static u32 lightrec_mfc2(struct lightrec_state *state, u8 reg);
static void lightrec_default_sb(struct lightrec_state *state, u32 opcode,
- void *host, u32 addr, u8 data)
+ void *host, u32 addr, u32 data)
{
- *(u8 *)host = data;
+ *(u8 *)host = (u8)data;
- if (!state->invalidate_from_dma_only)
+ if (!(state->opt_flags & LIGHTREC_OPT_INV_DMA_ONLY))
lightrec_invalidate(state, addr, 1);
}
static void lightrec_default_sh(struct lightrec_state *state, u32 opcode,
- void *host, u32 addr, u16 data)
+ void *host, u32 addr, u32 data)
{
- *(u16 *)host = HTOLE16(data);
+ *(u16 *)host = HTOLE16((u16)data);
- if (!state->invalidate_from_dma_only)
+ if (!(state->opt_flags & LIGHTREC_OPT_INV_DMA_ONLY))
lightrec_invalidate(state, addr, 2);
}
{
*(u32 *)host = HTOLE32(data);
- if (!state->invalidate_from_dma_only)
+ if (!(state->opt_flags & LIGHTREC_OPT_INV_DMA_ONLY))
lightrec_invalidate(state, addr, 4);
}
return LE32TOH(*(u32 *)host);
}
+static u32 lightrec_default_lwu(struct lightrec_state *state,
+ u32 opcode, void *host, u32 addr)
+{
+ u32 val;
+
+ memcpy(&val, host, 4);
+
+ return LE32TOH(val);
+}
+
+static void lightrec_default_swu(struct lightrec_state *state, u32 opcode,
+ void *host, u32 addr, u32 data)
+{
+ data = HTOLE32(data);
+
+ memcpy(host, &data, 4);
+
+ if (!(state->opt_flags & LIGHTREC_OPT_INV_DMA_ONLY))
+ lightrec_invalidate(state, addr & ~0x3, 8);
+}
+
static const struct lightrec_mem_map_ops lightrec_default_ops = {
.sb = lightrec_default_sb,
.sh = lightrec_default_sh,
.lb = lightrec_default_lb,
.lh = lightrec_default_lh,
.lw = lightrec_default_lw,
+ .lwu = lightrec_default_lwu,
+ .swu = lightrec_default_swu,
};
static void __segfault_cb(struct lightrec_state *state, u32 addr,
{
lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT);
pr_err("Segmentation fault in recompiled code: invalid "
- "load/store at address 0x%08x\n", addr);
+ "load/store at address "PC_FMT"\n", addr);
if (block)
- pr_err("Was executing block PC 0x%08x\n", block->pc);
+ pr_err("Was executing block "PC_FMT"\n", block->pc);
}
static void lightrec_swl(struct lightrec_state *state,
}
}
-enum psx_map
+static enum psx_map
lightrec_get_map_idx(struct lightrec_state *state, u32 kaddr)
{
const struct lightrec_mem_map *map;
return map;
}
-u32 lightrec_rw(struct lightrec_state *state, union code op,
- u32 addr, u32 data, u32 *flags, struct block *block)
+u32 lightrec_rw(struct lightrec_state *state, union code op, u32 base,
+ u32 data, u32 *flags, struct block *block, u16 offset)
{
const struct lightrec_mem_map *map;
const struct lightrec_mem_map_ops *ops;
u32 opcode = op.opcode;
+ bool was_tagged = true;
+ u16 old_flags;
+ u32 addr;
void *host;
- addr += (s16) op.i.imm;
+ addr = kunseg(base + (s16) op.i.imm);
- map = lightrec_get_map(state, &host, kunseg(addr));
+ map = lightrec_get_map(state, &host, addr);
if (!map) {
__segfault_cb(state, addr, block);
return 0;
}
+ if (flags)
+ was_tagged = LIGHTREC_FLAGS_GET_IO_MODE(*flags);
if (likely(!map->ops)) {
- if (flags && !LIGHTREC_FLAGS_GET_IO_MODE(*flags))
- *flags |= LIGHTREC_IO_MODE(LIGHTREC_IO_DIRECT);
+ if (flags && !LIGHTREC_FLAGS_GET_IO_MODE(*flags)) {
+ /* Force parallel port accesses as HW accesses, because
+ * the direct-I/O emitters can't differenciate it. */
+ if (unlikely(map == &state->maps[PSX_MAP_PARALLEL_PORT]))
+ *flags |= LIGHTREC_IO_MODE(LIGHTREC_IO_HW);
+ /* If the base register is 0x0, be extra suspicious.
