tcg: Convert div2 to TCGOutOpDivRem

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

tcg/aarch64/tcg-target.c.inc

@@ -2168,6 +2168,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rrr = tgen_divs,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu(TCGContext *s, TCGType type,
                       TCGReg a0, TCGReg a1, TCGReg a2)
 {

tcg/arm/tcg-target.c.inc

@@ -1880,6 +1880,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rrr = tgen_divs,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu(TCGContext *s, TCGType type,
                       TCGReg a0, TCGReg a1, TCGReg a2)
 {

tcg/i386/tcg-target.c.inc

@@ -2637,6 +2637,18 @@ static const TCGOutOpBinary outop_divs = {
     .base.static_constraint = C_NotImplemented,
 };
 
+static void tgen_divs2(TCGContext *s, TCGType type,
+                       TCGReg a0, TCGReg a1, TCGReg a4)
+{
+    int rexw = type == TCG_TYPE_I32 ? 0 : P_REXW;
+    tcg_out_modrm(s, OPC_GRP3_Ev + rexw, EXT3_IDIV, a4);
+}
+
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_O2_I3(a, d, 0, 1, r),
+    .out_rr01r = tgen_divs2,
+};
+
 static const TCGOutOpBinary outop_divu = {
     .base.static_constraint = C_NotImplemented,
 };
@@ -2847,9 +2859,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
         }
         break;
 
-    OP_32_64(div2):
-        tcg_out_modrm(s, OPC_GRP3_Ev + rexw, EXT3_IDIV, args[4]);
-        break;
     OP_32_64(divu2):
         tcg_out_modrm(s, OPC_GRP3_Ev + rexw, EXT3_DIV, args[4]);
         break;
@@ -3789,8 +3798,6 @@ tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
     case INDEX_op_movcond_i64:
         return C_O1_I4(r, r, reT, r, 0);
 
-    case INDEX_op_div2_i32:
-    case INDEX_op_div2_i64:
     case INDEX_op_divu2_i32:
     case INDEX_op_divu2_i64:
         return C_O2_I3(a, d, 0, 1, r);

tcg/loongarch64/tcg-target.c.inc

@@ -1343,6 +1343,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rrr = tgen_divs,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu(TCGContext *s, TCGType type,
                       TCGReg a0, TCGReg a1, TCGReg a2)
 {

tcg/mips/tcg-target.c.inc

@@ -1733,6 +1733,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rrr = tgen_divs,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu(TCGContext *s, TCGType type,
                       TCGReg a0, TCGReg a1, TCGReg a2)
 {

tcg/ppc/tcg-target.c.inc

@@ -2972,6 +2972,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rrr = tgen_divs,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu(TCGContext *s, TCGType type,
                       TCGReg a0, TCGReg a1, TCGReg a2)
 {

tcg/riscv/tcg-target.c.inc

@@ -2009,6 +2009,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rrr = tgen_divs,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu(TCGContext *s, TCGType type,
                       TCGReg a0, TCGReg a1, TCGReg a2)
 {

tcg/s390x/tcg-target.c.inc

@@ -2246,6 +2246,28 @@ static const TCGOutOpBinary outop_divs = {
     .base.static_constraint = C_NotImplemented,
 };
 
+static void tgen_divs2(TCGContext *s, TCGType type,
+                       TCGReg a0, TCGReg a1, TCGReg a4)
+{
+    tcg_debug_assert((a1 & 1) == 0);
+    tcg_debug_assert(a0 == a1 + 1);
+    if (type == TCG_TYPE_I32) {
+        tcg_out_insn(s, RR, DR, a1, a4);
+    } else {
+        /*
+         * TODO: Move the sign-extend of the numerator from a2 into a3
+         * into the tcg backend, instead of in early expansion.  It is
+         * required for 32-bit DR, but not 64-bit DSGR.
+         */
+        tcg_out_insn(s, RRE, DSGR, a1, a4);
+    }
+}
+
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_O2_I3(o, m, 0, 1, r),
+    .out_rr01r = tgen_divs2,
+};
+
 static const TCGOutOpBinary outop_divu = {
     .base.static_constraint = C_NotImplemented,
 };
@@ -2527,13 +2549,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
         tcg_out_st(s, TCG_TYPE_I32, args[0], args[1], args[2]);
         break;
 
-    case INDEX_op_div2_i32:
-        tcg_debug_assert(args[0] == args[2]);
-        tcg_debug_assert(args[1] == args[3]);
-        tcg_debug_assert((args[1] & 1) == 0);
-        tcg_debug_assert(args[0] == args[1] + 1);
-        tcg_out_insn(s, RR, DR, args[1], args[4]);
-        break;
     case INDEX_op_divu2_i32:
         tcg_debug_assert(args[0] == args[2]);
         tcg_debug_assert(args[1] == args[3]);
@@ -2702,19 +2717,6 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
         tcg_out_insn(s, RRE, LRVGR, args[0], args[1]);
         break;
 
