This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] i386: Don't use frame pointer without stack access
On Wed, Aug 9, 2017 at 10:28 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Wed, Aug 9, 2017 at 8:26 AM, Andi Kleen <andi@firstfloor.org> wrote:
>>> This must be much more specific. How does it impact:
>>>
>>> 1. LTO
>>> 2. Function inlining.
>>> 3. Partial function inlining.
>>> 4. Shrink-wrapping.
>>>
>>> Any of them can impact function stack frame.
>>
>> It doesn't. It's just to get back to the previous state.
>>
>> Also these others already have explicit options to disable them.
>>
>
> How about
>
> item -fkeep-frame-pointer
> @opindex fkeep-frame-pointer
> Keep the frame pointer in a register for functions. This option always
> forces a new stack frame for all functions regardless of whether
> @option{-fomit-frame-pointer} is enabled or not. Disabled by default.
>
Here is the updated patch with -fkeep-frame-pointer.
OK for trunk?
--
H.J.
From 6f5b42bbe2baa7977dcbf388219dabcb4a6b546d Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Sun, 6 Aug 2017 06:24:36 -0700
Subject: [PATCH] i386: Don't use frame pointer without stack access
When there is no stack access, there is no need to use frame pointer
even if -fno-omit-frame-pointer is used and caller's frame pointer is
unchanged. A command-line option, -fkeep-frame-pointer, is added to
always force a new stack frame.
gcc/
PR target/81736
* common.opt (-fomit-frame-pointer): Override
-fkeep-frame-pointer.
(-fkeep-frame-pointer): New command-line option.
* config/i386/i386.c (ix86_finalize_stack_realign_flags): Renamed
to ...
(ix86_finalize_stack_frame_flags): This. Also clear
frame_pointer_needed if -fkeep-frame-pointer isn't used and
,-fno-omit-frame-pointer is used without stack access.
(ix86_expand_prologue): Replace ix86_finalize_stack_realign_flags
with ix86_finalize_stack_frame_flags.
(ix86_expand_epilogue): Likewise.
(ix86_expand_split_stack_prologue): Likewise.
* doc/invoke.texi: Document -fkeep-frame-pointer. Update
-fomit-frame-pointer. Add a note for -fno-omit-frame-pointer.
* toplev.c (process_options): Disable -fomit-frame-pointer if
-fkeep-frame-pointer is used.
gcc/testsuite/
PR target/81736
* gcc.target/i386/pr81736-1a.c: New test.
* gcc.target/i386/pr81736-1b.c: Likewise.
* gcc.target/i386/pr81736-1c.c: Likewise.
* gcc.target/i386/pr81736-1d.c: Likewise.
* gcc.target/i386/pr81736-1e.c: Likewise.
* gcc.target/i386/pr81736-1f.c: Likewise.
* gcc.target/i386/pr81736-1g.c: Likewise.
* gcc.target/i386/pr81736-2.c: Likewise.
* gcc.target/i386/pr81736-3.c: Likewise.
* gcc.target/i386/pr81736-4.c: Likewise.
* gcc.target/i386/pr81736-5.c: Likewise.
* gcc.target/i386/pr81736-6.c: Likewise.
* gcc.target/i386/pr81736-7.c: Likewise.
---
gcc/common.opt | 6 +++++-
gcc/config/i386/i386.c | 24 +++++++++++++-----------
gcc/doc/invoke.texi | 13 ++++++++++++-
gcc/testsuite/gcc.target/i386/pr81736-1a.c | 13 +++++++++++++
gcc/testsuite/gcc.target/i386/pr81736-1b.c | 7 +++++++
gcc/testsuite/gcc.target/i386/pr81736-1c.c | 7 +++++++
gcc/testsuite/gcc.target/i386/pr81736-1d.c | 7 +++++++
gcc/testsuite/gcc.target/i386/pr81736-1e.c | 7 +++++++
gcc/testsuite/gcc.target/i386/pr81736-1f.c | 7 +++++++
gcc/testsuite/gcc.target/i386/pr81736-1g.c | 7 +++++++
gcc/testsuite/gcc.target/i386/pr81736-2.c | 14 ++++++++++++++
gcc/testsuite/gcc.target/i386/pr81736-3.c | 11 +++++++++++
gcc/testsuite/gcc.target/i386/pr81736-4.c | 11 +++++++++++
gcc/testsuite/gcc.target/i386/pr81736-5.c | 20 ++++++++++++++++++++
gcc/testsuite/gcc.target/i386/pr81736-6.c | 16 ++++++++++++++++
gcc/testsuite/gcc.target/i386/pr81736-7.c | 13 +++++++++++++
gcc/toplev.c | 4 ++++
17 files changed, 174 insertions(+), 13 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-1a.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-1b.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-1c.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-1d.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-1e.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-1f.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-1g.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-2.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-3.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-4.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-5.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-6.c
create mode 100644 gcc/testsuite/gcc.target/i386/pr81736-7.c
diff --git a/gcc/common.opt b/gcc/common.opt
index 1cb1c83d306..b73564d7145 100644
--- a/gcc/common.opt
+++ b/gcc/common.opt
@@ -1881,9 +1881,13 @@ EnumValue
Enum(offload_abi) String(lp64) Value(OFFLOAD_ABI_LP64)
fomit-frame-pointer
-Common Report Var(flag_omit_frame_pointer) Optimization
+Common Report Var(flag_omit_frame_pointer) Negative(fkeep-frame-pointer) Optimization
When possible do not generate stack frames.
