/*
* vim:shiftwidth=2:expandtab
+ *
+ * notes:
+ * - tcache, block descriptor, link buffer overflows result in sh2_translate()
+ * failure, followed by full tcache invalidation for that region
+ * - jumps between blocks are tracked for SMC handling (in block_links[])
*/
#include <stddef.h>
#include <stdio.h>
#define BLOCK_CYCLE_LIMIT 100
#define MAX_BLOCK_SIZE (BLOCK_CYCLE_LIMIT * 6 * 6)
+#define TCACHE_BUFFERS 3
// we have 3 translation cache buffers, split from one drc/cmn buffer.
// BIOS shares tcache with data array because it's only used for init
// and can be discarded early
// XXX: need to tune sizes
-static const int tcache_sizes[3] = {
+static const int tcache_sizes[TCACHE_BUFFERS] = {
DRC_TCACHE_SIZE * 6 / 8, // ROM, DRAM
DRC_TCACHE_SIZE / 8, // BIOS, data array in master sh2
DRC_TCACHE_SIZE / 8, // ... slave
};
-static u8 *tcache_bases[3];
-static u8 *tcache_ptrs[3];
+static u8 *tcache_bases[TCACHE_BUFFERS];
+static u8 *tcache_ptrs[TCACHE_BUFFERS];
// ptr for code emiters
static u8 *tcache_ptr;
+typedef struct block_desc_ {
+ u32 addr; // SH2 PC address
+ u32 end_addr; // TODO rm?
+ void *tcache_ptr; // translated block for above PC
+ struct block_desc_ *next; // next block with the same PC hash
+#if (DRC_DEBUG & 1)
+ int refcount;
+#endif
+} block_desc;
+
+typedef struct block_link_ {
+ u32 target_pc;
+ void *jump;
+// struct block_link_ *next;
+} block_link;
+
+static const int block_max_counts[TCACHE_BUFFERS] = {
+ 4*1024,
+ 256,
+ 256,
+};
+static block_desc *block_tables[TCACHE_BUFFERS];
+static block_link *block_links[TCACHE_BUFFERS];
+static int block_counts[TCACHE_BUFFERS];
+static int block_link_counts[TCACHE_BUFFERS];
+
// host register tracking
enum {
HR_FREE,
#define Q_SHIFT 8
#define M_SHIFT 9
-typedef struct block_desc_ {
- u32 addr; // SH2 PC address
- u32 end_addr; // TODO rm?
- void *tcache_ptr; // translated block for above PC
- struct block_desc_ *next; // next block with the same PC hash
-#if (DRC_DEBUG & 1)
- int refcount;
-#endif
-} block_desc;
-
-static const int block_max_counts[3] = {
- 4*1024,
- 256,
- 256,
-};
-static block_desc *block_tables[3];
-static int block_counts[3];
-
// ROM hash table
#define MAX_HASH_ENTRIES 1024
#define HASH_MASK (MAX_HASH_ENTRIES - 1)
#endif
}
+// add block links (tracked branches)
+static int dr_add_block_link(u32 target_pc, void *jump, int tcache_id)
+{
+ block_link *bl = block_links[tcache_id];
+ int cnt = block_link_counts[tcache_id];
+
+ if (cnt >= block_max_counts[tcache_id] * 2) {
+ printf("bl overflow for tcache %d\n", tcache_id);
+ return -1;
+ }
+
+ bl[cnt].target_pc = target_pc;
+ bl[cnt].jump = jump;
+ block_link_counts[tcache_id]++;
+
+ return 0;
+}
+
static void *dr_find_block(block_desc *tab, u32 addr)
{
for (tab = tab->next; tab != NULL; tab = tab->next)
int *bcount = &block_counts[tcache_id];
block_desc *bd;
- if (*bcount >= block_max_counts[tcache_id])
+ if (*bcount >= block_max_counts[tcache_id]) {
+ printf("bd overflow for tcache %d\n", tcache_id);
return NULL;
+ }
bd = &block_tables[tcache_id][*bcount];
bd->addr = addr;
return bd;
}
+#define ADD_TO_ARRAY(array, count, item, failcode) \
+ array[count++] = item; \
+ if (count >= ARRAY_SIZE(array)) { \
+ printf("warning: " #array " overflow\n"); \
+ failcode; \
+ }
+
int find_in_array(u32 *array, size_t size, u32 what)
{
size_t i;
return block;
}
+void dr_link_blocks(void *target, u32 pc, int tcache_id)
+{
+ block_link *bl = block_links[tcache_id];
+ int cnt = block_link_counts[tcache_id];
+ int i;
+
+ for (i = 0; i < cnt; i++) {
+ if (bl[i].target_pc == pc) {
+ dbg(1, "- link from %p", bl[i].jump);
+ emith_jump_patch(bl[i].jump, target);
+ // XXX: sync ARM caches (old jump should be fine)?
