diff --git a/src/asm.c b/src/asm.c
index 04e42b3..78917ae 100644
--- a/src/asm.c
+++ b/src/asm.c
@@ -13,8 +13,31 @@ void inst_jump(symbol sym) {
     int32_t disp = symbol_offset(sym, X86_JMP_DISP8_SIZE);
     if (disp >= INT8_MIN && disp <= INT8_MAX) {
         x86_inst_jmp_disp8(disp);
+    } else {
+        x86_inst_jmp_disp32_op();
+        relocate_pc32(sym);
     }
-    x86_inst_jmp_disp32_op();
-    relocate_pc32(sym);
     // TODO: support 64-bit jumps?
 }
+
+void inst_jump_if_zero(symbol sym, reg reg) {
+    x86_inst_test_r64_r64(reg, reg);
+    int32_t disp = symbol_offset(sym, X86_JZ_DISP8_SIZE);
+    if (disp >= INT8_MIN && disp <= INT8_MAX) {
+        x86_inst_jz_disp8(disp);
+    } else {
+        x86_inst_jz_disp32_op();
+        relocate_pc32(sym);
+    }
+}
+
+void inst_jump_if_not_zero(symbol sym, reg reg) {
+    x86_inst_test_r64_r64(reg, reg);
+    int32_t disp = symbol_offset(sym, X86_JNZ_DISP8_SIZE);
+    if (disp >= INT8_MIN && disp <= INT8_MAX) {
+        x86_inst_jnz_disp8(disp);
+    } else {
+        x86_inst_jnz_disp32_op();
+        relocate_pc32(sym);
+    }
+}
diff --git a/src/asm.h b/src/asm.h
index 4e7e196..235bccf 100644
--- a/src/asm.h
+++ b/src/asm.h
@@ -2,8 +2,15 @@
 #define _ASM_H
 
 #include "format.h"
+#include "x86encode.h"
 
 /// Jump to a known address.
 void inst_jump(symbol sym);
 
+/// Jump to a known address if the argument is zero.
+void inst_jump_if_zero(symbol sym, reg reg);
+
+/// Jump to a known address if the argument is not zero.
+void inst_jump_if_not_zero(symbol sym, reg reg);
+
 #endif
diff --git a/src/format.c b/src/format.c
index d5f8e43..4840410 100644
--- a/src/format.c
+++ b/src/format.c
@@ -100,7 +100,9 @@ void finish_executable(symbol entry_point) {
         assert(rel.type == REL_PC32);
         uint64_t vaddr = symbols[rel.symbol].vaddr;
         assert(vaddr != (uint64_t) -1);
-        int64_t disp = (int64_t) rel.offset - (int64_t) vaddr + 4;
+        // PC-relative is from the *end* of the instruction,
+        // and the displacement is 4 bytes (32 bits).
+        int64_t disp = (int64_t) vaddr - ((int64_t) rel.offset + 4);
         assert(disp >= INT32_MIN && disp <= INT32_MAX);
         patch_u32(rel.offset, (int32_t) disp);
     }
@@ -132,14 +134,17 @@ void append_data(size_t size, const void* buf) {
 }
 
 void append_u8(uint8_t x) {
+    file_here += 1;
     emit_u8(x);
 }
 
 void append_u32(uint32_t x) {
+    file_here += 4;
     emit_u32(x);
 }
 
 void append_u64(uint64_t x) {
+    file_here += 8;
     emit_u64(x);
 }
 
@@ -159,11 +164,11 @@ void relocate_pc32(symbol sym) {
         append_u32((uint32_t) offset);
         return;
     }
-    append_u32(0);
     struct relocation* rel = new_relocation();
     rel->type = REL_PC32;
     rel->offset = file_here;
     rel->symbol = sym;
+    append_u32(0);
 }
 
 int32_t symbol_offset(symbol sym, int8_t off) {
@@ -171,7 +176,7 @@ int32_t symbol_offset(symbol sym, int8_t off) {
     if (vaddr == (uint64_t) -1) {
         return INT32_MAX;
     }
-    int64_t disp = (int64_t) file_here - (int64_t) vaddr + off;
+    int64_t disp = (int64_t) vaddr - ((int64_t) file_here + off);
     if (disp >= INT32_MAX || disp <= INT32_MIN) {
         return INT32_MAX;
     }
diff --git a/src/ir.c b/src/ir.c
index f89cb42..73c064f 100644
--- a/src/ir.c
+++ b/src/ir.c
@@ -34,13 +34,15 @@ struct label {
 static uint32_t stack_depth = 0;
 static uint32_t stack_frame = 0;
 static struct stack_frame stack_frames[MAX_STACK_FRAMES];
-static uint32_t label_depth;
+static uint32_t label_depth = 0;
 static struct label labels[MAX_LABELS];
 
