mid/nctref
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Safety commit

Browse Source
This commit is contained in:
Mid 2024-11-20 16:36:17 +02:00
parent 07f6d57aba
commit ba6f9fb6c1
16 changed files with 708 additions and 317 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Makefile
View File

@ -11,7 +11,7 @@ ntc: $(SOURCES) $(HEADERS)
ifdef OW ifdef OW
wcl386 -ml $(if $(GAS),-DSYNTAX_GAS=1,) $(if $(DEBUG),-DDEBUG=1,) -fe="ntc.exe" -bt=dos -l=dos4g -ml $(if $(DEBUG),,-d0 -os -om -ob -oi -ol -ox) -lr -za99 -i=src $(SOURCES) wcl386 -ml $(if $(GAS),-DSYNTAX_GAS=1,) $(if $(DEBUG),-DDEBUG=1,) -fe="ntc.exe" -bt=dos -l=dos4g -ml $(if $(DEBUG),,-d0 -os -om -ob -oi -ol -ox) -lr -za99 -i=src $(SOURCES)
else else
cc $(if $(GAS),-DSYNTAX_GAS=1,) $(if $(DEBUG),-DDEBUG=1,) -Wall -o ntc -fno-PIE -no-pie -std=gnu11 $(if $(DEBUG),-O0 -g,-Os -s) -fms-extensions -Isrc $(SOURCES) cc $(if $(GAS),-DSYNTAX_GAS=1,) $(if $(DEBUGA),-DDEBUG=1,) -Wall -o ntc -fsanitize=address -fno-PIE -no-pie -std=gnu11 $(if $(DEBUG),-O0 -g,-Os -s) -fms-extensions -Isrc $(SOURCES) -Wno-array-bounds
endif endif
install: ntc install: ntc

