/* Size contains the flags as well. */
static sljit_ub* emit_x86_instruction(struct sljit_compiler *compiler, int size,
/* The register or immediate operand. */
int a, sljit_w imma,
/* The general operand (not immediate). */
int b, sljit_w immb)
{
sljit_ub *buf;
sljit_ub *buf_ptr;
int flags = size & ~0xf;
int inst_size;
/* 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));
#endif
size &= 0xf;
inst_size = size;
#if (defined SLJIT_SSE2 && SLJIT_SSE2)
if (flags & EX86_PREF_F2)
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 += sizeof(sljit_w);
else if (immb != 0 && !(b & 0xf0)) {
/* Immediate operand. */
if (immb <= 127 && immb >= -128)
inst_size += sizeof(sljit_b);
else
inst_size += sizeof(sljit_w);
}
if ((b & 0xf) == SLJIT_LOCALS_REG && !(b & 0xf0))
b |= SLJIT_LOCALS_REG << 4;
if ((b & 0xf0) != SLJIT_UNUSED)
inst_size += 1; /* SIB byte. */
}
/* Calculate size of a. */
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 &= 0x1f;
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_w);
}
else
SLJIT_ASSERT(!(flags & EX86_SHIFT_INS) || a == SLJIT_PREF_SHIFT_REG);
buf = (sljit_ub*)ensure_buf(compiler, 1 + inst_size);
PTR_FAIL_IF(!buf);
/* Encoding the byte. */
INC_SIZE(inst_size);
#if (defined SLJIT_SSE2 && SLJIT_SSE2)
if (flags & EX86_PREF_F2)
*buf++ = 0xf2;
#endif
if (flags & EX86_PREF_66)
*buf++ = 0x66;
buf_ptr = buf + size;
/* Encode mod/rm byte. */
if (!(flags & EX86_SHIFT_INS)) {
if ((flags & EX86_BIN_INS) && (a & SLJIT_IMM))
*buf = (flags & EX86_BYTE_ARG) ? 0x83 : 0x81;
if ((a & SLJIT_IMM) || (a == 0))
*buf_ptr = 0;
#if (defined SLJIT_SSE2 && SLJIT_SSE2)
else if (!(flags & EX86_SSE2))
*buf_ptr = reg_map[a] << 3;
else
*buf_ptr = a << 3;
#else
else
*buf_ptr = reg_map[a] << 3;
#endif
}
int sljit_emit_enter(struct sljit_compiler *compiler, int args, int temporaries, int generals, int local_size)
{
int size;
sljit_ub *buf;
CHECK_ERROR();
check_sljit_emit_enter(compiler, args, temporaries, generals, local_size);
compiler->temporaries = temporaries;
compiler->generals = generals;
compiler->args = args;
compiler->flags_saved = 0;
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
size = 1 + (generals <= 3 ? generals : 3) + (args > 0 ? (args * 2) : 0) + (args > 2 ? 2 : 0);
#else
size = 1 + (generals <= 3 ? generals : 3) + (args > 0 ? (2 + args * 3) : 0);
#endif
buf = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!buf);
INC_SIZE(size);
PUSH_REG(reg_map[TMP_REGISTER]);
#if !(defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (args > 0) {
*buf++ = 0x8b;
*buf++ = 0xc4 | (reg_map[TMP_REGISTER] << 3);
}
#endif
if (generals > 2)
PUSH_REG(reg_map[SLJIT_GENERAL_REG3]);
if (generals > 1)
PUSH_REG(reg_map[SLJIT_GENERAL_REG2]);
if (generals > 0)
PUSH_REG(reg_map[SLJIT_GENERAL_REG1]);
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (args > 0) {
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_map[SLJIT_GENERAL_REG1] << 3) | reg_map[SLJIT_TEMPORARY_REG3];
}
if (args > 1) {
*buf++ = 0x8b;
*buf++ = 0xc0 | (reg_map[SLJIT_GENERAL_REG2] << 3) | reg_map[SLJIT_TEMPORARY_REG2];
}
if (args > 2) {
*buf++ = 0x8b;
*buf++ = 0x44 | (reg_map[SLJIT_GENERAL_REG3] << 3);
*buf++ = 0x24;
*buf++ = sizeof(sljit_w) * (3 + 2); /* generals >= 3 as well. */
}
#else
if (args > 0) {
*buf++ = 0x8b;
*buf++ = 0x40 | (reg_map[SLJIT_GENERAL_REG1] << 3) | reg_map[TMP_REGISTER];
*buf++ = sizeof(sljit_w) * 2;
}
if (args > 1) {
*buf++ = 0x8b;
*buf++ = 0x40 | (reg_map[SLJIT_GENERAL_REG2] << 3) | reg_map[TMP_REGISTER];
*buf++ = sizeof(sljit_w) * 3;
}
if (args > 2) {
*buf++ = 0x8b;
*buf++ = 0x40 | (reg_map[SLJIT_GENERAL_REG3] << 3) | reg_map[TMP_REGISTER];
*buf++ = sizeof(sljit_w) * 4;
}
#endif
local_size = (local_size + sizeof(sljit_uw) - 1) & ~(sizeof(sljit_uw) - 1);
compiler->temporaries_start = local_size;
if (temporaries > 3)
