[PULL 00/29] tcg patch queue

[PULL 00/29] tcg patch queue

[Richard Henderson]

The following changes since commit aec6836c73403cffa56b9a4c5556451ee16071fe:

  Merge tag 'pull-vfio-20250626' of https://github.com/legoater/qemu into staging (2025-06-26 10:25:01 -0400)

are available in the Git repository at:

  https://gitlab.com/rth7680/qemu.git tags/pull-tcg-20250630

for you to fetch changes up to 0d0fc3f4658937fb81fcc16a89738e83bd8d4795:

  tcg: Fix constant propagation in tcg_reg_alloc_dup (2025-06-30 07:42:56 -0600)

----------------------------------------------------------------
tcg/optimize: Build and use one's mask in logical operations
tcg/optimize: Use fold_and in do_constant_folding_cond[12]
tcg/optimize: Fold and to extract during optimize
tcg/optimize: Simplify some fold constant checks
tcg/riscv: Fix typo in tgen_extract
tcg: Fix constant propagation in tcg_reg_alloc_dup

----------------------------------------------------------------
Richard Henderson (29):
      tcg/optimize: Introduce arg_const_val
      tcg/optimize: Add one's mask to TempOptInfo
      tcg/optimize: Introduce fold_masks_zosa
      tcg/optimize: Build and use o_bits in fold_and
      tcg/optimize: Build and use o_bits in fold_andc
      tcg/optimize: Build and use z_bits and o_bits in fold_eqv
      tcg/optimize: Build and use z_bits and o_bits in fold_nand
      tcg/optimize: Build and use z_bits and o_bits in fold_nor
      tcg/optimize: Build and use z_bits and o_bits in fold_not
      tcg/optimize: Build and use one and affected bits in fold_or
      tcg/optimize: Build and use zero, one and affected bits in fold_orc
      tcg/optimize: Build and use o_bits in fold_xor
      tcg/optimize: Build and use o_bits in fold_bswap
      tcg/optimize: Build and use o_bits in fold_deposit
      tcg/optimize: Build and use o_bits in fold_extract
      tcg/optimize: Build and use z_bits and o_bits in fold_extract2
      tcg/optimize: Build and use o_bits in fold_exts
      tcg/optimize: Build and use o_bits in fold_extu
      tcg/optimize: Build and use o_bits in fold_movcond
      tcg/optimize: Build and use o_bits in fold_sextract
      tcg/optimize: Build and use o_bits in fold_shift
      tcg/optimize: Use fold_and in do_constant_folding_cond[12]
      tcg/optimize: Fold and to extract during optimize
      tcg/optimize: Simplify fold_and constant checks
      tcg/optimize: Simplify fold_andc constant checks
      tcg/optimize: Simplify fold_orc constant checks
      tcg/optimize: Simplify fold_eqv constant checks
      tcg/riscv: Fix typo in tgen_extract
      tcg: Fix constant propagation in tcg_reg_alloc_dup

 tcg/optimize.c             | 456 +++++++++++++++++++++++++++------------------
 tcg/tcg.c                  |   2 +-
 tcg/riscv/tcg-target.c.inc |   2 +-
 3 files changed, 276 insertions(+), 184 deletions(-)

[PULL 01/29] tcg/optimize: Introduce arg_const_val

[Richard Henderson]

Use arg_const_val instead of direct access to the TempOptInfo val
member.  Rename both val and is_const to catch all direct accesses.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 78 ++++++++++++++++++++++++++------------------------
 1 file changed, 41 insertions(+), 37 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 10a76c5461..73a272eeb3 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -39,11 +39,11 @@ typedef struct MemCopyInfo {
 } MemCopyInfo;
 
 typedef struct TempOptInfo {
-    bool is_const;
+    bool is_const_;
     TCGTemp *prev_copy;
     TCGTemp *next_copy;
     QSIMPLEQ_HEAD(, MemCopyInfo) mem_copy;
-    uint64_t val;
+    uint64_t val_;
     uint64_t z_mask;  /* mask bit is 0 if and only if value bit is 0 */
     uint64_t s_mask;  /* mask bit is 1 if value bit matches msb */
 } TempOptInfo;
@@ -73,12 +73,12 @@ static inline TempOptInfo *arg_info(TCGArg arg)
 
 static inline bool ti_is_const(TempOptInfo *ti)
 {
-    return ti->is_const;
+    return ti->is_const_;
 }
 
 static inline uint64_t ti_const_val(TempOptInfo *ti)
 {
-    return ti->val;
+    return ti->val_;
 }
 
 static inline bool ti_is_const_val(TempOptInfo *ti, uint64_t val)
@@ -101,6 +101,11 @@ static inline bool arg_is_const(TCGArg arg)
     return ts_is_const(arg_temp(arg));
 }
 
+static inline uint64_t arg_const_val(TCGArg arg)
+{
+    return ti_const_val(arg_info(arg));
+}
+
 static inline bool arg_is_const_val(TCGArg arg, uint64_t val)
 {
     return ts_is_const_val(arg_temp(arg), val);
@@ -137,12 +142,12 @@ static void init_ts_info(OptContext *ctx, TCGTemp *ts)
     ti->prev_copy = ts;
     QSIMPLEQ_INIT(&ti->mem_copy);
     if (ts->kind == TEMP_CONST) {
-        ti->is_const = true;
-        ti->val = ts->val;
+        ti->is_const_ = true;
+        ti->val_ = ts->val;
         ti->z_mask = ts->val;
         ti->s_mask = INT64_MIN >> clrsb64(ts->val);
     } else {
-        ti->is_const = false;
+        ti->is_const_ = false;
         ti->z_mask = -1;
         ti->s_mask = 0;
     }
@@ -229,7 +234,7 @@ static void reset_ts(OptContext *ctx, TCGTemp *ts)
     pi->next_copy = ti->next_copy;
     ti->next_copy = ts;
     ti->prev_copy = ts;
-    ti->is_const = false;
+    ti->is_const_ = false;
     ti->z_mask = -1;
     ti->s_mask = 0;
 
@@ -394,8 +399,8 @@ static bool tcg_opt_gen_mov(OptContext *ctx, TCGOp *op, TCGArg dst, TCGArg src)
         di->prev_copy = src_ts;
         ni->prev_copy = dst_ts;
         si->next_copy = dst_ts;
-        di->is_const = si->is_const;
-        di->val = si->val;
+        di->is_const_ = si->is_const_;
+        di->val_ = si->val_;
 
         if (!QSIMPLEQ_EMPTY(&si->mem_copy)
             && cmp_better_copy(src_ts, dst_ts) == dst_ts) {
@@ -687,8 +692,8 @@ static int do_constant_folding_cond(TCGType type, TCGArg x,
                                     TCGArg y, TCGCond c)
 {
     if (arg_is_const(x) && arg_is_const(y)) {
-        uint64_t xv = arg_info(x)->val;
-        uint64_t yv = arg_info(y)->val;
+        uint64_t xv = arg_const_val(x);
+        uint64_t yv = arg_const_val(y);
 
         switch (type) {
         case TCG_TYPE_I32:
@@ -801,14 +806,14 @@ static int do_constant_folding_cond1(OptContext *ctx, TCGOp *op, TCGArg dest,
      * TSTNE x,i -> NE x,0 if i includes all nonzero bits of x
      */
     if (args_are_copies(*p1, *p2) ||
-        (arg_is_const(*p2) && (i1->z_mask & ~arg_info(*p2)->val) == 0)) {
+        (arg_is_const(*p2) && (i1->z_mask & ~arg_const_val(*p2)) == 0)) {
         *p2 = arg_new_constant(ctx, 0);
         *pcond = tcg_tst_eqne_cond(cond);
         return -1;
     }
 