+ * Some games (e.g. Sled Storm) actually do segmentation
+ * faults by using uninitialized pointers, which are
+ * later initialized to point to hardware registers. */
+ else if (op.i.rs && base == 0x0)
+ *flags |= LIGHTREC_IO_MODE(LIGHTREC_IO_HW);
+ else
+ *flags |= LIGHTREC_IO_MODE(LIGHTREC_IO_DIRECT);
+ }
ops = &lightrec_default_ops;
} else if (flags &&
ops = map->ops;
}
+ if (!was_tagged) {
+ old_flags = block_set_flags(block, BLOCK_SHOULD_RECOMPILE);
+
+ if (!(old_flags & BLOCK_SHOULD_RECOMPILE)) {
+ pr_debug("Opcode of block at "PC_FMT" has been tagged"
+ " - flag for recompilation\n", block->pc);
+
+ lut_write(state, lut_offset(block->pc), NULL);
+ }
+ }
+
switch (op.i.op) {
case OP_SB:
- ops->sb(state, opcode, host, addr, (u8) data);
+ ops->sb(state, opcode, host, addr, data);
return 0;
case OP_SH:
- ops->sh(state, opcode, host, addr, (u16) data);
+ ops->sh(state, opcode, host, addr, data);
return 0;
case OP_SWL:
lightrec_swl(state, ops, opcode, host, addr, data);
return lightrec_lwl(state, ops, opcode, host, addr, data);
case OP_LWR:
return lightrec_lwr(state, ops, opcode, host, addr, data);
+ case OP_META_LWU:
+ return ops->lwu(state, opcode, host, addr);
+ case OP_META_SWU:
+ ops->swu(state, opcode, host, addr, data);
+ return 0;
case OP_LW:
default:
return ops->lw(state, opcode, host, addr);
static void lightrec_rw_helper(struct lightrec_state *state,
union code op, u32 *flags,
- struct block *block)
+ struct block *block, u16 offset)
{
u32 ret = lightrec_rw(state, op, state->regs.gpr[op.i.rs],
- state->regs.gpr[op.i.rt], flags, block);
+ state->regs.gpr[op.i.rt], flags, block, offset);
switch (op.i.op) {
case OP_LB:
case OP_LWL:
case OP_LWR:
case OP_LW:
- if (op.i.rt)
+ case OP_META_LWU:
+ if (OPT_HANDLE_LOAD_DELAYS && unlikely(!state->in_delay_slot_n)) {
+ state->temp_reg = ret;
+ state->in_delay_slot_n = 0xff;
+ } else if (op.i.rt) {
state->regs.gpr[op.i.rt] = ret;
+ }
fallthrough;
default:
break;
static void lightrec_rw_cb(struct lightrec_state *state, u32 arg)
{
- lightrec_rw_helper(state, (union code) arg, NULL, NULL);
+ lightrec_rw_helper(state, (union code) arg, NULL, NULL, 0);
}
static void lightrec_rw_generic_cb(struct lightrec_state *state, u32 arg)
{
struct block *block;
struct opcode *op;
- bool was_tagged;
u16 offset = (u16)arg;
- u16 old_flags;
block = lightrec_find_block_from_lut(state->block_cache,
- arg >> 16, state->next_pc);
+ arg >> 16, state->curr_pc);
if (unlikely(!block)) {
- pr_err("rw_generic: No block found in LUT for PC 0x%x offset 0x%x\n",
- state->next_pc, offset);
+ pr_err("rw_generic: No block found in LUT for "PC_FMT" offset 0x%"PRIx16"\n",
+ state->curr_pc, offset);
lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT);
return;
}
op = &block->opcode_list[offset];
- was_tagged = LIGHTREC_FLAGS_GET_IO_MODE(op->flags);