local_size += (temporaries - 3) * sizeof(sljit_uw);
compiler->generals_start = local_size;
if (generals > 3)
local_size += (generals - 3) * sizeof(sljit_uw);
#ifdef _WIN32
if (local_size > 1024) {
FAIL_IF(emit_do_imm(compiler, 0xb8 + reg_map[SLJIT_TEMPORARY_REG1], local_size));
FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_touch_stack)));
}
#endif
compiler->local_size = local_size;
if (local_size > 0)
return emit_non_cum_binary(compiler, 0x2b, 0x29, 0x5 << 3, 0x2d,
SLJIT_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, local_size);
/* Mov arguments to general registers. */
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));
/* 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)) {
if (emit_load_imm64(compiler, TMP_REG3, immb))
return NULL;
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_si); /* 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_sb);
else
inst_size += sizeof(sljit_si);
}
else if (reg_lmap[b & REG_MASK] == 5)
inst_size += sizeof(sljit_sb);
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_si);
}
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_ub*)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_si*)buf_ptr = immb; /* 32 bit displacement. */
buf_ptr += sizeof(sljit_si);
}
}
}
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_si*)buf_ptr = immb; /* 32 bit displacement. */
buf_ptr += sizeof(sljit_si);
}
if (a & SLJIT_IMM) {
if (flags & EX86_BYTE_ARG)
*buf_ptr = imma;
else if (flags & EX86_HALF_ARG)
*(short*)buf_ptr = imma;
else if (!(flags & EX86_SHIFT_INS))
*(sljit_si*)buf_ptr = imma;
}
return !(flags & EX86_SHIFT_INS) ? inst : (inst + 1);
}
SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_enter(struct sljit_compiler *compiler,
sljit_si options, sljit_si args, sljit_si scratches, sljit_si saveds,
sljit_si fscratches, sljit_si fsaveds, sljit_si local_size)
{
sljit_si i, tmp, size, saved_register_size;
sljit_ub *inst;
CHECK_ERROR();
CHECK(check_sljit_emit_enter(compiler, options, args, scratches, saveds, fscratches, fsaveds, local_size));
set_emit_enter(compiler, options, args, scratches, saveds, fscratches, fsaveds, local_size);
compiler->flags_saved = 0;
/* Including the return address saved by the call instruction. */
saved_register_size = GET_SAVED_REGISTERS_SIZE(scratches, saveds, 1);
tmp = saveds < SLJIT_NUMBER_OF_SAVED_REGISTERS ? (SLJIT_S0 + 1 - saveds) : SLJIT_FIRST_SAVED_REG;
for (i = SLJIT_S0; i >= tmp; i--) {
size = reg_map[i] >= 8 ? 2 : 1;
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
if (reg_map[i] >= 8)
*inst++ = REX_B;
PUSH_REG(reg_lmap[i]);
}
for (i = scratches; i >= SLJIT_FIRST_SAVED_REG; i--) {
size = reg_map[i] >= 8 ? 2 : 1;
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
if (reg_map[i] >= 8)
*inst++ = REX_B;
PUSH_REG(reg_lmap[i]);
}
if (args > 0) {
size = args * 3;
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
#ifndef _WIN64
if (args > 0) {
*inst++ = REX_W;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_S0] << 3) | 0x7 /* rdi */;
}
if (args > 1) {
*inst++ = REX_W | REX_R;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_lmap[SLJIT_S1] << 3) | 0x6 /* rsi */;
}
if (args > 2) {
*inst++ = REX_W | REX_R;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_lmap[SLJIT_S2] << 3) | 0x2 /* rdx */;
}
#else
if (args > 0) {
*inst++ = REX_W;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_S0] << 3) | 0x1 /* rcx */;
}
if (args > 1) {
*inst++ = REX_W;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_S1] << 3) | 0x2 /* rdx */;
}
if (args > 2) {
*inst++ = REX_W | REX_B;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_S2] << 3) | 0x0 /* r8 */;
}
#endif
}
local_size = ((local_size + SLJIT_LOCALS_OFFSET + saved_register_size + 15) & ~15) - saved_register_size;
compiler->local_size = local_size;
#ifdef _WIN64
if (local_size > 1024) {
