diff --git a/lightening/lightening.c b/lightening/lightening.c
index 5a98eb43c..272c35bdd 100644
--- a/lightening/lightening.c
+++ b/lightening/lightening.c
@@ -467,3 +467,511 @@ jit_patch_there(jit_state_t* _jit, jit_reloc_t reloc, jit_pointer_t addr)
 #define IMPL_INSTRUCTION(kind, stem) JIT_IMPL_##kind(stem)
 FOR_EACH_INSTRUCTION(IMPL_INSTRUCTION)
 #undef IMPL_INSTRUCTION
+
+static void
+abi_imm_to_gpr(jit_state_t *_jit, enum jit_operand_abi abi, jit_gpr_t dst,
+               intptr_t imm)
+{
+  switch (abi) {
+  case JIT_OPERAND_ABI_UINT8:
+    ASSERT(0 <= imm);
+    ASSERT(imm <= UINT8_MAX);
+    break;
+  case JIT_OPERAND_ABI_INT8:
+    ASSERT(INT8_MIN <= imm);
+    ASSERT(imm <= INT8_MAX);
+    break;
+  case JIT_OPERAND_ABI_UINT16:
+    ASSERT(0 <= imm);
+    ASSERT(imm <= UINT16_MAX);
+    break;
+  case JIT_OPERAND_ABI_INT16:
+    ASSERT(INT16_MIN <= imm);
+    ASSERT(imm <= INT16_MAX);
+    break;
+  case JIT_OPERAND_ABI_UINT32:
+    ASSERT(0 <= imm);
+    ASSERT(imm <= UINT32_MAX);
+    break;
+  case JIT_OPERAND_ABI_INT32:
+    ASSERT(INT32_MIN <= imm);
+    ASSERT(imm <= INT32_MAX);
+    break;
+#if __WORDSIZE > 32
+  case JIT_OPERAND_ABI_UINT64:
+  case JIT_OPERAND_ABI_INT64:
+    break;
+#endif
+  case JIT_OPERAND_ABI_POINTER:
+    break;
+  default:
+    abort();
+  }
+  jit_movi (_jit, dst, imm);
+}
+
+static void
+abi_gpr_to_mem(jit_state_t *_jit, enum jit_operand_abi abi,
+               jit_gpr_t base, ptrdiff_t offset, jit_gpr_t src)
+{
+  switch (abi) {
+  case JIT_OPERAND_ABI_UINT8:
+  case JIT_OPERAND_ABI_INT8:
+    jit_stxi_c(_jit, offset, base, src);
+    break;
+  case JIT_OPERAND_ABI_UINT16:
+  case JIT_OPERAND_ABI_INT16:
+    jit_stxi_s(_jit, offset, base, src);
+    break;
+  case JIT_OPERAND_ABI_UINT32:
+  case JIT_OPERAND_ABI_INT32:
+#if __WORDSIZE == 32
+  case JIT_OPERAND_ABI_POINTER:
+#endif
+    jit_stxi_i(_jit, offset, base, src);
+    break;
+#if __WORDSIZE == 64
+  case JIT_OPERAND_ABI_UINT64:
+  case JIT_OPERAND_ABI_INT64:
+  case JIT_OPERAND_ABI_POINTER:
+    jit_stxi_l(_jit, offset, base, src);
+    break;
+#endif
+  default:
+    abort();
+  }
+}
+
+static void
+abi_fpr_to_mem(jit_state_t *_jit, enum jit_operand_abi abi,
+               jit_gpr_t base, ptrdiff_t offset, jit_fpr_t src)
+{
+  switch (abi) {
+  case JIT_OPERAND_ABI_FLOAT:
+    jit_stxi_f(_jit, offset, base, src);
+    break;
+  case JIT_OPERAND_ABI_DOUBLE:
+    jit_stxi_d(_jit, offset, base, src);
+    break;
+  default:
+    abort();
+  }
+}
+
+static void
+abi_mem_to_gpr(jit_state_t *_jit, enum jit_operand_abi abi,
+               jit_gpr_t dst, jit_gpr_t base, ptrdiff_t offset)
+{
+  switch (abi) {
+  case JIT_OPERAND_ABI_UINT8:
+    jit_ldxi_uc(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_INT8:
+    jit_ldxi_c(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_UINT16:
+    jit_ldxi_us(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_INT16:
+    jit_ldxi_s(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_UINT32:
+    jit_ldxi_ui(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_INT32:
+    jit_ldxi_i(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_UINT64:
+    jit_ldxi_l(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_INT64:
+    jit_ldxi_l(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_POINTER:
+    jit_ldxi_l(_jit, dst, base, offset);
+    break;
+  default:
+    abort();
+  }
+}
+
+static void
