P4 style multiplication eflags

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@481 c046a42c-6fe2-441c-8c8c-71466251a162

target-i386/cpu.h

@@ -184,7 +184,10 @@
 enum {
     CC_OP_DYNAMIC, /* must use dynamic code to get cc_op */
     CC_OP_EFLAGS,  /* all cc are explicitely computed, CC_SRC = flags */
-    CC_OP_MUL, /* modify all flags, C, O = (CC_SRC != 0) */
+
+    CC_OP_MULB, /* modify all flags, C, O = (CC_SRC != 0) */
+    CC_OP_MULW,
+    CC_OP_MULL,
 
     CC_OP_ADDB, /* modify all flags, CC_DST = res, CC_SRC = src1 */
     CC_OP_ADDW,

target-i386/op.c

@@ -169,11 +169,16 @@ void OPPROTO op_bswapl_T0(void)
 }
 
 /* multiply/divide */
+
+/* XXX: add eflags optimizations */
+/* XXX: add non P4 style flags */
+
 void OPPROTO op_mulb_AL_T0(void)
 {
     unsigned int res;
     res = (uint8_t)EAX * (uint8_t)T0;
     EAX = (EAX & 0xffff0000) | res;
+    CC_DST = res;
     CC_SRC = (res & 0xff00);
 }
 
@@ -182,6 +187,7 @@ void OPPROTO op_imulb_AL_T0(void)
     int res;
     res = (int8_t)EAX * (int8_t)T0;
     EAX = (EAX & 0xffff0000) | (res & 0xffff);
+    CC_DST = res;
     CC_SRC = (res != (int8_t)res);
 }
 
@@ -191,6 +197,7 @@ void OPPROTO op_mulw_AX_T0(void)
     res = (uint16_t)EAX * (uint16_t)T0;
     EAX = (EAX & 0xffff0000) | (res & 0xffff);
     EDX = (EDX & 0xffff0000) | ((res >> 16) & 0xffff);
+    CC_DST = res;
     CC_SRC = res >> 16;
 }
 
@@ -200,6 +207,7 @@ void OPPROTO op_imulw_AX_T0(void)
     res = (int16_t)EAX * (int16_t)T0;
     EAX = (EAX & 0xffff0000) | (res & 0xffff);
     EDX = (EDX & 0xffff0000) | ((res >> 16) & 0xffff);
+    CC_DST = res;
     CC_SRC = (res != (int16_t)res);
 }
 
@@ -209,6 +217,7 @@ void OPPROTO op_mull_EAX_T0(void)
     res = (uint64_t)((uint32_t)EAX) * (uint64_t)((uint32_t)T0);
     EAX = res;
     EDX = res >> 32;
+    CC_DST = res;
     CC_SRC = res >> 32;
 }
 
@@ -218,6 +227,7 @@ void OPPROTO op_imull_EAX_T0(void)
     res = (int64_t)((int32_t)EAX) * (int64_t)((int32_t)T0);
     EAX = res;
     EDX = res >> 32;
+    CC_DST = res;
     CC_SRC = (res != (int32_t)res);
 }
 
@@ -226,6 +236,7 @@ void OPPROTO op_imulw_T0_T1(void)
     int res;
     res = (int16_t)T0 * (int16_t)T1;
     T0 = res;
+    CC_DST = res;
     CC_SRC = (res != (int16_t)res);
 }
 
@@ -234,6 +245,7 @@ void OPPROTO op_imull_T0_T1(void)
     int64_t res;
     res = (int64_t)((int32_t)T0) * (int64_t)((int32_t)T1);
     T0 = res;
+    CC_DST = res;
     CC_SRC = (res != (int32_t)res);
 }
 
@@ -1293,31 +1305,14 @@ static int compute_c_eflags(void)
     return CC_SRC & CC_C;
 }
 
