#include <stdbool.h>
#include <stddef.h>
+#define LIGHTNING_UNALIGNED_32BIT 4
+
typedef void (*lightrec_rec_func_t)(struct lightrec_cstate *, const struct block *, u16);
/* Forward declarations */
const struct block *block, u16 offset,
u8 reg, u8 out_reg);
-static void unknown_opcode(struct lightrec_cstate *state, const struct block *block, u16 offset)
-{
- pr_warn("Unknown opcode: 0x%08x at PC 0x%08x\n",
- block->opcode_list[offset].c.opcode,
- block->pc + (offset << 2));
-}
-
static void
lightrec_jump_to_fn(jit_state_t *_jit, void (*fn)(void))
{
jit_state_t *_jit = block->_jit;
const struct opcode *op = &block->opcode_list[offset],
*ds = get_delay_slot(block->opcode_list, offset);
- u32 cycles = state->cycles + lightrec_cycles_of_opcode(op->c);
+ u32 cycles = state->cycles + lightrec_cycles_of_opcode(state->state, op->c);
+ bool has_ds = has_delay_slot(op->c);
jit_note(__FILE__, __LINE__);
update_ra_register(reg_cache, _jit, ra_reg, block->pc, link);
}
- if (has_delay_slot(op->c) &&
- !op_flag_no_ds(op->flags) && !op_flag_local_branch(op->flags)) {
- cycles += lightrec_cycles_of_opcode(ds->c);
+ if (has_ds && !op_flag_no_ds(op->flags) && !op_flag_local_branch(op->flags)) {
+ cycles += lightrec_cycles_of_opcode(state->state, ds->c);
/* Recompile the delay slot */
if (ds->c.opcode)
pr_debug("EOB: %u cycles\n", cycles);
}
- if (op_flag_load_delay(ds->flags)
- && opcode_is_load(ds->c) && !state->no_load_delay) {
+ if (has_ds && op_flag_load_delay(ds->flags)
+ && opcode_has_load_delay(ds->c) && !state->no_load_delay) {
/* If the delay slot is a load opcode, its target register
* will be written after the first opcode of the target is
* executed. Handle this by jumping to a special section of
} else {
lightrec_jump_to_eob(state, _jit);
}
+
+ lightrec_regcache_reset(reg_cache);
}
void lightrec_emit_jump_to_interpreter(struct lightrec_cstate *state,
* PC (which might have an offset) in JIT_V0. */
lightrec_load_imm(reg_cache, _jit, JIT_V0, block->pc,
block->pc + (offset << 2));
+ if (lightrec_store_next_pc()) {
+ jit_stxi_i(offsetof(struct lightrec_state, next_pc),
+ LIGHTREC_REG_STATE, JIT_V0);
+ }
+
jit_movi(JIT_V1, (uintptr_t)block);
jit_subi(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, state->cycles);
lightrec_load_imm(reg_cache, _jit, JIT_V0, block->pc,
block->pc + (offset << 2));
+ if (lightrec_store_next_pc()) {
+ jit_stxi_i(offsetof(struct lightrec_state, next_pc),
+ LIGHTREC_REG_STATE, JIT_V0);
+ }
+
jit_subi(LIGHTREC_REG_CYCLE, LIGHTREC_REG_CYCLE, state->cycles);
lightrec_jump_to_eob(state, _jit);
*ds = get_delay_slot(block->opcode_list, offset);
jit_node_t *addr;
bool is_forward = (s16)op->i.imm >= 0;
- int op_cycles = lightrec_cycles_of_opcode(op->c);
+ int op_cycles = lightrec_cycles_of_opcode(state->state, op->c);
u32 target_offset, cycles = state->cycles + op_cycles;
bool no_indirection = false;
u32 next_pc;
jit_note(__FILE__, __LINE__);
if (!op_flag_no_ds(op->flags))
- cycles += lightrec_cycles_of_opcode(ds->c);
+ cycles += lightrec_cycles_of_opcode(state->state, ds->c);
state->cycles = -op_cycles;
!op->i.rs, true);
}
+static void rec_alloc_rs_rd(struct regcache *reg_cache,
+ jit_state_t *_jit,
+ const struct opcode *op,
+ u8 rs, u8 rd,
+ u8 in_flags, u8 out_flags,
+ u8 *rs_out, u8 *rd_out)
+{
+ bool unload, discard;
+ u32 unload_flags;
+
+ if (OPT_EARLY_UNLOAD) {
+ unload_flags = LIGHTREC_FLAGS_GET_RS(op->flags);
+ unload = unload_flags == LIGHTREC_REG_UNLOAD;
+ discard = unload_flags == LIGHTREC_REG_DISCARD;
+ }
+
+ if (OPT_EARLY_UNLOAD && rs && rd != rs && (unload || discard)) {
