SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_fast_return(struct sljit_compiler *compiler, sljit_s32 src, sljit_sw srcw)
{
sljit_u8 *inst;
CHECK_ERROR();
CHECK(check_sljit_emit_fast_return(compiler, src, srcw));
ADJUST_LOCAL_OFFSET(src, srcw);
if ((src & SLJIT_IMM) && NOT_HALFWORD(srcw)) {
FAIL_IF(emit_load_imm64(compiler, TMP_REG1, srcw));
src = TMP_REG1;
}
if (FAST_IS_REG(src)) {
if (reg_map[src] < 8) {
inst = (sljit_u8*)ensure_buf(compiler, 1 + 1 + 1);
FAIL_IF(!inst);
INC_SIZE(1 + 1);
PUSH_REG(reg_lmap[src]);
}
else {
inst = (sljit_u8*)ensure_buf(compiler, 1 + 2 + 1);
FAIL_IF(!inst);
INC_SIZE(2 + 1);
*inst++ = REX_B;
PUSH_REG(reg_lmap[src]);
}
}
else if (src & SLJIT_MEM) {
/* REX_W is not necessary (src is not immediate). */
compiler->mode32 = 1;
inst = emit_x86_instruction(compiler, 1, 0, 0, src, srcw);
FAIL_IF(!inst);
*inst++ = GROUP_FF;
*inst |= PUSH_rm;
inst = (sljit_u8*)ensure_buf(compiler, 1 + 1);
FAIL_IF(!inst);
INC_SIZE(1);
}
else {
SLJIT_ASSERT(IS_HALFWORD(srcw));
/* SLJIT_IMM. */
inst = (sljit_u8*)ensure_buf(compiler, 1 + 5 + 1);
FAIL_IF(!inst);
INC_SIZE(5 + 1);
*inst++ = PUSH_i32;
sljit_unaligned_store_s32(inst, srcw);
inst += sizeof(sljit_s32);
}
RET();
return SLJIT_SUCCESS;
}
static sljit_s32 emit_mov_int(struct sljit_compiler *compiler, sljit_s32 sign,
sljit_s32 dst, sljit_sw dstw,
sljit_s32 src, sljit_sw srcw)
{
sljit_u8* inst;
sljit_s32 dst_r;
compiler->mode32 = 0;
if (dst == SLJIT_UNUSED && !(src & SLJIT_MEM))
return SLJIT_SUCCESS; /* Empty instruction. */
if (src & SLJIT_IMM) {
if (FAST_IS_REG(dst)) {
if (sign || ((sljit_uw)srcw <= 0x7fffffff)) {
inst = emit_x86_instruction(compiler, 1, SLJIT_IMM, (sljit_sw)(sljit_s32)srcw, dst, dstw);
FAIL_IF(!inst);
*inst = MOV_rm_i32;
return SLJIT_SUCCESS;
}
return emit_load_imm64(compiler, dst, srcw);
}
compiler->mode32 = 1;
inst = emit_x86_instruction(compiler, 1, SLJIT_IMM, (sljit_sw)(sljit_s32)srcw, dst, dstw);
FAIL_IF(!inst);
*inst = MOV_rm_i32;
compiler->mode32 = 0;
return SLJIT_SUCCESS;
}
dst_r = FAST_IS_REG(dst) ? dst : TMP_REG1;
if ((dst & SLJIT_MEM) && FAST_IS_REG(src))
dst_r = src;
else {
if (sign) {
inst = emit_x86_instruction(compiler, 1, dst_r, 0, src, srcw);
FAIL_IF(!inst);
*inst++ = MOVSXD_r_rm;
} else {
compiler->mode32 = 1;
FAIL_IF(emit_mov(compiler, dst_r, 0, src, srcw));
compiler->mode32 = 0;
}
}
if (dst & SLJIT_MEM) {
compiler->mode32 = 1;
inst = emit_x86_instruction(compiler, 1, dst_r, 0, dst, dstw);
FAIL_IF(!inst);
*inst = MOV_rm_r;
compiler->mode32 = 0;
}
return SLJIT_SUCCESS;
}
static sljit_ub* emit_x86_instruction(struct sljit_compiler *compiler, sljit_si size,
/* The register or immediate operand. */
sljit_si a, sljit_sw imma,
/* The general operand (not immediate). */
sljit_si b, sljit_sw immb)
{
sljit_ub *inst;
sljit_ub *buf_ptr;
sljit_ub rex = 0;
sljit_si flags = size & ~0xf;
sljit_si inst_size;
/* The immediate operand must be 32 bit. */
SLJIT_ASSERT(!(a & SLJIT_IMM) || compiler->mode32 || IS_HALFWORD(imma));
