From f96fa84e6a16f271515029cf9b05afb2e6099c03 Mon Sep 17 00:00:00 2001
From: tyfkda <tyfkda@gmail.com>
Date: Sat, 5 Oct 2024 09:36:27 +0900
Subject: [PATCH] Remove stack position counting

On x64,
stack pointer is aligned to 16
if the function contains function call in it.
---
 src/cc/arch/aarch64/ir_aarch64.c | 11 ++++++-----
 src/cc/arch/riscv64/ir_riscv64.c | 11 ++++++-----
 src/cc/arch/x64/emit_code.c      | 18 +++++-------------
 src/cc/arch/x64/emit_code.h      |  5 -----
 src/cc/arch/x64/ir_x64.c         | 24 +++++++-----------------
 src/cc/backend/codegen_expr.c    |  4 ----
 src/cc/backend/ir.c              |  2 +-
 7 files changed, 25 insertions(+), 50 deletions(-)

diff --git a/src/cc/arch/aarch64/ir_aarch64.c b/src/cc/arch/aarch64/ir_aarch64.c
index 50706df84..a2a981cde 100644
--- a/src/cc/arch/aarch64/ir_aarch64.c
+++ b/src/cc/arch/aarch64/ir_aarch64.c
@@ -670,8 +670,9 @@ static void ei_precall(IR *ir) {
   int align_stack = (16 - (ir->precall.stack_args_size)) & 15;
   ir->precall.stack_aligned = align_stack;
 
-  if (align_stack > 0) {
-    SUB(SP, SP, IM(align_stack));
+  int total = align_stack + ir->precall.stack_args_size;
+  if (total > 0) {
+    SUB(SP, SP, IM(total));
   }
 }
 
@@ -706,9 +707,9 @@ static void ei_call(IR *ir) {
   }
 
   IR *precall = ir->call.precall;
-  int align_stack = precall->precall.stack_aligned + precall->precall.stack_args_size;
-  if (align_stack != 0) {
-    ADD(SP, SP, IM(align_stack));
+  int total = precall->precall.stack_aligned + precall->precall.stack_args_size;
+  if (total != 0) {
+    ADD(SP, SP, IM(total));
   }
 
   // Resore caller save registers.
diff --git a/src/cc/arch/riscv64/ir_riscv64.c b/src/cc/arch/riscv64/ir_riscv64.c
index 75561a396..900c7445b 100644
--- a/src/cc/arch/riscv64/ir_riscv64.c
+++ b/src/cc/arch/riscv64/ir_riscv64.c
@@ -646,8 +646,9 @@ static void ei_precall(IR *ir) {
   int align_stack = (16 - (ir->precall.stack_args_size)) & 15;
   ir->precall.stack_aligned = align_stack;
 
-  if (align_stack > 0) {
-    ADDI(SP, SP, IM(-align_stack));
+  int total = align_stack + ir->precall.stack_args_size;
+  if (total > 0) {
+    ADDI(SP, SP, IM(-total));
   }
 }
 
@@ -691,9 +692,9 @@ static void ei_call(IR *ir) {
   }
 
   IR *precall = ir->call.precall;
-  int align_stack = precall->precall.stack_aligned + precall->precall.stack_args_size;
-  if (align_stack != 0) {
-    ADDI(SP, SP, IM(align_stack));
+  int total = precall->precall.stack_aligned + precall->precall.stack_args_size;
+  if (total != 0) {
+    ADDI(SP, SP, IM(total));
   }
 
   // Resore caller save registers.
diff --git a/src/cc/arch/x64/emit_code.c b/src/cc/arch/x64/emit_code.c
index dc6f7f226..f30e3a436 100644
--- a/src/cc/arch/x64/emit_code.c
+++ b/src/cc/arch/x64/emit_code.c
@@ -153,8 +153,6 @@ void emit_defun(Function *func) {
       func->extra == NULL)     // Code emission is omitted.
     return;
 
-  assert(stackpos == 8);
-
   emit_comment(NULL);
   _TEXT();
 
@@ -197,16 +195,17 @@ void emit_defun(Function *func) {
   FuncBackend *fnbe = func->extra;
   size_t frame_size = 0;
   bool rbp_saved = false;
+  int callee_saved_count = 0;
   if (!no_stmt) {
     // Callee save.
-    int callee_saved_count = push_callee_save_regs(fnbe->ra->used_reg_bits, fnbe->ra->used_freg_bits);
+    callee_saved_count = push_callee_save_regs(fnbe->ra->used_reg_bits, fnbe->ra->used_freg_bits);
 