+ }
+ }
+}
+
+void *dr_prepare_ext_branch(u32 pc, SH2 *sh2, int tcache_id)
+{
+ int target_tcache_id;
+ void *target;
+ int ret;
+
+ target = lookup_block(pc, sh2->is_slave, &target_tcache_id);
+ if (target_tcache_id == tcache_id) {
+ // allow linking blocks only from local cache
+ ret = dr_add_block_link(pc, tcache_ptr, tcache_id);
+ if (ret < 0)
+ return NULL;
+ }
+ if (target == NULL || target_tcache_id != tcache_id)
+ target = sh2_drc_dispatcher;
+
+ return target;
+}
+
#define DELAYED_OP \
drcf.delayed_op = 2
if (GET_Fx() >= n) \
goto default_
-#define MAX_LOCAL_BRANCHES 16
+#define MAX_LOCAL_BRANCHES 32
// op_flags: data from 1st pass
#define OP_FLAGS(pc) op_flags[((pc) - base_pc) / 2]
void *branch_patch_ptr[MAX_LOCAL_BRANCHES];
u32 branch_patch_pc[MAX_LOCAL_BRANCHES];
int branch_patch_count = 0;
- int branch_patch_cond = -1;
+ int pending_branch_cond = -1;
+ int pending_branch_pc = 0;
u8 op_flags[BLOCK_CYCLE_LIMIT + 1];
struct {
u32 delayed_op:2;
u32 use_saved_t:1; // delayed op modifies T
} drcf = { 0, };
+ // PC of current, first, last, last_target_blk SH2 insn
+ u32 pc, base_pc, end_pc, out_pc;
void *block_entry;
block_desc *this_block;
- u32 pc, base_pc, end_pc; // PC of current, first, last insn
int blkid_main = 0;
u32 tmp, tmp2;
int cycles;
tcache_ptr = tcache_ptrs[tcache_id];
this_block = dr_add_block(base_pc, tcache_id, &blkid_main);
+ if (this_block == NULL)
+ return NULL;
// predict tcache overflow
tmp = tcache_ptr - tcache_bases[tcache_id];
- if (tmp > tcache_sizes[tcache_id] - MAX_BLOCK_SIZE || this_block == NULL)
+ if (tmp > tcache_sizes[tcache_id] - MAX_BLOCK_SIZE) {
+ printf("tcache %d overflow\n", tcache_id);
return NULL;
+ }
block_entry = tcache_ptr;
dbg(1, "== %csh2 block #%d,%d %08x -> %p", sh2->is_slave ? 's' : 'm',
tcache_id, blkid_main, base_pc, block_entry);
+ dr_link_blocks(tcache_ptr, base_pc, tcache_id);
+
// 1st pass: scan forward for local branches
memset(op_flags, 0, sizeof(op_flags));
for (cycles = 0, pc = base_pc; cycles < BLOCK_CYCLE_LIMIT; cycles++, pc += 2) {
op = p32x_sh2_read16(pc, sh2);
if ((op & 0xf000) == 0xa000 || (op & 0xf000) == 0xb000) { // BRA, BSR
+ signed int offs = ((signed int)(op << 20) >> 19);
pc += 2;
OP_FLAGS(pc) |= OF_DELAY_OP;
+ ADD_TO_ARRAY(branch_target_pc, branch_target_count, pc + offs + 2,);
break;
}
if ((op & 0xf000) == 0) {
op &= 0xff;
- if (op == 0x23 || op == 0x03 || op == 0x0b) { // BRAF, BSRF, RTS
+ if (op == 0x1b) // SLEEP
+ break;
+ if (op == 0x23 || op == 0x03 || op == 0x0b || op == 0x2b) { // BRAF, BSRF, RTS, RTE
pc += 2;
OP_FLAGS(pc) |= OF_DELAY_OP;
break;
signed int offs = ((signed int)(op << 24) >> 23);
if (op & 0x0400)
OP_FLAGS(pc + 2) |= OF_DELAY_OP;
- branch_target_pc[branch_target_count++] = pc + offs + 4;
