fixed: 100*200=32, only for the ds390 port right now

git-svn-id: https://sdcc.svn.sourceforge.net/svnroot/sdcc/trunk/sdcc@644 4a8a32a2-be11-0410-ad9d-d568d2c75423

diff --git a/src/SDCCicode.c b/src/SDCCicode.c
index 0b74ff45a9344a0228e1b27fae27f0b738fc3dd6..e9f6b3b1401227e4ed564186f2c05802c646694c 100644 (file)
--- a/src/SDCCicode.c
+++ b/src/SDCCicode.c
@@ -1673,8 +1673,20 @@ geniCodeMultiply (operand * left, operand * right, bool ptrSizeCalculation)
       SPEC_SHORT (getSpec (resType)) = 0;
       options.ANSIint = saveOption;
     }
-  else
+  else {
     resType = usualBinaryConversions (&left, &right);
+    if (IS_DS390_PORT) {
+      /* jwk char*char=int
+        Now be can use the 16bit result of "mul a,b" instead of explicit
+        casts and support function calls as with --ansiint
+      */
+      if ((IS_CHAR(letype) || IS_SHORT(letype)) && 
+          (IS_CHAR(retype) || IS_SHORT(retype))) {
+       SPEC_NOUN(getSpec(resType))=V_INT;
+       SPEC_SHORT(getSpec(resType))=0;
+      }
+    }
+  }
 
   /* if the right is a literal & power of 2 */
   /* then make it a left shift              */

diff --git a/src/ds390/gen.c b/src/ds390/gen.c
index 1ec96079f9fdfc61c8c0b8347fb17c73b3172c2f..54117117655d202d78cd858ee309b1c6b1e948bf 100644 (file)
--- a/src/ds390/gen.c
+++ b/src/ds390/gen.c
@@ -965,7 +965,6 @@ aopGet (asmop * aop,
       if (aop->type == AOP_DPTR2)
        {
          genSetDPTR (1);
-
          if (!canClobberACC)
            {
              emitcode ("xch", "a, ap");
@@ -1000,14 +999,12 @@ aopGet (asmop * aop,
       if (aop->type == AOP_DPTR2)
        {
          genSetDPTR (0);
-
          if (!canClobberACC)
            {
              emitcode ("xch", "a, ap");
              return "ap";
            }
        }
-
       return (dname ? "acc" : "a");
 
     case AOP_IMMD:
@@ -3688,19 +3685,44 @@ genMultbits (operand * left,
 
 
 /*-----------------------------------------------------------------*/
-/* genMultOneByte : 8 bit multiplication & division                */
+/* genMultOneByte : 8*8=16 bit multiplication                      */
 /*-----------------------------------------------------------------*/
+void catchMe() {
+  emitcode(";catch","me");
+}
+
 static void
 genMultOneByte (operand * left,
                operand * right,
                operand * result)
 {
   sym_link *opetype = operandType (result);
-  char *l;
   symbol *lbl;
-  int size, offset;
 
-  /* (if two literals, the value is computed before) */
+  if (AOP_SIZE (result)!=2) {
+    // this should never happen
+      fprintf (stderr, "size!=2 in geniCodeMultOneByte\n");
+  }
+
+  if (SPEC_USIGN(opetype) || 
+      // ignore the sign of left and right, what else can we do?
+      (SPEC_USIGN(operandType(left)) && 
+       SPEC_USIGN(operandType(right)))) {
+    // just an unsigned 8*8=16 multiply
+    emitcode ("mov", "b,%s", aopGet (AOP (right), 0, FALSE, FALSE, TRUE));
+    MOVA (aopGet (AOP (left), 0, FALSE, FALSE, TRUE));
+    emitcode ("mul", "ab");
+    aopPut (AOP (result), "a", 0);
+    aopPut (AOP (result), "b", 1);
+    return;
+  }
+
+  // we have to do a signed multiply
+
+#if 0
+  // literals ignored for now
+
+  /* (if two literals: the value is computed before) */
   /* if one literal, literal on the right */
   if (AOP_TYPE (left) == AOP_LIT)
     {
@@ -3708,77 +3730,48 @@ genMultOneByte (operand * left,
       right = left;
       left = t;
     }
+#endif
 
