All of lore.kernel.org
 help / color / mirror / Atom feed
From: Richard Henderson <richard.henderson@linaro.org>
To: qemu-devel@nongnu.org
Cc: peter.maydell@linaro.org, "Philippe Mathieu-Daudé" <f4bug@amsat.org>
Subject: [PULL 15/33] tcg/tci: Change encoding to uint32_t units
Date: Sat, 19 Jun 2021 11:14:34 -0700	[thread overview]
Message-ID: <20210619181452.877683-16-richard.henderson@linaro.org> (raw)
In-Reply-To: <20210619181452.877683-1-richard.henderson@linaro.org>

This removes all of the problems with unaligned accesses
to the bytecode stream.

With an 8-bit opcode at the bottom, we have 24 bits remaining,
which are generally split into 6 4-bit slots.  This fits well
with the maximum length opcodes, e.g. INDEX_op_add2_i32, which
have 6 register operands.

We have, in previous patches, rearranged things such that there
are no operations with a label which have more than one other
operand.  Which leaves us with a 20-bit field in which to encode
a label, giving us a maximum TB size of 512k -- easily large.

Change the INDEX_op_tci_movi_{i32,i64} opcodes to tci_mov[il].
The former puts the immediate in the upper 20 bits of the insn,
like we do for the label displacement.  The later uses a label
to reference an entry in the constant pool.  Thus, in the worst
case we still have a single memory reference for any constant,
but now the constants are out-of-line of the bytecode and can
be shared between different moves saving space.

Change INDEX_op_call to use a label to reference a pair of
pointers in the constant pool.  This removes the only slightly
dodgy link with the layout of struct TCGHelperInfo.

The re-encode cannot be done in pieces.

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 include/tcg/tcg-opc.h    |   4 +-
 tcg/tci/tcg-target.h     |   3 +-
 tcg/tci.c                | 539 +++++++++++++++------------------------
 tcg/tci/tcg-target.c.inc | 379 ++++++++++++---------------
 tcg/tci/README           |  20 +-
 5 files changed, 383 insertions(+), 562 deletions(-)

diff --git a/include/tcg/tcg-opc.h b/include/tcg/tcg-opc.h
index bbb0884af8..5bbec858aa 100644
--- a/include/tcg/tcg-opc.h
+++ b/include/tcg/tcg-opc.h
@@ -277,8 +277,8 @@ DEF(last_generic, 0, 0, 0, TCG_OPF_NOT_PRESENT)
 
 #ifdef TCG_TARGET_INTERPRETER
 /* These opcodes are only for use between the tci generator and interpreter. */
-DEF(tci_movi_i32, 1, 0, 1, TCG_OPF_NOT_PRESENT)
-DEF(tci_movi_i64, 1, 0, 1, TCG_OPF_64BIT | TCG_OPF_NOT_PRESENT)
+DEF(tci_movi, 1, 0, 1, TCG_OPF_NOT_PRESENT)
+DEF(tci_movl, 1, 0, 1, TCG_OPF_NOT_PRESENT)
 #endif
 
 #undef TLADDR_ARGS
diff --git a/tcg/tci/tcg-target.h b/tcg/tci/tcg-target.h
index 80cafb7d18..c9cbe505a7 100644
--- a/tcg/tci/tcg-target.h
+++ b/tcg/tci/tcg-target.h
@@ -41,7 +41,7 @@
 #define TCG_TARGET_H
 
 #define TCG_TARGET_INTERPRETER 1
-#define TCG_TARGET_INSN_UNIT_SIZE 1
+#define TCG_TARGET_INSN_UNIT_SIZE 4
 #define TCG_TARGET_TLB_DISPLACEMENT_BITS 32
 #define MAX_CODE_GEN_BUFFER_SIZE  ((size_t)-1)
 
@@ -166,6 +166,7 @@ typedef enum {
 #define TCG_TARGET_STACK_ALIGN          8
 
 #define HAVE_TCG_QEMU_TB_EXEC
+#define TCG_TARGET_NEED_POOL_LABELS
 
 /* We could notice __i386__ or __s390x__ and reduce the barriers depending
    on the host.  But if you want performance, you use the normal backend.
diff --git a/tcg/tci.c b/tcg/tci.c
index 613b94997c..c82d5b2f51 100644
--- a/tcg/tci.c
+++ b/tcg/tci.c
@@ -51,49 +51,6 @@ static uint64_t tci_uint64(uint32_t high, uint32_t low)
     return ((uint64_t)high << 32) + low;
 }
 
-/* Read constant byte from bytecode. */
-static uint8_t tci_read_b(const uint8_t **tb_ptr)
-{
-    return *(tb_ptr[0]++);
-}
-
-/* Read register number from bytecode. */
-static TCGReg tci_read_r(const uint8_t **tb_ptr)
-{
-    uint8_t regno = tci_read_b(tb_ptr);
-    tci_assert(regno < TCG_TARGET_NB_REGS);
-    return regno;
-}
-
-/* Read constant (native size) from bytecode. */
-static tcg_target_ulong tci_read_i(const uint8_t **tb_ptr)
-{
-    tcg_target_ulong value = *(const tcg_target_ulong *)(*tb_ptr);
-    *tb_ptr += sizeof(value);
-    return value;
-}
-
-/* Read unsigned constant (32 bit) from bytecode. */
-static uint32_t tci_read_i32(const uint8_t **tb_ptr)
-{
-    uint32_t value = *(const uint32_t *)(*tb_ptr);
-    *tb_ptr += sizeof(value);
-    return value;
-}
-
-/* Read signed constant (32 bit) from bytecode. */
-static int32_t tci_read_s32(const uint8_t **tb_ptr)
-{
-    int32_t value = *(const int32_t *)(*tb_ptr);
-    *tb_ptr += sizeof(value);
-    return value;
-}
-
-static tcg_target_ulong tci_read_label(const uint8_t **tb_ptr)
-{
-    return tci_read_i(tb_ptr);
-}
-
 /*
  * Load sets of arguments all at once.  The naming convention is:
  *   tci_args_<arguments>
@@ -110,211 +67,128 @@ static tcg_target_ulong tci_read_label(const uint8_t **tb_ptr)
  *   s = signed ldst offset
  */
 
-static void check_size(const uint8_t *start, const uint8_t **tb_ptr)
+static void tci_args_l(uint32_t insn, const void *tb_ptr, void **l0)
 {
-    const uint8_t *old_code_ptr = start - 2;
-    uint8_t op_size = old_code_ptr[1];
-    tci_assert(*tb_ptr == old_code_ptr + op_size);
+    int diff = sextract32(insn, 12, 20);
+    *l0 = diff ? (void *)tb_ptr + diff : NULL;
 }
 
-static void tci_args_l(const uint8_t **tb_ptr, void **l0)
+static void tci_args_nl(uint32_t insn, const void *tb_ptr,
+                        uint8_t *n0, void **l1)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *l0 = (void *)tci_read_label(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *n0 = extract32(insn, 8, 4);
+    *l1 = sextract32(insn, 12, 20) + (void *)tb_ptr;
 }
 
-static void tci_args_nll(const uint8_t **tb_ptr, uint8_t *n0,
-                         void **l1, void **l2)
+static void tci_args_rl(uint32_t insn, const void *tb_ptr,
+                        TCGReg *r0, void **l1)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *n0 = tci_read_b(tb_ptr);
-    *l1 = (void *)tci_read_label(tb_ptr);
-    *l2 = (void *)tci_read_label(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *l1 = sextract32(insn, 12, 20) + (void *)tb_ptr;
 }
 