-    case INDEX_op_div2_i64:
-        /*
-         * ??? We get an unnecessary sign-extension of the dividend
-         * into op0 with this definition, but as we do in fact always
-         * produce both quotient and remainder using INDEX_op_div_i64
-         * instead requires jumping through even more hoops.
-         */
-        tcg_debug_assert(args[0] == args[2]);
-        tcg_debug_assert(args[1] == args[3]);
-        tcg_debug_assert((args[1] & 1) == 0);
-        tcg_debug_assert(args[0] == args[1] + 1);
-        tcg_out_insn(s, RRE, DSGR, args[1], args[4]);
-        break;
     case INDEX_op_divu2_i64:
         tcg_debug_assert(args[0] == args[2]);
         tcg_debug_assert(args[1] == args[3]);
@@ -3396,8 +3398,6 @@ tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
     case INDEX_op_movcond_i64:
         return C_O1_I4(r, r, rC, rI, r);
 
-    case INDEX_op_div2_i32:
-    case INDEX_op_div2_i64:
     case INDEX_op_divu2_i32:
     case INDEX_op_divu2_i64:
         return C_O2_I3(o, m, 0, 1, r);

tcg/sparc64/tcg-target.c.inc

@@ -1352,6 +1352,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rri = tgen_divsi,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu_rJ(TCGContext *s, TCGType type,
                          TCGReg a0, TCGReg a1, TCGArg a2, bool c2)
 {

tcg/tcg.c

@@ -986,6 +986,12 @@ typedef struct TCGOutOpBinary {
                     TCGReg a0, TCGReg a1, tcg_target_long a2);
 } TCGOutOpBinary;
 
+typedef struct TCGOutOpDivRem {
+    TCGOutOp base;
+    void (*out_rr01r)(TCGContext *s, TCGType type,
+                      TCGReg a0, TCGReg a1, TCGReg a4);
+} TCGOutOpDivRem;
+
 typedef struct TCGOutOpUnary {
     TCGOutOp base;
     void (*out_rr)(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1);
@@ -1022,6 +1028,8 @@ static const TCGOutOp * const all_outop[NB_OPS] = {
     OUTOP(INDEX_op_andc, TCGOutOpBinary, outop_andc),
     OUTOP(INDEX_op_divs, TCGOutOpBinary, outop_divs),
     OUTOP(INDEX_op_divu, TCGOutOpBinary, outop_divu),
+    OUTOP(INDEX_op_div2_i32, TCGOutOpDivRem, outop_divs2),
+    OUTOP(INDEX_op_div2_i64, TCGOutOpDivRem, outop_divs2),
     OUTOP(INDEX_op_eqv, TCGOutOpBinary, outop_eqv),
     OUTOP(INDEX_op_mul, TCGOutOpBinary, outop_mul),
     OUTOP(INDEX_op_mulsh, TCGOutOpBinary, outop_mulsh),
@@ -2265,7 +2273,6 @@ bool tcg_op_supported(TCGOpcode op, TCGType type, unsigned flags)
     case INDEX_op_rem_i32:
     case INDEX_op_remu_i32:
         return TCG_TARGET_HAS_rem_i32;
-    case INDEX_op_div2_i32:
     case INDEX_op_divu2_i32:
         return TCG_TARGET_HAS_div2_i32;
     case INDEX_op_rotl_i32:
@@ -2325,7 +2332,6 @@ bool tcg_op_supported(TCGOpcode op, TCGType type, unsigned flags)
     case INDEX_op_rem_i64:
     case INDEX_op_remu_i64:
         return TCG_TARGET_HAS_rem_i64;
-    case INDEX_op_div2_i64:
     case INDEX_op_divu2_i64:
         return TCG_TARGET_HAS_div2_i64;
     case INDEX_op_rotl_i64:
@@ -5467,6 +5473,20 @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op)
         }
         break;
 
+    case INDEX_op_div2_i32:
+    case INDEX_op_div2_i64:
+        {
+            const TCGOutOpDivRem *out =
+                container_of(all_outop[op->opc], TCGOutOpDivRem, base);
+
+            /* Only used by x86 and s390x, which use matching constraints. */
+            tcg_debug_assert(new_args[0] == new_args[2]);
+            tcg_debug_assert(new_args[1] == new_args[3]);
+            tcg_debug_assert(!const_args[4]);
+            out->out_rr01r(s, type, new_args[0], new_args[1], new_args[4]);
+        }
+        break;
+
     default:
         if (def->flags & TCG_OPF_VECTOR) {
             tcg_out_vec_op(s, op->opc, type - TCG_TYPE_V64,

tcg/tci/tcg-target.c.inc

@@ -658,6 +658,10 @@ static const TCGOutOpBinary outop_divs = {
     .out_rrr = tgen_divs,
 };
 
+static const TCGOutOpDivRem outop_divs2 = {
+    .base.static_constraint = C_NotImplemented,
+};
+
 static void tgen_divu(TCGContext *s, TCGType type,
                       TCGReg a0, TCGReg a1, TCGReg a2)
 {