-
- lightrec_rw_helper(state, op->c, &op->flags, block);
-
- if (!was_tagged) {
- old_flags = block_set_flags(block, BLOCK_SHOULD_RECOMPILE);
-
- if (!(old_flags & BLOCK_SHOULD_RECOMPILE)) {
- pr_debug("Opcode of block at PC 0x%08x has been tagged"
- " - flag for recompilation\n", block->pc);
-
- lut_write(state, lut_offset(block->pc), NULL);
- }
- }
+ lightrec_rw_helper(state, op->c, &op->flags, block, offset);
}
static u32 clamp_s32(s32 val, s32 min, s32 max)
if (op.i.op == OP_CP0)
return state->regs.cp0[op.r.rd];
- else if (op.r.rs == OP_CP2_BASIC_MFC2)
+
+ if (op.i.op == OP_SWC2) {
+ val = lightrec_mfc2(state, op.i.rt);
+ } else if (op.r.rs == OP_CP2_BASIC_MFC2)
val = lightrec_mfc2(state, op.r.rd);
else {
val = state->regs.cp2c[op.r.rd];
{
u32 rt = lightrec_mfc(state, op);
- if (op.r.rt)
+ if (op.i.op == OP_SWC2)
+ state->temp_reg = rt;
+ else if (op.r.rt)
state->regs.gpr[op.r.rt] = rt;
}
status = state->regs.cp0[12];
/* Handle software interrupts */
- if (!!(status & cause & 0x300) & status)
+ if ((!!(status & cause & 0x300)) & status)
lightrec_set_exit_flags(state, LIGHTREC_EXIT_CHECK_INTERRUPT);
/* Handle hardware interrupts */
}
}
-void lightrec_mtc(struct lightrec_state *state, union code op, u32 data)
+void lightrec_mtc(struct lightrec_state *state, union code op, u8 reg, u32 data)
{
if (op.i.op == OP_CP0) {
- lightrec_mtc0(state, op.r.rd, data);
+ lightrec_mtc0(state, reg, data);
} else {
- if (op.r.rs == OP_CP2_BASIC_CTC2)
- lightrec_ctc2(state, op.r.rd, data);
+ if (op.i.op == OP_LWC2 || op.r.rs != OP_CP2_BASIC_CTC2)
+ lightrec_mtc2(state, reg, data);
else
- lightrec_mtc2(state, op.r.rd, data);
+ lightrec_ctc2(state, reg, data);
if (state->ops.cop2_notify)
(*state->ops.cop2_notify)(state, op.opcode, data);
static void lightrec_mtc_cb(struct lightrec_state *state, u32 arg)
{
union code op = (union code) arg;
+ u32 data;
+ u8 reg;
- lightrec_mtc(state, op, state->regs.gpr[op.r.rt]);
+ if (op.i.op == OP_LWC2) {
+ data = state->temp_reg;
+ reg = op.i.rt;
+ } else {
+ data = state->regs.gpr[op.r.rt];
+ reg = op.r.rd;
+ }
+
+ lightrec_mtc(state, op, reg, data);
}
void lightrec_rfe(struct lightrec_state *state)
u8 old_flags;
if (block && lightrec_block_is_outdated(state, block)) {
- pr_debug("Block at PC 0x%08x is outdated!\n", block->pc);
+ pr_debug("Block at "PC_FMT" is outdated!\n", block->pc);
old_flags = block_set_flags(block, BLOCK_IS_DEAD);
if (!(old_flags & BLOCK_IS_DEAD)) {
if (!block) {
block = lightrec_precompile_block(state, pc);
if (!block) {
- pr_err("Unable to recompile block at PC 0x%x\n", pc);
+ pr_err("Unable to recompile block at "PC_FMT"\n", pc);
lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT);
return NULL;
}
void *func;
int err;
- for (;;) {
+ do {
func = lut_read(state, lut_offset(pc));
