Skip to content

[LA64_DYNAREC] Refacfored the handling of deferred flags #3288

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ptitSeb merged 1 commit into from
Dec 19, 2025
Merged

[LA64_DYNAREC] Refacfored the handling of deferred flags #3288

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion src/dynarec/dynarec_native_functions.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -720,7 +720,7 @@ static int flagsCacheNeedsTransform(dynarec_native_t* dyn, int ninst) {
int jmp = dyn->insts[ninst].x64.jmp_insts;
if(jmp<0)
return 0;
#ifdef ARM64
#if defined(ARM64) || defined(LA64)
// df_none is now a defered information
if(dyn->insts[ninst].f_exit==dyn->insts[jmp].f_entry)
return 0;
Expand Down
10 changes: 5 additions & 5 deletions src/dynarec/dynarec_native_pass.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
int rep = 0; // 0 none, 1=F2 prefix, 2=F3 prefix
int need_epilog = 1;
// Clean up (because there are multiple passes)
#ifdef ARM64
#if defined(ARM64) || defined(LA64)
dyn->f = status_unk;
#else
dyn->f.pending = 0;
Expand Down Expand Up @@ -111,7 +111,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
dyn->last_ip = 0;
if(reset_n==-2) {
MESSAGE(LOG_DEBUG, "Reset Caches to zero\n");
#ifdef ARM64
#if defined(ARM64) || defined(LA64)
dyn->f = status_unk;
#else
dyn->f.dfnone = 0;
Expand All @@ -128,7 +128,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
}
if(dyn->insts[ninst].x64.barrier&BARRIER_FLAGS) {
MESSAGE(LOG_DEBUG, "Apply Barrier Flags\n");
#ifdef ARM64
#if defined(ARM64) || defined(LA64)
dyn->f = status_unk;
#else
dyn->f.dfnone = 0;
Expand Down Expand Up @@ -262,7 +262,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
#endif
}
if(dyn->insts[next].x64.barrier&BARRIER_FLAGS) {
#ifdef ARM64
#if defined(ARM64) || defined(LA64)
dyn->f = status_unk;
#else
dyn->f.pending = 0;
Expand All @@ -288,7 +288,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
// we use the 1st predecessor here
if((ninst+1)<dyn->size && !dyn->insts[ninst+1].x64.alive) {
// reset fpu value...
#ifdef ARM64
#if defined(ARM64) || defined(LA64)
dyn->f = status_unk;
#else
dyn->f.dfnone = 0;
Expand Down
2 changes: 1 addition & 1 deletion src/dynarec/la64/dynarec_la64_00.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2679,7 +2679,7 @@ uintptr_t dynarec64_00(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
break;
case 0xCF:
INST_NAME("IRET");
SETFLAGS(X_ALL, SF_SET_NODF, NAT_FLAGS_NOFUSION); // Not a hack, EFLAGS are restored
SETFLAGS(X_ALL, SF_SET_DF, NAT_FLAGS_NOFUSION); // Not a hack, EFLAGS are restored
BARRIER(BARRIER_FLOAT);
iret_to_epilog(dyn, ip, ninst, rex.w);
*need_epilog = 0;
Expand Down
3 changes: 1 addition & 2 deletions src/dynarec/la64/dynarec_la64_0f.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -417,8 +417,7 @@ uintptr_t dynarec64_0F(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
} else {
INST_NAME("UCOMISS Gx, Ex");
}
SETFLAGS(X_ALL, SF_SET, NAT_FLAGS_NOFUSION);
SET_DFNONE();
SETFLAGS(X_ALL, SF_SET_DF, NAT_FLAGS_NOFUSION);
nextop = F8;
GETGX(d0, 0);
GETEXSS(v0, 0, 0);
Expand Down
9 changes: 6 additions & 3 deletions src/dynarec/la64/dynarec_la64_functions.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -599,17 +599,20 @@ static const char* Ft[] = { "fa0", "fa1", "fa2", "fa3", "fa4", "fa5", "fa6", "fa
static const char* Vt[] = { "vra0", "vra1", "vra2", "vra3", "vra4", "vra5", "vra6", "vra7", "vrt0", "vrt1", "vrt2", "vrt3", "vrt4", "vrt5", "vrt6", "vrt7", "vrt8", "vrt9", "vrt10", "vrt11", "vrt12", "vrt13", "vrt14", "vrt15", "vrs0", "vrs1", "vrs2", "vrs3", "vrs4", "vrs5", "vrs6", "vrs7" };
static const char* XVt[] = { "xvra0", "xvra1", "xvra2", "xvra3", "xvra4", "xvra5", "xvra6", "xvra7", "xvrt0", "xvrt1", "xvrt2", "xvrt3", "xvrt4", "xvrt5", "xvrt6", "xvrt7", "xvrt8", "xvrt9", "xvrt10", "xvrt11", "xvrt12", "xvrt13", "xvrt14", "xvrt15", "xvrs0", "xvrs1", "xvrs2", "xvrs3", "xvrs4", "xvrs5", "xvrs6", "xvrs7" };