- if (branch_target_count == MAX_LOCAL_BRANCHES) {
- printf("warning: branch target overflow\n");
- // will only spawn additional blocks
- break;
- }
+ ADD_TO_ARRAY(branch_target_pc, branch_target_count, pc + offs + 4, break);
}
+ if ((op & 0xff00) == 0xc300) // TRAPA
+ break;
}
end_pc = pc;
// -------------------------------------------------
// 2nd pass: actual compilation
+ out_pc = 0;
pc = base_pc;
for (cycles = 0; pc <= end_pc || drcf.delayed_op; )
{
*drcblk = (blkid << 1) | 1;
}
- dbg(1, "=== %csh2 subblock #%d,%d %08x -> %p", sh2->is_slave ? 's' : 'm',
+ dbg(1, "-- %csh2 subblock #%d,%d %08x -> %p", sh2->is_slave ? 's' : 'm',
tcache_id, blkid, pc, tcache_ptr);
+
+ // since we made a block entry, link any other blocks that jump to current pc
+ dr_link_blocks(tcache_ptr, pc, tcache_id);
}
branch_target_ptr[i] = tcache_ptr;
emith_move_r_imm(tmp3, pc + 2);
emith_add_r_r(tmp, tmp3);
}
+ out_pc = (u32)-1;
cycles++;
goto end_op;
case 0x04: // MOV.B Rm,@(R0,Rn) 0000nnnnmmmm0100
case 0: // RTS 0000000000001011
DELAYED_OP;
emit_move_r_r(SHR_PC, SHR_PR);
+ out_pc = (u32)-1;
cycles++;
break;
case 1: // SLEEP 0000000000011011
- emit_move_r_imm32(SHR_PC, pc - 2);
tmp = rcache_get_reg(SHR_SR, RC_GR_RMW);
emith_clear_msb(tmp, tmp, 20); // clear cycles
+ out_pc = out_pc - 2;
cycles = 1;
goto end_op;
case 2: // RTE 0000000000101011
tmp = rcache_get_reg(SHR_SP, RC_GR_RMW);
emith_add_r_imm(tmp, 4*2);
drcf.test_irq = 1;
+ out_pc = (u32)-1;
cycles += 3;
break;
default:
if (!(op & 0x20))
emit_move_r_imm32(SHR_PR, pc + 2);
emit_move_r_r(SHR_PC, (op >> 8) & 0x0f);
+ out_pc = (u32)-1;
cycles++;
break;
case 1: // TAS.B @Rn 0100nnnn00011011
DELAYED_OP;
cycles--;
// fallthrough
- case 0x0900: // BT label 10001001dddddddd
- case 0x0b00: { // BF label 10001011dddddddd
- // jmp_cond ~ cond when guest doesn't jump
- int jmp_cond = (op & 0x0200) ? DCOND_NE : DCOND_EQ;
- int insn_cond = (op & 0x0200) ? DCOND_EQ : DCOND_NE;
- signed int offs = ((signed int)(op << 24) >> 23);
- sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
- if (find_in_array(branch_target_pc, branch_target_count, pc + offs + 2) >= 0) {
- branch_patch_pc[branch_patch_count] = pc + offs + 2;
- branch_patch_cond = insn_cond;
- goto end_op;
- }
-
- // can't resolve branch, cause end of block
- tmp = rcache_get_reg(SHR_PC, RC_GR_WRITE);
- emith_move_r_imm(tmp, pc + (drcf.delayed_op ? 2 : 0));
- emith_tst_r_imm(sr, T);
- EMITH_SJMP_START(jmp_cond);
- if (!drcf.delayed_op)
- offs += 2;
- if (offs < 0) {
- emith_sub_r_imm_c(insn_cond, tmp, -offs);
- } else
- emith_add_r_imm_c(insn_cond, tmp, offs);
- EMITH_SJMP_END(jmp_cond);
+ case 0x0900: // BT label 10001001dddddddd
+ case 0x0b00: // BF label 10001011dddddddd
+ // will handle conditional branches later