if (func && func != state->get_next_block)
break;
}
should_recompile = block_has_flag(block, BLOCK_SHOULD_RECOMPILE) &&
+ !block_has_flag(block, BLOCK_NEVER_COMPILE) &&
!block_has_flag(block, BLOCK_IS_DEAD);
if (unlikely(should_recompile)) {
- pr_debug("Block at PC 0x%08x should recompile\n", pc);
+ pr_debug("Block at "PC_FMT" should recompile\n", pc);
if (ENABLE_THREADED_COMPILER) {
lightrec_recompiler_add(state->rec, block);
} else {
lightrec_recompiler_add(state->rec, block);
}
+ } while (state->exit_flags == LIGHTREC_EXIT_NORMAL
+ && state->current_cycle < state->target_cycle);
- if (state->exit_flags != LIGHTREC_EXIT_NORMAL ||
- state->current_cycle >= state->target_cycle)
- break;
- }
-
- state->next_pc = pc;
+ state->curr_pc = pc;
return func;
}
lightrec_code_alloc_unlock(state);
}
+static char lightning_code_data[0x80000];
+
static void * lightrec_emit_code(struct lightrec_state *state,
const struct block *block,
jit_state_t *_jit, unsigned int *size)
jit_realize();
- if (!ENABLE_DISASSEMBLER)
+ if (ENABLE_DISASSEMBLER)
+ jit_set_data(lightning_code_data, sizeof(lightning_code_data), 0);
+ else
jit_set_data(NULL, 0, JIT_DISABLE_DATA | JIT_DISABLE_NOTE);
if (has_code_buffer) {
*size = (unsigned int) new_code_size;
+ if (state->ops.code_inv)
+ state->ops.code_inv(code, new_code_size);
+
return code;
}
unsigned int i;
jit_node_t *addr[C_WRAPPERS_COUNT - 1];
jit_node_t *to_end[C_WRAPPERS_COUNT - 1];
+ u8 tmp = JIT_R1;
+
+#ifdef __sh__
+ /* On SH, GBR-relative loads target the r0 register.
+ * Use it as the temporary register to factorize the move to
+ * JIT_R1. */
+ if (LIGHTREC_REG_STATE == _GBR)
+ tmp = _R0;
+#endif
block = lightrec_malloc(state, MEM_FOR_IR, sizeof(*block));
if (!block)
/* Add entry points */
for (i = C_WRAPPERS_COUNT - 1; i > 0; i--) {
- jit_ldxi(JIT_R1, LIGHTREC_REG_STATE,
+ jit_ldxi(tmp, LIGHTREC_REG_STATE,
offsetof(struct lightrec_state, c_wrappers[i]));
to_end[i - 1] = jit_b();
addr[i - 1] = jit_indirect();
}
- jit_ldxi(JIT_R1, LIGHTREC_REG_STATE,
+ jit_ldxi(tmp, LIGHTREC_REG_STATE,
offsetof(struct lightrec_state, c_wrappers[0]));
for (i = 0; i < C_WRAPPERS_COUNT - 1; i++)
jit_patch(to_end[i]);
+ jit_movr(JIT_R1, tmp);
jit_epilog();
jit_prolog();
u32 length = state->regs.gpr[5] * 4;
if (!map) {
- pr_err("Unable to find memory map for memset target address "
- "0x%x\n", kunseg_pc);
+ pr_err("Unable to find memory map for memset target address "PC_FMT"\n",
+ kunseg_pc);
return 0;
}
- pr_debug("Calling host memset, PC 0x%x (host address 0x%" PRIxPTR ") for %u bytes\n",
+ pr_debug("Calling host memset, "PC_FMT" (host address 0x%"PRIxPTR") for %u bytes\n",
kunseg_pc, (uintptr_t)host, length);
memset(host, 0, length);
- if (!state->invalidate_from_dma_only)
+ if (!(state->opt_flags & LIGHTREC_OPT_INV_DMA_ONLY))
