]> git.xonotic.org Git - xonotic/gmqcc.git/blobdiff - ir.c
Update doc/specification.tex
[xonotic/gmqcc.git] / ir.c
diff --git a/ir.c b/ir.c
index 2ac7ff1bdc7500a5090785b45206dc8edf13af71..5d8085a18947e6fddbf3998b565afe71394c5eaa 100644 (file)
--- a/ir.c
+++ b/ir.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2012
+ * Copyright (C) 2012, 2013
  *     Wolfgang Bumiller
  *
  * Permission is hereby granted, free of charge, to any person obtaining a copy of
@@ -42,7 +42,10 @@ const char *type_name[TYPE_COUNT] = {
     "variant",
     "struct",
     "union",
-    "array"
+    "array",
+
+    "nil",
+    "<no-expression>"
 };
 
 size_t type_sizeof_[TYPE_COUNT] = {
@@ -59,6 +62,8 @@ size_t type_sizeof_[TYPE_COUNT] = {
     0, /* TYPE_STRUCT   */
     0, /* TYPE_UNION    */
     0, /* TYPE_ARRAY    */
+    0, /* TYPE_NIL      */
+    0, /* TYPE_NOESPR   */
 };
 
 uint16_t type_store_instr[TYPE_COUNT] = {
@@ -78,9 +83,11 @@ uint16_t type_store_instr[TYPE_COUNT] = {
 
     INSTR_STORE_V, /* variant, should never be accessed */
 
-    AINSTR_END, /* struct */
-    AINSTR_END, /* union  */
-    AINSTR_END, /* array  */
+    VINSTR_END, /* struct */
+    VINSTR_END, /* union  */
+    VINSTR_END, /* array  */
+    VINSTR_END, /* nil    */
+    VINSTR_END, /* noexpr */
 };
 
 uint16_t field_store_instr[TYPE_COUNT] = {
@@ -100,9 +107,11 @@ uint16_t field_store_instr[TYPE_COUNT] = {
 
     INSTR_STORE_V, /* variant, should never be accessed */
 
-    AINSTR_END, /* struct */
-    AINSTR_END, /* union  */
-    AINSTR_END, /* array  */
+    VINSTR_END, /* struct */
+    VINSTR_END, /* union  */
+    VINSTR_END, /* array  */
+    VINSTR_END, /* nil    */
+    VINSTR_END, /* noexpr */
 };
 
 uint16_t type_storep_instr[TYPE_COUNT] = {
@@ -122,9 +131,11 @@ uint16_t type_storep_instr[TYPE_COUNT] = {
 
     INSTR_STOREP_V, /* variant, should never be accessed */
 
-    AINSTR_END, /* struct */
-    AINSTR_END, /* union  */
-    AINSTR_END, /* array  */
+    VINSTR_END, /* struct */
+    VINSTR_END, /* union  */
+    VINSTR_END, /* array  */
+    VINSTR_END, /* nil    */
+    VINSTR_END, /* noexpr */
 };
 
 uint16_t type_eq_instr[TYPE_COUNT] = {
@@ -144,9 +155,11 @@ uint16_t type_eq_instr[TYPE_COUNT] = {
 
     INSTR_EQ_V, /* variant, should never be accessed */
 
-    AINSTR_END, /* struct */
-    AINSTR_END, /* union  */
-    AINSTR_END, /* array  */
+    VINSTR_END, /* struct */
+    VINSTR_END, /* union  */
+    VINSTR_END, /* array  */
+    VINSTR_END, /* nil    */
+    VINSTR_END, /* noexpr */
 };
 
 uint16_t type_ne_instr[TYPE_COUNT] = {
@@ -166,9 +179,11 @@ uint16_t type_ne_instr[TYPE_COUNT] = {
 
     INSTR_NE_V, /* variant, should never be accessed */
 
-    AINSTR_END, /* struct */
-    AINSTR_END, /* union  */
-    AINSTR_END, /* array  */
+    VINSTR_END, /* struct */
+    VINSTR_END, /* union  */
+    VINSTR_END, /* array  */
+    VINSTR_END, /* nil    */
+    VINSTR_END, /* noexpr */
 };
 
 uint16_t type_not_instr[TYPE_COUNT] = {
@@ -188,13 +203,16 @@ uint16_t type_not_instr[TYPE_COUNT] = {
 
     INSTR_NOT_V, /* variant, should never be accessed */
 
-    AINSTR_END, /* struct */
-    AINSTR_END, /* union  */
-    AINSTR_END, /* array  */
+    VINSTR_END, /* struct */
+    VINSTR_END, /* union  */
+    VINSTR_END, /* array  */
+    VINSTR_END, /* nil    */
+    VINSTR_END, /* noexpr */
 };
 
 /* protos */
-static void ir_gen_extparam(ir_builder *ir);
+static ir_value* ir_gen_extparam_proto(ir_builder *ir);
+static void      ir_gen_extparam      (ir_builder *ir);
 
 /* error functions */
 
@@ -220,7 +238,7 @@ static bool irwarning(lex_ctx ctx, int warntype, const char *fmt, ...)
  * Vector utility functions
  */
 
-bool GMQCC_WARN vec_ir_value_find(ir_value **vec, ir_value *what, size_t *idx)
+bool GMQCC_WARN vec_ir_value_find(ir_value **vec, const ir_value *what, size_t *idx)
 {
     size_t i;
     size_t len = vec_size(vec);
@@ -278,13 +296,20 @@ ir_builder* ir_builder_new(const char *modulename)
     self->functions   = NULL;
     self->globals     = NULL;
     self->fields      = NULL;
-    self->extparams   = NULL;
     self->filenames   = NULL;
     self->filestrings = NULL;
     self->htglobals   = util_htnew(IR_HT_SIZE);
     self->htfields    = util_htnew(IR_HT_SIZE);
     self->htfunctions = util_htnew(IR_HT_SIZE);
 
+    self->extparams       = NULL;
+    self->extparam_protos = NULL;
+
+    self->first_common_globaltemp = 0;
+    self->max_globaltemps         = 0;
+    self->first_common_local      = 0;
+    self->max_locals              = 0;
+
     self->str_immediate = 0;
     self->name = NULL;
     if (!ir_builder_set_name(self, modulename)) {
@@ -292,6 +317,11 @@ ir_builder* ir_builder_new(const char *modulename)
         return NULL;
     }
 
+    self->nil = ir_value_var("nil", store_value, TYPE_NIL);
+    self->nil->cvq = CV_CONST;
+
+    self->reserved_va_count = NULL;
+
     return self;
 }
 
@@ -317,6 +347,7 @@ void ir_builder_delete(ir_builder* self)
     for (i = 0; i != vec_size(self->fields); ++i) {
         ir_value_delete(self->fields[i]);
     }
+    ir_value_delete(self->nil);
     vec_free(self->fields);
     vec_free(self->filenames);
     vec_free(self->filestrings);
@@ -389,6 +420,13 @@ ir_value* ir_builder_create_global(ir_builder *self, const char *name, int vtype
     return ve;
 }
 
+ir_value* ir_builder_get_va_count(ir_builder *self)
+{
+    if (self->reserved_va_count)
+        return self->reserved_va_count;
+    return (self->reserved_va_count = ir_builder_create_global(self, "reserved:va_count", TYPE_FLOAT));
+}
+
 ir_value* ir_builder_get_field(ir_builder *self, const char *name)
 {
     return (ir_value*)util_htget(self->htfields, name);
@@ -433,6 +471,8 @@ ir_function* ir_function_new(ir_builder* owner, int outtype)
         mem_d(self);
         return NULL;
     }
+    self->flags = 0;
+
     self->owner = owner;
     self->context.file = "<@no context>";
     self->context.line = 0;
@@ -445,8 +485,11 @@ ir_function* ir_function_new(ir_builder* owner, int outtype)
     self->values = NULL;
     self->locals = NULL;
 
+    self->max_varargs = 0;
+
     self->code_function_def = -1;
     self->allocated_locals = 0;
+    self->globaltemps      = 0;
 
     self->run_id = 0;
     return self;
@@ -557,6 +600,13 @@ bool ir_function_pass_peephole(ir_function *self)
                 if (!instr_is_operation(oper->opcode))
                     continue;
 
+                if (OPTS_FLAG(LEGACY_VECTOR_MATHS)) {
+                    if (oper->opcode == INSTR_MUL_VF && oper->_ops[2]->memberof == oper->_ops[1])
+                        continue;
+                    if (oper->opcode == INSTR_MUL_FV && oper->_ops[1]->memberof == oper->_ops[2])
+                        continue;
+                }
+
                 value = oper->_ops[0];
 
                 /* only do it for SSA values */
@@ -605,7 +655,8 @@ bool ir_function_pass_peephole(ir_function *self)
                     if (inot->_ops[0] != value ||
                         inot->opcode < INSTR_NOT_F ||
                         inot->opcode > INSTR_NOT_FNC ||
-                        inot->opcode == INSTR_NOT_V) /* can't do this one */
+                        inot->opcode == INSTR_NOT_V || /* can't do these */
+                        inot->opcode == INSTR_NOT_S)
                     {
                         break;
                     }
@@ -728,6 +779,8 @@ bool ir_function_pass_tailrecursion(ir_function *self)
 
 bool ir_function_finalize(ir_function *self)
 {
+    size_t i;
+
     if (self->builtin)
         return true;
 
