static void unroll_op(pTHX_ struct sljit_compiler* compiler, OP* op)
{
DEBUGf((" ; %s [%p]\n", OP_NAME(op), op));
emit_label(compiler, op); // TODO(dgl): Can we avoid doing this for every op?
sljit_emit_ijump(compiler, SLJIT_CALL0, SLJIT_IMM, (sljit_w) op->op_ppaddr);
sljit_emit_op1(compiler, SLJIT_MOV, SLJIT_MEM, (sljit_w) &PL_op, SLJIT_RETURN_REG, 0);
}
SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_enter(struct sljit_compiler *compiler,
sljit_s32 options, sljit_s32 args, sljit_s32 scratches, sljit_s32 saveds,
sljit_s32 fscratches, sljit_s32 fsaveds, sljit_s32 local_size)
{
sljit_s32 i, tmp, size, saved_register_size;
sljit_u8 *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);
#ifdef _WIN64
/* Two/four register slots for parameters plus space for xmm6 register if needed. */
if (fscratches >= 6 || fsaveds >= 1)
compiler->locals_offset = 6 * sizeof(sljit_sw);
else
compiler->locals_offset = ((scratches > 2) ? 4 : 2) * sizeof(sljit_sw);
#endif
/* 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_u8*)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_u8*)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_u8*)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_u8*)ensure_buf(compiler, 1 + 4 + (3 + sizeof(sljit_s32)));
FAIL_IF(!inst);
INC_SIZE(4 + (3 + sizeof(sljit_s32)));
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | SUB | reg_map[SLJIT_SP];
/* 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_ASSERT(reg_map[SLJIT_R0] < 8);
*inst++ = REX_W;
*inst++ = MOV_rm_i32;
*inst++ = MOD_REG | reg_lmap[SLJIT_R0];
sljit_unaligned_store_s32(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
if (local_size > 0) {
if (local_size <= 127) {
inst = (sljit_u8*)ensure_buf(compiler, 1 + 4);
FAIL_IF(!inst);
INC_SIZE(4);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_83;
*inst++ = MOD_REG | SUB | reg_map[SLJIT_SP];
*inst++ = local_size;
}
else {
inst = (sljit_u8*)ensure_buf(compiler, 1 + 7);
FAIL_IF(!inst);
INC_SIZE(7);
*inst++ = REX_W;
*inst++ = GROUP_BINARY_81;
*inst++ = MOD_REG | SUB | reg_map[SLJIT_SP];
sljit_unaligned_store_s32(inst, local_size);
inst += sizeof(sljit_s32);
}
}
#ifdef _WIN64
/* Save xmm6 register: movaps [rsp + 0x20], xmm6 */
if (fscratches >= 6 || fsaveds >= 1) {
inst = (sljit_u8*)ensure_buf(compiler, 1 + 5);
FAIL_IF(!inst);
INC_SIZE(5);
*inst++ = GROUP_0F;
sljit_unaligned_store_s32(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;
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
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
locals_offset = 2 * sizeof(sljit_uw);
#else
SLJIT_COMPILE_ASSERT(FIXED_LOCALS_OFFSET >= 2 * sizeof(sljit_uw), require_at_least_two_words);
locals_offset = FIXED_LOCALS_OFFSET;
#endif
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));
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_SCRATCH_REG1], local_size));
#else
local_size -= FIXED_LOCALS_OFFSET;
FAIL_IF(emit_do_imm(compiler, MOV_r_i32 + reg_map[SLJIT_SCRATCH_REG1], local_size));
FAIL_IF(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, FIXED_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_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, local_size);
}
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;
}
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 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 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);
}