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, rs, in_flags);
+ lightrec_remap_reg(reg_cache, _jit, rs, rd, discard);
+ lightrec_set_reg_out_flags(reg_cache, rs, out_flags);
+ rd = rs;
+ } else {
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, rs, in_flags);
+ rd = lightrec_alloc_reg_out(reg_cache, _jit, rd, out_flags);
+ }
+
+ *rs_out = rs;
+ *rd_out = rd;
+}
+
static void rec_alu_imm(struct lightrec_cstate *state, const struct block *block,
u16 offset, jit_code_t code, bool slti)
{
out_flags |= REG_ZEXT;
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, REG_EXT);
- rt = lightrec_alloc_reg_out(reg_cache, _jit, c.i.rt, out_flags);
+
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.i.rs, c.i.rt, REG_EXT, out_flags, &rs, &rt);
jit_new_node_www(code, rt, rs, (s32)(s16) c.i.imm);
u8 rd, rt, rs;
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, REG_EXT);
+
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, REG_EXT);
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rd,
- out_ext ? REG_EXT | REG_ZEXT : 0);
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.r.rs, c.r.rd, REG_EXT,
+ out_ext ? REG_EXT | REG_ZEXT : 0, &rs, &rd);
jit_new_node_www(code, rd, rs, rt);
u8 rd, rt, rs, temp, flags = 0;
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, 0);
if (code == jit_code_rshr)
flags = REG_EXT;
else if (code == jit_code_rshr_u)
flags = REG_ZEXT;
- rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, flags);
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rd, flags);
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, 0);
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.r.rt, c.r.rd, flags, flags, &rt, &rd);
- if (rs != rd && rt != rd) {
+ if (rt != rd) {
jit_andi(rd, rs, 0x1f);
jit_new_node_www(code, rd, rt, rd);
} else {
union code c = block->opcode_list[offset].c;
jit_state_t *_jit = block->_jit;
u16 flags = REG_EXT;
+ s32 value = (s32)(s16) c.i.imm;
u8 rt;
- if (!(c.i.imm & 0x8000))
+ if (block->opcode_list[offset].flags & LIGHTREC_MOVI)
+ value += (s32)((u32)state->movi_temp[c.i.rt] << 16);
+
+ if (value >= 0)
flags |= REG_ZEXT;
rt = lightrec_alloc_reg_out(reg_cache, _jit, c.i.rt, flags);
- jit_movi(rt, (s32)(s16) c.i.imm);
+ jit_movi(rt, value);
lightrec_free_reg(reg_cache, rt);
}
static void rec_ADDIU(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
+ const struct opcode *op = &block->opcode_list[offset];
+
_jit_name(block->_jit, __func__);
- if (block->opcode_list[offset].c.i.rs)
+ if (op->i.rs && !(op->flags & LIGHTREC_MOVI))
rec_alu_imm(state, block, offset, jit_code_addi, false);
else
rec_movi(state, block, offset);
_jit_name(block->_jit, __func__);
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, 0);
- rt = lightrec_alloc_reg_out(reg_cache, _jit, c.i.rt,
- REG_EXT | REG_ZEXT);
+
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.i.rs, c.i.rt, 0, REG_EXT | REG_ZEXT, &rs, &rt);
/* PSX code uses ANDI 0xff / ANDI 0xffff a lot, which are basically
* casts to uint8_t / uint16_t. */
u8 rs, rt, flags;
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, 0);
- rt = lightrec_alloc_reg_out(reg_cache, _jit, c.i.rt, 0);
+
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.i.rs, c.i.rt, 0, 0, &rs, &rt);
flags = lightrec_get_reg_in_flags(reg_cache, rs);
lightrec_set_reg_out_flags(reg_cache, rt, flags);
static void rec_ORI(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
- _jit_name(block->_jit, __func__);
- rec_alu_or_xor(state, block, offset, jit_code_ori);
+ const struct opcode *op = &block->opcode_list[offset];
