block->pc + (offset << 2));
}
+static void
+lightrec_jump_to_eob(struct lightrec_cstate *state, jit_state_t *_jit)
+{
+ /* Prevent jit_jmpi() from using our cycles register as a temporary */
+ jit_live(LIGHTREC_REG_CYCLE);
+
+ jit_patch_abs(jit_jmpi(), state->state->eob_wrapper_func);
+}
+
static void lightrec_emit_end_of_block(struct lightrec_cstate *state,
const struct block *block, u16 offset,
s8 reg_new_pc, u32 imm, u8 ra_reg,
u32 link, bool update_cycles)
{
struct regcache *reg_cache = state->reg_cache;
- u32 cycles = state->cycles;
jit_state_t *_jit = block->_jit;
const struct opcode *op = &block->opcode_list[offset],
*next = &block->opcode_list[offset + 1];
+ u32 cycles = state->cycles + lightrec_cycles_of_opcode(op->c);
jit_note(__FILE__, __LINE__);
}
if (has_delay_slot(op->c) &&
- !(op->flags & (LIGHTREC_NO_DS | LIGHTREC_LOCAL_BRANCH))) {
+ !op_flag_no_ds(op->flags) && !op_flag_local_branch(op->flags)) {
cycles += lightrec_cycles_of_opcode(next->c);
/* Recompile the delay slot */
lightrec_rec_opcode(state, block, offset + 1);
}
- /* Store back remaining registers */
- lightrec_storeback_regs(reg_cache, _jit);
+ /* Clean the remaining registers */
+ lightrec_clean_regs(reg_cache, _jit);
jit_movr(JIT_V0, reg_new_pc);
pr_debug("EOB: %u cycles\n", cycles);
}
- if (offset + !!(op->flags & LIGHTREC_NO_DS) < block->nb_ops - 1)
- state->branches[state->nb_branches++] = jit_jmpi();
+ lightrec_jump_to_eob(state, _jit);
}
-void lightrec_emit_eob(struct lightrec_cstate *state, const struct block *block,
- u16 offset, bool after_op)
+void lightrec_emit_eob(struct lightrec_cstate *state,
+ const struct block *block, u16 offset)
{
struct regcache *reg_cache = state->reg_cache;
jit_state_t *_jit = block->_jit;
- union code c = block->opcode_list[offset].c;
- u32 cycles = state->cycles;
-
- if (!after_op)
- cycles -= lightrec_cycles_of_opcode(c);
- lightrec_storeback_regs(reg_cache, _jit);
+ lightrec_clean_regs(reg_cache, _jit);
jit_movi(JIT_V0, block->pc + (offset << 2));
- jit_subi(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, cycles);
+ jit_subi(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, state->cycles);
- state->branches[state->nb_branches++] = jit_jmpi();
+ lightrec_jump_to_eob(state, _jit);
}
static u8 get_jr_jalr_reg(struct lightrec_cstate *state, const struct block *block, u16 offset)
{
struct regcache *reg_cache = state->reg_cache;
jit_state_t *_jit = block->_jit;
- const struct opcode *op = &block->opcode_list[offset],
- *next = &block->opcode_list[offset + 1];
- u8 rs = lightrec_request_reg_in(reg_cache, _jit, op->r.rs, JIT_V0);
-
- /* If the source register is already mapped to JIT_R0 or JIT_R1, and the
- * delay slot is a I/O operation, unload the register, since JIT_R0 and
- * JIT_R1 are explicitely used by the I/O opcode generators. */
- if ((rs == JIT_R0 || rs == JIT_R1) &&
- !(op->flags & LIGHTREC_NO_DS) &&
- opcode_is_io(next->c) &&
- !(next->flags & (LIGHTREC_NO_INVALIDATE | LIGHTREC_DIRECT_IO))) {
- lightrec_unload_reg(reg_cache, _jit, rs);
- lightrec_free_reg(reg_cache, rs);
-
- rs = lightrec_request_reg_in(reg_cache, _jit, op->r.rs, JIT_V0);
- }
+ const struct opcode *op = &block->opcode_list[offset];
+ u8 rs;
+ rs = lightrec_request_reg_in(reg_cache, _jit, op->r.rs, JIT_V0);
lightrec_lock_reg(reg_cache, _jit, rs);
return rs;
31, get_branch_pc(block, offset, 2), true);
}
+static void lightrec_do_early_unload(struct lightrec_cstate *state,
+ const struct block *block, u16 offset)
+{
+ struct regcache *reg_cache = state->reg_cache;
+ const struct opcode *op = &block->opcode_list[offset];
+ jit_state_t *_jit = block->_jit;
+ unsigned int i;
+ u8 reg;
+ struct {
+ u8 reg, op;
+ } reg_ops[3] = {
+ { op->r.rd, LIGHTREC_FLAGS_GET_RD(op->flags), },
+ { op->i.rt, LIGHTREC_FLAGS_GET_RT(op->flags), },
+ { op->i.rs, LIGHTREC_FLAGS_GET_RS(op->flags), },
+ };
+
+ for (i = 0; i < ARRAY_SIZE(reg_ops); i++) {
+ reg = reg_ops[i].reg;
+
+ switch (reg_ops[i].op) {
+ case LIGHTREC_REG_UNLOAD:
+ lightrec_clean_reg_if_loaded(reg_cache, _jit, reg, true);
+ break;
+
+ case LIGHTREC_REG_DISCARD:
+ lightrec_discard_reg_if_loaded(reg_cache, reg);
+ break;
+
+ case LIGHTREC_REG_CLEAN:
+ lightrec_clean_reg_if_loaded(reg_cache, _jit, reg, false);
+ break;
+ default:
+ break;
+ };
+ }
+}
+
static void rec_b(struct lightrec_cstate *state, const struct block *block, u16 offset,
- jit_code_t code, u32 link, bool unconditional, bool bz)
+ jit_code_t code, jit_code_t code2, u32 link, bool unconditional, bool bz)
{
struct regcache *reg_cache = state->reg_cache;
struct native_register *regs_backup;
const struct opcode *op = &block->opcode_list[offset],
*next = &block->opcode_list[offset + 1];
jit_node_t *addr;
- u8 link_reg;
- u32 target_offset, cycles = state->cycles;
+ u8 link_reg, rs, rt;
bool is_forward = (s16)op->i.imm >= -1;
+ int op_cycles = lightrec_cycles_of_opcode(op->c);
+ u32 target_offset, cycles = state->cycles + op_cycles;
+ bool no_indirection = false;
u32 next_pc;
jit_note(__FILE__, __LINE__);
- if (!(op->flags & LIGHTREC_NO_DS))
+ if (!op_flag_no_ds(op->flags))
cycles += lightrec_cycles_of_opcode(next->c);
- state->cycles = 0;
+ state->cycles = -op_cycles;
+
+ if (!unconditional) {
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, op->i.rs, REG_EXT);
+ rt = bz ? 0 : lightrec_alloc_reg_in(reg_cache,
+ _jit, op->i.rt, REG_EXT);
+
+ /* Unload dead registers before evaluating the branch */
+ if (OPT_EARLY_UNLOAD)
+ lightrec_do_early_unload(state, block, offset);