+abi_mem_to_fpr(jit_state_t *_jit, enum jit_operand_abi abi,
+               jit_fpr_t dst, jit_gpr_t base, ptrdiff_t offset)
+{
+  switch (abi) {
+  case JIT_OPERAND_ABI_FLOAT:
+    jit_ldxi_f(_jit, dst, base, offset);
+    break;
+  case JIT_OPERAND_ABI_DOUBLE:
+    jit_ldxi_d(_jit, dst, base, offset);
+    break;
+  default:
+    abort();
+  }
+}
+
+static void
+abi_imm_to_mem(jit_state_t *_jit, enum jit_operand_abi abi, jit_gpr_t base,
+               ptrdiff_t offset, intmax_t imm)
+{
+  ASSERT(!is_fpr_arg(abi));
+
+  jit_gpr_t tmp = get_temp_gpr(_jit);
+  abi_imm_to_gpr(_jit, abi, tmp, imm);
+  abi_gpr_to_mem(_jit, abi, base, offset, tmp);
+  unget_temp_gpr(_jit);
+}
+
+static void
+abi_mem_to_mem(jit_state_t *_jit, enum jit_operand_abi abi, jit_gpr_t base,
+               ptrdiff_t offset, jit_gpr_t src_base, ptrdiff_t src_offset)
+{
+  if (is_gpr_arg (abi)) {
+    jit_gpr_t tmp = get_temp_gpr(_jit);
+    abi_mem_to_gpr(_jit, abi, tmp, src_base, src_offset);
+    abi_gpr_to_mem(_jit, abi, base, offset, tmp);
+    unget_temp_gpr(_jit);
+  } else {
+    jit_fpr_t tmp = get_temp_xpr(_jit);
+    abi_mem_to_fpr(_jit, abi, tmp, src_base, src_offset);
+    abi_fpr_to_mem(_jit, abi, base, offset, tmp);
+    unget_temp_xpr(_jit);
+  }
+}
+
+#define MOVE_KIND(a, b) ((((int) a) << 4) | ((int) b))
+
+#define MOVE_KIND_ENUM(a, b) \
+  MOVE_##a##_TO_##b = MOVE_KIND(JIT_OPERAND_KIND_##a, JIT_OPERAND_KIND_##b)
+enum move_kind {
+  MOVE_KIND_ENUM(IMM, GPR),
+  MOVE_KIND_ENUM(GPR, GPR),
+  MOVE_KIND_ENUM(MEM, GPR),
+  MOVE_KIND_ENUM(FPR, FPR),
+  MOVE_KIND_ENUM(MEM, FPR),
+  MOVE_KIND_ENUM(IMM, MEM),
+  MOVE_KIND_ENUM(GPR, MEM),
+  MOVE_KIND_ENUM(FPR, MEM),
+  MOVE_KIND_ENUM(MEM, MEM)
+};
+#undef MOVE_KIND_ENUM
+
+static void
+move_operand(jit_state_t *_jit, jit_operand_t dst, jit_operand_t src)
+{
+  switch (MOVE_KIND (src.kind, dst.kind)) {
+  case MOVE_IMM_TO_GPR:
+    return abi_imm_to_gpr(_jit, src.abi, dst.loc.gpr.gpr, src.loc.imm);
+
+  case MOVE_GPR_TO_GPR:
+    return jit_movr(_jit, dst.loc.gpr.gpr, src.loc.gpr.gpr);
+
+  case MOVE_MEM_TO_GPR:
+    return abi_mem_to_gpr(_jit, src.abi, dst.loc.gpr.gpr, src.loc.mem.base,
+                          src.loc.mem.offset);
+
+  case MOVE_FPR_TO_FPR:
+    return jit_movr_d(_jit, dst.loc.fpr, src.loc.fpr);
+
+  case MOVE_MEM_TO_FPR:
+    return abi_mem_to_fpr(_jit, src.abi, dst.loc.fpr, src.loc.mem.base,
+                          src.loc.mem.offset);
+
+  case MOVE_IMM_TO_MEM:
+    return abi_imm_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
+                          src.loc.imm);
+
+  case MOVE_GPR_TO_MEM:
+    return abi_gpr_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
+                          src.loc.gpr.gpr);
+
+  case MOVE_FPR_TO_MEM:
+    return abi_fpr_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
+                          src.loc.fpr);
+
+  case MOVE_MEM_TO_MEM:
+    return abi_mem_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
+                          src.loc.mem.base, src.loc.mem.offset);
+
+  default:
+    abort();
+  }
+}
+
+// A direct transliteration of "Tilting at windmills with Coq: formal
+// verification of a compilation algorithm for parallel moves" by
+// Laurence Rideau, Bernard Paul Serpette, and Xavier Leroy:
+// https://xavierleroy.org/publi/parallel-move.pdf
+
+enum move_status { TO_MOVE, BEING_MOVED, MOVED };
+
+static inline int
+already_in_place(jit_operand_t src, jit_operand_t dst)