+fkeep-frame-pointer
+Common Report Var(flag_keep_frame_pointer) Negative(fomit-frame-pointer)
+Always force a new stack frame
+
fopt-info
Common Report Var(flag_opt_info) Optimization
Enable all optimization info dumps on stderr.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 3f8519777f7..9c37e960dc8 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -14179,10 +14179,11 @@ output_probe_stack_range (rtx reg, rtx end)
return "";
}
-/* Finalize stack_realign_needed flag, which will guide prologue/epilogue
- to be generated in correct form. */
+/* Finalize stack_realign_needed and frame_pointer_needed flags, which
+ will guide prologue/epilogue to be generated in correct form. */
+
static void
-ix86_finalize_stack_realign_flags (void)
+ix86_finalize_stack_frame_flags (void)
{
/* Check if stack realign is really needed after reload, and
stores result in cfun */
@@ -14205,13 +14206,14 @@ ix86_finalize_stack_realign_flags (void)
}
/* If the only reason for frame_pointer_needed is that we conservatively
- assumed stack realignment might be needed, but in the end nothing that
- needed the stack alignment had been spilled, clear frame_pointer_needed
- and say we don't need stack realignment. */
- if (stack_realign
+ assumed stack realignment might be needed or -fno-omit-frame-pointer
+ is used, but in the end nothing that needed the stack alignment had
+ been spilled nor stack access, clear frame_pointer_needed and say we
+ don't need stack realignment. */
+ if ((stack_realign || !flag_omit_frame_pointer)
+ && !flag_keep_frame_pointer
&& frame_pointer_needed
&& crtl->is_leaf
- && flag_omit_frame_pointer
&& crtl->sp_is_unchanging
&& !ix86_current_function_calls_tls_descriptor
&& !crtl->accesses_prior_frames
@@ -14402,7 +14404,7 @@ ix86_expand_prologue (void)
if (ix86_function_naked (current_function_decl))
return;
- ix86_finalize_stack_realign_flags ();
+ ix86_finalize_stack_frame_flags ();
/* DRAP should not coexist with stack_realign_fp */
gcc_assert (!(crtl->drap_reg && stack_realign_fp));
@@ -15266,7 +15268,7 @@ ix86_expand_epilogue (int style)
return;
}
- ix86_finalize_stack_realign_flags ();
+ ix86_finalize_stack_frame_flags ();
frame = m->frame;
m->fs.sp_realigned = stack_realign_fp;
@@ -15801,7 +15803,7 @@ ix86_expand_split_stack_prologue (void)
gcc_assert (flag_split_stack && reload_completed);
- ix86_finalize_stack_realign_flags ();
+ ix86_finalize_stack_frame_flags ();
frame = cfun->machine->frame;
allocate = frame.stack_pointer_offset - INCOMING_FRAME_SP_OFFSET;
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index cc0f5a00a4f..7a3b3d53f17 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -397,7 +397,7 @@ Objective-C and Objective-C++ Dialects}.
-fno-peephole2 -fno-printf-return-value -fno-sched-interblock @gol
-fno-sched-spec -fno-signed-zeros @gol
-fno-toplevel-reorder -fno-trapping-math -fno-zero-initialized-in-bss @gol
--fomit-frame-pointer -foptimize-sibling-calls @gol
+-fomit-frame-pointer -fkeep-frame-pointer -foptimize-sibling-calls @gol
-fpartial-inlining -fpeel-loops -fpredictive-commoning @gol
-fprefetch-loop-arrays @gol
-fprofile-correction @gol
@@ -7365,9 +7365,20 @@ The default setting (when not optimizing for
size) for 32-bit GNU/Linux x86 and 32-bit Darwin x86 targets is
@option{-fomit-frame-pointer}. You can configure GCC with the
@option{--enable-frame-pointer} configure option to change the default.
+This option overrides @option{-fkeep-frame-pointer}.
+
+Note that @option{-fno-omit-frame-pointer} doesn't force a new stack
+frame for all functions if it isn't otherwise needed, and hence doesn't
+guarantee a new frame pointer for all functions.
Enabled at levels @option{-O}, @option{-O2}, @option{-O3}, @option{-Os}.
+@item -fkeep-frame-pointer
+@opindex fkeep-frame-pointer
+Keep the frame pointer in a register for functions. This option always
+forces a new stack frame for all functions. This option overrides
+@option{-fomit-frame-pointer}. Disabled by default.
+
@item -foptimize-sibling-calls
@opindex foptimize-sibling-calls
Optimize sibling and tail recursive calls.