+ struct regcache *reg_cache = state->reg_cache;
+ jit_state_t *_jit = block->_jit;
+ s32 val;
+ u8 rt;
+
+ _jit_name(_jit, __func__);
+
+ if (op->flags & LIGHTREC_MOVI) {
+ rt = lightrec_alloc_reg_out(reg_cache, _jit, op->i.rt, REG_EXT);
+
+ val = ((u32)state->movi_temp[op->i.rt] << 16) | op->i.imm;
+ jit_movi(rt, val);
+
+ lightrec_free_reg(reg_cache, rt);
+ } else {
+ rec_alu_or_xor(state, block, offset, jit_code_ori);
+ }
}
static void rec_XORI(struct lightrec_cstate *state,
jit_state_t *_jit = block->_jit;
u8 rt, flags = REG_EXT;
+ if (block->opcode_list[offset].flags & LIGHTREC_MOVI) {
+ state->movi_temp[c.i.rt] = c.i.imm;
+ return;
+ }
+
jit_name(__func__);
jit_note(__FILE__, __LINE__);
_jit_name(block->_jit, __func__);
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, 0);
+
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, 0);
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rd, 0);
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.r.rs, c.r.rd, 0, 0, &rs, &rd);
flags_rs = lightrec_get_reg_in_flags(reg_cache, rs);
flags_rt = lightrec_get_reg_in_flags(reg_cache, rt);
u8 rd, rt, rs, flags_rs, flags_rt, flags_rd = 0;
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, 0);
+
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, 0);
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rd, 0);
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.r.rs, c.r.rd, 0, 0, &rs, &rd);
flags_rs = lightrec_get_reg_in_flags(reg_cache, rs);
flags_rt = lightrec_get_reg_in_flags(reg_cache, rt);
_jit_name(block->_jit, __func__);
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rs, 0);
+
rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, 0);
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rd, 0);
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.r.rs, c.r.rd, 0, 0, &rs, &rd);
flags_rs = lightrec_get_reg_in_flags(reg_cache, rs);
flags_rt = lightrec_get_reg_in_flags(reg_cache, rt);
struct regcache *reg_cache = state->reg_cache;
union code c = block->opcode_list[offset].c;
jit_state_t *_jit = block->_jit;
- u8 rd, rt, flags = 0;
+ u8 rd, rt, flags = 0, out_flags = 0;
jit_note(__FILE__, __LINE__);
else if (code == jit_code_rshi_u)
flags = REG_ZEXT;
- rt = lightrec_alloc_reg_in(reg_cache, _jit, c.r.rt, flags);
-
/* Input reg is zero-extended, if we SRL at least by one bit, we know
* the output reg will be both zero-extended and sign-extended. */
+ out_flags = flags;
if (code == jit_code_rshi_u && c.r.imm)
- flags |= REG_EXT;
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.r.rd, flags);
+ out_flags |= REG_EXT;
+
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.r.rt, c.r.rd, flags, out_flags, &rt, &rd);
jit_new_node_www(code, rd, rt, c.r.imm);
u8 reg_hi = get_mult_div_hi(c);
jit_state_t *_jit = block->_jit;
u8 lo, hi, rs, rt, rflags = 0;
+ bool no_lo = op_flag_no_lo(flags);
+ bool no_hi = op_flag_no_hi(flags);
jit_note(__FILE__, __LINE__);
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 (!op_flag_no_lo(flags))
+ if (!no_lo)
lo = lightrec_alloc_reg_out(reg_cache, _jit, reg_lo, 0);
- else if (__WORDSIZE == 32)
- lo = lightrec_alloc_reg_temp(reg_cache, _jit);
- if (!op_flag_no_hi(flags))
+ if (!no_hi)
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 (!op_flag_no_hi(flags)) {
+ if (no_lo) {
if (is_signed)
- jit_qmulr(lo, hi, rs, rt);
+ jit_hmulr(hi, rs, rt);
else
- jit_qmulr_u(lo, hi, rs, rt);
- } else {
+ jit_hmulr_u(hi, rs, rt);
+ } else if (no_hi) {
jit_mulr(lo, rs, rt);
+ } else if (is_signed) {
+ jit_qmulr(lo, hi, rs, rt);
+ } else {
+ jit_qmulr_u(lo, hi, rs, rt);
}
} else {