/* Both cannot be switched on. */
SLJIT_ASSERT((flags & (EX86_BIN_INS | EX86_SHIFT_INS)) != (EX86_BIN_INS | EX86_SHIFT_INS));
/* Size flags not allowed for typed instructions. */
SLJIT_ASSERT(!(flags & (EX86_BIN_INS | EX86_SHIFT_INS)) || (flags & (EX86_BYTE_ARG | EX86_HALF_ARG)) == 0);
/* Both size flags cannot be switched on. */
SLJIT_ASSERT((flags & (EX86_BYTE_ARG | EX86_HALF_ARG)) != (EX86_BYTE_ARG | EX86_HALF_ARG));
#if (defined SLJIT_SSE2 && SLJIT_SSE2)
/* SSE2 and immediate is not possible. */
SLJIT_ASSERT(!(a & SLJIT_IMM) || !(flags & EX86_SSE2));
SLJIT_ASSERT((flags & (EX86_PREF_F2 | EX86_PREF_F3)) != (EX86_PREF_F2 | EX86_PREF_F3)
&& (flags & (EX86_PREF_F2 | EX86_PREF_66)) != (EX86_PREF_F2 | EX86_PREF_66)
&& (flags & (EX86_PREF_F3 | EX86_PREF_66)) != (EX86_PREF_F3 | EX86_PREF_66));
#endif
size &= 0xf;
inst_size = size;
if ((b & SLJIT_MEM) && !(b & 0xf0) && NOT_HALFWORD(immb)) {
if (emit_load_imm64(compiler, TMP_REG3, immb))
return NULL;
immb = 0;
if (b & 0xf)
b |= TMP_REG3 << 4;
else
b |= TMP_REG3;
}
if (!compiler->mode32 && !(flags & EX86_NO_REXW))
rex |= REX_W;
else if (flags & EX86_REX)
rex |= REX;
#if (defined SLJIT_SSE2 && SLJIT_SSE2)
if (flags & (EX86_PREF_F2 | EX86_PREF_F3))
inst_size++;
#endif
if (flags & EX86_PREF_66)
inst_size++;
/* Calculate size of b. */
inst_size += 1; /* mod r/m byte. */
if (b & SLJIT_MEM) {
if ((b & 0x0f) == SLJIT_UNUSED)
inst_size += 1 + sizeof(sljit_si); /* SIB byte required to avoid RIP based addressing. */
else {
if (reg_map[b & 0x0f] >= 8)
rex |= REX_B;
if (immb != 0 && !(b & 0xf0)) {
/* Immediate operand. */
if (immb <= 127 && immb >= -128)
inst_size += sizeof(sljit_sb);
else
inst_size += sizeof(sljit_si);
}
}
if ((b & 0xf) == SLJIT_LOCALS_REG && !(b & 0xf0))
b |= SLJIT_LOCALS_REG << 4;
if ((b & 0xf0) != SLJIT_UNUSED) {
inst_size += 1; /* SIB byte. */
if (reg_map[(b >> 4) & 0x0f] >= 8)
rex |= REX_X;
}
}
static sljit_u8* emit_x86_instruction(struct sljit_compiler *compiler, sljit_s32 size,
/* The register or immediate operand. */
sljit_s32 a, sljit_sw imma,
/* The general operand (not immediate). */
sljit_s32 b, sljit_sw immb)
{
sljit_u8 *inst;
sljit_u8 *buf_ptr;
sljit_u8 rex = 0;
sljit_s32 flags = size & ~0xf;
sljit_s32 inst_size;
/* The immediate operand must be 32 bit. */
SLJIT_ASSERT(!(a & SLJIT_IMM) || compiler->mode32 || IS_HALFWORD(imma));
/* Both cannot be switched on. */
SLJIT_ASSERT((flags & (EX86_BIN_INS | EX86_SHIFT_INS)) != (EX86_BIN_INS | EX86_SHIFT_INS));
/* Size flags not allowed for typed instructions. */
SLJIT_ASSERT(!(flags & (EX86_BIN_INS | EX86_SHIFT_INS)) || (flags & (EX86_BYTE_ARG | EX86_HALF_ARG)) == 0);
/* Both size flags cannot be switched on. */
SLJIT_ASSERT((flags & (EX86_BYTE_ARG | EX86_HALF_ARG)) != (EX86_BYTE_ARG | EX86_HALF_ARG));
/* SSE2 and immediate is not possible. */
SLJIT_ASSERT(!(a & SLJIT_IMM) || !(flags & EX86_SSE2));
SLJIT_ASSERT((flags & (EX86_PREF_F2 | EX86_PREF_F3)) != (EX86_PREF_F2 | EX86_PREF_F3)