451
src/ast.c
View File

@ -3,6 +3,12 @@
#include<stdint.h> #include<stdint.h>
#include<string.h> #include<string.h>
#include<stdlib.h> #include<stdlib.h>
#include<assert.h>
#include<stdarg.h>
const char *AST_KIND_STR[] = {
AST_KINDS(GEN_STRI)
};
AST *ast_expression_optimize(AST *ast) { AST *ast_expression_optimize(AST *ast) {
return ast; return ast;
@ -19,6 +25,8 @@ int ast_expression_equal(AST *a, AST *b) {
return a->exprUnOp.operator == b->exprUnOp.operator && ast_expression_equal(a->exprUnOp.operand, b->exprUnOp.operand); return a->exprUnOp.operator == b->exprUnOp.operator && ast_expression_equal(a->exprUnOp.operand, b->exprUnOp.operand);
} else if(a->nodeKind == AST_EXPR_BINARY_OP) { } else if(a->nodeKind == AST_EXPR_BINARY_OP) {
return a->exprBinOp.operator == b->exprBinOp.operator && ast_expression_equal(a->exprBinOp.operands[0], b->exprBinOp.operands[0]) && ast_expression_equal(a->exprBinOp.operands[1], b->exprBinOp.operands[1]); return a->exprBinOp.operator == b->exprBinOp.operator && ast_expression_equal(a->exprBinOp.operands[0], b->exprBinOp.operands[0]) && ast_expression_equal(a->exprBinOp.operands[1], b->exprBinOp.operands[1]);
} else if(a->nodeKind == AST_EXPR_STACK_POINTER) {
return 1;
} }
return 0; return 0;
@ -28,8 +36,8 @@ int ast_expression_equal(AST *a, AST *b) {
int ast_stmt_is_after(const AST *chunk, const AST *s1, const AST *s2) { int ast_stmt_is_after(const AST *chunk, const AST *s1, const AST *s2) {
const AST *s = chunk->chunk.statementFirst; const AST *s = chunk->chunk.statementFirst;
while(s) { while(1) {
if(s->nodeKind == AST_STMT_LOOP) { if(s && s->nodeKind == AST_STMT_LOOP) {
int i = ast_stmt_is_after(s->stmtLoop.body, s1, s2); int i = ast_stmt_is_after(s->stmtLoop.body, s1, s2);
if(i != -1) { if(i != -1) {
return i; return i;
@ -43,6 +51,8 @@ int ast_stmt_is_after(const AST *chunk, const AST *s1, const AST *s2) {
return 1; return 1;
} }
if(!s) break;
if(s->nodeKind == AST_STMT_IF) { if(s->nodeKind == AST_STMT_IF) {
int i = ast_stmt_is_after(s->stmtIf.then, s1, s2); int i = ast_stmt_is_after(s->stmtIf.then, s1, s2);
if(i != -1) { if(i != -1) {
@ -54,4 +64,439 @@ int ast_stmt_is_after(const AST *chunk, const AST *s1, const AST *s2) {
} }
return -1; return -1;
} }
/*
* This pass is necessary for the purposes of optimization and regalloc.
* Because an AST may hold outdated UD-chains, this pass MUST be called
* before using the UD-chains to make sure they are valid.
*
* Each local var (VTE of kind VARTABLEENTRY_VAR) holds its own UD-chain
* that specifies the exact nodes in the AST where:
* 1. It is used
* 2. The whole statement in which it is used
* 3. The definition that *might* be in use
*
* Because multiple definitions may be in use (reachable) at the point
* of the use, a unique UseDef for each possible definition is appended
* to the chain.
*
* Reachable definitions are kept track in a ReachingDefs, also held by
* each VTE. In the case of a single, simple block of code, we know
* exactly one definition (including undefined) can reach each variable,
* which would simplify the ReachingDefs structure to a single
* definition pointer.
*
* Unfortunately, conditional blocks and loops ruin this simplicity.
* If you have code like
* x = A
* if B {
* x = C
* }
* then afterward two definitions may apply to x.
*
* A solution here is to lay ReachingDefs as a graph, with each
* ReachingDefs having an optional parent. When we enter a new block of
* code, we create an empty ReachingDefs with the previous block as its
* parent. Any definitions replace the ones in the deepest
* ReachingDefs only.
*
* How we exit a block depends on its type. If it is conditional,
* the reaching definitions should join the parent (mergedefs).
* If the block is a loop, it is even worse. Given
* x = A
* loop {
* use x
* x = B
* }
* definitions can apply to uses that come before it!
*
* Also, a different case:
* x = A
* loop {
* use x
* y = B
* }
* Because, technically, the last use of x is before y = B, y and x may
* be assigned the same physical location, corrupting data as a result.
* To fix this, fake, "useless" statements are inserted during parsing
* that make the AST look as such:
* x = A
* loop {
* use x
* y = B
* }
* x;
* Until dead code removal is implemented, this will not be a problem.
*/
static void rawadduse(VarTableEntry *vte, UseDef *ud) {
assert(vte->kind == VARTABLEENTRY_VAR);
assert(ud->next == NULL);
assert(!!vte->data.var.usedefFirst == !!vte->data.var.usedefLast);
if(!vte->data.var.usedefFirst) {
vte->data.var.usedefFirst = vte->data.var.usedefLast = ud;
} else {
vte->data.var.usedefLast->next = ud;
vte->data.var.usedefLast = ud;
}
}
static void adduse(VarTableEntry *vte, AST *use, AST *whole) {
assert(vte->kind == VARTABLEENTRY_VAR);
assert(vte->data.var.reachingDefs != NULL);
for(size_t d = 0; d < vte->data.var.reachingDefs->defCount; d++) {
UseDef *ud = malloc(sizeof(*ud));
ud->def = vte->data.var.reachingDefs->defs[d];
ud->use = use;
ud->stmt = whole;
ud->next = NULL;
rawadduse(vte, ud);
}
}
static void overwritedefs(VarTableEntry *vte, AST *def) {
assert(vte->kind == VARTABLEENTRY_VAR);
if(!vte->data.var.reachingDefs) {
vte->data.var.reachingDefs = calloc(1, sizeof(*vte->data.var.reachingDefs));
}
vte->data.var.reachingDefs->defCount = 1;
if(!vte->data.var.reachingDefs->defs) {
vte->data.var.reachingDefs->defs = calloc(1, sizeof(*vte->data.var.reachingDefs->defs));
}
vte->data.var.reachingDefs->defs[0] = def;
}
static void mergedefs(VarTableEntry *vte) {
assert(vte->kind == VARTABLEENTRY_VAR);
ReachingDefs *rdefs = vte->data.var.reachingDefs;
assert(rdefs != NULL);
assert(rdefs->parent != NULL);
rdefs->parent->defs = realloc(rdefs->parent->defs, sizeof(*rdefs->parent->defs) * (rdefs->parent->defCount + rdefs->defCount));
memcpy(rdefs->parent->defs + rdefs->parent->defCount, rdefs->defs, rdefs->defCount * sizeof(*rdefs->defs));
vte->data.var.reachingDefs = rdefs->parent;
free(rdefs->defs);
free(rdefs);
}
static void pushdefs(VarTableEntry *vte) {
assert(vte->kind == VARTABLEENTRY_VAR);
ReachingDefs *rdefs = malloc(sizeof(*rdefs));
rdefs->defCount = 0;
rdefs->defs = NULL;
rdefs->excludeParent = 0;
rdefs->parent = vte->data.var.reachingDefs;
vte->data.var.reachingDefs = rdefs;
}
static void pushdefsall(AST *tlc) {
for(size_t i = 0; i < tlc->chunk.varCount; i++) {
pushdefs(tlc->chunk.vars[i]);
}
}
static void mergedefsall(AST *tlc) {
for(size_t i = 0; i < tlc->chunk.varCount; i++) {
mergedefs(tlc->chunk.vars[i]);
}
}
static void mergedefsloop(AST *tlc, VarTableEntry *vte, AST *daLoopStmt) {
assert(vte->kind == VARTABLEENTRY_VAR);
