Merge remote-tracking branch 'remotes/rth/tcg-next' into staging

* remotes/rth/tcg-next:
  TCG: Fix tcg_gen_extr_i64_tl for 32bit
  tcg: Remove TCG_TARGET_HAS_new_ldst
  tci: Convert to new ldst opcodes
  tcg-i386: Fix win64 qemu store

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

17 files changed, 237 insertions, 428 deletions

diff --git a/tcg/aarch64/tcg-target.h b/tcg/aarch64/tcg-target.h
index a32aea66fa..60c7493ac1 100644
--- a/tcg/aarch64/tcg-target.h
+++ b/tcg/aarch64/tcg-target.h
@@ -99,8 +99,6 @@ typedef enum {
 #define TCG_TARGET_HAS_muluh_i64        1
 #define TCG_TARGET_HAS_mulsh_i64        1
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
     __builtin___clear_cache((char *)start, (char *)stop);
diff --git a/tcg/arm/tcg-target.h b/tcg/arm/tcg-target.h
index 73f10c4424..1c719e2862 100644
--- a/tcg/arm/tcg-target.h
+++ b/tcg/arm/tcg-target.h
@@ -86,8 +86,6 @@ extern bool use_idiv_instructions;
 #define TCG_TARGET_HAS_div_i32          use_idiv_instructions
 #define TCG_TARGET_HAS_rem_i32          0
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 extern bool tcg_target_deposit_valid(int ofs, int len);
 #define TCG_TARGET_deposit_i32_valid  tcg_target_deposit_valid
 
diff --git a/tcg/i386/tcg-target.c b/tcg/i386/tcg-target.c
index d9102335f9..4133dcf1a9 100644
--- a/tcg/i386/tcg-target.c
+++ b/tcg/i386/tcg-target.c
@@ -1407,7 +1407,8 @@ static void tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
         } else {
             retaddr = TCG_REG_RAX;
             tcg_out_movi(s, TCG_TYPE_PTR, retaddr, (uintptr_t)l->raddr);
-            tcg_out_st(s, TCG_TYPE_PTR, retaddr, TCG_REG_ESP, 0);
+            tcg_out_st(s, TCG_TYPE_PTR, retaddr, TCG_REG_ESP,
+                       TCG_TARGET_CALL_STACK_OFFSET);
         }
     }
 
diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
index 6c94e5cde0..7a9980e70e 100644
--- a/tcg/i386/tcg-target.h
+++ b/tcg/i386/tcg-target.h
@@ -130,8 +130,6 @@ extern bool have_bmi1;
 #define TCG_TARGET_HAS_mulsh_i64        0
 #endif
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 #define TCG_TARGET_deposit_i32_valid(ofs, len) \
     (((ofs) == 0 && (len) == 8) || ((ofs) == 8 && (len) == 8) || \
      ((ofs) == 0 && (len) == 16))
diff --git a/tcg/ia64/tcg-target.h b/tcg/ia64/tcg-target.h
index 3a59b50349..d67558988a 100644
--- a/tcg/ia64/tcg-target.h
+++ b/tcg/ia64/tcg-target.h
@@ -160,8 +160,6 @@ typedef enum {
 #define TCG_TARGET_HAS_mulsh_i64        0
 #define TCG_TARGET_HAS_trunc_shr_i32    0
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 #define TCG_TARGET_deposit_i32_valid(ofs, len) ((len) <= 16)
 #define TCG_TARGET_deposit_i64_valid(ofs, len) ((len) <= 16)
 
diff --git a/tcg/mips/tcg-target.h b/tcg/mips/tcg-target.h
index b5face8b4d..c88a1c9272 100644
--- a/tcg/mips/tcg-target.h
+++ b/tcg/mips/tcg-target.h
@@ -118,8 +118,6 @@ extern bool use_mips32r2_instructions;
 #define TCG_TARGET_HAS_ext16s_i32       use_mips32r2_instructions
 #define TCG_TARGET_HAS_rot_i32          use_mips32r2_instructions
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 /* optional instructions automatically implemented */
 #define TCG_TARGET_HAS_neg_i32          0 /* sub  rd, zero, rt   */
 #define TCG_TARGET_HAS_ext8u_i32        0 /* andi rt, rs, 0xff   */
diff --git a/tcg/optimize.c b/tcg/optimize.c
index 77da2f942a..16cebbe16d 100644
--- a/tcg/optimize.c
+++ b/tcg/optimize.c
@@ -889,17 +889,12 @@ static TCGArg *tcg_constant_folding(TCGContext *s, uint16_t *tcg_opc_ptr,
             break;
 