&& (flags & (EX86_PREF_F2 | EX86_PREF_66)) != (EX86_PREF_F2 | EX86_PREF_66)
&& (flags & (EX86_PREF_F3 | EX86_PREF_66)) != (EX86_PREF_F3 | EX86_PREF_66));
size &= 0xf;
inst_size = size;
if (!compiler->mode32 && !(flags & EX86_NO_REXW))
rex |= REX_W;
else if (flags & EX86_REX)
rex |= REX;
if (flags & (EX86_PREF_F2 | EX86_PREF_F3))
inst_size++;
if (flags & EX86_PREF_66)
inst_size++;
/* Calculate size of b. */
inst_size += 1; /* mod r/m byte. */
if (b & SLJIT_MEM) {
if (!(b & OFFS_REG_MASK)) {
if (NOT_HALFWORD(immb)) {
PTR_FAIL_IF(emit_load_imm64(compiler, TMP_REG3, immb));
immb = 0;
if (b & REG_MASK)
b |= TO_OFFS_REG(TMP_REG3);
else
b |= TMP_REG3;
}
else if (reg_lmap[b & REG_MASK] == 4)
b |= TO_OFFS_REG(SLJIT_SP);
}
if ((b & REG_MASK) == SLJIT_UNUSED)
inst_size += 1 + sizeof(sljit_s32); /* SIB byte required to avoid RIP based addressing. */
else {
if (reg_map[b & REG_MASK] >= 8)
rex |= REX_B;
if (immb != 0 && (!(b & OFFS_REG_MASK) || (b & OFFS_REG_MASK) == TO_OFFS_REG(SLJIT_SP))) {
/* Immediate operand. */
if (immb <= 127 && immb >= -128)
inst_size += sizeof(sljit_s8);
else
inst_size += sizeof(sljit_s32);
}
else if (reg_lmap[b & REG_MASK] == 5)
inst_size += sizeof(sljit_s8);
if ((b & OFFS_REG_MASK) != SLJIT_UNUSED) {
inst_size += 1; /* SIB byte. */
if (reg_map[OFFS_REG(b)] >= 8)
rex |= REX_X;
}
}
}
else if (!(flags & EX86_SSE2_OP2) && reg_map[b] >= 8)
rex |= REX_B;
if (a & SLJIT_IMM) {
if (flags & EX86_BIN_INS) {
if (imma <= 127 && imma >= -128) {
inst_size += 1;
flags |= EX86_BYTE_ARG;
} else
inst_size += 4;
}
else if (flags & EX86_SHIFT_INS) {
imma &= compiler->mode32 ? 0x1f : 0x3f;
if (imma != 1) {
inst_size ++;
flags |= EX86_BYTE_ARG;
}
} else if (flags & EX86_BYTE_ARG)
inst_size++;
else if (flags & EX86_HALF_ARG)
inst_size += sizeof(short);
else
inst_size += sizeof(sljit_s32);
}
else {
SLJIT_ASSERT(!(flags & EX86_SHIFT_INS) || a == SLJIT_PREF_SHIFT_REG);
/* reg_map[SLJIT_PREF_SHIFT_REG] is less than 8. */
if (!(flags & EX86_SSE2_OP1) && reg_map[a] >= 8)
rex |= REX_R;
}
if (rex)
inst_size++;
inst = (sljit_u8*)ensure_buf(compiler, 1 + inst_size);
PTR_FAIL_IF(!inst);
/* Encoding the byte. */
INC_SIZE(inst_size);
if (flags & EX86_PREF_F2)
*inst++ = 0xf2;
if (flags & EX86_PREF_F3)
*inst++ = 0xf3;
if (flags & EX86_PREF_66)
*inst++ = 0x66;
if (rex)
*inst++ = rex;
buf_ptr = inst + size;
/* Encode mod/rm byte. */
if (!(flags & EX86_SHIFT_INS)) {
if ((flags & EX86_BIN_INS) && (a & SLJIT_IMM))
*inst = (flags & EX86_BYTE_ARG) ? GROUP_BINARY_83 : GROUP_BINARY_81;
if ((a & SLJIT_IMM) || (a == 0))
*buf_ptr = 0;
else if (!(flags & EX86_SSE2_OP1))
*buf_ptr = reg_lmap[a] << 3;
else
*buf_ptr = a << 3;
}
else {
if (a & SLJIT_IMM) {
if (imma == 1)
*inst = GROUP_SHIFT_1;
else
*inst = GROUP_SHIFT_N;
} else
*inst = GROUP_SHIFT_CL;
*buf_ptr = 0;
}
if (!(b & SLJIT_MEM))
*buf_ptr++ |= MOD_REG + ((!(flags & EX86_SSE2_OP2)) ? reg_lmap[b] : b);
else if ((b & REG_MASK) != SLJIT_UNUSED) {
if ((b & OFFS_REG_MASK) == SLJIT_UNUSED || (b & OFFS_REG_MASK) == TO_OFFS_REG(SLJIT_SP)) {
