[ARM64_DYNAREC] Reworked, again, strongmem emulation

4 files changed, 18 insertions, 8 deletions

diff --git a/src/dynarec/arm64/dynarec_arm64_00.c b/src/dynarec/arm64/dynarec_arm64_00.c
index 6a5c5b7e..62604eec 100644
--- a/src/dynarec/arm64/dynarec_arm64_00.c
+++ b/src/dynarec/arm64/dynarec_arm64_00.c
@@ -2160,10 +2160,12 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         call_n(dyn, ninst, *(void**)(addr+8), tmp);
                         addr+=8+8;
                     } else {
+                        WILLWRITE2();
                         GETIP(ip+1); // read the 0xCC
                         STORE_XEMU_CALL(xRIP);
                         ADDx_U12(x1, xEmu, (uint32_t)offsetof(x64emu_t, ip)); // setup addr as &emu->ip
                         CALL_S(x64Int3, -1);
+                        SMWRITE2();
                         LOAD_XEMU_CALL(xRIP);
                         addr+=8+8;
                         TABLE64(x3, addr); // expected return address
@@ -2834,6 +2836,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     } else {
                         TABLE64(x2, addr);
                     }
+                    WILLWRITE2();
                     PUSH1(x2);
                     MESSAGE(LOG_DUMP, "Native Call to %s (retn=%d)\n", GetNativeName(GetNativeFnc(dyn->insts[ninst].natcall-1)), dyn->insts[ninst].retn);
                     SKIPTEST(x1);    // disable test as this hack dos 2 instructions for 1
@@ -2850,6 +2853,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     if((box64_log<2 && !cycle_log) && dyn->insts[ninst].natcall && tmp) {
                         //GETIP(ip+3+8+8); // read the 0xCC
                         call_n(dyn, ninst, *(void**)(dyn->insts[ninst].natcall+2+8), tmp);
+                        SMWRITE2();
                         POP1(xRIP);   // pop the return address
                         dyn->last_ip = addr;
                     } else {
@@ -2857,6 +2861,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         STORE_XEMU_CALL(xRIP);
                         ADDx_U12(x1, xEmu, (uint32_t)offsetof(x64emu_t, ip)); // setup addr as &emu->ip
                         CALL_S(x64Int3, -1);
+                        SMWRITE2();
                         LOAD_XEMU_CALL(xRIP);
                         TABLE64(x3, dyn->insts[ninst].natcall);
                         ADDx_U12(x3, x3, 2+8+8);
diff --git a/src/dynarec/arm64/dynarec_arm64_helper.h b/src/dynarec/arm64/dynarec_arm64_helper.h
index 99894327..918b8b2a 100644
--- a/src/dynarec/arm64/dynarec_arm64_helper.h
+++ b/src/dynarec/arm64/dynarec_arm64_helper.h
@@ -33,15 +33,19 @@
 #define PKip(a)   *(uint8_t*)(ip+a)
 