lightrec_invalidate_map(state, map, kunseg_pc, length);
/* Rough estimation of the number of cycles consumed */
return 8 + 5 * (length + 3 / 4);
}
+static u32 lightrec_check_load_delay(struct lightrec_state *state, u32 pc, u8 reg)
+{
+ struct block *block;
+ union code first_op;
+
+ first_op = lightrec_read_opcode(state, pc);
+
+ if (likely(!opcode_reads_register(first_op, reg))) {
+ state->regs.gpr[reg] = state->temp_reg;
+ } else {
+ block = lightrec_get_block(state, pc);
+ if (unlikely(!block)) {
+ pr_err("Unable to get block at "PC_FMT"\n", pc);
+ lightrec_set_exit_flags(state, LIGHTREC_EXIT_SEGFAULT);
+ pc = 0;
+ } else {
+ pc = lightrec_handle_load_delay(state, block, pc, reg);
+ }
+ }
+
+ return pc;
+}
+
+static void update_cycle_counter_before_c(jit_state_t *_jit)
+{
+ /* update state->current_cycle */
+ jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE,
+ offsetof(struct lightrec_state, target_cycle));
+ jit_subr(JIT_R1, JIT_R2, LIGHTREC_REG_CYCLE);
+ jit_stxi_i(offsetof(struct lightrec_state, current_cycle),
+ LIGHTREC_REG_STATE, JIT_R1);
+}
+
+static void update_cycle_counter_after_c(jit_state_t *_jit)
+{
+ /* Recalc the delta */
+ jit_ldxi_i(JIT_R1, LIGHTREC_REG_STATE,
+ offsetof(struct lightrec_state, current_cycle));
+ jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE,
+ offsetof(struct lightrec_state, target_cycle));
+ jit_subr(LIGHTREC_REG_CYCLE, JIT_R2, JIT_R1);
+}
+
+static void sync_next_pc(jit_state_t *_jit)
+{
+ if (lightrec_store_next_pc()) {
+ jit_ldxi_ui(JIT_V0, LIGHTREC_REG_STATE,
+ offsetof(struct lightrec_state, next_pc));
+ }
+}
+
static struct block * generate_dispatcher(struct lightrec_state *state)
{
struct block *block;
jit_state_t *_jit;
- jit_node_t *to_end, *loop, *addr, *addr2, *addr3;
+ jit_node_t *to_end, *loop, *addr, *addr2, *addr3, *addr4, *addr5, *jmp, *jmp2;
unsigned int i;
u32 offset;
jit_prolog();
jit_frame(256);
+ jit_getarg(LIGHTREC_REG_STATE, jit_arg());
+ jit_getarg(JIT_V0, jit_arg());
jit_getarg(JIT_V1, jit_arg());
jit_getarg_i(LIGHTREC_REG_CYCLE, jit_arg());
for (i = 0; i < NUM_REGS; i++)
jit_movr(JIT_V(i + FIRST_REG), JIT_V(i + FIRST_REG));
- /* Pass lightrec_state structure to blocks, using the last callee-saved
- * register that Lightning provides */
- jit_movi(LIGHTREC_REG_STATE, (intptr_t) state);
-
loop = jit_label();
/* Call the block's code */
jit_prepare();
jit_pushargr(LIGHTREC_REG_STATE);
+
jit_finishi(lightrec_memset);
+ jit_retval(LIGHTREC_REG_CYCLE);
jit_ldxi_ui(JIT_V0, LIGHTREC_REG_STATE,
offsetof(struct lightrec_state, regs.gpr[31]));
-
- jit_retval(LIGHTREC_REG_CYCLE);
jit_subr(LIGHTREC_REG_CYCLE, JIT_V1, LIGHTREC_REG_CYCLE);
+
+ if (OPT_DETECT_IMPOSSIBLE_BRANCHES || OPT_HANDLE_LOAD_DELAYS)
+ jmp = jit_b();
+ }
+
+ if (OPT_DETECT_IMPOSSIBLE_BRANCHES) {