     /* TSTNE x,i -> LT x,0 if i only includes sign bit copies */
-    if (arg_is_const(*p2) && (arg_info(*p2)->val & ~i1->s_mask) == 0) {
+    if (arg_is_const(*p2) && (arg_const_val(*p2) & ~i1->s_mask) == 0) {
         *p2 = arg_new_constant(ctx, 0);
         *pcond = tcg_tst_ltge_cond(cond);
         return -1;
@@ -849,13 +854,13 @@ static int do_constant_folding_cond2(OptContext *ctx, TCGOp *op, TCGArg *args)
     bh = args[3];
 
     if (arg_is_const(bl) && arg_is_const(bh)) {
-        tcg_target_ulong blv = arg_info(bl)->val;
-        tcg_target_ulong bhv = arg_info(bh)->val;
+        tcg_target_ulong blv = arg_const_val(bl);
+        tcg_target_ulong bhv = arg_const_val(bh);
         uint64_t b = deposit64(blv, 32, 32, bhv);
 
         if (arg_is_const(al) && arg_is_const(ah)) {
-            tcg_target_ulong alv = arg_info(al)->val;
-            tcg_target_ulong ahv = arg_info(ah)->val;
+            tcg_target_ulong alv = arg_const_val(al);
+            tcg_target_ulong ahv = arg_const_val(ah);
             uint64_t a = deposit64(alv, 32, 32, ahv);
 
             r = do_constant_folding_cond_64(a, b, c);
@@ -989,9 +994,8 @@ static bool finish_folding(OptContext *ctx, TCGOp *op)
 static bool fold_const1(OptContext *ctx, TCGOp *op)
 {
     if (arg_is_const(op->args[1])) {
-        uint64_t t;
+        uint64_t t = arg_const_val(op->args[1]);
 
-        t = arg_info(op->args[1])->val;
         t = do_constant_folding(op->opc, ctx->type, t, 0);
         return tcg_opt_gen_movi(ctx, op, op->args[0], t);
     }
@@ -1001,8 +1005,8 @@ static bool fold_const1(OptContext *ctx, TCGOp *op)
 static bool fold_const2(OptContext *ctx, TCGOp *op)
 {
     if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) {
-        uint64_t t1 = arg_info(op->args[1])->val;
-        uint64_t t2 = arg_info(op->args[2])->val;
+        uint64_t t1 = arg_const_val(op->args[1]);
+        uint64_t t2 = arg_const_val(op->args[2]);
 
         t1 = do_constant_folding(op->opc, ctx->type, t1, t2);
         return tcg_opt_gen_movi(ctx, op, op->args[0], t1);
@@ -1486,8 +1490,8 @@ static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op)
     }
 
     if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])) {
-        uint64_t tv = arg_info(op->args[2])->val;
-        uint64_t fv = arg_info(op->args[3])->val;
+        uint64_t tv = arg_const_val(op->args[2]);
+        uint64_t fv = arg_const_val(op->args[3]);
 
         if (tv == -1 && fv == 0) {
             return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]);
@@ -1504,7 +1508,7 @@ static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op)
         }
     }
     if (arg_is_const(op->args[2])) {
-        uint64_t tv = arg_info(op->args[2])->val;
+        uint64_t tv = arg_const_val(op->args[2]);
         if (tv == -1) {
             op->opc = INDEX_op_or_vec;
             op->args[2] = op->args[3];
@@ -1518,7 +1522,7 @@ static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op)
         }
     }
     if (arg_is_const(op->args[3])) {
-        uint64_t fv = arg_info(op->args[3])->val;
+        uint64_t fv = arg_const_val(op->args[3]);
         if (fv == 0) {
             op->opc = INDEX_op_and_vec;
             return fold_and(ctx, op);
@@ -1876,7 +1880,7 @@ static bool fold_divide(OptContext *ctx, TCGOp *op)
 static bool fold_dup(OptContext *ctx, TCGOp *op)
 {
     if (arg_is_const(op->args[1])) {
-        uint64_t t = arg_info(op->args[1])->val;
+        uint64_t t = arg_const_val(op->args[1]);
         t = dup_const(TCGOP_VECE(op), t);
         return tcg_opt_gen_movi(ctx, op, op->args[0], t);
     }
@@ -1886,8 +1890,8 @@ static bool fold_dup(OptContext *ctx, TCGOp *op)
 static bool fold_dup2(OptContext *ctx, TCGOp *op)
 {
     if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) {
-        uint64_t t = deposit64(arg_info(op->args[1])->val, 32, 32,
-                               arg_info(op->args[2])->val);
+        uint64_t t = deposit64(arg_const_val(op->args[1]), 32, 32,
+                               arg_const_val(op->args[2]));
         return tcg_opt_gen_movi(ctx, op, op->args[0], t);
     }
 
@@ -1958,8 +1962,8 @@ static bool fold_extract(OptContext *ctx, TCGOp *op)
 static bool fold_extract2(OptContext *ctx, TCGOp *op)
 {
     if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) {
-        uint64_t v1 = arg_info(op->args[1])->val;
-        uint64_t v2 = arg_info(op->args[2])->val;
+        uint64_t v1 = arg_const_val(op->args[1]);
+        uint64_t v2 = arg_const_val(op->args[2]);
         int shr = op->args[3];
 
         if (ctx->type == TCG_TYPE_I32) {
@@ -2127,8 +2131,8 @@ static bool fold_multiply2(OptContext *ctx, TCGOp *op)
     swap_commutative(op->args[0], &op->args[2], &op->args[3]);
 
     if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])) {
-        uint64_t a = arg_info(op->args[2])->val;
-        uint64_t b = arg_info(op->args[3])->val;
+        uint64_t a = arg_const_val(op->args[2]);
+        uint64_t b = arg_const_val(op->args[3]);
         uint64_t h, l;
         TCGArg rl, rh;
         TCGOp *op2;
@@ -2330,7 +2334,7 @@ static int fold_setcond_zmask(OptContext *ctx, TCGOp *op, bool neg)
     }
 
     a_zmask = arg_info(op->args[1])->z_mask;
-    b_val = arg_info(op->args[2])->val;
+    b_val = arg_const_val(op->args[2]);
     cond = op->args[3];
 
     if (ctx->type == TCG_TYPE_I32) {
@@ -2418,7 +2422,7 @@ static void fold_setcond_tst_pow2(OptContext *ctx, TCGOp *op, bool neg)
     }
 
     src2 = op->args[2];
-    val = arg_info(src2)->val;
+    val = arg_const_val(src2);
     if (!is_power_of_2(val)) {
         return;
     }
@@ -2669,7 +2673,7 @@ static bool fold_sub_to_neg(OptContext *ctx, TCGOp *op)
     TCGOpcode neg_op;
     bool have_neg;
 
-    if (!arg_is_const(op->args[1]) || arg_info(op->args[1])->val != 0) {
+    if (!arg_is_const_val(op->args[1], 0)) {
         return false;
     }
 
@@ -2719,7 +2723,7 @@ static bool fold_sub(OptContext *ctx, TCGOp *op)
 
     /* Fold sub r,x,i to add r,x,-i */
     if (arg_is_const(op->args[2])) {
-        uint64_t val = arg_info(op->args[2])->val;
+        uint64_t val = arg_const_val(op->args[2]);
 
         op->opc = INDEX_op_add;
         op->args[2] = arg_new_constant(ctx, -val);
-- 
2.43.0

[PULL 02/29] tcg/optimize: Add one's mask to TempOptInfo

[Richard Henderson]

Add o_mask mirroring z_mask, but for 1's instead of 0's.
Drop is_const and val fields, which now logically overlap.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 51 ++++++++++++++++++++++++++++++++++----------------
 1 file changed, 35 insertions(+), 16 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 73a272eeb3..ce3cb4d7bc 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -39,12 +39,11 @@ typedef struct MemCopyInfo {
 } MemCopyInfo;
 
 typedef struct TempOptInfo {
-    bool is_const_;
     TCGTemp *prev_copy;
     TCGTemp *next_copy;
     QSIMPLEQ_HEAD(, MemCopyInfo) mem_copy;
-    uint64_t val_;
     uint64_t z_mask;  /* mask bit is 0 if and only if value bit is 0 */
+    uint64_t o_mask;  /* mask bit is 1 if and only if value bit is 1 */
     uint64_t s_mask;  /* mask bit is 1 if value bit matches msb */
 } TempOptInfo;
 