-void init(var* argc, var* argv, var* env) {
+void init_ir(var* argc, var* argv, var* env) {
     assert(stack_depth == 0);
+    // seems like this should be necessary. it really feels like there's some
+    // off-by-one arrows going on around here, but I can't figure it out.
+    //enter();
     x86_inst_mov_r64_r64(BP, SP);
-    // TODO: replace with add, once I implement add
     x86_inst_add_r64_imm8(BP, 8 * 3);
     *env = stack_depth++;
     *argv = stack_depth++;
@@ -58,10 +60,10 @@ void enter(void) {
 
 void leave(var* args) {
     assert(stack_frame > 0);
-    struct stack_frame frame = stack_frames[stack_frame];
-    define(frame.label_depth, args);
+    struct stack_frame frame = stack_frames[stack_frame - 1];
+    stack_frame--;
     stack_depth = frame.depth;
-    label_depth = frame.label_depth;
+    define(frame.label_depth, args);
 }
 
 label declare(uint32_t argc) {
@@ -73,7 +75,7 @@ label declare(uint32_t argc) {
 }
 
 label declare_exit(uint32_t argc) {
-    label label = stack_frames[stack_frame].label_depth;
+    label label = stack_frames[stack_frame - 1].label_depth;
     labels[label].argc = argc;
     return label;
 }
@@ -81,19 +83,17 @@ label declare_exit(uint32_t argc) {
 void define(label l, var* args) {
     struct label* label = &labels[l];
     define_executable_symbol(label->symbol);
-}
-
-void jump(label l, var* args) {
-    struct label* label = &labels[l];
-    inst_jump(label->symbol);
-}
-
-void jump_table(size_t branches, label* labels, var index, var* args) {
-    assert(0); // UNIMPLEMENTED
-}
-
-void jump_if(label label, var cond, var* args) {
-    //assert(0); // UNIMPLEMENTED
+    // possibly wrong. do I need to do any clean-up of the old frame here?
+    stack_frame = label->frame;
+    struct stack_frame* frame = &stack_frames[stack_frame - 1];
+    label_depth = frame->label_depth;
+    for (uint32_t i = 0; i < label->argc; i++) {
+        args[i] = frame->depth + i;
+    }
+    // probably wrong. seems like I ought to create a new frame or something?
+    // wouldn't this make the old frame too deep?
+    // but on the other hand, if I enter a new frame, how do I decide when to leave it?
+    frame->depth += label->argc;
 }
 
 void load_var(reg reg, var var) {
@@ -108,6 +108,45 @@ var push_var(reg reg) {
     return stack_depth++;
 }
 
+void load_args(struct label* label, var* args) {
+    struct stack_frame* cur_frame = &stack_frames[stack_frame - 1];
+    struct stack_frame* dest_frame = &stack_frames[label->frame];
+    uint32_t depth_diff = cur_frame->depth - dest_frame->depth;
+    if (depth_diff > 0) {
+        // FIXME: should be immX!!!
+        // FIXME: this should be necessary! stack depth is never getting decreased!
+        //x86_inst_add_r64_imm8(SP, depth_diff);
+    }
+    for (uint32_t arg = 0; arg < label->argc; arg++) {
+        load_var(AX, args[arg]);
+        x86_inst_push_r64(AX);
+    }
+}
+
+void jump(label l, var* args) {
+    struct label* label = &labels[l];
+    load_args(label, args);
+    inst_jump(label->symbol);
+}
+
+void jump_table(size_t branches, label* labels, var index, var* args) {
+    assert(0); // UNIMPLEMENTED
+}
+
+void jump_if(label l, var cond, var* args) {
+    struct label* label = &labels[l];
+    load_var(BX, cond);
+    load_args(label, args);
+    inst_jump_if_not_zero(label->symbol, BX);
+}
+
+void jump_unless(label l, var cond, var* args) {
+    struct label* label = &labels[l];
+    load_var(BX, cond);
+    load_args(label, args);
+    inst_jump_if_zero(label->symbol, BX);
+}
+
 var lit(uint64_t lit) {
     x86_inst_mov_r64_imm(AX, lit);
     x86_inst_push_r64(AX);
diff --git a/src/ir.h b/src/ir.h
index 9f8034e..408cd27 100644
--- a/src/ir.h
+++ b/src/ir.h
@@ -9,7 +9,7 @@ typedef uint32_t label;
 