+ /* Blocks will jump here when they reach a branch that should
+ * be executed with the interpreter, passing the branch's PC
+ * in JIT_V0 and the address of the block in JIT_V1. */
+ addr4 = jit_indirect();
+
+ sync_next_pc(_jit);
+ update_cycle_counter_before_c(_jit);
+
+ jit_prepare();
+ jit_pushargr(LIGHTREC_REG_STATE);
+ jit_pushargr(JIT_V1);
+ jit_pushargr(JIT_V0);
+ jit_finishi(lightrec_emulate_block);
+
+ jit_retval(JIT_V0);
+
+ update_cycle_counter_after_c(_jit);
+
+ if (OPT_HANDLE_LOAD_DELAYS)
+ jmp2 = jit_b();
+
+ }
+
+ if (OPT_HANDLE_LOAD_DELAYS) {
+ /* Blocks will jump here when they reach a branch with a load
+ * opcode in its delay slot. The delay slot has already been
+ * executed; the load value is in (state->temp_reg), and the
+ * register number is in JIT_V1.
+ * Jump to a C function which will evaluate the branch target's
+ * first opcode, to make sure that it does not read the register
+ * in question; and if it does, handle it accordingly. */
+ addr5 = jit_indirect();
+
+ sync_next_pc(_jit);
+ update_cycle_counter_before_c(_jit);
+
+ jit_prepare();
+ jit_pushargr(LIGHTREC_REG_STATE);
+ jit_pushargr(JIT_V0);
+ jit_pushargr(JIT_V1);
+ jit_finishi(lightrec_check_load_delay);
+
+ jit_retval(JIT_V0);
+
+ update_cycle_counter_after_c(_jit);
}
/* The block will jump here, with the number of cycles remaining in
* LIGHTREC_REG_CYCLE */
addr2 = jit_indirect();
- /* Store back the next_pc to the lightrec_state structure */
- offset = offsetof(struct lightrec_state, next_pc);
+ sync_next_pc(_jit);
+
+ if (OPT_HANDLE_LOAD_DELAYS && OPT_DETECT_IMPOSSIBLE_BRANCHES)
+ jit_patch(jmp2);
+
+ if (OPT_REPLACE_MEMSET
+ && (OPT_DETECT_IMPOSSIBLE_BRANCHES || OPT_HANDLE_LOAD_DELAYS)) {
+ jit_patch(jmp);
+ }
+
+ /* Store back the next PC to the lightrec_state structure */
+ offset = offsetof(struct lightrec_state, curr_pc);
jit_stxi_i(offset, LIGHTREC_REG_STATE, JIT_V0);
/* Jump to end if state->target_cycle < state->current_cycle */
/* If possible, use the code LUT */
if (!lut_is_32bit(state))
jit_lshi(JIT_V1, JIT_V1, 1);
- jit_addr(JIT_V1, JIT_V1, LIGHTREC_REG_STATE);
+ jit_add_state(JIT_V1, JIT_V1);
offset = offsetof(struct lightrec_state, code_lut);
if (lut_is_32bit(state))
if (ENABLE_FIRST_PASS || OPT_DETECT_IMPOSSIBLE_BRANCHES) {
/* We may call the interpreter - update state->current_cycle */
- jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, target_cycle));
- jit_subr(JIT_V1, JIT_R2, LIGHTREC_REG_CYCLE);
- jit_stxi_i(offsetof(struct lightrec_state, current_cycle),
- LIGHTREC_REG_STATE, JIT_V1);
+ update_cycle_counter_before_c(_jit);
}
jit_prepare();
if (ENABLE_FIRST_PASS || OPT_DETECT_IMPOSSIBLE_BRANCHES) {
/* The interpreter may have updated state->current_cycle and
* state->target_cycle - recalc the delta */