         CASE_OP_32_64(ld8u):
-        case INDEX_op_qemu_ld8u:
             mask = 0xff;
             break;
         CASE_OP_32_64(ld16u):
-        case INDEX_op_qemu_ld16u:
             mask = 0xffff;
             break;
         case INDEX_op_ld32u_i64:
-#if TCG_TARGET_REG_BITS == 64
-        case INDEX_op_qemu_ld32u:
-#endif
             mask = 0xffffffffu;
             break;
 
diff --git a/tcg/ppc/tcg-target.h b/tcg/ppc/tcg-target.h
index dd7e557948..05069ae2d8 100644
--- a/tcg/ppc/tcg-target.h
+++ b/tcg/ppc/tcg-target.h
@@ -100,8 +100,6 @@ typedef enum {
 #define TCG_TARGET_HAS_muluh_i32        0
 #define TCG_TARGET_HAS_mulsh_i32        0
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 #define TCG_AREG0 TCG_REG_R27
 
 #define tcg_qemu_tb_exec(env, tb_ptr) \
diff --git a/tcg/ppc64/tcg-target.h b/tcg/ppc64/tcg-target.h
index 29f479a3cf..f2360c869a 100644
--- a/tcg/ppc64/tcg-target.h
+++ b/tcg/ppc64/tcg-target.h
@@ -124,8 +124,6 @@ typedef enum {
 #define TCG_TARGET_HAS_muluh_i64        1
 #define TCG_TARGET_HAS_mulsh_i64        1
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 #define TCG_AREG0 TCG_REG_R27
 
 #define TCG_TARGET_EXTEND_ARGS 1
diff --git a/tcg/s390/tcg-target.h b/tcg/s390/tcg-target.h
index ad2c6ddaf4..5acc28ca6b 100644
--- a/tcg/s390/tcg-target.h
+++ b/tcg/s390/tcg-target.h
@@ -100,8 +100,6 @@ typedef enum TCGReg {
 #define TCG_TARGET_HAS_muluh_i64        0
 #define TCG_TARGET_HAS_mulsh_i64        0
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 extern bool tcg_target_deposit_valid(int ofs, int len);
 #define TCG_TARGET_deposit_i32_valid  tcg_target_deposit_valid
 #define TCG_TARGET_deposit_i64_valid  tcg_target_deposit_valid
diff --git a/tcg/sparc/tcg-target.h b/tcg/sparc/tcg-target.h
index 473bfc71fd..089f9761ca 100644
--- a/tcg/sparc/tcg-target.h
+++ b/tcg/sparc/tcg-target.h
@@ -140,8 +140,6 @@ typedef enum {
 #define TCG_TARGET_HAS_muluh_i64        0
 #define TCG_TARGET_HAS_mulsh_i64        0
 
-#define TCG_TARGET_HAS_new_ldst         1
-
 #define TCG_AREG0 TCG_REG_I0
 
 static inline void flush_icache_range(uintptr_t start, uintptr_t stop)
diff --git a/tcg/tcg-op.h b/tcg/tcg-op.h
index 719533ac39..019dd9b6fb 100644
--- a/tcg/tcg-op.h
+++ b/tcg/tcg-op.h
@@ -2718,7 +2718,7 @@ static inline void tcg_gen_qemu_st64(TCGv_i64 arg, TCGv addr, int mem_index)
 #define tcg_gen_bswap16_tl tcg_gen_bswap16_i32
 #define tcg_gen_bswap32_tl tcg_gen_bswap32_i32
 #define tcg_gen_concat_tl_i64 tcg_gen_concat_i32_i64
-#define tcg_gen_extr_tl_i64 tcg_gen_extr_i32_i64
+#define tcg_gen_extr_i64_tl tcg_gen_extr_i64_i32
 #define tcg_gen_andc_tl tcg_gen_andc_i32
 #define tcg_gen_eqv_tl tcg_gen_eqv_i32
 #define tcg_gen_nand_tl tcg_gen_nand_i32
diff --git a/tcg/tcg-opc.h b/tcg/tcg-opc.h
index 71ba64ac16..042d442c7e 100644
--- a/tcg/tcg-opc.h
+++ b/tcg/tcg-opc.h
@@ -185,106 +185,20 @@ DEF(debug_insn_start, 0, 0, 1, TCG_OPF_NOT_PRESENT)
 DEF(exit_tb, 0, 0, 1, TCG_OPF_BB_END)
 DEF(goto_tb, 0, 0, 1, TCG_OPF_BB_END)
 