+
+ if (op_flag_local_branch(op->flags) &&
+ (op_flag_no_ds(op->flags) || !next->opcode) &&
+ is_forward && !lightrec_has_dirty_regs(reg_cache))
+ no_indirection = true;
+
+ if (no_indirection)
+ pr_debug("Using no indirection for branch at offset 0x%hx\n", offset << 2);
+ }
if (cycles)
jit_subi(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, cycles);
if (!unconditional) {
- u8 rs = lightrec_alloc_reg_in(reg_cache, _jit, op->i.rs, REG_EXT),
- rt = bz ? 0 : lightrec_alloc_reg_in(reg_cache,
- _jit, op->i.rt, REG_EXT);
-
/* Generate the branch opcode */
- addr = jit_new_node_pww(code, NULL, rs, rt);
+ if (!no_indirection)
+ addr = jit_new_node_pww(code, NULL, rs, rt);
lightrec_free_regs(reg_cache);
regs_backup = lightrec_regcache_enter_branch(reg_cache);
}
- if (op->flags & LIGHTREC_LOCAL_BRANCH) {
- if (next && !(op->flags & LIGHTREC_NO_DS)) {
- /* Recompile the delay slot */
- if (next->opcode)
- lightrec_rec_opcode(state, block, offset + 1);
- }
+ if (op_flag_local_branch(op->flags)) {
+ /* Recompile the delay slot */
+ if (!op_flag_no_ds(op->flags) && next->opcode)
+ lightrec_rec_opcode(state, block, offset + 1);
if (link) {
/* Update the $ra register */
lightrec_free_reg(reg_cache, link_reg);
}
- /* Store back remaining registers */
- lightrec_storeback_regs(reg_cache, _jit);
+ /* Clean remaining registers */
+ lightrec_clean_regs(reg_cache, _jit);
target_offset = offset + 1 + (s16)op->i.imm
- - !!(OPT_SWITCH_DELAY_SLOTS && (op->flags & LIGHTREC_NO_DS));
+ - !!op_flag_no_ds(op->flags);
pr_debug("Adding local branch to offset 0x%x\n",
target_offset << 2);
branch = &state->local_branches[
state->nb_local_branches++];
branch->target = target_offset;
- if (is_forward)
- branch->branch = jit_jmpi();
+
+ if (no_indirection)
+ branch->branch = jit_new_node_pww(code2, NULL, rs, rt);
+ else if (is_forward)
+ branch->branch = jit_b();
else
branch->branch = jit_bgti(LIGHTREC_REG_CYCLE, 0);
}
- if (!(op->flags & LIGHTREC_LOCAL_BRANCH) || !is_forward) {
+ if (!op_flag_local_branch(op->flags) || !is_forward) {
next_pc = get_branch_pc(block, offset, 1 + (s16)op->i.imm);
lightrec_emit_end_of_block(state, block, offset, -1, next_pc,
31, link, false);
}
if (!unconditional) {
- jit_patch(addr);
+ if (!no_indirection)
+ jit_patch(addr);
+
lightrec_regcache_leave_branch(reg_cache, regs_backup);
if (bz && link) {
lightrec_free_reg(reg_cache, link_reg);
}
- if (!(op->flags & LIGHTREC_NO_DS) && next->opcode)
+ if (!op_flag_no_ds(op->flags) && next->opcode)
lightrec_rec_opcode(state, block, offset + 1);
}
}
_jit_name(block->_jit, __func__);
if (c.i.rt == 0)
- rec_b(state, block, offset, jit_code_beqi, 0, false, true);
+ rec_b(state, block, offset, jit_code_beqi, jit_code_bnei, 0, false, true);
else
- rec_b(state, block, offset, jit_code_beqr, 0, false, false);
+ rec_b(state, block, offset, jit_code_beqr, jit_code_bner, 0, false, false);
}
static void rec_BEQ(struct lightrec_cstate *state,
_jit_name(block->_jit, __func__);
if (c.i.rt == 0)
- rec_b(state, block, offset, jit_code_bnei, 0, c.i.rs == 0, true);
+ rec_b(state, block, offset, jit_code_bnei, jit_code_beqi, 0, c.i.rs == 0, true);
else
- rec_b(state, block, offset, jit_code_bner, 0, c.i.rs == c.i.rt, false);
+ rec_b(state, block, offset, jit_code_bner, jit_code_beqr, 0, c.i.rs == c.i.rt, false);
}
static void rec_BLEZ(struct lightrec_cstate *state,
union code c = block->opcode_list[offset].c;
_jit_name(block->_jit, __func__);
- rec_b(state, block, offset, jit_code_bgti, 0, c.i.rs == 0, true);
+ rec_b(state, block, offset, jit_code_bgti, jit_code_blei, 0, c.i.rs == 0, true);
}
static void rec_BGTZ(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_b(state, block, offset, jit_code_blei, 0, false, true);
+ rec_b(state, block, offset, jit_code_blei, jit_code_bgti, 0, false, true);
}
static void rec_regimm_BLTZ(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_b(state, block, offset, jit_code_bgei, 0, false, true);
+ rec_b(state, block, offset, jit_code_bgei, jit_code_blti, 0, false, true);
}
static void rec_regimm_BLTZAL(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_b(state, block, offset, jit_code_bgei,
+ rec_b(state, block, offset, jit_code_bgei, jit_code_blti,
get_branch_pc(block, offset, 2), false, true);
}
union code c = block->opcode_list[offset].c;
_jit_name(block->_jit, __func__);
- rec_b(state, block, offset, jit_code_blti, 0, !c.i.rs, true);
+ rec_b(state, block, offset, jit_code_blti, jit_code_bgei, 0, !c.i.rs, true);
}
static void rec_regimm_BGEZAL(struct lightrec_cstate *state,
{
const struct opcode *op = &block->opcode_list[offset];
_jit_name(block->_jit, __func__);
- rec_b(state, block, offset, jit_code_blti,
+ rec_b(state, block, offset, jit_code_blti, jit_code_bgei,
get_branch_pc(block, offset, 2),
!op->i.rs, true);
}
lightrec_free_reg(reg_cache, rd);
}
+static void rec_movi(struct lightrec_cstate *state,
+ const struct block *block, u16 offset)
+{
+ struct regcache *reg_cache = state->reg_cache;
+ union code c = block->opcode_list[offset].c;
+ jit_state_t *_jit = block->_jit;
+ u16 flags = REG_EXT;
+ u8 rt;
+
+ if (!(c.i.imm & 0x8000))
+ flags |= REG_ZEXT;
+
+ rt = lightrec_alloc_reg_out(reg_cache, _jit, c.i.rt, flags);
+
+ jit_movi(rt, (s32)(s16) c.i.imm);
+
+ lightrec_free_reg(reg_cache, rt);
+}
+
static void rec_ADDIU(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_alu_imm(state, block, offset, jit_code_addi, false);
+
+ if (block->opcode_list[offset].c.i.rs)