@@ -73,12 +72,14 @@ static inline TempOptInfo *arg_info(TCGArg arg)
 
 static inline bool ti_is_const(TempOptInfo *ti)
 {
-    return ti->is_const_;
+    /* If all bits that are not known zeros are known ones, it's constant. */
+    return ti->z_mask == ti->o_mask;
 }
 
 static inline uint64_t ti_const_val(TempOptInfo *ti)
 {
-    return ti->val_;
+    /* If constant, both z_mask and o_mask contain the value. */
+    return ti->z_mask;
 }
 
 static inline bool ti_is_const_val(TempOptInfo *ti, uint64_t val)
@@ -142,13 +143,12 @@ static void init_ts_info(OptContext *ctx, TCGTemp *ts)
     ti->prev_copy = ts;
     QSIMPLEQ_INIT(&ti->mem_copy);
     if (ts->kind == TEMP_CONST) {
-        ti->is_const_ = true;
-        ti->val_ = ts->val;
         ti->z_mask = ts->val;
+        ti->o_mask = ts->val;
         ti->s_mask = INT64_MIN >> clrsb64(ts->val);
     } else {
-        ti->is_const_ = false;
         ti->z_mask = -1;
+        ti->o_mask = 0;
         ti->s_mask = 0;
     }
 }
@@ -234,8 +234,8 @@ static void reset_ts(OptContext *ctx, TCGTemp *ts)
     pi->next_copy = ti->next_copy;
     ti->next_copy = ts;
     ti->prev_copy = ts;
-    ti->is_const_ = false;
     ti->z_mask = -1;
+    ti->o_mask = 0;
     ti->s_mask = 0;
 
     if (!QSIMPLEQ_EMPTY(&ti->mem_copy)) {
@@ -390,6 +390,7 @@ static bool tcg_opt_gen_mov(OptContext *ctx, TCGOp *op, TCGArg dst, TCGArg src)
     op->args[1] = src;
 
     di->z_mask = si->z_mask;
+    di->o_mask = si->o_mask;
     di->s_mask = si->s_mask;
 
     if (src_ts->type == dst_ts->type) {
@@ -399,13 +400,19 @@ static bool tcg_opt_gen_mov(OptContext *ctx, TCGOp *op, TCGArg dst, TCGArg src)
         di->prev_copy = src_ts;
         ni->prev_copy = dst_ts;
         si->next_copy = dst_ts;
-        di->is_const_ = si->is_const_;
-        di->val_ = si->val_;
 
         if (!QSIMPLEQ_EMPTY(&si->mem_copy)
             && cmp_better_copy(src_ts, dst_ts) == dst_ts) {
             move_mem_copies(dst_ts, src_ts);
         }
+    } else if (dst_ts->type == TCG_TYPE_I32) {
+        di->z_mask = (int32_t)di->z_mask;
+        di->o_mask = (int32_t)di->o_mask;
+        di->s_mask |= INT32_MIN;
+    } else {
+        di->z_mask |= MAKE_64BIT_MASK(32, 32);
+        di->o_mask = (uint32_t)di->o_mask;
+        di->s_mask = INT64_MIN;
     }
     return true;
 }
@@ -1032,8 +1039,8 @@ static bool fold_const2_commutative(OptContext *ctx, TCGOp *op)
  * If z_mask allows, fold the output to constant zero.
  * The passed s_mask may be augmented by z_mask.
  */
-static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
-                          uint64_t z_mask, int64_t s_mask)
+static bool fold_masks_zos(OptContext *ctx, TCGOp *op, uint64_t z_mask,
+                           uint64_t o_mask, int64_t s_mask)
 {
     const TCGOpDef *def = &tcg_op_defs[op->opc];
     TCGTemp *ts;
@@ -1052,11 +1059,16 @@ static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
      */
     if (ctx->type == TCG_TYPE_I32) {
         z_mask = (int32_t)z_mask;
+        o_mask = (int32_t)o_mask;
         s_mask |= INT32_MIN;
     }
 
-    if (z_mask == 0) {
-        return tcg_opt_gen_movi(ctx, op, op->args[0], 0);
+    /* Bits that are known 1 and bits that are known 0 must not overlap. */
+    tcg_debug_assert((o_mask & ~z_mask) == 0);
+
+    /* All bits that are not known zero are known one is a constant. */
+    if (z_mask == o_mask) {
+        return tcg_opt_gen_movi(ctx, op, op->args[0], o_mask);
     }
 
     ts = arg_temp(op->args[0]);
@@ -1068,20 +1080,27 @@ static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
     /* Canonicalize s_mask and incorporate data from z_mask. */
     rep = clz64(~s_mask);
     rep = MAX(rep, clz64(z_mask));
+    rep = MAX(rep, clz64(~o_mask));
     rep = MAX(rep - 1, 0);
     ti->s_mask = INT64_MIN >> rep;
 
     return true;
 }
 
+static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
+                          uint64_t z_mask, uint64_t s_mask)
+{
+    return fold_masks_zos(ctx, op, z_mask, 0, s_mask);
+}
+
 static bool fold_masks_z(OptContext *ctx, TCGOp *op, uint64_t z_mask)
 {
-    return fold_masks_zs(ctx, op, z_mask, 0);
+    return fold_masks_zos(ctx, op, z_mask, 0, 0);
 }
 
 static bool fold_masks_s(OptContext *ctx, TCGOp *op, uint64_t s_mask)
 {
-    return fold_masks_zs(ctx, op, -1, s_mask);
+    return fold_masks_zos(ctx, op, -1, 0, s_mask);
 }
 
 /*
-- 
2.43.0

[PULL 03/29] tcg/optimize: Introduce fold_masks_zosa

[Richard Henderson]

Add a new function with an affected mask.  This will allow
folding to a constant to happen before folding to a copy,
without having to mind the ordering in all users.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index ce3cb4d7bc..49ef039932 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1039,8 +1039,8 @@ static bool fold_const2_commutative(OptContext *ctx, TCGOp *op)
  * If z_mask allows, fold the output to constant zero.
  * The passed s_mask may be augmented by z_mask.
  */
-static bool fold_masks_zos(OptContext *ctx, TCGOp *op, uint64_t z_mask,
-                           uint64_t o_mask, int64_t s_mask)
+static bool fold_masks_zosa(OptContext *ctx, TCGOp *op, uint64_t z_mask,
+                            uint64_t o_mask, int64_t s_mask, uint64_t a_mask)
 {
     const TCGOpDef *def = &tcg_op_defs[op->opc];
     TCGTemp *ts;
@@ -1061,6 +1061,7 @@ static bool fold_masks_zos(OptContext *ctx, TCGOp *op, uint64_t z_mask,
         z_mask = (int32_t)z_mask;
         o_mask = (int32_t)o_mask;
         s_mask |= INT32_MIN;
+        a_mask = (uint32_t)a_mask;
     }
 
     /* Bits that are known 1 and bits that are known 0 must not overlap. */
@@ -1071,6 +1072,11 @@ static bool fold_masks_zos(OptContext *ctx, TCGOp *op, uint64_t z_mask,
         return tcg_opt_gen_movi(ctx, op, op->args[0], o_mask);
     }
 
+    /* If no bits are affected, the operation devolves to a copy. */
+    if (a_mask == 0) {
+        return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]);
+    }
+
     ts = arg_temp(op->args[0]);
     reset_ts(ctx, ts);
 
@@ -1090,17 +1096,17 @@ static bool fold_masks_zos(OptContext *ctx, TCGOp *op, uint64_t z_mask,
 static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
                           uint64_t z_mask, uint64_t s_mask)
 {
-    return fold_masks_zos(ctx, op, z_mask, 0, s_mask);
+    return fold_masks_zosa(ctx, op, z_mask, 0, s_mask, -1);
 }
 
 static bool fold_masks_z(OptContext *ctx, TCGOp *op, uint64_t z_mask)
 {
-    return fold_masks_zos(ctx, op, z_mask, 0, 0);
+    return fold_masks_zosa(ctx, op, z_mask, 0, 0, -1);
 }
 
 static bool fold_masks_s(OptContext *ctx, TCGOp *op, uint64_t s_mask)
 {
-    return fold_masks_zos(ctx, op, -1, 0, s_mask);
+    return fold_masks_zosa(ctx, op, -1, 0, s_mask, -1);
 }
 
 /*
-- 
2.43.0

[PULL 04/29] tcg/optimize: Build and use o_bits in fold_and

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 20 +++++++-------------
 1 file changed, 7 insertions(+), 13 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 49ef039932..d9ccbb36e2 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1422,7 +1422,7 @@ static bool fold_addco(OptContext *ctx, TCGOp *op)
 
 static bool fold_and(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z1, z2, z_mask, s_mask;
+    uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1, *t2;
 
     if (fold_const2_commutative(ctx, op) ||
@@ -1434,18 +1434,9 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
 
     t1 = arg_info(op->args[1]);
     t2 = arg_info(op->args[2]);
-    z1 = t1->z_mask;
-    z2 = t2->z_mask;
 
