More x87 fixes ([DYNAREC] too)

ptitSeb · May 22, 2024 · 69f0c97 · 69f0c97
1 parent ff702c0
commit 69f0c97
Show file tree

Hide file tree

Showing 2 changed files with 29 additions and 9 deletions.
diff --git a/src/dynarec/dynarec_arm_d9.c b/src/dynarec/dynarec_arm_d9.c
@@ -138,6 +138,31 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     MOV32(x14, 0b100000100000000);
                     B_MARK3(c__);
                 } else {                // not in cache, so check Empty status and load it
+                    // load tag
+                    LDRH_IMM8(x3, xEmu, offsetof(x86emu_t, fpu_tags));
+                    if(i2) {
+                        if(i2<0) {
+                            MOV_REG_LSL_IMM5(x3, x3, -i2*2);
+                        } else {
+                            MOV32(x14, 0xffff);
+                            ORR_REG_LSL_IMM5(x3, x3, x14, 16);
+                            MOV_REG_LSR_IMM5(x3, x3, i2*2);
+                        }
+                    }
+                    TSTS_IMM8(x3, 0b11);
+                    MOVW(x14, 0b100000100000000);   // empty
+                    B_MARK3(cNE);
+                    // check stack  TODO: this test should not be needed, something is wrong with tag handling
+                    LDR_IMM9(x3, xEmu, offsetof(x86emu_t, fpu_stack));
+                    if(i2) {
+                        if(i2<0) {
+                            ADD_IMM8(x3, x3, -i2);
+                        } else {
+                            SUB_IMM8(x3, x3, i2);
+                        }
+                    }
+                    CMPS_IMM8(x3, 0);
+                    B_MARK3(cLE);
                     // x14 will be the actual top
                     LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
                     i2 = -dyn->n.x87stack;
@@ -149,11 +174,6 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         }
                         AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
                     }
-                    // load tag
-                    LDRH_IMM8(x3, xEmu, offsetof(x86emu_t, fpu_tags));
-                    TSTS_IMM8(x3, 0b11);
-                    MOVW_COND(cNE, x14, 0b100000100000000);
-                    B_MARK3(cNE);
                     ADD_REG_LSL_IMM5(x1, xEmu, x14, 3);
                     LDRD_IMM8(x2, x1, offsetof(x86emu_t, x87)); // load r2/r3 with ST0 anyway, for sign extraction
                 }

diff --git a/src/emu/x87emu_private.c b/src/emu/x87emu_private.c
@@ -295,15 +295,15 @@ void fpu_fxsave(x86emu_t* emu, void* ed)
     int top = emu->top&7;
     int stack = 8-top;
     if(top==0)  // check if stack is full or empty, based on tag[0]
-        stack = (emu->fpu_tags&0b11)?8:0;
+        stack = (emu->fpu_tags)?8:0;
     emu->sw.f.F87_TOP = top;
     p->ControlWord = emu->cw.x16;
     p->StatusWord = emu->sw.x16;
     p->MxCsr = emu->mxcsr.x32;
     p->MxCsr_Mask = 0;
     uint8_t tags = 0;
     for (int i=0; i<8; ++i)
-        tags |= ((emu->fpu_tags>>(i*2))&0b11)?0:1;
+        tags |= (((emu->fpu_tags>>(i*2))&0b11)?0:1)<<i;
     p->TagWord = tags;
     p->ErrorOpcode = 0;
     p->ErrorOffset = 0;
@@ -327,12 +327,12 @@ void fpu_fxrstor(x86emu_t* emu, void* ed)
     uint8_t tags = p->TagWord;
     emu->fpu_tags = 0;
     for(int i=0; i<8; ++i)
-        emu->fpu_tags |= (((tags>>(i*2))&0b1)?0:0b11)<<(i*2);
+        emu->fpu_tags |= (((tags>>i)&0b1)?0:0b11)<<(i*2);
     // copy back MMX regs...
     int top = emu->top&7;
     int stack = 8-top;
     if(top==0)  // check if stack is full or empty, based on tag[0]
-        stack = (emu->fpu_tags&0b11)?8:0;
+        stack = (emu->fpu_tags)?8:0;
     for(int i=0; i<8; ++i)
         memcpy((i<stack)?&ST(i):&emu->mmx[i], &p->FloatRegisters[i].q[0], sizeof(mmx87_regs_t));
     // copy SSE regs