if (immb != 0 || reg_lmap[b & REG_MASK] == 5) {
if (immb <= 127 && immb >= -128)
*buf_ptr |= 0x40;
else
*buf_ptr |= 0x80;
}
if ((b & OFFS_REG_MASK) == SLJIT_UNUSED)
*buf_ptr++ |= reg_lmap[b & REG_MASK];
else {
*buf_ptr++ |= 0x04;
*buf_ptr++ = reg_lmap[b & REG_MASK] | (reg_lmap[OFFS_REG(b)] << 3);
}
if (immb != 0 || reg_lmap[b & REG_MASK] == 5) {
if (immb <= 127 && immb >= -128)
*buf_ptr++ = immb; /* 8 bit displacement. */
else {
sljit_unaligned_store_s32(buf_ptr, immb); /* 32 bit displacement. */
buf_ptr += sizeof(sljit_s32);
}
}
}
else {
if (reg_lmap[b & REG_MASK] == 5)
*buf_ptr |= 0x40;
*buf_ptr++ |= 0x04;
*buf_ptr++ = reg_lmap[b & REG_MASK] | (reg_lmap[OFFS_REG(b)] << 3) | (immb << 6);
if (reg_lmap[b & REG_MASK] == 5)
*buf_ptr++ = 0;
}
}
else {
*buf_ptr++ |= 0x04;
*buf_ptr++ = 0x25;
sljit_unaligned_store_s32(buf_ptr, immb); /* 32 bit displacement. */
buf_ptr += sizeof(sljit_s32);
}
if (a & SLJIT_IMM) {
if (flags & EX86_BYTE_ARG)
*buf_ptr = imma;
else if (flags & EX86_HALF_ARG)
sljit_unaligned_store_s16(buf_ptr, imma);
else if (!(flags & EX86_SHIFT_INS))
sljit_unaligned_store_s32(buf_ptr, imma);
}
return !(flags & EX86_SHIFT_INS) ? inst : (inst + 1);
}
SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_enter(struct sljit_compiler *compiler, int args, int temporaries, int saveds, int local_size)
{
int size, pushed_size;
sljit_ub *buf;
CHECK_ERROR();
check_sljit_emit_enter(compiler, args, temporaries, saveds, local_size);
compiler->temporaries = temporaries;
compiler->saveds = saveds;
compiler->flags_saved = 0;
size = saveds;
/* Including the return address saved by the call instruction. */
pushed_size = (saveds + 1) * sizeof(sljit_w);
#ifndef _WIN64
if (saveds >= 2)
size += saveds - 1;
#else
/* Saving the virtual stack pointer. */
compiler->has_locals = local_size > 0;
if (local_size > 0) {
size += 2;
pushed_size += sizeof(sljit_w);
}
if (saveds >= 4)
size += saveds - 3;
if (temporaries >= 5) {
size += (5 - 4) * 2;
pushed_size += sizeof(sljit_w);
}
#endif
size += args * 3;
if (size > 0) {
buf = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!buf);
INC_SIZE(size);
if (saveds >= 5) {
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_EREG2] >= 8, saved_ereg2_is_hireg);
*buf++ = REX_B;
PUSH_REG(reg_lmap[SLJIT_SAVED_EREG2]);
}
if (saveds >= 4) {
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_EREG1] >= 8, saved_ereg1_is_hireg);
*buf++ = REX_B;
PUSH_REG(reg_lmap[SLJIT_SAVED_EREG1]);
}
if (saveds >= 3) {
#ifndef _WIN64
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_REG3] >= 8, saved_reg3_is_hireg);
*buf++ = REX_B;
#else
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_REG3] < 8, saved_reg3_is_loreg);
#endif
PUSH_REG(reg_lmap[SLJIT_SAVED_REG3]);
}
if (saveds >= 2) {
#ifndef _WIN64
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_REG2] >= 8, saved_reg2_is_hireg);
*buf++ = REX_B;
#else
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_REG2] < 8, saved_reg2_is_loreg);
