static sljit_s32 emit_mov_int(struct sljit_compiler *compiler, sljit_s32 sign,
sljit_s32 dst, sljit_sw dstw,
sljit_s32 src, sljit_sw srcw)
{
sljit_u8* inst;
sljit_s32 dst_r;
compiler->mode32 = 0;
if (dst == SLJIT_UNUSED && !(src & SLJIT_MEM))
return SLJIT_SUCCESS; /* Empty instruction. */
if (src & SLJIT_IMM) {
if (FAST_IS_REG(dst)) {
if (sign || ((sljit_uw)srcw <= 0x7fffffff)) {
inst = emit_x86_instruction(compiler, 1, SLJIT_IMM, (sljit_sw)(sljit_s32)srcw, dst, dstw);
FAIL_IF(!inst);
*inst = MOV_rm_i32;
return SLJIT_SUCCESS;
}
return emit_load_imm64(compiler, dst, srcw);
}
compiler->mode32 = 1;
inst = emit_x86_instruction(compiler, 1, SLJIT_IMM, (sljit_sw)(sljit_s32)srcw, dst, dstw);
FAIL_IF(!inst);
*inst = MOV_rm_i32;
compiler->mode32 = 0;
return SLJIT_SUCCESS;
}
dst_r = FAST_IS_REG(dst) ? dst : TMP_REG1;
if ((dst & SLJIT_MEM) && FAST_IS_REG(src))
dst_r = src;
else {
if (sign) {
inst = emit_x86_instruction(compiler, 1, dst_r, 0, src, srcw);
FAIL_IF(!inst);
*inst++ = MOVSXD_r_rm;
} else {
compiler->mode32 = 1;
FAIL_IF(emit_mov(compiler, dst_r, 0, src, srcw));
compiler->mode32 = 0;
}
}
if (dst & SLJIT_MEM) {
compiler->mode32 = 1;
inst = emit_x86_instruction(compiler, 1, dst_r, 0, dst, dstw);
FAIL_IF(!inst);
*inst = MOV_rm_r;
compiler->mode32 = 0;
}
return SLJIT_SUCCESS;
}
int sljit_emit_return(struct sljit_compiler *compiler, int src, sljit_w srcw)
{
int size;
sljit_ub *buf;
CHECK_ERROR();
check_sljit_emit_return(compiler, src, srcw);
SLJIT_ASSERT(compiler->args >= 0);
compiler->flags_saved = 0;
CHECK_EXTRA_REGS(src, srcw, (void)0);
if (src != SLJIT_UNUSED && src != SLJIT_RETURN_REG)
FAIL_IF(emit_mov(compiler, SLJIT_RETURN_REG, 0, src, srcw));
if (compiler->local_size > 0)
FAIL_IF(emit_cum_binary(compiler, 0x03, 0x01, 0x0 << 3, 0x05,
SLJIT_LOCALS_REG, 0, SLJIT_LOCALS_REG, 0, SLJIT_IMM, compiler->local_size));
size = 2 + (compiler->generals <= 3 ? compiler->generals : 3);
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (compiler->args > 2)
size += 2;
#else
if (compiler->args > 0)
size += 2;
#endif
buf = (sljit_ub*)ensure_buf(compiler, 1 + size);
FAIL_IF(!buf);
INC_SIZE(size);
if (compiler->generals > 0)
POP_REG(reg_map[SLJIT_GENERAL_REG1]);
if (compiler->generals > 1)
POP_REG(reg_map[SLJIT_GENERAL_REG2]);
if (compiler->generals > 2)
POP_REG(reg_map[SLJIT_GENERAL_REG3]);
POP_REG(reg_map[TMP_REGISTER]);
#if (defined SLJIT_X86_32_FASTCALL && SLJIT_X86_32_FASTCALL)
if (compiler->args > 2)
RETN(sizeof(sljit_w));
else
RET();
#else
if (compiler->args > 0)
RETN(compiler->args * sizeof(sljit_w));
else
RET();
#endif
return SLJIT_SUCCESS;
}
static void x86_emit_inst(Inst* inst, int* pc2addr, int rodata_addr) {
switch (inst->op) {
case MOV:
emit_mov(inst->dst.reg, &inst->src);
break;
case ADD:
if (inst->src.type == REG) {
emit_1(0x01);
emit_reg2(inst->dst.reg, inst->src.reg);
} else {
emit_2(0x81, 0xc0 + REGNO[inst->dst.reg]);
emit_le(inst->src.imm);
}
emit_2(0x81, 0xe0 + REGNO[inst->dst.reg]);
emit_le(0xffffff);
break;
case SUB:
if (inst->src.type == REG) {
emit_1(0x29);
emit_reg2(inst->dst.reg, inst->src.reg);
} else {
emit_2(0x81, 0xe8 + REGNO[inst->dst.reg]);
emit_le(inst->src.imm);
}
emit_2(0x81, 0xe0 + REGNO[inst->dst.reg]);
emit_le(0xffffff);
break;
