Changes from v2:
(xtensa_expand_prologue): Changed to check conditions for suppressing emit insns in advance, instead of tracking emitted and later replacing them with NOPs if they are found to be unnecessary.
---
In the example below, 'x' is once placed on the stack frame and then read
into registers as the argument value of bar():
/* example */
struct foo {
int a, b;
};
extern struct foo bar(struct foo);
struct foo test(void) {
struct foo x = { 0, 1 };
return bar(x);
}
Thanks to the dead store elimination, the initialization of 'x' turns into
merely loading the immediates to registers, but corresponding stack frame
growth is not rolled back. As a result:
;; prereq: the CALL0 ABI
;; before
test:
addi sp, sp, -16 // unused stack frame allocation/freeing
movi.n a2, 0
movi.n a3, 1
addi sp, sp, 16 // because no instructions that refer to
j.l bar, a9 // the stack pointer between the two
This patch eliminates such unused stack frame allocation/freeing:
;; after
test:
movi.n a2, 0
movi.n a3, 1
j.l bar, a9
gcc/ChangeLog:
* config/xtensa/xtensa.cc (machine_function): New boolean member as
a flag that controls whether to emit the insns for stack pointer
adjustment inside of the pro/epilogue.
(xtensa_emit_adjust_stack_ptr): New function to share the common
codes and to emit insns if not inhibited.
(xtensa_expand_epilogue): Change to use the function mentioned
above when using the CALL0 ABI.
(xtensa_expand_prologue): Ditto.
And also change to set the inhibit flag used by
xtensa_emit_adjust_stack_ptr() to true if the stack pointer is only
used for its own adjustment.
---
gcc/config/xtensa/xtensa.cc | 162 +++++++++++++++++-------------------
1 file changed, 78 insertions(+), 84 deletions(-)
Hi Suwa-san,
On Wed, Sep 7, 2022 at 2:08 AM Takayuki 'January June' Suwa
<jjsuwa_sys3175@yahoo.co.jp> wrote:
>
> Changes from v2:
> (xtensa_expand_prologue): Changed to check conditions for suppressing emit insns in advance, instead of tracking emitted and later replacing them with NOPs if they are found to be unnecessary.
>
> ---
>
> In the example below, 'x' is once placed on the stack frame and then read
> into registers as the argument value of bar():
>
> /* example */
> struct foo {
> int a, b;
> };
> extern struct foo bar(struct foo);
> struct foo test(void) {
> struct foo x = { 0, 1 };
> return bar(x);
> }
>
> Thanks to the dead store elimination, the initialization of 'x' turns into
> merely loading the immediates to registers, but corresponding stack frame
> growth is not rolled back. As a result:
>
> ;; prereq: the CALL0 ABI
> ;; before
> test:
> addi sp, sp, -16 // unused stack frame allocation/freeing
> movi.n a2, 0
> movi.n a3, 1
> addi sp, sp, 16 // because no instructions that refer to
> j.l bar, a9 // the stack pointer between the two
>
> This patch eliminates such unused stack frame allocation/freeing:
>
> ;; after
> test:
> movi.n a2, 0
> movi.n a3, 1
> j.l bar, a9
>
> gcc/ChangeLog:
>
> * config/xtensa/xtensa.cc (machine_function): New boolean member as
> a flag that controls whether to emit the insns for stack pointer
> adjustment inside of the pro/epilogue.
> (xtensa_emit_adjust_stack_ptr): New function to share the common
> codes and to emit insns if not inhibited.
> (xtensa_expand_epilogue): Change to use the function mentioned
> above when using the CALL0 ABI.
> (xtensa_expand_prologue): Ditto.
> And also change to set the inhibit flag used by
> xtensa_emit_adjust_stack_ptr() to true if the stack pointer is only
> used for its own adjustment.
> ---
> gcc/config/xtensa/xtensa.cc | 162 +++++++++++++++++-------------------
> 1 file changed, 78 insertions(+), 84 deletions(-)
There's still a minor issue here: this change introduces a new regression
in the following test:
g++.dg/opt/pr100469.C
AFAICS it generates different code with and without the '-g' option: no
stack modification without -g, but adjustment in prologue and epilogue
with -g.