/* Allocate stack for the callback, which grows the stack. */
inst = (sljit_ub*)ensure_buf(compiler, 1 + 4 + (3 + sizeof(sljit_si)));
FAIL_IF(!inst);
INC_SIZE(4 + (3 + sizeof(sljit_si)));
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | SUB | 4;
/* Allocated size for registers must be divisible by 8. */
SLJIT_ASSERT(!(saved_register_size & 0x7));
/* Aligned to 16 byte. */
if (saved_register_size & 0x8) {
*inst++ = 5 * sizeof(sljit_sw);
local_size -= 5 * sizeof(sljit_sw);
} else {
*inst++ = 4 * sizeof(sljit_sw);
local_size -= 4 * sizeof(sljit_sw);
}
/* Second instruction */
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_R0] < 8, temporary_reg1_is_loreg);
*inst++ = REX_W;
*inst++ = MOV_rm_i32;
*inst++ = MOD_REG | reg_lmap[SLJIT_R0];
*(sljit_si*)inst = local_size;
#if (defined SLJIT_VERBOSE && SLJIT_VERBOSE) \
|| (defined SLJIT_ARGUMENT_CHECKS && SLJIT_ARGUMENT_CHECKS)
compiler->skip_checks = 1;
#endif
FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_grow_stack)));
}
#endif
SLJIT_ASSERT(local_size > 0);
if (local_size <= 127) {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!inst);
INC_SIZE(4);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | SUB | 4;
*inst++ = local_size;
}
else {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 7);
FAIL_IF(!inst);
INC_SIZE(7);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_81;
*inst++ = MOD_REG | SUB | 4;
*(sljit_si*)inst = local_size;
inst += sizeof(sljit_si);
}
#ifdef _WIN64
/* Save xmm6 register: movaps [rsp + 0x20], xmm6 */
if (fscratches >= 6 || fsaveds >= 1) {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 5);
FAIL_IF(!inst);
INC_SIZE(5);
*inst++ = GROUP_0F;
*(sljit_si*)inst = 0x20247429;
}
#endif
return SLJIT_SUCCESS;
}
SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_enter(struct sljit_compiler *compiler, sljit_si args, sljit_si scratches, sljit_si saveds, sljit_si local_size)
{
sljit_si size, pushed_size;
sljit_ub *inst;
CHECK_ERROR();
check_sljit_emit_enter(compiler, args, scratches, saveds, local_size);
compiler->scratches = scratches;
compiler->saveds = saveds;
compiler->flags_saved = 0;
#if (defined SLJIT_DEBUG && SLJIT_DEBUG)
compiler->logical_local_size = local_size;
#endif
size = saveds;
/* Including the return address saved by the call instruction. */
pushed_size = (saveds + 1) * sizeof(sljit_sw);
#ifndef _WIN64
if (saveds >= 2)
size += saveds - 1;
#else
if (saveds >= 4)
size += saveds - 3;
if (scratches >= 5) {
size += (5 - 4) * 2;
pushed_size += sizeof(sljit_sw);
}
#endif
size += args * 3;
if (size > 0) {
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
if (saveds >= 5) {
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SAVED_EREG2] >= 8, saved_ereg2_is_hireg);
*inst++ = 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);
*inst++ = 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);
*inst++ = 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);
*inst++ = 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 (scratches >= 5) {
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_TEMPORARY_EREG2] >= 8, temporary_ereg2_is_hireg);
*inst++ = REX_B;
PUSH_REG(reg_lmap[SLJIT_TEMPORARY_EREG2]);
}
#endif
#ifndef _WIN64
if (args > 0) {
*inst++ = REX_W;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_SAVED_REG1] << 3) | 0x7 /* rdi */;
}
if (args > 1) {
*inst++ = REX_W | REX_R;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_lmap[SLJIT_SAVED_REG2] << 3) | 0x6 /* rsi */;
}
if (args > 2) {
*inst++ = REX_W | REX_R;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_lmap[SLJIT_SAVED_REG3] << 3) | 0x2 /* rdx */;
}
#else
if (args > 0) {
*inst++ = REX_W;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_SAVED_REG1] << 3) | 0x1 /* rcx */;
}
if (args > 1) {
