_RSI, REG_RDI, REG_RSI, 3);
if (size != 1)
{
host_x86_TEST32_REG_IMM(block, REG_ECX, size-1);
misaligned_offset = host_x86_JNZ_short(block);
}
host_x86_CMP64_REG_IMM(block, REG_RSI, (uint32_t)-1);
branch_offset = host_x86_JZ_short(block);
if (size == 1 && !is_float)
host_x86_MOVZX_BASE_INDEX_32_8(block, REG_ECX, REG_RSI, REG_RCX);
else if (size == 2 && !is_float)
host_x86_MOVZX_BASE_INDEX_32_16(block, REG_ECX, REG_RSI, REG_RCX);
else if (size == 4 && !is_float)
host_x86_MOV32_REG_BASE_INDEX(block, REG_ECX, REG_RSI, REG_RCX);
else if (size == 4 && is_float)
host_x86_CVTSS2SD_XREG_BASE_INDEX(block, REG_XMM_TEMP, REG_RSI, REG_RCX);
else if (size == 8)
host_x86_MOVQ_XREG_BASE_INDEX(block, REG_XMM_TEMP, REG_RSI, REG_RCX);
else
fatal("build_load_routine: size=%i\n", size);
host_x86_XOR32_REG_REG(block, REG_ESI, REG_ESI);
host_x86_RET(block);
*branch_offset = (uint8_t)((uintptr_t)&block_write_data[block_pos] - (uintptr_t)branch_offset) - 1;
if (size != 1)
*misaligned_offset = (uint8_t)((uintptr_t)&block_write_data[block_pos] - (uintptr_t)misaligned_offset) - 1;
host_x86_PUSH(block, REG_RAX);
host_x86_PUSH(block, REG_RDX);
#if WIN64
host_x86_SUB64_REG_IMM(block, REG_RSP, 0x20);
//host_x86_MOV32_REG_REG(block, REG_ECX, uop->imm_data);
#else
host_x86_MOV32_REG_REG(block, REG_EDI, REG_ECX);
#endif
if (size == 1 && !is_float)
{
host_x86_CALL(block, (void *)readmembl);
host_x86_MOVZX_REG_32_8(block, REG_ECX, REG_EAX);
}
else if (size == 2 && !is_float)
{
host_x86_CALL(block, (void *)readmemwl);
host_x86_MOVZX_REG_32_16(block, REG_ECX, REG_EAX);
}
else if (size == 4 && !is_float)
{
host_x86_CALL(block, (void *)readmemll);
host_x86_MOV32_REG_REG(block, REG_ECX, REG_EAX);
}
else if (size == 4 && is_float)
{
host_x86_CALL(block, (void *)readmemll);
host_x86_MOVD_XREG_REG(block, REG_XMM_TEMP, REG_EAX);
host_x86_CVTSS2SD_XREG_XREG(block, REG_XMM_TEMP, REG_XMM_TEMP);
}
else if (size == 8)
{
host_x86_CALL(block, (void *)readmemql);
host_x86_MOVQ_XREG_REG(block, REG_XMM_TEMP, REG_RAX);
}
#if WIN64
host_x86_ADD64_REG_IMM(block, REG_RSP, 0x20);
#endif
host_x86_POP(block, REG_RDX);
host_x86_POP(block, REG_RAX);
host_x86_MOVZX_REG_ABS_32_8(block, REG_ESI, &cpu_state.abrt);
host_x86_RET(block);
}
static void build_store_routine(codeblock_t *block, int size, int is_float)
{
uint8_t *branch_offset;
uint8_t *misaligned_offset;
/*In - ECX = data, ESI = address
Out - ESI = abrt
Corrupts EDI*/
/*MOV EDI, ESI
SHR ESI, 12
MOV ESI, [writelookup2+ESI*4]
CMP ESI, -1
JNZ +
MOV [RSI+RDI], ECX
XOR ESI,ESI
RET
* PUSH EAX
PUSH EDX
PUSH ECX
CALL writemembl
POP ECX
POP EDX
POP EAX
MOVZX ECX, AL
RET
*/
host_x86_MOV32_REG_REG(block, REG_EDI, REG_ESI);
host_x86_SHR32_IMM(block, REG_ESI, 12);
host_x86_MOV64_REG_IMM(block, REG_R8, (uint64_t)(uintptr_t)writelookup2);
host_x86_MOV64_REG_BASE_INDEX_SHIFT(block, REG_RSI, REG_R8, REG_RSI, 3);
if (size != 1)
{
host_x86_TEST32_REG_IMM(block, REG_EDI, size-1);
misaligned_offset = host_x86_JNZ_short(block);
}
host_x86_CMP64_REG_IMM(block, REG_RSI, (uint32_t)-1);
branch_offset = host_x86_JZ_short(block);
if (size == 1 && !is_float)
host_x86_MOV8_BASE_INDEX_REG(block, REG_RSI, REG_RDI, REG_ECX);