-    /*
-     * Known-zeros does not imply known-ones.  Therefore unless
-     * arg2 is constant, we can't infer affected bits from it.
-     */
-    if (ti_is_const(t2) && fold_affected_mask(ctx, op, z1 & ~z2)) {
-        return true;
-    }
-
-    z_mask = z1 & z2;
+    z_mask = t1->z_mask & t2->z_mask;
+    o_mask = t1->o_mask & t2->o_mask;
 
     /*
      * Sign repetitions are perforce all identical, whether they are 1 or 0.
@@ -1453,7 +1444,10 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
      */
     s_mask = t1->s_mask & t2->s_mask;
 
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+    /* Affected bits are those not known zero, masked by those known one. */
+    a_mask = t1->z_mask & ~t2->o_mask;
+
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, a_mask);
 }
 
 static bool fold_andc(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 05/29] tcg/optimize: Build and use o_bits in fold_andc

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 23 ++++++++---------------
 1 file changed, 8 insertions(+), 15 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index d9ccbb36e2..123734b167 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1452,7 +1452,7 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
 
 static bool fold_andc(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask, s_mask;
+    uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1, *t2;
 
     if (fold_const2(ctx, op) ||
@@ -1464,7 +1464,6 @@ static bool fold_andc(OptContext *ctx, TCGOp *op)
 
     t1 = arg_info(op->args[1]);
     t2 = arg_info(op->args[2]);
-    z_mask = t1->z_mask;
 
     if (ti_is_const(t2)) {
         /* Fold andc r,x,i to and r,x,~i. */
@@ -1485,20 +1484,14 @@ static bool fold_andc(OptContext *ctx, TCGOp *op)
         return fold_and(ctx, op);
     }
 
-    /*
-     * Known-zeros does not imply known-ones.  Therefore unless
-     * arg2 is constant, we can't infer anything from it.
-     */
-    if (ti_is_const(t2)) {
-        uint64_t v2 = ti_const_val(t2);
-        if (fold_affected_mask(ctx, op, z_mask & v2)) {
-            return true;
-        }
-        z_mask &= ~v2;
-    }
-
+    z_mask = t1->z_mask & ~t2->o_mask;
+    o_mask = t1->o_mask & ~t2->z_mask;
     s_mask = t1->s_mask & t2->s_mask;
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+
+    /* Affected bits are those not known zero, masked by those known zero. */
+    a_mask = t1->z_mask & t2->z_mask;
+
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, a_mask);
 }
 
 static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 06/29] tcg/optimize: Build and use z_bits and o_bits in fold_eqv

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 123734b167..6d35a2e58b 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1093,6 +1093,12 @@ static bool fold_masks_zosa(OptContext *ctx, TCGOp *op, uint64_t z_mask,
     return true;
 }
 
+static bool fold_masks_zos(OptContext *ctx, TCGOp *op,
+                           uint64_t z_mask, uint64_t o_mask, uint64_t s_mask)
+{
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, -1);
+}
+
 static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
                           uint64_t z_mask, uint64_t s_mask)
 {
@@ -1916,7 +1922,7 @@ static bool fold_dup2(OptContext *ctx, TCGOp *op)
 
 static bool fold_eqv(OptContext *ctx, TCGOp *op)
 {
-    uint64_t s_mask;
+    uint64_t z_mask, o_mask, s_mask;
     TempOptInfo *t1, *t2;
 
     if (fold_const2_commutative(ctx, op) ||
@@ -1946,8 +1952,12 @@ static bool fold_eqv(OptContext *ctx, TCGOp *op)
     }
 
     t1 = arg_info(op->args[1]);
+
+    z_mask = (t1->z_mask | ~t2->o_mask) & (t2->z_mask | ~t1->o_mask);
+    o_mask = ~(t1->z_mask | t2->z_mask) | (t1->o_mask & t2->o_mask);
     s_mask = t1->s_mask & t2->s_mask;
-    return fold_masks_s(ctx, op, s_mask);
+
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 static bool fold_extract(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 07/29] tcg/optimize: Build and use z_bits and o_bits in fold_nand

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 6d35a2e58b..758f7b142e 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2197,16 +2197,22 @@ static bool fold_multiply2(OptContext *ctx, TCGOp *op)
 
 static bool fold_nand(OptContext *ctx, TCGOp *op)
 {
-    uint64_t s_mask;
+    uint64_t z_mask, o_mask, s_mask;
+    TempOptInfo *t1, *t2;
 
     if (fold_const2_commutative(ctx, op) ||
         fold_xi_to_not(ctx, op, -1)) {
         return true;
     }
 
-    s_mask = arg_info(op->args[1])->s_mask
-           & arg_info(op->args[2])->s_mask;
-    return fold_masks_s(ctx, op, s_mask);
+    t1 = arg_info(op->args[1]);
+    t2 = arg_info(op->args[2]);
+
+    z_mask = ~(t1->o_mask & t2->o_mask);
+    o_mask = ~(t1->z_mask & t2->z_mask);
+    s_mask = t1->s_mask & t2->s_mask;
+
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 static bool fold_neg_no_const(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 08/29] tcg/optimize: Build and use z_bits and o_bits in fold_nor

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 758f7b142e..29d1f29124 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2231,16 +2231,22 @@ static bool fold_neg(OptContext *ctx, TCGOp *op)
 
 static bool fold_nor(OptContext *ctx, TCGOp *op)
 {
-    uint64_t s_mask;
+    uint64_t z_mask, o_mask, s_mask;
+    TempOptInfo *t1, *t2;
 
     if (fold_const2_commutative(ctx, op) ||
         fold_xi_to_not(ctx, op, 0)) {
         return true;
     }
 
-    s_mask = arg_info(op->args[1])->s_mask
-           & arg_info(op->args[2])->s_mask;
-    return fold_masks_s(ctx, op, s_mask);
+    t1 = arg_info(op->args[1]);
+    t2 = arg_info(op->args[2]);
+
+    z_mask = ~(t1->o_mask | t2->o_mask);
+    o_mask = ~(t1->z_mask | t2->z_mask);
+    s_mask = t1->s_mask & t2->s_mask;
+
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 static bool fold_not(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 09/29] tcg/optimize: Build and use z_bits and o_bits in fold_not

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 29d1f29124..d22396f6d7 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2251,10 +2251,14 @@ static bool fold_nor(OptContext *ctx, TCGOp *op)
 
 static bool fold_not(OptContext *ctx, TCGOp *op)
 {
+    TempOptInfo *t1;
+
     if (fold_const1(ctx, op)) {
         return true;
     }
-    return fold_masks_s(ctx, op, arg_info(op->args[1])->s_mask);
+
+    t1 = arg_info(op->args[1]);
+    return fold_masks_zos(ctx, op, ~t1->o_mask, ~t1->z_mask, t1->s_mask);
 }
 
 static bool fold_or(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 10/29] tcg/optimize: Build and use one and affected bits in fold_or

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index d22396f6d7..ce065d0e22 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2263,7 +2263,7 @@ static bool fold_not(OptContext *ctx, TCGOp *op)
 
 static bool fold_or(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask, s_mask;
+    uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1, *t2;
 
     if (fold_const2_commutative(ctx, op) ||
@@ -2274,9 +2274,15 @@ static bool fold_or(OptContext *ctx, TCGOp *op)
 
     t1 = arg_info(op->args[1]);
     t2 = arg_info(op->args[2]);
+
     z_mask = t1->z_mask | t2->z_mask;
+    o_mask = t1->o_mask | t2->o_mask;
     s_mask = t1->s_mask & t2->s_mask;
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+
+    /* Affected bits are those not known one, masked by those known zero. */
+    a_mask = ~t1->o_mask & t2->z_mask;
+
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, a_mask);
 }
 
 static bool fold_orc(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 11/29] tcg/optimize: Build and use zero, one and affected bits in fold_orc