+ rec_alu_imm(state, block, offset, jit_code_addi, false);
+ else
+ rec_movi(state, block, offset);
}
static void rec_ADDI(struct lightrec_cstate *state,
{
/* TODO: Handle the exception? */
_jit_name(block->_jit, __func__);
- rec_alu_imm(state, block, offset, jit_code_addi, false);
+ rec_ADDIU(state, block, offset);
}
static void rec_SLTIU(struct lightrec_cstate *state,
if (!nor)
flags_rd = REG_ZEXT & flags_rs & flags_rt;
- /* E(rd) = (E(rs) & E(rt)) | (E(rt) & !Z(rt)) | (E(rs) & !Z(rs)) */
- if ((REG_EXT & flags_rs & flags_rt) ||
- (flags_rt & (REG_EXT | REG_ZEXT) == REG_EXT) ||
- (flags_rs & (REG_EXT | REG_ZEXT) == REG_EXT))
+ /* E(rd) = E(rs) & E(rt) */
+ if (REG_EXT & flags_rs & flags_rt)
flags_rd |= REG_EXT;
lightrec_set_reg_out_flags(reg_cache, rd, flags_rd);
{
struct regcache *reg_cache = state->reg_cache;
union code c = block->opcode_list[offset].c;
- u16 flags = block->opcode_list[offset].flags;
+ u32 flags = block->opcode_list[offset].flags;
u8 reg_lo = get_mult_div_lo(c);
u8 reg_hi = get_mult_div_hi(c);
jit_state_t *_jit = block->_jit;
rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, rflags);
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, rflags);
- if (!(flags & LIGHTREC_NO_LO))
+ if (!op_flag_no_lo(flags))
lo = lightrec_alloc_reg_out(reg_cache, _jit, reg_lo, 0);
else if (__WORDSIZE == 32)
lo = lightrec_alloc_reg_temp(reg_cache, _jit);
- if (!(flags & LIGHTREC_NO_HI))
+ if (!op_flag_no_hi(flags))
hi = lightrec_alloc_reg_out(reg_cache, _jit, reg_hi, REG_EXT);
if (__WORDSIZE == 32) {
/* On 32-bit systems, do a 32*32->64 bit operation, or a 32*32->32 bit
* operation if the MULT was detected a 32-bit only. */
- if (!(flags & LIGHTREC_NO_HI)) {
+ if (!op_flag_no_hi(flags)) {
if (is_signed)
jit_qmulr(lo, hi, rs, rt);
else
}
} else {
/* On 64-bit systems, do a 64*64->64 bit operation. */
- if (flags & LIGHTREC_NO_LO) {
+ if (op_flag_no_lo(flags)) {
jit_mulr(hi, rs, rt);
jit_rshi(hi, hi, 32);
} else {
jit_mulr(lo, rs, rt);
/* The 64-bit output value is in $lo, store the upper 32 bits in $hi */
- if (!(flags & LIGHTREC_NO_HI))
+ if (!op_flag_no_hi(flags))
jit_rshi(hi, lo, 32);
}
}
lightrec_free_reg(reg_cache, rs);
lightrec_free_reg(reg_cache, rt);
- if (!(flags & LIGHTREC_NO_LO) || __WORDSIZE == 32)
+ if (!op_flag_no_lo(flags) || __WORDSIZE == 32)
lightrec_free_reg(reg_cache, lo);
- if (!(flags & LIGHTREC_NO_HI))
+ if (!op_flag_no_hi(flags))
lightrec_free_reg(reg_cache, hi);
}
{
struct regcache *reg_cache = state->reg_cache;
union code c = block->opcode_list[offset].c;
- u16 flags = block->opcode_list[offset].flags;
- bool no_check = flags & LIGHTREC_NO_DIV_CHECK;
+ u32 flags = block->opcode_list[offset].flags;
+ bool no_check = op_flag_no_div_check(flags);
u8 reg_lo = get_mult_div_lo(c);
u8 reg_hi = get_mult_div_hi(c);
jit_state_t *_jit = block->_jit;
rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, rflags);
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, rflags);
- if (!(flags & LIGHTREC_NO_LO))
+ if (!op_flag_no_lo(flags))
lo = lightrec_alloc_reg_out(reg_cache, _jit, reg_lo, 0);
- if (!(flags & LIGHTREC_NO_HI))
+ if (!op_flag_no_hi(flags))
hi = lightrec_alloc_reg_out(reg_cache, _jit, reg_hi, 0);
/* Jump to special handler if dividing by zero */
if (!no_check)
branch = jit_beqi(rt, 0);
- if (flags & LIGHTREC_NO_LO) {
+ if (op_flag_no_lo(flags)) {
if (is_signed)
jit_remr(hi, rs, rt);
else
jit_remr_u(hi, rs, rt);
- } else if (flags & LIGHTREC_NO_HI) {
+ } else if (op_flag_no_hi(flags)) {
if (is_signed)
jit_divr(lo, rs, rt);
else
}
if (!no_check) {
- lightrec_regcache_mark_live(reg_cache, _jit);
-
/* Jump above the div-by-zero handler */
- to_end = jit_jmpi();
+ to_end = jit_b();
jit_patch(branch);
- if (!(flags & LIGHTREC_NO_LO)) {
+ if (!op_flag_no_lo(flags)) {
if (is_signed) {
jit_lti(lo, rs, 0);
jit_lshi(lo, lo, 1);
}
}
- if (!(flags & LIGHTREC_NO_HI))
+ if (!op_flag_no_hi(flags))
jit_movr(hi, rs);
jit_patch(to_end);
lightrec_free_reg(reg_cache, rs);
lightrec_free_reg(reg_cache, rt);
- if (!(flags & LIGHTREC_NO_LO))
+ if (!op_flag_no_lo(flags))
lightrec_free_reg(reg_cache, lo);
- if (!(flags & LIGHTREC_NO_HI))
+ if (!op_flag_no_hi(flags))
lightrec_free_reg(reg_cache, hi);
}
rec_alu_mv_lo_hi(state, block, REG_LO, c.r.rs);
}
-static void call_to_c_wrapper(struct lightrec_cstate *state, const struct block *block,
- u32 arg, bool with_arg, enum c_wrappers wrapper)
+static void call_to_c_wrapper(struct lightrec_cstate *state,
+ const struct block *block, u32 arg,
+ enum c_wrappers wrapper)
{
struct regcache *reg_cache = state->reg_cache;
jit_state_t *_jit = block->_jit;
- u8 tmp, tmp3;
+ s8 tmp, tmp2;
- if (with_arg)
- tmp3 = lightrec_alloc_reg(reg_cache, _jit, JIT_R1);
- tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
+ /* Make sure JIT_R1 is not mapped; it will be used in the C wrapper. */
+ tmp2 = lightrec_alloc_reg(reg_cache, _jit, JIT_R1);
+
+ tmp = lightrec_get_reg_with_value(reg_cache,
+ (intptr_t) state->state->wrappers_eps[wrapper]);
+ if (tmp < 0) {
+ tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
+ jit_ldxi(tmp, LIGHTREC_REG_STATE,
+ offsetof(struct lightrec_state, wrappers_eps[wrapper]));
- jit_ldxi(tmp, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, wrappers_eps[wrapper]));
- if (with_arg)
- jit_movi(tmp3, arg);
+ lightrec_temp_set_value(reg_cache, tmp,
+ (intptr_t) state->state->wrappers_eps[wrapper]);
+ }
+
+ lightrec_free_reg(reg_cache, tmp2);
+#ifdef __mips__
+ /* On MIPS, register t9 is always used as the target register for JALR.
