tcg/{i386, s390x}: Add earlyclobber to the op_add2's first output

i386 and s390x implementations of op_add2 require an earlyclobber,
which is currently missing. This breaks VCKSM in s390x guests. E.g., on
x86_64 the following op:

    add2_i32 tmp2,tmp3,tmp2,tmp3,tmp3,tmp2   dead: 0 2 3 4 5  pref=none,0xffff

is translated to:

    addl     %ebx, %r12d
    adcl     %r12d, %ebx

Introduce a new C_N1_O1_I4 constraint, and make sure that earlyclobber
of aliased outputs is honored.

Cc: qemu-stable@nongnu.org
Fixes: 82790a870992 ("tcg: Add markup for output requires new register")
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20230719221310.1968845-7-iii@linux.ibm.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

5 files changed, 19 insertions, 8 deletions

diff --git a/tcg/i386/tcg-target-con-set.h b/tcg/i386/tcg-target-con-set.h
index 91ceb0e..5ea3a29 100644
--- a/tcg/i386/tcg-target-con-set.h
+++ b/tcg/i386/tcg-target-con-set.h
@@ -11,6 +11,9 @@
  *
  * C_N1_Im(...) defines a constraint set with 1 output and <m> inputs,
  * except that the output must use a new register.
+ *
+ * C_Nn_Om_Ik(...) defines a constraint set with <n + m> outputs and <k>
+ * inputs, except that the first <n> outputs must use new registers.
  */
 C_O0_I1(r)
 C_O0_I2(L, L)
@@ -53,4 +56,4 @@ C_O2_I1(r, r, L)
 C_O2_I2(a, d, a, r)
 C_O2_I2(r, r, L, L)
 C_O2_I3(a, d, 0, 1, r)
-C_O2_I4(r, r, 0, 1, re, re)
+C_N1_O1_I4(r, r, 0, 1, re, re)
diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc
index ab997b5..77482da 100644
--- a/tcg/i386/tcg-target.c.inc
+++ b/tcg/i386/tcg-target.c.inc
@@ -3335,7 +3335,7 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
     case INDEX_op_add2_i64:
     case INDEX_op_sub2_i32:
     case INDEX_op_sub2_i64:
-        return C_O2_I4(r, r, 0, 1, re, re);
+        return C_N1_O1_I4(r, r, 0, 1, re, re);
 
     case INDEX_op_ctz_i32:
     case INDEX_op_ctz_i64:
diff --git a/tcg/s390x/tcg-target-con-set.h b/tcg/s390x/tcg-target-con-set.h
index cbad91b..9a42037 100644
--- a/tcg/s390x/tcg-target-con-set.h
+++ b/tcg/s390x/tcg-target-con-set.h
@@ -8,6 +8,9 @@
  * C_On_Im(...) defines a constraint set with <n> outputs and <m> inputs.
  * Each operand should be a sequence of constraint letters as defined by
  * tcg-target-con-str.h; the constraint combination is inclusive or.
+ *
+ * C_Nn_Om_Ik(...) defines a constraint set with <n + m> outputs and <k>
+ * inputs, except that the first <n> outputs must use new registers.
  */
 C_O0_I1(r)
 C_O0_I2(r, r)
@@ -41,6 +44,5 @@ C_O2_I1(o, m, r)
 C_O2_I2(o, m, 0, r)
 C_O2_I2(o, m, r, r)
 C_O2_I3(o, m, 0, 1, r)
-C_O2_I4(r, r, 0, 1, rA, r)
-C_O2_I4(r, r, 0, 1, ri, r)
-C_O2_I4(r, r, 0, 1, r, r)
+C_N1_O1_I4(r, r, 0, 1, ri, r)
+C_N1_O1_I4(r, r, 0, 1, rA, r)
diff --git a/tcg/s390x/tcg-target.c.inc b/tcg/s390x/tcg-target.c.inc
index a878acd..a94f790 100644
--- a/tcg/s390x/tcg-target.c.inc
+++ b/tcg/s390x/tcg-target.c.inc
@@ -3229,11 +3229,11 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
 