- jit_ldxi_i(JIT_R1, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, current_cycle));
- jit_ldxi_i(JIT_R2, LIGHTREC_REG_STATE,
- offsetof(struct lightrec_state, target_cycle));
- jit_subr(LIGHTREC_REG_CYCLE, JIT_R2, JIT_R1);
+ update_cycle_counter_after_c(_jit);
} else {
jit_movr(LIGHTREC_REG_CYCLE, JIT_V0);
}
+ /* Reset JIT_V0 to the next PC */
+ jit_ldxi_ui(JIT_V0, LIGHTREC_REG_STATE,
+ offsetof(struct lightrec_state, curr_pc));
+
/* If we get non-NULL, loop */
jit_patch_at(jit_bnei(JIT_V1, 0), loop);
goto err_free_block;
state->eob_wrapper_func = jit_address(addr2);
+ if (OPT_DETECT_IMPOSSIBLE_BRANCHES)
+ state->interpreter_func = jit_address(addr4);
+ if (OPT_HANDLE_LOAD_DELAYS)
+ state->ds_check_func = jit_address(addr5);
if (OPT_REPLACE_MEMSET)
state->memset_func = jit_address(addr3);
state->get_next_block = jit_address(addr);
return (union code) LE32TOH(*code);
}
-unsigned int lightrec_cycles_of_opcode(union code code)
+unsigned int lightrec_cycles_of_opcode(const struct lightrec_state *state,
+ union code code)
{
- return 2;
+ return state->cycles_per_op;
}
void lightrec_free_opcode_list(struct lightrec_state *state, struct opcode *ops)
pr_debug("Block size: %hu opcodes\n", block->nb_ops);
- /* If the first opcode is an 'impossible' branch, never compile the
- * block */
- if (should_emulate(block->opcode_list))
- block_flags |= BLOCK_NEVER_COMPILE;
-
fully_tagged = lightrec_block_is_fully_tagged(block);
if (fully_tagged)
block_flags |= BLOCK_FULLY_TAGGED;
addr = state->get_next_block;
lut_write(state, lut_offset(pc), addr);
- pr_debug("Recompile count: %u\n", state->nb_precompile++);
+ pr_debug("Blocks created: %u\n", ++state->nb_precompile);
return block;
}
for (i = 0; i < block->nb_ops; i++) {
op = &block->opcode_list[i];
- /* Verify that all load/stores of the opcode list
- * Check all loads/stores of the opcode list and mark the
+ /* If we have one branch that must be emulated, we cannot trash
+ * the opcode list. */
+ if (should_emulate(op))
+ return false;
+
+ /* Check all loads/stores of the opcode list and mark the
* block as fully compiled if they all have been tagged. */
switch (op->c.i.op) {
case OP_LB:
case OP_SWR:
case OP_LWC2:
case OP_SWC2:
+ case OP_META_LWU:
+ case OP_META_SWU:
if (!LIGHTREC_FLAGS_GET_IO_MODE(op->flags))
return false;
fallthrough;
{
struct block *block = data;
- pr_debug("Reap dead block at PC 0x%08x\n", block->pc);
+ pr_debug("Reap dead block at "PC_FMT"\n", block->pc);
lightrec_unregister_block(state->block_cache, block);
lightrec_free_block(state, block);
}
block->_jit = _jit;
lightrec_regcache_reset(cstate->reg_cache);
+
+ if (OPT_PRELOAD_PC && (block->flags & BLOCK_PRELOAD_PC))
+ lightrec_preload_pc(cstate->reg_cache, _jit);
+
cstate->cycles = 0;
cstate->nb_local_branches = 0;
cstate->nb_targets = 0;
+ cstate->no_load_delay = false;