/* On 64-bit systems, do a 64*64->64 bit operation. */
- if (op_flag_no_lo(flags)) {
+ if (no_lo) {
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 (!op_flag_no_hi(flags))
+ if (!no_hi)
jit_rshi(hi, lo, 32);
}
}
lightrec_free_reg(reg_cache, rs);
lightrec_free_reg(reg_cache, rt);
- if (!op_flag_no_lo(flags) || __WORDSIZE == 32)
+ if (!no_lo)
lightrec_free_reg(reg_cache, lo);
- if (!op_flag_no_hi(flags))
+ if (!no_hi)
lightrec_free_reg(reg_cache, hi);
}
}
static void rec_alu_mv_lo_hi(struct lightrec_cstate *state,
- const struct block *block, u8 dst, u8 src)
+ const struct block *block, u16 offset,
+ u8 dst, u8 src)
{
struct regcache *reg_cache = state->reg_cache;
jit_state_t *_jit = block->_jit;
jit_note(__FILE__, __LINE__);
- src = lightrec_alloc_reg_in(reg_cache, _jit, src, 0);
- dst = lightrec_alloc_reg_out(reg_cache, _jit, dst, REG_EXT);
+
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ src, dst, 0, REG_EXT, &src, &dst);
jit_extr_i(dst, src);
union code c = block->opcode_list[offset].c;
_jit_name(block->_jit, __func__);
- rec_alu_mv_lo_hi(state, block, c.r.rd, REG_HI);
+ rec_alu_mv_lo_hi(state, block, offset, c.r.rd, REG_HI);
}
static void rec_special_MTHI(struct lightrec_cstate *state,
union code c = block->opcode_list[offset].c;
_jit_name(block->_jit, __func__);
- rec_alu_mv_lo_hi(state, block, REG_HI, c.r.rs);
+ rec_alu_mv_lo_hi(state, block, offset, REG_HI, c.r.rs);
}
static void rec_special_MFLO(struct lightrec_cstate *state,
union code c = block->opcode_list[offset].c;
_jit_name(block->_jit, __func__);
- rec_alu_mv_lo_hi(state, block, c.r.rd, REG_LO);
+ rec_alu_mv_lo_hi(state, block, offset, c.r.rd, REG_LO);
}
static void rec_special_MTLO(struct lightrec_cstate *state,
union code c = block->opcode_list[offset].c;
_jit_name(block->_jit, __func__);
- rec_alu_mv_lo_hi(state, block, REG_LO, c.r.rs);
+ rec_alu_mv_lo_hi(state, block, offset, REG_LO, c.r.rs);
}
static void call_to_c_wrapper(struct lightrec_cstate *state,
}
}
-static u32 rec_ram_mask(struct lightrec_state *state)
+static u32 rec_ram_mask(const struct lightrec_state *state)
{
return (RAM_SIZE << (state->mirrors_mapped * 2)) - 1;
}
{
u32 length = state->maps[PSX_MAP_HW_REGISTERS].length;
- return GENMASK(31 - clz32(length - 1), 0);
+ return 0x1f800000 | GENMASK(31 - clz32(length - 1), 0);
}
static void rec_store_memory(struct lightrec_cstate *cstate,
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;
+ u8 rs, rt, tmp = 0, tmp2 = 0, 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 &&
+ (c.i.op == OP_META_SWU
+ || (!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;
+ bool need_tmp = !no_mask || add_imm || invalidate;
bool swc2 = c.i.op == OP_SWC2;
u8 in_reg = swc2 ? REG_TEMP : c.i.rt;
+ s8 reg_imm;
- rt = lightrec_alloc_reg_in(reg_cache, _jit, in_reg, 0);
rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, 0);
if (need_tmp)
tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
if (add_imm) {
jit_addi(tmp, addr_reg, (s16)c.i.imm);
+ lightrec_free_reg(reg_cache, rs);
addr_reg = tmp;
imm = 0;
} else if (simm) {
}
if (!no_mask) {
- jit_andi(tmp, addr_reg, addr_mask);
+ reg_imm = lightrec_alloc_reg_temp_with_value(reg_cache, _jit,
+ addr_mask);
+
+ jit_andr(tmp, addr_reg, reg_imm);
addr_reg = tmp;
+
+ lightrec_free_reg(reg_cache, reg_imm);
}
if (addr_offset) {
+ reg_imm = lightrec_alloc_reg_temp_with_value(reg_cache, _jit,
+ addr_offset);
tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
- jit_addi(tmp2, addr_reg, addr_offset);
+ jit_addr(tmp2, addr_reg, reg_imm);
addr_reg2 = tmp2;
+
+ lightrec_free_reg(reg_cache, reg_imm);