static const char* df_status[] = { "unknown", "set", "none_pending", "none" };

void printf_x64_instruction(dynarec_native_t* dyn, zydis_dec_t* dec, instruction_x64_t* inst, const char* name);
void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t rex)
{
if (!dyn->need_dump && !BOX64ENV(dynarec_gdbjit) && !BOX64ENV(dynarec_perf_map)) return;

static char buf[4096];
int length = sprintf(buf, "barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X, fuse=%d/%d, sm=%d(%d/%d)",
int length = sprintf(buf, "barrier=%d state=%d/%s(%s->%s), %s=%X/%X, use=%X, need=%X/%X, fuse=%d/%d, sm=%d(%d/%d)",
dyn->insts[ninst].x64.barrier,
dyn->insts[ninst].x64.state_flags,
dyn->f.pending,
dyn->f.dfnone,
df_status[dyn->f],
df_status[dyn->insts[ninst].f_entry],
df_status[dyn->insts[ninst].f_exit],
dyn->insts[ninst].x64.may_set ? "may" : "set",
dyn->insts[ninst].x64.set_flags,
dyn->insts[ninst].x64.gen_flags,
Expand Down
46 changes: 33 additions & 13 deletions src/dynarec/la64/dynarec_la64_helper.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -349,6 +349,7 @@ void jump_to_epilog(dynarec_la64_t* dyn, uintptr_t ip, int reg, int ninst)
}
TABLE64C(x2, const_epilog);
SMEND();
CHECK_DFNONE(0);
BR(x2);
}

Expand All @@ -368,6 +369,7 @@ void jump_to_epilog_fast(dynarec_la64_t* dyn, uintptr_t ip, int reg, int ninst)
}
TABLE64C(x2, const_epilog_fast);
SMEND();
CHECK_DFNONE(0);
BR(x2);
}

Expand Down Expand Up @@ -407,6 +409,7 @@ void jump_to_next(dynarec_la64_t* dyn, uintptr_t ip, int reg, int ninst, int is3
if (is32bits)
ip &= 0xffffffffLL;