-  size = AOP_SIZE (result);
-  /* signed or unsigned */
-  emitcode ("mov", "b,%s", aopGet (AOP (right), 0, FALSE, FALSE, FALSE));
-  l = aopGet (AOP (left), 0, FALSE, FALSE, TRUE);
-  MOVA (l);
-  emitcode ("mul", "ab");
-  /* if result size = 1, mul signed = mul unsigned */
-  aopPut (AOP (result), "a", 0);
-  if (size > 1)
-    {
-      if (SPEC_USIGN (opetype))
-       {
-         aopPut (AOP (result), "b", 1);
-         if (size > 2)
-           /* for filling the MSBs */
-           emitcode ("clr", "a");
-       }
-      else
-       {
-         emitcode ("mov", "a,b");
-
-         /* adjust the MSB if left or right neg */
-
-         /* if one literal */
-         if (AOP_TYPE (right) == AOP_LIT)
-           {
-             /* AND literal negative */
-             if ((int) floatFromVal (AOP (right)->aopu.aop_lit) < 0)
-               {
-                 /* adjust MSB (c==0 after mul) */
-                 emitcode ("subb", "a,%s", aopGet (AOP (left), 0, FALSE, FALSE, FALSE));
-               }
-           }
-         else
-           {
-             lbl = newiTempLabel (NULL);
-             emitcode ("xch", "a,%s", aopGet (AOP (right), 0, FALSE, FALSE, FALSE));
-             emitcode ("cjne", "a,#0x80,%05d$", (lbl->key + 100));
-             emitcode ("", "%05d$:", (lbl->key + 100));
-             emitcode ("xch", "a,%s", aopGet (AOP (right), 0, FALSE, FALSE, FALSE));
-             lbl = newiTempLabel (NULL);
-             emitcode ("jc", "%05d$", (lbl->key + 100));
-             emitcode ("subb", "a,%s", aopGet (AOP (left), 0, FALSE, FALSE, FALSE));
-             emitcode ("", "%05d$:", (lbl->key + 100));
-           }
+  emitcode ("mov", "ap,#0"); // can't we use some flag bit here?
+  MOVA (aopGet (AOP (right), 0, FALSE, FALSE, TRUE));
+  lbl=newiTempLabel(NULL);
+  emitcode ("jnb", "acc.7,%05d$",  lbl->key+100);
+  // right side is negative, should test for litteral too
+  emitcode ("inc", "ap"); // opRight is negative
+  emitcode ("cpl", "a"); // 8-bit two's complement, this fails for -128
+  emitcode ("inc", "a");
+  emitcode ("", "%05d$:", lbl->key+100);
+  emitcode ("mov", "b,a");
 
-         lbl = newiTempLabel (NULL);
-         emitcode ("xch", "a,%s", aopGet (AOP (left), 0, FALSE, FALSE, FALSE));
-         emitcode ("cjne", "a,#0x80,%05d$", (lbl->key + 100));
-         emitcode ("", "%05d$:", (lbl->key + 100));
-         emitcode ("xch", "a,%s", aopGet (AOP (left), 0, FALSE, FALSE, FALSE));
-         lbl = newiTempLabel (NULL);
-         emitcode ("jc", "%05d$", (lbl->key + 100));
-         emitcode ("subb", "a,%s", aopGet (AOP (right), 0, FALSE, FALSE, FALSE));
-         emitcode ("", "%05d$:", (lbl->key + 100));
+  MOVA (aopGet (AOP (left), 0, FALSE, FALSE, TRUE));
+  lbl=newiTempLabel(NULL);
+  emitcode ("jnb", "acc.7,%05d$", lbl->key+100);
+  // left side is negative
+  emitcode ("inc", "ap"); // opLeft is negative
+  emitcode ("cpl", "a"); // 8-bit two's complement, this fails for -128
+  emitcode ("inc", "a");
+  emitcode ("", "%05d$:", lbl->key+100);
+  emitcode ("mul", "ab");
+    
+  lbl=newiTempLabel(NULL);
+  emitcode ("xch", "a,ap");
+  emitcode ("rrc", "a");
+  emitcode ("xch", "a,ap");
+  emitcode ("jnc", "%05d$", lbl->key+100);
+  // only ONE op was negative, we have to do a 16-bit two's complement
+  emitcode ("setb", "c");
+  emitcode ("cpl", "a");
+  emitcode ("addc", "a,#0");
+  emitcode ("push", "acc"); // lsb
+  emitcode ("mov", "a,b");
+  emitcode ("cpl", "a");
+  emitcode ("addc", "a,#0");
+  emitcode ("mov", "b,a"); // msb
+  emitcode ("pop", "acc"); // lsb
 
-         aopPut (AOP (result), "a", 1);
-         if (size > 2)
-           {
-             /* get the sign */
-             emitcode ("rlc", "a");
-             emitcode ("subb", "a,acc");
-           }
-       }
-      size -= 2;
-      offset = 2;
-      if (size > 0)
-       while (size--)
-         aopPut (AOP (result), "a", offset++);
-    }
+  emitcode ("", "%05d$:", lbl->key+100);
+  aopPut (AOP (result), "a", 0);
+  aopPut (AOP (result), "b", 1);
 }
 
 /*-----------------------------------------------------------------*/