@@ -745,8 +798,31 @@ bool ir_function_finalize(ir_function *self)
         }
     }
 
-    if (!ir_function_naive_phi(self))
+    if (!ir_function_naive_phi(self)) {
+        irerror(self->context, "internal error: ir_function_naive_phi failed");
         return false;
+    }
+
+    for (i = 0; i < vec_size(self->locals); ++i) {
+        ir_value *v = self->locals[i];
+        if (v->vtype == TYPE_VECTOR ||
+            (v->vtype == TYPE_FIELD && v->outtype == TYPE_VECTOR))
+        {
+            ir_value_vector_member(v, 0);
+            ir_value_vector_member(v, 1);
+            ir_value_vector_member(v, 2);
+        }
+    }
+    for (i = 0; i < vec_size(self->values); ++i) {
+        ir_value *v = self->values[i];
+        if (v->vtype == TYPE_VECTOR ||
+            (v->vtype == TYPE_FIELD && v->outtype == TYPE_VECTOR))
+        {
+            ir_value_vector_member(v, 0);
+            ir_value_vector_member(v, 1);
+            ir_value_vector_member(v, 2);
+        }
+    }
 
     ir_function_enumerate(self);
 
@@ -769,6 +845,8 @@ ir_value* ir_function_create_local(ir_function *self, const char *name, int vtyp
     }
 
     ve = ir_value_var(name, (param ? store_param : store_local), vtype);
+    if (param)
+        ve->locked = true;
     vec_push(self->locals, ve);
     return ve;
 }
@@ -958,6 +1036,7 @@ ir_value* ir_value_var(const char *name, int storetype, int vtype)
     self->fieldtype = TYPE_VOID;
     self->outtype = TYPE_VOID;
     self->store = storetype;
+    self->flags = 0;
 
     self->reads  = NULL;
     self->writes = NULL;
@@ -982,6 +1061,8 @@ ir_value* ir_value_var(const char *name, int storetype, int vtype)
     self->memberof = NULL;
 
     self->unique_life = false;
+    self->locked      = false;
+    self->callparam   = false;
 
     self->life = NULL;
     return self;
@@ -998,16 +1079,22 @@ ir_value* ir_value_vector_member(ir_value *self, unsigned int member)
     if (self->members[member])
         return self->members[member];
 
-    len = strlen(self->name);
-    name = (char*)mem_a(len + 3);
-    memcpy(name, self->name, len);
-    name[len+0] = '_';
-    name[len+1] = 'x' + member;
-    name[len+2] = '\0';
+    if (self->name) {
+        len = strlen(self->name);
+        name = (char*)mem_a(len + 3);
+        memcpy(name, self->name, len);
+        name[len+0] = '_';
+        name[len+1] = 'x' + member;
+        name[len+2] = '\0';
+    }
+    else
+        name = NULL;
+
     if (self->vtype == TYPE_VECTOR)
     {
         m = ir_value_var(name, self->store, TYPE_FLOAT);
-        mem_d(name);
+        if (name)
+            mem_d(name);
         if (!m)
             return NULL;
         m->context = self->context;
@@ -1020,7 +1107,8 @@ ir_value* ir_value_vector_member(ir_value *self, unsigned int member)
         if (self->fieldtype != TYPE_VECTOR)
             return NULL;
         m = ir_value_var(name, self->store, TYPE_FIELD);
-        mem_d(name);
+        if (name)
+            mem_d(name);
         if (!m)
             return NULL;
         m->fieldtype = TYPE_FLOAT;
@@ -1177,12 +1265,13 @@ bool ir_value_life_insert(ir_value *self, size_t idx, ir_life_entry_t e)
 bool ir_value_life_merge(ir_value *self, size_t s)
 {
     size_t i;
+    const size_t vs = vec_size(self->life);
     ir_life_entry_t *life = NULL;
     ir_life_entry_t *before = NULL;
     ir_life_entry_t new_entry;
 
     /* Find the first range >= s */
-    for (i = 0; i < vec_size(self->life); ++i)
+    for (i = 0; i < vs; ++i)
     {
         before = life;
         life = &self->life[i];
@@ -1190,7 +1279,7 @@ bool ir_value_life_merge(ir_value *self, size_t s)
             break;
     }
     /* nothing found? append */
-    if (i == vec_size(self->life)) {
+    if (i == vs) {
         ir_life_entry_t e;
         if (life && life->end+1 == s)
         {
@@ -1396,7 +1485,7 @@ bool ir_block_create_store_op(ir_block *self, lex_ctx ctx, int op, ir_value *tar
     if (!in)
         return false;
 
-    if (!ir_instr_op(in, 0, target, true) ||
+    if (!ir_instr_op(in, 0, target, (op < INSTR_STOREP_F || op > INSTR_STOREP_FNC)) ||
         !ir_instr_op(in, 1, what, false))
     {
         ir_instr_delete(in);
@@ -1522,20 +1611,8 @@ bool ir_block_create_jump(ir_block *self, lex_ctx ctx, ir_block *to)
 
 bool ir_block_create_goto(ir_block *self, lex_ctx ctx, ir_block *to)
 {
-    ir_instr *in;
-    if (!ir_check_unreachable(self))
-        return false;
-    self->final = true;
-    in = ir_instr_new(ctx, self, INSTR_GOTO);
-    if (!in)
-        return false;
-
-    in->bops[0] = to;
-    vec_push(self->instr, in);
-
-    vec_push(self->exits, to);
-    vec_push(to->entries, self);
-    return true;
+    self->owner->flags |= IR_FLAG_HAS_GOTO;
+    return ir_block_create_jump(self, ctx, to);
 }
 
 ir_instr* ir_block_create_phi(ir_block *self, lex_ctx ctx, const char *label, int ot)
@@ -1839,175 +1916,6 @@ ir_value* ir_block_create_load_from_ent(ir_block *self, lex_ctx ctx, const char
     return ir_block_create_general_instr(self, ctx, label, op, ent, field, outype);
 }
 
-ir_value* ir_block_create_add(ir_block *self, lex_ctx ctx,
-                              const char *label,
-                              ir_value *left, ir_value *right)
-{
-    int op = 0;
-    int l = left->vtype;
-    int r = right->vtype;
-    if (l == r) {
-        switch (l) {
-            default:
-                irerror(self->context, "invalid type for ir_block_create_add: %s", type_name[l]);
-                return NULL;
-            case TYPE_FLOAT:
-                op = INSTR_ADD_F;
-                break;
-#if 0
-            case TYPE_INTEGER:
-                op = INSTR_ADD_I;
-                break;
-#endif
-            case TYPE_VECTOR:
-                op = INSTR_ADD_V;
-                break;
-        }
-    } else {
-#if 0
-        if ( (l == TYPE_FLOAT && r == TYPE_INTEGER) )
-            op = INSTR_ADD_FI;
-        else if ( (l == TYPE_INTEGER && r == TYPE_FLOAT) )
-            op = INSTR_ADD_IF;
-        else
-#endif
-        {
-            irerror(self->context, "invalid type for ir_block_create_add: %s", type_name[l]);
-            return NULL;
-        }
-    }
-    return ir_block_create_binop(self, ctx, label, op, left, right);
-}
-
-ir_value* ir_block_create_sub(ir_block *self, lex_ctx ctx,
-                              const char *label,
-                              ir_value *left, ir_value *right)
-{
-    int op = 0;
-    int l = left->vtype;
-    int r = right->vtype;
-    if (l == r) {
-
-        switch (l) {
-            default:
-                irerror(self->context, "invalid type for ir_block_create_sub: %s", type_name[l]);
-                return NULL;
-            case TYPE_FLOAT:
-                op = INSTR_SUB_F;
-                break;
-#if 0
-            case TYPE_INTEGER:
-                op = INSTR_SUB_I;
-                break;
-#endif
-            case TYPE_VECTOR:
-                op = INSTR_SUB_V;
-                break;
-        }
-    } else {
-#if 0
-        if ( (l == TYPE_FLOAT && r == TYPE_INTEGER) )
-            op = INSTR_SUB_FI;
-        else if ( (l == TYPE_INTEGER && r == TYPE_FLOAT) )
-            op = INSTR_SUB_IF;
-        else
-#endif
-        {
-            irerror(self->context, "invalid type for ir_block_create_sub: %s", type_name[l]);
-            return NULL;
-        }
-    }
-    return ir_block_create_binop(self, ctx, label, op, left, right);
-}
-
-ir_value* ir_block_create_mul(ir_block *self, lex_ctx ctx,
-                              const char *label,
-                              ir_value *left, ir_value *right)
-{
-    int op = 0;
-    int l = left->vtype;
-    int r = right->vtype;
-    if (l == r) {
-
-        switch (l) {
-            default:
-                irerror(self->context, "invalid type for ir_block_create_mul: %s", type_name[l]);
-                return NULL;
-            case TYPE_FLOAT:
-                op = INSTR_MUL_F;
-                break;
-#if 0
-            case TYPE_INTEGER:
-                op = INSTR_MUL_I;
-                break;
-#endif
-            case TYPE_VECTOR:
-                op = INSTR_MUL_V;
-                break;
-        }
-    } else {
-        if ( (l == TYPE_VECTOR && r == TYPE_FLOAT) )
-            op = INSTR_MUL_VF;
-        else if ( (l == TYPE_FLOAT && r == TYPE_VECTOR) )
-            op = INSTR_MUL_FV;
-#if 0
-        else if ( (l == TYPE_VECTOR && r == TYPE_INTEGER) )
-            op = INSTR_MUL_VI;
-        else if ( (l == TYPE_INTEGER && r == TYPE_VECTOR) )
-            op = INSTR_MUL_IV;
-        else if ( (l == TYPE_FLOAT && r == TYPE_INTEGER) )
-            op = INSTR_MUL_FI;
-        else if ( (l == TYPE_INTEGER && r == TYPE_FLOAT) )
-            op = INSTR_MUL_IF;
-#endif
-        else {
-            irerror(self->context, "invalid type for ir_block_create_mul: %s", type_name[l]);
-            return NULL;
-        }
-    }
-    return ir_block_create_binop(self, ctx, label, op, left, right);
-}
-
-ir_value* ir_block_create_div(ir_block *self, lex_ctx ctx,
-                              const char *label,
-                              ir_value *left, ir_value *right)
-{
-    int op = 0;
-    int l = left->vtype;
-    int r = right->vtype;
-    if (l == r) {
-
-        switch (l) {
-            default:
-                irerror(self->context, "invalid type for ir_block_create_div: %s", type_name[l]);
-                return NULL;
-            case TYPE_FLOAT:
-                op = INSTR_DIV_F;
-                break;
-#if 0
-            case TYPE_INTEGER:
-                op = INSTR_DIV_I;
-                break;
-#endif
-        }
-    } else {
-#if 0
-        if ( (l == TYPE_VECTOR && r == TYPE_FLOAT) )
-            op = INSTR_DIV_VF;
-        else if ( (l == TYPE_FLOAT && r == TYPE_INTEGER) )
-            op = INSTR_DIV_FI;
-        else if ( (l == TYPE_INTEGER && r == TYPE_FLOAT) )
-            op = INSTR_DIV_IF;
-        else
-#endif
-        {
-            irerror(self->context, "invalid type for ir_block_create_div: %s", type_name[l]);
-            return NULL;
-        }
-    }
-    return ir_block_create_binop(self, ctx, label, op, left, right);
-}
-
 /* PHI resolving breaks the SSA, and must thus be the last
  * step before life-range calculation.
  */
@@ -2025,26 +1933,6 @@ bool ir_function_naive_phi(ir_function *self)
     return true;
 }
 