     // When function is called, return address is pused onto the stack by caller,
     // so default offset is 8.
     size_t frame_offset = 8;
 
     if (fnbe->frame_size > 0 || fnbe->ra->flag & RAF_STACK_FRAME) {
-      PUSH(RBP); PUSH_STACK_POS();
+      PUSH(RBP);
       MOV(RSP, RBP);
       rbp_saved = true;
       // RBP is pushed so the 16-bytes-align offset becomes 0.
@@ -215,13 +214,12 @@ void emit_defun(Function *func) {
 
     size_t callee_saved_size = callee_saved_count * TARGET_POINTER_SIZE;
     frame_size = fnbe->frame_size;
-    if (func->flag & FUNCF_HAS_FUNCALL) {
+    if (func->flag & (FUNCF_HAS_FUNCALL | FUNCF_STACK_MODIFIED)) {
       // Align frame size to 16 only it contains funcall.
       frame_size += -(fnbe->frame_size + callee_saved_size + frame_offset) & 15;
     }
     if (frame_size > 0) {
       SUB(IM(frame_size), RSP);
-      stackpos += frame_size;
     }
 
     move_params_to_assigned(func);
@@ -234,21 +232,15 @@ void emit_defun(Function *func) {
     if (!no_stmt) {
       if (rbp_saved) {
         MOV(RBP, RSP);
-        stackpos -= frame_size;
-        POP(RBP); POP_STACK_POS();
+        POP(RBP);
       } else if (frame_size > 0) {
         ADD(IM(frame_size), RSP);
-        stackpos -= frame_size;
       }
 
       pop_callee_save_regs(fnbe->ra->used_reg_bits, fnbe->ra->used_freg_bits);
     }
 
     RET();
-
-    assert(stackpos == 8);
-  } else {
-    stackpos = 8;
   }
 
   // Static variables are emitted through global variables.
diff --git a/src/cc/arch/x64/emit_code.h b/src/cc/arch/x64/emit_code.h
index 238b1f196..96d4a63d7 100644
--- a/src/cc/arch/x64/emit_code.h
+++ b/src/cc/arch/x64/emit_code.h
@@ -6,11 +6,6 @@
 
 typedef struct Vector Vector;
 
-extern int stackpos;
-
-#define PUSH_STACK_POS()  do { stackpos += TARGET_POINTER_SIZE; } while (0)
-#define POP_STACK_POS()   do { stackpos -= TARGET_POINTER_SIZE; } while (0)
-
 char *im(int64_t x);  // $x
 char *indirect(const char *base, const char *index, int scale);
 char *offset_indirect(int offset, const char *base, const char *index, int scale);
diff --git a/src/cc/arch/x64/ir_x64.c b/src/cc/arch/x64/ir_x64.c
index f65ca39d1..f1b7a9549 100644
--- a/src/cc/arch/x64/ir_x64.c
+++ b/src/cc/arch/x64/ir_x64.c
@@ -16,8 +16,6 @@
 static Vector *push_caller_save_regs(unsigned long living);
 static void pop_caller_save_regs(Vector *saves);
 
-int stackpos = 8;
-
 // Register allocator
 
 const char *kRegSizeTable[][PHYSICAL_REG_MAX] = {
@@ -715,12 +713,12 @@ static void ei_precall(IR *ir) {
   // so safely saved before calculating argument values.
   ir->precall.caller_saves = push_caller_save_regs(ir->precall.living_pregs);
 
-  int align_stack = (16 - (stackpos + ir->precall.stack_args_size)) & 15;
+  int align_stack = (16 - (ir->precall.caller_saves->len * TARGET_POINTER_SIZE + ir->precall.stack_args_size)) & 15;
   ir->precall.stack_aligned = align_stack;
 
-  if (align_stack > 0) {
-    SUB(IM(align_stack), RSP);
-    stackpos += align_stack;
+  int total = align_stack + ir->precall.stack_args_size;
+  if (total > 0) {
+    SUB(IM(total), RSP);
   }
 }
 
