Fixed a regression introduced with box32 that broke Bladur's Gate 3 (and probably a few other)

5 files changed, 15 insertions, 13 deletions

diff --git a/src/dynarec/arm64/dynarec_arm64_0f.c b/src/dynarec/arm64/dynarec_arm64_0f.c
index 225161ce..5a752519 100644
--- a/src/dynarec/arm64/dynarec_arm64_0f.c
+++ b/src/dynarec/arm64/dynarec_arm64_0f.c
@@ -1781,7 +1781,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         fpu_purgecache(dyn, ninst, 0, x1, x2, x3);

                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

                         if(ed!=x1) {MOVx_REG(x1, ed);}

-                        CALL(rex.w?((void*)fpu_fxsave64):((void*)fpu_fxsave32), -1);

+                        CALL(rex.is32bits?((void*)fpu_fxsave32):((void*)fpu_fxsave64), -1);

                         break;

                     case 1:

                         INST_NAME("FXRSTOR Ed");

@@ -1789,7 +1789,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         fpu_purgecache(dyn, ninst, 0, x1, x2, x3);

                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);

                         if(ed!=x1) {MOVx_REG(x1, ed);}

-                        CALL(rex.w?((void*)fpu_fxrstor64):((void*)fpu_fxrstor32), -1);

+                        CALL(rex.is32bits?((void*)fpu_fxrstor32):((void*)fpu_fxrstor64), -1);

                         break;

                     case 2:

                         INST_NAME("LDMXCSR Md");

diff --git a/src/dynarec/la64/dynarec_la64_0f.c b/src/dynarec/la64/dynarec_la64_0f.c
index 46fea7ed..eadd17aa 100644
--- a/src/dynarec/la64/dynarec_la64_0f.c
+++ b/src/dynarec/la64/dynarec_la64_0f.c
@@ -764,7 +764,7 @@ uintptr_t dynarec64_0F(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         } else {
                             addr = geted(dyn, addr, ninst, nextop, &ed, x1, x3, &fixedaddress, rex, NULL, 0, 0);
                             if (ed != x1) { MV(x1, ed); }
-                            CALL(rex.w ? ((void*)fpu_fxsave64) : ((void*)fpu_fxsave32), -1);
+                            CALL(rex.is32bits ? ((void*)fpu_fxsave32) : ((void*)fpu_fxsave64), -1);
                         }
                         break;
                     case 2:
diff --git a/src/dynarec/rv64/dynarec_rv64_0f.c b/src/dynarec/rv64/dynarec_rv64_0f.c
index 434b5a99..a4b02d48 100644
--- a/src/dynarec/rv64/dynarec_rv64_0f.c
+++ b/src/dynarec/rv64/dynarec_rv64_0f.c
@@ -1860,7 +1860,7 @@ uintptr_t dynarec64_0F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, x3, &fixedaddress, rex, NULL, 0, 0);
                         if (ed != x1) { MV(x1, ed); }
-                        CALL(rex.w ? ((void*)fpu_fxsave64) : ((void*)fpu_fxsave32), -1);
+                        CALL(rex.is32bits ? ((void*)fpu_fxsave32) : ((void*)fpu_fxsave64), -1);
                         break;
                     case 1:
                         INST_NAME("FXRSTOR Ed");
@@ -1869,7 +1869,7 @@ uintptr_t dynarec64_0F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                         addr = geted(dyn, addr, ninst, nextop, &ed, x1, x3, &fixedaddress, rex, NULL, 0, 0);
                         if (ed != x1) { MV(x1, ed); }
-                        CALL(rex.w ? ((void*)fpu_fxrstor64) : ((void*)fpu_fxrstor32), -1);
+                        CALL(rex.is32bits ? ((void*)fpu_fxrstor32) : ((void*)fpu_fxrstor64), -1);
                         break;
                     case 2:
                         INST_NAME("LDMXCSR Md");
diff --git a/src/emu/x64run0f.c b/src/emu/x64run0f.c
index 1e54db04..d81a9d12 100644
--- a/src/emu/x64run0f.c
+++ b/src/emu/x64run0f.c
@@ -1242,18 +1242,18 @@ uintptr_t Run0F(x64emu_t *emu, rex_t rex, uintptr_t addr, int *step)
                     #ifdef TEST_INTERPRETER

                     emu->sw.f.F87_TOP = emu->top&7;

                     #else

-                    if(rex.w)

-                        fpu_fxsave64(emu, ED);

-                    else

+                    if(rex.is32bits)

                         fpu_fxsave32(emu, ED);

+                    else

+                        fpu_fxsave64(emu, ED);

                     #endif

                     break;

                 case 1:                 /* FXRSTOR Ed */

                     _GETED(0);

-                    if(rex.w)

-                        fpu_fxrstor64(emu, ED);

-                    else

+                    if(rex.is32bits)

                         fpu_fxrstor32(emu, ED);

+                    else

+                        fpu_fxrstor64(emu, ED);

                     break;

                 case 2:                 /* LDMXCSR Md */

                     GETED(0);

diff --git a/src/emu/x87emu_private.c b/src/emu/x87emu_private.c
index c64f49cb..b5894d97 100644
--- a/src/emu/x87emu_private.c
+++ b/src/emu/x87emu_private.c
@@ -295,6 +295,7 @@ void fpu_savenv(x64emu_t* emu, char* p, int b16)
     // other stuff are not pushed....
 }
 
+// this is the 64bits version (slightly different than the 32bits!)
 typedef struct xsave32_s {
     uint16_t ControlWord;        /* 000 */
     uint16_t StatusWord;         /* 002 */
@@ -311,9 +312,8 @@ typedef struct xsave32_s {
     uint32_t MxCsr_Mask;         /* 01c */
     sse_regs_t FloatRegisters[8];/* 020 */  // fpu/mmx are store in 128bits here
     sse_regs_t XmmRegisters[8];  /* 0a0 */
-    uint8_t  Reserved4[56*4];    /* 1a0 */
+    uint8_t  Reserved4[56*4];    /* 120 */
 } xsave32_t;
-// this is the 64bits version (slightly different than the 32bits!)
 typedef struct xsave64_s {
     uint16_t ControlWord;        /* 000 */
     uint16_t StatusWord;         /* 002 */
@@ -359,6 +359,7 @@ void fpu_fxsave32(x64emu_t* emu, void* ed)
 
 void fpu_fxsave64(x64emu_t* emu, void* ed)
 {
+    // the subtelties of the REX.W are not handled in fxsave64/fxrstor64
     xsave64_t *p = (xsave64_t*)ed;
     // should save flags & all
     int top = emu->top&7;
@@ -409,6 +410,7 @@ void fpu_fxrstor32(x64emu_t* emu, void* ed)
 
 void fpu_fxrstor64(x64emu_t* emu, void* ed)
 {
+    // the subtelties of the REX.W are not handled in fxsave64/fxrstor64
     xsave64_t *p = (xsave64_t*)ed;
     emu->cw.x16 = p->ControlWord;
     emu->sw.x16 = p->StatusWord;