-#if 0
-static bool ir_naive_phi_emit_store(ir_block *block, size_t iid, ir_value *old, ir_value *what)
-{
-    ir_instr *instr;
-    size_t i;
-
-    /* create a store */
-    if (!ir_block_create_store(block, old, what))
-        return false;
-
-    /* we now move it up */
-    instr = vec_last(block->instr);
-    for (i = vec_size(block->instr)-1; i > iid; --i)
-        block->instr[i] = block->instr[i-1];
-    block->instr[i] = instr;
-
-    return true;
-}
-#endif
-
 static bool ir_block_naive_phi(ir_block *self)
 {
     size_t i, p; /*, w;*/
@@ -2087,58 +1975,6 @@ static bool ir_block_naive_phi(ir_block *self)
                 vec_push(b->instr, prevjump);
                 b->final = true;
             }
-
-#if 0
-            ir_value *v = instr->phi[p].value;
-            for (w = 0; w < vec_size(v->writes); ++w) {
-                ir_value *old;
-
-                if (!v->writes[w]->_ops[0])
-                    continue;
-
-                /* When the write was to a global, we have to emit a mov */
-                old = v->writes[w]->_ops[0];
-
-                /* The original instruction now writes to the PHI target local */
-                if (v->writes[w]->_ops[0] == v)
-                    v->writes[w]->_ops[0] = instr->_ops[0];
-
-                if (old->store != store_value && old->store != store_local && old->store != store_param)
-                {
-                    /* If it originally wrote to a global we need to store the value
-                     * there as welli
-                     */
-                    if (!ir_naive_phi_emit_store(self, i+1, old, v))
-                        return false;
-                    if (i+1 < vec_size(self->instr))
-                        instr = self->instr[i+1];
-                    else
-                        instr = NULL;
-                    /* In case I forget and access instr later, it'll be NULL
-                     * when it's a problem, to make sure we crash, rather than accessing
-                     * invalid data.
-                     */
-                }
-                else
-                {
-                    /* If it didn't, we can replace all reads by the phi target now. */
-                    size_t r;
-                    for (r = 0; r < vec_size(old->reads); ++r)
-                    {
-                        size_t op;
-                        ir_instr *ri = old->reads[r];
-                        for (op = 0; op < vec_size(ri->phi); ++op) {
-                            if (ri->phi[op].value == old)
-                                ri->phi[op].value = v;
-                        }
-                        for (op = 0; op < 3; ++op) {
-                            if (ri->_ops[op] == old)
-                                ri->_ops[op] = v;
-                        }
-                    }
-                }
-            }
-#endif
         }
         ir_instr_delete(instr);
     }
@@ -2178,6 +2014,12 @@ void ir_function_enumerate(ir_function *self)
     size_t instruction_id = 0;
     for (i = 0; i < vec_size(self->blocks); ++i)
     {
+        /* each block now gets an additional "entry" instruction id
+         * we can use to avoid point-life issues
+         */
+        self->blocks[i]->entry_id = instruction_id;
+        ++instruction_id;
+
         self->blocks[i]->eid = i;
         self->blocks[i]->run_id = 0;
         ir_block_enumerate(self->blocks[i], &instruction_id);
@@ -2187,9 +2029,13 @@ void ir_function_enumerate(ir_function *self)
 static bool ir_block_life_propagate(ir_block *b, ir_block *prev, bool *changed);
 bool ir_function_calculate_liferanges(ir_function *self)
 {
-    size_t i;
+    size_t i, s;
     bool changed;
 
+    /* parameters live at 0 */
+    for (i = 0; i < vec_size(self->params); ++i)
+        ir_value_life_merge(self->locals[i], 0);
+
     do {
         self->run_id++;
         changed = false;
@@ -2209,21 +2055,44 @@ bool ir_function_calculate_liferanges(ir_function *self)
             ir_value *v = block->living[i];
             if (v->store != store_local)
                 continue;
-            if ((v->members[0] && v->members[1] && v->members[2])) {
-                /* all vector members have been accessed - only treat this as uninitialized
-                 * if any of them is also uninitialized.
-                 */
-                if (!vec_ir_value_find(block->living, v->members[0], NULL) &&
-                    !vec_ir_value_find(block->living, v->members[1], NULL) &&
-                    !vec_ir_value_find(block->living, v->members[2], NULL))
+            if (v->vtype == TYPE_VECTOR)
+                continue;
+            self->flags |= IR_FLAG_HAS_UNINITIALIZED;
+            /* find the instruction reading from it */
+            for (s = 0; s < vec_size(v->reads); ++s) {
+                if (v->reads[s]->eid == v->life[0].end)
+                    break;
+            }
+            if (s < vec_size(v->reads)) {
+                if (irwarning(v->context, WARN_USED_UNINITIALIZED,
+                              "variable `%s` may be used uninitialized in this function\n"
+                              " -> %s:%i",
+                              v->name,
+                              v->reads[s]->context.file, v->reads[s]->context.line)
+                   )
                 {
-                    continue;
+                    return false;
                 }
+                continue;
             }
             if (v->memberof) {
-                /* A member is only uninitialized if the whole vector is also uninitialized */
-                if (!vec_ir_value_find(block->living, v->memberof, NULL))
+                ir_value *vec = v->memberof;
+                for (s = 0; s < vec_size(vec->reads); ++s) {
+                    if (vec->reads[s]->eid == v->life[0].end)
+                        break;
+                }
+                if (s < vec_size(vec->reads)) {
+                    if (irwarning(v->context, WARN_USED_UNINITIALIZED,
+                                  "variable `%s` may be used uninitialized in this function\n"
+                                  " -> %s:%i",
+                                  v->name,
+                                  vec->reads[s]->context.file, vec->reads[s]->context.line)
+                       )
+                    {
+                        return false;
+                    }
                     continue;
+                }
             }
             if (irwarning(v->context, WARN_USED_UNINITIALIZED,
                           "variable `%s` may be used uninitialized in this function", v->name))
@@ -2247,11 +2116,13 @@ typedef struct {
     bool      *unique;
 } function_allocator;
 
-static bool function_allocator_alloc(function_allocator *alloc, const ir_value *var)
+static bool function_allocator_alloc(function_allocator *alloc, ir_value *var)
 {
     ir_value *slot;
     size_t vsize = ir_value_sizeof(var);
 
+    var->code.local = vec_size(alloc->locals);
+
     slot = ir_value_var("reg", store_global, var->vtype);
     if (!slot)
         return false;
@@ -2270,32 +2141,94 @@ localerror:
     return false;
 }
 