     case INDEX_op_add2_i32:
     case INDEX_op_sub2_i32:
-        return C_O2_I4(r, r, 0, 1, ri, r);
+        return C_N1_O1_I4(r, r, 0, 1, ri, r);
 
     case INDEX_op_add2_i64:
     case INDEX_op_sub2_i64:
-        return C_O2_I4(r, r, 0, 1, rA, r);
+        return C_N1_O1_I4(r, r, 0, 1, rA, r);
 
     case INDEX_op_st_vec:
         return C_O0_I2(v, r);
diff --git a/tcg/tcg.c b/tcg/tcg.c
index 652e8ea..ddfe9a9 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -648,6 +648,7 @@ static void tcg_out_movext3(TCGContext *s, const TCGMovExtend *i1,
 #define C_O2_I2(O1, O2, I1, I2)         C_PFX4(c_o2_i2_, O1, O2, I1, I2),
 #define C_O2_I3(O1, O2, I1, I2, I3)     C_PFX5(c_o2_i3_, O1, O2, I1, I2, I3),
 #define C_O2_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_o2_i4_, O1, O2, I1, I2, I3, I4),
+#define C_N1_O1_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_n1_o1_i4_, O1, O2, I1, I2, I3, I4),
 
 typedef enum {
 #include "tcg-target-con-set.h"
@@ -668,6 +669,7 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode);
 #undef C_O2_I2
 #undef C_O2_I3
 #undef C_O2_I4
+#undef C_N1_O1_I4
 
 /* Put all of the constraint sets into an array, indexed by the enum. */
 
@@ -687,6 +689,7 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode);
 #define C_O2_I2(O1, O2, I1, I2)         { .args_ct_str = { #O1, #O2, #I1, #I2 } },
 #define C_O2_I3(O1, O2, I1, I2, I3)     { .args_ct_str = { #O1, #O2, #I1, #I2, #I3 } },
 #define C_O2_I4(O1, O2, I1, I2, I3, I4) { .args_ct_str = { #O1, #O2, #I1, #I2, #I3, #I4 } },
+#define C_N1_O1_I4(O1, O2, I1, I2, I3, I4) { .args_ct_str = { "&" #O1, #O2, #I1, #I2, #I3, #I4 } },
 
 static const TCGTargetOpDef constraint_sets[] = {
 #include "tcg-target-con-set.h"
@@ -706,6 +709,7 @@ static const TCGTargetOpDef constraint_sets[] = {
 #undef C_O2_I2
 #undef C_O2_I3
 #undef C_O2_I4
+#undef C_N1_O1_I4
 
 /* Expand the enumerator to be returned from tcg_target_op_def(). */
 
@@ -725,6 +729,7 @@ static const TCGTargetOpDef constraint_sets[] = {
 #define C_O2_I2(O1, O2, I1, I2)         C_PFX4(c_o2_i2_, O1, O2, I1, I2)
 #define C_O2_I3(O1, O2, I1, I2, I3)     C_PFX5(c_o2_i3_, O1, O2, I1, I2, I3)
 #define C_O2_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_o2_i4_, O1, O2, I1, I2, I3, I4)
+#define C_N1_O1_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_n1_o1_i4_, O1, O2, I1, I2, I3, I4)
 
 #include "tcg-target.c.inc"
 
@@ -4703,7 +4708,8 @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op)
                  * dead after the instruction, we must allocate a new
                  * register and move it.
                  */
-                if (temp_readonly(ts) || !IS_DEAD_ARG(i)) {
+                if (temp_readonly(ts) || !IS_DEAD_ARG(i)
+                    || def->args_ct[arg_ct->alias_index].newreg) {
                     allocate_new_reg = true;
                 } else if (ts->val_type == TEMP_VAL_REG) {
                     /*