case LOAD:
emit_1(0x8b);
goto store_load_common;
case STORE:
emit_1(0x89);
store_load_common:
if (inst->src.type == REG) {
emit_2(4 + (REGNO[inst->dst.reg] * 8),
0x86 + (REGNO[inst->src.reg] * 8));
} else {
emit_1(0x86 + (REGNO[inst->dst.reg] * 8));
emit_le(inst->src.imm * 4);
}
break;
case PUTC:
// push EDI
emit_1(0x57);
emit_mov(EDI, &inst->src);
// push EAX, ECX, EDX, EBX, EDI
emit_5(0x50, 0x51, 0x52, 0x53, 0x57);
emit_mov_imm(B, 1); // stdout
emit_mov_reg(C, ESP);
emit_mov_imm(D, 1);
emit_mov_imm(A, 4); // write
emit_int80();
// pop EDI, EBX, EDX, ECX, EAX
emit_5(0x5f, 0x5b, 0x5a, 0x59, 0x58);
// pop EDI
emit_1(0x5f);
break;
case GETC:
// push EDI
emit_1(0x57);
// push EAX, ECX, EDX, EBX
emit_4(0x50, 0x51, 0x52, 0x53);
// push 0
emit_2(0x6a, 0x00);
emit_mov_imm(B, 0); // stdin
emit_mov_reg(C, ESP);
emit_mov_imm(D, 1);
emit_mov_imm(A, 3); // read
emit_int80();
// pop EDI
emit_1(0x5f);
emit_mov_imm(B, 0);
// cmp EAX, 1
emit_3(0x83, 0xf8, 0x01);
// cmovnz EDI, EBX
emit_3(0x0f, 0x45, 0xfb);
// pop EBX, EDX, ECX, EAX
emit_4(0x5b, 0x5a, 0x59, 0x58);
emit_mov_reg(inst->dst.reg, EDI);
// pop EDI
emit_1(0x5f);
break;
case EXIT:
emit_mov_imm(B, 0);
emit_mov_imm(A, 1); // exit
emit_int80();
break;
case DUMP:
break;
case EQ:
emit_setcc(inst, 0x94);
break;
case NE:
emit_setcc(inst, 0x95);
break;
case LT:
emit_setcc(inst, 0x9c);
break;
case GT:
emit_setcc(inst, 0x9f);
break;
case LE:
emit_setcc(inst, 0x9e);
break;
case GE:
emit_setcc(inst, 0x9d);
break;
case JEQ:
emit_jcc(inst, 0x75, pc2addr, rodata_addr);
break;
case JNE:
emit_jcc(inst, 0x74, pc2addr, rodata_addr);
break;
case JLT:
emit_jcc(inst, 0x7d, pc2addr, rodata_addr);
break;
case JGT:
emit_jcc(inst, 0x7e, pc2addr, rodata_addr);
break;
case JLE:
emit_jcc(inst, 0x7f, pc2addr, rodata_addr);
break;
case JGE:
emit_jcc(inst, 0x7c, pc2addr, rodata_addr);
break;
case JMP:
emit_jcc(inst, 0, pc2addr, rodata_addr);
break;
default:
error("oops");
}
}
static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
{
#define MAX_IFSN 32
#define MAX_LOOP_DEPTH 32
struct brw_instruction *if_inst[MAX_IFSN], *loop_inst[MAX_LOOP_DEPTH];
struct brw_instruction *inst0, *inst1;
int i, if_insn = 0, loop_insn = 0;
struct brw_compile *p = &c->func;
struct brw_indirect stack_index = brw_indirect(0, 0);
c->reg_index = 0;
prealloc_reg(c);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));
for (i = 0; i < c->nr_fp_insns; i++) {
struct prog_instruction *inst = &c->prog_instructions[i];
struct prog_instruction *orig_inst;
if ((orig_inst = inst->Data) != 0)
orig_inst->Data = current_insn(p);
if (inst->CondUpdate)
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
else
brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
switch (inst->Opcode) {
case WM_PIXELXY:
emit_pixel_xy(c, inst);
break;
case WM_DELTAXY:
emit_delta_xy(c, inst);
break;
case WM_PIXELW:
emit_pixel_w(c, inst);
break;
case WM_LINTERP:
emit_linterp(c, inst);
break;
case WM_PINTERP:
emit_pinterp(c, inst);
break;
case WM_CINTERP:
emit_cinterp(c, inst);
break;
case WM_WPOSXY:
emit_wpos_xy(c, inst);
break;
case WM_FB_WRITE:
emit_fb_write(c, inst);
break;
case OPCODE_ABS:
emit_abs(c, inst);
break;
case OPCODE_ADD:
emit_add(c, inst);
break;
case OPCODE_SUB:
emit_sub(c, inst);