+static bool ir_function_allocator_assign(ir_function *self, function_allocator *alloc, ir_value *v)
+{
+    size_t a;
+    ir_value *slot;
+
+    if (v->unique_life)
+        return function_allocator_alloc(alloc, v);
+
+    for (a = 0; a < vec_size(alloc->locals); ++a)
+    {
+        /* if it's reserved for a unique liferange: skip */
+        if (alloc->unique[a])
+            continue;
+
+        slot = alloc->locals[a];
+
+        /* never resize parameters
+         * will be required later when overlapping temps + locals
+         */
+        if (a < vec_size(self->params) &&
+            alloc->sizes[a] < ir_value_sizeof(v))
+        {
+            continue;
+        }
+
+        if (ir_values_overlap(v, slot))
+            continue;
+
+        if (!ir_value_life_merge_into(slot, v))
+            return false;
+
+        /* adjust size for this slot */
+        if (alloc->sizes[a] < ir_value_sizeof(v))
+            alloc->sizes[a] = ir_value_sizeof(v);
+
+        v->code.local = a;
+        return true;
+    }
+    if (a >= vec_size(alloc->locals)) {
+        if (!function_allocator_alloc(alloc, v))
+            return false;
+    }
+    return true;
+}
+
 bool ir_function_allocate_locals(ir_function *self)
 {
-    size_t i, a;
+    size_t i;
     bool   retval = true;
     size_t pos;
+    bool   opt_gt = OPTS_OPTIMIZATION(OPTIM_GLOBAL_TEMPS);
 
-    ir_value *slot;
-    const ir_value *v;
+    ir_value *v;
 
-    function_allocator alloc;
+    function_allocator lockalloc, globalloc;
 
     if (!vec_size(self->locals) && !vec_size(self->values))
         return true;
 
-    alloc.locals    = NULL;
-    alloc.sizes     = NULL;
-    alloc.positions = NULL;
-    alloc.unique    = NULL;
+    globalloc.locals    = NULL;
+    globalloc.sizes     = NULL;
+    globalloc.positions = NULL;
+    globalloc.unique    = NULL;
+    lockalloc.locals    = NULL;
+    lockalloc.sizes     = NULL;
+    lockalloc.positions = NULL;
+    lockalloc.unique    = NULL;
 
     for (i = 0; i < vec_size(self->locals); ++i)
     {
-#if 0
-        if (!OPTS_OPTIMIZATION(OPTIM_LOCALTEMPS))
-#endif
-            self->locals[i]->unique_life = true;
-        if (!function_allocator_alloc(&alloc, self->locals[i]))
+        v = self->locals[i];
+        if ((self->flags & IR_FLAG_MASK_NO_LOCAL_TEMPS) || !OPTS_OPTIMIZATION(OPTIM_LOCAL_TEMPS)) {
+            v->locked      = true;
+            v->unique_life = true;
+        }
+        else if (i >= vec_size(self->params))
+            break;
+        else
+            v->locked = true; /* lock parameters locals */
+        if (!function_allocator_alloc((v->locked || !opt_gt ? &lockalloc : &globalloc), v))
+            goto error;
+    }
+    for (; i < vec_size(self->locals); ++i)
+    {
+        v = self->locals[i];
+        if (!vec_size(v->life))
+            continue;
+        if (!ir_function_allocator_assign(self, (v->locked || !opt_gt ? &lockalloc : &globalloc), v))
             goto error;
     }
 
@@ -2307,69 +2240,100 @@ bool ir_function_allocate_locals(ir_function *self)
         if (!vec_size(v->life))
             continue;
 
-        for (a = 0; a < vec_size(alloc.locals); ++a)
-        {
-            /* if it's reserved for a unique liferange: skip */
-            if (alloc.unique[a])
-                continue;
-
-            slot = alloc.locals[a];
-
-            /* never resize parameters
-             * will be required later when overlapping temps + locals
-             */
-            if (a < vec_size(self->params) &&
-                alloc.sizes[a] < ir_value_sizeof(v))
+        /* CALL optimization:
+         * If the value is a parameter-temp: 1 write, 1 read from a CALL
+         * and it's not "locked", write it to the OFS_PARM directly.
+         */
+        if (OPTS_OPTIMIZATION(OPTIM_CALL_STORES) && !v->locked && !v->unique_life) {
+            if (vec_size(v->reads) == 1 && vec_size(v->writes) == 1 &&
+                (v->reads[0]->opcode == VINSTR_NRCALL ||
+                 (v->reads[0]->opcode >= INSTR_CALL0 && v->reads[0]->opcode <= INSTR_CALL8)
+                )
+               )
             {
+                size_t    param;
+                ir_instr *call = v->reads[0];
+                if (!vec_ir_value_find(call->params, v, &param)) {
+                    irerror(call->context, "internal error: unlocked parameter %s not found", v->name);
+                    goto error;
+                }
+                ++opts_optimizationcount[OPTIM_CALL_STORES];
+                v->callparam = true;
+                if (param < 8)
+                    ir_value_code_setaddr(v, OFS_PARM0 + 3*param);
+                else {
+                    size_t nprotos = vec_size(self->owner->extparam_protos);
+                    ir_value *ep;
+                    param -= 8;
+                    if (nprotos > param)
+                        ep = self->owner->extparam_protos[param];
+                    else
+                    {
+                        ep = ir_gen_extparam_proto(self->owner);
+                        while (++nprotos <= param)
+                            ep = ir_gen_extparam_proto(self->owner);
+                    }
+                    ir_instr_op(v->writes[0], 0, ep, true);
+                    call->params[param+8] = ep;
+                }
                 continue;
             }
-
-            if (ir_values_overlap(v, slot))
+            if (vec_size(v->writes) == 1 && v->writes[0]->opcode == INSTR_CALL0)
+            {
+                v->store = store_return;
+                if (v->members[0]) v->members[0]->store = store_return;
+                if (v->members[1]) v->members[1]->store = store_return;
+                if (v->members[2]) v->members[2]->store = store_return;
+                ++opts_optimizationcount[OPTIM_CALL_STORES];
                 continue;
-
-            if (!ir_value_life_merge_into(slot, v))
-                goto error;
-
-            /* adjust size for this slot */
-            if (alloc.sizes[a] < ir_value_sizeof(v))
-                alloc.sizes[a] = ir_value_sizeof(v);
-
-            self->values[i]->code.local = a;
-            break;
-        }
-        if (a >= vec_size(alloc.locals)) {
-            self->values[i]->code.local = vec_size(alloc.locals);
-            if (!function_allocator_alloc(&alloc, v))
-                goto error;
+            }
         }
+
+        if (!ir_function_allocator_assign(self, (v->locked || !opt_gt ? &lockalloc : &globalloc), v))
+            goto error;
     }
 
-    if (!alloc.sizes) {
+    if (!lockalloc.sizes && !globalloc.sizes) {
         goto cleanup;
     }
+    vec_push(lockalloc.positions, 0);
+    vec_push(globalloc.positions, 0);
 
     /* Adjust slot positions based on sizes */
-    vec_push(alloc.positions, 0);
-
-    if (vec_size(alloc.sizes))
-        pos = alloc.positions[0] + alloc.sizes[0];
-    else
-        pos = 0;
-    for (i = 1; i < vec_size(alloc.sizes); ++i)
-    {
-        pos = alloc.positions[i-1] + alloc.sizes[i-1];
-        vec_push(alloc.positions, pos);
+    if (lockalloc.sizes) {
+        pos = (vec_size(lockalloc.sizes) ? lockalloc.positions[0] : 0);
+        for (i = 1; i < vec_size(lockalloc.sizes); ++i)
+        {
+            pos = lockalloc.positions[i-1] + lockalloc.sizes[i-1];
+            vec_push(lockalloc.positions, pos);
+        }
+        self->allocated_locals = pos + vec_last(lockalloc.sizes);
+    }
+    if (globalloc.sizes) {
+        pos = (vec_size(globalloc.sizes) ? globalloc.positions[0] : 0);
+        for (i = 1; i < vec_size(globalloc.sizes); ++i)
+        {
+            pos = globalloc.positions[i-1] + globalloc.sizes[i-1];
+            vec_push(globalloc.positions, pos);
+        }
+        self->globaltemps = pos + vec_last(globalloc.sizes);
     }
-
-    self->allocated_locals = pos + vec_last(alloc.sizes);
 
