More cases to test17, and added -NAN generation to divpd ([DYNAREC] too)

src/dynarec/arm64/dynarec_arm64_660f.c

@@ -769,8 +769,21 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("DIVPD Gx, Ex");
             nextop = F8;
             GETEX(q0, 0);
-            GETGX(v0);
-            VFDIVQD(v0, v0, q0);
+            GETGX(q1);
+            if(!box64_dynarec_fastnan) {
+                v0 = fpu_get_scratch(dyn);
+                v1 = fpu_get_scratch(dyn);
+                // check if any input value was NAN
+                VFMAXQD(v0, q0, q1);    // propagate NAN
+                FCMEQQD(v0, v0, v0);    // 0 if NAN, 1 if not NAN
+            }
+            VFDIVQD(q1, q1, q0);
+            if(!box64_dynarec_fastnan) {
+                FCMEQQD(v1, q1, q1);    // 0 => out is NAN
+                VBICQ(v1, v0, v1);      // forget it in any input was a NAN already
+                VSHLQ_64(v1, v1, 63);   // only keep the sign bit
+                VORRQ(q1, q1, v1);      // NAN -> -NAN
+            }
             break;
         case 0x5F:
             INST_NAME("MAXPD Gx, Ex");

src/emu/x64run660f.c

@@ -55,6 +55,9 @@ int Run660F(x64emu_t *emu, rex_t rex)
     uint32_t tmp32u;
     uint64_t tmp64u;
     float tmpf;
+    #ifndef NOALIGN
+    int is_nan;
+    #endif
     reg64_t *oped, *opgd;
     sse_regs_t *opex, *opgx, eax1, *opex2;
     mmx87_regs_t *opem, *opgm;
@@ -912,8 +915,16 @@ int Run660F(x64emu_t *emu, rex_t rex)
         nextop = F8;
         GETEX(0);
         GETGX;
-        GX->d[0] /= EX->d[0];
-        GX->d[1] /= EX->d[1];
+        for (int i=0; i<2; ++i) {
+            #ifndef NOALIGN
+            is_nan = isnan(GX->d[i]) || isnan(EX->d[i]);
+            #endif
+            GX->d[i] /= EX->d[i];
+            #ifndef NOALIGN
+            if(!is_nan && isnan(GX->d[i]))
+                GX->d[i] = -NAN;
+            #endif
+        }
         break;
     case 0x5F:                      /* MAXPD Gx, Ex */
         nextop = F8;

tests/ref17.txt

@@ -234,3 +234,27 @@ mulpd(1 2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -0
 mulpd(0 -2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 0 
 mulpd(inf -inf , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 0xfff8000000000000 
 mulpd(0x7ff8000000000000 -0 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 0 
+subpd(1 2 , 0 -2 ) = 1 4 
+subpd(0 -2 , inf -inf ) = -inf inf 
+subpd(1 2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 2 
+subpd(0 -2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -2 
+subpd(inf -inf , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -inf 
+subpd(0x7ff8000000000000 -0 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 0 
+minpd(1 2 , 0 -2 ) = 0 -2 
+minpd(0 -2 , inf -inf ) = 0 -inf 
+minpd(1 2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -0 
+minpd(0 -2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -2 
+minpd(inf -inf , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -inf 
+minpd(0x7ff8000000000000 -0 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -0 
+divpd(1 2 , 0 -2 ) = inf -1 
+divpd(0 -2 , inf -inf ) = 0 0 
+divpd(1 2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -inf 
+divpd(0 -2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 inf 
+divpd(inf -inf , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 inf 
+divpd(0x7ff8000000000000 -0 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 0xfff8000000000000 
+maxpd(1 2 , 0 -2 ) = 1 2 
+maxpd(0 -2 , inf -inf ) = inf -2 
+maxpd(1 2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 2 
+maxpd(0 -2 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -0 
+maxpd(inf -inf , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -0 
+maxpd(0x7ff8000000000000 -0 , 0x7ff8000000000000 -0 ) = 0x7ff8000000000000 -0 

tests/test17.c

@@ -347,6 +347,10 @@ printf(N " %g, %g => %g\n", b, a, *(float*)&r);
  MULITGO2pd(xor, xorpd)
  MULITGO2pd(add, addpd)
  MULITGO2pd(mul, mulpd)
+ MULITGO2pd(sub, subpd)
+ MULITGO2pd(min, minpd)
+ MULITGO2pd(div, divpd)
+ MULITGO2pd(max, maxpd)
 
  return 0;
 }