-static void tci_args_rl(const uint8_t **tb_ptr, TCGReg *r0, void **l1)
+static void tci_args_rr(uint32_t insn, TCGReg *r0, TCGReg *r1)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *l1 = (void *)tci_read_label(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
 }
 
-static void tci_args_rr(const uint8_t **tb_ptr,
-                        TCGReg *r0, TCGReg *r1)
+static void tci_args_ri(uint32_t insn, TCGReg *r0, tcg_target_ulong *i1)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *i1 = sextract32(insn, 12, 20);
 }
 
-static void tci_args_ri(const uint8_t **tb_ptr,
-                        TCGReg *r0, tcg_target_ulong *i1)
+static void tci_args_rrm(uint32_t insn, TCGReg *r0,
+                         TCGReg *r1, TCGMemOpIdx *m2)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *i1 = tci_read_i32(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *m2 = extract32(insn, 20, 12);
 }
 
-#if TCG_TARGET_REG_BITS == 64
-static void tci_args_rI(const uint8_t **tb_ptr,
-                        TCGReg *r0, tcg_target_ulong *i1)
+static void tci_args_rrr(uint32_t insn, TCGReg *r0, TCGReg *r1, TCGReg *r2)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *i1 = tci_read_i(tb_ptr);
-
-    check_size(start, tb_ptr);
-}
-#endif
-
-static void tci_args_rrm(const uint8_t **tb_ptr,
-                         TCGReg *r0, TCGReg *r1, TCGMemOpIdx *m2)
-{
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *m2 = tci_read_i32(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
 }
 
-static void tci_args_rrr(const uint8_t **tb_ptr,
-                         TCGReg *r0, TCGReg *r1, TCGReg *r2)
+static void tci_args_rrs(uint32_t insn, TCGReg *r0, TCGReg *r1, int32_t *i2)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *i2 = sextract32(insn, 16, 16);
 }
 
-static void tci_args_rrs(const uint8_t **tb_ptr,
-                         TCGReg *r0, TCGReg *r1, int32_t *i2)
-{
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *i2 = tci_read_s32(tb_ptr);
-
-    check_size(start, tb_ptr);
-}
-
-static void tci_args_rrrc(const uint8_t **tb_ptr,
+static void tci_args_rrrc(uint32_t insn,
                           TCGReg *r0, TCGReg *r1, TCGReg *r2, TCGCond *c3)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-    *c3 = tci_read_b(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
+    *c3 = extract32(insn, 20, 4);
 }
 
-static void tci_args_rrrm(const uint8_t **tb_ptr,
+static void tci_args_rrrm(uint32_t insn,
                           TCGReg *r0, TCGReg *r1, TCGReg *r2, TCGMemOpIdx *m3)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-    *m3 = tci_read_i32(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
+    *m3 = extract32(insn, 20, 12);
 }
 
-static void tci_args_rrrbb(const uint8_t **tb_ptr, TCGReg *r0, TCGReg *r1,
+static void tci_args_rrrbb(uint32_t insn, TCGReg *r0, TCGReg *r1,
                            TCGReg *r2, uint8_t *i3, uint8_t *i4)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-    *i3 = tci_read_b(tb_ptr);
-    *i4 = tci_read_b(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
+    *i3 = extract32(insn, 20, 6);
+    *i4 = extract32(insn, 26, 6);
 }
 
-static void tci_args_rrrrm(const uint8_t **tb_ptr, TCGReg *r0, TCGReg *r1,
-                           TCGReg *r2, TCGReg *r3, TCGMemOpIdx *m4)
+static void tci_args_rrrrr(uint32_t insn, TCGReg *r0, TCGReg *r1,
+                           TCGReg *r2, TCGReg *r3, TCGReg *r4)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-    *r3 = tci_read_r(tb_ptr);
-    *m4 = tci_read_i32(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
+    *r3 = extract32(insn, 20, 4);
+    *r4 = extract32(insn, 24, 4);
 }
 
 #if TCG_TARGET_REG_BITS == 32
-static void tci_args_rrrr(const uint8_t **tb_ptr,
+static void tci_args_rrrr(uint32_t insn,
                           TCGReg *r0, TCGReg *r1, TCGReg *r2, TCGReg *r3)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-    *r3 = tci_read_r(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
+    *r3 = extract32(insn, 20, 4);
 }
 
-static void tci_args_rrrrrc(const uint8_t **tb_ptr, TCGReg *r0, TCGReg *r1,
+static void tci_args_rrrrrc(uint32_t insn, TCGReg *r0, TCGReg *r1,
                             TCGReg *r2, TCGReg *r3, TCGReg *r4, TCGCond *c5)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-    *r3 = tci_read_r(tb_ptr);
-    *r4 = tci_read_r(tb_ptr);
-    *c5 = tci_read_b(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
+    *r3 = extract32(insn, 20, 4);
+    *r4 = extract32(insn, 24, 4);
+    *c5 = extract32(insn, 28, 4);
 }
 
-static void tci_args_rrrrrr(const uint8_t **tb_ptr, TCGReg *r0, TCGReg *r1,
+static void tci_args_rrrrrr(uint32_t insn, TCGReg *r0, TCGReg *r1,
                             TCGReg *r2, TCGReg *r3, TCGReg *r4, TCGReg *r5)
 {
-    const uint8_t *start = *tb_ptr;
-
-    *r0 = tci_read_r(tb_ptr);
-    *r1 = tci_read_r(tb_ptr);
-    *r2 = tci_read_r(tb_ptr);
-    *r3 = tci_read_r(tb_ptr);
-    *r4 = tci_read_r(tb_ptr);
-    *r5 = tci_read_r(tb_ptr);
-
-    check_size(start, tb_ptr);
+    *r0 = extract32(insn, 8, 4);
+    *r1 = extract32(insn, 12, 4);
+    *r2 = extract32(insn, 16, 4);
+    *r3 = extract32(insn, 20, 4);
+    *r4 = extract32(insn, 24, 4);
+    *r5 = extract32(insn, 28, 4);
 }
 #endif
 
@@ -452,7 +326,7 @@ static bool tci_compare64(uint64_t u0, uint64_t u1, TCGCond condition)
 uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
                                             const void *v_tb_ptr)
 {
-    const uint8_t *tb_ptr = v_tb_ptr;
+    const uint32_t *tb_ptr = v_tb_ptr;
     tcg_target_ulong regs[TCG_TARGET_NB_REGS];
     uint64_t stack[(TCG_STATIC_CALL_ARGS_SIZE + TCG_STATIC_FRAME_SIZE)
                    / sizeof(uint64_t)];
@@ -465,8 +339,9 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
     tci_assert(tb_ptr);
 
     for (;;) {
-        TCGOpcode opc = tb_ptr[0];
-        TCGReg r0, r1, r2, r3;
+        uint32_t insn;
+        TCGOpcode opc;
+        TCGReg r0, r1, r2, r3, r4;
         tcg_target_ulong t1;
         TCGCond condition;
         target_ulong taddr;
@@ -474,15 +349,15 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
         uint32_t tmp32;
         uint64_t tmp64;
 #if TCG_TARGET_REG_BITS == 32
-        TCGReg r4, r5;
+        TCGReg r5;
         uint64_t T1, T2;
 #endif
         TCGMemOpIdx oi;
         int32_t ofs;
-        void *ptr, *cif;
+        void *ptr;
 
-        /* Skip opcode and size entry. */
-        tb_ptr += 2;
+        insn = *tb_ptr++;
+        opc = extract32(insn, 0, 8);
 
         switch (opc) {
         case INDEX_op_call:
@@ -498,12 +373,15 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
                 }
             }
 