jit_prolog();
jit_tramp(256);
pr_debug("Branch at offset 0x%x will be emulated\n",
i << 2);
- lightrec_emit_eob(cstate, block, i);
+ lightrec_emit_jump_to_interpreter(cstate, block, i);
skip_next = !op_flag_no_ds(elm->flags);
} else {
lightrec_rec_opcode(cstate, block, i);
#endif
}
- cstate->cycles += lightrec_cycles_of_opcode(elm->c);
+ cstate->cycles += lightrec_cycles_of_opcode(state, elm->c);
}
for (i = 0; i < cstate->nb_local_branches; i++) {
old_flags = block_set_flags(block, BLOCK_NO_OPCODE_LIST);
if (fully_tagged && !(old_flags & BLOCK_NO_OPCODE_LIST)) {
- pr_debug("Block PC 0x%08x is fully tagged"
+ pr_debug("Block "PC_FMT" is fully tagged"
" - free opcode list\n", block->pc);
if (ENABLE_THREADED_COMPILER) {
lightrec_unregister(MEM_FOR_CODE, old_code_size);
}
+ pr_debug("Blocks compiled: %u\n", ++state->nb_compile);
+
return 0;
}
u32 lightrec_execute(struct lightrec_state *state, u32 pc, u32 target_cycle)
{
- s32 (*func)(void *, s32) = (void *)state->dispatcher->function;
+ s32 (*func)(struct lightrec_state *, u32, void *, s32) = (void *)state->dispatcher->function;
void *block_trace;
s32 cycles_delta;
target_cycle = UINT_MAX;
state->target_cycle = target_cycle;
- state->next_pc = pc;
+ state->curr_pc = pc;
block_trace = get_next_block_func(state, pc);
if (block_trace) {
cycles_delta = state->target_cycle - state->current_cycle;
- cycles_delta = (*func)(block_trace, cycles_delta);
+ cycles_delta = (*func)(state, state->curr_pc,
+ block_trace, cycles_delta);
state->current_cycle = state->target_cycle - cycles_delta;
}
if (LOG_LEVEL >= INFO_L)
lightrec_print_info(state);
- return state->next_pc;
+ return state->curr_pc;
}
u32 lightrec_run_interpreter(struct lightrec_state *state, u32 pc,
else
lut_size = CODE_LUT_SIZE * sizeof(void *);
- init_jit(argv0);
+ init_jit_with_debug(argv0, stdout);
state = calloc(1, sizeof(*state) + lut_size);
if (!state)
state->tlsf = tlsf;
state->with_32bit_lut = with_32bit_lut;
+ state->in_delay_slot_n = 0xff;
+ state->cycles_per_op = 2;
state->block_cache = lightrec_blockcache_init(state);
if (!state->block_cache)
memset(state->code_lut, 0, lut_elm_size(state) * CODE_LUT_SIZE);
}
-void lightrec_set_invalidate_mode(struct lightrec_state *state, bool dma_only)
+void lightrec_set_unsafe_opt_flags(struct lightrec_state *state, u32 flags)
{
- if (state->invalidate_from_dma_only != dma_only)
+ if ((flags ^ state->opt_flags) & LIGHTREC_OPT_INV_DMA_ONLY)
lightrec_invalidate_all(state);
- state->invalidate_from_dma_only = dma_only;
+ state->opt_flags = flags;
}
void lightrec_set_exit_flags(struct lightrec_state *state, u32 flags)
{
return &state->regs;
}
+
+void lightrec_set_cycles_per_opcode(struct lightrec_state *state, u32 cycles)
+{
+ state->cycles_per_op = cycles;
+}
notaz.gp2x.de / pcsx_rearmed.git / blobdiff