*inst++ = REX_W;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_SAVED_REG2] << 3) | 0x2 /* rdx */;
}
if (args > 2) {
*inst++ = REX_W | REX_B;
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_SAVED_REG3] << 3) | 0x0 /* r8 */;
}
#endif
}
local_size = ((local_size + FIXED_LOCALS_OFFSET + pushed_size + 16 - 1) & ~(16 - 1)) - pushed_size;
compiler->local_size = local_size;
#ifdef _WIN64
if (local_size > 1024) {
/* Allocate stack for the callback, which grows the stack. */
inst = (sljit_ub*)ensure_buf(compiler, 1 + 4 + (3 + sizeof(sljit_si)));
FAIL_IF(!inst);
INC_SIZE(4 + (3 + sizeof(sljit_si)));
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | SUB | 4;
/* Pushed size must be divisible by 8. */
SLJIT_ASSERT(!(pushed_size & 0x7));
if (pushed_size & 0x8) {
*inst++ = 5 * sizeof(sljit_sw);
local_size -= 5 * sizeof(sljit_sw);
} else {
*inst++ = 4 * sizeof(sljit_sw);
local_size -= 4 * sizeof(sljit_sw);
}
/* Second instruction */
SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SCRATCH_REG1] < 8, temporary_reg1_is_loreg);
*inst++ = REX_W;
*inst++ = MOV_rm_i32;
*inst++ = MOD_REG | reg_lmap[SLJIT_SCRATCH_REG1];
*(sljit_si*)inst = local_size;
#if (defined SLJIT_VERBOSE && SLJIT_VERBOSE) || (defined SLJIT_DEBUG && SLJIT_DEBUG)
compiler->skip_checks = 1;
#endif
FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_grow_stack)));
}
#endif
SLJIT_ASSERT(local_size > 0);
if (local_size <= 127) {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!inst);
INC_SIZE(4);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | SUB | 4;
*inst++ = local_size;
}
else {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 7);
FAIL_IF(!inst);
INC_SIZE(7);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_81;
*inst++ = MOD_REG | SUB | 4;
*(sljit_si*)inst = local_size;
inst += sizeof(sljit_si);
}
#ifdef _WIN64
/* Save xmm6 with MOVAPS instruction. */
inst = (sljit_ub*)ensure_buf(compiler, 1 + 5);
FAIL_IF(!inst);
INC_SIZE(5);
*inst++ = GROUP_0F;
*(sljit_si*)inst = 0x20247429;
#endif
return SLJIT_SUCCESS;
}
SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_return(struct sljit_compiler *compiler, sljit_si op, sljit_si src, sljit_sw srcw)
{
sljit_si i, tmp, size;
sljit_ub *inst;
CHECK_ERROR();
CHECK(check_sljit_emit_return(compiler, op, src, srcw));
compiler->flags_saved = 0;
FAIL_IF(emit_mov_before_return(compiler, op, src, srcw));
#ifdef _WIN64
/* Restore xmm6 register: movaps xmm6, [rsp + 0x20] */
if (compiler->fscratches >= 6 || compiler->fsaveds >= 1) {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 5);
FAIL_IF(!inst);
INC_SIZE(5);
*inst++ = GROUP_0F;
*(sljit_si*)inst = 0x20247428;
}
#endif
SLJIT_ASSERT(compiler->local_size > 0);
if (compiler->local_size <= 127) {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!inst);
INC_SIZE(4);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | ADD | 4;
*inst = compiler->local_size;
}
else {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 7);
FAIL_IF(!inst);
INC_SIZE(7);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_81;
*inst++ = MOD_REG | ADD | 4;
*(sljit_si*)inst = compiler->local_size;
}
tmp = compiler->scratches;
for (i = SLJIT_FIRST_SAVED_REG; i <= tmp; i++) {
size = reg_map[i] >= 8 ? 2 : 1;
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
if (reg_map[i] >= 8)
*inst++ = REX_B;
POP_REG(reg_lmap[i]);
}
tmp = compiler->saveds < SLJIT_NUMBER_OF_SAVED_REGISTERS ? (SLJIT_S0 + 1 - compiler->saveds) : SLJIT_FIRST_SAVED_REG;
for (i = tmp; i <= SLJIT_S0; i++) {
size = reg_map[i] >= 8 ? 2 : 1;
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
if (reg_map[i] >= 8)
*inst++ = REX_B;
POP_REG(reg_lmap[i]);
}
inst = (sljit_ub*)ensure_buf(compiler, 1 + 1);
FAIL_IF(!inst);
INC_SIZE(1);
RET();
return SLJIT_SUCCESS;
}
SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_enter(struct sljit_compiler *compiler, sljit_si args, sljit_si scratches, sljit_si saveds, sljit_si local_size)
