diff options
Diffstat (limited to 'thirdparty/pcre2/src/sljit/sljitNativeX86_64.c')
| -rw-r--r-- | thirdparty/pcre2/src/sljit/sljitNativeX86_64.c | 838 |
1 files changed, 452 insertions, 386 deletions
diff --git a/thirdparty/pcre2/src/sljit/sljitNativeX86_64.c b/thirdparty/pcre2/src/sljit/sljitNativeX86_64.c index e85b56a61a..f37df6e1bf 100644 --- a/thirdparty/pcre2/src/sljit/sljitNativeX86_64.c +++ b/thirdparty/pcre2/src/sljit/sljitNativeX86_64.c @@ -26,6 +26,10 @@ /* x86 64-bit arch dependent functions. */ +/* --------------------------------------------------------------------- */ +/* Operators */ +/* --------------------------------------------------------------------- */ + static sljit_s32 emit_load_imm64(struct sljit_compiler *compiler, sljit_s32 reg, sljit_sw imm) { sljit_u8 *inst; @@ -34,14 +38,246 @@ static sljit_s32 emit_load_imm64(struct sljit_compiler *compiler, sljit_s32 reg, FAIL_IF(!inst); INC_SIZE(2 + sizeof(sljit_sw)); *inst++ = REX_W | ((reg_map[reg] <= 7) ? 0 : REX_B); - *inst++ = MOV_r_i32 + (reg_map[reg] & 0x7); + *inst++ = U8(MOV_r_i32 | (reg_map[reg] & 0x7)); sljit_unaligned_store_sw(inst, imm); return SLJIT_SUCCESS; } +static sljit_s32 emit_do_imm32(struct sljit_compiler *compiler, sljit_u8 rex, sljit_u8 opcode, sljit_sw imm) +{ + sljit_u8 *inst; + sljit_uw length = (rex ? 2 : 1) + sizeof(sljit_s32); + + inst = (sljit_u8*)ensure_buf(compiler, 1 + length); + FAIL_IF(!inst); + INC_SIZE(length); + if (rex) + *inst++ = rex; + *inst++ = opcode; + sljit_unaligned_store_s32(inst, (sljit_s32)imm); + return SLJIT_SUCCESS; +} + +static sljit_u8* emit_x86_instruction(struct sljit_compiler *compiler, sljit_uw 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_u8 reg_lmap_b; + sljit_uw flags = size; + sljit_uw 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_REG2, immb)); + immb = 0; + if (b & REG_MASK) + b |= TO_OFFS_REG(TMP_REG2); + else + b |= TMP_REG2; + } + else if (reg_lmap[b & REG_MASK] == 4) + b |= TO_OFFS_REG(SLJIT_SP); + } + + if (!(b & REG_MASK)) + 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) { + inst_size += 1; /* SIB byte. */ + if (reg_map[OFFS_REG(b)] >= 8) + rex |= REX_X; + } + } + } + else if (!(flags & EX86_SSE2_OP2)) { + if (reg_map[b] >= 8) + rex |= REX_B; + } + else if (freg_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)) { + if (reg_map[a] >= 8) + rex |= REX_R; + } + else if (freg_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) + *buf_ptr = 0; + else if (!(flags & EX86_SSE2_OP1)) + *buf_ptr = U8(reg_lmap[a] << 3); + else + *buf_ptr = U8(freg_lmap[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 = U8(*buf_ptr | MOD_REG | (!(flags & EX86_SSE2_OP2) ? reg_lmap[b] : freg_lmap[b])); + buf_ptr++; + } else if (b & REG_MASK) { + reg_lmap_b = reg_lmap[b & REG_MASK]; + + if (!(b & OFFS_REG_MASK) || (b & OFFS_REG_MASK) == TO_OFFS_REG(SLJIT_SP) || reg_lmap_b == 5) { + if (immb != 0 || reg_lmap_b == 5) { + if (immb <= 127 && immb >= -128) + *buf_ptr |= 0x40; + else + *buf_ptr |= 0x80; + } + + if (!