+{
+  switch (MOVE_KIND(src.kind, dst.kind)) {
+  case MOVE_GPR_TO_GPR:
+    return jit_same_gprs (src.loc.gpr.gpr, dst.loc.gpr.gpr);
+  case MOVE_FPR_TO_FPR:
+    return jit_same_fprs (src.loc.fpr, dst.loc.fpr);
+  case MOVE_MEM_TO_MEM:
+    return jit_same_gprs (src.loc.mem.base, dst.loc.mem.base) &&
+      src.loc.mem.offset == dst.loc.mem.offset;
+  default:
+    return 0;
+  }
+}
+
+static inline int
+write_would_clobber(jit_operand_t src, jit_operand_t dst)
+{
+  if (already_in_place (src, dst))
+    return 1;
+
+  if (MOVE_KIND(src.kind, dst.kind) == MOVE_MEM_TO_GPR)
+    return jit_same_gprs(src.loc.mem.base, dst.loc.gpr.gpr);
+
+  return 0;
+}
+
+static inline ptrdiff_t
+operand_addend(jit_operand_t op)
+{
+  switch (op.kind) {
+  case JIT_OPERAND_KIND_GPR:
+    return op.loc.gpr.addend;
+  case JIT_OPERAND_KIND_MEM:
+    return op.loc.mem.addend;
+  default:
+    abort();
+  }
+}
+
+static void
+move_one(jit_state_t *_jit, jit_operand_t *dst, jit_operand_t *src,
+         size_t argc, enum move_status *status, size_t i)
+{
+  int tmp_gpr = 0, tmp_fpr = 0;
+
+  if (already_in_place(src[i], dst[i]))
+    return;
+
+  status[i] = BEING_MOVED;
+  for (size_t j = 0; j < argc; j++) {
+    if (write_would_clobber(src[j], dst[i])) {
+      switch (status[j]) {
+      case TO_MOVE:
+        move_one(_jit, dst, src, argc, status, j);
+        break;
+      case BEING_MOVED: {
+        jit_operand_t tmp;
+        if (is_fpr_arg (src[j].kind)) {
+          tmp_fpr = 1;
+          tmp = jit_operand_fpr(src[j].abi, get_temp_xpr(_jit));
+        } else {
+          tmp_gpr = 1;
+          /* Preserve addend, if any, from source operand, to be applied
+             at the end.  */
+          tmp = jit_operand_gpr_with_addend(src[j].abi, get_temp_gpr(_jit),
+                                            operand_addend(src[j]));
+        }
+        move_operand (_jit, tmp, src[j]);
+        src[j] = tmp;
+        break;
+      }
+      case MOVED:
+        break;
+      default:
+        abort ();
+      }
+    }
+  }
+
+  move_operand (_jit, dst[i], src[i]);
+  status[i] = MOVED;
+  if (tmp_gpr)
+    unget_temp_gpr(_jit);
+  else if (tmp_fpr)
+    unget_temp_xpr(_jit);
+}
+
+static void
+apply_addend(jit_state_t *_jit, jit_operand_t dst, jit_operand_t src)
+{
+  switch (MOVE_KIND(src.kind, dst.kind)) {
+  case MOVE_GPR_TO_GPR:
+  case MOVE_MEM_TO_GPR:
+    if (operand_addend(src))
+      jit_addi(_jit, dst.loc.gpr.gpr, dst.loc.gpr.gpr, operand_addend(src));
+    break;
+  case MOVE_GPR_TO_MEM:
+  case MOVE_MEM_TO_MEM:
+    if (operand_addend(src)) {
+      jit_gpr_t tmp = get_temp_gpr(_jit);
+      abi_mem_to_gpr(_jit, dst.abi, tmp, dst.loc.mem.base, dst.loc.mem.offset);
+      jit_addi(_jit, tmp, tmp, operand_addend(src));
+      abi_gpr_to_mem(_jit, dst.abi, dst.loc.mem.base, dst.loc.mem.offset, tmp);
+    }
+    break;
+  default:
+    break;
+  }
+}
+
+/* Preconditions: No dest operand is IMM.  No dest operand aliases
+   another dest operand.  No dest MEM operand uses a base register which
+   is used as a dest GPR.  No dst operand has an addend.  The registers
+   returned by get_temp_gpr and get_temp_fpr do not appear in source or
+   dest args.  */
+void
+jit_move_operands(jit_state_t *_jit, jit_operand_t *dst, jit_operand_t *src,
+                  size_t argc)
+{
+  // Check preconditions, except the condition about tmp registers.