else if (size == 2 && !is_float)
host_x86_MOV16_BASE_INDEX_REG(block, REG_RSI, REG_RDI, REG_ECX);
else if (size == 4 && !is_float)
host_x86_MOV32_BASE_INDEX_REG(block, REG_RSI, REG_RDI, REG_ECX);
else if (size == 4 && is_float)
host_x86_MOVD_BASE_INDEX_XREG(block, REG_RSI, REG_RDI, REG_XMM_TEMP);
else if (size == 8)
host_x86_MOVQ_BASE_INDEX_XREG(block, REG_RSI, REG_RDI, REG_XMM_TEMP);
else
fatal("build_store_routine: size=%i\n", size);
host_x86_XOR32_REG_REG(block, REG_ESI, REG_ESI);
host_x86_RET(block);
*branch_offset = (uint8_t)((uintptr_t)&block_write_data[block_pos] - (uintptr_t)branch_offset) - 1;
if (size != 1)
*misaligned_offset = (uint8_t)((uintptr_t)&block_write_data[block_pos] - (uintptr_t)misaligned_offset) - 1;
host_x86_PUSH(block, REG_RAX);
host_x86_PUSH(block, REG_RDX);
#if WIN64
host_x86_SUB64_REG_IMM(block, REG_RSP, 0x28);
if (size == 4 && is_float)
host_x86_MOVD_REG_XREG(block, REG_EDX, REG_XMM_TEMP); //data
else if (size == 8)
host_x86_MOVQ_REG_XREG(block, REG_RDX, REG_XMM_TEMP); //data
else
host_x86_MOV32_REG_REG(block, REG_EDX, REG_ECX); //data
host_x86_MOV32_REG_REG(block, REG_ECX, REG_EDI); //address
#else
host_x86_SUB64_REG_IMM(block, REG_RSP, 0x8);
//host_x86_MOV32_REG_REG(block, REG_EDI, REG_ECX); //address
if (size == 4 && is_float)
host_x86_MOVD_REG_XREG(block, REG_ESI, REG_XMM_TEMP); //data
else if (size == 8)
host_x86_MOVQ_REG_XREG(block, REG_RSI, REG_XMM_TEMP); //data
else
host_x86_MOV32_REG_REG(block, REG_ESI, REG_ECX); //data
#endif
if (size == 1)
host_x86_CALL(block, (void *)writemembl);
else if (size == 2)
host_x86_CALL(block, (void *)writememwl);
else if (size == 4)
host_x86_CALL(block, (void *)writememll);
else if (size == 8)
host_x86_CALL(block, (void *)writememql);
#if WIN64
host_x86_ADD64_REG_IMM(block, REG_RSP, 0x28);
#else
host_x86_ADD64_REG_IMM(block, REG_RSP, 0x8);
#endif
host_x86_POP(block, REG_RDX);
host_x86_POP(block, REG_RAX);
host_x86_MOVZX_REG_ABS_32_8(block, REG_ESI, &cpu_state.abrt);
host_x86_RET(block);
}
static void build_loadstore_routines(codeblock_t *block)
{
codegen_mem_load_byte = &codeblock[block_current].data[block_pos];
build_load_routine(block, 1, 0);
codegen_mem_load_word = &codeblock[block_current].data[block_pos];
build_load_routine(block, 2, 0);
codegen_mem_load_long = &codeblock[block_current].data[block_pos];
build_load_routine(block, 4, 0);
codegen_mem_load_quad = &codeblock[block_current].data[block_pos];
build_load_routine(block, 8, 0);
codegen_mem_load_single = &codeblock[block_current].data[block_pos];
build_load_routine(block, 4, 1);
codegen_mem_load_double = &codeblock[block_current].data[block_pos];
build_load_routine(block, 8, 1);
codegen_mem_store_byte = &codeblock[block_current].data[block_pos];
build_store_routine(block, 1, 0);
codegen_mem_store_word = &codeblock[block_current].data[block_pos];
build_store_routine(block, 2, 0);
codegen_mem_store_long = &codeblock[block_current].data[block_pos];
build_store_routine(block, 4, 0);
codegen_mem_store_quad = &codeblock[block_current].data[block_pos];
build_store_routine(block, 8, 0);
codegen_mem_store_single = &codeblock[block_current].data[block_pos];
build_store_routine(block, 4, 1);