-            tci_args_nll(&tb_ptr, &len, &ptr, &cif);
+            tci_args_nl(insn, tb_ptr, &len, &ptr);
 
             /* Helper functions may need to access the "return address" */
             tci_tb_ptr = (uintptr_t)tb_ptr;
 
-            ffi_call(cif, ptr, stack, call_slots);
+            {
+                void **pptr = ptr;
+                ffi_call(pptr[1], pptr[0], stack, call_slots);
+            }
 
             /* Any result winds up "left-aligned" in the stack[0] slot. */
             switch (len) {
@@ -532,76 +410,80 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
             break;
 
         case INDEX_op_br:
-            tci_args_l(&tb_ptr, &ptr);
+            tci_args_l(insn, tb_ptr, &ptr);
             tb_ptr = ptr;
             continue;
         case INDEX_op_setcond_i32:
-            tci_args_rrrc(&tb_ptr, &r0, &r1, &r2, &condition);
+            tci_args_rrrc(insn, &r0, &r1, &r2, &condition);
             regs[r0] = tci_compare32(regs[r1], regs[r2], condition);
             break;
 #if TCG_TARGET_REG_BITS == 32
         case INDEX_op_setcond2_i32:
-            tci_args_rrrrrc(&tb_ptr, &r0, &r1, &r2, &r3, &r4, &condition);
+            tci_args_rrrrrc(insn, &r0, &r1, &r2, &r3, &r4, &condition);
             T1 = tci_uint64(regs[r2], regs[r1]);
             T2 = tci_uint64(regs[r4], regs[r3]);
             regs[r0] = tci_compare64(T1, T2, condition);
             break;
 #elif TCG_TARGET_REG_BITS == 64
         case INDEX_op_setcond_i64:
-            tci_args_rrrc(&tb_ptr, &r0, &r1, &r2, &condition);
+            tci_args_rrrc(insn, &r0, &r1, &r2, &condition);
             regs[r0] = tci_compare64(regs[r1], regs[r2], condition);
             break;
 #endif
         CASE_32_64(mov)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = regs[r1];
             break;
-        case INDEX_op_tci_movi_i32:
-            tci_args_ri(&tb_ptr, &r0, &t1);
+        case INDEX_op_tci_movi:
+            tci_args_ri(insn, &r0, &t1);
             regs[r0] = t1;
             break;
+        case INDEX_op_tci_movl:
+            tci_args_rl(insn, tb_ptr, &r0, &ptr);
+            regs[r0] = *(tcg_target_ulong *)ptr;
+            break;
 
             /* Load/store operations (32 bit). */
 
         CASE_32_64(ld8u)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             regs[r0] = *(uint8_t *)ptr;
             break;
         CASE_32_64(ld8s)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             regs[r0] = *(int8_t *)ptr;
             break;
         CASE_32_64(ld16u)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             regs[r0] = *(uint16_t *)ptr;
             break;
         CASE_32_64(ld16s)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             regs[r0] = *(int16_t *)ptr;
             break;
         case INDEX_op_ld_i32:
         CASE_64(ld32u)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             regs[r0] = *(uint32_t *)ptr;
             break;
         CASE_32_64(st8)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             *(uint8_t *)ptr = regs[r0];
             break;
         CASE_32_64(st16)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             *(uint16_t *)ptr = regs[r0];
             break;
         case INDEX_op_st_i32:
         CASE_64(st32)
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             *(uint32_t *)ptr = regs[r0];
             break;
@@ -609,171 +491,166 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
             /* Arithmetic operations (mixed 32/64 bit). */
 
         CASE_32_64(add)
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] + regs[r2];
             break;
         CASE_32_64(sub)
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] - regs[r2];
             break;
         CASE_32_64(mul)
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] * regs[r2];
             break;
         CASE_32_64(and)
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] & regs[r2];
             break;
         CASE_32_64(or)
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] | regs[r2];
             break;
         CASE_32_64(xor)
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] ^ regs[r2];
             break;
 
             /* Arithmetic operations (32 bit). */
 
         case INDEX_op_div_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (int32_t)regs[r1] / (int32_t)regs[r2];
             break;
         case INDEX_op_divu_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (uint32_t)regs[r1] / (uint32_t)regs[r2];
             break;
         case INDEX_op_rem_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (int32_t)regs[r1] % (int32_t)regs[r2];
             break;
         case INDEX_op_remu_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (uint32_t)regs[r1] % (uint32_t)regs[r2];
             break;
 
             /* Shift/rotate operations (32 bit). */
 
         case INDEX_op_shl_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (uint32_t)regs[r1] << (regs[r2] & 31);
             break;
         case INDEX_op_shr_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (uint32_t)regs[r1] >> (regs[r2] & 31);
             break;
         case INDEX_op_sar_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (int32_t)regs[r1] >> (regs[r2] & 31);
             break;
 #if TCG_TARGET_HAS_rot_i32
         case INDEX_op_rotl_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = rol32(regs[r1], regs[r2] & 31);
             break;
         case INDEX_op_rotr_i32:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = ror32(regs[r1], regs[r2] & 31);
             break;
 #endif
 #if TCG_TARGET_HAS_deposit_i32
         case INDEX_op_deposit_i32:
-            tci_args_rrrbb(&tb_ptr, &r0, &r1, &r2, &pos, &len);
+            tci_args_rrrbb(insn, &r0, &r1, &r2, &pos, &len);
             regs[r0] = deposit32(regs[r1], pos, len, regs[r2]);
             break;
 #endif
         case INDEX_op_brcond_i32:
-            tci_args_rl(&tb_ptr, &r0, &ptr);
+            tci_args_rl(insn, tb_ptr, &r0, &ptr);
             if ((uint32_t)regs[r0]) {
                 tb_ptr = ptr;
             }
             break;
 #if TCG_TARGET_REG_BITS == 32
         case INDEX_op_add2_i32:
-            tci_args_rrrrrr(&tb_ptr, &r0, &r1, &r2, &r3, &r4, &r5);
+            tci_args_rrrrrr(insn, &r0, &r1, &r2, &r3, &r4, &r5);
             T1 = tci_uint64(regs[r3], regs[r2]);
             T2 = tci_uint64(regs[r5], regs[r4]);
             tci_write_reg64(regs, r1, r0, T1 + T2);
             break;
         case INDEX_op_sub2_i32:
-            tci_args_rrrrrr(&tb_ptr, &r0, &r1, &r2, &r3, &r4, &r5);
+            tci_args_rrrrrr(insn, &r0, &r1, &r2, &r3, &r4, &r5);
             T1 = tci_uint64(regs[r3], regs[r2]);
             T2 = tci_uint64(regs[r5], regs[r4]);
             tci_write_reg64(regs, r1, r0, T1 - T2);
             break;
         case INDEX_op_mulu2_i32:
-            tci_args_rrrr(&tb_ptr, &r0, &r1, &r2, &r3);
+            tci_args_rrrr(insn, &r0, &r1, &r2, &r3);
             tci_write_reg64(regs, r1, r0, (uint64_t)regs[r2] * regs[r3]);
             break;
 #endif /* TCG_TARGET_REG_BITS == 32 */
 #if TCG_TARGET_HAS_ext8s_i32 || TCG_TARGET_HAS_ext8s_i64
         CASE_32_64(ext8s)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = (int8_t)regs[r1];
             break;
 #endif
 #if TCG_TARGET_HAS_ext16s_i32 || TCG_TARGET_HAS_ext16s_i64
         CASE_32_64(ext16s)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = (int16_t)regs[r1];
             break;
 #endif
 #if TCG_TARGET_HAS_ext8u_i32 || TCG_TARGET_HAS_ext8u_i64
         CASE_32_64(ext8u)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = (uint8_t)regs[r1];
             break;
 #endif
 #if TCG_TARGET_HAS_ext16u_i32 || TCG_TARGET_HAS_ext16u_i64
         CASE_32_64(ext16u)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = (uint16_t)regs[r1];
             break;
 #endif
 #if TCG_TARGET_HAS_bswap16_i32 || TCG_TARGET_HAS_bswap16_i64
         CASE_32_64(bswap16)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = bswap16(regs[r1]);
             break;
 #endif
 #if TCG_TARGET_HAS_bswap32_i32 || TCG_TARGET_HAS_bswap32_i64
         CASE_32_64(bswap32)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = bswap32(regs[r1]);
             break;
 #endif
 #if TCG_TARGET_HAS_not_i32 || TCG_TARGET_HAS_not_i64
         CASE_32_64(not)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = ~regs[r1];
             break;
 #endif
 #if TCG_TARGET_HAS_neg_i32 || TCG_TARGET_HAS_neg_i64
         CASE_32_64(neg)
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = -regs[r1];
             break;
 #endif
 #if TCG_TARGET_REG_BITS == 64
-        case INDEX_op_tci_movi_i64:
-            tci_args_rI(&tb_ptr, &r0, &t1);
-            regs[r0] = t1;
-            break;
-
             /* Load/store operations (64 bit). */
 