{
sljit_si size;
sljit_si locals_offset;
sljit_ub *inst;
CHECK_ERROR();
check_sljit_emit_enter(compiler, args, scratches, saveds, local_size);
compiler->scratches = scratches;
compiler->saveds = saveds;
compiler->args = args;
compiler->flags_saved = 0;
#if (defined SLJIT_DEBUG && SLJIT_DEBUG)
compiler->logical_local_size = local_size;
#endif
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
size = 1 + (saveds <= 3 ? saveds : 3) + (args > 0 ? (args * 2) : 0) + (args > 2 ? 2 : 0);
#else
size = 1 + (saveds <= 3 ? saveds : 3) + (args > 0 ? (2 + args * 3) : 0);
#endif
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
PUSH_REG(reg_map[TMP_REGISTER]);
#if !(defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (args > 0) {
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[TMP_REGISTER] << 3) | 0x4 /* esp */;
}
#endif
if (saveds > 2)
PUSH_REG(reg_map[SLJIT_SAVED_REG3]);
if (saveds > 1)
PUSH_REG(reg_map[SLJIT_SAVED_REG2]);
if (saveds > 0)
PUSH_REG(reg_map[SLJIT_SAVED_REG1]);
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (args > 0) {
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_SAVED_REG1] << 3) | reg_map[SLJIT_SCRATCH_REG3];
}
if (args > 1) {
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_SAVED_REG2] << 3) | reg_map[SLJIT_SCRATCH_REG2];
}
if (args > 2) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_SAVED_REG3] << 3) | 0x4 /* esp */;
*inst++ = 0x24;
*inst++ = sizeof(sljit_sw) * (3 + 2); /* saveds >= 3 as well. */
}
#else
if (args > 0) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_SAVED_REG1] << 3) | reg_map[TMP_REGISTER];
*inst++ = sizeof(sljit_sw) * 2;
}
if (args > 1) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_SAVED_REG2] << 3) | reg_map[TMP_REGISTER];
*inst++ = sizeof(sljit_sw) * 3;
}
if (args > 2) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_SAVED_REG3] << 3) | reg_map[TMP_REGISTER];
*inst++ = sizeof(sljit_sw) * 4;
}
#endif
locals_offset = 2 * sizeof(sljit_uw);
compiler->scratches_start = locals_offset;
if (scratches > 3)
locals_offset += (scratches - 3) * sizeof(sljit_uw);
compiler->saveds_start = locals_offset;
if (saveds > 3)
locals_offset += (saveds - 3) * sizeof(sljit_uw);
compiler->locals_offset = locals_offset;
local_size = locals_offset + ((local_size + sizeof(sljit_uw) - 1) & ~(sizeof(sljit_uw) - 1));
#ifdef _WIN32
if (local_size > 1024) {
FAIL_IF(emit_do_imm(compiler, MOV_r_i32 + reg_map[SLJIT_SCRATCH_REG1], local_size));
FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_grow_stack)));
}
#endif
compiler->local_size = local_size;
SLJIT_ASSERT(local_size > 0);
return emit_non_cum_binary(compiler, SUB_r_rm, SUB_rm_r, SUB, SUB_EAX_i32,
SLJIT_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, local_size);
return SLJIT_SUCCESS;
}
SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_return(struct sljit_compiler *compiler, sljit_si op, sljit_si src, sljit_sw srcw)
{
sljit_si size;
sljit_ub *inst;
CHECK_ERROR();
check_sljit_emit_return(compiler, op, src, srcw);
compiler->flags_saved = 0;
FAIL_IF(emit_mov_before_return(compiler, op, src, srcw));
#ifdef _WIN64
/* Restore xmm6 with MOVAPS instruction. */
inst = (sljit_ub*)ensure_buf(compiler, 1 + 5);
FAIL_IF(!inst);
INC_SIZE(5);
*inst++ = GROUP_0F;
*(sljit_si*)inst = 0x20247428;
#endif
SLJIT_ASSERT(compiler->local_size > 0);
if (compiler->local_size <= 127) {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!inst);
INC_SIZE(4);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | ADD | 4;
*inst = compiler->local_size;
}
else {
inst = (sljit_ub*)ensure_buf(compiler, 1 + 7);
FAIL_IF(!inst);
INC_SIZE(7);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_81;
*inst++ = MOD_REG | ADD | 4;
*(sljit_si*)inst = compiler->local_size;
}