 /// Call this at the beginning of execution.
 /// It performs initialization and stuff.
-void init(var* argc, var* argv, var* env);
+void init_ir(var* argc, var* argv, var* env);
 
 /// Enter a new block.
 ///
@@ -64,6 +64,9 @@ void jump_table(size_t branches, label* labels, var index, var* args);
 /// Jump to label if `cond` is not zero.
 void jump_if(label label, var cond, var* args);
 
+/// Jump to label if `cond` is zero.
+void jump_unless(label label, var cond, var* args);
+
 /// Integer literal.
 var lit(uint64_t lit);
 
diff --git a/src/lang.c b/src/lang.c
index 99bf983..9337271 100644
--- a/src/lang.c
+++ b/src/lang.c
@@ -286,7 +286,7 @@ static struct label_and_arity lookup_label(enum label_type type, char* name) {
     exit(1);
 }
 
-void reduce_expression_binop(struct expr_crumb* exprc, var (*emit)(var arg1, var arg2)) {
+static void reduce_expression_binop(struct expr_crumb* exprc, var (*emit)(var arg1, var arg2)) {
     assert(exprc->argument_count >= 2);
     var arg1 = exprc->arguments[0];
     var arg2 = exprc->arguments[1];
@@ -295,7 +295,7 @@ void reduce_expression_binop(struct expr_crumb* exprc, var (*emit)(var arg1, var
     exprc->argument_count--;
 }
 
-var reduce_expression(struct expr_crumb* exprc) {
+static var reduce_expression(struct expr_crumb* exprc) {
     // TODO FIXME: operator precedence
     if (exprc->operator_count > 0 || exprc->argument_count > 1) {
         fprintf(stderr, "warning: expression reduction may be incorrect\n");
@@ -350,6 +350,7 @@ void enter_block(void) {
             struct loop_crumb* loopc = &ctx->data.loop;
             assert(loopc->state == LOOP_CLEAN);
             loopc->state = LOOP_BODY;
+            loopc->next = declare(loopc->assignment_count);
             var args[MAX_ASSIGNMENTS];
             define(loopc->next, args);
             // TODO NOTE: is this the correct order?
@@ -440,8 +441,7 @@ void exit_expr(void) {
     assert(ctx->type == EXPR_CRUMB);
     struct expr_crumb* exprc = &ctx->data.expr;
     assert(exprc->argument_count > 0);
-    reduce_expression(exprc);
-    var ret = exprc->arguments[0];
+    var ret = reduce_expression(exprc);
     context_depth--;
     ctx = &context[context_depth - 1];
     switch (ctx->type) {
@@ -465,8 +465,8 @@ void exit_expr(void) {
         case IF_CRUMB: {
             struct if_crumb* ifc = &ctx->data.if_;
             assert(ifc->state == IF_COND);
-            jump_if(ifc->then, ret, NULL);
-            jump(ifc->else_, NULL);
+            jump_unless(ifc->else_, ret, NULL);
+            //jump(ifc->then_, NULL);
             ifc->state = IF_THEN;
             break;
         }
diff --git a/src/main.c b/src/main.c
index 5530d50..d426591 100644
--- a/src/main.c
+++ b/src/main.c
@@ -14,14 +14,14 @@ symbol compile(void) {
     symbol entry_point = new_symbol();
     define_executable_symbol(entry_point);
     var argc, argv, env;
-    init(&argc, &argv, &env);
+    init_ir(&argc, &argv, &env);
+    parse();
     var a = lit(52);
     var b = lit(10);
     var exit_code = sub(a, b);
     var sys_exit = lit(60);
     var args[2] = { sys_exit, exit_code };
     syscall(2, args);
-    parse();
     return entry_point;
 }
 
diff --git a/src/x86encode.c b/src/x86encode.c
index f806ec4..f4c7dc2 100644
--- a/src/x86encode.c
+++ b/src/x86encode.c
@@ -226,6 +226,10 @@ void x86_inst_pop_r64(reg reg) {
     x86_enc_opr(0x58, reg);
 }
 