+  {
+    uint64_t src_gprs = 0;
+    uint64_t dst_gprs = 0;
+    uint64_t dst_fprs = 0;
+    uint64_t dst_mem_base_gprs = 0;
+    for (size_t i = 0; i < argc; i++) {
+      switch (src[i].kind) {
+      case JIT_OPERAND_KIND_GPR:
+        src_gprs |= 1ULL << jit_gpr_regno(src[i].loc.gpr.gpr);
+        break;
+      case JIT_OPERAND_KIND_FPR:
+      case JIT_OPERAND_KIND_IMM:
+      case JIT_OPERAND_KIND_MEM:
+        break;
+      default:
+        abort();
+      }
+      switch (dst[i].kind) {
+      case JIT_OPERAND_KIND_GPR: {
+        ASSERT(dst[i].loc.gpr.addend == 0);
+        uint64_t bit = 1ULL << jit_gpr_regno(dst[i].loc.gpr.gpr);
+        ASSERT((dst_gprs & bit) == 0);
+        dst_gprs |= bit;
+        break;
+      }
+      case JIT_OPERAND_KIND_FPR: {
+        uint64_t bit = 1ULL << jit_fpr_regno(dst[i].loc.fpr);
+        ASSERT((dst_fprs & bit) == 0);
+        dst_fprs |= bit;
+        break;
+      }
+      case JIT_OPERAND_KIND_MEM: {
+        ASSERT(dst[i].loc.mem.addend == 0);
+        uint64_t bit = 1ULL << jit_gpr_regno(dst[i].loc.mem.base);
+        dst_mem_base_gprs |= bit;
+        break;
+      }
+      case JIT_OPERAND_KIND_IMM:
+      default:
+        abort();
+        break;
+      }
+    }
+    ASSERT(((src_gprs | dst_gprs) & dst_mem_base_gprs) == 0);
+  }
+
+  enum move_status status[argc];
+  for (size_t i = 0; i < argc; i++)
+    status[i] = TO_MOVE;
+  for (size_t i = 0; i < argc; i++)
+    if (status[i] == TO_MOVE)
+      move_one(_jit, dst, src, argc, status, i);
+
+  // Apply addends at the end.  We could do it earlier in some cases but
+  // at least at the end we know that an in-place increment of one
+  // operand won't alias another.
+  for (size_t i = 0; i < argc; i++)
+    apply_addend(_jit, dst[i], src[i]);
+}
+
+// Precondition: stack is already aligned.
+static size_t
+prepare_call_args(jit_state_t *_jit, size_t argc, jit_operand_t args[])
+{
+  jit_operand_t dst[argc];
+  struct abi_arg_iterator iter;
+  
+  // Compute shuffle destinations and space for spilled arguments.
+  reset_abi_arg_iterator(&iter, argc, args);
+  for (size_t i = 0; i < argc; i++)
+    next_abi_arg(&iter, &dst[i]);
+
+  size_t stack_size = iter.stack_size;
+
+  // Reserve space for spilled arguments, and fix up SP-relative
+  // operands.
+  if (stack_size)
+    {
+      // Align stack to 16-byte boundaries on 64-bit targets.
+      if (__WORDSIZE == 64)
+        stack_size = (stack_size + 15) & ~15;
+      jit_subi(_jit, JIT_SP, JIT_SP, stack_size);
+      for (size_t i = 0; i < argc; i++) {
+        switch(args[i].kind) {
+        case JIT_OPERAND_KIND_GPR:
+          if (jit_same_gprs (args[i].loc.mem.base, JIT_SP))
+            args[i].loc.gpr.addend += stack_size;
+          break;
+        case JIT_OPERAND_KIND_MEM:
+          if (jit_same_gprs (args[i].loc.mem.base, JIT_SP))
+            args[i].loc.mem.offset += stack_size;
+          break;
+        default:
+          break;
+        }
+      }
+    }
+
+  jit_move_operands(_jit, dst, args, argc);
+
+  return stack_size;
+}
+
+void
+jit_calli(jit_state_t *_jit, jit_pointer_t f, size_t argc, jit_operand_t args[])
+{
+  size_t spill_size = prepare_call_args(_jit, argc, args);
+
+  calli(_jit, (jit_word_t)f);
+
+  if (spill_size)
+    jit_addi(_jit, JIT_SP, JIT_SP, spill_size);
+}
+
+void
+jit_callr(jit_state_t *_jit, jit_gpr_t f, size_t argc, jit_operand_t args[])
+{
+  size_t spill_size = prepare_call_args(_jit, argc, args);
+
+  callr(_jit, jit_gpr_regno(f));
+
+  if (spill_size)
+    jit_addi(_jit, JIT_SP, JIT_SP, spill_size);
+}