-#define IMPL_NEW_LDST \
-    (TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS \
-     | IMPL(TCG_TARGET_HAS_new_ldst))
-
-#if TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
-DEF(qemu_ld_i32, 1, 1, 2, IMPL_NEW_LDST)
-DEF(qemu_st_i32, 0, 2, 2, IMPL_NEW_LDST)
-# if TCG_TARGET_REG_BITS == 64
-DEF(qemu_ld_i64, 1, 1, 2, IMPL_NEW_LDST | TCG_OPF_64BIT)
-DEF(qemu_st_i64, 0, 2, 2, IMPL_NEW_LDST | TCG_OPF_64BIT)
-# else
-DEF(qemu_ld_i64, 2, 1, 2, IMPL_NEW_LDST | TCG_OPF_64BIT)
-DEF(qemu_st_i64, 0, 3, 2, IMPL_NEW_LDST | TCG_OPF_64BIT)
-# endif
-#else
-DEF(qemu_ld_i32, 1, 2, 2, IMPL_NEW_LDST)
-DEF(qemu_st_i32, 0, 3, 2, IMPL_NEW_LDST)
-DEF(qemu_ld_i64, 2, 2, 2, IMPL_NEW_LDST | TCG_OPF_64BIT)
-DEF(qemu_st_i64, 0, 4, 2, IMPL_NEW_LDST | TCG_OPF_64BIT)
-#endif
-
-#undef IMPL_NEW_LDST
-
-#define IMPL_OLD_LDST \
-    (TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS \
-     | IMPL(!TCG_TARGET_HAS_new_ldst))
-
-#if TCG_TARGET_REG_BITS == 32
-#if TARGET_LONG_BITS == 32
-DEF(qemu_ld8u, 1, 1, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_ld8u, 1, 2, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_ld8s, 1, 1, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_ld8s, 1, 2, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_ld16u, 1, 1, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_ld16u, 1, 2, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_ld16s, 1, 1, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_ld16s, 1, 2, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_ld32, 1, 1, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_ld32, 1, 2, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_ld64, 2, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-#else
-DEF(qemu_ld64, 2, 2, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-#endif
-
-#if TARGET_LONG_BITS == 32
-DEF(qemu_st8, 0, 2, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_st8, 0, 3, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_st16, 0, 2, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_st16, 0, 3, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_st32, 0, 2, 1, IMPL_OLD_LDST)
-#else
-DEF(qemu_st32, 0, 3, 1, IMPL_OLD_LDST)
-#endif
-#if TARGET_LONG_BITS == 32
-DEF(qemu_st64, 0, 3, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-#else
-DEF(qemu_st64, 0, 4, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-#endif
-
-#else /* TCG_TARGET_REG_BITS == 32 */
-
-DEF(qemu_ld8u, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_ld8s, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_ld16u, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_ld16s, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_ld32, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_ld32u, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_ld32s, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_ld64, 1, 1, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-
-DEF(qemu_st8, 0, 2, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_st16, 0, 2, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_st32, 0, 2, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-DEF(qemu_st64, 0, 2, 1, IMPL_OLD_LDST | TCG_OPF_64BIT)
-
-#endif /* TCG_TARGET_REG_BITS != 32 */
-
-#undef IMPL_OLD_LDST
-
+#define TLADDR_ARGS    (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? 1 : 2)
+#define DATA64_ARGS  (TCG_TARGET_REG_BITS == 64 ? 1 : 2)
+
+DEF(qemu_ld_i32, 1, TLADDR_ARGS, 2,
+    TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS)
+DEF(qemu_st_i32, 0, TLADDR_ARGS + 1, 2,
+    TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS)
+DEF(qemu_ld_i64, DATA64_ARGS, TLADDR_ARGS, 2,
+    TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS | TCG_OPF_64BIT)
+DEF(qemu_st_i64, 0, TLADDR_ARGS + DATA64_ARGS, 2,
+    TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS | TCG_OPF_64BIT)
+
+#undef TLADDR_ARGS
+#undef DATA64_ARGS
 #undef IMPL
 #undef IMPL64
 #undef DEF
diff --git a/tcg/tcg.c b/tcg/tcg.c
index 2c5732da17..dae4b7cb00 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -941,97 +941,26 @@ static inline TCGMemOp tcg_canonicalize_memop(TCGMemOp op, bool is64, bool st)
     return op;
 }
 
-static const TCGOpcode old_ld_opc[8] = {
-    [MO_UB] = INDEX_op_qemu_ld8u,
-    [MO_SB] = INDEX_op_qemu_ld8s,
-    [MO_UW] = INDEX_op_qemu_ld16u,
-    [MO_SW] = INDEX_op_qemu_ld16s,
-#if TCG_TARGET_REG_BITS == 32
-    [MO_UL] = INDEX_op_qemu_ld32,
-    [MO_SL] = INDEX_op_qemu_ld32,
-#else
-    [MO_UL] = INDEX_op_qemu_ld32u,
-    [MO_SL] = INDEX_op_qemu_ld32s,
-#endif
-    [MO_Q]  = INDEX_op_qemu_ld64,
-};
-
-static const TCGOpcode old_st_opc[4] = {
-    [MO_UB] = INDEX_op_qemu_st8,
-    [MO_UW] = INDEX_op_qemu_st16,
-    [MO_UL] = INDEX_op_qemu_st32,
-    [MO_Q]  = INDEX_op_qemu_st64,
-};
-
 void tcg_gen_qemu_ld_i32(TCGv_i32 val, TCGv addr, TCGArg idx, TCGMemOp memop)
 {
     memop = tcg_canonicalize_memop(memop, 0, 0);
 