+ * Therefore if it does not contain the target address we must
+ * invalidate it. */
+ if (tmp != _T9)
+ lightrec_unload_reg(reg_cache, _jit, _T9);
+#endif
+
+ jit_prepare();
+ jit_pushargi(arg);
+
+ lightrec_regcache_mark_live(reg_cache, _jit);
jit_callr(tmp);
lightrec_free_reg(reg_cache, tmp);
- if (with_arg)
- lightrec_free_reg(reg_cache, tmp3);
lightrec_regcache_mark_live(reg_cache, _jit);
}
struct regcache *reg_cache = state->reg_cache;
jit_state_t *_jit = block->_jit;
union code c = block->opcode_list[offset].c;
- u16 flags = block->opcode_list[offset].flags;
- bool is_tagged = flags & (LIGHTREC_HW_IO | LIGHTREC_DIRECT_IO);
+ u32 flags = block->opcode_list[offset].flags;
+ bool is_tagged = LIGHTREC_FLAGS_GET_IO_MODE(flags);
u32 lut_entry;
jit_note(__FILE__, __LINE__);
lightrec_clean_reg_if_loaded(reg_cache, _jit, c.i.rt, false);
if (is_tagged) {
- call_to_c_wrapper(state, block, c.opcode, true, C_WRAPPER_RW);
+ call_to_c_wrapper(state, block, c.opcode, C_WRAPPER_RW);
} else {
lut_entry = lightrec_get_lut_entry(block);
call_to_c_wrapper(state, block, (lut_entry << 16) | offset,
- true, C_WRAPPER_RW_GENERIC);
+ C_WRAPPER_RW_GENERIC);
}
}
+static u32 rec_ram_mask(struct lightrec_state *state)
+{
+ return (RAM_SIZE << (state->mirrors_mapped * 2)) - 1;
+}
+
+static u32 rec_io_mask(const struct lightrec_state *state)
+{
+ u32 length = state->maps[PSX_MAP_HW_REGISTERS].length;
+
+ return GENMASK(31 - clz32(length - 1), 0);
+}
+
+static void rec_store_memory(struct lightrec_cstate *cstate,
+ const struct block *block,
+ u16 offset, jit_code_t code,
+ jit_code_t swap_code,
+ uintptr_t addr_offset, u32 addr_mask,
+ bool invalidate)
+{
+ const struct lightrec_state *state = cstate->state;
+ struct regcache *reg_cache = cstate->reg_cache;
+ struct opcode *op = &block->opcode_list[offset];
+ jit_state_t *_jit = block->_jit;
+ union code c = op->c;
+ u8 rs, rt, tmp, tmp2, tmp3, addr_reg, addr_reg2;
+ s16 imm = (s16)c.i.imm;
+ s32 simm = (s32)imm << (1 - lut_is_32bit(state));
+ s32 lut_offt = offsetof(struct lightrec_state, code_lut);
+ bool no_mask = op_flag_no_mask(op->flags);
+ bool add_imm = c.i.imm &&
+ ((!state->mirrors_mapped && !no_mask) || (invalidate &&
+ ((imm & 0x3) || simm + lut_offt != (s16)(simm + lut_offt))));
+ bool need_tmp = !no_mask || addr_offset || add_imm || invalidate;
+
+ rt = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rt, 0);
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, 0);
+ if (need_tmp)
+ tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
+
+ addr_reg = rs;
+
+ if (add_imm) {
+ jit_addi(tmp, addr_reg, (s16)c.i.imm);
+ addr_reg = tmp;
+ imm = 0;
+ } else if (simm) {
+ lut_offt += simm;
+ }
+
+ if (!no_mask) {
+ jit_andi(tmp, addr_reg, addr_mask);
+ addr_reg = tmp;
+ }
+
+ if (addr_offset) {
+ tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
+ jit_addi(tmp2, addr_reg, addr_offset);
+ addr_reg2 = tmp2;
+ } else {
+ addr_reg2 = addr_reg;
+ }
+
+ if (is_big_endian() && swap_code && c.i.rt) {
+ tmp3 = lightrec_alloc_reg_temp(reg_cache, _jit);
+
+ jit_new_node_ww(swap_code, tmp3, rt);
+ jit_new_node_www(code, imm, addr_reg2, tmp3);
+
+ lightrec_free_reg(reg_cache, tmp3);
+ } else {
+ jit_new_node_www(code, imm, addr_reg2, rt);
+ }
+
+ lightrec_free_reg(reg_cache, rt);
+
+ if (invalidate) {
+ tmp3 = lightrec_alloc_reg_in(reg_cache, _jit, 0, 0);
+
+ if (c.i.op != OP_SW) {
+ jit_andi(tmp, addr_reg, ~3);
+ addr_reg = tmp;
+ }
+
+ if (!lut_is_32bit(state)) {
+ jit_lshi(tmp, addr_reg, 1);
+ addr_reg = tmp;
+ }
+
+ if (addr_reg == rs && c.i.rs == 0) {
+ addr_reg = LIGHTREC_REG_STATE;
+ } else {
+ jit_addr(tmp, addr_reg, LIGHTREC_REG_STATE);
+ addr_reg = tmp;
+ }
+
+ if (lut_is_32bit(state))
+ jit_stxi_i(lut_offt, addr_reg, tmp3);
+ else
+ jit_stxi(lut_offt, addr_reg, tmp3);
+
+ lightrec_free_reg(reg_cache, tmp3);
+ }
+
+ if (addr_offset)
+ lightrec_free_reg(reg_cache, tmp2);
+ if (need_tmp)
+ lightrec_free_reg(reg_cache, tmp);
+ lightrec_free_reg(reg_cache, rs);
+}
+
+static void rec_store_ram(struct lightrec_cstate *cstate,
+ const struct block *block,
+ u16 offset, jit_code_t code,
+ jit_code_t swap_code, bool invalidate)
+{
+ struct lightrec_state *state = cstate->state;
+
+ _jit_note(block->_jit, __FILE__, __LINE__);
+
+ return rec_store_memory(cstate, block, offset, code, swap_code,
+ state->offset_ram, rec_ram_mask(state),
+ invalidate);
+}
+
+static void rec_store_scratch(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code)
+{
+ _jit_note(block->_jit, __FILE__, __LINE__);
+
+ return rec_store_memory(cstate, block, offset, code, swap_code,
+ cstate->state->offset_scratch,
+ 0x1fffffff, false);
+}
+
+static void rec_store_io(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code)
+{
+ _jit_note(block->_jit, __FILE__, __LINE__);
+
+ return rec_store_memory(cstate, block, offset, code, swap_code,
+ cstate->state->offset_io,
+ rec_io_mask(cstate->state), false);
+}
+
static void rec_store_direct_no_invalidate(struct lightrec_cstate *cstate,
const struct block *block,
- u16 offset, jit_code_t code)
+ u16 offset, jit_code_t code,
+ jit_code_t swap_code)
{
struct lightrec_state *state = cstate->state;
struct regcache *reg_cache = cstate->reg_cache;
if (state->offset_ram != state->offset_scratch) {
to_not_ram = jit_bmsi(tmp, BIT(28));
- lightrec_regcache_mark_live(reg_cache, _jit);
-
jit_movi(tmp2, state->offset_ram);
- to_end = jit_jmpi();
+ to_end = jit_b();
jit_patch(to_not_ram);
jit_movi(tmp2, state->offset_scratch);
}
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rt, 0);
- jit_new_node_www(code, imm, tmp, rt);
+
+ if (is_big_endian() && swap_code && c.i.rt) {
+ tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
+
+ jit_new_node_ww(swap_code, tmp2, rt);
+ jit_new_node_www(code, imm, tmp, tmp2);
+
+ lightrec_free_reg(reg_cache, tmp2);
+ } else {
+ jit_new_node_www(code, imm, tmp, rt);
+ }
lightrec_free_reg(reg_cache, rt);
lightrec_free_reg(reg_cache, tmp);
}
static void rec_store_direct(struct lightrec_cstate *cstate, const struct block *block,
- u16 offset, jit_code_t code)
+ u16 offset, jit_code_t code, jit_code_t swap_code)
{
struct lightrec_state *state = cstate->state;
u32 ram_size = state->mirrors_mapped ? RAM_SIZE * 4 : RAM_SIZE;
to_not_ram = jit_bgti(tmp2, ram_size);
- lightrec_regcache_mark_live(reg_cache, _jit);
-
/* Compute the offset to the code LUT */
jit_andi(tmp, tmp2, (RAM_SIZE - 1) & ~3);
- if (__WORDSIZE == 64)
+ if (!lut_is_32bit(state))
jit_lshi(tmp, tmp, 1);
jit_addr(tmp, LIGHTREC_REG_STATE, tmp);
/* Write NULL to the code LUT to invalidate any block that's there */
- jit_stxi(offsetof(struct lightrec_state, code_lut), tmp, tmp3);
+ if (lut_is_32bit(state))
+ jit_stxi_i(offsetof(struct lightrec_state, code_lut), tmp, tmp3);
+ else
+ jit_stxi(offsetof(struct lightrec_state, code_lut), tmp, tmp3);
if (state->offset_ram != state->offset_scratch) {
jit_movi(tmp, state->offset_ram);
- to_end = jit_jmpi();
+ to_end = jit_b();
}
jit_patch(to_not_ram);
lightrec_free_reg(reg_cache, tmp3);
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rt, 0);
- jit_new_node_www(code, 0, tmp2, rt);
+
+ if (is_big_endian() && swap_code && c.i.rt) {
+ tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
+
+ jit_new_node_ww(swap_code, tmp, rt);
+ jit_new_node_www(code, 0, tmp2, tmp);
+
+ lightrec_free_reg(reg_cache, tmp);
+ } else {
+ jit_new_node_www(code, 0, tmp2, rt);
+ }
lightrec_free_reg(reg_cache, rt);
lightrec_free_reg(reg_cache, tmp2);
}
static void rec_store(struct lightrec_cstate *state,
- const struct block *block, u16 offset, jit_code_t code)
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code)
{
- u16 flags = block->opcode_list[offset].flags;
+ u32 flags = block->opcode_list[offset].flags;
+ bool no_invalidate = op_flag_no_invalidate(flags) ||
+ state->state->invalidate_from_dma_only;
- if (flags & LIGHTREC_NO_INVALIDATE) {
- rec_store_direct_no_invalidate(state, block, offset, code);
- } else if (flags & LIGHTREC_DIRECT_IO) {
- if (state->state->invalidate_from_dma_only)
- rec_store_direct_no_invalidate(state, block, offset, code);
- else
- rec_store_direct(state, block, offset, code);
- } else {
+ switch (LIGHTREC_FLAGS_GET_IO_MODE(flags)) {
+ case LIGHTREC_IO_RAM:
+ rec_store_ram(state, block, offset, code,
+ swap_code, !no_invalidate);
+ break;
+ case LIGHTREC_IO_SCRATCH:
+ rec_store_scratch(state, block, offset, code, swap_code);
+ break;
+ case LIGHTREC_IO_DIRECT:
+ if (no_invalidate) {
+ rec_store_direct_no_invalidate(state, block, offset,
+ code, swap_code);
+ } else {
+ rec_store_direct(state, block, offset, code, swap_code);
+ }
+ break;
+ case LIGHTREC_IO_DIRECT_HW:
+ rec_store_io(state, block, offset, code, swap_code);
+ break;
+ default:
rec_io(state, block, offset, true, false);
+ break;
}
}
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_store(state, block, offset, jit_code_stxi_c);
+ rec_store(state, block, offset, jit_code_stxi_c, 0);
}
static void rec_SH(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_store(state, block, offset, jit_code_stxi_s);
+ rec_store(state, block, offset,
+ jit_code_stxi_s, jit_code_bswapr_us);
}
static void rec_SW(struct lightrec_cstate *state,
{
_jit_name(block->_jit, __func__);
- rec_store(state, block, offset, jit_code_stxi_i);
+ rec_store(state, block, offset,
+ jit_code_stxi_i, jit_code_bswapr_ui);
}
static void rec_SWL(struct lightrec_cstate *state,
rec_io(state, block, offset, false, false);
}
-static void rec_load_direct(struct lightrec_cstate *cstate, const struct block *block,
- u16 offset, jit_code_t code, bool is_unsigned)
+static void rec_load_memory(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code, bool is_unsigned,
+ uintptr_t addr_offset, u32 addr_mask)
+{
+ struct regcache *reg_cache = cstate->reg_cache;
+ struct opcode *op = &block->opcode_list[offset];
+ jit_state_t *_jit = block->_jit;
+ u8 rs, rt, addr_reg, flags = REG_EXT;
+ bool no_mask = op_flag_no_mask(op->flags);
+ union code c = op->c;
+ s16 imm;
+
+ if (!c.i.rt)
+ return;
+
+ if (is_unsigned)
+ flags |= REG_ZEXT;
+
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, 0);
+ rt = lightrec_alloc_reg_out(reg_cache, _jit, c.i.rt, flags);
+
+ if (!cstate->state->mirrors_mapped && c.i.imm && !no_mask) {
+ jit_addi(rt, rs, (s16)c.i.imm);
+ addr_reg = rt;
+ imm = 0;
+ } else {
+ addr_reg = rs;
+ imm = (s16)c.i.imm;
+ }
+
+ if (!no_mask) {
+ jit_andi(rt, addr_reg, addr_mask);
+ addr_reg = rt;
+ }
+
+ if (addr_offset) {
+ jit_addi(rt, addr_reg, addr_offset);
+ addr_reg = rt;
+ }
+
+ jit_new_node_www(code, rt, addr_reg, imm);
+
+ if (is_big_endian() && swap_code) {
+ jit_new_node_ww(swap_code, rt, rt);
+
+ if (c.i.op == OP_LH)
+ jit_extr_s(rt, rt);
+ else if (c.i.op == OP_LW && __WORDSIZE == 64)
+ jit_extr_i(rt, rt);
+ }
+
+ lightrec_free_reg(reg_cache, rs);
+ lightrec_free_reg(reg_cache, rt);
+}
+
+static void rec_load_ram(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code, bool is_unsigned)
+{
+ _jit_note(block->_jit, __FILE__, __LINE__);
+
+ rec_load_memory(cstate, block, offset, code, swap_code, is_unsigned,
+ cstate->state->offset_ram, rec_ram_mask(cstate->state));
+}
+
+static void rec_load_bios(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code, bool is_unsigned)
+{
+ _jit_note(block->_jit, __FILE__, __LINE__);
+
+ rec_load_memory(cstate, block, offset, code, swap_code, is_unsigned,