+ pending_branch_cond = (op & 0x0200) ? DCOND_EQ : DCOND_NE;
+ i = ((signed int)(op << 24) >> 23);
+ pending_branch_pc = pc + i + 2;
cycles += 2;
- if (!drcf.delayed_op)
- goto end_block_btf;
goto end_op;
- }}
+ }
goto default_;
/////////////////////////////////////////////
case 0x0a:
// BRA label 1010dddddddddddd
DELAYED_OP;
- do_bra:
+ sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
tmp = ((signed int)(op << 20) >> 19);
- emit_move_r_imm32(SHR_PC, pc + tmp + 2);
+ out_pc = pc + tmp + 2;
+ if (tmp == (u32)-4)
+ emith_clear_msb(sr, sr, 20); // burn cycles
cycles++;
break;
// BSR label 1011dddddddddddd
DELAYED_OP;
emit_move_r_imm32(SHR_PR, pc + 2);
- goto do_bra;
+ tmp = ((signed int)(op << 20) >> 19);
+ out_pc = pc + tmp + 2;
+ cycles++;
+ break;
/////////////////////////////////////////////
case 0x0c:
tmp2 = rcache_get_reg(SHR_PC, RC_GR_WRITE);
emith_move_r_r(tmp2, tmp);
rcache_free_tmp(tmp);
+ out_pc = (u32)-1;
cycles += 7;
- goto end_block_btf;
+ goto end_op;
case 0x0700: // MOVA @(disp,PC),R0 11000111dddddddd
emit_move_r_imm32(SHR_R0, (pc + (op & 0xff) * 4 + 2) & ~3);
goto end_op;
}
end_op:
- // block-local conditional branch handling (with/without delay)
- if (branch_patch_cond != -1 && drcf.delayed_op != 2) {
+ // conditional branch handling (with/without delay)
+ if (pending_branch_cond != -1 && drcf.delayed_op != 2)
+ {
+ u32 target_pc = pending_branch_pc;
+ void *target;
+
sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
// handle cycles
FLUSH_CYCLES(sr);
rcache_clean();
-
if (drcf.use_saved_t)
emith_tst_r_imm(sr, T_save);
else
emith_tst_r_imm(sr, T);
- branch_patch_ptr[branch_patch_count] = tcache_ptr;
- emith_jump_patchable(branch_patch_cond);
- drcf.use_saved_t = 0;
- branch_patch_cond = -1;
- branch_patch_count++;
- drcf.delayed_op = 0; // XXX: delayed_op ends block, so must override
- if (branch_patch_count == MAX_LOCAL_BRANCHES) {
- printf("too many local branches\n");
- break;
+ if (find_in_array(branch_target_pc, branch_target_count, target_pc) >= 0) {
+ // local branch
+ // XXX: jumps back can be linked already
+ branch_patch_pc[branch_patch_count] = target_pc;
+ branch_patch_ptr[branch_patch_count] = tcache_ptr;
+ emith_jump_cond_patchable(pending_branch_cond, tcache_ptr);
+
+ branch_patch_count++;
+ if (branch_patch_count == MAX_LOCAL_BRANCHES) {
+ printf("warning: too many local branches\n");
+ break;
+ }
+ }
+ else {
+ // can't resolve branch locally, make a block exit
+ emit_move_r_imm32(SHR_PC, target_pc);
+ rcache_clean();
+
+ target = dr_prepare_ext_branch(target_pc, sh2, tcache_id);
+ if (target == NULL)
+ return NULL;
+ emith_jump_cond_patchable(pending_branch_cond, target);
}
+
+ drcf.use_saved_t = 0;
+ pending_branch_cond = -1;
}
+
// test irq?
// XXX: delay slots..