size = 1 + compiler->saveds;
#ifndef _WIN64
if (compiler->saveds >= 2)
size += compiler->saveds - 1;
#else
if (compiler->saveds >= 4)
size += compiler->saveds - 3;
if (compiler->scratches >= 5)
size += (5 - 4) * 2;
#endif
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
#ifdef _WIN64
if (compiler->scratches >= 5) {
*inst++ = REX_B;
POP_REG(reg_lmap[SLJIT_TEMPORARY_EREG2]);
}
#endif
if (compiler->saveds >= 1)
POP_REG(reg_map[SLJIT_SAVED_REG1]);
if (compiler->saveds >= 2) {
#ifndef _WIN64
*inst++ = REX_B;
#endif
POP_REG(reg_lmap[SLJIT_SAVED_REG2]);
}
if (compiler->saveds >= 3) {
#ifndef _WIN64
*inst++ = REX_B;
#endif
POP_REG(reg_lmap[SLJIT_SAVED_REG3]);
}
if (compiler->saveds >= 4) {
*inst++ = REX_B;
POP_REG(reg_lmap[SLJIT_SAVED_EREG1]);
}
if (compiler->saveds >= 5) {
*inst++ = REX_B;
POP_REG(reg_lmap[SLJIT_SAVED_EREG2]);
}
RET();
return SLJIT_SUCCESS;
}
gboolean
gegl_buffer_iterator_next (GeglBufferIterator *iterator)
{
GeglBufferIterators *i = (gpointer)iterator;
gboolean result = FALSE;
gint no;
if (i->is_finished)
g_error ("%s called on finished buffer iterator", G_STRFUNC);
if (i->iteration_no == 0)
{
for (no=0; no<i->iterators;no++)
{
gint j;
gboolean found = FALSE;
for (j=0; j<no; j++)
if (i->buffer[no]==i->buffer[j])
{
found = TRUE;
break;
}
if (!found)
gegl_buffer_lock (i->buffer[no]);
if (gegl_cl_is_accelerated ())
gegl_buffer_cl_cache_flush (i->buffer[no], &i->rect[no]);
}
}
else
{
/* complete pending write work */
for (no=0; no<i->iterators;no++)
{
if (i->flags[no] & GEGL_BUFFER_WRITE)
{
if (i->flags[no] & GEGL_BUFFER_SCAN_COMPATIBLE &&
i->flags[no] & GEGL_BUFFER_FORMAT_COMPATIBLE &&
i->roi[no].width == i->i[no].buffer->tile_storage->tile_width && (i->flags[no] & GEGL_BUFFER_FORMAT_COMPATIBLE))
{ /* direct access, don't need to do anything */
#if DEBUG_DIRECT
direct_write += i->roi[no].width * i->roi[no].height;
#endif
}
else
{
#if DEBUG_DIRECT
in_direct_write += i->roi[no].width * i->roi[no].height;
#endif
ensure_buf (i, no);
/* XXX: should perhaps use _set_unlocked, and keep the lock in the
* iterator.
*/
gegl_buffer_set (i->buffer[no], &(i->roi[no]), 0, i->format[no], i->buf[no], GEGL_AUTO_ROWSTRIDE); /* XXX: use correct level */
}
}
}
}
g_assert (i->iterators > 0);
/* then we iterate all */
for (no=0; no<i->iterators;no++)
{
if (i->flags[no] & GEGL_BUFFER_SCAN_COMPATIBLE)
{
gboolean res;
res = gegl_buffer_tile_iterator_next (&i->i[no]);
if (no == 0)
{
result = res;
}
i->roi[no] = i->i[no].roi2;
/* since they were scan compatible this should be true */
if (res != result)
{
g_print ("%i==%i != 0==%i\n", no, res, result);
}
g_assert (res == result);
if ((i->flags[no] & GEGL_BUFFER_FORMAT_COMPATIBLE) &&
i->roi[no].width == i->i[no].buffer->tile_storage->tile_width
)
{
/* direct access */
i->data[no]=i->i[no].sub_data;
#if DEBUG_DIRECT
direct_read += i->roi[no].width * i->roi[no].height;
#endif
}
else
{
ensure_buf (i, no);
if (i->flags[no] & GEGL_BUFFER_READ)
{
gegl_buffer_get_unlocked (i->buffer[no], 1.0, &(i->roi[no]), i->format[no], i->buf[no], GEGL_AUTO_ROWSTRIDE);
}
i->data[no]=i->buf[no];
#if DEBUG_DIRECT
in_direct_read += i->roi[no].width * i->roi[no].height;
#endif
}
}
else
{
/* we copy the roi from iterator 0 */
i->roi[no] = i->roi[0];
i->roi[no].x += (i->rect[no].x-i->rect[0].x);
i->roi[no].y += (i->rect[no].y-i->rect[0].y);
ensure_buf (i, no);
if (i->flags[no] & GEGL_BUFFER_READ)
{
gegl_buffer_get_unlocked (i->buffer[no], 1.0, &(i->roi[no]), i->format[no], i->buf[no], GEGL_AUTO_ROWSTRIDE);
}
i->data[no]=i->buf[no];
#if DEBUG_DIRECT
in_direct_read += i->roi[no].width * i->roi[no].height;
#endif
}
i->length = i->roi[no].width * i->roi[no].height;
}
i->iteration_no++;
if (result == FALSE)
gegl_buffer_iterator_stop (iterator);
return result;
}
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;
}