         case INDEX_op_ld32s_i64:
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             regs[r0] = *(int32_t *)ptr;
             break;
         case INDEX_op_ld_i64:
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             regs[r0] = *(uint64_t *)ptr;
             break;
         case INDEX_op_st_i64:
-            tci_args_rrs(&tb_ptr, &r0, &r1, &ofs);
+            tci_args_rrs(insn, &r0, &r1, &ofs);
             ptr = (void *)(regs[r1] + ofs);
             *(uint64_t *)ptr = regs[r0];
             break;
@@ -781,71 +658,71 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
             /* Arithmetic operations (64 bit). */
 
         case INDEX_op_div_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (int64_t)regs[r1] / (int64_t)regs[r2];
             break;
         case INDEX_op_divu_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (uint64_t)regs[r1] / (uint64_t)regs[r2];
             break;
         case INDEX_op_rem_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (int64_t)regs[r1] % (int64_t)regs[r2];
             break;
         case INDEX_op_remu_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (uint64_t)regs[r1] % (uint64_t)regs[r2];
             break;
 
             /* Shift/rotate operations (64 bit). */
 
         case INDEX_op_shl_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] << (regs[r2] & 63);
             break;
         case INDEX_op_shr_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = regs[r1] >> (regs[r2] & 63);
             break;
         case INDEX_op_sar_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = (int64_t)regs[r1] >> (regs[r2] & 63);
             break;
 #if TCG_TARGET_HAS_rot_i64
         case INDEX_op_rotl_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = rol64(regs[r1], regs[r2] & 63);
             break;
         case INDEX_op_rotr_i64:
-            tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+            tci_args_rrr(insn, &r0, &r1, &r2);
             regs[r0] = ror64(regs[r1], regs[r2] & 63);
             break;
 #endif
 #if TCG_TARGET_HAS_deposit_i64
         case INDEX_op_deposit_i64:
-            tci_args_rrrbb(&tb_ptr, &r0, &r1, &r2, &pos, &len);
+            tci_args_rrrbb(insn, &r0, &r1, &r2, &pos, &len);
             regs[r0] = deposit64(regs[r1], pos, len, regs[r2]);
             break;
 #endif
         case INDEX_op_brcond_i64:
-            tci_args_rl(&tb_ptr, &r0, &ptr);
+            tci_args_rl(insn, tb_ptr, &r0, &ptr);
             if (regs[r0]) {
                 tb_ptr = ptr;
             }
             break;
         case INDEX_op_ext32s_i64:
         case INDEX_op_ext_i32_i64:
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = (int32_t)regs[r1];
             break;
         case INDEX_op_ext32u_i64:
         case INDEX_op_extu_i32_i64:
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = (uint32_t)regs[r1];
             break;
 #if TCG_TARGET_HAS_bswap64_i64
         case INDEX_op_bswap64_i64:
-            tci_args_rr(&tb_ptr, &r0, &r1);
+            tci_args_rr(insn, &r0, &r1);
             regs[r0] = bswap64(regs[r1]);
             break;
 #endif
@@ -854,20 +731,20 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
             /* QEMU specific operations. */
 
         case INDEX_op_exit_tb:
-            tci_args_l(&tb_ptr, &ptr);
+            tci_args_l(insn, tb_ptr, &ptr);
             return (uintptr_t)ptr;
 
         case INDEX_op_goto_tb:
-            tci_args_l(&tb_ptr, &ptr);
+            tci_args_l(insn, tb_ptr, &ptr);
             tb_ptr = *(void **)ptr;
             break;
 
         case INDEX_op_qemu_ld_i32:
             if (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS) {
-                tci_args_rrm(&tb_ptr, &r0, &r1, &oi);
+                tci_args_rrm(insn, &r0, &r1, &oi);
                 taddr = regs[r1];
             } else {
-                tci_args_rrrm(&tb_ptr, &r0, &r1, &r2, &oi);
+                tci_args_rrrm(insn, &r0, &r1, &r2, &oi);
                 taddr = tci_uint64(regs[r2], regs[r1]);
             }
             switch (get_memop(oi) & (MO_BSWAP | MO_SSIZE)) {
@@ -903,14 +780,15 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
 
         case INDEX_op_qemu_ld_i64:
             if (TCG_TARGET_REG_BITS == 64) {
-                tci_args_rrm(&tb_ptr, &r0, &r1, &oi);
+                tci_args_rrm(insn, &r0, &r1, &oi);
                 taddr = regs[r1];
             } else if (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS) {
-                tci_args_rrrm(&tb_ptr, &r0, &r1, &r2, &oi);
+                tci_args_rrrm(insn, &r0, &r1, &r2, &oi);
                 taddr = regs[r2];
             } else {
-                tci_args_rrrrm(&tb_ptr, &r0, &r1, &r2, &r3, &oi);
+                tci_args_rrrrr(insn, &r0, &r1, &r2, &r3, &r4);
                 taddr = tci_uint64(regs[r3], regs[r2]);
+                oi = regs[r4];
             }
             switch (get_memop(oi) & (MO_BSWAP | MO_SSIZE)) {
             case MO_UB:
@@ -961,10 +839,10 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
 
         case INDEX_op_qemu_st_i32:
             if (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS) {
-                tci_args_rrm(&tb_ptr, &r0, &r1, &oi);
+                tci_args_rrm(insn, &r0, &r1, &oi);
                 taddr = regs[r1];
             } else {
-                tci_args_rrrm(&tb_ptr, &r0, &r1, &r2, &oi);
+                tci_args_rrrm(insn, &r0, &r1, &r2, &oi);
                 taddr = tci_uint64(regs[r2], regs[r1]);
             }
             tmp32 = regs[r0];
@@ -991,16 +869,17 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
 