     /* Locals need to know their new position */
     for (i = 0; i < vec_size(self->locals); ++i) {
-        self->locals[i]->code.local = alloc.positions[i];
+        v = self->locals[i];
+        if (v->locked || !opt_gt)
+            v->code.local = lockalloc.positions[v->code.local];
+        else
+            v->code.local = globalloc.positions[v->code.local];
     }
     /* Take over the actual slot positions on values */
     for (i = 0; i < vec_size(self->values); ++i) {
-        self->values[i]->code.local = alloc.positions[self->values[i]->code.local];
+        v = self->values[i];
+        if (v->locked || !opt_gt)
+            v->code.local = lockalloc.positions[v->code.local];
+        else
+            v->code.local = globalloc.positions[v->code.local];
     }
 
     goto cleanup;
@@ -2377,12 +2341,18 @@ bool ir_function_allocate_locals(ir_function *self)
 error:
     retval = false;
 cleanup:
-    for (i = 0; i < vec_size(alloc.locals); ++i)
-        ir_value_delete(alloc.locals[i]);
-    vec_free(alloc.unique);
-    vec_free(alloc.locals);
-    vec_free(alloc.sizes);
-    vec_free(alloc.positions);
+    for (i = 0; i < vec_size(lockalloc.locals); ++i)
+        ir_value_delete(lockalloc.locals[i]);
+    for (i = 0; i < vec_size(globalloc.locals); ++i)
+        ir_value_delete(globalloc.locals[i]);
+    vec_free(globalloc.unique);
+    vec_free(globalloc.locals);
+    vec_free(globalloc.sizes);
+    vec_free(globalloc.positions);
+    vec_free(lockalloc.unique);
+    vec_free(lockalloc.locals);
+    vec_free(lockalloc.sizes);
+    vec_free(lockalloc.positions);
     return retval;
 }
 
@@ -2426,18 +2396,28 @@ static void ir_op_read_write(int op, size_t *read, size_t *write)
 }
 
 static bool ir_block_living_add_instr(ir_block *self, size_t eid)
+{
+    size_t       i;
+    const size_t vs = vec_size(self->living);
+    bool         changed = false;
+    for (i = 0; i != vs; ++i)
+    {
+        if (ir_value_life_merge(self->living[i], eid))
+            changed = true;
+    }
+    return changed;
+}
+
+static bool ir_block_living_lock(ir_block *self)
 {
     size_t i;
     bool changed = false;
-    bool tempbool;
     for (i = 0; i != vec_size(self->living); ++i)
     {
-        tempbool = ir_value_life_merge(self->living[i], eid);
-        /* debug
-        if (tempbool)
-            irerror(self->context, "block_living_add_instr() value instruction added %s: %i", self->living[i]->_name, (int)eid);
-        */
-        changed = changed || tempbool;
+        if (!self->living[i]->locked) {
+            self->living[i]->locked = true;
+            changed = true;
+        }
     }
     return changed;
 }
@@ -2481,11 +2461,12 @@ static bool ir_block_life_propagate(ir_block *self, ir_block *prev, bool *change
 {
     ir_instr *instr;
     ir_value *value;
-    bool  tempbool;
-    size_t i, o, p;
+    size_t i, o, p, mem;
     /* bitmasks which operands are read from or written to */
     size_t read, write;
-    char dbg_ind[16] = { '#', '0' };
+    char dbg_ind[16];
+    dbg_ind[0] = '#';
+    dbg_ind[1] = '0';
     (void)dbg_ind;
 
     if (prev)
@@ -2499,39 +2480,12 @@ static bool ir_block_life_propagate(ir_block *self, ir_block *prev, bool *change
     { --i;
         instr = self->instr[i];
 
-        /* PHI operands are always read operands */
-        for (p = 0; p < vec_size(instr->phi); ++p)
-        {
-            value = instr->phi[p].value;
-            if (!vec_ir_value_find(self->living, value, NULL))
-                vec_push(self->living, value);
-        }
-
-        /* call params are read operands too */
-        for (p = 0; p < vec_size(instr->params); ++p)
-        {
-            value = instr->params[p];
-            if (!vec_ir_value_find(self->living, value, NULL))
-                vec_push(self->living, value);
-        }
-
         /* See which operands are read and write operands */
         ir_op_read_write(instr->opcode, &read, &write);
 
-        if (instr->opcode == INSTR_MUL_VF)
-        {
-            /* the float source will get an additional lifetime */
-            tempbool = ir_value_life_merge(instr->_ops[2], instr->eid+1);
-            *changed = *changed || tempbool;
-        }
-        else if (instr->opcode == INSTR_MUL_FV)
-        {
-            /* the float source will get an additional lifetime */
-            tempbool = ir_value_life_merge(instr->_ops[1], instr->eid+1);
-            *changed = *changed || tempbool;
-        }
-
-        /* Go through the 3 main operands */
+        /* Go through the 3 main operands
+         * writes first, then reads
+         */
         for (o = 0; o < 3; ++o)
         {
             if (!instr->_ops[o]) /* no such operand */
@@ -2547,13 +2501,6 @@ static bool ir_block_life_propagate(ir_block *self, ir_block *prev, bool *change
                 value->store != store_param)
                 continue;
 
-            /* read operands */
-            if (read & (1<<o))
-            {
-                if (!vec_ir_value_find(self->living, value, NULL))
-                    vec_push(self->living, value);
-            }
-
             /* write operands */
             /* When we write to a local, we consider it "dead" for the
              * remaining upper part of the function, since in SSA a value
@@ -2572,36 +2519,133 @@ static bool ir_block_life_propagate(ir_block *self, ir_block *prev, bool *change
                      * and make sure it's only printed once
                      * since this function is run multiple times.
                      */
-                    /* For now: debug info: */
                     /* con_err( "Value only written %s\n", value->name); */
-                    tempbool = ir_value_life_merge(value, instr->eid);
-                    *changed = *changed || tempbool;
-                    /*
-                    ir_instr_dump(instr, dbg_ind, printf);
-                    abort();
-                    */
+                    if (ir_value_life_merge(value, instr->eid))
+                        *changed = true;
                 } else {
                     /* since 'living' won't contain it
                      * anymore, merge the value, since
                      * (A) doesn't.
                      */
-                    tempbool = ir_value_life_merge(value, instr->eid);
-                    /*
-                    if (tempbool)
-                        con_err( "value added id %s %i\n", value->name, (int)instr->eid);
-                    */
-                    *changed = *changed || tempbool;
+                    if (ir_value_life_merge(value, instr->eid))
+                        *changed = true;
                     /* Then remove */
                     vec_remove(self->living, idx, 1);
                 }
+                /* Removing a vector removes all members */
+                for (mem = 0; mem < 3; ++mem) {
+                    if (value->members[mem] && vec_ir_value_find(self->living, value->members[mem], &idx)) {
+                        if (ir_value_life_merge(value->members[mem], instr->eid))
+                            *changed = true;
+                        vec_remove(self->living, idx, 1);
+                    }
+                }
+                /* Removing the last member removes the vector */
+                if (value->memberof) {
+                    value = value->memberof;
+                    for (mem = 0; mem < 3; ++mem) {
+                        if (value->members[mem] && vec_ir_value_find(self->living, value->members[mem], NULL))
+                            break;
+                    }
+                    if (mem == 3 && vec_ir_value_find(self->living, value, &idx)) {
+                        if (ir_value_life_merge(value, instr->eid))
+                            *changed = true;
+                        vec_remove(self->living, idx, 1);
+                    }
+                }
             }
         }
-        /* (A) */
-        tempbool = ir_block_living_add_instr(self, instr->eid);
-        /*con_err( "living added values\n");*/
-        *changed = *changed || tempbool;
 
+        if (instr->opcode == INSTR_MUL_VF)
+        {
+            value = instr->_ops[2];
+            /* the float source will get an additional lifetime */
+            if (ir_value_life_merge(value, instr->eid+1))
+                *changed = true;
+            if (value->memberof && ir_value_life_merge(value->memberof, instr->eid+1))
+                *changed = true;
+        }
+        else if (instr->opcode == INSTR_MUL_FV || instr->opcode == INSTR_LOAD_V)
+        {
+            value = instr->_ops[1];
+            /* the float source will get an additional lifetime */
+            if (ir_value_life_merge(value, instr->eid+1))
+                *changed = true;
+            if (value->memberof && ir_value_life_merge(value->memberof, instr->eid+1))
+                *changed = true;
+        }
+
+        for (o = 0; o < 3; ++o)
+        {
+            if (!instr->_ops[o]) /* no such operand */
+                continue;
+
+            value = instr->_ops[o];
+
+            /* We only care about locals */
+            /* we also calculate parameter liferanges so that locals
+             * can take up parameter slots */
+            if (value->store != store_value &&
+                value->store != store_local &&
+                value->store != store_param)
+                continue;
+
+            /* read operands */
+            if (read & (1<<o))
+            {
+                if (!vec_ir_value_find(self->living, value, NULL))
+                    vec_push(self->living, value);
+                /* reading adds the full vector */
+                if (value->memberof && !vec_ir_value_find(self->living, value->memberof, NULL))
+                    vec_push(self->living, value->memberof);
+                for (mem = 0; mem < 3; ++mem) {
+                    if (value->members[mem] && !vec_ir_value_find(self->living, value->members[mem], NULL))
+                        vec_push(self->living, value->members[mem]);
+                }
+            }
+        }
+        /* PHI operands are always read operands */
+        for (p = 0; p < vec_size(instr->phi); ++p)
+        {
+            value = instr->phi[p].value;
+            if (!vec_ir_value_find(self->living, value, NULL))
+                vec_push(self->living, value);
+            /* reading adds the full vector */
+            if (value->memberof && !vec_ir_value_find(self->living, value->memberof, NULL))
+                vec_push(self->living, value->memberof);
+            for (mem = 0; mem < 3; ++mem) {
+                if (value->members[mem] && !vec_ir_value_find(self->living, value->members[mem], NULL))
+                    vec_push(self->living, value->members[mem]);
+            }
+        }
+
+        /* on a call, all these values must be "locked" */
+        if (instr->opcode >= INSTR_CALL0 && instr->opcode <= INSTR_CALL8) {
+            if (ir_block_living_lock(self))
+                *changed = true;
+        }
+        /* call params are read operands too */
+        for (p = 0; p < vec_size(instr->params); ++p)
+        {
+            value = instr->params[p];
+            if (!vec_ir_value_find(self->living, value, NULL))
+                vec_push(self->living, value);
+            /* reading adds the full vector */
+            if (value->memberof && !vec_ir_value_find(self->living, value->memberof, NULL))
+                vec_push(self->living, value->memberof);
+            for (mem = 0; mem < 3; ++mem) {
+                if (value->members[mem] && !vec_ir_value_find(self->living, value->members[mem], NULL))
+                    vec_push(self->living, value->members[mem]);
+            }
+        }
+
+        /* (A) */
+        if (ir_block_living_add_instr(self, instr->eid))
+            *changed = true;
     }
+    /* the "entry" instruction ID */
+    if (ir_block_living_add_instr(self, self->entry_id))
+        *changed = true;
 