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index ce065d0e22..795f1c900e 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2287,7 +2287,7 @@ static bool fold_or(OptContext *ctx, TCGOp *op)
 
 static bool fold_orc(OptContext *ctx, TCGOp *op)
 {
-    uint64_t s_mask;
+    uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1, *t2;
 
     if (fold_const2(ctx, op) ||
@@ -2318,8 +2318,15 @@ static bool fold_orc(OptContext *ctx, TCGOp *op)
     }
 
     t1 = arg_info(op->args[1]);
+
+    z_mask = t1->z_mask | ~t2->o_mask;
+    o_mask = t1->o_mask | ~t2->z_mask;
     s_mask = t1->s_mask & t2->s_mask;
-    return fold_masks_s(ctx, op, s_mask);
+
+    /* Affected bits are those not known one, masked by those known one. */
+    a_mask = ~t1->o_mask & t2->o_mask;
+
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, a_mask);
 }
 
 static bool fold_qemu_ld_1reg(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 12/29] tcg/optimize: Build and use o_bits in fold_xor

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 795f1c900e..572d314578 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -3039,7 +3039,7 @@ static bool fold_tcg_st_memcopy(OptContext *ctx, TCGOp *op)
 
 static bool fold_xor(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask, s_mask;
+    uint64_t z_mask, o_mask, s_mask;
     TempOptInfo *t1, *t2;
 
     if (fold_const2_commutative(ctx, op) ||
@@ -3051,9 +3051,12 @@ static bool fold_xor(OptContext *ctx, TCGOp *op)
 
     t1 = arg_info(op->args[1]);
     t2 = arg_info(op->args[2]);
-    z_mask = t1->z_mask | t2->z_mask;
+
+    z_mask = (t1->z_mask | t2->z_mask) & ~(t1->o_mask & t2->o_mask);
+    o_mask = (t1->o_mask & ~t2->z_mask) | (t2->o_mask & ~t1->z_mask);
     s_mask = t1->s_mask & t2->s_mask;
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 /* Propagate constants and copies, fold constant expressions. */
-- 
2.43.0

[PULL 13/29] tcg/optimize: Build and use o_bits in fold_bswap

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 49 ++++++++++++++++++++++++-------------------------
 1 file changed, 24 insertions(+), 25 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 572d314578..c9c53f796f 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1671,53 +1671,52 @@ static bool fold_brcond2(OptContext *ctx, TCGOp *op)
 
 static bool fold_bswap(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask, s_mask, sign;
+    uint64_t z_mask, o_mask, s_mask;
     TempOptInfo *t1 = arg_info(op->args[1]);
+    int flags = op->args[2];
 
     if (ti_is_const(t1)) {
         return tcg_opt_gen_movi(ctx, op, op->args[0],
                                 do_constant_folding(op->opc, ctx->type,
-                                                    ti_const_val(t1),
-                                                    op->args[2]));
+                                                    ti_const_val(t1), flags));
     }
 
     z_mask = t1->z_mask;
+    o_mask = t1->o_mask;
+    s_mask = 0;
+
     switch (op->opc) {
     case INDEX_op_bswap16:
         z_mask = bswap16(z_mask);
-        sign = INT16_MIN;
+        o_mask = bswap16(o_mask);
+        if (flags & TCG_BSWAP_OS) {
+            z_mask = (int16_t)z_mask;
+            o_mask = (int16_t)o_mask;
+            s_mask = INT16_MIN;
+        } else if (!(flags & TCG_BSWAP_OZ)) {
+            z_mask |= MAKE_64BIT_MASK(16, 48);
+        }
         break;
     case INDEX_op_bswap32:
         z_mask = bswap32(z_mask);
-        sign = INT32_MIN;
+        o_mask = bswap32(o_mask);
+        if (flags & TCG_BSWAP_OS) {
+            z_mask = (int32_t)z_mask;
+            o_mask = (int32_t)o_mask;
+            s_mask = INT32_MIN;
+        } else if (!(flags & TCG_BSWAP_OZ)) {
+            z_mask |= MAKE_64BIT_MASK(32, 32);
+        }
         break;
     case INDEX_op_bswap64:
         z_mask = bswap64(z_mask);
-        sign = INT64_MIN;
+        o_mask = bswap64(o_mask);
         break;
     default:
         g_assert_not_reached();
     }
 
-    s_mask = 0;
-    switch (op->args[2] & (TCG_BSWAP_OZ | TCG_BSWAP_OS)) {
-    case TCG_BSWAP_OZ:
-        break;
-    case TCG_BSWAP_OS:
-        /* If the sign bit may be 1, force all the bits above to 1. */
-        if (z_mask & sign) {
-            z_mask |= sign;
-        }
-        /* The value and therefore s_mask is explicitly sign-extended. */
-        s_mask = sign;
-        break;
-    default:
-        /* The high bits are undefined: force all bits above the sign to 1. */
-        z_mask |= sign << 1;
-        break;
-    }
-
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 static bool fold_call(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 14/29] tcg/optimize: Build and use o_bits in fold_deposit

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index c9c53f796f..043568a10d 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1847,7 +1847,7 @@ static bool fold_deposit(OptContext *ctx, TCGOp *op)
     int ofs = op->args[3];
     int len = op->args[4];
     int width = 8 * tcg_type_size(ctx->type);
-    uint64_t z_mask, s_mask;
+    uint64_t z_mask, o_mask, s_mask;
 
     if (ti_is_const(t1) && ti_is_const(t2)) {
         return tcg_opt_gen_movi(ctx, op, op->args[0],
@@ -1882,7 +1882,9 @@ static bool fold_deposit(OptContext *ctx, TCGOp *op)
     }
 
     z_mask = deposit64(t1->z_mask, ofs, len, t2->z_mask);
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+    o_mask = deposit64(t1->o_mask, ofs, len, t2->o_mask);
+
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 static bool fold_divide(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 15/29] tcg/optimize: Build and use o_bits in fold_extract

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 043568a10d..f5fc0cfff9 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1963,7 +1963,7 @@ static bool fold_eqv(OptContext *ctx, TCGOp *op)
 
 static bool fold_extract(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask_old, z_mask;
+    uint64_t z_mask, o_mask, a_mask;
     TempOptInfo *t1 = arg_info(op->args[1]);
     int pos = op->args[2];
     int len = op->args[3];
@@ -1973,13 +1973,11 @@ static bool fold_extract(OptContext *ctx, TCGOp *op)
                                 extract64(ti_const_val(t1), pos, len));
     }
 
-    z_mask_old = t1->z_mask;
-    z_mask = extract64(z_mask_old, pos, len);
-    if (pos == 0 && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) {
-        return true;
-    }
+    z_mask = extract64(t1->z_mask, pos, len);
+    o_mask = extract64(t1->o_mask, pos, len);
+    a_mask = pos ? -1 : t1->z_mask ^ z_mask;
 
-    return fold_masks_z(ctx, op, z_mask);
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, 0, a_mask);
 }
 
 static bool fold_extract2(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 16/29] tcg/optimize: Build and use z_bits and o_bits in fold_extract2

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 38 +++++++++++++++++++++++++-------------
 1 file changed, 25 insertions(+), 13 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index f5fc0cfff9..86d958267a 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1099,6 +1099,12 @@ static bool fold_masks_zos(OptContext *ctx, TCGOp *op,
     return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, -1);
 }
 
+static bool fold_masks_zo(OptContext *ctx, TCGOp *op,
+                          uint64_t z_mask, uint64_t o_mask)
+{
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, 0, -1);
+}
+
 static bool fold_masks_zs(OptContext *ctx, TCGOp *op,
                           uint64_t z_mask, uint64_t s_mask)
 {
@@ -1982,21 +1988,27 @@ static bool fold_extract(OptContext *ctx, TCGOp *op)
 
 static bool fold_extract2(OptContext *ctx, TCGOp *op)
 {
-    if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) {
-        uint64_t v1 = arg_const_val(op->args[1]);
-        uint64_t v2 = arg_const_val(op->args[2]);
-        int shr = op->args[3];
+    TempOptInfo *t1 = arg_info(op->args[1]);
+    TempOptInfo *t2 = arg_info(op->args[2]);
+    uint64_t z1 = t1->z_mask;
+    uint64_t z2 = t2->z_mask;
+    uint64_t o1 = t1->o_mask;
+    uint64_t o2 = t2->o_mask;
+    int shr = op->args[3];
 