+ cstate->state->offset_bios, 0x1fffffff);
+}
+
+static void rec_load_scratch(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code, bool is_unsigned)
+{
+ _jit_note(block->_jit, __FILE__, __LINE__);
+
+ rec_load_memory(cstate, block, offset, code, swap_code, is_unsigned,
+ cstate->state->offset_scratch, 0x1fffffff);
+}
+
+static void rec_load_io(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code, bool is_unsigned)
+{
+ _jit_note(block->_jit, __FILE__, __LINE__);
+
+ rec_load_memory(cstate, block, offset, code, swap_code, is_unsigned,
+ cstate->state->offset_io, rec_io_mask(cstate->state));
+}
+
+static void rec_load_direct(struct lightrec_cstate *cstate,
+ const struct block *block, u16 offset,
+ jit_code_t code, jit_code_t swap_code,
+ bool is_unsigned)
{
struct lightrec_state *state = cstate->state;
struct regcache *reg_cache = cstate->reg_cache;
} else {
to_not_ram = jit_bmsi(addr_reg, BIT(28));
- lightrec_regcache_mark_live(reg_cache, _jit);
-
/* Convert to KUNSEG and avoid RAM mirrors */
jit_andi(rt, addr_reg, RAM_SIZE - 1);
if (state->offset_ram)
jit_movi(tmp, state->offset_ram);
- to_end = jit_jmpi();
+ to_end = jit_b();
jit_patch(to_not_ram);
jit_movi(tmp, state->offset_bios);
if (state->offset_bios != state->offset_scratch) {
- to_end2 = jit_jmpi();
+ to_end2 = jit_b();
jit_patch(to_not_bios);
jit_new_node_www(code, rt, rt, imm);
+ if (is_big_endian() && swap_code) {
+ jit_new_node_ww(swap_code, rt, rt);
+
+ if (c.i.op == OP_LH)
+ jit_extr_s(rt, rt);
+ else if (c.i.op == OP_LW && __WORDSIZE == 64)
+ jit_extr_i(rt, rt);
+ }
+
lightrec_free_reg(reg_cache, addr_reg);
lightrec_free_reg(reg_cache, rt);
lightrec_free_reg(reg_cache, tmp);
}
static void rec_load(struct lightrec_cstate *state, const struct block *block,
- u16 offset, jit_code_t code, bool is_unsigned)
+ u16 offset, jit_code_t code, jit_code_t swap_code,
+ bool is_unsigned)
{
- u16 flags = block->opcode_list[offset].flags;
+ u32 flags = block->opcode_list[offset].flags;
- if (flags & LIGHTREC_DIRECT_IO)
- rec_load_direct(state, block, offset, code, is_unsigned);
- else
+ switch (LIGHTREC_FLAGS_GET_IO_MODE(flags)) {
+ case LIGHTREC_IO_RAM:
+ rec_load_ram(state, block, offset, code, swap_code, is_unsigned);
+ break;
+ case LIGHTREC_IO_BIOS:
+ rec_load_bios(state, block, offset, code, swap_code, is_unsigned);
+ break;
+ case LIGHTREC_IO_SCRATCH:
+ rec_load_scratch(state, block, offset, code, swap_code, is_unsigned);
+ break;
+ case LIGHTREC_IO_DIRECT_HW:
+ rec_load_io(state, block, offset, code, swap_code, is_unsigned);
+ break;
+ case LIGHTREC_IO_DIRECT:
+ rec_load_direct(state, block, offset, code, swap_code, is_unsigned);
+ break;
+ default:
rec_io(state, block, offset, false, true);
+ break;
+ }
}
static void rec_LB(struct lightrec_cstate *state, const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_load(state, block, offset, jit_code_ldxi_c, false);
+ rec_load(state, block, offset, jit_code_ldxi_c, 0, false);
}
static void rec_LBU(struct lightrec_cstate *state, const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_load(state, block, offset, jit_code_ldxi_uc, true);
+ rec_load(state, block, offset, jit_code_ldxi_uc, 0, true);
}
static void rec_LH(struct lightrec_cstate *state, const struct block *block, u16 offset)
{
+ jit_code_t code = is_big_endian() ? jit_code_ldxi_us : jit_code_ldxi_s;
+
_jit_name(block->_jit, __func__);
- rec_load(state, block, offset, jit_code_ldxi_s, false);
+ rec_load(state, block, offset, code, jit_code_bswapr_us, false);
}
static void rec_LHU(struct lightrec_cstate *state, const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_load(state, block, offset, jit_code_ldxi_us, true);
+ rec_load(state, block, offset, jit_code_ldxi_us, jit_code_bswapr_us, true);
}
static void rec_LWL(struct lightrec_cstate *state, const struct block *block, u16 offset)
static void rec_LW(struct lightrec_cstate *state, const struct block *block, u16 offset)
{
+ jit_code_t code;
+
+ if (is_big_endian() && __WORDSIZE == 64)
+ code = jit_code_ldxi_ui;
+ else
+ code = jit_code_ldxi_i;
+
_jit_name(block->_jit, __func__);
- rec_load(state, block, offset, jit_code_ldxi_i, false);
+ rec_load(state, block, offset, code, jit_code_bswapr_ui, false);
}
static void rec_LWC2(struct lightrec_cstate *state, const struct block *block, u16 offset)
}
static void rec_break_syscall(struct lightrec_cstate *state,
- const struct block *block, u16 offset, bool is_break)
+ const struct block *block, u16 offset,
+ u32 exit_code)
{
+ struct regcache *reg_cache = state->reg_cache;
+ jit_state_t *_jit = block->_jit;
+ u8 tmp;
+
_jit_note(block->_jit, __FILE__, __LINE__);
- if (is_break)
- call_to_c_wrapper(state, block, 0, false, C_WRAPPER_BREAK);
- else
- call_to_c_wrapper(state, block, 0, false, C_WRAPPER_SYSCALL);
+ tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
+
+ jit_movi(tmp, exit_code);
+ jit_stxi_i(offsetof(struct lightrec_state, exit_flags),
+ LIGHTREC_REG_STATE, tmp);
+
+ lightrec_free_reg(reg_cache, tmp);
/* TODO: the return address should be "pc - 4" if we're a delay slot */
lightrec_emit_end_of_block(state, block, offset, -1,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_break_syscall(state, block, offset, false);
+ rec_break_syscall(state, block, offset, LIGHTREC_EXIT_SYSCALL);
}
static void rec_special_BREAK(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_break_syscall(state, block, offset, true);
+ rec_break_syscall(state, block, offset, LIGHTREC_EXIT_BREAK);
+}
+
+static void rec_mfc(struct lightrec_cstate *state, const struct block *block, u16 offset)
+{
+ struct regcache *reg_cache = state->reg_cache;