     if (self->run_id == self->owner->run_id)
         return true;
@@ -2715,7 +2759,6 @@ static bool gen_blocks_recursive(ir_function *func, ir_block *block)
     size_t    stidx;
     size_t    i;
 
-tailcall:
     block->generated = true;
     block->code_start = vec_size(code_statements);
     for (i = 0; i < vec_size(block->instr); ++i)
@@ -2732,10 +2775,8 @@ tailcall:
             /* for uncoditional jumps, if the target hasn't been generated
              * yet, we generate them right here.
              */
-            if (!target->generated) {
-                block = target;
-                goto tailcall;
-            }
+            if (!target->generated)
+                return gen_blocks_recursive(func, target);
 
             /* otherwise we generate a jump instruction */
             stmt.opcode = INSTR_GOTO;
@@ -2773,16 +2814,12 @@ tailcall:
                     code_push_statement(&stmt, instr->context.line);
             }
             if (!ontrue->generated) {
-                if (onfalse->generated) {
-                    block = ontrue;
-                    goto tailcall;
-                }
+                if (onfalse->generated)
+                    return gen_blocks_recursive(func, ontrue);
             }
             if (!onfalse->generated) {
-                if (ontrue->generated) {
-                    block = onfalse;
-                    goto tailcall;
-                }
+                if (ontrue->generated)
+                    return gen_blocks_recursive(func, onfalse);
             }
             /* neither ontrue nor onfalse exist */
             stmt.opcode = INSTR_IFNOT;
@@ -2804,7 +2841,7 @@ tailcall:
             if (onfalse->generated) {
                 /* fixup the jump address */
                 code_statements[stidx].o2.s1 = (onfalse->code_start) - (stidx);
-                if (code_statements[stidx].o2.s1 == 1) {
+                if (stidx+2 == vec_size(code_statements) && code_statements[stidx].o2.s1 == 1) {
                     code_statements[stidx] = code_statements[stidx+1];
                     if (code_statements[stidx].o1.s1 < 0)
                         code_statements[stidx].o1.s1++;
@@ -2829,33 +2866,19 @@ tailcall:
                     code_push_statement(&stmt, instr->context.line);
                 return true;
             }
-            else if (code_statements[stidx].o2.s1 == 1) {
+            else if (stidx+2 == vec_size(code_statements) && code_statements[stidx].o2.s1 == 1) {
                 code_statements[stidx] = code_statements[stidx+1];
                 if (code_statements[stidx].o1.s1 < 0)
                     code_statements[stidx].o1.s1++;
                 code_pop_statement();
             }
             /* if not, generate now */
-            block = onfalse;
-            goto tailcall;
+            return gen_blocks_recursive(func, onfalse);
         }
 
         if ( (instr->opcode >= INSTR_CALL0 && instr->opcode <= INSTR_CALL8)
            || instr->opcode == VINSTR_NRCALL)
         {
-            /* Trivial call translation:
-             * copy all params to OFS_PARM*
-             * if the output's storetype is not store_return,
-             * add append a STORE instruction!
-             *
-             * NOTES on how to do it better without much trouble:
-             * -) The liferanges!
-             *      Simply check the liferange of all parameters for
-             *      other CALLs. For each param with no CALL in its
-             *      liferange, we can store it in an OFS_PARM at
-             *      generation already. This would even include later
-             *      reuse.... probably... :)
-             */
             size_t p, first;
             ir_value *retvalue;
 
@@ -2865,12 +2888,16 @@ tailcall:
             for (p = 0; p < first; ++p)
             {
                 ir_value *param = instr->params[p];
+                if (param->callparam)
+                    continue;
 
                 stmt.opcode = INSTR_STORE_F;
                 stmt.o3.u1 = 0;
 
                 if (param->vtype == TYPE_FIELD)
                     stmt.opcode = field_store_instr[param->fieldtype];
+                else if (param->vtype == TYPE_NIL)
+                    stmt.opcode = INSTR_STORE_V;
                 else
                     stmt.opcode = type_store_instr[param->vtype];
                 stmt.o1.u1 = ir_value_code_addr(param);
@@ -2885,6 +2912,9 @@ tailcall:
                 ir_value *param = instr->params[p];
                 ir_value *targetparam;
 
+                if (param->callparam)
+                    continue;
+
                 if (p-8 >= vec_size(ir->extparams))
                     ir_gen_extparam(ir);
 
@@ -2895,6 +2925,8 @@ tailcall:
 
                 if (param->vtype == TYPE_FIELD)
                     stmt.opcode = field_store_instr[param->fieldtype];
+                else if (param->vtype == TYPE_NIL)
+                    stmt.opcode = INSTR_STORE_V;
                 else
                     stmt.opcode = type_store_instr[param->vtype];
                 stmt.o1.u1 = ir_value_code_addr(param);
@@ -2911,7 +2943,8 @@ tailcall:
             code_push_statement(&stmt, instr->context.line);
 
             retvalue = instr->_ops[0];
-            if (retvalue && retvalue->store != store_return && (vec_size(retvalue->life) || retvalue->store == store_global))
+            if (retvalue && retvalue->store != store_return &&
+                (retvalue->store == store_global || vec_size(retvalue->life)))
             {
                 /* not to be kept in OFS_RETURN */
                 if (retvalue->vtype == TYPE_FIELD && OPTS_FLAG(ADJUST_VECTOR_FIELDS))
@@ -2979,7 +3012,7 @@ tailcall:
 static bool gen_function_code(ir_function *self)
 {
     ir_block *block;
-    prog_section_statement stmt;
+    prog_section_statement stmt, *retst;
 
     /* Starting from entry point, we generate blocks "as they come"
      * for now. Dead blocks will not be translated obviously.
@@ -2999,11 +3032,21 @@ static bool gen_function_code(ir_function *self)
     }
 
     /* code_write and qcvm -disasm need to know that the function ends here */
-    stmt.opcode = INSTR_DONE;
-    stmt.o1.u1 = 0;
-    stmt.o2.u1 = 0;
-    stmt.o3.u1 = 0;
-    code_push_statement(&stmt, vec_last(code_linenums));
+    retst = &vec_last(code_statements);
+    if (OPTS_OPTIMIZATION(OPTIM_VOID_RETURN) &&
+        self->outtype == TYPE_VOID &&
+        retst->opcode == INSTR_RETURN &&
+        !retst->o1.u1 && !retst->o2.u1 && !retst->o3.u1)
+    {
+        retst->opcode = INSTR_DONE;
+        ++opts_optimizationcount[OPTIM_VOID_RETURN];
+    } else {
+        stmt.opcode = INSTR_DONE;
+        stmt.o1.u1 = 0;
+        stmt.o2.u1 = 0;
+        stmt.o3.u1 = 0;
+        code_push_statement(&stmt, vec_last(code_linenums));
+    }
     return true;
 }
 
@@ -3055,24 +3098,8 @@ static bool gen_global_function(ir_builder *ir, ir_value *global)
             fun.argsize[i] = type_sizeof_[irfun->params[i]];
     }
 
-    fun.firstlocal = vec_size(code_globals);
-
-    fun.locals = irfun->allocated_locals;
-    for (i = 0; i < vec_size(irfun->locals); ++i) {
-        if (!ir_builder_gen_global(ir, irfun->locals[i], true)) {
-            irerror(irfun->locals[i]->context, "Failed to generate local %s", irfun->locals[i]->name);
-            return false;
-        }
-        ir_value_code_setaddr(irfun->locals[i], fun.firstlocal + irfun->locals[i]->code.local);
-    }
-    for (i = 0; i < vec_size(irfun->values); ++i)
-    {
-        /* generate code.globaladdr for ssa values */
-        ir_value *v = irfun->values[i];
-        ir_value_code_setaddr(v, fun.firstlocal + v->code.local);
-    }
-    for (i = vec_size(code_globals); i < fun.firstlocal + irfun->allocated_locals; ++i)
-        vec_push(code_globals, 0);
+    fun.firstlocal = 0;
+    fun.locals     = irfun->allocated_locals;
 
     if (irfun->builtin)
         fun.entry = irfun->builtin+1;
@@ -3085,16 +3112,29 @@ static bool gen_global_function(ir_builder *ir, ir_value *global)
     return true;
 }
 
