===================================================================
@@ -13579,8 +13579,9 @@ ix86_expand_prologue (void)
}
m->fs.sp_offset += allocate;
- /* Use stack_pointer_rtx for relative addressing so that code
- works for realigned stack, too. */
+ /* Use stack_pointer_rtx for relative addressing so that code works for
+ realigned stack. But this means that we need a blockage to prevent
+ stores based on the frame pointer from being scheduled before. */
if (r10_live && eax_live)
{
t = gen_rtx_PLUS (Pmode, stack_pointer_rtx, eax);
@@ -13589,6 +13590,7 @@ ix86_expand_prologue (void)
t = plus_constant (Pmode, t, UNITS_PER_WORD);
emit_move_insn (gen_rtx_REG (word_mode, AX_REG),
gen_frame_mem (word_mode, t));
+ emit_insn (gen_memory_blockage ());
}
else if (eax_live || r10_live)
{
@@ -13596,6 +13598,7 @@ ix86_expand_prologue (void)
emit_move_insn (gen_rtx_REG (word_mode,
(eax_live ? AX_REG : R10_REG)),
gen_frame_mem (word_mode, t));
+ emit_insn (gen_memory_blockage ());
}
}
gcc_assert (m->fs.sp_offset == frame.stack_pointer_offset);
Hi, this is a regression present on all active branches: if you compile the attached Ada testcase with -O2 -gnatp -fno-omit-frame-pointer for 32-bit Windows, you'll see that the compiler swaps a load based on the stack pointer with a store based on the frame pointer, thus clobbering a saved argument: pushl %ebp movl %esp, %ebp pushl %esi pushl %ebx pushl %eax <- %eax save movl $4108, %eax call ___chkstk_ms leal 8(%ebp), %esi subl %eax, %esp movl 8(%ebp), %ebx movl %edx, -20(%ebp) movl %esi, -12(%ebp) <- fp-based store movl (%esp,%eax), %eax <- sp-based load ... wrong value in eax... The load and the store are swapped because there are not based on the same register and the offset between them is seen as variable. The proposed fix is to add a memory blockage, like in other frame-related constructs. Tested on x86/Windows and x86-64/Windows, OK for all active branches? 2019-02-06 Eric Botcazou <ebotcazou@adacore.com> * config/i386/i386.c (ix86_expand_prologue): Generate a memory blockage after restoring registers saved to allocate the frame on Windows. 2019-02-06 Eric Botcazou <ebotcazou@adacore.com> * gnat.dg/opt76.adb: New test.