-        if (ctx->type == TCG_TYPE_I32) {
-            v1 = (uint32_t)v1 >> shr;
-            v2 = (uint64_t)((int32_t)v2 << (32 - shr));
-        } else {
-            v1 >>= shr;
-            v2 <<= 64 - shr;
-        }
-        return tcg_opt_gen_movi(ctx, op, op->args[0], v1 | v2);
+    if (ctx->type == TCG_TYPE_I32) {
+        z1 = (uint32_t)z1 >> shr;
+        o1 = (uint32_t)o1 >> shr;
+        z2 = (uint64_t)((int32_t)z2 << (32 - shr));
+        o2 = (uint64_t)((int32_t)o2 << (32 - shr));
+    } else {
+        z1 >>= shr;
+        o1 >>= shr;
+        z2 <<= 64 - shr;
+        o2 <<= 64 - shr;
     }
-    return finish_folding(ctx, op);
+
+    return fold_masks_zo(ctx, op, z1 | z2, o1 | o2);
 }
 
 static bool fold_exts(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 17/29] tcg/optimize: Build and use o_bits in fold_exts

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 86d958267a..103c94b12f 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2013,7 +2013,7 @@ static bool fold_extract2(OptContext *ctx, TCGOp *op)
 
 static bool fold_exts(OptContext *ctx, TCGOp *op)
 {
-    uint64_t s_mask, z_mask;
+    uint64_t z_mask, o_mask, s_mask;
     TempOptInfo *t1;
 
     if (fold_const1(ctx, op)) {
@@ -2022,17 +2022,19 @@ static bool fold_exts(OptContext *ctx, TCGOp *op)
 
     t1 = arg_info(op->args[1]);
     z_mask = t1->z_mask;
+    o_mask = t1->o_mask;
     s_mask = t1->s_mask;
 
     switch (op->opc) {
     case INDEX_op_ext_i32_i64:
         s_mask |= INT32_MIN;
         z_mask = (int32_t)z_mask;
+        o_mask = (int32_t)o_mask;
         break;
     default:
         g_assert_not_reached();
     }
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 static bool fold_extu(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 18/29] tcg/optimize: Build and use o_bits in fold_extu

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 103c94b12f..42d5ee23c0 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2039,25 +2039,31 @@ static bool fold_exts(OptContext *ctx, TCGOp *op)
 
 static bool fold_extu(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask;
+    uint64_t z_mask, o_mask;
+    TempOptInfo *t1;
 
     if (fold_const1(ctx, op)) {
         return true;
     }
 
-    z_mask = arg_info(op->args[1])->z_mask;
+    t1 = arg_info(op->args[1]);
+    z_mask = t1->z_mask;
+    o_mask = t1->o_mask;
+
     switch (op->opc) {
     case INDEX_op_extrl_i64_i32:
     case INDEX_op_extu_i32_i64:
         z_mask = (uint32_t)z_mask;
+        o_mask = (uint32_t)o_mask;
         break;
     case INDEX_op_extrh_i64_i32:
         z_mask >>= 32;
+        o_mask >>= 32;
         break;
     default:
         g_assert_not_reached();
     }
-    return fold_masks_z(ctx, op, z_mask);
+    return fold_masks_zo(ctx, op, z_mask, o_mask);
 }
 
 static bool fold_mb(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 19/29] tcg/optimize: Build and use o_bits in fold_movcond

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 42d5ee23c0..abcbee9111 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2097,7 +2097,7 @@ static bool fold_mov(OptContext *ctx, TCGOp *op)
 
 static bool fold_movcond(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask, s_mask;
+    uint64_t z_mask, o_mask, s_mask;
     TempOptInfo *tt, *ft;
     int i;
 
@@ -2123,6 +2123,7 @@ static bool fold_movcond(OptContext *ctx, TCGOp *op)
     tt = arg_info(op->args[3]);
     ft = arg_info(op->args[4]);
     z_mask = tt->z_mask | ft->z_mask;
+    o_mask = tt->o_mask & ft->o_mask;
     s_mask = tt->s_mask & ft->s_mask;
 
     if (ti_is_const(tt) && ti_is_const(ft)) {
@@ -2145,7 +2146,7 @@ static bool fold_movcond(OptContext *ctx, TCGOp *op)
         }
     }
 
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+    return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
 }
 
 static bool fold_mul(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 20/29] tcg/optimize: Build and use o_bits in fold_sextract

[Richard Henderson]

This was the last use of fold_affected_mask,
now fully replaced by fold_masks_zosa.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 30 ++++++------------------------
 1 file changed, 6 insertions(+), 24 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index abcbee9111..673849f07a 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1121,22 +1121,6 @@ static bool fold_masks_s(OptContext *ctx, TCGOp *op, uint64_t s_mask)
     return fold_masks_zosa(ctx, op, -1, 0, s_mask, -1);
 }
 
-/*
- * An "affected" mask bit is 0 if and only if the result is identical
- * to the first input.  Thus if the entire mask is 0, the operation
- * is equivalent to a copy.
- */
-static bool fold_affected_mask(OptContext *ctx, TCGOp *op, uint64_t a_mask)
-{
-    if (ctx->type == TCG_TYPE_I32) {
-        a_mask = (uint32_t)a_mask;
-    }
-    if (a_mask == 0) {
-        return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]);
-    }
-    return false;
-}
-
 /*
  * Convert @op to NOT, if NOT is supported by the host.
  * Return true f the conversion is successful, which will still
@@ -2669,7 +2653,7 @@ static bool fold_setcond2(OptContext *ctx, TCGOp *op)
 
 static bool fold_sextract(OptContext *ctx, TCGOp *op)
 {
-    uint64_t z_mask, s_mask, s_mask_old;
+    uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1 = arg_info(op->args[1]);
     int pos = op->args[2];
     int len = op->args[3];
@@ -2679,16 +2663,14 @@ static bool fold_sextract(OptContext *ctx, TCGOp *op)
                                 sextract64(ti_const_val(t1), pos, len));
     }
 
-    s_mask_old = t1->s_mask;
-    s_mask = s_mask_old >> pos;
+    s_mask = t1->s_mask >> pos;
     s_mask |= -1ull << (len - 1);
-
-    if (pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) {
-        return true;
-    }
+    a_mask = pos ? -1 : s_mask & ~t1->s_mask;
 
     z_mask = sextract64(t1->z_mask, pos, len);
-    return fold_masks_zs(ctx, op, z_mask, s_mask);
+    o_mask = sextract64(t1->o_mask, pos, len);
+
+    return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, a_mask);
 }
 
 static bool fold_shift(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 21/29] tcg/optimize: Build and use o_bits in fold_shift

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 673849f07a..0b441bc611 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2675,7 +2675,7 @@ static bool fold_sextract(OptContext *ctx, TCGOp *op)
 
 static bool fold_shift(OptContext *ctx, TCGOp *op)
 {
-    uint64_t s_mask, z_mask;
+    uint64_t s_mask, z_mask, o_mask;
     TempOptInfo *t1, *t2;
 
     if (fold_const2(ctx, op) ||
@@ -2688,14 +2688,16 @@ static bool fold_shift(OptContext *ctx, TCGOp *op)
     t2 = arg_info(op->args[2]);
     s_mask = t1->s_mask;
     z_mask = t1->z_mask;
+    o_mask = t1->o_mask;
 
     if (ti_is_const(t2)) {
         int sh = ti_const_val(t2);
 
         z_mask = do_constant_folding(op->opc, ctx->type, z_mask, sh);
+        o_mask = do_constant_folding(op->opc, ctx->type, o_mask, sh);
         s_mask = do_constant_folding(op->opc, ctx->type, s_mask, sh);
 
-        return fold_masks_zs(ctx, op, z_mask, s_mask);
+        return fold_masks_zos(ctx, op, z_mask, o_mask, s_mask);
     }
 
     switch (op->opc) {
-- 
2.43.0

[PULL 22/29] tcg/optimize: Use fold_and in do_constant_folding_cond[12]

[Richard Henderson]

When lowering tst comparisons, completely fold the and
opcode that we generate.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 0b441bc611..aa64f15779 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -784,6 +784,7 @@ static bool swap_commutative2(TCGArg *p1, TCGArg *p2)
  * Return -1 if the condition can't be simplified,
  * and the result of the condition (0 or 1) if it can.
  */
+static bool fold_and(OptContext *ctx, TCGOp *op);
 static int do_constant_folding_cond1(OptContext *ctx, TCGOp *op, TCGArg dest,
                                      TCGArg *p1, TCGArg *p2, TCGArg *pcond)
 {
@@ -834,6 +835,7 @@ static int do_constant_folding_cond1(OptContext *ctx, TCGOp *op, TCGArg dest,
         op2->args[0] = tmp;
         op2->args[1] = *p1;
         op2->args[2] = *p2;
+        fold_and(ctx, op2);
 