+static ir_value* ir_gen_extparam_proto(ir_builder *ir)
+{
+    ir_value *global;
+    char      name[128];
+
+    snprintf(name, sizeof(name), "EXTPARM#%i", (int)(vec_size(ir->extparam_protos)));
+    global = ir_value_var(name, store_global, TYPE_VECTOR);
+
+    vec_push(ir->extparam_protos, global);
+    return global;
+}
+
 static void ir_gen_extparam(ir_builder *ir)
 {
     prog_section_def def;
     ir_value        *global;
-    char             name[128];
 
-    snprintf(name, sizeof(name), "EXTPARM#%i", (int)(vec_size(ir->extparams)+8));
-    global = ir_value_var(name, store_global, TYPE_VECTOR);
+    if (vec_size(ir->extparam_protos) < vec_size(ir->extparams)+1)
+        global = ir_gen_extparam_proto(ir);
+    else
+        global = ir->extparam_protos[vec_size(ir->extparams)];
 
-    def.name = code_genstring(name);
+    def.name = code_genstring(global->name);
     def.type = TYPE_VECTOR;
     def.offset = vec_size(code_globals);
 
@@ -3142,6 +3182,91 @@ static bool gen_function_extparam_copy(ir_function *self)
     return true;
 }
 
+static bool gen_function_varargs_copy(ir_function *self)
+{
+    size_t i, ext, numparams, maxparams;
+
+    ir_builder *ir = self->owner;
+    ir_value   *ep;
+    prog_section_statement stmt;
+
+    numparams = vec_size(self->params);
+    if (!numparams)
+        return true;
+
+    stmt.opcode = INSTR_STORE_V;
+    stmt.o3.s1 = 0;
+    maxparams = numparams + self->max_varargs;
+    for (i = numparams; i < maxparams; ++i) {
+        if (i <= 8) {
+            stmt.o1.u1 = OFS_PARM0 + 3*i;
+            stmt.o2.u1 = ir_value_code_addr(self->locals[i]);
+            code_push_statement(&stmt, self->context.line);
+            continue;
+        }
+        ext = i - 9;
+        if (ext >= vec_size(ir->extparams))
+            ir_gen_extparam(ir);
+
+        ep = ir->extparams[ext];
+
+        stmt.o1.u1 = ir_value_code_addr(ep);
+        stmt.o2.u1 = ir_value_code_addr(self->locals[i]);
+        code_push_statement(&stmt, self->context.line);
+    }
+
+    return true;
+}
+
+static bool gen_function_locals(ir_builder *ir, ir_value *global)
+{
+    prog_section_function *def;
+    ir_function           *irfun;
+    size_t                 i;
+    uint32_t               firstlocal, firstglobal;
+
+    irfun = global->constval.vfunc;
+    def   = code_functions + irfun->code_function_def;
+
+    if (OPTS_OPTION_BOOL(OPTION_G) ||
+        !OPTS_OPTIMIZATION(OPTIM_OVERLAP_LOCALS)        ||
+        (irfun->flags & IR_FLAG_MASK_NO_OVERLAP))
+    {
+        firstlocal = def->firstlocal = vec_size(code_globals);
+    } else {
+        firstlocal = def->firstlocal = ir->first_common_local;
+        ++opts_optimizationcount[OPTIM_OVERLAP_LOCALS];
+    }
+
+    firstglobal = (OPTS_OPTIMIZATION(OPTIM_GLOBAL_TEMPS) ? ir->first_common_globaltemp : firstlocal);
+
+    for (i = vec_size(code_globals); i < firstlocal + irfun->allocated_locals; ++i)
+        vec_push(code_globals, 0);
+    for (i = 0; i < vec_size(irfun->locals); ++i) {
+        ir_value *v = irfun->locals[i];
+        if (v->locked || !OPTS_OPTIMIZATION(OPTIM_GLOBAL_TEMPS)) {
+            ir_value_code_setaddr(v, firstlocal + v->code.local);
+            if (!ir_builder_gen_global(ir, irfun->locals[i], true)) {
+                irerror(irfun->locals[i]->context, "failed to generate local %s", irfun->locals[i]->name);
+                return false;
+            }
+        }
+        else
+            ir_value_code_setaddr(v, firstglobal + v->code.local);
+    }
+    for (i = 0; i < vec_size(irfun->values); ++i)
+    {
+        ir_value *v = irfun->values[i];
+        if (v->callparam)
+            continue;
+        if (v->locked)
+            ir_value_code_setaddr(v, firstlocal + v->code.local);
+        else
+            ir_value_code_setaddr(v, firstglobal + v->code.local);
+    }
+    return true;
+}
+
 static bool gen_global_function_code(ir_builder *ir, ir_value *global)
 {
     prog_section_function *fundef;
@@ -3169,10 +3294,18 @@ static bool gen_global_function_code(ir_builder *ir, ir_value *global)
     fundef = &code_functions[irfun->code_function_def];
 
     fundef->entry = vec_size(code_statements);
+    if (!gen_function_locals(ir, global)) {
+        irerror(irfun->context, "Failed to generate locals for function %s", irfun->name);
+        return false;
+    }
     if (!gen_function_extparam_copy(irfun)) {
         irerror(irfun->context, "Failed to generate extparam-copy code for function %s", irfun->name);
         return false;
     }
+    if (irfun->max_varargs && !gen_function_varargs_copy(irfun)) {
+        irerror(irfun->context, "Failed to generate vararg-copy code for function %s", irfun->name);
+        return false;
+    }
     if (!gen_function_code(irfun)) {
         irerror(irfun->context, "Failed to generate code for function %s", irfun->name);
         return false;
@@ -3185,7 +3318,7 @@ static void gen_vector_defs(prog_section_def def, const char *name)
     char  *component;
     size_t len, i;
 
-    if (!name || OPTS_FLAG(SINGLE_VECTOR_DEFS))
+    if (!name || name[0] == '#' || OPTS_FLAG(SINGLE_VECTOR_DEFS))
         return;
 
     def.type = TYPE_FLOAT;
@@ -3241,21 +3374,45 @@ static bool ir_builder_gen_global(ir_builder *self, ir_value *global, bool isloc
     size_t           i;
     int32_t         *iptr;
     prog_section_def def;
+    bool             pushdef = false;
 
     def.type   = global->vtype;
     def.offset = vec_size(code_globals);
+    def.name   = 0;
+    if (OPTS_OPTION_BOOL(OPTION_G) || !islocal)
+    {
+        pushdef = true;
+
+        if (OPTS_OPTIMIZATION(OPTIM_STRIP_CONSTANT_NAMES) &&
+            !(global->flags & IR_FLAG_INCLUDE_DEF) &&
+            (global->name[0] == '#' || global->cvq == CV_CONST))
+        {
+            pushdef = false;
+        }
 
-    if (global->name) {
-        if (global->name[0] == '#') {
-            if (!self->str_immediate)
-                self->str_immediate = code_genstring("IMMEDIATE");
-            def.name = global->code.name = self->str_immediate;
+        if (pushdef && global->name) {
+            if (global->name[0] == '#') {
+                if (!self->str_immediate)
+                    self->str_immediate = code_genstring("IMMEDIATE");
+                def.name = global->code.name = self->str_immediate;
+            }
+            else
+                def.name = global->code.name = code_genstring(global->name);
         }
         else
-            def.name = global->code.name = code_genstring(global->name);
+            def.name   = 0;
+        if (islocal) {
+            def.offset = ir_value_code_addr(global);
+            vec_push(code_defs, def);
+            if (global->vtype == TYPE_VECTOR)
+                gen_vector_defs(def, global->name);
+            else if (global->vtype == TYPE_FIELD && global->fieldtype == TYPE_VECTOR)
+                gen_vector_defs(def, global->name);
+            return true;
+        }
     }
-    else
-        def.name   = 0;
+    if (islocal)
+        return true;
 
     switch (global->vtype)
     {
@@ -3280,14 +3437,17 @@ static bool ir_builder_gen_global(ir_builder *self, ir_value *global, bool isloc
         ir_value_code_setaddr(global, vec_size(code_globals));
         vec_push(code_globals, 0);
         /* Add the def */
-        vec_push(code_defs, def);
+        if (pushdef) vec_push(code_defs, def);
         return true;
     case TYPE_POINTER:
-        vec_push(code_defs, def);
+        if (pushdef) vec_push(code_defs, def);
         return gen_global_pointer(global);
     case TYPE_FIELD:
-        vec_push(code_defs, def);
-        gen_vector_defs(def, global->name);
+        if (pushdef) {
+            vec_push(code_defs, def);
+            if (global->fieldtype == TYPE_VECTOR)
+                gen_vector_defs(def, global->name);
+        }
         return gen_global_field(global);
     case TYPE_ENTITY:
         /* fall through */
@@ -3302,7 +3462,7 @@ static bool ir_builder_gen_global(ir_builder *self, ir_value *global, bool isloc
         }
         if (!islocal && global->cvq != CV_CONST)
             def.type |= DEF_SAVEGLOBAL;
-        vec_push(code_defs, def);
+        if (pushdef) vec_push(code_defs, def);
 
         return global->code.globaladdr >= 0;
     }
@@ -3316,7 +3476,7 @@ static bool ir_builder_gen_global(ir_builder *self, ir_value *global, bool isloc
         }
         if (!islocal && global->cvq != CV_CONST)
             def.type |= DEF_SAVEGLOBAL;
-        vec_push(code_defs, def);
+        if (pushdef) vec_push(code_defs, def);
         return global->code.globaladdr >= 0;
     }
     case TYPE_VECTOR:
@@ -3342,9 +3502,11 @@ static bool ir_builder_gen_global(ir_builder *self, ir_value *global, bool isloc
         if (!islocal && global->cvq != CV_CONST)
             def.type |= DEF_SAVEGLOBAL;
 