break;
case OPCODE_FRC:
emit_frc(c, inst);
break;
case OPCODE_FLR:
emit_flr(c, inst);
break;
case OPCODE_LRP:
emit_lrp(c, inst);
break;
case OPCODE_INT:
emit_int(c, inst);
break;
case OPCODE_MOV:
emit_mov(c, inst);
break;
case OPCODE_DP3:
emit_dp3(c, inst);
break;
case OPCODE_DP4:
emit_dp4(c, inst);
break;
case OPCODE_XPD:
emit_xpd(c, inst);
break;
case OPCODE_DPH:
emit_dph(c, inst);
break;
case OPCODE_RCP:
emit_rcp(c, inst);
break;
case OPCODE_RSQ:
emit_rsq(c, inst);
break;
case OPCODE_SIN:
emit_sin(c, inst);
break;
case OPCODE_COS:
emit_cos(c, inst);
break;
case OPCODE_EX2:
emit_ex2(c, inst);
break;
case OPCODE_LG2:
emit_lg2(c, inst);
break;
case OPCODE_MAX:
emit_max(c, inst);
break;
case OPCODE_MIN:
emit_min(c, inst);
break;
case OPCODE_DDX:
emit_ddx(c, inst);
break;
case OPCODE_DDY:
emit_ddy(c, inst);
break;
case OPCODE_SLT:
emit_slt(c, inst);
break;
case OPCODE_SLE:
emit_sle(c, inst);
break;
case OPCODE_SGT:
emit_sgt(c, inst);
break;
case OPCODE_SGE:
emit_sge(c, inst);
break;
case OPCODE_SEQ:
emit_seq(c, inst);
break;
case OPCODE_SNE:
emit_sne(c, inst);
break;
case OPCODE_MUL:
emit_mul(c, inst);
break;
case OPCODE_POW:
emit_pow(c, inst);
break;
case OPCODE_MAD:
emit_mad(c, inst);
break;
case OPCODE_TEX:
emit_tex(c, inst);
break;
case OPCODE_TXB:
emit_txb(c, inst);
break;
case OPCODE_KIL_NV:
emit_kil(c);
break;
case OPCODE_IF:
assert(if_insn < MAX_IFSN);
if_inst[if_insn++] = brw_IF(p, BRW_EXECUTE_8);
break;
case OPCODE_ELSE:
if_inst[if_insn-1] = brw_ELSE(p, if_inst[if_insn-1]);
break;
case OPCODE_ENDIF:
assert(if_insn > 0);
brw_ENDIF(p, if_inst[--if_insn]);
break;
case OPCODE_BGNSUB:
case OPCODE_ENDSUB:
break;
case OPCODE_CAL:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_ADD(p, deref_1ud(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(4));
orig_inst = inst->Data;
orig_inst->Data = &p->store[p->nr_insn];
brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
brw_pop_insn_state(p);
break;
case OPCODE_RET:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(-4));
brw_set_access_mode(p, BRW_ALIGN_1);
brw_MOV(p, brw_ip_reg(), deref_1ud(stack_index, 0));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_pop_insn_state(p);
break;
case OPCODE_BGNLOOP:
loop_inst[loop_insn++] = brw_DO(p, BRW_EXECUTE_8);
break;
case OPCODE_BRK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_CONT:
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_ENDLOOP:
loop_insn--;
inst0 = inst1 = brw_WHILE(p, loop_inst[loop_insn]);
/* patch all the BREAK instructions from
last BEGINLOOP */
while (inst0 > loop_inst[loop_insn]) {
inst0--;
if (inst0->header.opcode == BRW_OPCODE_BREAK) {
inst0->bits3.if_else.jump_count = inst1 - inst0 + 1;
inst0->bits3.if_else.pop_count = 0;
} else if (inst0->header.opcode == BRW_OPCODE_CONTINUE) {
inst0->bits3.if_else.jump_count = inst1 - inst0;
inst0->bits3.if_else.pop_count = 0;
}
}
break;
default:
_mesa_printf("unsupported IR in fragment shader %d\n",
inst->Opcode);
}
if (inst->CondUpdate)
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
else
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
post_wm_emit(c);
for (i = 0; i < c->fp->program.Base.NumInstructions; i++)
c->fp->program.Base.Instructions[i].Data = NULL;
}