(b & OFFS_REG_MASK)) + *buf_ptr++ |= reg_lmap_b; + else { + *buf_ptr++ |= 0x04; + *buf_ptr++ = U8(reg_lmap_b | (reg_lmap[OFFS_REG(b)] << 3)); + } + + if (immb != 0 || reg_lmap_b == 5) { + if (immb <= 127 && immb >= -128) + *buf_ptr++ = U8(immb); /* 8 bit displacement. */ + else { + sljit_unaligned_store_s32(buf_ptr, (sljit_s32)immb); /* 32 bit displacement. */ + buf_ptr += sizeof(sljit_s32); + } + } + } + else { + *buf_ptr++ |= 0x04; + *buf_ptr++ = U8(reg_lmap_b | (reg_lmap[OFFS_REG(b)] << 3) | (immb << 6)); + } + } + else { + *buf_ptr++ |= 0x04; + *buf_ptr++ = 0x25; + sljit_unaligned_store_s32(buf_ptr, (sljit_s32)immb); /* 32 bit displacement. */ + buf_ptr += sizeof(sljit_s32); + } + + if (a & SLJIT_IMM) { + if (flags & EX86_BYTE_ARG) + *buf_ptr = U8(imma); + else if (flags & EX86_HALF_ARG) + sljit_unaligned_store_s16(buf_ptr, (sljit_s16)imma); + else if (!(flags & EX86_SHIFT_INS)) + sljit_unaligned_store_s32(buf_ptr, (sljit_s32)imma); + } + + return !(flags & EX86_SHIFT_INS) ? inst : (inst + 1); +} + +/* --------------------------------------------------------------------- */ +/* Enter / return */ +/* --------------------------------------------------------------------- */ + static sljit_u8* generate_far_jump_code(struct sljit_jump *jump, sljit_u8 *code_ptr) { - sljit_s32 type = jump->flags >> TYPE_SHIFT; + sljit_uw type = jump->flags >> TYPE_SHIFT; int short_addr = !(jump->flags & SLJIT_REWRITABLE_JUMP) && !(jump->flags & JUMP_LABEL) && (jump->u.target <= 0xffffffff); @@ -50,7 +286,7 @@ static sljit_u8* generate_far_jump_code(struct sljit_jump *jump, sljit_u8 *code_ if (type < SLJIT_JUMP) { /* Invert type. */ - *code_ptr++ = get_jump_code(type ^ 0x1) - 0x10; + *code_ptr++ = U8(get_jump_code(type ^ 0x1) - 0x10); *code_ptr++ = short_addr ? (6 + 3) : (10 + 3); } @@ -63,13 +299,13 @@ static sljit_u8* generate_far_jump_code(struct sljit_jump *jump, sljit_u8 *code_ else if (short_addr) sljit_unaligned_store_s32(code_ptr, (sljit_s32)jump->u.target); else - sljit_unaligned_store_sw(code_ptr, jump->u.target); + sljit_unaligned_store_sw(code_ptr, (sljit_sw)jump->u.target); code_ptr += short_addr ? sizeof(sljit_s32) : sizeof(sljit_sw); *code_ptr++ = REX_B; *code_ptr++ = GROUP_FF; - *code_ptr++ = MOD_REG | (type >= SLJIT_FAST_CALL ? CALL_rm : JMP_rm) | reg_lmap[TMP_REG2]; + *code_ptr++ = U8(MOD_REG | (type >= SLJIT_FAST_CALL ? CALL_rm : JMP_rm) | reg_lmap[TMP_REG2]); return code_ptr; } @@ -90,7 +326,7 @@ static sljit_u8* generate_put_label_code(struct sljit_put_label *put_label, slji SLJIT_ASSERT((code_ptr[1] & 0xf8) == MOV_r_i32); if ((code_ptr[0] & 0x07) != 0) { - code_ptr[0] = (sljit_u8)(code_ptr[0] & ~0x08); + code_ptr[0] = U8(code_ptr[0] & ~0x08); code_ptr += 2 + sizeof(sljit_s32); } else { @@ -114,9 +350,9 @@ static sljit_u8* generate_put_label_code(struct sljit_put_label *put_label, slji SLJIT_ASSERT(code_ptr[1] == MOV_rm_r); - code_ptr[0] = (sljit_u8)(code_ptr[0] & ~0x4); + code_ptr[0] = U8(code_ptr[0] & ~0x4); code_ptr[1] = MOV_rm_i32; - code_ptr[2] = (sljit_u8)(code_ptr[2] & ~(0x7 << 3)); + code_ptr[2] = U8(code_ptr[2] & ~(0x7 << 3)); code_ptr = (sljit_u8*)(put_label->addr - (2 + sizeof(sljit_uw)) + sizeof(sljit_s32)); put_label->addr = (sljit_uw)code_ptr; @@ -128,7 +364,15 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_enter(struct sljit_compiler *compi sljit_s32 options, sljit_s32 arg_types, sljit_s32 scratches, sljit_s32 saveds, sljit_s32 fscratches, sljit_s32 fsaveds, sljit_s32 local_size) { - sljit_s32 args, i, tmp, size, saved_register_size; + sljit_uw size; + sljit_s32 word_arg_count = 0; + sljit_s32 saved_arg_count = 0; + sljit_s32 saved_regs_size, tmp, i; +#ifdef _WIN64 + sljit_s32 saved_float_regs_size; + sljit_s32 saved_float_regs_offset = 0; + sljit_s32 float_arg_count = 0; +#endif /* _WIN64 */ sljit_u8 *inst; CHECK_ERROR(); @@ -140,19 +384,11 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_enter(struct sljit_compiler *compi compiler->mode32 = 0; -#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); + saved_regs_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--) { + tmp = SLJIT_S0 - saveds; + 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); @@ -172,55 +408,75 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_enter(struct sljit_compiler *compi PUSH_REG(reg_lmap[i]); } - args = get_arg_count(arg_types); +#ifdef _WIN64 + local_size += SLJIT_LOCALS_OFFSET; + saved_float_regs_size = GET_SAVED_FLOAT_REGISTERS_SIZE(fscratches, fsaveds, 16); - if (args > 0) { - size = args * 3; - inst = (sljit_u8*)ensure_buf(compiler, 1 + size); - FAIL_IF(!inst); + if (saved_float_regs_size > 0) { + saved_float_regs_offset = ((local_size + 0xf) & ~0xf); + local_size = saved_float_regs_offset + saved_float_regs_size; + } +#else /* !_WIN64 */ + SLJIT_ASSERT(SLJIT_LOCALS_OFFSET == 0); +#endif /* _WIN64 */ - INC_SIZE(size); + arg_types >>= SLJIT_ARG_SHIFT; + while (arg_types > 0) { + if ((arg_types & SLJIT_ARG_MASK) < SLJIT_ARG_TYPE_F64) { + tmp = 0; #ifndef _WIN64 - if (args > 0) { - inst[0] = REX_W; - inst[1] = MOV_r_rm; - inst[2] = MOD_REG | (reg_map[SLJIT_S0] << 3) | 0x7 /* rdi */; - inst += 3; - } - if (args > 1) { - inst[0] = REX_W | REX_R; - inst[1] = MOV_r_rm; - inst[2] = MOD_REG | (reg_lmap[SLJIT_S1] << 3) | 0x6 /* rsi */; - inst += 3; - } - if (args > 2) { - inst[0] = REX_W | REX_R; - inst[1] = MOV_r_rm; - inst[2] = MOD_REG | (reg_lmap[SLJIT_S2] << 3) | 0x2 /* rdx */; - } -#else - if (args > 0) { - inst[0] = REX_W; - inst[1] = MOV_r_rm; - inst[2] = MOD_REG | (reg_map[SLJIT_S0] << 3) | 0x1 /* rcx */; - inst += 3; - } - if (args > 1) { - inst[0] = REX_W; - inst[1] = MOV_r_rm; - inst[2] = MOD_REG | (reg_map[SLJIT_S1] << 3) | 0x2 /* rdx */; - inst += 3; - } - if (args > 2) { - inst[0] = REX_W | REX_B; - inst[1] = MOV_r_rm; - inst[2] = MOD_REG | (reg_map[SLJIT_S2] << 3) | 0x0 /* r8 */; + switch (word_arg_count) { + case 0: + tmp = SLJIT_R2; + break; + case 1: + tmp = SLJIT_R1; + break; + case 2: + tmp = TMP_REG1; + break; + default: + tmp = SLJIT_R3; + break; + } +#else /* !