-    if (TCG_TARGET_HAS_new_ldst) {
-        *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_ld_i32;
-        tcg_add_param_i32(val);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = memop;
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-        return;
-    }
-
-    /* The old opcodes only support target-endian memory operations.  */
-    assert((memop & MO_BSWAP) == MO_TE || (memop & MO_SIZE) == MO_8);
-    assert(old_ld_opc[memop & MO_SSIZE] != 0);
-
-    if (TCG_TARGET_REG_BITS == 32) {
-        *tcg_ctx.gen_opc_ptr++ = old_ld_opc[memop & MO_SSIZE];
-        tcg_add_param_i32(val);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-    } else {
-        TCGv_i64 val64 = tcg_temp_new_i64();
-
-        *tcg_ctx.gen_opc_ptr++ = old_ld_opc[memop & MO_SSIZE];
-        tcg_add_param_i64(val64);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-
-        tcg_gen_trunc_i64_i32(val, val64);
-        tcg_temp_free_i64(val64);
-    }
+    *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_ld_i32;
+    tcg_add_param_i32(val);
+    tcg_add_param_tl(addr);
+    *tcg_ctx.gen_opparam_ptr++ = memop;
+    *tcg_ctx.gen_opparam_ptr++ = idx;
 }
 
 void tcg_gen_qemu_st_i32(TCGv_i32 val, TCGv addr, TCGArg idx, TCGMemOp memop)
 {
     memop = tcg_canonicalize_memop(memop, 0, 1);
 
-    if (TCG_TARGET_HAS_new_ldst) {
-        *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_st_i32;
-        tcg_add_param_i32(val);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = memop;
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-        return;
-    }
-
-    /* The old opcodes only support target-endian memory operations.  */
-    assert((memop & MO_BSWAP) == MO_TE || (memop & MO_SIZE) == MO_8);
-    assert(old_st_opc[memop & MO_SIZE] != 0);
-
-    if (TCG_TARGET_REG_BITS == 32) {
-        *tcg_ctx.gen_opc_ptr++ = old_st_opc[memop & MO_SIZE];
-        tcg_add_param_i32(val);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-    } else {
-        TCGv_i64 val64 = tcg_temp_new_i64();
-
-        tcg_gen_extu_i32_i64(val64, val);
-
-        *tcg_ctx.gen_opc_ptr++ = old_st_opc[memop & MO_SIZE];
-        tcg_add_param_i64(val64);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-
-        tcg_temp_free_i64(val64);
-    }
+    *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_st_i32;
+    tcg_add_param_i32(val);
+    tcg_add_param_tl(addr);
+    *tcg_ctx.gen_opparam_ptr++ = memop;
+    *tcg_ctx.gen_opparam_ptr++ = idx;
 }
 
 void tcg_gen_qemu_ld_i64(TCGv_i64 val, TCGv addr, TCGArg idx, TCGMemOp memop)
@@ -1050,22 +979,10 @@ void tcg_gen_qemu_ld_i64(TCGv_i64 val, TCGv addr, TCGArg idx, TCGMemOp memop)
     }
 #endif
 
-    if (TCG_TARGET_HAS_new_ldst) {
-        *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_ld_i64;
-        tcg_add_param_i64(val);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = memop;
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-        return;
-    }
-
-    /* The old opcodes only support target-endian memory operations.  */
-    assert((memop & MO_BSWAP) == MO_TE || (memop & MO_SIZE) == MO_8);
-    assert(old_ld_opc[memop & MO_SSIZE] != 0);
-
-    *tcg_ctx.gen_opc_ptr++ = old_ld_opc[memop & MO_SSIZE];
+    *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_ld_i64;
     tcg_add_param_i64(val);
     tcg_add_param_tl(addr);
+    *tcg_ctx.gen_opparam_ptr++ = memop;
     *tcg_ctx.gen_opparam_ptr++ = idx;
 }
 
@@ -1080,22 +997,10 @@ void tcg_gen_qemu_st_i64(TCGv_i64 val, TCGv addr, TCGArg idx, TCGMemOp memop)
     }
 #endif
 