+void x86_inst_test_r64_r64(reg r1, reg r2) {
+    x86_enc_rexw_modrr(0x85, r1, r2);
+}
+
 void x86_inst_jmp_disp8(int8_t disp) {
     x86_enc_disp8(0xeb, disp);
 }
@@ -247,6 +251,52 @@ void x86_inst_jmp_disp32_op(void) {
     append_u8(0xe9);
 }
 
+void x86_inst_jnz_disp8(int8_t disp) {
+    x86_enc_disp8(0x75, disp);
+}
+
+void x86_inst_jnz_disp32(int32_t disp) {
+    append_u8(0x0f);
+    x86_enc_disp32(0x85, disp);
+}
+
+void x86_inst_jnz_disp(int32_t disp) {
+    int32_t disp8 = disp + X86_JNZ_DISP8_SIZE;
+    if (disp8 >= INT8_MIN && disp8 <= INT8_MAX) {
+        x86_inst_jnz_disp8((int8_t) disp);
+    } else {
+        x86_inst_jnz_disp32(disp + X86_JNZ_DISP32_SIZE);
+    }
+}
+
+void x86_inst_jnz_disp32_op(void) {
+    append_u8(0x0f);
+    append_u8(0x84);
+}
+
+void x86_inst_jz_disp8(int8_t disp) {
+    x86_enc_disp8(0x74, disp);
+}
+
+void x86_inst_jz_disp32(int32_t disp) {
+    append_u8(0x0f);
+    x86_enc_disp32(0x84, disp);
+}
+
+void x86_inst_jz_disp(int32_t disp) {
+    int32_t disp8 = disp + X86_JZ_DISP8_SIZE;
+    if (disp8 >= INT8_MIN && disp8 <= INT8_MAX) {
+        x86_inst_jz_disp8((int8_t) disp);
+    } else {
+        x86_inst_jz_disp32(disp + X86_JZ_DISP32_SIZE);
+    }
+}
+
+void x86_inst_jz_disp32_op(void) {
+    append_u8(0x0f);
+    append_u8(0x84);
+}
+
 // TODO: special instructions for AX
 void x86_inst_sub_r64_imm8(reg dest, int8_t imm) {
     x86_enc_rexw_modxm_imm8(0x83, 5, dest, (uint8_t) imm);
@@ -270,7 +320,7 @@ void x86_inst_add_r64_imm8(reg dest, int8_t imm) {
 }
 
 void x86_inst_add_r64_r64(reg dest, reg src) {
-    x86_enc_rexw_modrr(0x0, src, dest);
+    x86_enc_rexw_modrr(0x01, src, dest);
 }
 
 void x86_inst_syscall(void) {
diff --git a/src/x86encode.h b/src/x86encode.h
index 0ff4570..96f7091 100644
--- a/src/x86encode.h
+++ b/src/x86encode.h
@@ -42,6 +42,8 @@ void x86_inst_mov_m64_r64_disp(reg dest, reg src, int32_t disp);
 void x86_inst_push_r64(reg reg);
 void x86_inst_pop_r64(reg reg);
 
+void x86_inst_test_r64_r64(reg r1, reg r2);
+
 #define X86_JMP_DISP8_SIZE 2
 void x86_inst_jmp_disp8(int8_t disp);
 #define X86_JMP_DISP32_SIZE 5
@@ -49,6 +51,20 @@ void x86_inst_jmp_disp32(int32_t disp);
 void x86_inst_jmp_disp(int32_t disp);
 void x86_inst_jmp_disp32_op(void);
 
+#define X86_JNZ_DISP8_SIZE 2
+void x86_inst_jnz_disp8(int8_t disp);
+#define X86_JNZ_DISP32_SIZE 6
+void x86_inst_jnz_disp32(int32_t disp);
+void x86_inst_jnz_disp(int32_t disp);
+void x86_inst_jnz_disp32_op(void);
+
+#define X86_JZ_DISP8_SIZE 2
+void x86_inst_jz_disp8(int8_t disp);
+#define X86_JZ_DISP32_SIZE 6
+void x86_inst_jz_disp32(int32_t disp);
+void x86_inst_jz_disp(int32_t disp);
+void x86_inst_jz_disp32_op(void);
+
 void x86_inst_sub_r64_imm8(reg dest, int8_t imm);
 void x86_inst_sub_r64_imm32(reg dest, int32_t imm);
 void x86_inst_sub_r64_imm(reg dest, int32_t imm);