[Tinycc-devel] PE target: shorter function prologs

[Zdenek Pavlas]

Hello everybody,

Because of Windows' draconic handling of page faults 'sub esp, 4096' is 
a no-no.  Rare functions needing page or more of stack space must call 
chkstk(stacksize) first.  TCC needs 10 bytes for such call while the 
standard function prolog was only 9 so it pads them to 10 bytes.

This patch uses a shorter chkstk() calling convention that fits in 9 
bytes so no more padding.  Please check and apply if ok.

regards

--
Zdenek Pavlas



diff -ur tinycc-1eefdf914c2d/i386/i386-gen.c tinycc/i386/i386-gen.c
--- tinycc-1eefdf914c2d/i386/i386-gen.c 2007-05-12 06:33:24.000000000 +0200
+++ tinycc/i386/i386-gen.c      2007-05-14 11:52:58.000000000 +0200
@@ -407,11 +407,7 @@
     vtop--;
 }
 
-#ifdef TCC_TARGET_PE
-#define FUNC_PROLOG_SIZE 10
-#else
 #define FUNC_PROLOG_SIZE 9
-#endif
 
 /* generate function prolog of type 't' */
 void gfunc_prolog(CType *func_type)
@@ -541,7 +537,6 @@
 #ifdef TCC_TARGET_PE
     if (v >= 4096) {
         Sym *sym = external_global_sym(TOK___chkstk, &func_old_type, 0);
-        oad(0xb8, v); /* mov stacksize, %eax */
         oad(0xe8, -4); /* call __chkstk, (does the stackframe too) */
         greloc(cur_text_section, sym, ind-4, R_386_PC32);
     } else
@@ -549,11 +544,8 @@
     {
         o(0xe58955);  /* push %ebp, mov %esp, %ebp */
         o(0xec81);  /* sub esp, stacksize */
-        gen_le32(v);
-#if FUNC_PROLOG_SIZE == 10
-        o(0x90);  /* adjust to FUNC_PROLOG_SIZE */
-#endif
     }
+    gen_le32(v);
     ind = saved_ind;
 }
 
diff -ur tinycc-1eefdf914c2d/win32/lib/chkstk.S tinycc/win32/lib/chkstk.S
--- tinycc-1eefdf914c2d/win32/lib/chkstk.S      2007-05-12 06:33:24.000000000 
+0200
+++ tinycc/win32/lib/chkstk.S   2007-05-14 11:52:29.000000000 +0200
@@ -6,6 +6,8 @@
 
 __chkstk:
        xchg    (%esp), %ebp   // store ebp, get ret.addr
+       mov     (%ebp), %eax   // get stacksize
+       lea     4(%ebp), %ebp  // skip it
        push    %ebp           // push ret.addr
        lea     4(%esp), %ebp  // setup frame ptr
        push    %ecx           // save ecx