for(size_t d = 0; d < vte->data.var.reachingDefs->defCount; d++) {
UseDef *ud = vte->data.var.usedefFirst;
while(ud) {
if(ast_stmt_is_after(daLoopStmt->stmtLoop.body, NULL, ud->stmt) == 1 && ud->def != vte->data.var.reachingDefs->defs[d]) {
UseDef *udnew = calloc(1, sizeof(*udnew));
udnew->next = ud->next;
ud->next = udnew;
udnew->def = vte->data.var.reachingDefs->defs[d];
udnew->use = ud->use;
udnew->stmt = ud->stmt;
if(udnew->next == NULL) {
vte->data.var.usedefLast = udnew;
}
}
ud = ud->next;
}
}
mergedefs(vte);
}
static void mergedefsloopall(AST *tlc, AST *daLoopStmt) {
for(size_t i = 0; i < tlc->chunk.varCount; i++) {
mergedefsloop(tlc, tlc->chunk.vars[i], daLoopStmt);
}
}
static void ast_usedef_pass(AST *tlc, AST *a, AST *wholestmt) {
if(a->nodeKind == AST_CHUNK) {
for(AST *s = a->chunk.statementFirst; s; s = s->statement.next) {
ast_usedef_pass(tlc, s, s);
}
} else if(a->nodeKind == AST_STMT_IF) {
pushdefsall(tlc);
ast_usedef_pass(tlc, a->stmtIf.expression, wholestmt);
ast_usedef_pass(tlc, a->stmtIf.then, wholestmt);
mergedefsall(tlc);
} else if(a->nodeKind == AST_STMT_LOOP) {
pushdefsall(tlc);
ast_usedef_pass(tlc, a->stmtLoop.body, wholestmt);
mergedefsloopall(tlc, a);
} else if(a->nodeKind == AST_STMT_ASSIGN) {
if(a->stmtAssign.what->nodeKind == AST_EXPR_VAR && a->stmtAssign.what->exprVar.thing->kind == VARTABLEENTRY_VAR) {
overwritedefs(a->stmtAssign.what->exprVar.thing, a);
}
ast_usedef_pass(tlc, a->stmtAssign.what, wholestmt);
if(a->stmtAssign.to) {
ast_usedef_pass(tlc, a->stmtAssign.to, wholestmt);
}
} else if(a->nodeKind == AST_STMT_EXPR) {
ast_usedef_pass(tlc, a->stmtExpr.expr, wholestmt);
} else if(a->nodeKind == AST_EXPR_VAR) {
if(a->exprVar.thing->kind == VARTABLEENTRY_VAR) {
adduse(a->exprVar.thing, a, wholestmt);
}
} else if(a->nodeKind == AST_EXPR_BINARY_OP) {
ast_usedef_pass(tlc, a->exprBinOp.operands[0], wholestmt);
ast_usedef_pass(tlc, a->exprBinOp.operands[1], wholestmt);
} else if(a->nodeKind == AST_EXPR_UNARY_OP) {
ast_usedef_pass(tlc, a->exprUnOp.operand, wholestmt);
} else if(a->nodeKind == AST_EXPR_CALL) {
ast_usedef_pass(tlc, a->exprCall.what, wholestmt);
for(size_t p = 0; p < a->exprCall.what->expression.type->function.argCount; p++) {
ast_usedef_pass(tlc, a->exprCall.args[p], wholestmt);
}
} else if(a->nodeKind == AST_EXPR_PRIMITIVE) {
} else if(a->nodeKind == AST_EXPR_STRING_LITERAL) {
} else if(a->nodeKind == AST_EXPR_CAST) {
ast_usedef_pass(tlc, a->exprCast.what, wholestmt);
} else if(a->nodeKind == AST_EXPR_STACK_POINTER) {
} else if(a->nodeKind == AST_STMT_BREAK) {
} else if(a->nodeKind == AST_STMT_CONTINUE) {
} else if(a->nodeKind == AST_STMT_EXT_ALIGN) {
} else if(a->nodeKind == AST_STMT_EXT_ORG) {
} else if(a->nodeKind == AST_STMT_EXT_SECTION) {
} else if(a->nodeKind == AST_STMT_DECL) {
assert(a->stmtDecl.thing->kind != VARTABLEENTRY_VAR || a->stmtDecl.expression);
} else {
abort();
}
}
void ast_usedef_reset(AST *chu) {
for(size_t i = 0; i < chu->chunk.varCount; i++) {
VarTableEntry *vte = chu->chunk.vars[i];
assert(vte->kind == VARTABLEENTRY_VAR);
vte->data.var.reachingDefs = NULL;
vte->data.var.usedefFirst = NULL;
vte->data.var.usedefLast = NULL;
}
pushdefsall(chu);
return ast_usedef_pass(chu, chu, NULL);
}
static char *cat(char *a, const char *b) {
if(!a) {
return strdup(b);
}
a = realloc(a, strlen(a) + strlen(b) + 1);
strcpy(a + strlen(a), b);
return a;
}
__attribute__((format(printf, 1, 2))) static char *malp(const char *fmt, ...) {
va_list v1, v2;
va_start(v1, fmt);
va_copy(v2, v1);
size_t len = vsnprintf(NULL, 0, fmt, v1);
va_end(v1);
va_start(v2, fmt);
char *str = malloc(len + 1);
vsnprintf(str, len + 1, fmt, v2);
str[len] = 0;
va_end(v2);
return str;
}
char *type_to_string(Type *t) {
if(t->type == TYPE_TYPE_PRIMITIVE) {
char ret[16] = {};
int i = 0;
ret[i++] = t->primitive.isFloat ? 'f' : (t->primitive.isUnsigned ? 'u' : 'i');
snprintf(ret + i, sizeof(ret) - i, "%i", t->primitive.width);
return strdup(ret);
} else if(t->type == TYPE_TYPE_POINTER) {
char *c = type_to_string(t->pointer.of);
char *r = malp("%s*", c);
free(c);
return r;
}
return strdup("@unimp");
}
static char *ast_dumpe(AST *e) {
if(e->nodeKind == AST_EXPR_PRIMITIVE) {
return malp("%i", e->exprPrim.val);
} else if(e->nodeKind == AST_EXPR_VAR) {
VarTableEntry *vte = e->exprVar.thing;
if(vte->kind == VARTABLEENTRY_VAR) {
return strdup(vte->data.var.name);
} else if(vte->kind == VARTABLEENTRY_SYMBOL) {
return strdup(vte->data.symbol.name);
} else abort();
} else if(e->nodeKind == AST_EXPR_UNARY_OP) {
const char *op;
switch(e->exprUnOp.operator) {
case UNOP_REF:
op = "&";
break;
case UNOP_DEREF:
op = "*";
break;
case UNOP_BITWISE_NOT:
op = "~";
break;
case UNOP_NEGATE:
op = "-";
break;
}
char *c = ast_dumpe(e->exprUnOp.operand);
char *r = malp("%s%s", op, c);
free(c);
return r;
} else if(e->nodeKind == AST_EXPR_BINARY_OP) {
char *a = ast_dumpe(e->exprBinOp.operands[0]);
char *b = ast_dumpe(e->exprBinOp.operands[1]);
const char *op;
switch(e->exprBinOp.operator) {
case BINOP_ADD:
op = "+";
break;
case BINOP_SUB:
op = "-";
break;
case BINOP_MUL:
op = "*";
break;
case BINOP_DIV:
op = "/";
break;
case BINOP_BITWISE_AND:
op = "&";
break;
case BINOP_BITWISE_OR:
op = "|";
break;
case BINOP_BITWISE_XOR:
op = "^";
break;
case BINOP_EQUAL:
op = "==";
break;
default:
abort();
}
char *r = malp("(%s %s %s)", a, op, b);
free(a);
free(b);
return r;
}
return malp("@unimp:%s", AST_KIND_STR[e->nodeKind]);
}
static char *ast_dumps(AST *s) {
if(s->nodeKind == AST_STMT_DECL) {
VarTableEntry *vte = s->stmtDecl.thing;
if(vte->kind == VARTABLEENTRY_SYMBOL) {
char *t = type_to_string(vte->type);
char *e = s->stmtDecl.expression ? ast_dumpe(s->stmtDecl.expression) : strdup("");
char *r = malp("%s%s %s: %s;\n", vte->data.symbol.isExternal ? "external " : "", t, vte->data.symbol.name, e);
free(t);
free(e);
return r;
}
} else if(s->nodeKind == AST_STMT_ASSIGN) {
char *a = ast_dumpe(s->stmtAssign.what);
char *b = ast_dumpe(s->stmtAssign.to);
char *r = malp("%s = %s;\n", a, b);
free(a);
free(b);
return r;
}
return malp("@unimp:%s\n", AST_KIND_STR[s->nodeKind]);
}
char *ast_dump(AST *tlc) {
AST *stmt = tlc->chunk.statementFirst;
char *ret = NULL;
#define CAT(s) do { char *b = s; ret = cat(ret, (b)); free(b); } while(0)
while(stmt) {
CAT(ast_dumps(stmt));
stmt = stmt->statement.next;
}
return ret;
}
static void *memdup(void *a, size_t len) {
void *r = malloc(len);
memcpy(r, a, len);
return r;
}
AST *ast_deep_copy(AST *src) {
if(src->nodeKind == AST_EXPR_VAR) {
return memdup(src, sizeof(ASTExprVar));
} else if(src->nodeKind == AST_EXPR_PRIMITIVE) {
return memdup(src, sizeof(ASTExprPrimitive));
}
abort();
}