} else {
addr_reg2 = addr_reg;
}
+ rt = lightrec_alloc_reg_in(reg_cache, _jit, in_reg, 0);
+
if (is_big_endian() && swap_code && in_reg) {
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);
+
+ if (c.i.op == OP_META_SWU)
+ jit_unstr(addr_reg2, tmp3, LIGHTNING_UNALIGNED_32BIT);
+ else
+ jit_new_node_www(code, imm, addr_reg2, tmp3);
lightrec_free_reg(reg_cache, tmp3);
+ } else if (c.i.op == OP_META_SWU) {
+ jit_unstr(addr_reg2, rt, LIGHTNING_UNALIGNED_32BIT);
} else {
jit_new_node_www(code, imm, addr_reg2, rt);
}
u16 offset, jit_code_t code,
jit_code_t swap_code, bool invalidate)
{
- struct lightrec_state *state = cstate->state;
+ const struct lightrec_state *state = cstate->state;
_jit_note(block->_jit, __FILE__, __LINE__);
u16 offset, jit_code_t code,
jit_code_t swap_code)
{
- struct lightrec_state *state = cstate->state;
+ const struct lightrec_state *state = cstate->state;
struct regcache *reg_cache = cstate->reg_cache;
union code c = block->opcode_list[offset].c;
jit_state_t *_jit = block->_jit;
jit_node_t *to_not_ram, *to_end;
bool swc2 = c.i.op == OP_SWC2;
- bool offset_ram_or_scratch = state->offset_ram || state->offset_scratch;
- u8 tmp, tmp2, rs, rt, in_reg = swc2 ? REG_TEMP : c.i.rt;
+ u8 tmp, tmp2 = 0, rs, rt, in_reg = swc2 ? REG_TEMP : c.i.rt;
+ u32 addr_mask;
+ s32 reg_imm;
s16 imm;
jit_note(__FILE__, __LINE__);
rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, 0);
tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
- if (offset_ram_or_scratch)
- tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
+ if (state->mirrors_mapped)
+ addr_mask = 0x1f800000 | (4 * RAM_SIZE - 1);
+ else
+ addr_mask = 0x1f800000 | (RAM_SIZE - 1);
+
+ reg_imm = lightrec_alloc_reg_temp_with_value(reg_cache, _jit, addr_mask);
/* Convert to KUNSEG and avoid RAM mirrors */
- if (state->mirrors_mapped) {
- imm = (s16)c.i.imm;
- jit_andi(tmp, rs, 0x1f800000 | (4 * RAM_SIZE - 1));
- } else if (c.i.imm) {
+ if ((c.i.op == OP_META_SWU || !state->mirrors_mapped) && c.i.imm) {
imm = 0;
jit_addi(tmp, rs, (s16)c.i.imm);
- jit_andi(tmp, tmp, 0x1f800000 | (RAM_SIZE - 1));
+ jit_andr(tmp, tmp, reg_imm);
} else {
- imm = 0;
- jit_andi(tmp, rs, 0x1f800000 | (RAM_SIZE - 1));
+ imm = (s16)c.i.imm;
+ jit_andr(tmp, rs, reg_imm);
}
lightrec_free_reg(reg_cache, rs);
+ lightrec_free_reg(reg_cache, reg_imm);
if (state->offset_ram != state->offset_scratch) {
+ tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
+
to_not_ram = jit_bmsi(tmp, BIT(28));
jit_movi(tmp2, state->offset_ram);
jit_movi(tmp2, state->offset_scratch);
jit_patch(to_end);
} else if (state->offset_ram) {
- jit_movi(tmp2, state->offset_ram);
+ tmp2 = lightrec_alloc_reg_temp_with_value(reg_cache, _jit,
+ state->offset_ram);
}
- if (offset_ram_or_scratch) {
+ if (state->offset_ram || state->offset_scratch) {
jit_addr(tmp, tmp, tmp2);
lightrec_free_reg(reg_cache, tmp2);
}
tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
jit_new_node_ww(swap_code, tmp2, rt);
- jit_new_node_www(code, imm, tmp, tmp2);
+
+ if (c.i.op == OP_META_SWU)
+ jit_unstr(tmp, tmp2, LIGHTNING_UNALIGNED_32BIT);
+ else
+ jit_new_node_www(code, imm, tmp, tmp2);
lightrec_free_reg(reg_cache, tmp2);
+ } else if (c.i.op == OP_META_SWU) {
+ jit_unstr(tmp, rt, LIGHTNING_UNALIGNED_32BIT);
} else {
jit_new_node_www(code, imm, tmp, rt);
}
static void rec_store_direct(struct lightrec_cstate *cstate, const struct block *block,
u16 offset, jit_code_t code, jit_code_t swap_code)
{
- struct lightrec_state *state = cstate->state;
+ const struct lightrec_state *state = cstate->state;