         case INDEX_op_qemu_st_i64:
             if (TCG_TARGET_REG_BITS == 64) {
-                tci_args_rrm(&tb_ptr, &r0, &r1, &oi);
+                tci_args_rrm(insn, &r0, &r1, &oi);
                 taddr = regs[r1];
                 tmp64 = regs[r0];
             } else {
                 if (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS) {
-                    tci_args_rrrm(&tb_ptr, &r0, &r1, &r2, &oi);
+                    tci_args_rrrm(insn, &r0, &r1, &r2, &oi);
                     taddr = regs[r2];
                 } else {
-                    tci_args_rrrrm(&tb_ptr, &r0, &r1, &r2, &r3, &oi);
+                    tci_args_rrrrr(insn, &r0, &r1, &r2, &r3, &r4);
                     taddr = tci_uint64(regs[r3], regs[r2]);
+                    oi = regs[r4];
                 }
                 tmp64 = tci_uint64(regs[r1], regs[r0]);
             }
@@ -1084,87 +963,69 @@ static const char *str_c(TCGCond c)
 /* Disassemble TCI bytecode. */
 int print_insn_tci(bfd_vma addr, disassemble_info *info)
 {
-    uint8_t buf[256];
-    int length, status;
+    const uint32_t *tb_ptr = (const void *)(uintptr_t)addr;
     const TCGOpDef *def;
     const char *op_name;
+    uint32_t insn;
     TCGOpcode op;
-    TCGReg r0, r1, r2, r3;
+    TCGReg r0, r1, r2, r3, r4;
 #if TCG_TARGET_REG_BITS == 32
-    TCGReg r4, r5;
+    TCGReg r5;
 #endif
     tcg_target_ulong i1;
     int32_t s2;
     TCGCond c;
     TCGMemOpIdx oi;
     uint8_t pos, len;
-    void *ptr, *cif;
-    const uint8_t *tb_ptr;
+    void *ptr;
 
-    status = info->read_memory_func(addr, buf, 2, info);
-    if (status != 0) {
-        info->memory_error_func(status, addr, info);
-        return -1;
-    }
-    op = buf[0];
-    length = buf[1];
+    /* TCI is always the host, so we don't need to load indirect. */
+    insn = *tb_ptr++;
 
-    if (length < 2) {
-        info->fprintf_func(info->stream, "invalid length %d", length);
-        return 1;
-    }
-
-    status = info->read_memory_func(addr + 2, buf + 2, length - 2, info);
-    if (status != 0) {
-        info->memory_error_func(status, addr + 2, info);
-        return -1;
-    }
+    info->fprintf_func(info->stream, "%08x  ", insn);
 
+    op = extract32(insn, 0, 8);
     def = &tcg_op_defs[op];
     op_name = def->name;
-    tb_ptr = buf + 2;
 
     switch (op) {
     case INDEX_op_br:
     case INDEX_op_exit_tb:
     case INDEX_op_goto_tb:
-        tci_args_l(&tb_ptr, &ptr);
+        tci_args_l(insn, tb_ptr, &ptr);
         info->fprintf_func(info->stream, "%-12s  %p", op_name, ptr);
         break;
 
     case INDEX_op_call:
-        tci_args_nll(&tb_ptr, &len, &ptr, &cif);
-        info->fprintf_func(info->stream, "%-12s  %d, %p, %p",
-                           op_name, len, ptr, cif);
+        tci_args_nl(insn, tb_ptr, &len, &ptr);
+        info->fprintf_func(info->stream, "%-12s  %d, %p", op_name, len, ptr);
         break;
 
     case INDEX_op_brcond_i32:
     case INDEX_op_brcond_i64:
-        tci_args_rl(&tb_ptr, &r0, &ptr);
+        tci_args_rl(insn, tb_ptr, &r0, &ptr);
         info->fprintf_func(info->stream, "%-12s  %s, 0, ne, %p",
                            op_name, str_r(r0), ptr);
         break;
 
     case INDEX_op_setcond_i32:
     case INDEX_op_setcond_i64:
-        tci_args_rrrc(&tb_ptr, &r0, &r1, &r2, &c);
+        tci_args_rrrc(insn, &r0, &r1, &r2, &c);
         info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %s",
                            op_name, str_r(r0), str_r(r1), str_r(r2), str_c(c));
         break;
 
-    case INDEX_op_tci_movi_i32:
-        tci_args_ri(&tb_ptr, &r0, &i1);
+    case INDEX_op_tci_movi:
+        tci_args_ri(insn, &r0, &i1);
         info->fprintf_func(info->stream, "%-12s  %s, 0x%" TCG_PRIlx,
                            op_name, str_r(r0), i1);
         break;
 
-#if TCG_TARGET_REG_BITS == 64
-    case INDEX_op_tci_movi_i64:
-        tci_args_rI(&tb_ptr, &r0, &i1);
-        info->fprintf_func(info->stream, "%-12s  %s, 0x%" TCG_PRIlx,
-                           op_name, str_r(r0), i1);
+    case INDEX_op_tci_movl:
+        tci_args_rl(insn, tb_ptr, &r0, &ptr);
+        info->fprintf_func(info->stream, "%-12s  %s, %p",
+                           op_name, str_r(r0), ptr);
         break;
-#endif
 
     case INDEX_op_ld8u_i32:
     case INDEX_op_ld8u_i64:
@@ -1185,7 +1046,7 @@ int print_insn_tci(bfd_vma addr, disassemble_info *info)
     case INDEX_op_st32_i64:
     case INDEX_op_st_i32:
     case INDEX_op_st_i64:
-        tci_args_rrs(&tb_ptr, &r0, &r1, &s2);
+        tci_args_rrs(insn, &r0, &r1, &s2);
         info->fprintf_func(info->stream, "%-12s  %s, %s, %d",
                            op_name, str_r(r0), str_r(r1), s2);
         break;
@@ -1212,7 +1073,7 @@ int print_insn_tci(bfd_vma addr, disassemble_info *info)
     case INDEX_op_not_i64:
     case INDEX_op_neg_i32:
     case INDEX_op_neg_i64:
-        tci_args_rr(&tb_ptr, &r0, &r1);
+        tci_args_rr(insn, &r0, &r1);
         info->fprintf_func(info->stream, "%-12s  %s, %s",
                            op_name, str_r(r0), str_r(r1));
         break;
@@ -1247,28 +1108,28 @@ int print_insn_tci(bfd_vma addr, disassemble_info *info)
     case INDEX_op_rotl_i64:
     case INDEX_op_rotr_i32:
     case INDEX_op_rotr_i64:
-        tci_args_rrr(&tb_ptr, &r0, &r1, &r2);
+        tci_args_rrr(insn, &r0, &r1, &r2);
         info->fprintf_func(info->stream, "%-12s  %s, %s, %s",
                            op_name, str_r(r0), str_r(r1), str_r(r2));
         break;
 
     case INDEX_op_deposit_i32:
     case INDEX_op_deposit_i64:
-        tci_args_rrrbb(&tb_ptr, &r0, &r1, &r2, &pos, &len);
+        tci_args_rrrbb(insn, &r0, &r1, &r2, &pos, &len);
         info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %d, %d",
                            op_name, str_r(r0), str_r(r1), str_r(r2), pos, len);
         break;
 