SLJIT_API_FUNC_ATTRIBUTE int sljit_emit_return(struct sljit_compiler *compiler, int op, int src, sljit_w srcw)
{
int size;
sljit_ub *buf;
CHECK_ERROR();
check_sljit_emit_return(compiler, op, src, srcw);
compiler->flags_saved = 0;
FAIL_IF(emit_mov_before_return(compiler, op, src, srcw));
if (compiler->local_size > 0) {
if (compiler->local_size <= 127) {
buf = (sljit_ub*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!buf);
INC_SIZE(4);
*buf++ = REX_W;
*buf++ = 0x83;
*buf++ = 0xc0 | (0 << 3) | 4;
*buf = compiler->local_size;
}
else {
buf = (sljit_ub*)ensure_buf(compiler, 1 + 7);
FAIL_IF(!buf);
INC_SIZE(7);
*buf++ = REX_W;
*buf++ = 0x81;
*buf++ = 0xc0 | (0 << 3) | 4;
*(sljit_hw*)buf = compiler->local_size;
}
}
size = 1 + compiler->saveds;
#ifndef _WIN64
if (compiler->saveds >= 2)
size += compiler->saveds - 1;
#else
if (compiler->has_locals)
size += 2;
if (compiler->saveds >= 4)
size += compiler->saveds - 3;
if (compiler->temporaries >= 5)
size += (5 - 4) * 2;
#endif
buf = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!buf);
INC_SIZE(size);
#ifdef _WIN64
if (compiler->has_locals) {
*buf++ = REX_B;
POP_REG(reg_lmap[SLJIT_LOCALS_REG]);
}
if (compiler->temporaries >= 5) {
*buf++ = REX_B;
POP_REG(reg_lmap[SLJIT_TEMPORARY_EREG2]);
}
#endif
if (compiler->saveds >= 1)
POP_REG(reg_map[SLJIT_SAVED_REG1]);
if (compiler->saveds >= 2) {
#ifndef _WIN64
*buf++ = REX_B;
#endif
POP_REG(reg_lmap[SLJIT_SAVED_REG2]);
}
if (compiler->saveds >= 3) {
#ifndef _WIN64
*buf++ = REX_B;
#endif
POP_REG(reg_lmap[SLJIT_SAVED_REG3]);
}
if (compiler->saveds >= 4) {
*buf++ = REX_B;
POP_REG(reg_lmap[SLJIT_SAVED_EREG1]);
}
if (compiler->saveds >= 5) {
*buf++ = REX_B;
POP_REG(reg_lmap[SLJIT_SAVED_EREG2]);
}
RET();
return SLJIT_SUCCESS;
}
SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_enter(struct sljit_compiler *compiler,
sljit_si options, sljit_si args, sljit_si scratches, sljit_si saveds,
sljit_si fscratches, sljit_si fsaveds, sljit_si local_size)
{
sljit_si size;
sljit_ub *inst;
CHECK_ERROR();
CHECK(check_sljit_emit_enter(compiler, options, args, scratches, saveds, fscratches, fsaveds, local_size));
set_emit_enter(compiler, options, args, scratches, saveds, fscratches, fsaveds, local_size);
compiler->args = args;
compiler->flags_saved = 0;
size = 1 + (scratches > 7 ? (scratches - 7) : 0) + (saveds <= 3 ? saveds : 3);
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
size += (args > 0 ? (args * 2) : 0) + (args > 2 ? 2 : 0);
#else
size += (args > 0 ? (2 + args * 3) : 0);
#endif
inst = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!inst);
INC_SIZE(size);
PUSH_REG(reg_map[TMP_REG1]);
#if !(defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (args > 0) {
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[TMP_REG1] << 3) | 0x4 /* esp */;
}
#endif
if (saveds > 2 || scratches > 7)
PUSH_REG(reg_map[SLJIT_S2]);
if (saveds > 1 || scratches > 8)
PUSH_REG(reg_map[SLJIT_S1]);
if (saveds > 0 || scratches > 9)
PUSH_REG(reg_map[SLJIT_S0]);
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (args > 0) {
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_S0] << 3) | reg_map[SLJIT_R2];
}
if (args > 1) {
*inst++ = MOV_r_rm;
*inst++ = MOD_REG | (reg_map[SLJIT_S1] << 3) | reg_map[SLJIT_R1];
}
if (args > 2) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_S2] << 3) | 0x4 /* esp */;
*inst++ = 0x24;
*inst++ = sizeof(sljit_sw) * (3 + 2); /* saveds >= 3 as well. */
}
#else
if (args > 0) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_S0] << 3) | reg_map[TMP_REG1];
*inst++ = sizeof(sljit_sw) * 2;
}
if (args > 1) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_S1] << 3) | reg_map[TMP_REG1];
*inst++ = sizeof(sljit_sw) * 3;
}
if (args > 2) {
*inst++ = MOV_r_rm;
*inst++ = MOD_DISP8 | (reg_map[SLJIT_S2] << 3) | reg_map[TMP_REG1];
*inst++ = sizeof(sljit_sw) * 4;