codegen_mem_store_double = &codeblock[block_current].data[block_pos];
build_store_routine(block, 8, 1);
}
void codegen_backend_init()
{
codeblock_t *block;
int c;
#if defined(__linux__) || defined(__APPLE__)
void *start;
size_t len;
long pagesize = sysconf(_SC_PAGESIZE);
long pagemask = ~(pagesize - 1);
#endif
codeblock = malloc(BLOCK_SIZE * sizeof(codeblock_t));
codeblock_hash = malloc(HASH_SIZE * sizeof(codeblock_t *));
memset(codeblock, 0, BLOCK_SIZE * sizeof(codeblock_t));
memset(codeblock_hash, 0, HASH_SIZE * sizeof(codeblock_t *));
for (c = 0; c < BLOCK_SIZE; c++)
codeblock[c].pc = BLOCK_PC_INVALID;
block_current = 0;
block_pos = 0;
block = &codeblock[block_current];
codeblock[block_current].head_mem_block = codegen_allocator_allocate(NULL, block_current);
codeblock[block_current].data = codeblock_allocator_get_ptr(codeblock[block_current].head_mem_block);
block_write_data = codeblock[block_current].data;
build_loadstore_routines(&codeblock[block_current]);
codegen_gpf_rout = &codeblock[block_current].data[block_pos];
#if WIN64
host_x86_XOR32_REG_REG(block, REG_ECX, REG_ECX);
host_x86_XOR32_REG_REG(block, REG_EDX, REG_EDX);
#else
host_x86_XOR32_REG_REG(block, REG_EDI, REG_EDI);
host_x86_XOR32_REG_REG(block, REG_ESI, REG_ESI);
#endif
/* host_x86_CALL(block, (uintptr_t)x86gpf); */
host_x86_CALL(block, (void *)x86gpf);
codegen_exit_rout = &codeblock[block_current].data[block_pos];
host_x86_ADD64_REG_IMM(block, REG_RSP, 0x38);
host_x86_POP(block, REG_R15);
host_x86_POP(block, REG_R14);
host_x86_POP(block, REG_R13);
host_x86_POP(block, REG_R12);
host_x86_POP(block, REG_RDI);
host_x86_POP(block, REG_RSI);
host_x86_POP(block, REG_RBP);
host_x86_POP(block, REG_RDX);
host_x86_RET(block);
block_write_data = NULL;
asm(
"stmxcsr %0\n"
: "=m" (cpu_state.old_fp_control)
);
cpu_state.trunc_fp_control = cpu_state.old_fp_control | 0x6000;
}
void codegen_set_rounding_mode(int mode)
{
cpu_state.new_fp_control = (cpu_state.old_fp_control & ~0x6000) | (mode << 13);
}
void codegen_backend_prologue(codeblock_t *block)
{
block_pos = BLOCK_START; /*Entry code*/
host_x86_PUSH(block, REG_RBX);
host_x86_PUSH(block, REG_RBP);
host_x86_PUSH(block, REG_RSI);
host_x86_PUSH(block, REG_RDI);
host_x86_PUSH(block, REG_R12);
host_x86_PUSH(block, REG_R13);
host_x86_PUSH(block, REG_R14);
host_x86_PUSH(block, REG_R15);
host_x86_SUB64_REG_IMM(block, REG_RSP, 0x38);
host_x86_MOV64_REG_IMM(block, REG_RBP, ((uintptr_t)&cpu_state) + 128);
if (block->flags & CODEBLOCK_HAS_FPU)
{
host_x86_MOV32_REG_ABS(block, REG_EAX, &cpu_state.TOP);
host_x86_SUB32_REG_IMM(block, REG_EAX, block->TOP);
host_x86_MOV32_BASE_OFFSET_REG(block, REG_RSP, IREG_TOP_diff_stack_offset, REG_EAX);
}
if (block->flags & CODEBLOCK_NO_IMMEDIATES)
host_x86_MOV64_REG_IMM(block, REG_R12, (uintptr_t)ram);
}
void codegen_backend_epilogue(codeblock_t *block)
{
host_x86_ADD64_REG_IMM(block, REG_RSP, 0x38);
host_x86_POP(block, REG_R15);
host_x86_POP(block, REG_R14);
host_x86_POP(block, REG_R13);
host_x86_POP(block, REG_R12);
host_x86_POP(block, REG_RDI);
host_x86_POP(block, REG_RSI);
host_x86_POP(block, REG_RBP);
host_x86_POP(block, REG_RDX);
host_x86_RET(block);
}
#endif