         *p1 = tmp;
         *p2 = arg_new_constant(ctx, 0);
@@ -929,9 +931,12 @@ static int do_constant_folding_cond2(OptContext *ctx, TCGOp *op, TCGArg *args)
         op1->args[0] = t1;
         op1->args[1] = al;
         op1->args[2] = bl;
+        fold_and(ctx, op1);
+
         op2->args[0] = t2;
         op2->args[1] = ah;
         op2->args[2] = bh;
+        fold_and(ctx, op1);
 
         args[0] = t1;
         args[1] = t2;
-- 
2.43.0

[PULL 23/29] tcg/optimize: Fold and to extract during optimize

[Richard Henderson]

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 33 ++++++++++++++++++++++++++++++---
 1 file changed, 30 insertions(+), 3 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index aa64f15779..06ccf39d64 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1044,8 +1044,9 @@ static bool fold_const2_commutative(OptContext *ctx, TCGOp *op)
  * If z_mask allows, fold the output to constant zero.
  * The passed s_mask may be augmented by z_mask.
  */
-static bool fold_masks_zosa(OptContext *ctx, TCGOp *op, uint64_t z_mask,
-                            uint64_t o_mask, int64_t s_mask, uint64_t a_mask)
+static bool fold_masks_zosa_int(OptContext *ctx, TCGOp *op,
+                                uint64_t z_mask, uint64_t o_mask,
+                                int64_t s_mask, uint64_t a_mask)
 {
     const TCGOpDef *def = &tcg_op_defs[op->opc];
     TCGTemp *ts;
@@ -1095,6 +1096,13 @@ static bool fold_masks_zosa(OptContext *ctx, TCGOp *op, uint64_t z_mask,
     rep = MAX(rep - 1, 0);
     ti->s_mask = INT64_MIN >> rep;
 
+    return false;
+}
+
+static bool fold_masks_zosa(OptContext *ctx, TCGOp *op, uint64_t z_mask,
+                            uint64_t o_mask, int64_t s_mask, uint64_t a_mask)
+{
+    fold_masks_zosa_int(ctx, op, z_mask, o_mask, s_mask, -1);
     return true;
 }
 
@@ -1448,7 +1456,26 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
     /* Affected bits are those not known zero, masked by those known one. */
     a_mask = t1->z_mask & ~t2->o_mask;
 
-    return fold_masks_zosa(ctx, op, z_mask, o_mask, s_mask, a_mask);
+    if (!fold_masks_zosa_int(ctx, op, z_mask, o_mask, s_mask, a_mask)) {
+        if (ti_is_const(t2)) {
+            /*
+             * Canonicalize on extract, if valid.  This aids x86 with its
+             * 2 operand MOVZBL and 2 operand AND, selecting the TCGOpcode
+             * which does not require matching operands.  Other backends can
+             * trivially expand the extract to AND during code generation.
+             */
+            uint64_t val = ti_const_val(t2);
+            if (!(val & (val + 1))) {
+                unsigned len = ctz64(~val);
+                if (TCG_TARGET_extract_valid(ctx->type, 0, len)) {
+                    op->opc = INDEX_op_extract;
+                    op->args[2] = 0;
+                    op->args[3] = len;
+                }
+            }
+        }
+    }
+    return true;
 }
 
 static bool fold_andc(OptContext *ctx, TCGOp *op)
-- 
2.43.0

[PULL 24/29] tcg/optimize: Simplify fold_and constant checks

[Richard Henderson]

If operand 2 is constant, then the computation of z_mask
and a_mask will produce the same results as the explicit
checks via fold_xi_to_i and fold_xi_to_x.  Shift the call
of fold_xx_to_x down below the ti_is_const(t2) check.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 06ccf39d64..f3a2328fe4 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1434,10 +1434,7 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
     uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1, *t2;
 
-    if (fold_const2_commutative(ctx, op) ||
-        fold_xi_to_i(ctx, op, 0) ||
-        fold_xi_to_x(ctx, op, -1) ||
-        fold_xx_to_x(ctx, op)) {
+    if (fold_const2_commutative(ctx, op)) {
         return true;
     }
 
@@ -1473,6 +1470,8 @@ static bool fold_and(OptContext *ctx, TCGOp *op)
                     op->args[3] = len;
                 }
             }
+        } else {
+            fold_xx_to_x(ctx, op);
         }
     }
     return true;
-- 
2.43.0

[PULL 25/29] tcg/optimize: Simplify fold_andc constant checks

[Richard Henderson]

If operand 2 is constant, then the computation of z_mask and a_mask
will produce the same results as the explicit check via fold_xi_to_i.
Shift the calls of fold_xx_to_i and fold_ix_to_not down below the
i2->is_const check.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index f3a2328fe4..8d14a38f9d 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1482,10 +1482,7 @@ static bool fold_andc(OptContext *ctx, TCGOp *op)
     uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1, *t2;
 
-    if (fold_const2(ctx, op) ||
-        fold_xx_to_i(ctx, op, 0) ||
-        fold_xi_to_x(ctx, op, 0) ||
-        fold_ix_to_not(ctx, op, -1)) {
+    if (fold_const2(ctx, op)) {
         return true;
     }
 
@@ -1510,6 +1507,10 @@ static bool fold_andc(OptContext *ctx, TCGOp *op)
         op->args[2] = arg_new_constant(ctx, ~ti_const_val(t2));
         return fold_and(ctx, op);
     }
+    if (fold_xx_to_i(ctx, op, 0) ||
+        fold_ix_to_not(ctx, op, -1)) {
+        return true;
+    }
 
     z_mask = t1->z_mask & ~t2->o_mask;
     o_mask = t1->o_mask & ~t2->z_mask;
-- 
2.43.0

[PULL 26/29] tcg/optimize: Simplify fold_orc constant checks

[Richard Henderson]

If operand 2 is constant, then the computation of z_mask and a_mask
will produce the same results as the explicit check via fold_xi_to_i.
Shift the calls of fold_xx_to_i and fold_ix_to_not down below the
i2->is_const check.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index 8d14a38f9d..a48ddd9171 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -2326,10 +2326,7 @@ static bool fold_orc(OptContext *ctx, TCGOp *op)
     uint64_t z_mask, o_mask, s_mask, a_mask;
     TempOptInfo *t1, *t2;
 
-    if (fold_const2(ctx, op) ||
-        fold_xx_to_i(ctx, op, -1) ||
-        fold_xi_to_x(ctx, op, -1) ||
-        fold_ix_to_not(ctx, op, 0)) {
+    if (fold_const2(ctx, op)) {
         return true;
     }
 
@@ -2352,7 +2349,10 @@ static bool fold_orc(OptContext *ctx, TCGOp *op)
         op->args[2] = arg_new_constant(ctx, ~ti_const_val(t2));
         return fold_or(ctx, op);
     }
-
+    if (fold_xx_to_i(ctx, op, -1) ||
+        fold_ix_to_not(ctx, op, 0)) {
+        return true;
+    }
     t1 = arg_info(op->args[1]);
 
     z_mask = t1->z_mask | ~t2->o_mask;
-- 
2.43.0

[PULL 27/29] tcg/optimize: Simplify fold_eqv constant checks

[Richard Henderson]

Both cases are handled by fold_xor after conversion.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/optimize.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/tcg/optimize.c b/tcg/optimize.c
index a48ddd9171..62a128bc9b 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -1948,9 +1948,7 @@ static bool fold_eqv(OptContext *ctx, TCGOp *op)
     uint64_t z_mask, o_mask, s_mask;
     TempOptInfo *t1, *t2;
 
-    if (fold_const2_commutative(ctx, op) ||
-        fold_xi_to_x(ctx, op, -1) ||
-        fold_xi_to_not(ctx, op, 0)) {
+    if (fold_const2_commutative(ctx, op)) {
         return true;
     }
 
-- 
2.43.0

[PULL 28/29] tcg/riscv: Fix typo in tgen_extract

[Richard Henderson]

Fix the direction of the shift, introduced when converting
the codebase to TCGOutOp* and small tgen_* helpers.