#endif
PUSH_REG(reg_lmap[SLJIT_SAVED_REG2]);
}
if (saveds >= 1) {
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_REG1] < 8, saved_reg1_is_loreg);
PUSH_REG(reg_lmap[SLJIT_SAVED_REG1]);
}
#ifdef _WIN64
if (temporaries >= 5) {
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_TEMPORARY_EREG2] >= 8, temporary_ereg2_is_hireg);
*buf++ = REX_B;
PUSH_REG(reg_lmap[SLJIT_TEMPORARY_EREG2]);
}
if (local_size > 0) {
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_LOCALS_REG] >= 8, locals_reg_is_hireg);
*buf++ = REX_B;
PUSH_REG(reg_lmap[SLJIT_LOCALS_REG]);
}
#endif
#ifndef _WIN64
if (args > 0) {
*buf++ = REX_W;
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_map[SLJIT_SAVED_REG1] << 3) | 0x7;
}
if (args > 1) {
*buf++ = REX_W | REX_R;
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_lmap[SLJIT_SAVED_REG2] << 3) | 0x6;
}
if (args > 2) {
*buf++ = REX_W | REX_R;
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_lmap[SLJIT_SAVED_REG3] << 3) | 0x2;
}
#else
if (args > 0) {
*buf++ = REX_W;
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_map[SLJIT_SAVED_REG1] << 3) | 0x1;
}
if (args > 1) {
*buf++ = REX_W;
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_map[SLJIT_SAVED_REG2] << 3) | 0x2;
}
if (args > 2) {
*buf++ = REX_W | REX_B;
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_map[SLJIT_SAVED_REG3] << 3) | 0x0;
}
#endif
}
local_size = ((local_size + pushed_size + 16 - 1) & ~(16 - 1)) - pushed_size;
#ifdef _WIN64
local_size += 4 * sizeof(sljit_w);
compiler->local_size = local_size;
if (local_size > 1024) {
/* Allocate the stack for the function itself. */
buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!buf);
INC_SIZE(4);
*buf++ = REX_W;
*buf++ = 0x83;
*buf++ = 0xc0 | (5 << 3) | 4;
/* Pushed size must be divisible by 8. */
SLJIT_ASSERT(!(pushed_size & 0x7));
if (pushed_size & 0x8) {
*buf++ = 5 * sizeof(sljit_w);
local_size -= 5 * sizeof(sljit_w);
} else {
*buf++ = 4 * sizeof(sljit_w);
local_size -= 4 * sizeof(sljit_w);
}
FAIL_IF(emit_load_imm64(compiler, SLJIT_TEMPORARY_REG1, local_size));
FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_touch_stack)));
}
#else
compiler->local_size = local_size;
if (local_size > 0) {
#endif
/* In case of Win64, local_size is always > 4 * sizeof(sljit_w) */
if (local_size <= 127) {
buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!buf);
INC_SIZE(4);
*buf++ = REX_W;
*buf++ = 0x83;
*buf++ = 0xc0 | (5 << 3) | 4;
*buf++ = local_size;
}
else {
buf = (sljit_ub*)ensure_buf(compiler, 1 + 7);
FAIL_IF(!buf);
INC_SIZE(7);
*buf++ = REX_W;
*buf++ = 0x81;
*buf++ = 0xc0 | (5 << 3) | 4;
*(sljit_hw*)buf = local_size;
buf += sizeof(sljit_hw);
}
#ifndef _WIN64
}
#endif
#ifdef _WIN64
if (compiler->has_locals) {
buf = (sljit_ub*)ensure_buf(compiler, 1 + 5);
FAIL_IF(!buf);
INC_SIZE(5);
*buf++ = REX_W | REX_R;
*buf++ = 0x8d;
*buf++ = 0x40 | (reg_lmap[SLJIT_LOCALS_REG] << 3) | 0x4;
*buf++ = 0x24;
*buf = 4 * sizeof(sljit_w);
}
#endif
return SLJIT_SUCCESS;
}