64
src/ast.h
View File

@ -18,29 +18,35 @@
#define ENUMPAK #define ENUMPAK
#endif #endif
typedef enum ENUMPAK { #define GEN_ENUM(x) x,
AST_CHUNK, #define GEN_STRI(x) #x,
AST_STMT_DECL,
AST_TYPE_IDENTIFIER, #define AST_KINDS(K) \
AST_EXPR_PRIMITIVE, K(AST_CHUNK) \
AST_STMT_IF, K(AST_STMT_DECL) \
AST_EXPR_BINARY_OP, K(AST_TYPE_IDENTIFIER) \
AST_EXPR_VAR, K(AST_EXPR_PRIMITIVE) \
AST_TYPE_POINTER, K(AST_STMT_IF) \
AST_EXPR_UNARY_OP, K(AST_EXPR_BINARY_OP) \
AST_STMT_LOOP, K(AST_EXPR_VAR) \
AST_STMT_BREAK, K(AST_EXPR_STACK_POINTER) \
AST_STMT_CONTINUE, K(AST_TYPE_POINTER) \
AST_EXPR_CALL, K(AST_EXPR_UNARY_OP) \
AST_STMT_EXPR, K(AST_STMT_LOOP) \
AST_STMT_ASSIGN, K(AST_STMT_BREAK) \
AST_STMT_EXT_ALIGN, K(AST_STMT_CONTINUE) \
AST_EXPR_STRING_LITERAL, K(AST_EXPR_CALL) \
AST_EXPR_CAST, K(AST_STMT_EXPR) \
AST_EXPR_ARRAY, K(AST_STMT_ASSIGN) \
AST_STMT_EXT_ORG, K(AST_STMT_EXT_ALIGN) \
AST_STMT_EXT_SECTION, K(AST_EXPR_STRING_LITERAL) \
} ASTKind; K(AST_EXPR_CAST) \
K(AST_EXPR_ARRAY) \
K(AST_STMT_EXT_ORG) \
K(AST_STMT_EXT_SECTION)
typedef enum ENUMPAK { AST_KINDS(GEN_ENUM) } ASTKind;
extern const char *AST_KIND_STR[];
typedef enum ENUMPAK { typedef enum ENUMPAK {
BINOP_ADD = 0, BINOP_ADD = 0,
@ -126,6 +132,10 @@ typedef struct {
char *data; char *data;
} ASTExprStringLiteral; } ASTExprStringLiteral;
typedef struct {
ASTExpr;
} ASTExprStackPointer;
typedef struct { typedef struct {
ASTKind nodeKind; ASTKind nodeKind;
@ -167,6 +177,8 @@ typedef struct {
union AST *statementFirst; union AST *statementFirst;
union AST *statementLast; union AST *statementLast;
size_t stackReservation;
} ASTChunk; } ASTChunk;
typedef struct { typedef struct {
@ -270,4 +282,10 @@ int ast_expression_equal(AST*, AST*);
int ast_stmt_is_after(const AST *chunk, const AST *s1, const AST *s2); int ast_stmt_is_after(const AST *chunk, const AST *s1, const AST *s2);
void ast_usedef_reset(AST *chu);
char *ast_dump(AST *tlc);
AST *ast_deep_copy(AST*);
#endif #endif