-    if (TCG_TARGET_HAS_new_ldst) {
-        *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_st_i64;
-        tcg_add_param_i64(val);
-        tcg_add_param_tl(addr);
-        *tcg_ctx.gen_opparam_ptr++ = memop;
-        *tcg_ctx.gen_opparam_ptr++ = idx;
-        return;
-    }
-
-    /* The old opcodes only support target-endian memory operations.  */
-    assert((memop & MO_BSWAP) == MO_TE || (memop & MO_SIZE) == MO_8);
-    assert(old_st_opc[memop & MO_SIZE] != 0);
-
-    *tcg_ctx.gen_opc_ptr++ = old_st_opc[memop & MO_SIZE];
+    *tcg_ctx.gen_opc_ptr++ = INDEX_op_qemu_st_i64;
     tcg_add_param_i64(val);
     tcg_add_param_tl(addr);
+    *tcg_ctx.gen_opparam_ptr++ = memop;
     *tcg_ctx.gen_opparam_ptr++ = idx;
 }
 
diff --git a/tcg/tci/tcg-target.c b/tcg/tci/tcg-target.c
index 375e590d2b..03a7b46958 100644
--- a/tcg/tci/tcg-target.c
+++ b/tcg/tci/tcg-target.c
@@ -227,21 +227,11 @@ static const TCGTargetOpDef tcg_target_op_defs[] = {
 #endif
 #endif /* TCG_TARGET_REG_BITS == 64 */
 
-    { INDEX_op_qemu_ld8u, { R, L } },
-    { INDEX_op_qemu_ld8s, { R, L } },
-    { INDEX_op_qemu_ld16u, { R, L } },
-    { INDEX_op_qemu_ld16s, { R, L } },
-    { INDEX_op_qemu_ld32, { R, L } },
-#if TCG_TARGET_REG_BITS == 64
-    { INDEX_op_qemu_ld32u, { R, L } },
-    { INDEX_op_qemu_ld32s, { R, L } },
-#endif
-    { INDEX_op_qemu_ld64, { R64, L } },
+    { INDEX_op_qemu_ld_i32, { R, L } },
+    { INDEX_op_qemu_ld_i64, { R64, L } },
 
-    { INDEX_op_qemu_st8, { R, S } },
-    { INDEX_op_qemu_st16, { R, S } },
-    { INDEX_op_qemu_st32, { R, S } },
-    { INDEX_op_qemu_st64, { R64, S } },
+    { INDEX_op_qemu_st_i32, { R, S } },
+    { INDEX_op_qemu_st_i64, { R64, S } },
 
 #if TCG_TARGET_HAS_ext8s_i32
     { INDEX_op_ext8s_i32, { R, R } },
@@ -767,58 +757,52 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
         tcg_out8(s, args[2]);           /* condition */
         tci_out_label(s, args[3]);
         break;
-    case INDEX_op_qemu_ld8u:
-    case INDEX_op_qemu_ld8s:
-    case INDEX_op_qemu_ld16u:
-    case INDEX_op_qemu_ld16s:
-    case INDEX_op_qemu_ld32:
-#if TCG_TARGET_REG_BITS == 64
-    case INDEX_op_qemu_ld32s:
-    case INDEX_op_qemu_ld32u:
-#endif
+    case INDEX_op_qemu_ld_i32:
         tcg_out_r(s, *args++);
         tcg_out_r(s, *args++);
-#if TARGET_LONG_BITS > TCG_TARGET_REG_BITS
-        tcg_out_r(s, *args++);
-#endif
+        if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
+            tcg_out_r(s, *args++);
+        }
+        tcg_out_i(s, *args++);
 #ifdef CONFIG_SOFTMMU
         tcg_out_i(s, *args);
 #endif
         break;
-    case INDEX_op_qemu_ld64:
-        tcg_out_r(s, *args++);
-#if TCG_TARGET_REG_BITS == 32
+    case INDEX_op_qemu_ld_i64:
         tcg_out_r(s, *args++);
-#endif
-        tcg_out_r(s, *args++);
-#if TARGET_LONG_BITS > TCG_TARGET_REG_BITS
+        if (TCG_TARGET_REG_BITS == 32) {
+            tcg_out_r(s, *args++);
+        }
         tcg_out_r(s, *args++);
-#endif
+        if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
+            tcg_out_r(s, *args++);
+        }
+        tcg_out_i(s, *args++);
 #ifdef CONFIG_SOFTMMU
         tcg_out_i(s, *args);
 #endif
         break;
-    case INDEX_op_qemu_st8:
-    case INDEX_op_qemu_st16:
-    case INDEX_op_qemu_st32:
+    case INDEX_op_qemu_st_i32:
         tcg_out_r(s, *args++);
         tcg_out_r(s, *args++);
-#if TARGET_LONG_BITS > TCG_TARGET_REG_BITS
-        tcg_out_r(s, *args++);
-#endif
+        if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
+            tcg_out_r(s, *args++);
+        }
+        tcg_out_i(s, *args++);
 #ifdef CONFIG_SOFTMMU
         tcg_out_i(s, *args);
 #endif
         break;
-    case INDEX_op_qemu_st64:
-        tcg_out_r(s, *args++);
-#if TCG_TARGET_REG_BITS == 32
-        tcg_out_r(s, *args++);
-#endif
+    case INDEX_op_qemu_st_i64:
         tcg_out_r(s, *args++);
-#if TARGET_LONG_BITS > TCG_TARGET_REG_BITS
+        if (TCG_TARGET_REG_BITS == 32) {
+            tcg_out_r(s, *args++);
+        }
         tcg_out_r(s, *args++);
-#endif
+        if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
+            tcg_out_r(s, *args++);
+        }
+        tcg_out_i(s, *args++);
 #ifdef CONFIG_SOFTMMU
         tcg_out_i(s, *args);
 #endif
diff --git a/tcg/tci/tcg-target.h b/tcg/tci/tcg-target.h
index 0be5acdb81..bd1e97468c 100644
--- a/tcg/tci/tcg-target.h
+++ b/tcg/tci/tcg-target.h
@@ -118,8 +118,6 @@
 #define TCG_TARGET_HAS_mulu2_i32        1
 #endif /* TCG_TARGET_REG_BITS == 64 */
 