u32 ram_size = state->mirrors_mapped ? RAM_SIZE * 4 : RAM_SIZE;
struct regcache *reg_cache = cstate->reg_cache;
union code c = block->opcode_list[offset].c;
bool swc2 = c.i.op == OP_SWC2;
u8 tmp, tmp2, tmp3, masked_reg, rs, rt;
u8 in_reg = swc2 ? REG_TEMP : c.i.rt;
+ u32 addr_mask = 0x1f800000 | (ram_size - 1);
+ bool different_offsets = state->offset_ram != state->offset_scratch;
+ s32 reg_imm;
jit_note(__FILE__, __LINE__);
tmp2 = lightrec_alloc_reg_temp(reg_cache, _jit);
tmp3 = lightrec_alloc_reg_in(reg_cache, _jit, 0, 0);
+ reg_imm = lightrec_alloc_reg_temp_with_value(reg_cache, _jit, addr_mask);
+
/* Convert to KUNSEG and avoid RAM mirrors */
if (c.i.imm) {
jit_addi(tmp2, rs, (s16)c.i.imm);
- jit_andi(tmp2, tmp2, 0x1f800000 | (ram_size - 1));
+ jit_andr(tmp2, tmp2, reg_imm);
} else {
- jit_andi(tmp2, rs, 0x1f800000 | (ram_size - 1));
+ jit_andr(tmp2, rs, reg_imm);
}
lightrec_free_reg(reg_cache, rs);
+ lightrec_free_reg(reg_cache, reg_imm);
tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
- if (state->offset_ram != state->offset_scratch) {
+ if (different_offsets) {
to_not_ram = jit_bgti(tmp2, ram_size);
masked_reg = tmp2;
} else {
else
jit_stxi(offsetof(struct lightrec_state, code_lut), tmp, tmp3);
- if (state->offset_ram != state->offset_scratch) {
+ if (c.i.op == OP_META_SWU) {
+ /* With a SWU opcode, we might have touched the following 32-bit
+ * word, so invalidate it as well */
+ if (lut_is_32bit(state)) {
+ jit_stxi_i(offsetof(struct lightrec_state, code_lut) + 4,
+ tmp, tmp3);
+ } else {
+ jit_stxi(offsetof(struct lightrec_state, code_lut)
+ + sizeof(uintptr_t), tmp, tmp3);
+ }
+ }
+
+ if (different_offsets) {
jit_movi(tmp, state->offset_ram);
to_end = jit_b();
if (state->offset_ram || state->offset_scratch)
jit_movi(tmp, state->offset_scratch);
- if (state->offset_ram != state->offset_scratch)
+ if (different_offsets)
jit_patch(to_end);
if (state->offset_ram || state->offset_scratch)
tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
jit_new_node_ww(swap_code, tmp, rt);
- jit_new_node_www(code, 0, tmp2, tmp);
+
+ if (c.i.op == OP_META_SWU)
+ jit_unstr(tmp2, tmp, LIGHTNING_UNALIGNED_32BIT);
+ else
+ jit_new_node_www(code, 0, tmp2, tmp);
lightrec_free_reg(reg_cache, tmp);
+ } else if (c.i.op == OP_META_SWU) {
+ jit_unstr(tmp2, rt, LIGHTNING_UNALIGNED_32BIT);
} else {
jit_new_node_www(code, 0, tmp2, rt);
}
u32 flags = block->opcode_list[offset].flags;
u32 mode = LIGHTREC_FLAGS_GET_IO_MODE(flags);
bool no_invalidate = op_flag_no_invalidate(flags) ||
- state->state->invalidate_from_dma_only;
+ (state->state->opt_flags & LIGHTREC_OPT_INV_DMA_ONLY);
union code c = block->opcode_list[offset].c;
bool is_swc2 = c.i.op == OP_SWC2;
u8 rs, rt, out_reg, addr_reg, flags = REG_EXT;
bool no_mask = op_flag_no_mask(op->flags);
union code c = op->c;
+ s8 reg_imm;
s16 imm;
if (load_delay || c.i.op == OP_LWC2)
rs = lightrec_alloc_reg_in(reg_cache, _jit, c.i.rs, 0);
rt = lightrec_alloc_reg_out(reg_cache, _jit, out_reg, flags);
- if (!cstate->state->mirrors_mapped && c.i.imm && !no_mask) {
+ if ((op->i.op == OP_META_LWU && c.i.imm)
+ || (!cstate->state->mirrors_mapped && c.i.imm && !no_mask)) {
jit_addi(rt, rs, (s16)c.i.imm);
addr_reg = rt;
imm = 0;
imm = (s16)c.i.imm;
}
+ if (op->i.op == OP_META_LWU)
+ imm = LIGHTNING_UNALIGNED_32BIT;
+
if (!no_mask) {
- jit_andi(rt, addr_reg, addr_mask);
+ reg_imm = lightrec_alloc_reg_temp_with_value(reg_cache, _jit,
+ addr_mask);
+
+ jit_andr(rt, addr_reg, reg_imm);
addr_reg = rt;
+
+ lightrec_free_reg(reg_cache, reg_imm);
}
if (addr_offset) {
- jit_addi(rt, addr_reg, addr_offset);