-        vec_push(code_defs, def);
-        def.type &= ~DEF_SAVEGLOBAL;
-        gen_vector_defs(def, global->name);
+        if (pushdef) {
+            vec_push(code_defs, def);
+            def.type &= ~DEF_SAVEGLOBAL;
+            gen_vector_defs(def, global->name);
+        }
         return global->code.globaladdr >= 0;
     }
     case TYPE_FUNCTION:
@@ -3360,7 +3522,7 @@ static bool ir_builder_gen_global(ir_builder *self, ir_value *global, bool isloc
         }
         if (!islocal && global->cvq != CV_CONST)
             def.type |= DEF_SAVEGLOBAL;
-        vec_push(code_defs, def);
+        if (pushdef) vec_push(code_defs, def);
         return true;
     case TYPE_VARIANT:
         /* assume biggest type */
@@ -3393,7 +3555,7 @@ static bool ir_builder_gen_field(ir_builder *self, ir_value *field)
     def.offset = (uint16_t)vec_size(code_globals);
 
     /* create a global named the same as the field */
-    if (opts.standard == COMPILER_GMQCC) {
+    if (OPTS_OPTION_U32(OPTION_STANDARD) == COMPILER_GMQCC) {
         /* in our standard, the global gets a dot prefix */
         size_t len = strlen(field->name);
         char name[1024];
@@ -3456,7 +3618,7 @@ bool ir_builder_generate(ir_builder *self, const char *filename)
 {
     prog_section_statement stmt;
     size_t i;
-    char   *lnofile = NULL;
+    char  *lnofile = NULL;
 
     code_init();
 
@@ -3470,6 +3632,16 @@ bool ir_builder_generate(ir_builder *self, const char *filename)
         if (!ir_builder_gen_global(self, self->globals[i], false)) {
             return false;
         }
+        if (self->globals[i]->vtype == TYPE_FUNCTION) {
+            ir_function *func = self->globals[i]->constval.vfunc;
+            if (func && self->max_locals < func->allocated_locals &&
+                !(func->flags & IR_FLAG_MASK_NO_OVERLAP))
+            {
+                self->max_locals = func->allocated_locals;
+            }
+            if (func && self->max_globaltemps < func->globaltemps)
+                self->max_globaltemps = func->globaltemps;
+        }
     }
 
     for (i = 0; i < vec_size(self->fields); ++i)
@@ -3479,6 +3651,23 @@ bool ir_builder_generate(ir_builder *self, const char *filename)
         }
     }
 
+    /* generate nil */
+    ir_value_code_setaddr(self->nil, vec_size(code_globals));
+    vec_push(code_globals, 0);
+    vec_push(code_globals, 0);
+    vec_push(code_globals, 0);
+
+    /* generate global temps */
+    self->first_common_globaltemp = vec_size(code_globals);
+    for (i = 0; i < self->max_globaltemps; ++i) {
+        vec_push(code_globals, 0);
+    }
+    /* generate common locals */
+    self->first_common_local = vec_size(code_globals);
+    for (i = 0; i < self->max_locals; ++i) {
+        vec_push(code_globals, 0);
+    }
+
     /* generate function code */
     for (i = 0; i < vec_size(self->globals); ++i)
     {
@@ -3504,7 +3693,7 @@ bool ir_builder_generate(ir_builder *self, const char *filename)
         code_push_statement(&stmt, vec_last(code_linenums));
     }
 
-    if (opts.pp_only)
+    if (OPTS_OPTION_BOOL(OPTION_PP_ONLY))
         return true;
 
     if (vec_size(code_statements) != vec_size(code_linenums)) {
@@ -3525,10 +3714,12 @@ bool ir_builder_generate(ir_builder *self, const char *filename)
         memcpy(vec_add(lnofile, 5), ".lno", 5);
     }
 
-    if (lnofile)
-        con_out("writing '%s' and '%s'...\n", filename, lnofile);
-    else
-        con_out("writing '%s'\n", filename);
+    if (!OPTS_OPTION_BOOL(OPTION_QUIET)) {
+        if (lnofile)
+            con_out("writing '%s' and '%s'...\n", filename, lnofile);
+        else
+            con_out("writing '%s'\n", filename);
+    }
     if (!code_write(filename, lnofile)) {
         vec_free(lnofile);
         return false;
@@ -3581,6 +3772,10 @@ void ir_builder_dump(ir_builder *b, int (*oprintf)(const char*, ...))
     oprintf("endmodule %s\n", b->name);
 }
 
+static const char *storenames[] = {
+    "[global]", "[local]", "[param]", "[value]", "[return]"
+};
+
 void ir_function_dump(ir_function *f, char *ind,
                       int (*oprintf)(const char*, ...))
 {
@@ -3602,22 +3797,72 @@ void ir_function_dump(ir_function *f, char *ind,
     }
     oprintf("%sliferanges:\n", ind);
     for (i = 0; i < vec_size(f->locals); ++i) {
-        size_t l;
+        const char *attr = "";
+        size_t l, m;
         ir_value *v = f->locals[i];
-        oprintf("%s\t%s: %s@%i ", ind, v->name, (v->unique_life ? "unique " : ""), (int)v->code.local);
+        if (v->unique_life && v->locked)
+            attr = "unique,locked ";
+        else if (v->unique_life)
+            attr = "unique ";
+        else if (v->locked)
+            attr = "locked ";
+        oprintf("%s\t%s: %s %s %s%s@%i ", ind, v->name, type_name[v->vtype],
+                storenames[v->store],
+                attr, (v->callparam ? "callparam " : ""),
+                (int)v->code.local);
+        if (!v->life)
+            oprintf("[null]");
         for (l = 0; l < vec_size(v->life); ++l) {
             oprintf("[%i,%i] ", v->life[l].start, v->life[l].end);
         }
         oprintf("\n");
+        for (m = 0; m < 3; ++m) {
+            ir_value *vm = v->members[m];
+            if (!vm)
+                continue;
+            oprintf("%s\t%s: @%i ", ind, vm->name, (int)vm->code.local);
+            for (l = 0; l < vec_size(vm->life); ++l) {
+                oprintf("[%i,%i] ", vm->life[l].start, vm->life[l].end);
+            }
+            oprintf("\n");
+        }
     }
     for (i = 0; i < vec_size(f->values); ++i) {
-        size_t l;
+        const char *attr = "";
+        size_t l, m;
         ir_value *v = f->values[i];
-        oprintf("%s\t%s: @%i ", ind, v->name, (int)v->code.local);
+        if (v->unique_life && v->locked)
+            attr = "unique,locked ";
+        else if (v->unique_life)
+            attr = "unique ";
+        else if (v->locked)
+            attr = "locked ";
+        oprintf("%s\t%s: %s %s %s%s@%i ", ind, v->name, type_name[v->vtype],
+                storenames[v->store],
+                attr, (v->callparam ? "callparam " : ""),
+                (int)v->code.local);
+        if (!v->life)
+            oprintf("[null]");
         for (l = 0; l < vec_size(v->life); ++l) {
             oprintf("[%i,%i] ", v->life[l].start, v->life[l].end);
         }
         oprintf("\n");
+        for (m = 0; m < 3; ++m) {
+            ir_value *vm = v->members[m];
+            if (!vm)
+                continue;
+            if (vm->unique_life && vm->locked)
+                attr = "unique,locked ";
+            else if (vm->unique_life)
+                attr = "unique ";
+            else if (vm->locked)
+                attr = "locked ";
+            oprintf("%s\t%s: %s@%i ", ind, vm->name, attr, (int)vm->code.local);
+            for (l = 0; l < vec_size(vm->life); ++l) {
+                oprintf("[%i,%i] ", vm->life[l].start, vm->life[l].end);
+            }
+            oprintf("\n");
+        }
     }
     if (vec_size(f->blocks))
     {
@@ -3641,6 +3886,8 @@ void ir_block_dump(ir_block* b, char *ind,
     oprintf("%s:%s\n", ind, b->label);
     strncat(ind, "\t", IND_BUFSZ);
 
+    if (b->instr && b->instr[0])
+        oprintf("%s (%i) [entry]\n", ind, (int)(b->instr[0]->eid-1));
     for (i = 0; i < vec_size(b->instr); ++i)
         ir_instr_dump(b->instr[i], ind, oprintf);
     ind[strlen(ind)-1] = 0;