CHECK_DFNONE(0);
int dest;
if (reg) {
if (reg != xRIP) {
Expand Down Expand Up @@ -441,6 +444,7 @@ void ret_to_epilog(dynarec_la64_t* dyn, uintptr_t ip, int ninst, rex_t rex)
MAYUSE(dyn);
MAYUSE(ninst);
MESSAGE(LOG_DUMP, "Ret to epilog\n");
CHECK_DFNONE(0);
POP1z(xRIP);
MVz(x1, xRIP);
SMEND();
Expand All @@ -465,6 +469,7 @@ void retn_to_epilog(dynarec_la64_t* dyn, uintptr_t ip, int ninst, rex_t rex, int
MAYUSE(dyn);
MAYUSE(ninst);
MESSAGE(LOG_DUMP, "Retn to epilog\n");
CHECK_DFNONE(0);
POP1z(xRIP);
if (n > 0x7ff) {
MOV64x(x1, n);
Expand Down Expand Up @@ -513,7 +518,7 @@ void iret_to_epilog(dynarec_la64_t* dyn, uintptr_t ip, int ninst, int is64bits)
AND(xFlags, xFlags, x1);
ORI(xFlags, xFlags, 0x2);
SPILL_EFLAGS();
SET_DFNONE();
FORCE_DFNONE();
// POP RSP
if (is64bits) {
POP1(x3); // rsp
Expand All @@ -540,6 +545,7 @@ void iret_to_epilog(dynarec_la64_t* dyn, uintptr_t ip, int ninst, int is64bits)
void call_c(dynarec_la64_t* dyn, int ninst, la64_consts_t fnc, int reg, int ret, int saveflags, int savereg, int arg1, int arg2, int arg3, int arg4, int arg5, int arg6)
{
MAYUSE(fnc);
CHECK_DFNONE(1);
if (savereg == 0)
savereg = x87pc;
if (saveflags) {
Expand Down Expand Up @@ -608,6 +614,7 @@ void call_c(dynarec_la64_t* dyn, int ninst, la64_consts_t fnc, int reg, int ret,
void call_n(dynarec_la64_t* dyn, int ninst, void* fnc, int w)
{
MAYUSE(fnc);
CHECK_DFNONE(1);
fpu_pushcache(dyn, ninst, x3, 1);
ST_D(xRSP, xEmu, offsetof(x64emu_t, regs[_SP]));
ST_D(xRBP, xEmu, offsetof(x64emu_t, regs[_BP]));
Expand Down Expand Up @@ -2269,31 +2276,44 @@ static void flagsCacheTransform(dynarec_la64_t* dyn, int ninst, int s1)
int jmp = dyn->insts[ninst].x64.jmp_insts;
if (jmp < 0)
return;
if (dyn->f.dfnone || ((dyn->insts[jmp].f_exit.dfnone && !dyn->insts[jmp].f_entry.dfnone) && !dyn->insts[jmp].x64.use_flags)) // flags are fully known, nothing we can do more
if (dyn->insts[jmp].f_exit == dyn->insts[jmp].f_entry) // flags will be fully known, nothing we can do more
return;
MESSAGE(LOG_DUMP, "\tFlags fetch ---- ninst=%d -> %d\n", ninst, jmp);
int go = (dyn->insts[jmp].f_entry.dfnone && !dyn->f.dfnone && !dyn->insts[jmp].df_notneeded) ? 1 : 0;
switch (dyn->insts[jmp].f_entry.pending) {
case SF_UNKNOWN:
go = 0;
int go_fetch = 0;
switch (dyn->insts[jmp].f_entry) {
case status_unk:
if (dyn->insts[ninst].f_exit == status_none_pending) {
FORCE_DFNONE();
}
break;
default:
if (go && !(dyn->insts[jmp].x64.need_before & X_PEND) && (dyn->f.pending != SF_UNKNOWN)) {
// just clear df flags
go = 0;
ST_W(xZR, xEmu, offsetof(x64emu_t, df));
case status_set:
if (dyn->insts[ninst].f_exit == status_none_pending) {
FORCE_DFNONE();
}
if (dyn->insts[ninst].f_exit == status_unk)
go_fetch = 1;
break;
case status_none_pending:
if (dyn->insts[ninst].f_exit != status_none)
go_fetch = 1;
break;
case status_none:
if (dyn->insts[ninst].f_exit == status_none_pending) {
FORCE_DFNONE();
} else
go_fetch = 1;
break;
}
if (go) {
if (dyn->f.pending != SF_PENDING) {
if (go_fetch) {
if (dyn->f == status_unk) {
LD_WU(s1, xEmu, offsetof(x64emu_t, df));
j64 = (GETMARKF2) - (dyn->native_size);
BEQZ(s1, j64);
}
CALL_(const_updateflags, -1, 0, 0, 0);
MARKF2;
}
MESSAGE(LOG_DUMP, "\t---- Flags fetch\n");
}

void CacheTransform(dynarec_la64_t* dyn, int ninst, int cacheupd, int s1, int s2, int s3)
Expand Down
106 changes: 44 additions & 62 deletions src/dynarec/la64/dynarec_la64_helper.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -831,36 +831,29 @@
LOAD_REG(R15);

#define FORCE_DFNONE() ST_W(xZR, xEmu, offsetof(x64emu_t, df))
#define CHECK_DFNONE(N) \
do { \
if (dyn->f == status_none_pending) { \
FORCE_DFNONE(); \
if (N) dyn->f = status_none; \
} \
} while (0)

#define SET_DFNONE() \
do { \
if (!dyn->f.dfnone) { \
FORCE_DFNONE(); \
} \
if (!dyn->insts[ninst].x64.may_set) { \
dyn->f.dfnone = 1; \
} \
#define SET_DFNONE() \
do { \
if (!dyn->insts[ninst].x64.may_set && (dyn->f != status_none)) { \
dyn->f = status_none_pending; \
} \
} while (0)