 #if TCG_TARGET_REG_BITS == 32
     case INDEX_op_setcond2_i32:
-        tci_args_rrrrrc(&tb_ptr, &r0, &r1, &r2, &r3, &r4, &c);
+        tci_args_rrrrrc(insn, &r0, &r1, &r2, &r3, &r4, &c);
         info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %s, %s, %s",
                            op_name, str_r(r0), str_r(r1), str_r(r2),
                            str_r(r3), str_r(r4), str_c(c));
         break;
 
     case INDEX_op_mulu2_i32:
-        tci_args_rrrr(&tb_ptr, &r0, &r1, &r2, &r3);
+        tci_args_rrrr(insn, &r0, &r1, &r2, &r3);
         info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %s",
                            op_name, str_r(r0), str_r(r1),
                            str_r(r2), str_r(r3));
@@ -1276,7 +1137,7 @@ int print_insn_tci(bfd_vma addr, disassemble_info *info)
 
     case INDEX_op_add2_i32:
     case INDEX_op_sub2_i32:
-        tci_args_rrrrrr(&tb_ptr, &r0, &r1, &r2, &r3, &r4, &r5);
+        tci_args_rrrrrr(insn, &r0, &r1, &r2, &r3, &r4, &r5);
         info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %s, %s, %s",
                            op_name, str_r(r0), str_r(r1), str_r(r2),
                            str_r(r3), str_r(r4), str_r(r5));
@@ -1294,30 +1155,38 @@ int print_insn_tci(bfd_vma addr, disassemble_info *info)
         len += DIV_ROUND_UP(TARGET_LONG_BITS, TCG_TARGET_REG_BITS);
         switch (len) {
         case 2:
-            tci_args_rrm(&tb_ptr, &r0, &r1, &oi);
+            tci_args_rrm(insn, &r0, &r1, &oi);
             info->fprintf_func(info->stream, "%-12s  %s, %s, %x",
                                op_name, str_r(r0), str_r(r1), oi);
             break;
         case 3:
-            tci_args_rrrm(&tb_ptr, &r0, &r1, &r2, &oi);
+            tci_args_rrrm(insn, &r0, &r1, &r2, &oi);
             info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %x",
                                op_name, str_r(r0), str_r(r1), str_r(r2), oi);
             break;
         case 4:
-            tci_args_rrrrm(&tb_ptr, &r0, &r1, &r2, &r3, &oi);
-            info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %s, %x",
+            tci_args_rrrrr(insn, &r0, &r1, &r2, &r3, &r4);
+            info->fprintf_func(info->stream, "%-12s  %s, %s, %s, %s, %s",
                                op_name, str_r(r0), str_r(r1),
-                               str_r(r2), str_r(r3), oi);
+                               str_r(r2), str_r(r3), str_r(r4));
             break;
         default:
             g_assert_not_reached();
         }
         break;
 
+    case 0:
+        /* tcg_out_nop_fill uses zeros */
+        if (insn == 0) {
+            info->fprintf_func(info->stream, "align");
+            break;
+        }
+        /* fall through */
+
     default:
         info->fprintf_func(info->stream, "illegal opcode %d", op);
         break;
     }
 
-    return length;
+    return sizeof(insn);
 }
diff --git a/tcg/tci/tcg-target.c.inc b/tcg/tci/tcg-target.c.inc
index a1b5177abb..f74328dcbd 100644
--- a/tcg/tci/tcg-target.c.inc
+++ b/tcg/tci/tcg-target.c.inc
@@ -22,20 +22,7 @@
  * THE SOFTWARE.
  */
 
-/* TODO list:
- * - See TODO comments in code.
- */
-
-/* Marker for missing code. */
-#define TODO() \
-    do { \
-        fprintf(stderr, "TODO %s:%u: %s()\n", \
-                __FILE__, __LINE__, __func__); \
-        tcg_abort(); \
-    } while (0)
-
-/* Bitfield n...m (in 32 bit value). */
-#define BITS(n, m) (((0xffffffffU << (31 - n)) >> (31 - n + m)) << m)
+#include "../tcg-pool.c.inc"
 
 static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
 {
@@ -226,52 +213,16 @@ static const char *const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
                         intptr_t value, intptr_t addend)
 {
-    /* tcg_out_reloc always uses the same type, addend. */
-    tcg_debug_assert(type == sizeof(tcg_target_long));
+    intptr_t diff = value - (intptr_t)(code_ptr + 1);
+
     tcg_debug_assert(addend == 0);
-    tcg_debug_assert(value != 0);
-    if (TCG_TARGET_REG_BITS == 32) {
-        tcg_patch32(code_ptr, value);
-    } else {
-        tcg_patch64(code_ptr, value);
-    }
-    return true;
-}
-
-/* Write value (native size). */
-static void tcg_out_i(TCGContext *s, tcg_target_ulong v)
-{
-    if (TCG_TARGET_REG_BITS == 32) {
-        tcg_out32(s, v);
-    } else {
-        tcg_out64(s, v);
-    }
-}
-
-/* Write opcode. */
-static void tcg_out_op_t(TCGContext *s, TCGOpcode op)
-{
-    tcg_out8(s, op);
-    tcg_out8(s, 0);
-}
-
-/* Write register. */
-static void tcg_out_r(TCGContext *s, TCGArg t0)
-{
-    tcg_debug_assert(t0 < TCG_TARGET_NB_REGS);
-    tcg_out8(s, t0);
-}
-
-/* Write label. */
-static void tci_out_label(TCGContext *s, TCGLabel *label)
-{
-    if (label->has_value) {
-        tcg_out_i(s, label->u.value);
-        tcg_debug_assert(label->u.value);
-    } else {
-        tcg_out_reloc(s, s->code_ptr, sizeof(tcg_target_ulong), label, 0);
-        s->code_ptr += sizeof(tcg_target_ulong);
+    tcg_debug_assert(type == 20);
+
+    if (diff == sextract32(diff, 0, type)) {
+        tcg_patch32(code_ptr, deposit32(*code_ptr, 32 - type, type, diff));
+        return true;
     }
+    return false;
 }
 
 static void stack_bounds_check(TCGReg base, target_long offset)
@@ -285,239 +236,236 @@ static void stack_bounds_check(TCGReg base, target_long offset)
 
 static void tcg_out_op_l(TCGContext *s, TCGOpcode op, TCGLabel *l0)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tci_out_label(s, l0);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_out_reloc(s, s->code_ptr, 20, l0, 0);
+    insn = deposit32(insn, 0, 8, op);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_p(TCGContext *s, TCGOpcode op, void *p0)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
+    intptr_t diff;
 
-    tcg_out_op_t(s, op);
-    tcg_out_i(s, (uintptr_t)p0);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    /* Special case for exit_tb: map null -> 0. */
+    if (p0 == NULL) {
+        diff = 0;
+    } else {
+        diff = p0 - (void *)(s->code_ptr + 1);
+        tcg_debug_assert(diff != 0);
+        if (diff != sextract32(diff, 0, 20)) {
+            tcg_raise_tb_overflow(s);
+        }
+    }
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 12, 20, diff);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_v(TCGContext *s, TCGOpcode op)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
-
-    tcg_out_op_t(s, op);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_out32(s, (uint8_t)op);
 }
 
 static void tcg_out_op_ri(TCGContext *s, TCGOpcode op, TCGReg r0, int32_t i1)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out32(s, i1);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_debug_assert(i1 == sextract32(i1, 0, 20));
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 20, i1);
+    tcg_out32(s, insn);
 }
 