103
src/cg.c
View File

@ -8,7 +8,7 @@
#include"x86.h" #include"x86.h"
#define REGS 4 #define REGS 4
static const char *regs[REGS][3] = {{"al", "ax", "eax"}, {"bl", "bx", "ebx"}, {"cl", "cx", "ecx"}, {"dl", "dx", "edx"}}; static const char *regs[][3] = {{"al", "ax", "eax"}, {"bl", "bx", "ebx"}, {"cl", "cx", "ecx"}, {"dl", "dx", "edx"}, {"fu", "fk", "fck"}};
static const char *BINOP_SIMPLE_INSTRS[] = {[BINOP_ADD] = "add", [BINOP_SUB] = "sub", [BINOP_BITWISE_AND] = "and", [BINOP_BITWISE_OR] = "or", [BINOP_BITWISE_XOR] = "xor"}; static const char *BINOP_SIMPLE_INSTRS[] = {[BINOP_ADD] = "add", [BINOP_SUB] = "sub", [BINOP_BITWISE_AND] = "and", [BINOP_BITWISE_OR] = "or", [BINOP_BITWISE_XOR] = "xor"};
@ -94,7 +94,9 @@ static const char *xop_sz(AST *e, int sz) {
char *ret = bufs[bufidx]; char *ret = bufs[bufidx];
if(e->nodeKind == AST_EXPR_VAR) { if(e->nodeKind == AST_EXPR_STACK_POINTER) {
snprintf(ret, XOPBUFSZ, "esp");
} else if(e->nodeKind == AST_EXPR_VAR) {
VarTableEntry *v = e->exprVar.thing; VarTableEntry *v = e->exprVar.thing;
if(v->kind == VARTABLEENTRY_VAR) { if(v->kind == VARTABLEENTRY_VAR) {
@ -204,6 +206,10 @@ void cg_chunk(AST *a) {
static const char *instrs[] = {"inc", "dec"}; static const char *instrs[] = {"inc", "dec"};
printf("%s %s %s\n", instrs[s->stmtAssign.to->exprBinOp.operator == BINOP_SUB], specexpr(s->stmtAssign.what), xop(s->stmtAssign.what)); printf("%s %s %s\n", instrs[s->stmtAssign.to->exprBinOp.operator == BINOP_SUB], specexpr(s->stmtAssign.what), xop(s->stmtAssign.what));
} else if(s->stmtAssign.to->nodeKind == AST_EXPR_BINARY_OP && ast_expression_equal(s->stmtAssign.what, s->stmtAssign.to->exprBinOp.operands[0]) && s->stmtAssign.to->exprBinOp.operator < BINOP_SIMPLES) {
printf("%s %s, %s\n", BINOP_SIMPLE_INSTRS[s->stmtAssign.to->exprBinOp.operator], xop(s->stmtAssign.what), xop(s->stmtAssign.to->exprBinOp.operands[1]));
} else if(s->stmtAssign.what->nodeKind == AST_EXPR_VAR && s->stmtAssign.to->nodeKind == AST_EXPR_BINARY_OP && s->stmtAssign.to->exprBinOp.operator == BINOP_ADD && s->stmtAssign.to->exprBinOp.operands[0]->nodeKind == AST_EXPR_VAR && s->stmtAssign.to->exprBinOp.operands[1]->nodeKind == AST_EXPR_VAR && s->stmtAssign.to->exprBinOp.operands[0]->exprVar.thing->kind == VARTABLEENTRY_VAR && s->stmtAssign.to->exprBinOp.operands[1]->exprVar.thing->kind == VARTABLEENTRY_VAR) { } else if(s->stmtAssign.what->nodeKind == AST_EXPR_VAR && s->stmtAssign.to->nodeKind == AST_EXPR_BINARY_OP && s->stmtAssign.to->exprBinOp.operator == BINOP_ADD && s->stmtAssign.to->exprBinOp.operands[0]->nodeKind == AST_EXPR_VAR && s->stmtAssign.to->exprBinOp.operands[1]->nodeKind == AST_EXPR_VAR && s->stmtAssign.to->exprBinOp.operands[0]->exprVar.thing->kind == VARTABLEENTRY_VAR && s->stmtAssign.to->exprBinOp.operands[1]->exprVar.thing->kind == VARTABLEENTRY_VAR) {
printf("lea %s, [%s + %s]\n", printf("lea %s, [%s + %s]\n",
@ -228,6 +234,10 @@ void cg_chunk(AST *a) {
} else if(is_xop(s->stmtAssign.what) != NULL && s->stmtAssign.to->nodeKind == AST_EXPR_UNARY_OP && s->stmtAssign.to->exprUnOp.operator == UNOP_NEGATE && ast_expression_equal(s->stmtAssign.what, s->stmtAssign.to->exprUnOp.operand)) { } else if(is_xop(s->stmtAssign.what) != NULL && s->stmtAssign.to->nodeKind == AST_EXPR_UNARY_OP && s->stmtAssign.to->exprUnOp.operator == UNOP_NEGATE && ast_expression_equal(s->stmtAssign.what, s->stmtAssign.to->exprUnOp.operand)) {
printf("neg %s\n", xop(s->stmtAssign.what)); printf("neg %s\n", xop(s->stmtAssign.what));
} else if(is_xop(s->stmtAssign.what) && s->stmtAssign.to->nodeKind == AST_EXPR_CAST) {
printf("movzx %s, %s\n", xop(s->stmtAssign.what), xop(s->stmtAssign.to->exprCast.what));
} else { } else {
@ -316,20 +326,91 @@ void cg_chunk(AST *a) {
} }
} }
static void spill_pass(AST *tlc, AST *a, VarTableEntry *vte) {
if(a->nodeKind == AST_CHUNK) {
for(AST *s = a->chunk.statementFirst; s; s = s->statement.next) {
ast_usedef_pass(tlc, s, s);
}
} else if(a->nodeKind == AST_STMT_IF) {
pushdefsall(tlc);
ast_usedef_pass(tlc, a->stmtIf.expression, wholestmt);
ast_usedef_pass(tlc, a->stmtIf.then, wholestmt);
mergedefsall(tlc);
} else if(a->nodeKind == AST_STMT_LOOP) {
pushdefsall(tlc);
ast_usedef_pass(tlc, a->stmtLoop.body, wholestmt);
mergedefsloopall(tlc, a);
} else if(a->nodeKind == AST_STMT_ASSIGN) {
if(a->stmtAssign.what->nodeKind == AST_EXPR_VAR && a->stmtAssign.what->exprVar.thing->kind == VARTABLEENTRY_VAR) {
overwritedefs(a->stmtAssign.what->exprVar.thing, a);
}
ast_usedef_pass(tlc, a->stmtAssign.what, wholestmt);
if(a->stmtAssign.to) {
ast_usedef_pass(tlc, a->stmtAssign.to, wholestmt);
}
} else if(a->nodeKind == AST_STMT_EXPR) {
ast_usedef_pass(tlc, a->stmtExpr.expr, wholestmt);
} else if(a->nodeKind == AST_EXPR_VAR) {
if(a->exprVar.thing->kind == VARTABLEENTRY_VAR) {
adduse(a->exprVar.thing, a, wholestmt);
}
} else if(a->nodeKind == AST_EXPR_BINARY_OP) {
ast_usedef_pass(tlc, a->exprBinOp.operands[0], wholestmt);
ast_usedef_pass(tlc, a->exprBinOp.operands[1], wholestmt);
} else if(a->nodeKind == AST_EXPR_UNARY_OP) {
ast_usedef_pass(tlc, a->exprUnOp.operand, wholestmt);
} else if(a->nodeKind == AST_EXPR_CALL) {
ast_usedef_pass(tlc, a->exprCall.what, wholestmt);
for(size_t p = 0; p < a->exprCall.what->expression.type->function.argCount; p++) {
ast_usedef_pass(tlc, a->exprCall.args[p], wholestmt);
}
} else if(a->nodeKind == AST_EXPR_PRIMITIVE) {
} else if(a->nodeKind == AST_EXPR_STRING_LITERAL) {
} else if(a->nodeKind == AST_EXPR_CAST) {
ast_usedef_pass(tlc, a->exprCast.what, wholestmt);
} else if(a->nodeKind == AST_EXPR_STACK_POINTER) {
} else if(a->nodeKind == AST_STMT_BREAK) {
} else if(a->nodeKind == AST_STMT_CONTINUE) {
} else if(a->nodeKind == AST_STMT_EXT_ALIGN) {
} else if(a->nodeKind == AST_STMT_EXT_ORG) {
} else if(a->nodeKind == AST_STMT_EXT_SECTION) {
} else if(a->nodeKind == AST_STMT_DECL) {
assert(a->stmtDecl.thing->kind != VARTABLEENTRY_VAR || a->stmtDecl.expression);
} else {
abort();
}
}
/* Welsh-Powell graph coloring */ /* Welsh-Powell graph coloring */
static int comparator(const void *A, const void *B) { static int comparator(const void *A, const void *B) {
VarTableEntry *const *a = A; VarTableEntry *const *a = A;
VarTableEntry *const *b = B; VarTableEntry *const *b = B;
return ((*a)->data.var.degree * (*a)->data.var.priority) - ((*b)->data.var.degree * (*b)->data.var.priority); return ((*a)->data.var.degree * (*a)->data.var.priority) - ((*b)->data.var.degree * (*b)->data.var.priority);
} }
void cg_go(AST *a) { int cg_go(AST *a) {
typedef VarTableEntry *Adjacency[2]; typedef VarTableEntry *Adjacency[2];
ast_usedef_reset(a);
size_t adjCount = 0; size_t adjCount = 0;
Adjacency *adjs = malloc(sizeof(*adjs) * adjCount); Adjacency *adjs = malloc(sizeof(*adjs) * adjCount);
VarTableEntry **vars = a->chunk.vars; VarTableEntry **vars = a->chunk.vars;
for(size_t vi = 0; vi < a->chunk.varCount; vi++) {
vars[vi]->data.var.priority = 1;
vars[vi]->data.var.color = 0;
vars[vi]->data.var.degree = 0;
vars[vi]->data.var.markedForSpill = 0;
}
for(size_t v1i = 0; v1i < a->chunk.varCount; v1i++) { for(size_t v1i = 0; v1i < a->chunk.varCount; v1i++) {
for(size_t v2i = 0; v2i < a->chunk.varCount; v2i++) { for(size_t v2i = 0; v2i < a->chunk.varCount; v2i++) {
if(v1i == v2i) continue; if(v1i == v2i) continue;
@ -372,6 +453,8 @@ cont:;
qsort(vars, a->chunk.varCount, sizeof(*vars), comparator); qsort(vars, a->chunk.varCount, sizeof(*vars), comparator);
int lastColor = 0;
/* Welsh plow my ass */ /* Welsh plow my ass */
for(int v = 0; v < a->chunk.varCount; v++) { for(int v = 0; v < a->chunk.varCount; v++) {
for(int c = 0;; c++) { for(int c = 0;; c++) {
@ -384,6 +467,9 @@ cont:;
} }
vars[v]->data.var.color = c; vars[v]->data.var.color = c;
lastColor = lastColor < c ? c : lastColor;
break; break;
nextColor:; nextColor:;
@ -392,7 +478,16 @@ nextColor:;
free(adjs); free(adjs);
if(lastColor >= 4) {
// Spill node with highest degree
spill_pass(a, a, vars[a->chunk.varCount - 1]);
return 0;
}
cg_chunk(a); cg_chunk(a);
free(vars); return 1;
} }