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-1a.c b/gcc/testsuite/gcc.target/i386/pr81736-1a.c
new file mode 100644
index 00000000000..92c7bc97a0d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-1a.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer" } */
+
+extern int i;
+
+int
+foo (void)
+{
+ return i;
+}
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-1b.c b/gcc/testsuite/gcc.target/i386/pr81736-1b.c
new file mode 100644
index 00000000000..692ca9b7a30
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-1b.c
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fkeep-frame-pointer -fno-omit-frame-pointer" } */
+
+#include "pr81736-1a.c"
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-1c.c b/gcc/testsuite/gcc.target/i386/pr81736-1c.c
new file mode 100644
index 00000000000..36ee6bc63f0
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-1c.c
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fomit-frame-pointer -fno-omit-frame-pointer" } */
+
+#include "pr81736-1a.c"
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-1d.c b/gcc/testsuite/gcc.target/i386/pr81736-1d.c
new file mode 100644
index 00000000000..143985852f7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-1d.c
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fkeep-frame-pointer -fomit-frame-pointer" } */
+
+#include "pr81736-1a.c"
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-1e.c b/gcc/testsuite/gcc.target/i386/pr81736-1e.c
new file mode 100644
index 00000000000..fe98e96074c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-1e.c
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fomit-frame-pointer -fkeep-frame-pointer" } */
+
+#include "pr81736-1a.c"
+
+/* Need to use a frame pointer. */
+/* { dg-final { scan-assembler "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-1f.c b/gcc/testsuite/gcc.target/i386/pr81736-1f.c
new file mode 100644
index 00000000000..682cb3b4030
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-1f.c
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fkeep-frame-pointer" } */
+
+#include "pr81736-1a.c"
+
+/* Need to use a frame pointer. */
+/* { dg-final { scan-assembler "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-1g.c b/gcc/testsuite/gcc.target/i386/pr81736-1g.c
new file mode 100644
index 00000000000..b6d19985c8c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-1g.c
@@ -0,0 +1,7 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer -fkeep-frame-pointer" } */
+
+#include "pr81736-1a.c"
+
+/* Need to use a frame pointer. */
+/* { dg-final { scan-assembler "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-2.c b/gcc/testsuite/gcc.target/i386/pr81736-2.c
new file mode 100644
index 00000000000..a3720879937
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-2.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer" } */
+
+int
+#ifndef __x86_64__
+__attribute__((regparm(3)))
+#endif
+foo (int i)
+{
+ return i;
+}
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-3.c b/gcc/testsuite/gcc.target/i386/pr81736-3.c
new file mode 100644
index 00000000000..c3bde7dd933
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-3.c
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer" } */
+
+void
+foo (void)
+{
+ asm ("# " : : : "ebx");
+}
+
+/* Need to use a frame pointer. */
+/* { dg-final { scan-assembler "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-4.c b/gcc/testsuite/gcc.target/i386/pr81736-4.c
new file mode 100644
index 00000000000..25f50016a64
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-4.c
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer" } */
+
+int
+foo (int i1, int i2, int i3, int i4, int i5, int i6, int i7)
+{
+ return i7;
+}
+
+/* Need to use a frame pointer. */
+/* { dg-final { scan-assembler "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-5.c b/gcc/testsuite/gcc.target/i386/pr81736-5.c
new file mode 100644
index 00000000000..e1602cf25ba
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-5.c
@@ -0,0 +1,20 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer -mavx" } */
+
+typedef int v8si __attribute__ ((vector_size (32)));
+
+void
+#ifndef __x86_64__
+__attribute__((regparm(3)))
+#endif
+foo (v8si *out_start, v8si *out_end, v8si *regions)
+{
+ v8si base = regions[3];
+ *out_start = base;
+ *out_end = base;
+}
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
+/* Verify no dynamic realignment is performed. */
+/* { dg-final { scan-assembler-not "and\[^\n\r]*sp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-6.c b/gcc/testsuite/gcc.target/i386/pr81736-6.c
new file mode 100644
index 00000000000..6198574c8cc
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-6.c
@@ -0,0 +1,16 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer" } */
+
+struct foo
+{
+ int head;
+} a;
+
+int
+bar (void)
+{
+ return a.head != 0;
+}
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
diff --git a/gcc/testsuite/gcc.target/i386/pr81736-7.c b/gcc/testsuite/gcc.target/i386/pr81736-7.c
new file mode 100644
index 00000000000..f947886e642
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr81736-7.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fno-omit-frame-pointer" } */
+
+extern int foo (void);
+
+int
+bar (void)
+{
+ return foo ();
+}
+
+/* No need to use a frame pointer. */
+/* { dg-final { scan-assembler-not "%\[re\]bp" } } */
diff --git a/gcc/toplev.c b/gcc/toplev.c
index d23714c4773..55069784bdd 100644
--- a/gcc/toplev.c
+++ b/gcc/toplev.c
@@ -1297,6 +1297,10 @@ process_options (void)
"-floop-parallelize-all)");
#endif
+ /* Don't omit frame pointer if -fkeep-frame-pointer is used. */
+ if (flag_keep_frame_pointer)
+ flag_omit_frame_pointer = 0;
+
if (flag_check_pointer_bounds)
{
if (targetm.chkp_bound_mode () == VOIDmode)
--
2.13.4