@@ -775,10 +773,9 @@ static void ei_call(IR *ir) {
   }
 
   IR *precall = ir->call.precall;
-  int align_stack = precall->precall.stack_aligned + precall->precall.stack_args_size;
-  if (align_stack != 0) {
-    ADD(IM(align_stack), RSP);
-    stackpos -= precall->precall.stack_aligned;
+  int total = precall->precall.stack_aligned + precall->precall.stack_args_size;
+  if (total != 0) {
+    ADD(IM(total), RSP);
   }
 
   // Resore caller save registers.
@@ -834,7 +831,6 @@ static void ei_subsp(IR *ir) {
       SUB(IM(ir->opr1->fixnum), RSP);
     else if (ir->opr1->fixnum < 0)
       ADD(IM(-ir->opr1->fixnum), RSP);
-    // stackpos += ir->opr1->fixnum;
   } else {
     SUB(kReg64s[ir->opr1->phys], RSP);
   }
@@ -1008,7 +1004,6 @@ int push_callee_save_regs(unsigned long used, unsigned long fused) {
     int ireg = kCalleeSaveRegs[i];
     if (used & (1 << ireg)) {
       PUSH(kReg64s[ireg]);
-      PUSH_STACK_POS();
       ++count;
     }
   }
@@ -1023,7 +1018,6 @@ void pop_callee_save_regs(unsigned long used, unsigned long fused) {
     int ireg = kCalleeSaveRegs[i];
     if (used & (1 << ireg)) {
       POP(kReg64s[ireg]);
-      POP_STACK_POS();
     }
   }
 }
@@ -1044,7 +1038,6 @@ static Vector *push_caller_save_regs(unsigned long living) {
     if (living & (1UL << ireg)) {
       const char *reg = kReg64s[ireg];
       PUSH(reg);
-      PUSH_STACK_POS();
       vec_push(saves, reg);
     }
   }
@@ -1062,7 +1055,6 @@ static Vector *push_caller_save_regs(unsigned long living) {
     if (n > 0) {
       int ofs = n * TARGET_POINTER_SIZE;
       SUB(IM(ofs), RSP);
-      stackpos += ofs;
       for (int i = 0; i < n; ++i) {
         ofs -= TARGET_POINTER_SIZE;
         MOVSD(saves->data[i + fstart], OFFSET_INDIRECT(ofs, RSP, NULL, 1));
@@ -1085,14 +1077,12 @@ static void pop_caller_save_regs(Vector *saves) {
   }
   if (ofs > 0) {
     ADD(IM(ofs), RSP);
-    stackpos -= ofs;
   }
   ++i;
 
   while (--i >= 0) {
     const char *reg = saves->data[i];
     POP(reg);
-    POP_STACK_POS();
   }
 }
 
diff --git a/src/cc/backend/codegen_expr.c b/src/cc/backend/codegen_expr.c
index d46aad6d0..3880cd27f 100644
--- a/src/cc/backend/codegen_expr.c
+++ b/src/cc/backend/codegen_expr.c
@@ -410,13 +410,9 @@ static VReg *gen_funcall(Expr *expr) {
       }
     }
   }
-  offset = ALIGN(offset, 16);
 
   IR *precall = new_ir_precall(arg_count - stack_arg_count, offset);
 
-  if (offset > 0)
-    new_ir_subsp(new_const_vreg(offset, to_vsize(&tySSize)), NULL);
-
   int total_arg_count = arg_count + (ret_varinfo != NULL ? 1 : 0);
   VReg **arg_vregs = total_arg_count == 0 ? NULL : calloc_or_die(total_arg_count * sizeof(*arg_vregs));
 
diff --git a/src/cc/backend/ir.c b/src/cc/backend/ir.c
index 8634804d1..fd8fbdae5 100644
--- a/src/cc/backend/ir.c
+++ b/src/cc/backend/ir.c
@@ -270,7 +270,7 @@ IR *new_ir_precall(int arg_count, int stack_args_size) {
   IR *ir = new_ir(IR_PRECALL);
   ir->precall.arg_count = arg_count;
   ir->precall.stack_args_size = stack_args_size;
-  ir->precall.stack_aligned = false;
+  ir->precall.stack_aligned = 0;
   ir->precall.living_pregs = 0;
   ir->precall.caller_saves = NULL;
   return ir;