2
src/cg.h
View File

@ -3,6 +3,6 @@
#include"ast.h" #include"ast.h"
void cg_go(union AST*); int cg_go(union AST*);
#endif #endif

191
src/dumberdowner.c
View File

@ -8,87 +8,11 @@
// Complex expressions are to be "broken down" into simpler ones until the AST // Complex expressions are to be "broken down" into simpler ones until the AST
// can be trivially translated to the target architecture. // can be trivially translated to the target architecture.
// //
// This pass, along with CG is strictly dependent on x86 and will fail for // This file along with CG is strictly for IA-32 and will fail for other
// any other architecture. // architecture.
#include"x86.h" #include"x86.h"
static void varify_change_usedefs(AST *e, AST *oldStmt, AST *newStmt) {
if(e->nodeKind == AST_EXPR_VAR) {
if(e->exprVar.thing->kind != VARTABLEENTRY_VAR) {
return;
}
UseDef *swapWithPrev = NULL;
UseDef *swapWith = NULL;
UseDef *udPrev = NULL;
UseDef *ud = e->exprVar.thing->data.var.usedefFirst;
while(ud && ud->use != e) {
if(ud->stmt == oldStmt && !swapWith) {
swapWithPrev = udPrev;
swapWith = ud;
}
udPrev = ud;
ud = ud->next;
}
if(ud) {
assert(ud->stmt == oldStmt);
ud->stmt = newStmt;
// We must make sure all newStmt usedefs are before the oldStmt ones, so swap
if(swapWith) {
assert(!!swapWithPrev + !!udPrev != 0);
if(swapWithPrev) {
swapWithPrev->next = ud;
} else {
e->exprVar.thing->data.var.usedefFirst = ud;
}
if(udPrev) {
udPrev->next = swapWith;
} else {
e->exprVar.thing->data.var.usedefFirst = swapWith;
}
UseDef *temp = ud->next;
ud->next = swapWith->next;
swapWith->next = temp;
assert(!!ud->next + !!swapWith->next != 0);
if(!swapWith->next) {
e->exprVar.thing->data.var.usedefLast = swapWith;
}
if(!ud->next) {
e->exprVar.thing->data.var.usedefLast = ud;
}
}
}
} else if(e->nodeKind == AST_EXPR_BINARY_OP) {
varify_change_usedefs(e->exprBinOp.operands[0], oldStmt, newStmt);
varify_change_usedefs(e->exprBinOp.operands[1], oldStmt, newStmt);
} else if(e->nodeKind == AST_EXPR_UNARY_OP) {
varify_change_usedefs(e->exprUnOp.operand, oldStmt, newStmt);
} else if(e->nodeKind == AST_EXPR_CALL) {
varify_change_usedefs(e->exprCall.what, oldStmt, newStmt);
int argCount = e->exprCall.what->expression.type->function.argCount;
for(int i = 0; i < argCount; i++) {
varify_change_usedefs(e->exprCall.args[i], oldStmt, newStmt);
}
} else if(e->nodeKind == AST_EXPR_ARRAY) {
int sz = e->exprArray.type->array.length;
for(int i = 0; i < sz; i++) {
varify_change_usedefs(e->exprArray.items[i], oldStmt, newStmt);
}
} else if(e->nodeKind == AST_EXPR_CAST) {
varify_change_usedefs(e->exprCast.what, oldStmt, newStmt);
}
}
/* Split away complex expression into a new local variable */ /* Split away complex expression into a new local variable */
static AST *varify(AST *tlc, AST *chunk, AST *stmtPrev, AST *stmt, AST *e) { static AST *varify(AST *tlc, AST *chunk, AST *stmtPrev, AST *stmt, AST *e) {
VarTableEntry *vte = malloc(sizeof(*vte)); VarTableEntry *vte = malloc(sizeof(*vte));
@ -125,28 +49,10 @@ static AST *varify(AST *tlc, AST *chunk, AST *stmtPrev, AST *stmt, AST *e) {
} }
assign->next = stmt; assign->next = stmt;
// Assemble ud-chain // Reset ud-chain
UseDef *ud[2]; vte->data.var.usedefFirst = NULL;
ud[0] = malloc(sizeof(UseDef)); vte->data.var.usedefLast = NULL;
ud[1] = malloc(sizeof(UseDef));
ud[0]->stmt = (AST*) assign;
ud[0]->use = assign->what;
ud[0]->def = (AST*) assign;
ud[0]->next = ud[1];
ud[1]->stmt = stmt;
ud[1]->use = (AST*) ev[1];
ud[1]->def = (AST*) assign;
ud[1]->next = NULL;
vte->data.var.usedefFirst = ud[0];
vte->data.var.usedefLast = ud[1];
// Correct the ud-chain of variables used in varified expression `e`
varify_change_usedefs(e, stmt, (AST*) assign);
return (AST*) ev[1]; return (AST*) ev[1];
} }
@ -155,28 +61,6 @@ static AST *xopify(AST *tlc, AST *chunk, AST *stmtPrev, AST *stmt, AST *e) {
return varify(tlc, chunk, stmtPrev, stmt, e); return varify(tlc, chunk, stmtPrev, stmt, e);
} }
static UseDef *get_last_usedef_before_stmt(AST *tlc, VarTableEntry *v, AST *stmt) {
assert(v->kind == VARTABLEENTRY_VAR);
UseDef *ud = v->data.var.usedefFirst;
if(!ud) {
return NULL;
}
UseDef *udPrev = NULL;
while(ud) {
if(ast_stmt_is_after(tlc, ud->stmt, stmt)) {
break;
}
udPrev = ud;
ud = ud->next;
}
return udPrev;
}
static int dumben_chunk(AST *tlc, AST *chu) { static int dumben_chunk(AST *tlc, AST *chu) {
AST *sPrev = NULL; AST *sPrev = NULL;
AST *s = chu->chunk.statementFirst; AST *s = chu->chunk.statementFirst;
@ -230,24 +114,17 @@ static int dumben_chunk(AST *tlc, AST *chu) {
effective = 1; effective = 1;
} }
if(s->stmtAssign.what->nodeKind == AST_EXPR_VAR && s->stmtAssign.what->exprVar.thing->kind == VARTABLEENTRY_VAR && s->stmtAssign.to->nodeKind == AST_EXPR_UNARY_OP && s->stmtAssign.to->exprUnOp.operator == UNOP_NEGATE && (s->stmtAssign.to->exprUnOp.operand->nodeKind != AST_EXPR_VAR || s->stmtAssign.what->exprVar.thing != s->stmtAssign.to->exprUnOp.operand->exprVar.thing)) { if(s->stmtAssign.to && s->stmtAssign.to->nodeKind == AST_EXPR_UNARY_OP && s->stmtAssign.to->exprUnOp.operator == UNOP_NEGATE && !ast_expression_equal(s->stmtAssign.what, s->stmtAssign.to->exprUnOp.operand)) {
// Turn this: // Turn this:
// a = -b // a = -b
// into // into
// a = b // a = b
// a = -a // a = -a
// While keeping UD-chain valid.
// Currently only for variables. While this could work for any XOP, it requires the ability to deep-copy `a`, which is not trivial when taking into account UD-chains. AST *ev[2] = {
// TODO: Scrap incremental UD-chain modifications. Simply rebuild them after each pass. ast_deep_copy(s->stmtAssign.what),
ast_deep_copy(s->stmtAssign.what)
ASTExprVar *ev[2]; };
for(int i = 0; i < 2; i++) {
ev[i] = malloc(sizeof(ASTExprVar));
ev[i]->nodeKind = AST_EXPR_VAR;
ev[i]->type = s->stmtAssign.what->expression.type;
ev[i]->thing = s->stmtAssign.what->exprVar.thing;
}
AST *negation = s->stmtAssign.to; AST *negation = s->stmtAssign.to;
@ -261,28 +138,40 @@ static int dumben_chunk(AST *tlc, AST *chu) {
assign2->statement.next = s->statement.next; assign2->statement.next = s->statement.next;
s->statement.next = assign2; s->statement.next = assign2;
UseDef *link = get_last_usedef_before_stmt(tlc, ev[0]->thing, assign2->statement.next);
UseDef *ud[2];
ud[0] = malloc(sizeof(UseDef));
ud[1] = malloc(sizeof(UseDef));
ud[0]->stmt = assign2;
ud[0]->use = (AST*) ev[1];
ud[0]->def = s;
ud[0]->next = ud[1];
ud[1]->stmt = assign2;
ud[1]->use = (AST*) ev[0];
ud[1]->def = s;
ud[1]->next = link->next;
link->next = ud[0];
effective = 1; effective = 1;
} }
if(is_xop(s->stmtAssign.to) == XOP_NOT_XOP) {
if(s->stmtAssign.to->nodeKind == AST_EXPR_UNARY_OP && s->stmtAssign.to->exprUnOp.operator == UNOP_DEREF) {
s->stmtAssign.to->exprUnOp.operand = varify(tlc, chu, sPrev, s, s->stmtAssign.to->exprUnOp.operand);
effective = 1;
} else if(s->stmtAssign.to->nodeKind == AST_EXPR_BINARY_OP && s->stmtAssign.to->exprBinOp.operator < BINOP_SIMPLES && !ast_expression_equal(s->stmtAssign.what, s->stmtAssign.to->exprBinOp.operands[0])) {
// Turn this:
// a = b op c
// into
// a = b
// a = a op c
AST *assign2 = malloc(sizeof(ASTStmtAssign));
assign2->nodeKind = AST_STMT_ASSIGN;
assign2->stmtAssign.what = ast_deep_copy(s->stmtAssign.what);
assign2->stmtAssign.to = s->stmtAssign.to->exprBinOp.operands[0];
sPrev->statement.next = assign2;
assign2->statement.next = s;
s->stmtAssign.to->exprBinOp.operands[0] = ast_deep_copy(s->stmtAssign.what);
effective = 1;
} else if(s->stmtAssign.to->nodeKind == AST_EXPR_BINARY_OP && !is_xop(s->stmtAssign.to->exprBinOp.operands[0])) {
s->stmtAssign.to->exprBinOp.operands[0] = xopify(tlc, chu, sPrev, s, s->stmtAssign.to->exprBinOp.operands[0]);
} else if(s->stmtAssign.to->nodeKind == AST_EXPR_BINARY_OP && !is_xop(s->stmtAssign.to->exprBinOp.operands[1])) {
s->stmtAssign.to->exprBinOp.operands[1] = xopify(tlc, chu, sPrev, s, s->stmtAssign.to->exprBinOp.operands[1]);
}
}
} else if(s->nodeKind == AST_STMT_EXPR && s->stmtExpr.expr->nodeKind == AST_EXPR_CALL) { } else if(s->nodeKind == AST_STMT_EXPR && s->stmtExpr.expr->nodeKind == AST_EXPR_CALL) {
int argCount = s->stmtExpr.expr->exprCall.what->expression.type->function.argCount; int argCount = s->stmtExpr.expr->exprCall.what->expression.type->function.argCount;