 // Strong mem emulation helpers
-#define SMREAD_MIN  2
+#define SMREAD_MIN  3
+#define SMWRITE2_MIN 1
 #define SMFIRST_MIN 1
+#define SMSEQ_MIN 2
+#define SMSEQ_MAX 3
 #if STEP == 0
 // pass 0 will store is opcode write memory
 #define SMWRITE()   dyn->insts[ninst].will_write = 1; dyn->smwrite = 1
 #define SMREAD()
 #define SMREADLOCK(lock)
 #define SMMIGHTREAD()
-#define SMWRITE2()   if(box64_dynarec_strongmem>SMREAD_MIN) {SMWRITE();}
+#define WILLWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {WILLWRITE();}
+#define SMWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {SMWRITE();}
 #define SMWRITELOCK(lock)   SMWRITE()
 #define WILLWRITELOCK(lock)
 #define WILLWRITE()
@@ -60,11 +64,12 @@
 // Opcode might read (depend on nextop)
 #define SMMIGHTREAD()   if(!MODREG) {SMREAD();}
 // Opcode has wrote
-#define SMWRITE()   if((box64_dynarec_strongmem>=SMFIRST_MIN) && dyn->smwrite==0) {SMDMB();} dyn->smwrite=1
+#define SMWRITE()   if((box64_dynarec_strongmem>=SMFIRST_MIN) && dyn->smwrite==0) {SMDMB();} if(box64_dynarec_strongmem>SMSEQ_MIN) {if(++dyn->smwrite>=SMSEQ_MAX) {SMDMB(); dyn->smwrite=1;}} else dyn->smwrite=1
 // Opcode has wrote (strongmem>1 only)
-#define SMWRITE2()   if(box64_dynarec_strongmem>SMREAD_MIN) {SMWRITE();}
+#define WILLWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {WILLWRITE();}
+#define SMWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {SMWRITE();}
 // Opcode has wrote with option forced lock
-#define SMWRITELOCK(lock)   if(lock) {SMDMB();} else {SMWRITE();}
+#define SMWRITELOCK(lock)   if(lock) {SMDMB(); dyn->smwrite=1;} else {SMWRITE();}
 // Opcode has wrote with option forced lock
 #define WILLWRITELOCK(lock)   if(lock) {DMB_ISH();} else {WILLWRITE();}
 // Opcode might have wrote (depend on nextop)
diff --git a/src/main.c b/src/main.c
index 9ed59359..31075af4 100644
--- a/src/main.c
+++ b/src/main.c
@@ -608,11 +608,11 @@ void LoadLogEnv()
     p = getenv("BOX64_DYNAREC_STRONGMEM");
     if(p) {
         if(strlen(p)==1) {
-            if(p[0]>='0' && p[0]<='3')
+            if(p[0]>='0' && p[0]<='4')
                 box64_dynarec_strongmem = p[0]-'0';
         }
         if(box64_dynarec_strongmem)
-            printf_log(LOG_INFO, "Dynarec will try to emulate a strong memory model%s\n", (box64_dynarec_strongmem==1)?" with limited performance loss":((box64_dynarec_strongmem==3)?" with more performance loss":""));
+            printf_log(LOG_INFO, "Dynarec will try to emulate a strong memory model%s\n", (box64_dynarec_strongmem==1)?" with limited performance loss":((box64_dynarec_strongmem>1)?" with more performance loss":""));
     }
     p = getenv("BOX64_DYNAREC_X87DOUBLE");
     if(p) {
diff --git a/src/tools/rcfile.c b/src/tools/rcfile.c
index e7dcd939..b628fb6b 100644
--- a/src/tools/rcfile.c
+++ b/src/tools/rcfile.c
@@ -145,7 +145,7 @@ ENTRYINT(BOX64_DYNAREC_DUMP, box64_dynarec_dump, 0, 2, 2)           \
 ENTRYINT(BOX64_DYNAREC_LOG, box64_dynarec_log, 0, 3, 2)             \
 ENTRYINT(BOX64_DYNAREC_BIGBLOCK, box64_dynarec_bigblock, 0, 3, 2)   \
 ENTRYSTRING_(BOX64_DYNAREC_FORWARD, box64_dynarec_forward)          \
-ENTRYINT(BOX64_DYNAREC_STRONGMEM, box64_dynarec_strongmem, 0, 3, 2) \
+ENTRYINT(BOX64_DYNAREC_STRONGMEM, box64_dynarec_strongmem, 0, 4, 3) \
 ENTRYBOOL(BOX64_DYNAREC_X87DOUBLE, box64_dynarec_x87double)         \
 ENTRYBOOL(BOX64_DYNAREC_DIV0, box64_dynarec_div0)                   \
 ENTRYBOOL(BOX64_DYNAREC_FASTNAN, box64_dynarec_fastnan)             \