-#define TCG_TARGET_HAS_new_ldst         0
-
 /* Number of registers available.
    For 32 bit hosts, we need more than 8 registers (call arguments). */
 /* #define TCG_TARGET_NB_REGS 8 */
diff --git a/tci.c b/tci.c
index 6523ab82f4..a744760266 100644
--- a/tci.c
+++ b/tci.c
@@ -116,16 +116,6 @@ static void tci_write_reg(TCGReg index, tcg_target_ulong value)
     tci_reg[index] = value;
 }
 
-static void tci_write_reg8s(TCGReg index, int8_t value)
-{
-    tci_write_reg(index, value);
-}
-
-static void tci_write_reg16s(TCGReg index, int16_t value)
-{
-    tci_write_reg(index, value);
-}
-
 #if TCG_TARGET_REG_BITS == 64
 static void tci_write_reg32s(TCGReg index, int32_t value)
 {
@@ -138,11 +128,6 @@ static void tci_write_reg8(TCGReg index, uint8_t value)
     tci_write_reg(index, value);
 }
 
-static void tci_write_reg16(TCGReg index, uint16_t value)
-{
-    tci_write_reg(index, value);
-}
-
 static void tci_write_reg32(TCGReg index, uint32_t value)
 {
     tci_write_reg(index, value);
@@ -433,6 +418,53 @@ static bool tci_compare64(uint64_t u0, uint64_t u1, TCGCond condition)
     return result;
 }
 
+#ifdef CONFIG_SOFTMMU
+# define mmuidx          tci_read_i(&tb_ptr)
+# define qemu_ld_ub \
+    helper_ret_ldub_mmu(env, taddr, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_ld_leuw \
+    helper_le_lduw_mmu(env, taddr, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_ld_leul \
+    helper_le_ldul_mmu(env, taddr, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_ld_leq \
+    helper_le_ldq_mmu(env, taddr, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_ld_beuw \
+    helper_be_lduw_mmu(env, taddr, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_ld_beul \
+    helper_be_ldul_mmu(env, taddr, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_ld_beq \
+    helper_be_ldq_mmu(env, taddr, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_st_b(X) \
+    helper_ret_stb_mmu(env, taddr, X, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_st_lew(X) \
+    helper_le_stw_mmu(env, taddr, X, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_st_lel(X) \
+    helper_le_stl_mmu(env, taddr, X, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_st_leq(X) \
+    helper_le_stq_mmu(env, taddr, X, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_st_bew(X) \
+    helper_be_stw_mmu(env, taddr, X, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_st_bel(X) \
+    helper_be_stl_mmu(env, taddr, X, mmuidx, (uintptr_t)tb_ptr)
+# define qemu_st_beq(X) \
+    helper_be_stq_mmu(env, taddr, X, mmuidx, (uintptr_t)tb_ptr)
+#else
+# define qemu_ld_ub      ldub_p(g2h(taddr))
+# define qemu_ld_leuw    lduw_le_p(g2h(taddr))
+# define qemu_ld_leul    (uint32_t)ldl_le_p(g2h(taddr))
+# define qemu_ld_leq     ldq_le_p(g2h(taddr))
+# define qemu_ld_beuw    lduw_be_p(g2h(taddr))
+# define qemu_ld_beul    (uint32_t)ldl_be_p(g2h(taddr))
+# define qemu_ld_beq     ldq_be_p(g2h(taddr))
+# define qemu_st_b(X)    stb_p(g2h(taddr), X)
+# define qemu_st_lew(X)  stw_le_p(g2h(taddr), X)
+# define qemu_st_lel(X)  stl_le_p(g2h(taddr), X)
+# define qemu_st_leq(X)  stq_le_p(g2h(taddr), X)
+# define qemu_st_bew(X)  stw_be_p(g2h(taddr), X)
+# define qemu_st_bel(X)  stl_be_p(g2h(taddr), X)
+# define qemu_st_beq(X)  stq_be_p(g2h(taddr), X)
+#endif
+
 /* Interpret pseudo code in tb. */
 uintptr_t tcg_qemu_tb_exec(CPUArchState *env, uint8_t *tb_ptr)
 {
@@ -456,9 +488,6 @@ uintptr_t tcg_qemu_tb_exec(CPUArchState *env, uint8_t *tb_ptr)
         tcg_target_ulong label;
         TCGCond condition;
         target_ulong taddr;
-#ifndef CONFIG_SOFTMMU
-        tcg_target_ulong host_addr;
-#endif
         uint8_t tmp8;
         uint16_t tmp16;
         uint32_t tmp32;
@@ -466,6 +495,7 @@ uintptr_t tcg_qemu_tb_exec(CPUArchState *env, uint8_t *tb_ptr)
 #if TCG_TARGET_REG_BITS == 32
         uint64_t v64;
 #endif
+        TCGMemOp memop;
 