-#if TCG_TARGET_REG_BITS == 64
-static void tcg_out_op_rI(TCGContext *s, TCGOpcode op,
-                          TCGReg r0, uint64_t i1)
-{
-    uint8_t *old_code_ptr = s->code_ptr;
-
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out64(s, i1);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
-}
-#endif
-
 static void tcg_out_op_rl(TCGContext *s, TCGOpcode op, TCGReg r0, TCGLabel *l1)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tci_out_label(s, l1);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_out_reloc(s, s->code_ptr, 20, l1, 0);
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rr(TCGContext *s, TCGOpcode op, TCGReg r0, TCGReg r1)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrm(TCGContext *s, TCGOpcode op,
                            TCGReg r0, TCGReg r1, TCGArg m2)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out32(s, m2);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_debug_assert(m2 == extract32(m2, 0, 12));
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 20, 12, m2);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrr(TCGContext *s, TCGOpcode op,
                            TCGReg r0, TCGReg r1, TCGReg r2)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrs(TCGContext *s, TCGOpcode op,
                            TCGReg r0, TCGReg r1, intptr_t i2)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_debug_assert(i2 == (int32_t)i2);
-    tcg_out32(s, i2);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_debug_assert(i2 == sextract32(i2, 0, 16));
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 16, i2);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrrc(TCGContext *s, TCGOpcode op,
                             TCGReg r0, TCGReg r1, TCGReg r2, TCGCond c3)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-    tcg_out8(s, c3);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    insn = deposit32(insn, 20, 4, c3);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrrm(TCGContext *s, TCGOpcode op,
                             TCGReg r0, TCGReg r1, TCGReg r2, TCGArg m3)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-    tcg_out32(s, m3);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_debug_assert(m3 == extract32(m3, 0, 12));
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    insn = deposit32(insn, 20, 12, m3);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrrbb(TCGContext *s, TCGOpcode op, TCGReg r0,
                              TCGReg r1, TCGReg r2, uint8_t b3, uint8_t b4)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-    tcg_out8(s, b3);
-    tcg_out8(s, b4);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    tcg_debug_assert(b3 == extract32(b3, 0, 6));
+    tcg_debug_assert(b4 == extract32(b4, 0, 6));
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    insn = deposit32(insn, 20, 6, b3);
+    insn = deposit32(insn, 26, 6, b4);
+    tcg_out32(s, insn);
 }
 
-static void tcg_out_op_rrrrm(TCGContext *s, TCGOpcode op, TCGReg r0,
-                             TCGReg r1, TCGReg r2, TCGReg r3, TCGArg m4)
+static void tcg_out_op_rrrrr(TCGContext *s, TCGOpcode op, TCGReg r0,
+                             TCGReg r1, TCGReg r2, TCGReg r3, TCGReg r4)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-    tcg_out_r(s, r3);
-    tcg_out32(s, m4);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    insn = deposit32(insn, 20, 4, r3);
+    insn = deposit32(insn, 24, 4, r4);
+    tcg_out32(s, insn);
 }
 
 #if TCG_TARGET_REG_BITS == 32
 static void tcg_out_op_rrrr(TCGContext *s, TCGOpcode op,
                             TCGReg r0, TCGReg r1, TCGReg r2, TCGReg r3)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-    tcg_out_r(s, r3);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    insn = deposit32(insn, 20, 4, r3);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrrrrc(TCGContext *s, TCGOpcode op,
                               TCGReg r0, TCGReg r1, TCGReg r2,
                               TCGReg r3, TCGReg r4, TCGCond c5)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-    tcg_out_r(s, r3);
-    tcg_out_r(s, r4);
-    tcg_out8(s, c5);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    insn = deposit32(insn, 20, 4, r3);
+    insn = deposit32(insn, 24, 4, r4);
+    insn = deposit32(insn, 28, 4, c5);
+    tcg_out32(s, insn);
 }
 
 static void tcg_out_op_rrrrrr(TCGContext *s, TCGOpcode op,
                               TCGReg r0, TCGReg r1, TCGReg r2,
                               TCGReg r3, TCGReg r4, TCGReg r5)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
 
-    tcg_out_op_t(s, op);
-    tcg_out_r(s, r0);
-    tcg_out_r(s, r1);
-    tcg_out_r(s, r2);
-    tcg_out_r(s, r3);
-    tcg_out_r(s, r4);
-    tcg_out_r(s, r5);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    insn = deposit32(insn, 0, 8, op);
+    insn = deposit32(insn, 8, 4, r0);
+    insn = deposit32(insn, 12, 4, r1);
+    insn = deposit32(insn, 16, 4, r2);
+    insn = deposit32(insn, 20, 4, r3);
+    insn = deposit32(insn, 24, 4, r4);
+    insn = deposit32(insn, 28, 4, r5);
+    tcg_out32(s, insn);
 }
 #endif
 
+static void tcg_out_ldst(TCGContext *s, TCGOpcode op, TCGReg val,
+                         TCGReg base, intptr_t offset)
+{
+    stack_bounds_check(base, offset);
+    if (offset != sextract32(offset, 0, 16)) {
+        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, offset);
+        tcg_out_op_rrr(s, (TCG_TARGET_REG_BITS == 32
+                           ? INDEX_op_add_i32 : INDEX_op_add_i64),
+                       TCG_REG_TMP, TCG_REG_TMP, base);
+        base = TCG_REG_TMP;
+        offset = 0;
+    }
+    tcg_out_op_rrs(s, op, val, base, offset);
+}
+
 static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg val, TCGReg base,
                        intptr_t offset)
 {
-    stack_bounds_check(base, offset);
     switch (type) {
     case TCG_TYPE_I32:
-        tcg_out_op_rrs(s, INDEX_op_ld_i32, val, base, offset);
+        tcg_out_ldst(s, INDEX_op_ld_i32, val, base, offset);
         break;
 #if TCG_TARGET_REG_BITS == 64
     case TCG_TYPE_I64:
-        tcg_out_op_rrs(s, INDEX_op_ld_i64, val, base, offset);
+        tcg_out_ldst(s, INDEX_op_ld_i64, val, base, offset);
         break;
 #endif
     default:
@@ -547,22 +495,32 @@ static void tcg_out_movi(TCGContext *s, TCGType type,
 {
     switch (type) {
     case TCG_TYPE_I32:
-        tcg_out_op_ri(s, INDEX_op_tci_movi_i32, ret, arg);
-        break;
 #if TCG_TARGET_REG_BITS == 64
+        arg = (int32_t)arg;
+        /* fall through */
     case TCG_TYPE_I64:
-        tcg_out_op_rI(s, INDEX_op_tci_movi_i64, ret, arg);
-        break;
 #endif
+        break;
     default:
         g_assert_not_reached();
     }
+
+    if (arg == sextract32(arg, 0, 20)) {
+        tcg_out_op_ri(s, INDEX_op_tci_movi, ret, arg);
+    } else {
+        tcg_insn_unit insn = 0;
+
+        new_pool_label(s, arg, 20, s->code_ptr, 0);
+        insn = deposit32(insn, 0, 8, INDEX_op_tci_movl);
+        insn = deposit32(insn, 8, 4, ret);
+        tcg_out32(s, insn);
+    }
 }
 
 static void tcg_out_call(TCGContext *s, const tcg_insn_unit *func,
                          ffi_cif *cif)
 {
-    uint8_t *old_code_ptr = s->code_ptr;
+    tcg_insn_unit insn = 0;
     uint8_t which;
 
     if (cif->rtype == &ffi_type_void) {
@@ -573,12 +531,10 @@ static void tcg_out_call(TCGContext *s, const tcg_insn_unit *func,
         tcg_debug_assert(cif->rtype->size == 8);
         which = 2;
     }
-    tcg_out_op_t(s, INDEX_op_call);
-    tcg_out8(s, which);
-    tcg_out_i(s, (uintptr_t)func);
-    tcg_out_i(s, (uintptr_t)cif);
-
-    old_code_ptr[1] = s->code_ptr - old_code_ptr;
+    new_pool_l2(s, 20, s->code_ptr, 0, (uintptr_t)func, (uintptr_t)cif);
+    insn = deposit32(insn, 0, 8, INDEX_op_call);
+    insn = deposit32(insn, 8, 4, which);
+    tcg_out32(s, insn);
 }
 
 #if TCG_TARGET_REG_BITS == 64
@@ -637,8 +593,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
     case INDEX_op_st_i32:
     CASE_64(st32)
     CASE_64(st)
-        stack_bounds_check(args[1], args[2]);
-        tcg_out_op_rrs(s, opc, args[0], args[1], args[2]);
+        tcg_out_ldst(s, opc, args[0], args[1], args[2]);
         break;
 