_WIN64 */ + switch (word_arg_count + float_arg_count) { + case 0: + tmp = SLJIT_R3; + break; + case 1: + tmp = SLJIT_R1; + break; + case 2: + tmp = SLJIT_R2; + break; + default: + tmp = TMP_REG1; + break; + } +#endif /* _WIN64 */ + if (arg_types & SLJIT_ARG_TYPE_SCRATCH_REG) { + if (tmp != SLJIT_R0 + word_arg_count) + EMIT_MOV(compiler, SLJIT_R0 + word_arg_count, 0, tmp, 0); + } else { + EMIT_MOV(compiler, SLJIT_S0 - saved_arg_count, 0, tmp, 0); + saved_arg_count++; + } + word_arg_count++; + } else { +#ifdef _WIN64 + SLJIT_COMPILE_ASSERT(SLJIT_FR0 == 1, float_register_index_start); + float_arg_count++; + if (float_arg_count != float_arg_count + word_arg_count) + FAIL_IF(emit_sse2_load(compiler, (arg_types & SLJIT_ARG_MASK) == SLJIT_ARG_TYPE_F32, + float_arg_count, float_arg_count + word_arg_count, 0)); +#endif /* _WIN64 */ } -#endif + arg_types >>= SLJIT_ARG_SHIFT; } - local_size = ((local_size + SLJIT_LOCALS_OFFSET + saved_register_size + 15) & ~15) - saved_register_size; + local_size = ((local_size + saved_regs_size + 0xf) & ~0xf) - saved_regs_size; compiler->local_size = local_size; #ifdef _WIN64 @@ -234,44 +490,49 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_enter(struct sljit_compiler *compi EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_MEM1(SLJIT_SP), -4096 * 3); } else { - EMIT_MOV(compiler, SLJIT_R0, 0, SLJIT_SP, 0); - EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_IMM, (local_size - 1) >> 12); + EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_IMM, local_size >> 12); - SLJIT_ASSERT (reg_map[SLJIT_R0] == 0); - - EMIT_MOV(compiler, TMP_REG2, 0, SLJIT_MEM1(SLJIT_R0), -4096); - FAIL_IF(emit_non_cum_binary(compiler, BINARY_OPCODE(SUB), - SLJIT_R0, 0, SLJIT_R0, 0, SLJIT_IMM, 4096)); - FAIL_IF(emit_non_cum_binary(compiler, BINARY_OPCODE(SUB), - TMP_REG1, 0, TMP_REG1, 0, SLJIT_IMM, 1)); + EMIT_MOV(compiler, TMP_REG2, 0, SLJIT_MEM1(SLJIT_SP), -4096); + BINARY_IMM32(SUB, 4096, SLJIT_SP, 0); + BINARY_IMM32(SUB, 1, TMP_REG1, 0); inst = (sljit_u8*)ensure_buf(compiler, 1 + 2); FAIL_IF(!inst); INC_SIZE(2); inst[0] = JNE_i8; - inst[1] = (sljit_s8) -19; + inst[1] = (sljit_u8)-21; + local_size &= 0xfff; } - EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_MEM1(SLJIT_SP), -local_size); + if (local_size > 0) + EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_MEM1(SLJIT_SP), -local_size); } -#endif +#endif /* _WIN64 */ - if (local_size > 0) { - FAIL_IF(emit_non_cum_binary(compiler, BINARY_OPCODE(SUB), - SLJIT_SP, 0, SLJIT_SP, 0, SLJIT_IMM, local_size)); - } + if (local_size > 0) + BINARY_IMM32(SUB, local_size, SLJIT_SP, 0); #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); + if (saved_float_regs_size > 0) { + compiler->mode32 = 1; + + tmp = SLJIT_FS0 - fsaveds; + for (i = SLJIT_FS0; i > tmp; i--) { + inst = emit_x86_instruction(compiler, 2 | EX86_SSE2, i, 0, SLJIT_MEM1(SLJIT_SP), saved_float_regs_offset); + *inst++ = GROUP_0F; + *inst = MOVAPS_xm_x; + saved_float_regs_offset += 16; + } + + for (i = fscratches; i >= SLJIT_FIRST_SAVED_FLOAT_REG; i--) { + inst = emit_x86_instruction(compiler, 2 | EX86_SSE2, i, 0, SLJIT_MEM1(SLJIT_SP), saved_float_regs_offset); + *inst++ = GROUP_0F; + *inst = MOVAPS_xm_x; + saved_float_regs_offset += 16; + } } -#endif +#endif /* _WIN64 */ return SLJIT_SUCCESS; } @@ -280,46 +541,65 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_set_context(struct sljit_compiler *comp sljit_s32 options, sljit_s32 arg_types, sljit_s32 scratches, sljit_s32 saveds, sljit_s32 fscratches, sljit_s32 fsaveds, sljit_s32 local_size) { - sljit_s32 saved_register_size; + sljit_s32 saved_regs_size; +#ifdef _WIN64 + sljit_s32 saved_float_regs_size; +#endif /* _WIN64 */ CHECK_ERROR(); CHECK(check_sljit_set_context(compiler, options, arg_types, scratches, saveds, fscratches, fsaveds, local_size)); set_set_context(compiler, options, arg_types, 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 + local_size += SLJIT_LOCALS_OFFSET; + saved_float_regs_size = GET_SAVED_FLOAT_REGISTERS_SIZE(fscratches, fsaveds, 16); + + if (saved_float_regs_size > 0) + local_size = ((local_size + 0xf) & ~0xf) + saved_float_regs_size; +#else /* !_WIN64 */ + SLJIT_ASSERT(SLJIT_LOCALS_OFFSET == 0); +#endif /* _WIN64 */ /* Including the return address saved by the call instruction. */ - saved_register_size = GET_SAVED_REGISTERS_SIZE(scratches, saveds, 1); - compiler->local_size = ((local_size + SLJIT_LOCALS_OFFSET + saved_register_size + 15) & ~15) - saved_register_size; + saved_regs_size = GET_SAVED_REGISTERS_SIZE(scratches, saveds, 1); + compiler->local_size = ((local_size + saved_regs_size + 0xf) & ~0xf) - saved_regs_size; return SLJIT_SUCCESS; } -SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_return(struct sljit_compiler *compiler, sljit_s32 op, sljit_s32 src, sljit_sw srcw) +static sljit_s32 emit_stack_frame_release(struct sljit_compiler *compiler) { - sljit_s32 i, tmp, size; + sljit_uw size; + sljit_s32 i, tmp; sljit_u8 *inst; +#ifdef _WIN64 + sljit_s32 saved_float_regs_offset; + sljit_s32 fscratches = compiler->fscratches; + sljit_s32 fsaveds = compiler->fsaveds; +#endif /* _WIN64 */ - CHECK_ERROR(); - CHECK(check_sljit_emit_return(compiler, op, src, srcw)); +#ifdef _WIN64 + saved_float_regs_offset = GET_SAVED_FLOAT_REGISTERS_SIZE(fscratches, fsaveds, 16); - FAIL_IF(emit_mov_before_return(compiler, op, src, srcw)); + if (saved_float_regs_offset > 0) { + compiler->mode32 = 1; + saved_float_regs_offset = (compiler->local_size - saved_float_regs_offset) & ~0xf; + + tmp = SLJIT_FS0 - fsaveds; + for (i = SLJIT_FS0; i > tmp; i--) { + inst = emit_x86_instruction(compiler, 2 | EX86_SSE2, i, 0, SLJIT_MEM1(SLJIT_SP), saved_float_regs_offset); + *inst++ = GROUP_0F; + *inst = MOVAPS_x_xm; + saved_float_regs_offset += 16; + } -#ifdef _WIN64 - /* Restore xmm6 register: movaps xmm6, [rsp + 0x20] */ - if (compiler->fscratches >= 6 || compiler->fsaveds >= 1) { - inst = (sljit_u8*)ensure_buf(compiler, 1 + 5); - FAIL_IF(!inst); - INC_SIZE(5); - *inst++ = GROUP_0F; - sljit_unaligned_store_s32(inst, 0x20247428); + for (i = fscratches; i >= SLJIT_FIRST_SAVED_FLOAT_REG; i--) { + inst = emit_x86_instruction(compiler, 2 | EX86_SSE2, i, 