+ union code c = block->opcode_list[offset].c;
+ jit_state_t *_jit = block->_jit;
+
+ jit_note(__FILE__, __LINE__);
+ lightrec_clean_reg_if_loaded(reg_cache, _jit, c.i.rt, true);
+
+ call_to_c_wrapper(state, block, c.opcode, C_WRAPPER_MFC);
}
static void rec_mtc(struct lightrec_cstate *state, const struct block *block, u16 offset)
lightrec_clean_reg_if_loaded(reg_cache, _jit, c.i.rs, false);
lightrec_clean_reg_if_loaded(reg_cache, _jit, c.i.rt, false);
- call_to_c_wrapper(state, block, c.opcode, true, C_WRAPPER_MTC);
+ call_to_c_wrapper(state, block, c.opcode, C_WRAPPER_MTC);
if (c.i.op == OP_CP0 &&
- !(block->opcode_list[offset].flags & LIGHTREC_NO_DS) &&
+ !op_flag_no_ds(block->opcode_list[offset].flags) &&
(c.r.rd == 12 || c.r.rd == 13))
lightrec_emit_end_of_block(state, block, offset, -1,
get_ds_pc(block, offset, 1),
const union code c = block->opcode_list[offset].c;
jit_state_t *_jit = block->_jit;
u8 rt, tmp = 0, tmp2, status;
+ jit_node_t *to_end;
jit_note(__FILE__, __LINE__);
jit_orr(tmp, tmp, tmp2);
}
+ lightrec_free_reg(reg_cache, rt);
+
if (c.r.rd == 12 || c.r.rd == 13) {
- jit_stxi_i(offsetof(struct lightrec_state, exit_flags),
- LIGHTREC_REG_STATE, tmp);
+ to_end = jit_beqi(tmp, 0);
- lightrec_free_reg(reg_cache, tmp);
- lightrec_free_reg(reg_cache, tmp2);
- }
+ jit_ldxi_i(tmp2, LIGHTREC_REG_STATE,
+ offsetof(struct lightrec_state, target_cycle));
+ jit_subr(tmp2, tmp2, LIGHTREC_REG_CYCLE);
+ jit_movi(LIGHTREC_REG_CYCLE, 0);
+ jit_stxi_i(offsetof(struct lightrec_state, target_cycle),
+ LIGHTREC_REG_STATE, tmp2);
+ jit_stxi_i(offsetof(struct lightrec_state, current_cycle),
+ LIGHTREC_REG_STATE, tmp2);
- lightrec_free_reg(reg_cache, rt);
- if (!(block->opcode_list[offset].flags & LIGHTREC_NO_DS) &&
- (c.r.rd == 12 || c.r.rd == 13))
- lightrec_emit_eob(state, block, offset + 1, true);
+ jit_patch(to_end);
+ }
+
+ if (!op_flag_no_ds(block->opcode_list[offset].flags) &&
+ (c.r.rd == 12 || c.r.rd == 13)) {
+ state->cycles += lightrec_cycles_of_opcode(c);
+ lightrec_emit_eob(state, block, offset + 1);
+ }
}
static void rec_cp0_MFC0(struct lightrec_cstate *state,
rec_mtc0(state, block, offset);
}
+static unsigned int cp2d_i_offset(u8 reg)
+{
+ return offsetof(struct lightrec_state, regs.cp2d[reg]);
+}
+
+static unsigned int cp2d_s_offset(u8 reg)
+{
+ return cp2d_i_offset(reg) + is_big_endian() * 2;
+}
+
+static unsigned int cp2c_i_offset(u8 reg)
+{
+ return offsetof(struct lightrec_state, regs.cp2c[reg]);
+}
+
+static unsigned int cp2c_s_offset(u8 reg)
+{
+ return cp2c_i_offset(reg) + is_big_endian() * 2;
+}
+
static void rec_cp2_basic_MFC2(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
+ if (state->state->ops.cop2_notify) {
+ /* We must call cop2_notify, handle that in C. */
+ rec_mfc(state, block, offset);
+ return;
+ }
+
flags = (zext_regs & BIT(reg)) ? REG_ZEXT : REG_EXT;
rt = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rt, flags);
case 9:
case 10:
case 11:
- jit_ldxi_s(rt, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2d[reg]));
+ jit_ldxi_s(rt, LIGHTREC_REG_STATE, cp2d_s_offset(reg));
break;
case 7:
case 16:
case 17:
case 18:
case 19:
- jit_ldxi_us(rt, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2d[reg]));
+ jit_ldxi_us(rt, LIGHTREC_REG_STATE, cp2d_s_offset(reg));
break;
case 28:
case 29:
for (i = 0; i < 3; i++) {
out = i == 0 ? rt : tmp;
- jit_ldxi_s(tmp, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2d[9 + i]));
+ jit_ldxi_s(tmp, LIGHTREC_REG_STATE, cp2d_s_offset(9 + i));
jit_movi(tmp2, 0x1f);
jit_rshi(out, tmp, 7);
lightrec_free_reg(reg_cache, tmp3);
break;
default:
- jit_ldxi_i(rt, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2d[reg]));
+ jit_ldxi_i(rt, LIGHTREC_REG_STATE, cp2d_i_offset(reg));
break;
}
_jit_name(block->_jit, __func__);
+ if (state->state->ops.cop2_notify) {
+ /* We must call cop2_notify, handle that in C. */
+ rec_mfc(state, block, offset);
+ return;
+ }
+
switch (c.r.rd) {
case 4:
case 12:
case 29:
case 30:
rt = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rt, REG_EXT);
- jit_ldxi_s(rt, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2c[c.r.rd]));
+ jit_ldxi_s(rt, LIGHTREC_REG_STATE, cp2c_s_offset(c.r.rd));
break;
default:
rt = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rt, REG_ZEXT);
- jit_ldxi_i(rt, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2c[c.r.rd]));
+ jit_ldxi_ui(rt, LIGHTREC_REG_STATE, cp2c_i_offset(c.r.rd));
break;
}
_jit_name(block->_jit, __func__);
+ if (state->state->ops.cop2_notify) {
+ /* We must call cop2_notify, handle that in C. */
+ rec_mtc(state, block, offset);
+ return;
+ }
+
if (c.r.rd == 31)
return;
switch (c.r.rd) {
case 15:
tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
- jit_ldxi_i(tmp, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2d[13]));
+ jit_ldxi_i(tmp, LIGHTREC_REG_STATE, cp2d_i_offset(13));
tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
- jit_ldxi_i(tmp2, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, regs.cp2d[14]));
+ jit_ldxi_i(tmp2, LIGHTREC_REG_STATE, cp2d_i_offset(14));
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2d[12]),
- LIGHTREC_REG_STATE, tmp);
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2d[13]),
- LIGHTREC_REG_STATE, tmp2);
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2d[14]),
- LIGHTREC_REG_STATE, rt);
+ jit_stxi_i(cp2d_i_offset(12), LIGHTREC_REG_STATE, tmp);
+ jit_stxi_i(cp2d_i_offset(13), LIGHTREC_REG_STATE, tmp2);
+ jit_stxi_i(cp2d_i_offset(14), LIGHTREC_REG_STATE, rt);
lightrec_free_reg(reg_cache, tmp);
lightrec_free_reg(reg_cache, tmp2);
jit_lshi(tmp, rt, 7);
jit_andi(tmp, tmp, 0xf80);
- jit_stxi_s(offsetof(struct lightrec_state, regs.cp2d[9]),