+ reg_imm = lightrec_alloc_reg_temp_with_value(reg_cache, _jit,
+ addr_offset);
+
+ jit_addr(rt, addr_reg, reg_imm);
addr_reg = rt;
+
+ lightrec_free_reg(reg_cache, reg_imm);
}
jit_new_node_www(code, rt, addr_reg, imm);
jit_code_t code, jit_code_t swap_code,
bool is_unsigned)
{
- struct lightrec_state *state = cstate->state;
+ const struct lightrec_state *state = cstate->state;
struct regcache *reg_cache = cstate->reg_cache;
struct opcode *op = &block->opcode_list[offset];
bool load_delay = op_flag_load_delay(op->flags) && !cstate->no_load_delay;
jit_state_t *_jit = block->_jit;
jit_node_t *to_not_ram, *to_not_bios, *to_end, *to_end2;
u8 tmp, rs, rt, out_reg, addr_reg, flags = REG_EXT;
+ bool different_offsets = state->offset_bios != state->offset_scratch;
union code c = op->c;
+ s32 addr_mask;
+ u32 reg_imm;
+ s8 offt_reg;
s16 imm;
if (load_delay || c.i.op == OP_LWC2)
if ((state->offset_ram == state->offset_bios &&
state->offset_ram == state->offset_scratch &&
- state->mirrors_mapped) || !c.i.imm) {
+ state->mirrors_mapped && c.i.op != OP_META_LWU)
+ || !c.i.imm) {
addr_reg = rs;
imm = (s16)c.i.imm;
} else {
lightrec_free_reg(reg_cache, rs);
}
+ if (op->i.op == OP_META_LWU)
+ imm = LIGHTNING_UNALIGNED_32BIT;
+
tmp = lightrec_alloc_reg_temp(reg_cache, _jit);
if (state->offset_ram == state->offset_bios &&
state->offset_ram == state->offset_scratch) {
+ if (!state->mirrors_mapped)
+ addr_mask = 0x1f800000 | (RAM_SIZE - 1);
+ else
+ addr_mask = 0x1fffffff;
+
+ reg_imm = lightrec_alloc_reg_temp_with_value(reg_cache, _jit,
+ addr_mask);
if (!state->mirrors_mapped) {
jit_andi(tmp, addr_reg, BIT(28));
jit_rshi_u(tmp, tmp, 28 - 22);
- jit_ori(tmp, tmp, 0x1f800000 | (RAM_SIZE - 1));
+ jit_orr(tmp, tmp, reg_imm);
jit_andr(rt, addr_reg, tmp);
} else {
- jit_andi(rt, addr_reg, 0x1fffffff);
+ jit_andr(rt, addr_reg, reg_imm);
}
- if (state->offset_ram)
- jit_movi(tmp, state->offset_ram);
+ lightrec_free_reg(reg_cache, reg_imm);
+
+ if (state->offset_ram) {
+ offt_reg = lightrec_get_reg_with_value(reg_cache,
+ state->offset_ram);
+ if (offt_reg < 0) {
+ jit_movi(tmp, state->offset_ram);
+ lightrec_temp_set_value(reg_cache, tmp,
+ state->offset_ram);
+ } else {
+ lightrec_free_reg(reg_cache, tmp);
+ tmp = offt_reg;
+ }
+ }
} else {
to_not_ram = jit_bmsi(addr_reg, BIT(28));
jit_patch(to_not_ram);
- if (state->offset_bios != state->offset_scratch)
+ if (different_offsets)
to_not_bios = jit_bmci(addr_reg, BIT(22));
/* Convert to KUNSEG */
jit_movi(tmp, state->offset_bios);
- if (state->offset_bios != state->offset_scratch) {
+ if (different_offsets) {
to_end2 = jit_b();
jit_patch(to_not_bios);
rec_load(state, block, offset, code, jit_code_bswapr_ui, false);
}
-static void rec_break_syscall(struct lightrec_cstate *state,
- const struct block *block, u16 offset,
- u32 exit_code)
+static void rec_exit_early(struct lightrec_cstate *state,
+ const struct block *block, u16 offset,
+ u32 exit_code, u32 pc)
{
struct regcache *reg_cache = state->reg_cache;
jit_state_t *_jit = block->_jit;
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,
- get_ds_pc(block, offset, 0),
- 31, 0, true);
+ lightrec_emit_end_of_block(state, block, offset, -1, pc, 31, 0, true);
}
static void rec_special_SYSCALL(struct lightrec_cstate *state,
const struct block *block, u16 offset)
{
_jit_name(block->_jit, __func__);
- rec_break_syscall(state, block, offset, LIGHTREC_EXIT_SYSCALL);
+
+ /* TODO: the return address should be "pc - 4" if we're a delay slot */
+ rec_exit_early(state, block, offset, LIGHTREC_EXIT_SYSCALL,
+ get_ds_pc(block, offset, 0));