 #if defined(GETPC)
         tci_tb_ptr = (uintptr_t)tb_ptr;
@@ -1086,145 +1116,145 @@ uintptr_t tcg_qemu_tb_exec(CPUArchState *env, uint8_t *tb_ptr)
             assert(tb_ptr == old_code_ptr + op_size);
             tb_ptr += (int32_t)t0;
             continue;
-        case INDEX_op_qemu_ld8u:
-            t0 = *tb_ptr++;
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp8 = helper_ldb_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp8 = *(uint8_t *)(host_addr + GUEST_BASE);
-#endif
-            tci_write_reg8(t0, tmp8);
-            break;
-        case INDEX_op_qemu_ld8s:
+        case INDEX_op_qemu_ld_i32:
             t0 = *tb_ptr++;
             taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp8 = helper_ldb_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp8 = *(uint8_t *)(host_addr + GUEST_BASE);
-#endif
-            tci_write_reg8s(t0, tmp8);
-            break;
-        case INDEX_op_qemu_ld16u:
-            t0 = *tb_ptr++;
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp16 = helper_ldw_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp16 = tswap16(*(uint16_t *)(host_addr + GUEST_BASE));
-#endif
-            tci_write_reg16(t0, tmp16);
-            break;
-        case INDEX_op_qemu_ld16s:
-            t0 = *tb_ptr++;
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp16 = helper_ldw_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp16 = tswap16(*(uint16_t *)(host_addr + GUEST_BASE));
-#endif
-            tci_write_reg16s(t0, tmp16);
-            break;
-#if TCG_TARGET_REG_BITS == 64
-        case INDEX_op_qemu_ld32u:
-            t0 = *tb_ptr++;
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp32 = helper_ldl_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp32 = tswap32(*(uint32_t *)(host_addr + GUEST_BASE));
-#endif
-            tci_write_reg32(t0, tmp32);
-            break;
-        case INDEX_op_qemu_ld32s:
-            t0 = *tb_ptr++;
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp32 = helper_ldl_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp32 = tswap32(*(uint32_t *)(host_addr + GUEST_BASE));
-#endif
-            tci_write_reg32s(t0, tmp32);
-            break;
-#endif /* TCG_TARGET_REG_BITS == 64 */
-        case INDEX_op_qemu_ld32:
-            t0 = *tb_ptr++;
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp32 = helper_ldl_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp32 = tswap32(*(uint32_t *)(host_addr + GUEST_BASE));
-#endif
-            tci_write_reg32(t0, tmp32);
+            memop = tci_read_i(&tb_ptr);
+            switch (memop) {
+            case MO_UB:
+                tmp32 = qemu_ld_ub;
+                break;
+            case MO_SB:
+                tmp32 = (int8_t)qemu_ld_ub;
+                break;
+            case MO_LEUW:
+                tmp32 = qemu_ld_leuw;
+                break;
+            case MO_LESW:
+                tmp32 = (int16_t)qemu_ld_leuw;
+                break;
+            case MO_LEUL:
+                tmp32 = qemu_ld_leul;
+                break;
+            case MO_BEUW:
+                tmp32 = qemu_ld_beuw;
+                break;
+            case MO_BESW:
+                tmp32 = (int16_t)qemu_ld_beuw;
+                break;
+            case MO_BEUL:
+                tmp32 = qemu_ld_beul;
+                break;
+            default:
+                tcg_abort();
+            }
+            tci_write_reg(t0, tmp32);
             break;
-        case INDEX_op_qemu_ld64:
+        case INDEX_op_qemu_ld_i64:
             t0 = *tb_ptr++;
-#if TCG_TARGET_REG_BITS == 32
-            t1 = *tb_ptr++;
-#endif
+            if (TCG_TARGET_REG_BITS == 32) {
+                t1 = *tb_ptr++;
+            }
             taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            tmp64 = helper_ldq_mmu(env, taddr, tci_read_i(&tb_ptr));
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            tmp64 = tswap64(*(uint64_t *)(host_addr + GUEST_BASE));
-#endif
+            memop = tci_read_i(&tb_ptr);
+            switch (memop) {
+            case MO_UB:
+                tmp64 = qemu_ld_ub;
+                break;
+            case MO_SB:
+                tmp64 = (int8_t)qemu_ld_ub;
+                break;