- LIGHTREC_REG_STATE, tmp);
+ jit_stxi_s(cp2d_s_offset(9), LIGHTREC_REG_STATE, tmp);
jit_lshi(tmp, rt, 2);
jit_andi(tmp, tmp, 0xf80);
- jit_stxi_s(offsetof(struct lightrec_state, regs.cp2d[10]),
- LIGHTREC_REG_STATE, tmp);
+ jit_stxi_s(cp2d_s_offset(10), LIGHTREC_REG_STATE, tmp);
jit_rshi(tmp, rt, 3);
jit_andi(tmp, tmp, 0xf80);
- jit_stxi_s(offsetof(struct lightrec_state, regs.cp2d[11]),
- LIGHTREC_REG_STATE, tmp);
+ jit_stxi_s(cp2d_s_offset(11), LIGHTREC_REG_STATE, tmp);
lightrec_free_reg(reg_cache, tmp);
break;
jit_patch_at(to_loop, loop);
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2d[31]),
- LIGHTREC_REG_STATE, tmp2);
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2d[30]),
- LIGHTREC_REG_STATE, rt);
+ jit_stxi_i(cp2d_i_offset(31), LIGHTREC_REG_STATE, tmp2);
+ jit_stxi_i(cp2d_i_offset(30), LIGHTREC_REG_STATE, rt);
lightrec_free_reg(reg_cache, tmp);
lightrec_free_reg(reg_cache, tmp2);
break;
default:
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2d[c.r.rd]),
- LIGHTREC_REG_STATE, rt);
+ jit_stxi_i(cp2d_i_offset(c.r.rd), LIGHTREC_REG_STATE, rt);
break;
}
_jit_name(block->_jit, __func__);
+ if (state->state->ops.cop2_notify) {
+ /* We must call cop2_notify, handle that in C. */
+ rec_mtc(state, block, offset);
+ return;
+ }
+
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, 0);
switch (c.r.rd) {
case 27:
case 29:
case 30:
- jit_stxi_s(offsetof(struct lightrec_state, regs.cp2c[c.r.rd]),
- LIGHTREC_REG_STATE, rt);
+ jit_stxi_s(cp2c_s_offset(c.r.rd), LIGHTREC_REG_STATE, rt);
break;
case 31:
tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
jit_andi(tmp2, rt, 0x7ffff000);
jit_orr(tmp, tmp2, tmp);
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2c[31]),
- LIGHTREC_REG_STATE, tmp);
+ jit_stxi_i(cp2c_i_offset(31), LIGHTREC_REG_STATE, tmp);
lightrec_free_reg(reg_cache, tmp);
lightrec_free_reg(reg_cache, tmp2);
break;
default:
- jit_stxi_i(offsetof(struct lightrec_state, regs.cp2c[c.r.rd]),
- LIGHTREC_REG_STATE, rt);
+ jit_stxi_i(cp2c_i_offset(c.r.rd), LIGHTREC_REG_STATE, rt);
}
lightrec_free_reg(reg_cache, rt);
jit_name(__func__);
jit_note(__FILE__, __LINE__);
- call_to_c_wrapper(state, block, c.opcode, true, C_WRAPPER_CP);
+ call_to_c_wrapper(state, block, c.opcode, C_WRAPPER_CP);
}
static void rec_meta_MOV(struct lightrec_cstate *state,
lightrec_free_reg(reg_cache, rt);
}
+static void rec_meta_MULT2(struct lightrec_cstate *state,
+ const struct block *block,
+ u16 offset)
+{
+ struct regcache *reg_cache = state->reg_cache;
+ union code c = block->opcode_list[offset].c;
+ jit_state_t *_jit = block->_jit;
+ u8 reg_lo = get_mult_div_lo(c);
+ u8 reg_hi = get_mult_div_hi(c);
+ u32 flags = block->opcode_list[offset].flags;
+ bool is_signed = c.i.op == OP_META_MULT2;
+ u8 rs, lo, hi, rflags = 0, hiflags = 0;
+ unsigned int i;
+
+ if (!op_flag_no_hi(flags) && c.r.op < 32) {
+ rflags = is_signed ? REG_EXT : REG_ZEXT;
+ hiflags = is_signed ? REG_EXT : (REG_EXT | REG_ZEXT);
+ }
+
+ _jit_name(block->_jit, __func__);
+ jit_note(__FILE__, __LINE__);
+
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, rflags);
+
+ /*
+ * We must handle the case where one of the output registers is our rs
+ * input register. Thanksfully, computing LO/HI can be done in any
+ * order. Here, we make sure that the computation that overwrites the
+ * input register is always performed last.
+ */
+ for (i = 0; i < 2; i++) {
+ if ((!i ^ (reg_lo == c.i.rs)) && !op_flag_no_lo(flags)) {
+ lo = lightrec_alloc_reg_out(reg_cache, _jit, reg_lo, 0);
+
+ if (c.r.op < 32)
+ jit_lshi(lo, rs, c.r.op);
+ else
+ jit_movi(lo, 0);
+
+ lightrec_free_reg(reg_cache, lo);
+ continue;
+ }
+
+ if ((!!i ^ (reg_lo == c.i.rs)) && !op_flag_no_hi(flags)) {
+ hi = lightrec_alloc_reg_out(reg_cache, _jit,
+ reg_hi, hiflags);
+
+ if (c.r.op >= 32)
+ jit_lshi(hi, rs, c.r.op - 32);
+ else if (is_signed)
+ jit_rshi(hi, rs, 32 - c.r.op);
+ else
+ jit_rshi_u(hi, rs, 32 - c.r.op);
+
+ lightrec_free_reg(reg_cache, hi);
+ }
+ }
+
+ lightrec_free_reg(reg_cache, rs);
+
+ _jit_name(block->_jit, __func__);
+ jit_note(__FILE__, __LINE__);
+}
+
static const lightrec_rec_func_t rec_standard[64] = {
SET_DEFAULT_ELM(rec_standard, unknown_opcode),
[OP_SPECIAL] = rec_SPECIAL,
[OP_META_MOV] = rec_meta_MOV,
[OP_META_EXTC] = rec_meta_EXTC_EXTS,
[OP_META_EXTS] = rec_meta_EXTC_EXTS,
+ [OP_META_MULT2] = rec_meta_MULT2,
+ [OP_META_MULTU2] = rec_meta_MULT2,
};
static const lightrec_rec_func_t rec_special[64] = {
const struct opcode *op = &block->opcode_list[offset];
jit_state_t *_jit = block->_jit;
lightrec_rec_func_t f;
+ u16 unload_offset;
- if (op->flags & LIGHTREC_SYNC) {
- jit_subi(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, state->cycles);
+ if (op_flag_sync(op->flags)) {
+ if (state->cycles)
+ jit_subi(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, state->cycles);
state->cycles = 0;
lightrec_storeback_regs(reg_cache, _jit);
(*f)(state, block, offset);
}
- if (unlikely(op->flags & LIGHTREC_UNLOAD_RD)) {
- lightrec_clean_reg_if_loaded(reg_cache, _jit, op->r.rd, true);
- pr_debug("Cleaning RD reg %s\n", lightrec_reg_name(op->r.rd));
- }
- if (unlikely(op->flags & LIGHTREC_UNLOAD_RS)) {
- lightrec_clean_reg_if_loaded(reg_cache, _jit, op->i.rs, true);
- pr_debug("Cleaning RS reg %s\n", lightrec_reg_name(op->i.rt));
- }
- if (unlikely(op->flags & LIGHTREC_UNLOAD_RT)) {
- lightrec_clean_reg_if_loaded(reg_cache, _jit, op->i.rt, true);
- pr_debug("Cleaning RT reg %s\n", lightrec_reg_name(op->i.rt));
+ if (OPT_EARLY_UNLOAD) {
+ unload_offset = offset +
+ (has_delay_slot(op->c) && !op_flag_no_ds(op->flags));
+
+ lightrec_do_early_unload(state, block, unload_offset);
}
}