}
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, LIGHTREC_EXIT_BREAK);
+ rec_exit_early(state, block, offset, LIGHTREC_EXIT_BREAK,
+ get_ds_pc(block, offset, 0));
}
static void rec_mfc(struct lightrec_cstate *state, const struct block *block, u16 offset)
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);
+ state->cycles += lightrec_cycles_of_opcode(state->state, c);
lightrec_emit_eob(state, block, offset + 1);
}
}
union code c = op->c;
jit_state_t *_jit = block->_jit;
bool unload_rd;
+ bool unload_rs, discard_rs;
u8 rs, rd;
_jit_name(block->_jit, __func__);
jit_note(__FILE__, __LINE__);
+ unload_rs = OPT_EARLY_UNLOAD
+ && LIGHTREC_FLAGS_GET_RS(op->flags) == LIGHTREC_REG_UNLOAD;
+ discard_rs = OPT_EARLY_UNLOAD
+ && LIGHTREC_FLAGS_GET_RS(op->flags) == LIGHTREC_REG_DISCARD;
+
+ if ((unload_rs || discard_rs) && c.m.rs) {
+ /* If the source register is going to be unloaded or discarded,
+ * then we can simply mark its host register as now pointing to
+ * the destination register. */
+ pr_debug("Remap %s to %s at offset 0x%x\n",
+ lightrec_reg_name(c.m.rs), lightrec_reg_name(c.m.rd),
+ offset << 2);
+ rs = lightrec_alloc_reg_in(reg_cache, _jit, c.m.rs, 0);
+ lightrec_remap_reg(reg_cache, _jit, rs, c.m.rd, discard_rs);
+ lightrec_free_reg(reg_cache, rs);
+ return;
+ }
+
unload_rd = OPT_EARLY_UNLOAD
&& LIGHTREC_FLAGS_GET_RD(op->flags) == LIGHTREC_REG_UNLOAD;
_jit_name(block->_jit, __func__);
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.m.rs, 0);
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.m.rd, REG_EXT);
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.m.rs, c.m.rd, 0, REG_EXT, &rs, &rd);
if (c.m.op == OP_META_EXTC)
jit_extr_c(rd, rs);
hi = lightrec_alloc_reg_out(reg_cache, _jit,
reg_hi, hiflags);
- if (c.r.op >= 32)
+ 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);
+ } else if (is_signed) {
+ if (c.r.op)
+ jit_rshi(hi, rs, 32 - c.r.op);
+ else
+ jit_rshi(hi, rs, 31);
+ } else {
+ if (c.r.op)
+ jit_rshi_u(hi, rs, 32 - c.r.op);
+ else
+ jit_movi(hi, 0);
+ }
lightrec_free_reg(reg_cache, hi);
}
u8 rd, rs, flags;
jit_note(__FILE__, __LINE__);
- rs = lightrec_alloc_reg_in(reg_cache, _jit, c.m.rs, 0);
- rd = lightrec_alloc_reg_out(reg_cache, _jit, c.m.rd, 0);
+
+ rec_alloc_rs_rd(reg_cache, _jit, &block->opcode_list[offset],
+ c.m.rs, c.m.rd, 0, 0, &rs, &rd);
flags = lightrec_get_reg_in_flags(reg_cache, rs);
lightrec_free_reg(reg_cache, rd);
}
+static void rec_meta_LWU(struct lightrec_cstate *state,
+ const struct block *block,
+ u16 offset)
+{
+ jit_code_t code;
+
+ if (is_big_endian() && __WORDSIZE == 64)
+ code = jit_code_unldr_u;
+ else
+ code = jit_code_unldr;
+
+ _jit_name(block->_jit, __func__);
+ rec_load(state, block, offset, code, jit_code_bswapr_ui, false);
+}
+
+static void rec_meta_SWU(struct lightrec_cstate *state,
+ const struct block *block,
+ u16 offset)
+{
+ _jit_name(block->_jit, __func__);
+ rec_store(state, block, offset, jit_code_unstr, jit_code_bswapr_ui);
+}
+
+static void unknown_opcode(struct lightrec_cstate *state,
+ const struct block *block, u16 offset)
+{
+ rec_exit_early(state, block, offset, LIGHTREC_EXIT_UNKNOWN_OP,
+ block->pc + (offset << 2));
+}
+
static const lightrec_rec_func_t rec_standard[64] = {
SET_DEFAULT_ELM(rec_standard, unknown_opcode),
[OP_SPECIAL] = rec_SPECIAL,
[OP_META] = rec_META,
[OP_META_MULT2] = rec_meta_MULT2,
[OP_META_MULTU2] = rec_meta_MULT2,
+ [OP_META_LWU] = rec_meta_LWU,
+ [OP_META_SWU] = rec_meta_SWU,
};
static const lightrec_rec_func_t rec_special[64] = {
notaz.gp2x.de / pcsx_rearmed.git / blobdiff