7
src/ntc.c
View File

@ -39,7 +39,12 @@ int main(int argc_, char **argv_) {
dumben_go(chunk); dumben_go(chunk);
cg_go(chunk); while(!cg_go(chunk)) {
puts(ast_dump(chunk));
dumben_go(chunk);
}
return 0; return 0;
} }

170
src/parse.c
View File

@ -24,11 +24,6 @@ static inline int __builtin_clzl(unsigned long x) {
} }
#endif #endif
typedef struct {
UseDef ud;
VarTableEntry *to;
} UseDefToAdd;
typedef struct { typedef struct {
Token *tokens; Token *tokens;
intmax_t i; intmax_t i;
@ -41,10 +36,6 @@ typedef struct {
// Used by pushstat to add statements // Used by pushstat to add statements
ASTChunk *currentChunk; ASTChunk *currentChunk;
// Used by pushstmt to assemble use-def chain
size_t udsToAddCount;
UseDefToAdd *udsToAdd;
// Used to place guard variable uses after loops to stop reg allocation from fucking up // Used to place guard variable uses after loops to stop reg allocation from fucking up
VarTable *loopScope; VarTable *loopScope;
size_t guardedVarCount; size_t guardedVarCount;
@ -88,28 +79,6 @@ static int maybe(Parser *P, TokenKind t) {
} }
static void pushstat(Parser *P, void *a) { static void pushstat(Parser *P, void *a) {
for(size_t i = 0; i < P->udsToAddCount; i++) {
VarTableEntry *to = P->udsToAdd[i].to;
UseDef *ud = malloc(sizeof(*ud));
memcpy(ud, &P->udsToAdd[i].ud, sizeof(*ud));
ud->stmt = a;
/*if(ud->stmt->nodeKind == AST_STMT_ASSIGN && ud->stmt->stmtAssign.what == ud->use) {
// Forget. The x in `x = y` is not a use of x.
continue;
}*/
if(to->data.var.usedefFirst) {
to->data.var.usedefLast->next = ud;
to->data.var.usedefLast = ud;
} else {
to->data.var.usedefFirst = to->data.var.usedefLast = ud;
}
}
P->udsToAddCount = 0;
if(P->currentChunk->statementFirst) { if(P->currentChunk->statementFirst) {
P->currentChunk->statementLast->statement.next = a; P->currentChunk->statementLast->statement.next = a;
P->currentChunk->statementLast = a; P->currentChunk->statementLast = a;
@ -141,7 +110,7 @@ static ASTExprPrimitive *parse_prim(Parser *P) {
return ret; return ret;
} }
static void newusedef(Parser *P, VarTableEntry *v, AST *expr) { /*static void newusedef(Parser *P, VarTableEntry *v, AST *expr) {
ReachingDefs *defs = v->data.var.reachingDefs; ReachingDefs *defs = v->data.var.reachingDefs;
while(defs) { while(defs) {
@ -160,7 +129,7 @@ static void newusedef(Parser *P, VarTableEntry *v, AST *expr) {
defs = defs->parent; defs = defs->parent;
} }
} }*/
static AST *exprvar(Parser *P, VarTableEntry *v) { static AST *exprvar(Parser *P, VarTableEntry *v) {
AST *a = malloc(sizeof(ASTExprVar)); AST *a = malloc(sizeof(ASTExprVar));
@ -168,10 +137,6 @@ static AST *exprvar(Parser *P, VarTableEntry *v) {
a->exprVar.type = v->type; a->exprVar.type = v->type;
a->exprVar.thing = v; a->exprVar.thing = v;
if(v->kind == VARTABLEENTRY_VAR) {
newusedef(P, v, a);
}
if(P->loopScope) { if(P->loopScope) {
// XXX: O(n)!!!!!!!!! // XXX: O(n)!!!!!!!!!
@ -254,9 +219,6 @@ AST *nct_cast_expr(AST *what, Type *to) {
ret->val = what->exprPrim.val & (((int64_t) 1 << to->primitive.width) - 1); ret->val = what->exprPrim.val & (((int64_t) 1 << to->primitive.width) - 1);
return (AST*) ret; return (AST*) ret;
} else { } else {
// Silently fail :)
return what;
ASTExprCast *ret = malloc(sizeof(*ret)); ASTExprCast *ret = malloc(sizeof(*ret));
ret->nodeKind = AST_EXPR_CAST; ret->nodeKind = AST_EXPR_CAST;
ret->type = to; ret->type = to;
@ -273,6 +235,15 @@ AST *nct_parse_expression(Parser *P, int lOP) {
if(peek(P, 0).type == TOKEN_NUMBER) { if(peek(P, 0).type == TOKEN_NUMBER) {
return (AST*) parse_prim(P); return (AST*) parse_prim(P);
} else if(peek(P, 0).type == TOKEN_IDENTIFIER) { } else if(peek(P, 0).type == TOKEN_IDENTIFIER) {
if(!strcmp(peek(P, 0).content, "@stack")) {
get(P);
ASTExprStackPointer *ret = malloc(sizeof(*ret));
ret->nodeKind = AST_EXPR_STACK_POINTER;
ret->type = primitive_parse("u32");
return ret;
}
return exprvar(P, vartable_find(P->scope, get(P).content)); return exprvar(P, vartable_find(P->scope, get(P).content));
} else if(peek(P, 0).type == TOKEN_STRING) { } else if(peek(P, 0).type == TOKEN_STRING) {
ASTExprStringLiteral *ret = malloc(sizeof(*ret)); ASTExprStringLiteral *ret = malloc(sizeof(*ret));
@ -478,10 +449,20 @@ AST *nct_parse_expression(Parser *P, int lOP) {
ASTExpr *operand = &(astop->operands[1] = nct_parse_expression(P, lOP + 1))->expression; ASTExpr *operand = &(astop->operands[1] = nct_parse_expression(P, lOP + 1))->expression;
if(operand->type->type != TYPE_TYPE_PRIMITIVE) { if(!type_is_number(astop->operands[0]->expression.type)
stahp(P->tokens[P->i].row, P->tokens[P->i].column, "Invalid combination of operator and operand types."); || !type_is_number(astop->operands[1]->expression.type)) {
}
stahp(P->tokens[P->i].row, P->tokens[P->i].column, "Attempt to perform arithmetic on non-number types.");
}
if(type_size(astop->operands[0]->expression.type) < type_size(astop->operands[1]->expression.type)) {
astop->operands[0] = nct_cast_expr(astop->operands[0], astop->operands[1]->expression.type);
}
if(type_size(astop->operands[1]->expression.type) < type_size(astop->operands[0]->expression.type)) {
astop->operands[1] = nct_cast_expr(astop->operands[1], astop->operands[0]->expression.type);
}
if(!astop->type) { if(!astop->type) {
astop->type = operand->type; astop->type = operand->type;
} else { } else {
@ -655,8 +636,8 @@ static AST *parse_declaration(Parser *P) {
assign->nodeKind = AST_STMT_ASSIGN; assign->nodeKind = AST_STMT_ASSIGN;
assign->next = NULL; assign->next = NULL;
entry->data.var.reachingDefs = reachingdefs_push(NULL); //entry->data.var.reachingDefs = reachingdefs_push(NULL);
reachingdefs_set(entry->data.var.reachingDefs, (AST*) assign); //reachingdefs_set(entry->data.var.reachingDefs, (AST*) assign);
assign->what = exprvar(P, entry); assign->what = exprvar(P, entry);
assign->to = peek(P, 0).type == TOKEN_SEMICOLON ? NULL : nct_parse_expression(P, 0); assign->to = peek(P, 0).type == TOKEN_SEMICOLON ? NULL : nct_parse_expression(P, 0);
@ -710,77 +691,6 @@ backtrack:
return NULL; return NULL;
} }
/* Imagine the following pseudocode:
x = ...
while ... {
(do something with x)
x = ...
}
The second definition of x reaches the code *before* it.
Therefore we must go over the loop body a second time with known reaching definitions and add the necessary usedefs.*/
static void loop_second_pass(AST *a) {
if(a->nodeKind == AST_CHUNK) {
for(AST *s = a->chunk.statementFirst; s; s = s->statement.next) {
loop_second_pass(s);
}
} else if(a->nodeKind == AST_STMT_IF) {
loop_second_pass(a->stmtIf.expression);
loop_second_pass(a->stmtIf.then);
} else if(a->nodeKind == AST_STMT_LOOP) {
loop_second_pass(a->stmtLoop.body);
} else if(a->nodeKind == AST_STMT_ASSIGN) {
loop_second_pass(a->stmtAssign.what);
loop_second_pass(a->stmtAssign.to);
/* TODO: Per-variable flag.to cease additional usedefs after assignment?
It is okay to have too many usedefs, so this isn't high priority.*/
} else if(a->nodeKind == AST_STMT_EXPR) {
loop_second_pass(a->stmtExpr.expr);
} else if(a->nodeKind == AST_EXPR_VAR) {
UseDef *udPrev = NULL;
UseDef *ud = a->exprVar.thing->data.var.usedefFirst;
while(ud && ud->use != a) {
udPrev = ud;
ud = ud->next;
}
if(ud) {
ReachingDefs *defs = a->exprVar.thing->data.var.reachingDefs;
while(defs) {
for(size_t i = 0; i < defs->defCount; i++) {
UseDef *newud = calloc(1, sizeof(*newud));
memcpy(newud, ud, sizeof(*ud));
newud->def = defs->defs[i];
udPrev->next = newud;
newud->next = ud;
ud = newud;
}
if(defs->excludeParent) {
break;
}
defs = defs->parent;
}
}
} else if(a->nodeKind == AST_EXPR_BINARY_OP) {
loop_second_pass(a->exprBinOp.operands[0]);
loop_second_pass(a->exprBinOp.operands[1]);
} else if(a->nodeKind == AST_EXPR_UNARY_OP) {
loop_second_pass(a->exprUnOp.operand);
} else if(a->nodeKind == AST_EXPR_CAST) {
loop_second_pass(a->exprCast.what);
} else if(a->nodeKind == AST_EXPR_CALL) {
loop_second_pass(a->exprCall.what);
size_t argCount = a->exprCall.what->expression.type->function.argCount;
for(size_t i = 0; i < argCount; i++) {
loop_second_pass(a->exprCall.args[i]);
}
}
}
ASTChunk *nct_parse_chunk(Parser*, int, int); ASTChunk *nct_parse_chunk(Parser*, int, int);
void nct_parse_statement(Parser *P) { void nct_parse_statement(Parser *P) {
if(maybe(P, TOKEN_IF)) { if(maybe(P, TOKEN_IF)) {
@ -815,22 +725,26 @@ void nct_parse_statement(Parser *P) {
ret->body = (AST*) nct_parse_chunk(P, 0, 1); ret->body = (AST*) nct_parse_chunk(P, 0, 1);
expect(P, TOKEN_SQUIGGLY_R); expect(P, TOKEN_SQUIGGLY_R);
pushstat(P, ret);
if(isFirstLoop) { if(isFirstLoop) {
P->loopScope = NULL; P->loopScope = NULL;
for(size_t i = 0; i < P->guardedVarCount; i++) { for(size_t i = 0; i < P->guardedVarCount; i++) {
ASTExprVar *ev = P->guardedVars[i]; ASTExprVar *ev = P->guardedVars[i];
newusedef(P, ev->thing, ev); AST *es = calloc(1, sizeof(ASTStmtExpr));
es->nodeKind = AST_STMT_EXPR;
es->stmtExpr.expr = ev;
pushstat(P, es);
} }
P->guardedVarCount = 0; P->guardedVarCount = 0;
free(P->guardedVars); free(P->guardedVars);
P->guardedVars = NULL; P->guardedVars = NULL;
} }
loop_second_pass(ret->body);
pushstat(P, ret);
return; return;
} else if(maybe(P, TOKEN_BREAK)) { } else if(maybe(P, TOKEN_BREAK)) {
ASTStmtBreak *ret = malloc(sizeof(*ret)); ASTStmtBreak *ret = malloc(sizeof(*ret));
@ -840,6 +754,7 @@ void nct_parse_statement(Parser *P) {
expect(P, TOKEN_SEMICOLON); expect(P, TOKEN_SEMICOLON);
pushstat(P, ret); pushstat(P, ret);
return; return;
} else if(maybe(P, TOKEN_CONTINUE)) { } else if(maybe(P, TOKEN_CONTINUE)) {
ASTStmtContinue *ret = malloc(sizeof(*ret)); ASTStmtContinue *ret = malloc(sizeof(*ret));
@ -849,6 +764,7 @@ void nct_parse_statement(Parser *P) {
expect(P, TOKEN_SEMICOLON); expect(P, TOKEN_SEMICOLON);
pushstat(P, ret); pushstat(P, ret);
return; return;
} else if(peek(P, 0).type == TOKEN_IDENTIFIER) { } else if(peek(P, 0).type == TOKEN_IDENTIFIER) {
if(!strcmp(peek(P, 0).content, "@align")) { if(!strcmp(peek(P, 0).content, "@align")) {
@ -920,9 +836,9 @@ void nct_parse_statement(Parser *P) {
ret->what = e; ret->what = e;
ret->to = nct_cast_expr(nct_parse_expression(P, 0), ret->what->expression.type); ret->to = nct_cast_expr(nct_parse_expression(P, 0), ret->what->expression.type);
if(ret->what->nodeKind == AST_EXPR_VAR) { //if(ret->what->nodeKind == AST_EXPR_VAR) {
reachingdefs_set(ret->what->exprVar.thing->data.var.reachingDefs, (AST*) ret); // reachingdefs_set(ret->what->exprVar.thing->data.var.reachingDefs, (AST*) ret);
} //}
expect(P, TOKEN_SEMICOLON); expect(P, TOKEN_SEMICOLON);
@ -942,7 +858,7 @@ void nct_parse_statement(Parser *P) {
} }
ASTChunk *nct_parse_chunk(Parser *P, int isTopLevel, int varPrioritize) { ASTChunk *nct_parse_chunk(Parser *P, int isTopLevel, int varPrioritize) {
AST *ret = malloc(sizeof(ASTChunk)); AST *ret = calloc(1, sizeof(ASTChunk));
ret->nodeKind = AST_CHUNK; ret->nodeKind = AST_CHUNK;
ret->chunk.statementFirst = ret->chunk.statementLast = NULL; ret->chunk.statementFirst = ret->chunk.statementLast = NULL;
ret->chunk.varCount = 0; ret->chunk.varCount = 0;
@ -957,7 +873,7 @@ ASTChunk *nct_parse_chunk(Parser *P, int isTopLevel, int varPrioritize) {
P->topLevel = &ret->chunk; P->topLevel = &ret->chunk;
} }
vartable_new_reachingdefs_for_all_vars(P->scope); //vartable_new_reachingdefs_for_all_vars(P->scope);
/* Find all symbol names and struct types ahead of time. Searches for colons as those can only mean symbol declarations */ /* Find all symbol names and struct types ahead of time. Searches for colons as those can only mean symbol declarations */
@ -1001,7 +917,7 @@ ASTChunk *nct_parse_chunk(Parser *P, int isTopLevel, int varPrioritize) {
nct_parse_statement(P); nct_parse_statement(P);
} }
vartable_coalesce_reachingdefs_for_all_vars(P->scope); //vartable_coalesce_reachingdefs_for_all_vars(P->scope);
size_t nonSymbols = 0; size_t nonSymbols = 0;
for(size_t i = 0; i < P->scope->count; i++) { for(size_t i = 0; i < P->scope->count; i++) {