+
+void
+jit_locate_args(jit_state_t *_jit, size_t argc, jit_operand_t args[])
+{
+  struct abi_arg_iterator iter;
+    
+  reset_abi_arg_iterator(&iter, argc, args);
+  for (size_t i = 0; i < argc; i++)
+    next_abi_arg(&iter, &args[i]);
+}
+
+/* Precondition: args are distinct locations of type GPR or FPR.  All
+   addends of arg operands are zero.  No GPR arg is SP.  */
+void
+jit_load_args(jit_state_t *_jit, size_t argc, jit_operand_t args[])
+{
+  jit_operand_t src[argc];
+
+  memcpy(src, args, sizeof(src[0]) * argc);
+
+  jit_locate_args(_jit, argc, src);
+  jit_move_operands(_jit, args, src, argc);
+}
diff --git a/lightening/x86.c b/lightening/x86.c
index 144816dfd..153997d45 100644
--- a/lightening/x86.c
+++ b/lightening/x86.c
@@ -249,426 +249,6 @@ is_gpr_arg(enum jit_operand_abi arg)
   return !is_fpr_arg(arg);
 }
 
-static void
-abi_imm_to_gpr(jit_state_t *_jit, enum jit_operand_abi abi, jit_gpr_t dst,
-               intptr_t imm)
-{
-  switch (abi) {
-  case JIT_OPERAND_ABI_UINT8:
-    ASSERT(0 <= imm);
-    ASSERT(imm <= UINT8_MAX);
-    break;
-  case JIT_OPERAND_ABI_INT8:
-    ASSERT(INT8_MIN <= imm);
-    ASSERT(imm <= INT8_MAX);
-    break;
-  case JIT_OPERAND_ABI_UINT16:
-    ASSERT(0 <= imm);
-    ASSERT(imm <= UINT16_MAX);
-    break;
-  case JIT_OPERAND_ABI_INT16:
-    ASSERT(INT16_MIN <= imm);
-    ASSERT(imm <= INT16_MAX);
-    break;
-  case JIT_OPERAND_ABI_UINT32:
-    ASSERT(0 <= imm);
-    ASSERT(imm <= UINT32_MAX);
-    break;
-  case JIT_OPERAND_ABI_INT32:
-    ASSERT(INT32_MIN <= imm);
-    ASSERT(imm <= INT32_MAX);
-    break;
-#if __WORDSIZE > 32
-  case JIT_OPERAND_ABI_UINT64:
-  case JIT_OPERAND_ABI_INT64:
-    break;
-#endif
-  case JIT_OPERAND_ABI_POINTER:
-    break;
-  default:
-    abort();
-  }
-  jit_movi (_jit, dst, imm);
-}
-
-static void
-abi_gpr_to_mem(jit_state_t *_jit, enum jit_operand_abi abi,
-               jit_gpr_t base, ptrdiff_t offset, jit_gpr_t src)
-{
-  switch (abi) {
-  case JIT_OPERAND_ABI_UINT8:
-  case JIT_OPERAND_ABI_INT8:
-    jit_stxi_c(_jit, offset, base, src);
-    break;
-  case JIT_OPERAND_ABI_UINT16:
-  case JIT_OPERAND_ABI_INT16:
-    jit_stxi_s(_jit, offset, base, src);
-    break;
-  case JIT_OPERAND_ABI_UINT32:
-  case JIT_OPERAND_ABI_INT32:
-#if __WORDSIZE == 32
-  case JIT_OPERAND_ABI_POINTER:
-#endif
-    jit_stxi_i(_jit, offset, base, src);
-    break;
-#if __WORDSIZE == 64
-  case JIT_OPERAND_ABI_UINT64:
-  case JIT_OPERAND_ABI_INT64:
-  case JIT_OPERAND_ABI_POINTER:
-    jit_stxi_l(_jit, offset, base, src);
-    break;
-#endif
-  default:
-    abort();
-  }
-}
-
-static void
-abi_fpr_to_mem(jit_state_t *_jit, enum jit_operand_abi abi,
-               jit_gpr_t base, ptrdiff_t offset, jit_fpr_t src)
-{
-  switch (abi) {
-  case JIT_OPERAND_ABI_FLOAT:
-    jit_stxi_f(_jit, offset, base, src);
-    break;
-  case JIT_OPERAND_ABI_DOUBLE:
-    jit_stxi_d(_jit, offset, base, src);
-    break;
-  default:
-    abort();
-  }
-}
-
-static void
-abi_mem_to_gpr(jit_state_t *_jit, enum jit_operand_abi abi,
-               jit_gpr_t dst, jit_gpr_t base, ptrdiff_t offset)
-{
-  switch (abi) {
-  case JIT_OPERAND_ABI_UINT8:
-    jit_ldxi_uc(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_INT8:
-    jit_ldxi_c(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_UINT16:
-    jit_ldxi_us(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_INT16:
-    jit_ldxi_s(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_UINT32:
-    jit_ldxi_ui(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_INT32:
-    jit_ldxi_i(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_UINT64:
-    jit_ldxi_l(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_INT64:
-    jit_ldxi_l(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_POINTER:
-    jit_ldxi_l(_jit, dst, base, offset);
-    break;
-  default:
-    abort();
-  }
-}
-
-static void
-abi_mem_to_fpr(jit_state_t *_jit, enum jit_operand_abi abi,
-               jit_fpr_t dst, jit_gpr_t base, ptrdiff_t offset)
-{
-  switch (abi) {
-  case JIT_OPERAND_ABI_FLOAT:
-    jit_ldxi_f(_jit, dst, base, offset);
-    break;
-  case JIT_OPERAND_ABI_DOUBLE:
-    jit_ldxi_d(_jit, dst, base, offset);
-    break;
-  default:
-    abort();
-  }
-}
-
-static void
-abi_imm_to_mem(jit_state_t *_jit, enum jit_operand_abi abi, jit_gpr_t base,
-               ptrdiff_t offset, intmax_t imm)
-{
-  ASSERT(!is_fpr_arg(abi));
-
-  jit_gpr_t tmp = get_temp_gpr(_jit);
-  abi_imm_to_gpr(_jit, abi, tmp, imm);
-  abi_gpr_to_mem(_jit, abi, base, offset, tmp);
-  unget_temp_gpr(_jit);
-}
-
-static void
-abi_mem_to_mem(jit_state_t *_jit, enum jit_operand_abi abi, jit_gpr_t base,
-               ptrdiff_t offset, jit_gpr_t src_base, ptrdiff_t src_offset)
-{
-  if (is_gpr_arg (abi)) {
-    jit_gpr_t tmp = get_temp_gpr(_jit);
-    abi_mem_to_gpr(_jit, abi, tmp, src_base, src_offset);
-    abi_gpr_to_mem(_jit, abi, base, offset, tmp);
-    unget_temp_gpr(_jit);
-  } else {
-    jit_fpr_t tmp = get_temp_xpr(_jit);
-    abi_mem_to_fpr(_jit, abi, tmp, src_base, src_offset);
-    abi_fpr_to_mem(_jit, abi, base, offset, tmp);
-    unget_temp_xpr(_jit);
-  }
-}
-
-#define MOVE_KIND(a, b) ((((int) a) << 4) | ((int) b))
-
-#define MOVE_KIND_ENUM(a, b) \
-  MOVE_##a##_TO_##b = MOVE_KIND(JIT_OPERAND_KIND_##a, JIT_OPERAND_KIND_##b)
-enum move_kind {
-  MOVE_KIND_ENUM(IMM, GPR),
-  MOVE_KIND_ENUM(GPR, GPR),
-  MOVE_KIND_ENUM(MEM, GPR),
-  MOVE_KIND_ENUM(FPR, FPR),
-  MOVE_KIND_ENUM(MEM, FPR),
-  MOVE_KIND_ENUM(IMM, MEM),
-  MOVE_KIND_ENUM(GPR, MEM),
-  MOVE_KIND_ENUM(FPR, MEM),
-  MOVE_KIND_ENUM(MEM, MEM)
-};
-#undef MOVE_KIND_ENUM
-
-static void
-move_operand(jit_state_t *_jit, jit_operand_t dst, jit_operand_t src)
-{
-  switch (MOVE_KIND (src.kind, dst.kind)) {
-  case MOVE_IMM_TO_GPR:
-    return abi_imm_to_gpr(_jit, src.abi, dst.loc.gpr.gpr, src.loc.imm);
-
-  case MOVE_GPR_TO_GPR:
-    return jit_movr(_jit, dst.loc.gpr.gpr, src.loc.gpr.gpr);
-
-  case MOVE_MEM_TO_GPR:
-    return abi_mem_to_gpr(_jit, src.abi, dst.loc.gpr.gpr, src.loc.mem.base,
-                          src.loc.mem.offset);
-
-  case MOVE_FPR_TO_FPR:
-    return jit_movr_d(_jit, dst.loc.fpr, src.loc.fpr);
-
-  case MOVE_MEM_TO_FPR:
-    return abi_mem_to_fpr(_jit, src.abi, dst.loc.fpr, src.loc.mem.base,
-                          src.loc.mem.offset);
-
-  case MOVE_IMM_TO_MEM:
-    return abi_imm_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
-                          src.loc.imm);
-
-  case MOVE_GPR_TO_MEM:
-    return abi_gpr_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
-                          src.loc.gpr.gpr);
-
-  case MOVE_FPR_TO_MEM:
-    return abi_fpr_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
-                          src.loc.fpr);
-
-  case MOVE_MEM_TO_MEM:
-    return abi_mem_to_mem(_jit, src.abi, dst.loc.mem.base, dst.loc.mem.offset,
-                          src.loc.mem.base, src.loc.mem.offset);
-
-  default:
-    abort();
-  }
-}
-
-// A direct transliteration of "Tilting at windmills with Coq: formal
-// verification of a compilation algorithm for parallel moves" by
-// Laurence Rideau, Bernard Paul Serpette, and Xavier Leroy:
-// https://xavierleroy.org/publi/parallel-move.pdf
-
-enum move_status { TO_MOVE, BEING_MOVED, MOVED };
-
-static inline int
-already_in_place(jit_operand_t src, jit_operand_t dst)
-{
-  switch (MOVE_KIND(src.kind, dst.kind)) {
-  case MOVE_GPR_TO_GPR:
-    return jit_same_gprs (src.loc.gpr.gpr, dst.loc.gpr.gpr);
-  case MOVE_FPR_TO_FPR:
-    return jit_same_fprs (src.loc.fpr, dst.loc.fpr);
-  case MOVE_MEM_TO_MEM:
-    return jit_same_gprs (src.loc.mem.base, dst.loc.mem.base) &&
-      src.loc.mem.offset == dst.loc.mem.offset;
-  default:
-    return 0;
-  }
-}
-
-static inline int
-write_would_clobber(jit_operand_t src, jit_operand_t dst)
-{
-  if (already_in_place (src, dst))
-    return 1;
-
-  if (MOVE_KIND(src.kind, dst.kind) == MOVE_MEM_TO_GPR)
-    return jit_same_gprs(src.loc.mem.base, dst.loc.gpr.gpr);
-
-  return 0;
-}
-
-static inline ptrdiff_t
-operand_addend(jit_operand_t op)
-{
-  switch (op.kind) {
-  case JIT_OPERAND_KIND_GPR:
-    return op.loc.gpr.addend;
-  case JIT_OPERAND_KIND_MEM:
-    return op.loc.mem.addend;
-  default:
-    abort();
-  }
-}
-
-static void
-move_one(jit_state_t *_jit, jit_operand_t *dst, jit_operand_t *src,
-         size_t argc, enum move_status *status, size_t i)
-{
-  int tmp_gpr = 0, tmp_fpr = 0;
-
-  if (already_in_place(src[i], dst[i]))
-    return;
-
-  status[i] = BEING_MOVED;
-  for (size_t j = 0; j < argc; j++) {
-    if (write_would_clobber(src[j], dst[i])) {
-      switch (status[j]) {
-      case TO_MOVE:
-        move_one(_jit, dst, src, argc, status, j);
-        break;
-      case BEING_MOVED: {
-        jit_operand_t tmp;
-        if (is_fpr_arg (src[j].kind)) {
-          tmp_fpr = 1;
-          tmp = jit_operand_fpr(src[j].abi, get_temp_xpr(_jit));
-        } else {
-          tmp_gpr = 1;
-          /* Preserve addend, if any, from source operand, to be applied
-             at the end.  */
-          tmp = jit_operand_gpr_with_addend(src[j].abi, get_temp_gpr(_jit),
-                                            operand_addend(src[j]));
-        }
-        move_operand (_jit, tmp, src[j]);
-        src[j] = tmp;
-        break;
-      }
-      case MOVED:
-        break;
-      default:
-        abort ();
-      }
-    }
-  }
-
-  move_operand (_jit, dst[i], src[i]);
-  status[i] = MOVED;
-  if (tmp_gpr)
-    unget_temp_gpr(_jit);
-  else if (tmp_fpr)
-    unget_temp_xpr(_jit);
-}
-
-static void
-apply_addend(jit_state_t *_jit, jit_operand_t dst, jit_operand_t src)
-{
-  switch (MOVE_KIND(src.kind, dst.kind)) {
-  case MOVE_GPR_TO_GPR:
-  case MOVE_MEM_TO_GPR:
-    if (operand_addend(src))
-      jit_addi(_jit, dst.loc.gpr.gpr, dst.loc.gpr.gpr, operand_addend(src));
-    break;
-  case MOVE_GPR_TO_MEM:
-  case MOVE_MEM_TO_MEM:
-    if (operand_addend(src)) {
-      jit_gpr_t tmp = get_temp_gpr(_jit);
-      abi_mem_to_gpr(_jit, dst.abi, tmp, dst.loc.mem.base, dst.loc.mem.offset);
-      jit_addi(_jit, tmp, tmp, operand_addend(src));
-      abi_gpr_to_mem(_jit, dst.abi, dst.loc.mem.base, dst.loc.mem.offset, tmp);
-    }
-    break;
-  default:
-    break;
-  }
-}
-
-/* Preconditions: No dest operand is IMM.  No dest operand aliases
-   another dest operand.  No dest MEM operand uses a base register which
-   is used as a dest GPR.  No dst operand has an addend.  The registers
-   returned by get_temp_gpr and get_temp_fpr do not appear in source or
-   dest args.  */
-void
-jit_move_operands(jit_state_t *_jit, jit_operand_t *dst, jit_operand_t *src,
-                  size_t argc)
-{
-  // Check preconditions, except the condition about tmp registers.