+            case MO_LEUW:
+                tmp64 = qemu_ld_leuw;
+                break;
+            case MO_LESW:
+                tmp64 = (int16_t)qemu_ld_leuw;
+                break;
+            case MO_LEUL:
+                tmp64 = qemu_ld_leul;
+                break;
+            case MO_LESL:
+                tmp64 = (int32_t)qemu_ld_leul;
+                break;
+            case MO_LEQ:
+                tmp64 = qemu_ld_leq;
+                break;
+            case MO_BEUW:
+                tmp64 = qemu_ld_beuw;
+                break;
+            case MO_BESW:
+                tmp64 = (int16_t)qemu_ld_beuw;
+                break;
+            case MO_BEUL:
+                tmp64 = qemu_ld_beul;
+                break;
+            case MO_BESL:
+                tmp64 = (int32_t)qemu_ld_beul;
+                break;
+            case MO_BEQ:
+                tmp64 = qemu_ld_beq;
+                break;
+            default:
+                tcg_abort();
+            }
             tci_write_reg(t0, tmp64);
-#if TCG_TARGET_REG_BITS == 32
-            tci_write_reg(t1, tmp64 >> 32);
-#endif
-            break;
-        case INDEX_op_qemu_st8:
-            t0 = tci_read_r8(&tb_ptr);
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            t2 = tci_read_i(&tb_ptr);
-            helper_stb_mmu(env, taddr, t0, t2);
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            *(uint8_t *)(host_addr + GUEST_BASE) = t0;
-#endif
-            break;
-        case INDEX_op_qemu_st16:
-            t0 = tci_read_r16(&tb_ptr);
-            taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            t2 = tci_read_i(&tb_ptr);
-            helper_stw_mmu(env, taddr, t0, t2);
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            *(uint16_t *)(host_addr + GUEST_BASE) = tswap16(t0);
-#endif
+            if (TCG_TARGET_REG_BITS == 32) {
+                tci_write_reg(t1, tmp64 >> 32);
+            }
             break;
-        case INDEX_op_qemu_st32:
-            t0 = tci_read_r32(&tb_ptr);
+        case INDEX_op_qemu_st_i32:
+            t0 = tci_read_r(&tb_ptr);
             taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            t2 = tci_read_i(&tb_ptr);
-            helper_stl_mmu(env, taddr, t0, t2);
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            *(uint32_t *)(host_addr + GUEST_BASE) = tswap32(t0);
-#endif
+            memop = tci_read_i(&tb_ptr);
+            switch (memop) {
+            case MO_UB:
+                qemu_st_b(t0);
+                break;
+            case MO_LEUW:
+                qemu_st_lew(t0);
+                break;
+            case MO_LEUL:
+                qemu_st_lel(t0);
+                break;
+            case MO_BEUW:
+                qemu_st_bew(t0);
+                break;
+            case MO_BEUL:
+                qemu_st_bel(t0);
+                break;
+            default:
+                tcg_abort();
+            }
             break;
-        case INDEX_op_qemu_st64:
+        case INDEX_op_qemu_st_i64:
             tmp64 = tci_read_r64(&tb_ptr);
             taddr = tci_read_ulong(&tb_ptr);
-#ifdef CONFIG_SOFTMMU
-            t2 = tci_read_i(&tb_ptr);
-            helper_stq_mmu(env, taddr, tmp64, t2);
-#else
-            host_addr = (tcg_target_ulong)taddr;
-            *(uint64_t *)(host_addr + GUEST_BASE) = tswap64(tmp64);
-#endif
+            memop = tci_read_i(&tb_ptr);
+            switch (memop) {
+            case MO_UB:
+                qemu_st_b(tmp64);
+                break;
+            case MO_LEUW:
+                qemu_st_lew(tmp64);
+                break;
+            case MO_LEUL:
+                qemu_st_lel(tmp64);
+                break;
+            case MO_LEQ:
+                qemu_st_leq(tmp64);
+                break;
+            case MO_BEUW:
+                qemu_st_bew(tmp64);
+                break;
+            case MO_BEUL:
+                qemu_st_bel(tmp64);
+                break;
+            case MO_BEQ:
+                qemu_st_beq(tmp64);
+                break;
+            default:
+                tcg_abort();
+            }
             break;
         default:
             TODO();