2
src/reporting.h
View File

@ -1,7 +1,7 @@
#ifndef NCTREF_REPORTING_H #ifndef NCTREF_REPORTING_H
#define NCTREF_REPORTING_H #define NCTREF_REPORTING_H
#ifndef _GNUC #ifndef __GNUC__
#define __attribute__(x) #define __attribute__(x)
#endif #endif

6
src/types.c
View File

@ -132,6 +132,10 @@ Type *type_pointer_wrap(Type *t) {
return (Type*) ret; return (Type*) ret;
} }
int type_is_number(Type *t) {
return t->type == TYPE_TYPE_POINTER || t->type == TYPE_TYPE_PRIMITIVE;
}
int type_is_castable(Type *from, Type *to) { int type_is_castable(Type *from, Type *to) {
if(type_equal(from, to)) return 2; if(type_equal(from, to)) return 2;
@ -152,4 +156,4 @@ int type_is_castable(Type *from, Type *to) {
} }
return 0; return 0;
} }

4
src/types.h
View File

@ -71,4 +71,8 @@ Type *type_pointer_wrap(Type*);
/* 0 = not castable, 1 = explicitly castable, 2 = implicitly castable */ /* 0 = not castable, 1 = explicitly castable, 2 = implicitly castable */
int type_is_castable(Type *from, Type *to); int type_is_castable(Type *from, Type *to);
int type_is_number(Type *t);
char *type_to_string(Type*);
#endif #endif

5
src/vartable.h
View File

@ -2,9 +2,10 @@
#define NCTREF_VARTABLE_H #define NCTREF_VARTABLE_H
#include"types.h" #include"types.h"
#include<stdbool.h>
typedef enum { typedef enum {
VARTABLEENTRY_SYMBOL, VARTABLEENTRY_TYPE, VARTABLEENTRY_VAR VARTABLEENTRY_SYMBOL, VARTABLEENTRY_STACK, VARTABLEENTRY_VAR
} VarTableEntryKind; } VarTableEntryKind;
union AST; union AST;
@ -49,7 +50,7 @@ typedef struct VarTableEntry {
// Register allocation // Register allocation
// vars in loops have higher priority // vars in loops have higher priority
// a more advanced approach would be to use weights for different colors (e.g. vars for mul "should" be in eax) // a more advanced approach would be to use weights for different colors (e.g. multipliers "should" be in eax)
uint8_t priority; uint8_t priority;
uint16_t color, degree; uint16_t color, degree;

2
src/x86.h
View File

@ -29,6 +29,8 @@ static inline int is_xop(AST *e) {
} }
} }
} }
} else if(e->nodeKind == AST_EXPR_STACK_POINTER) {
return XOP_NOT_MEM;
} }
return XOP_NOT_XOP; return XOP_NOT_XOP;

9
tests/bit-rounding.nct
View File

@ -1,6 +1,11 @@
u32 x: 123; u32 x: 123;
u33 y: 5; u33 y: 5;
u3 *o = 5000; u3* o = 5000;
u3 z = *o; u32 z = *o;
u3 p = *(o + z); u3 p = *(o + z);
u3 *g = o - z;
p = *g;
u32 c = 566;
u16 j = c;

6
tests/spill.nct Normal file
View File

@ -0,0 +1,6 @@
u8 a = 4;
u8 b = 1;
u8 c = 9;
u8 d = 5;
u8 e = 22;
u8 f = a + b + c + d + e;

1
tests/stack.nct Normal file
View File

@ -0,0 +1 @@
@stack = @stack - 123;