Fixes: 5a4d034f3cb ("tcg: Convert extract to TCGOutOpExtract")
Reported-by: Andrea Bolognani <abologna@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Andrea Bolognani <abologna@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
---
 tcg/riscv/tcg-target.c.inc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc
index 1800fd5077..31b9f7d87a 100644
--- a/tcg/riscv/tcg-target.c.inc
+++ b/tcg/riscv/tcg-target.c.inc
@@ -2502,7 +2502,7 @@ static void tgen_extract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
         }
     }
     if (ofs + len == 32) {
-        tgen_shli(s, TCG_TYPE_I32, a0, a1, ofs);
+        tgen_shri(s, TCG_TYPE_I32, a0, a1, ofs);
         return;
     }
     if (len == 1) {
-- 
2.43.0

[PULL 29/29] tcg: Fix constant propagation in tcg_reg_alloc_dup

[Richard Henderson]

The scalar constant must be replicated for dup.

Cc: qemu-stable@nongnu.org
Fixes: bab1671f0fa ("tcg: Manually expand INDEX_op_dup_vec")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3002
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tcg/tcg.c b/tcg/tcg.c
index d714ae2889..50d40b9cbe 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -5154,7 +5154,7 @@ static void tcg_reg_alloc_dup(TCGContext *s, const TCGOp *op)
 
     if (its->val_type == TEMP_VAL_CONST) {
         /* Propagate constant via movi -> dupi.  */
-        tcg_target_ulong val = its->val;
+        tcg_target_ulong val = dup_const(vece, its->val);
         if (IS_DEAD_ARG(1)) {
             temp_dead(s, its);
         }
-- 
2.43.0

Re: [PULL 29/29] tcg: Fix constant propagation in tcg_reg_alloc_dup

[Michael Tokarev]

On 30.06.2025 18:28, Richard Henderson wrote:
> The scalar constant must be replicated for dup.
> 
> Cc: qemu-stable@nongnu.org
> Fixes: bab1671f0fa ("tcg: Manually expand INDEX_op_dup_vec")
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3002
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   tcg/tcg.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tcg/tcg.c b/tcg/tcg.c
> index d714ae2889..50d40b9cbe 100644
> --- a/tcg/tcg.c
> +++ b/tcg/tcg.c
> @@ -5154,7 +5154,7 @@ static void tcg_reg_alloc_dup(TCGContext *s, const TCGOp *op)
>   
>       if (its->val_type == TEMP_VAL_CONST) {
>           /* Propagate constant via movi -> dupi.  */
> -        tcg_target_ulong val = its->val;
> +        tcg_target_ulong val = dup_const(vece, its->val);
>           if (IS_DEAD_ARG(1)) {
>               temp_dead(s, its);
>           }

Hi Richard!

Does this apply to 7.2.x series, where temps had to be freed
explicitly?

Thanks,

/mjt

Re: [PULL 29/29] tcg: Fix constant propagation in tcg_reg_alloc_dup

[Richard Henderson]

On 7/1/25 13:04, Michael Tokarev wrote:
> On 30.06.2025 18:28, Richard Henderson wrote:
>> The scalar constant must be replicated for dup.
>>
>> Cc: qemu-stable@nongnu.org
>> Fixes: bab1671f0fa ("tcg: Manually expand INDEX_op_dup_vec")
>> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/3002
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>>   tcg/tcg.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/tcg/tcg.c b/tcg/tcg.c
>> index d714ae2889..50d40b9cbe 100644
>> --- a/tcg/tcg.c
>> +++ b/tcg/tcg.c
>> @@ -5154,7 +5154,7 @@ static void tcg_reg_alloc_dup(TCGContext *s, const TCGOp *op)
>>       if (its->val_type == TEMP_VAL_CONST) {
>>           /* Propagate constant via movi -> dupi.  */
>> -        tcg_target_ulong val = its->val;
>> +        tcg_target_ulong val = dup_const(vece, its->val);
>>           if (IS_DEAD_ARG(1)) {
>>               temp_dead(s, its);
>>           }
> 
> Hi Richard!
> 
> Does this apply to 7.2.x series, where temps had to be freed
> explicitly?

Yep.

r~

Re: [PULL 00/29] tcg patch queue

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 116 bytes --]

Applied, thanks.

Please update the changelog at https://wiki.qemu.org/ChangeLog/10.1 for any user-visible changes.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[PULL 00/29] tcg patch queue

[Richard Henderson]

The following changes since commit ec6f9f135d5e5596ab0258da2ddd048f1fd8c359:

  Merge tag 'migration-20231017-pull-request' of https://gitlab.com/juan.quintela/qemu into staging (2023-10-17 10:06:21 -0400)

are available in the Git repository at:

  https://gitlab.com/rth7680/qemu.git tags/pull-tcg-20231018

for you to fetch changes up to b540757b7f711eaf069f530916005cf8cfe7c00f:

  target/i386: Use i128 for 128 and 256-bit loads and stores (2023-10-18 12:33:08 -0700)

----------------------------------------------------------------
tcg: Drop unused tcg_temp_free define
tcg: Introduce tcg_use_softmmu
tcg: Optimize past conditional branches
tcg: Use constant zero when expanding with divu2
tcg/ppc: Enable direct branching tcg_out_goto_tb with TCG_REG_TB
tcg/ppc: Use ADDPCIS for power9
tcg/ppc: Use prefixed instructions for power10
tcg/ppc: Disable TCG_REG_TB for Power9/Power10

----------------------------------------------------------------
Jordan Niethe (1):
      tcg/ppc: Enable direct branching tcg_out_goto_tb with TCG_REG_TB

Mike Frysinger (1):
      tcg: drop unused tcg_temp_free define

Richard Henderson (27):
      tcg/ppc: Untabify tcg-target.c.inc
      tcg/ppc: Reinterpret tb-relative to TB+4
      tcg/ppc: Use ADDPCIS in tcg_out_tb_start
      tcg/ppc: Use ADDPCIS in tcg_out_movi_int
      tcg/ppc: Use ADDPCIS for the constant pool
      tcg/ppc: Use ADDPCIS in tcg_out_goto_tb
      tcg/ppc: Use PADDI in tcg_out_movi
      tcg/ppc: Use prefixed instructions in tcg_out_mem_long
      tcg/ppc: Use PLD in tcg_out_movi for constant pool
      tcg/ppc: Use prefixed instructions in tcg_out_dupi_vec
      tcg/ppc: Use PLD in tcg_out_goto_tb
      tcg/ppc: Disable TCG_REG_TB for Power9/Power10
      tcg: Introduce tcg_use_softmmu
      tcg: Provide guest_base fallback for system mode
      tcg/arm: Use tcg_use_softmmu
      tcg/aarch64: Use tcg_use_softmmu
      tcg/i386: Use tcg_use_softmmu
      tcg/loongarch64: Use tcg_use_softmmu
      tcg/mips: Use tcg_use_softmmu
      tcg/ppc: Use tcg_use_softmmu
      tcg/riscv: Do not reserve TCG_GUEST_BASE_REG for guest_base zero
      tcg/riscv: Use tcg_use_softmmu
      tcg/s390x: Use tcg_use_softmmu
      tcg: Use constant zero when expanding with divu2
      tcg: Optimize past conditional branches
      tcg: Add tcg_gen_{ld,st}_i128
      target/i386: Use i128 for 128 and 256-bit loads and stores

 include/tcg/tcg-op-common.h      |   3 +
 include/tcg/tcg-op.h             |   2 -
 include/tcg/tcg.h                |   8 +-
 target/i386/tcg/translate.c      |  63 ++---
 tcg/optimize.c                   |   8 +-
 tcg/tcg-op-ldst.c                |  14 +-
 tcg/tcg-op.c                     |  38 ++-
 tcg/tcg.c                        |  13 +-
 tcg/aarch64/tcg-target.c.inc     | 177 ++++++------
 tcg/arm/tcg-target.c.inc         | 203 +++++++-------
 tcg/i386/tcg-target.c.inc        | 184 +++++++------
 tcg/loongarch64/tcg-target.c.inc | 126 +++++----
 tcg/mips/tcg-target.c.inc        | 231 ++++++++--------
 tcg/ppc/tcg-target.c.inc         | 561 ++++++++++++++++++++++++++-------------
 tcg/riscv/tcg-target.c.inc       | 189 ++++++-------
 tcg/s390x/tcg-target.c.inc       | 161 ++++++-----
 16 files changed, 1093 insertions(+), 888 deletions(-)