-  {
-    uint64_t src_gprs = 0;
-    uint64_t dst_gprs = 0;
-    uint64_t dst_fprs = 0;
-    uint64_t dst_mem_base_gprs = 0;
-    for (size_t i = 0; i < argc; i++) {
-      switch (src[i].kind) {
-      case JIT_OPERAND_KIND_GPR:
-        src_gprs |= 1ULL << jit_gpr_regno(src[i].loc.gpr.gpr);
-        break;
-      case JIT_OPERAND_KIND_FPR:
-      case JIT_OPERAND_KIND_IMM:
-      case JIT_OPERAND_KIND_MEM:
-        break;
-      default:
-        abort();
-      }
-      switch (dst[i].kind) {
-      case JIT_OPERAND_KIND_GPR: {
-        ASSERT(dst[i].loc.gpr.addend == 0);
-        uint64_t bit = 1ULL << jit_gpr_regno(dst[i].loc.gpr.gpr);
-        ASSERT((dst_gprs & bit) == 0);
-        dst_gprs |= bit;
-        break;
-      }
-      case JIT_OPERAND_KIND_FPR: {
-        uint64_t bit = 1ULL << jit_fpr_regno(dst[i].loc.fpr);
-        ASSERT((dst_fprs & bit) == 0);
-        dst_fprs |= bit;
-        break;
-      }
-      case JIT_OPERAND_KIND_MEM: {
-        ASSERT(dst[i].loc.mem.addend == 0);
-        uint64_t bit = 1ULL << jit_gpr_regno(dst[i].loc.mem.base);
-        dst_mem_base_gprs |= bit;
-        break;
-      }
-      case JIT_OPERAND_KIND_IMM:
-      default:
-        abort();
-        break;
-      }
-    }
-    ASSERT(((src_gprs | dst_gprs) & dst_mem_base_gprs) == 0);
-  }
-
-  enum move_status status[argc];
-  for (size_t i = 0; i < argc; i++)
-    status[i] = TO_MOVE;
-  for (size_t i = 0; i < argc; i++)
-    if (status[i] == TO_MOVE)
-      move_one(_jit, dst, src, argc, status, i);
-
-  // Apply addends at the end.  We could do it earlier in some cases but
-  // at least at the end we know that an in-place increment of one
-  // operand won't alias another.
-  for (size_t i = 0; i < argc; i++)
-    apply_addend(_jit, dst[i], src[i]);
-}
-
 static const jit_gpr_t abi_gpr_args[] = {
 #if __X32
   /* No GPRs in args.  */
@@ -771,94 +351,6 @@ next_abi_arg(struct abi_arg_iterator *iter, jit_operand_t *arg)
   iter->arg_idx++;
 }
 
-// Precondition: stack is already aligned.
-static size_t
-prepare_call_args(jit_state_t *_jit, size_t argc, jit_operand_t args[])
-{
-  jit_operand_t dst[argc];
-  struct abi_arg_iterator iter;
-  
-  // Compute shuffle destinations and space for spilled arguments.
-  reset_abi_arg_iterator(&iter, argc, args);
-  for (size_t i = 0; i < argc; i++)
-    next_abi_arg(&iter, &dst[i]);
-
-  size_t stack_size = iter.stack_size;
-
-  // Reserve space for spilled arguments, and fix up SP-relative
-  // operands.
-  if (stack_size)
-    {
-      // Align stack to 16-byte boundaries on 64-bit targets.
-      if (__WORDSIZE == 64)
-        stack_size = (stack_size + 15) & ~15;
-      jit_subi(_jit, JIT_SP, JIT_SP, stack_size);
-      for (size_t i = 0; i < argc; i++) {
-        switch(args[i].kind) {
-        case JIT_OPERAND_KIND_GPR:
-          if (jit_same_gprs (args[i].loc.mem.base, JIT_SP))
-            args[i].loc.gpr.addend += stack_size;
-          break;
-        case JIT_OPERAND_KIND_MEM:
-          if (jit_same_gprs (args[i].loc.mem.base, JIT_SP))
-            args[i].loc.mem.offset += stack_size;
-          break;
-        default:
-          break;
-        }
-      }
-    }
-
-  jit_move_operands(_jit, dst, args, argc);
-
-  return stack_size;
-}
-
-void
-jit_calli(jit_state_t *_jit, jit_pointer_t f, size_t argc, jit_operand_t args[])
-{
-  size_t spill_size = prepare_call_args(_jit, argc, args);
-
-  calli(_jit, (jit_word_t)f);
-
-  if (spill_size)
-    jit_addi(_jit, JIT_SP, JIT_SP, spill_size);
-}
-
-void
-jit_callr(jit_state_t *_jit, jit_gpr_t f, size_t argc, jit_operand_t args[])
-{
-  size_t spill_size = prepare_call_args(_jit, argc, args);
-
-  callr(_jit, jit_gpr_regno(f));
-
-  if (spill_size)
-    jit_addi(_jit, JIT_SP, JIT_SP, spill_size);
-}
-
-void
-jit_locate_args(jit_state_t *_jit, size_t argc, jit_operand_t args[])
-{
-  struct abi_arg_iterator iter;
-    
-  reset_abi_arg_iterator(&iter, argc, args);
-  for (size_t i = 0; i < argc; i++)
-    next_abi_arg(&iter, &args[i]);
-}
-
-/* Precondition: args are distinct locations of type GPR or FPR.  All
-   addends of arg operands are zero.  No GPR arg is SP.  */
-void
-jit_load_args(jit_state_t *_jit, size_t argc, jit_operand_t args[])
-{
-  jit_operand_t src[argc];
-
-  memcpy(src, args, sizeof(src[0]) * argc);
-
-  jit_locate_args(_jit, argc, src);
-  jit_move_operands(_jit, args, src, argc);
-}
-
 void
 jit_flush(void *fptr, void *tptr)
 {