-static int compute_c_mul(void)
-{
-    int cf;
-    cf = (CC_SRC != 0);
-    return cf;
-}
-
-static int compute_all_mul(void)
-{
-    int cf, pf, af, zf, sf, of;
-    cf = (CC_SRC != 0);
-    pf = 0; /* undefined */
-    af = 0; /* undefined */
-    zf = 0; /* undefined */
-    sf = 0; /* undefined */
-    of = cf << 11;
-    return cf | pf | af | zf | sf | of;
-}
-    
 CCTable cc_table[CC_OP_NB] = {
     [CC_OP_DYNAMIC] = { /* should never happen */ },
 
     [CC_OP_EFLAGS] = { compute_all_eflags, compute_c_eflags },
 
-    [CC_OP_MUL] = { compute_all_mul, compute_c_mul },
+    [CC_OP_MULB] = { compute_all_mulb, compute_c_mull },
+    [CC_OP_MULW] = { compute_all_mulw, compute_c_mull },
+    [CC_OP_MULL] = { compute_all_mull, compute_c_mull },
 
     [CC_OP_ADDB] = { compute_all_addb, compute_c_addb },
     [CC_OP_ADDW] = { compute_all_addw, compute_c_addw  },

target-i386/ops_template.h

@@ -229,6 +229,29 @@ static int glue(compute_all_sar, SUFFIX)(void)
     return cf | pf | af | zf | sf | of;
 }
 
+#if DATA_BITS == 32
+static int glue(compute_c_mul, SUFFIX)(void)
+{
+    int cf;
+    cf = (CC_SRC != 0);
+    return cf;
+}
+#endif
+
+/* NOTE: we compute the flags like the P4. On olders CPUs, only OF and
+   CF are modified and it is slower to do that. */
+static int glue(compute_all_mul, SUFFIX)(void)
+{
+    int cf, pf, af, zf, sf, of;
+    cf = (CC_SRC != 0);
+    pf = parity_table[(uint8_t)CC_DST];
+    af = 0; /* undefined */
+    zf = ((DATA_TYPE)CC_DST == 0) << 6;
+    sf = lshift(CC_DST, 8 - DATA_BITS) & 0x80;
+    of = cf << 11;
+    return cf | pf | af | zf | sf | of;
+}
+
 /* various optimized jumps cases */
 
 void OPPROTO glue(op_jb_sub, SUFFIX)(void)

target-i386/translate.c

@@ -2016,31 +2016,35 @@ static uint8_t *disas_insn(DisasContext *s, uint8_t *pc_start)
             switch(ot) {
             case OT_BYTE:
                 gen_op_mulb_AL_T0();
+                s->cc_op = CC_OP_MULB;
                 break;
             case OT_WORD:
                 gen_op_mulw_AX_T0();
+                s->cc_op = CC_OP_MULW;
                 break;
             default:
             case OT_LONG:
                 gen_op_mull_EAX_T0();
+                s->cc_op = CC_OP_MULL;
                 break;
             }
-            s->cc_op = CC_OP_MUL;
             break;
         case 5: /* imul */
             switch(ot) {
             case OT_BYTE:
                 gen_op_imulb_AL_T0();
+                s->cc_op = CC_OP_MULB;
                 break;
             case OT_WORD:
                 gen_op_imulw_AX_T0();
+                s->cc_op = CC_OP_MULW;
                 break;
             default:
             case OT_LONG:
                 gen_op_imull_EAX_T0();
+                s->cc_op = CC_OP_MULL;
                 break;
             }
-            s->cc_op = CC_OP_MUL;
             break;
         case 6: /* div */
             switch(ot) {
@@ -2235,7 +2239,7 @@ static uint8_t *disas_insn(DisasContext *s, uint8_t *pc_start)
             gen_op_imulw_T0_T1();
         }
         gen_op_mov_reg_T0[ot][reg]();
-        s->cc_op = CC_OP_MUL;
+        s->cc_op = CC_OP_MULB + ot;
         break;
     case 0x1c0:
     case 0x1c1: /* xadd Ev, Gv */