if (drcf.test_irq && drcf.delayed_op != 2) {
emith_call(sh2_drc_test_irq);
drcf.test_irq = 0;
}
- if (drcf.delayed_op == 1)
- break;
do_host_disasm(tcache_id);
- }
-
- // delayed_op means some kind of branch - PC already handled
- if (!drcf.delayed_op)
- emit_move_r_imm32(SHR_PC, pc);
-end_block_btf:
- this_block->end_addr = pc;
+ if (out_pc != 0 && drcf.delayed_op != 2)
+ break;
+ }
tmp = rcache_get_reg(SHR_SR, RC_GR_RMW);
FLUSH_CYCLES(tmp);
rcache_flush();
- emith_jump(sh2_drc_dispatcher);
+
+ if (out_pc == (u32)-1) {
+ // indirect jump -> back to dispatcher
+ emith_jump(sh2_drc_dispatcher);
+ } else {
+ void *target;
+ if (out_pc == 0)
+ out_pc = pc;
+ emit_move_r_imm32(SHR_PC, out_pc);
+ rcache_flush();
+
+ target = dr_prepare_ext_branch(out_pc, sh2, tcache_id);
+ if (target == NULL)
+ return NULL;
+ emith_jump_patchable(target);
+ }
// link local branches
for (i = 0; i < branch_patch_count; i++) {
void *target;
int t;
- //printf("patch %08x %p\n", branch_patch_pc[i], branch_patch_ptr[i]);
t = find_in_array(branch_target_pc, branch_target_count, branch_patch_pc[i]);
- if (branch_target_ptr[t] != NULL)
- target = branch_target_ptr[t];
- else {
- // flush pc and go back to dispatcher (for now)
+ target = branch_target_ptr[t];
+ if (target == NULL) {
+ // flush pc and go back to dispatcher (should no longer happen)
printf("stray branch to %08x %p\n", branch_patch_pc[i], tcache_ptr);
target = tcache_ptr;
emit_move_r_imm32(SHR_PC, branch_patch_pc[i]);
emith_jump_patch(branch_patch_ptr[i], target);
}
+ this_block->end_addr = pc;
+
// mark memory blocks as containing compiled code
if (tcache_id != 0) {
// data array, BIOS
#endif
return block_entry;
-/*
-unimplemented:
- // last op
- do_host_disasm(tcache_id);
- exit(1);
-*/
}
static void sh2_generate_utils(void)
block_desc *bd = btab + id;
// FIXME: skip subblocks; do both directions
+ // FIXME: collect all branches
dbg(1, " killing block %08x", bd->addr);
bd->addr = bd->end_addr = 0;
int sh2_drc_init(SH2 *sh2)
{
- if (block_tables[0] == NULL) {
- int i, cnt;
-
- drc_cmn_init();
+ int i;
- cnt = block_max_counts[0] + block_max_counts[1] + block_max_counts[2];
- block_tables[0] = calloc(cnt, sizeof(*block_tables[0]));
- if (block_tables[0] == NULL)
- return -1;
+ if (block_tables[0] == NULL)
+ {
+ for (i = 0; i < TCACHE_BUFFERS; i++) {
+ block_tables[i] = calloc(block_max_counts[i], sizeof(*block_tables[0]));
+ if (block_tables[i] == NULL)
+ goto fail;
+ // max 2 block links (exits) per block
+ block_links[i] = calloc(block_max_counts[i] * 2, sizeof(*block_links[0]));
+ if (block_links[i] == NULL)
+ goto fail;
+ }
+ memset(block_counts, 0, sizeof(block_counts));
+ memset(block_link_counts, 0, sizeof(block_link_counts));
+ drc_cmn_init();
tcache_ptr = tcache;
sh2_generate_utils();
#ifdef ARM
cache_flush_d_inval_i(tcache, tcache_ptr);
#endif
- memset(block_counts, 0, sizeof(block_counts));
tcache_bases[0] = tcache_ptrs[0] = tcache_ptr;
-
- for (i = 1; i < ARRAY_SIZE(block_tables); i++) {
- block_tables[i] = block_tables[i - 1] + block_max_counts[i - 1];
+ for (i = 1; i < ARRAY_SIZE(tcache_bases); i++)
tcache_bases[i] = tcache_ptrs[i] = tcache_bases[i - 1] + tcache_sizes[i - 1];
- }
// tmp
PicoOpt |= POPT_DIS_VDP_FIFO;
if (hash_table == NULL) {
hash_table = calloc(sizeof(hash_table[0]), MAX_HASH_ENTRIES);
if (hash_table == NULL)
- return -1;
+ goto fail;
}
return 0;
+
+fail:
+ sh2_drc_finish(sh2);
+ return -1;
}
void sh2_drc_finish(SH2 *sh2)
{
+ int i;
+
if (block_tables[0] != NULL) {
block_stats();
- free(block_tables[0]);
- memset(block_tables, 0, sizeof(block_tables));
+
+ for (i = 0; i < TCACHE_BUFFERS; i++) {
+#if (DRC_DEBUG & 2)
+ printf("~~~ tcache %d\n", i);
+ tcache_dsm_ptrs[i] = tcache_bases[i];
+ tcache_ptr = tcache_ptrs[i];
+ do_host_disasm(i);
+#endif
+
+ if (block_tables[i] != NULL)
+ free(block_tables[i]);
+ block_tables[i] = NULL;
+ if (block_links[i] == NULL)
+ free(block_links[i]);
+ block_links[i] = NULL;
+ }
drc_cmn_cleanup();
}
notaz.gp2x.de / picodrive.git / commitdiff