     CASE_32_64(add)
@@ -731,8 +686,9 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
         } else if (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS) {
             tcg_out_op_rrrm(s, opc, args[0], args[1], args[2], args[3]);
         } else {
-            tcg_out_op_rrrrm(s, opc, args[0], args[1],
-                             args[2], args[3], args[4]);
+            tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP, args[4]);
+            tcg_out_op_rrrrr(s, opc, args[0], args[1],
+                             args[2], args[3], TCG_REG_TMP);
         }
         break;
 
@@ -778,6 +734,11 @@ static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
     return ct & TCG_CT_CONST;
 }
 
+static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
+{
+    memset(p, 0, sizeof(*p) * count);
+}
+
 static void tcg_target_init(TCGContext *s)
 {
 #if defined(CONFIG_DEBUG_TCG_INTERPRETER)
diff --git a/tcg/tci/README b/tcg/tci/README
index 9bb7d7a5d3..f72a40a395 100644
--- a/tcg/tci/README
+++ b/tcg/tci/README
@@ -23,10 +23,12 @@ This is what TCI (Tiny Code Interpreter) does.
 Like each TCG host frontend, TCI implements the code generator in
 tcg-target.c.inc, tcg-target.h. Both files are in directory tcg/tci.
 
-The additional file tcg/tci.c adds the interpreter.
+The additional file tcg/tci.c adds the interpreter and disassembler.
 
-The bytecode consists of opcodes (same numeric values as those used by
-TCG), command length and arguments of variable size and number.
+The bytecode consists of opcodes (with only a few exceptions, with
+the same same numeric values and semantics as used by TCG), and up
+to six arguments packed into a 32-bit integer.  See comments in tci.c
+for details on the encoding.
 
 3) Usage
 
@@ -39,11 +41,6 @@ suggest using this option. Setting it automatically would need
 additional code in configure which must be fixed when new native TCG
 implementations are added.
 
-System emulation should work on any 32 or 64 bit host.
-User mode emulation might work. Maybe a new linker script (*.ld)
-is needed. Byte order might be wrong (on big endian hosts)
-and need fixes in configure.
-
 For hosts with native TCG, the interpreter TCI can be enabled by
 
         configure --enable-tcg-interpreter
@@ -118,13 +115,6 @@ u1 = linux-user-test works
   in the interpreter. These opcodes raise a runtime exception, so it is
   possible to see where code must be added.
 
-* The pseudo code is not optimized and still ugly. For hosts with special
-  alignment requirements, it needs some fixes (maybe aligned bytecode
-  would also improve speed for hosts which support byte alignment).
-
-* A better disassembler for the pseudo code would be nice (a very primitive
-  disassembler is included in tcg-target.c.inc).
-
 * It might be useful to have a runtime option which selects the native TCG
   or TCI, so QEMU would have to include two TCGs. Today, selecting TCI
   is a configure option, so you need two compilations of QEMU.
-- 
2.25.1



  parent reply	other threads:[~2021-06-19 18:27 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-06-19 18:14 [PULL 00/33] tcg patch queue Richard Henderson
2021-06-19 18:14 ` [PULL 01/33] tcg: Combine dh_is_64bit and dh_is_signed to dh_typecode Richard Henderson
2021-06-19 18:14 ` [PULL 02/33] tcg: Add tcg_call_flags Richard Henderson
2021-06-19 18:14 ` [PULL 03/33] accel/tcg/plugin-gen: Drop inline markers Richard Henderson
2021-06-19 18:14 ` [PULL 04/33] plugins: Drop tcg_flags from struct qemu_plugin_dyn_cb Richard Henderson
2021-06-19 18:14 ` [PULL 05/33] accel/tcg: Add tcg call flags to plugins helpers Richard Henderson
2021-06-19 18:14 ` [PULL 06/33] tcg: Store the TCGHelperInfo in the TCGOp for call Richard Henderson
2021-06-19 18:14 ` [PULL 07/33] tcg: Add tcg_call_func Richard Henderson
2021-06-19 18:14 ` [PULL 08/33] tcg: Build ffi data structures for helpers Richard Henderson
2021-06-19 18:14 ` [PULL 09/33] tcg/tci: Improve tcg_target_call_clobber_regs Richard Henderson
2021-06-19 18:14 ` [PULL 10/33] tcg/tci: Move call-return regs to end of tcg_target_reg_alloc_order Richard Henderson
2021-06-19 18:14 ` [PULL 11/33] tcg/tci: Use ffi for calls Richard Henderson
2021-06-19 18:14 ` [PULL 12/33] tcg/tci: Reserve r13 for a temporary Richard Henderson
2021-06-19 18:14 ` [PULL 13/33] tcg/tci: Emit setcond before brcond Richard Henderson
2021-06-19 18:14 ` [PULL 14/33] tcg/tci: Remove tci_write_reg Richard Henderson
2021-06-19 18:14 ` Richard Henderson [this message]
2021-06-19 18:14 ` [PULL 16/33] tcg/tci: Implement goto_ptr Richard Henderson
2021-06-19 18:14 ` [PULL 17/33] tcg/tci: Implement movcond Richard Henderson
2021-06-19 18:14 ` [PULL 18/33] tcg/tci: Implement andc, orc, eqv, nand, nor Richard Henderson
2021-06-19 18:14 ` [PULL 19/33] tcg/tci: Implement extract, sextract Richard Henderson
2021-06-19 18:14 ` [PULL 20/33] tcg/tci: Implement clz, ctz, ctpop Richard Henderson
2021-06-19 18:14 ` [PULL 21/33] tcg/tci: Implement mulu2, muls2 Richard Henderson
2021-06-19 18:14 ` [PULL 22/33] tcg/tci: Implement add2, sub2 Richard Henderson
2021-06-19 18:14 ` [PULL 23/33] tcg/tci: Split out tci_qemu_ld, tci_qemu_st Richard Henderson
2021-06-19 18:14 ` [PULL 24/33] Revert "tcg/tci: Use exec/cpu_ldst.h interfaces" Richard Henderson
2021-06-19 18:14 ` [PULL 25/33] tcg/tci: Remove the qemu_ld/st_type macros Richard Henderson
2021-06-19 18:14 ` [PULL 26/33] tcg/tci: Use {set,clear}_helper_retaddr Richard Henderson
2021-06-19 18:14 ` [PULL 27/33] tests/tcg: Increase timeout for TCI Richard Henderson
2021-06-19 18:14 ` [PULL 28/33] accel/tcg: Probe the proper permissions for atomic ops Richard Henderson
2021-06-19 18:14 ` [PULL 29/33] tcg/sparc: Fix temp_allocate_frame vs sparc stack bias Richard Henderson
2021-06-19 18:14 ` [PULL 30/33] tcg: Allocate sufficient storage in temp_allocate_frame Richard Henderson
2021-09-01 10:52   ` Richard W.M. Jones
2021-09-01 12:55     ` Daniel P. Berrangé
2021-06-19 18:14 ` [PULL 31/33] tcg: Restart when exhausting the stack frame Richard Henderson
2021-06-19 18:14 ` [PULL 32/33] tcg: expose TCGCond manipulation routines Richard Henderson
2021-06-19 18:14 ` [PULL 33/33] util/oslib-win32: Fix fatal assertion in qemu_try_memalign Richard Henderson
2021-06-19 21:50 ` [PULL 00/33] tcg patch queue Richard Henderson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20210619181452.877683-16-richard.henderson@linaro.org \
    --to=richard.henderson@linaro.org \
    --cc=f4bug@amsat.org \
    --cc=peter.maydell@linaro.org \
    --cc=qemu-devel@nongnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.