0, SLJIT_MEM1(SLJIT_SP), saved_float_regs_offset); + *inst++ = GROUP_0F; + *inst = MOVAPS_x_xm; + saved_float_regs_offset += 16; + } } -#endif +#endif /* _WIN64 */ if (compiler->local_size > 0) { if (compiler->local_size <= 127) { @@ -329,7 +609,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_return(struct sljit_compiler *comp *inst++ = REX_W; *inst++ = GROUP_BINARY_83; *inst++ = MOD_REG | ADD | 4; - *inst = compiler->local_size; + *inst = U8(compiler->local_size); } else { inst = (sljit_u8*)ensure_buf(compiler, 1 + 7); @@ -364,243 +644,23 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_return(struct sljit_compiler *comp POP_REG(reg_lmap[i]); } - inst = (sljit_u8*)ensure_buf(compiler, 1 + 1); - FAIL_IF(!inst); - INC_SIZE(1); - RET(); - return SLJIT_SUCCESS; -} - -/* --------------------------------------------------------------------- */ -/* Operators */ -/* --------------------------------------------------------------------- */ - -static sljit_s32 emit_do_imm32(struct sljit_compiler *compiler, sljit_u8 rex, sljit_u8 opcode, sljit_sw imm) -{ - sljit_u8 *inst; - sljit_s32 length = 1 + (rex ? 1 : 0) + sizeof(sljit_s32); - - inst = (sljit_u8*)ensure_buf(compiler, 1 + length); - FAIL_IF(!inst); - INC_SIZE(length); - if (rex) - *inst++ = rex; - *inst++ = opcode; - sljit_unaligned_store_s32(inst, imm); return SLJIT_SUCCESS; } -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_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_return_void(struct sljit_compiler *compiler) { 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_REG2, immb)); - immb = 0; - if (b & REG_MASK) - b |= TO_OFFS_REG(TMP_REG2); - else - b |= TMP_REG2; - } - 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)) { - if (reg_map[b] >= 8) - rex |= REX_B; - } - else if (freg_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)) { - if (reg_map[a] >= 8) - rex |= REX_R; - } - else if (freg_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) - *buf_ptr = 0; - else if (!(flags & EX86_SSE2_OP1)) - *buf_ptr = reg_lmap[a] << 3; - else - *buf_ptr = freg_lmap[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] : freg_lmap[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); - } + CHECK_ERROR(); + CHECK(check_sljit_emit_return_void(compiler)); - 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); - } + FAIL_IF(emit_stack_frame_release(compiler)); - return !(flags & EX86_SHIFT_INS) ? inst : (inst + 1); + inst = (sljit_u8*)ensure_buf(compiler, 1 + 1); + FAIL_IF(!inst); + INC_SIZE(1); + RET(); + return SLJIT_SUCCESS; } /* --------------------------------------------------------------------- */ @@ -609,43 +669,38 @@ static sljit_u8* emit_x86_instruction(struct sljit_compiler *compiler, sljit_s32 #ifndef _WIN64 -static sljit_s32 call_with_args(struct sljit_compiler *compiler, sljit_s32 arg_types, sljit_s32 *src_ptr, sljit_sw srcw) +static sljit_s32 call_with_args(struct sljit_compiler *compiler, sljit_s32 arg_types, sljit_s32 *src_ptr) { sljit_s32 src = src_ptr ? (*src_ptr) : 0; sljit_s32 word_arg_count = 0; SLJIT_ASSERT(reg_map[SLJIT_R1] == 6 && reg_map[SLJIT_R3] == 1 && reg_map[TMP_REG1] == 2); - - compiler->mode32 = 0; + SLJIT_ASSERT(!