#define SET_DF(S, N) \
if ((N) != d_none) { \
MOV32w(S, (N)); \
ST_W(S, xEmu, offsetof(x64emu_t, df)); \
if (dyn->f.pending == SF_PENDING \
&& dyn->insts[ninst].x64.need_after \
&& !(dyn->insts[ninst].x64.need_after & X_PEND)) { \
CALL_(const_updateflags, -1, 0, 0, 0); \
dyn->f.pending = SF_SET; \
SET_NODF(); \
} \
dyn->f.dfnone = 0; \
} else \
#define SET_DF(S, N) \
if ((N) != d_none) { \
MOV32w(S, (N)); \
ST_W(S, xEmu, offsetof(x64emu_t, df)); \
dyn->f = status_set; \
} else \
SET_DFNONE()

#define SET_NODF() dyn->f.dfnone = 0
#define SET_DFOK() \
dyn->f.dfnone = 1

#define CLEAR_FLAGS_(s) \
MOV64x(s, (1UL << F_AF) | (1UL << F_CF) | (1UL << F_OF) | (1UL << F_ZF) | (1UL << F_SF) | (1UL << F_PF)); \
ANDN(xFlags, xFlags, s);
Expand Down Expand Up @@ -942,18 +935,11 @@
#endif

#ifndef READFLAGS
#define READFLAGS(A) \
if (((A) != X_PEND && dyn->f.pending != SF_SET) \
&& (dyn->f.pending != SF_SET_PENDING)) { \
if (dyn->f.pending != SF_PENDING) { \
LD_WU(x3, xEmu, offsetof(x64emu_t, df)); \
j64 = (GETMARKF) - (dyn->native_size); \
BEQ(x3, xZR, j64); \
} \
CALL_(const_updateflags, -1, 0, 0, 0); \
MARKF; \
dyn->f.pending = SF_SET; \
SET_DFOK(); \
#define READFLAGS(A) \
if ((A) != X_PEND \
&& (dyn->f == status_unk)) { \
CALL_(const_updateflags, -1, 0, 0, 0); \
dyn->f = status_none; \
}
#endif

Expand Down Expand Up @@ -987,32 +973,28 @@
#define NAT_FLAGS_ENABLE_SIGN() dyn->insts[ninst].nat_flags_sign = 1

#ifndef SETFLAGS
#define SETFLAGS(A, B, FUSION) \
if (dyn->f.pending != SF_SET \
&& ((B) & SF_SUB) \
&& (dyn->insts[ninst].x64.gen_flags & (~(A)))) \
READFLAGS(((dyn->insts[ninst].x64.gen_flags & X_PEND) ? X_ALL : dyn->insts[ninst].x64.gen_flags) & (~(A))); \
if (dyn->insts[ninst].x64.gen_flags) switch (B) { \
case SF_SUBSET: \
case SF_SET: dyn->f.pending = SF_SET; break; \
case SF_SET_DF: \
dyn->f.pending = SF_SET; \
dyn->f.dfnone = 1; \
break; \
case SF_SET_NODF: \
dyn->f.pending = SF_SET; \
dyn->f.dfnone = 0; \
break; \
case SF_PENDING: dyn->f.pending = SF_PENDING; break; \
case SF_SUBSET_PENDING: \
case SF_SET_PENDING: \
dyn->f.pending = (dyn->insts[ninst].x64.gen_flags & X_PEND) ? SF_SET_PENDING : SF_SET; \
break; \
} \
else \
dyn->f.pending = SF_SET; \
dyn->insts[ninst].nat_flags_nofusion = (FUSION)
#define SETFLAGS(A, B, FUSION) \
do { \
if (((B) & SF_SUB) \
&& (dyn->insts[ninst].x64.gen_flags & (~(A)))) \
READFLAGS(((dyn->insts[ninst].x64.gen_flags & X_PEND) ? X_ALL : dyn->insts[ninst].x64.gen_flags) & (~(A))); \
if (dyn->insts[ninst].x64.gen_flags) switch (B) { \
case SF_SET_DF: dyn->f = status_set; break; \
case SF_SET_NODF: SET_DFNONE(); break; \
case SF_SUBSET: \
case SF_SUBSET_PENDING: \
case SF_SET: \
case SF_PENDING: \
case SF_SET_PENDING: \
SET_DFNONE(); \
break; \
} \
else \
SET_DFNONE(); \
dyn->insts[ninst].nat_flags_nofusion = (FUSION); \
} while (0)
#endif

#ifndef JUMP
#define JUMP(A, C)
#endif
Expand Down
Loading
Loading