}
#endif
SLJIT_COMPILE_ASSERT(SLJIT_LOCALS_OFFSET >= (2 + 4) * sizeof(sljit_uw), require_at_least_two_words);
#if defined(__APPLE__)
/* Ignore pushed registers and SLJIT_LOCALS_OFFSET when computing the aligned local size. */
saveds = (2 + (scratches > 7 ? (scratches - 7) : 0) + (saveds <= 3 ? saveds : 3)) * sizeof(sljit_uw);
local_size = ((SLJIT_LOCALS_OFFSET + saveds + local_size + 15) & ~15) - saveds;
#else
if (options & SLJIT_DOUBLE_ALIGNMENT) {
local_size = SLJIT_LOCALS_OFFSET + ((local_size + 7) & ~7);
inst = (sljit_ub*)ensure_buf(compiler, 1 + 17);
FAIL_IF(!inst);
INC_SIZE(17);
inst[0] = MOV_r_rm;
inst[1] = MOD_REG | (reg_map[TMP_REG1] << 3) | reg_map[SLJIT_SP];
inst[2] = GROUP_F7;
inst[3] = MOD_REG | (0 << 3) | reg_map[SLJIT_SP];
*(sljit_sw*)(inst + 4) = 0x4;
inst[8] = JNE_i8;
inst[9] = 6;
inst[10] = GROUP_BINARY_81;
inst[11] = MOD_REG | (5 << 3) | reg_map[SLJIT_SP];
*(sljit_sw*)(inst + 12) = 0x4;
inst[16] = PUSH_r + reg_map[TMP_REG1];
}
else
local_size = SLJIT_LOCALS_OFFSET + ((local_size + 3) & ~3);
#endif
compiler->local_size = local_size;
#ifdef _WIN32
if (local_size > 1024) {
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
FAIL_IF(emit_do_imm(compiler, MOV_r_i32 + reg_map[SLJIT_R0], local_size));
#else
local_size -= SLJIT_LOCALS_OFFSET;
FAIL_IF(emit_do_imm(compiler, MOV_r_i32 + reg_map[SLJIT_R0], local_size));
FAIL_IF(emit_non_cum_binary(compiler, SUB_r_rm, SUB_rm_r, SUB, SUB_EAX_i32,
SLJIT_SP, 0, SLJIT_SP, 0, SLJIT_IMM, SLJIT_LOCALS_OFFSET));
#endif
FAIL_IF(sljit_emit_ijump(compiler, SLJIT_CALL1, SLJIT_IMM, SLJIT_FUNC_OFFSET(sljit_grow_stack)));
}
#endif
SLJIT_ASSERT(local_size > 0);
return emit_non_cum_binary(compiler, SUB_r_rm, SUB_rm_r, SUB, SUB_EAX_i32,
SLJIT_SP, 0, SLJIT_SP, 0, SLJIT_IMM, local_size);
}
/* Size contains the flags as well. */
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_si flags = size & ~0xf;
sljit_si inst_size;
/* 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 (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 & REG_MASK) == SLJIT_UNUSED)
inst_size += sizeof(sljit_sw);
else if (immb != 0 && !(b & OFFS_REG_MASK)) {
/* Immediate operand. */
if (immb <= 127 && immb >= -128)
inst_size += sizeof(sljit_sb);
else
inst_size += sizeof(sljit_sw);
}
if ((b & REG_MASK) == SLJIT_SP && !(b & OFFS_REG_MASK))
b |= TO_OFFS_REG(SLJIT_SP);
if ((b & OFFS_REG_MASK) != SLJIT_UNUSED)
inst_size += 1; /* SIB byte. */
}
/* Calculate size of a. */
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 &= 0x1f;
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_sw);
}
else
SLJIT_ASSERT(!(flags & EX86_SHIFT_INS) || a == SLJIT_PREF_SHIFT_REG);
inst = (sljit_ub*)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;
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_map[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_map[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) {
if (immb <= 127 && immb >= -128)
*buf_ptr |= 0x40;
else
*buf_ptr |= 0x80;
}
if ((b & OFFS_REG_MASK) == SLJIT_UNUSED)
*buf_ptr++ |= reg_map[b & REG_MASK];
else {
*buf_ptr++ |= 0x04;
*buf_ptr++ = reg_map[b & REG_MASK] | (reg_map[OFFS_REG(b)] << 3);
}
if (immb != 0) {
if (immb <= 127 && immb >= -128)
*buf_ptr++ = immb; /* 8 bit displacement. */
else {
*(sljit_sw*)buf_ptr = immb; /* 32 bit displacement. */
buf_ptr += sizeof(sljit_sw);
}
}
}
else {
*buf_ptr++ |= 0x04;
*buf_ptr++ = reg_map[b & REG_MASK] | (reg_map[OFFS_REG(b)] << 3) | (immb << 6);
}
}
else {
*buf_ptr++ |= 0x05;
*(sljit_sw*)buf_ptr = immb; /* 32 bit displacement. */
buf_ptr += sizeof(sljit_sw);
}
if (a & SLJIT_IMM) {
if (flags & EX86_BYTE_ARG)
*buf_ptr = imma;
else if (flags & EX86_HALF_ARG)
*(short*)buf_ptr = imma;
else if (!(flags & EX86_SHIFT_INS))
*(sljit_sw*)buf_ptr = imma;
}
return !(flags & EX86_SHIFT_INS) ? inst : (inst + 1);
}