(src & SLJIT_MEM)); /* Remove return value. */ - arg_types >>= SLJIT_DEF_SHIFT; + arg_types >>= SLJIT_ARG_SHIFT; while (arg_types) { - if ((arg_types & SLJIT_DEF_MASK) < SLJIT_ARG_TYPE_F32) + if ((arg_types & SLJIT_ARG_MASK) < SLJIT_ARG_TYPE_F64) word_arg_count++; - arg_types >>= SLJIT_DEF_SHIFT; + arg_types >>= SLJIT_ARG_SHIFT; } if (word_arg_count == 0) return SLJIT_SUCCESS; - if (src & SLJIT_MEM) { - ADJUST_LOCAL_OFFSET(src, srcw); - EMIT_MOV(compiler, TMP_REG2, 0, src, srcw); - *src_ptr = TMP_REG2; + if (word_arg_count >= 3) { + if (src == SLJIT_R2) + *src_ptr = TMP_REG1; + EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_R2, 0); } - else if (src == SLJIT_R2 && word_arg_count >= SLJIT_R2) - *src_ptr = TMP_REG1; - if (word_arg_count >= 3) - EMIT_MOV(compiler, TMP_REG1, 0, SLJIT_R2, 0); return emit_mov(compiler, SLJIT_R2, 0, SLJIT_R0, 0); } #else -static sljit_s32 call_with_args(struct sljit_compiler *compiler, sljit_s32 arg_types, sljit_s32 *src_ptr, sljit_sw srcw) +static sljit_s32 call_with_args(struct sljit_compiler *compiler, sljit_s32 arg_types, sljit_s32 *src_ptr) { sljit_s32 src = src_ptr ? (*src_ptr) : 0; sljit_s32 arg_count = 0; @@ -656,16 +711,16 @@ static sljit_s32 call_with_args(struct sljit_compiler *compiler, sljit_s32 arg_t static sljit_u8 word_arg_regs[5] = { 0, SLJIT_R3, SLJIT_R1, SLJIT_R2, TMP_REG1 }; SLJIT_ASSERT(reg_map[SLJIT_R3] == 1 && reg_map[SLJIT_R1] == 2 && reg_map[SLJIT_R2] == 8 && reg_map[TMP_REG1] == 9); + SLJIT_ASSERT(!(src & SLJIT_MEM)); - compiler->mode32 = 0; - arg_types >>= SLJIT_DEF_SHIFT; + arg_types >>= SLJIT_ARG_SHIFT; while (arg_types) { - types = (types << SLJIT_DEF_SHIFT) | (arg_types & SLJIT_DEF_MASK); + types = (types << SLJIT_ARG_SHIFT) | (arg_types & SLJIT_ARG_MASK); - switch (arg_types & SLJIT_DEF_MASK) { - case SLJIT_ARG_TYPE_F32: + switch (arg_types & SLJIT_ARG_MASK) { case SLJIT_ARG_TYPE_F64: + case SLJIT_ARG_TYPE_F32: arg_count++; float_arg_count++; @@ -687,29 +742,23 @@ static sljit_s32 call_with_args(struct sljit_compiler *compiler, sljit_s32 arg_t break; } - arg_types >>= SLJIT_DEF_SHIFT; + arg_types >>= SLJIT_ARG_SHIFT; } if (!data_trandfer) return SLJIT_SUCCESS; - if (src & SLJIT_MEM) { - ADJUST_LOCAL_OFFSET(src, srcw); - EMIT_MOV(compiler, TMP_REG2, 0, src, srcw); - *src_ptr = TMP_REG2; - } - while (types) { - switch (types & SLJIT_DEF_MASK) { - case SLJIT_ARG_TYPE_F32: + switch (types & SLJIT_ARG_MASK) { + case SLJIT_ARG_TYPE_F64: if (arg_count != float_arg_count) - FAIL_IF(emit_sse2_load(compiler, 1, arg_count, float_arg_count, 0)); + FAIL_IF(emit_sse2_load(compiler, 0, arg_count, float_arg_count, 0)); arg_count--; float_arg_count--; break; - case SLJIT_ARG_TYPE_F64: + case SLJIT_ARG_TYPE_F32: if (arg_count != float_arg_count) - FAIL_IF(emit_sse2_load(compiler, 0, arg_count, float_arg_count, 0)); + FAIL_IF(emit_sse2_load(compiler, 1, arg_count, float_arg_count, 0)); arg_count--; float_arg_count--; break; @@ -721,7 +770,7 @@ static sljit_s32 call_with_args(struct sljit_compiler *compiler, sljit_s32 arg_t break; } - types >>= SLJIT_DEF_SHIFT; + types >>= SLJIT_ARG_SHIFT; } return SLJIT_SUCCESS; @@ -735,13 +784,19 @@ SLJIT_API_FUNC_ATTRIBUTE struct sljit_jump* sljit_emit_call(struct sljit_compile CHECK_ERROR_PTR(); CHECK_PTR(check_sljit_emit_call(compiler, type, arg_types)); - PTR_FAIL_IF(call_with_args(compiler, arg_types, NULL, 0)); + compiler->mode32 = 0; + + PTR_FAIL_IF(call_with_args(compiler, arg_types, NULL)); + + if (type & SLJIT_CALL_RETURN) { + PTR_FAIL_IF(emit_stack_frame_release(compiler)); + type = SLJIT_JUMP | (type & SLJIT_REWRITABLE_JUMP); + } #if (defined SLJIT_VERBOSE && SLJIT_VERBOSE) \ || (defined SLJIT_ARGUMENT_CHECKS && SLJIT_ARGUMENT_CHECKS) compiler->skip_checks = 1; #endif - return sljit_emit_jump(compiler, type); } @@ -752,7 +807,25 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_icall(struct sljit_compiler *compi CHECK_ERROR(); CHECK(check_sljit_emit_icall(compiler, type, arg_types, src, srcw)); - FAIL_IF(call_with_args(compiler, arg_types, &src, srcw)); + compiler->mode32 = 0; + + if (src & SLJIT_MEM) { + ADJUST_LOCAL_OFFSET(src, srcw); + EMIT_MOV(compiler, TMP_REG2, 0, src, srcw); + src = TMP_REG2; + } + + if (type & SLJIT_CALL_RETURN) { + if (src >= SLJIT_FIRST_SAVED_REG && src <= SLJIT_S0) { + EMIT_MOV(compiler, TMP_REG2, 0, src, srcw); + src = TMP_REG2; + } + + FAIL_IF(emit_stack_frame_release(compiler)); + type = SLJIT_JUMP; + } + + FAIL_IF(call_with_args(compiler, arg_types, &src)); #if (defined SLJIT_VERBOSE && SLJIT_VERBOSE) \ || (defined SLJIT_ARGUMENT_CHECKS && SLJIT_ARGUMENT_CHECKS) @@ -770,10 +843,6 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_emit_fast_enter(struct sljit_compiler * CHECK(check_sljit_emit_fast_enter(compiler, dst, dstw)); ADJUST_LOCAL_OFFSET(dst, dstw); - /* For UNUSED dst. Uncommon, but possible. */ - if (dst == SLJIT_UNUSED) - dst = TMP_REG1; - if (FAST_IS_REG(dst)) { if (reg_map[dst] < 8) { inst = (sljit_u8*)ensure_buf(compiler, 1 + 1); @@ -850,9 +919,6 @@ static sljit_s32 emit_mov_int(struct sljit_compiler *compiler, sljit_s32 sign, 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)) { @@ -903,16 +969,16 @@ static sljit_s32 skip_frames_before_return(struct sljit_compiler *compiler) sljit_s32 tmp, size; /* Don't adjust shadow stack if it isn't enabled. */ - if (!cpu_has_shadow_stack ()) + if (!cpu_has_shadow_stack()) return SLJIT_SUCCESS; size = compiler->local_size; tmp = compiler->scratches; if (tmp >= SLJIT_FIRST_SAVED_REG) - size += (tmp - SLJIT_FIRST_SAVED_REG + 1) * sizeof(sljit_uw); + size += (tmp - SLJIT_FIRST_SAVED_REG + 1) * SSIZE_OF(sw); tmp = compiler->saveds < SLJIT_NUMBER_OF_SAVED_REGISTERS ? (SLJIT_S0 + 1 - compiler->saveds) : SLJIT_FIRST_SAVED_REG; if (SLJIT_S0 >= tmp) - size += (SLJIT_S0 - tmp + 1) * sizeof(sljit_uw); + size += (SLJIT_S0 - tmp + 1) * SSIZE_OF(sw); - return adjust_shadow_stack(compiler, SLJIT_UNUSED, 0, SLJIT_SP, size); + return adjust_shadow_stack(compiler, SLJIT_MEM1(SLJIT_SP), size); } |