use call-clobbered reg to disalign the stack
diff mbox series

Message ID orh84kt0t1.fsf@lxoliva.fsfla.org
State New
Headers show
Series
  • use call-clobbered reg to disalign the stack
Related show

Commit Message

Alexandre Oliva Oct. 8, 2019, 2:42 a.m. UTC
Some x86 tests of stack realignment, that disaligned the stack with
pushes and pops, failed when the compiler was configured to tune for a
target that preferred to accumulate outgoing arguments: the stack
space is reserved before the asm push, the call sequence overwrites
the saved register, and then the asm pop restores the overwritten
value.  Since that's a call-preserved register in 32-bit mode, it
should be preserved unchanged, but isn't.

Merely changing the register to a call-clobbered one would be enough,
but the tests would remain fragile and prone to failure due to other
optimizations, so I arranged for the compiler to be made aware of the
register used for the push and the pop, so it won't use it for
something else, and forced the function to use a frame pointer, so
that it won't use stack pointer offsets for local variables: the
offsets would likely be wrong between the asm push and pop.

Tested on x86_64-linux-gnu with -m64 and -m32.  Ok to install?


for  gcc/testsuite/ChangeLog

	* gcc.target/i386/20060512-1.c (sse2_test): Use a
	call-clobbered register variable for stack-disaligning push
	and pop.  Require a frame pointer.
	* gcc.target/i386/20060512-3.c (sse2_test): Likewise.
---
 gcc/testsuite/gcc.target/i386/20060512-1.c |   16 +++++++++-------
 gcc/testsuite/gcc.target/i386/20060512-3.c |    7 ++++---
 2 files changed, 13 insertions(+), 10 deletions(-)

Comments

Jeff Law Oct. 14, 2019, 9:24 p.m. UTC | #1
On 10/7/19 8:42 PM, Alexandre Oliva wrote:
> Some x86 tests of stack realignment, that disaligned the stack with
> pushes and pops, failed when the compiler was configured to tune for a
> target that preferred to accumulate outgoing arguments: the stack
> space is reserved before the asm push, the call sequence overwrites
> the saved register, and then the asm pop restores the overwritten
> value.  Since that's a call-preserved register in 32-bit mode, it
> should be preserved unchanged, but isn't.
> 
> Merely changing the register to a call-clobbered one would be enough,
> but the tests would remain fragile and prone to failure due to other
> optimizations, so I arranged for the compiler to be made aware of the
> register used for the push and the pop, so it won't use it for
> something else, and forced the function to use a frame pointer, so
> that it won't use stack pointer offsets for local variables: the
> offsets would likely be wrong between the asm push and pop.
> 
> Tested on x86_64-linux-gnu with -m64 and -m32.  Ok to install?
> 
> 
> for  gcc/testsuite/ChangeLog
> 
> 	* gcc.target/i386/20060512-1.c (sse2_test): Use a
> 	call-clobbered register variable for stack-disaligning push
> 	and pop.  Require a frame pointer.
> 	* gcc.target/i386/20060512-3.c (sse2_test): Likewise.
OK
jeff
> ---

Patch
diff mbox series

diff --git a/gcc/testsuite/gcc.target/i386/20060512-1.c b/gcc/testsuite/gcc.target/i386/20060512-1.c
index ec163a9bc51e9..fe95f6d52fa98 100644
--- a/gcc/testsuite/gcc.target/i386/20060512-1.c
+++ b/gcc/testsuite/gcc.target/i386/20060512-1.c
@@ -7,11 +7,11 @@ 
 #include <emmintrin.h>
 
 #ifdef __x86_64__
-# define PUSH "pushq %rsi"
-# define POP "popq %rsi"
+# define REG "rcx"
+# define WIDTH "q"
 #else
-# define PUSH "pushl %esi"
-# define POP "popl %esi"
+# define REG "ecx"
+# define WIDTH "l"
 #endif
 
 __m128i __attribute__ ((__noinline__))
@@ -30,13 +30,15 @@  self_aligning_function (int x, int y)
 int g_1 = 20;
 int g_2 = 22;
 
-static void
+static void __attribute__ ((__optimize__ ("-fno-omit-frame-pointer")))
 sse2_test (void)
 {
   int result;
-  asm (PUSH);                  /* Misalign runtime stack.  */
+  register int __attribute__ ((__mode__ (__word__))) reg asm (REG);
+  asm volatile ("push" WIDTH "\t%0"  /* Disalign runtime stack.  */
+		: : "r" (reg) : "memory");
   result = self_aligning_function (g_1, g_2);
   if (result != 42)
     abort ();
-  asm (POP);
+  asm volatile ("pop" WIDTH "\t%0" : "=r" (reg));
 }
diff --git a/gcc/testsuite/gcc.target/i386/20060512-3.c b/gcc/testsuite/gcc.target/i386/20060512-3.c
index 3370b9ec25afb..0cebb47f6e9bc 100644
--- a/gcc/testsuite/gcc.target/i386/20060512-3.c
+++ b/gcc/testsuite/gcc.target/i386/20060512-3.c
@@ -23,13 +23,14 @@  self_aligning_function (int x, int y)
 int g_1 = 20;
 int g_2 = 22;
 
-static void
+static void __attribute__ ((__optimize__ ("-fno-omit-frame-pointer")))
 sse2_test (void)
 {
   int result;
-  asm ("pushl %esi");		/* Disalign runtime stack.  */
+  register int reg asm ("ecx");
+  asm ("pushl\t%0": : "r" (reg) : "memory"); /* Disalign runtime stack.  */
   result = self_aligning_function (g_1, g_2);
   if (result != 42)
     abort ();
-  asm ("popl %esi");
+  asm ("popl\t%0" : "=r" (reg));
 }