@@ -102,6 +102,7 @@ struct GTY(()) machine_function
int callee_save_size;
bool frame_laid_out;
bool epilogue_done;
+ bool inhibit_logues_a1_adjusts;
};
/* Vector, indexed by hard register number, which contains 1 for a
@@ -3048,7 +3049,7 @@ xtensa_output_literal (FILE *file, rtx x, machine_mode mode, int labelno)
}
static bool
-xtensa_call_save_reg(int regno)
+xtensa_call_save_reg (int regno)
{
if (TARGET_WINDOWED_ABI)
return false;
@@ -3084,7 +3085,7 @@ compute_frame_size (poly_int64 size)
cfun->machine->callee_save_size = 0;
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; ++regno)
{
- if (xtensa_call_save_reg(regno))
+ if (xtensa_call_save_reg (regno))
cfun->machine->callee_save_size += UNITS_PER_WORD;
}
@@ -3139,6 +3140,49 @@ xtensa_initial_elimination_offset (int from, int to ATTRIBUTE_UNUSED)
return offset;
}
+#define ADJUST_SP_NONE 0x0
+#define ADJUST_SP_NEED_NOTE 0x1
+#define ADJUST_SP_FRAME_PTR 0x2
+static void
+xtensa_emit_adjust_stack_ptr (HOST_WIDE_INT offset, int flags)
+{
+ rtx_insn *insn;
+ rtx ptr = (flags & ADJUST_SP_FRAME_PTR) ? hard_frame_pointer_rtx
+ : stack_pointer_rtx;
+
+ if (cfun->machine->inhibit_logues_a1_adjusts)
+ return;
+
+ if (xtensa_simm8 (offset)
+ || xtensa_simm8x256 (offset))
+ insn = emit_insn (gen_addsi3 (stack_pointer_rtx, ptr, GEN_INT (offset)));
+ else
+ {
+ rtx tmp_reg = gen_rtx_REG (Pmode, A9_REG);
+
+ if (offset < 0)
+ {
+ emit_move_insn (tmp_reg, GEN_INT (-offset));
+ insn = emit_insn (gen_subsi3 (stack_pointer_rtx, ptr, tmp_reg));
+ }
+ else
+ {
+ emit_move_insn (tmp_reg, GEN_INT (offset));
+ insn = emit_insn (gen_addsi3 (stack_pointer_rtx, ptr, tmp_reg));
+ }
+ }
+
+ if (flags & ADJUST_SP_NEED_NOTE)
+ {
+ rtx note_rtx = gen_rtx_SET (stack_pointer_rtx,
+ plus_constant (Pmode, stack_pointer_rtx,
+ offset));
+
+ RTX_FRAME_RELATED_P (insn) = 1;
+ add_reg_note (insn, REG_FRAME_RELATED_EXPR, note_rtx);
+ }
+}
+
/* minimum frame = reg save area (4 words) plus static chain (1 word)
and the total number of words must be a multiple of 128 bits. */
#define MIN_FRAME_SIZE (8 * UNITS_PER_WORD)
@@ -3174,17 +3218,28 @@ xtensa_expand_prologue (void)
int regno;
HOST_WIDE_INT offset = 0;
int callee_save_size = cfun->machine->callee_save_size;
+ df_ref ref;
+ bool stack_pointer_really_used = false;
+
+ /* Check if the function body really needs the stack pointer. */
+ for (ref = DF_REG_USE_CHAIN (A1_REG);
+ ref; ref = DF_REF_NEXT_REG (ref))
+ if (DF_REF_CLASS (ref) == DF_REF_REGULAR)
+ stack_pointer_really_used = true;
+ /* Check if callee-saved registers really need saving to the stack. */
+ for (regno = 0; regno < FIRST_PSEUDO_REGISTER; ++regno)
+ if (xtensa_call_save_reg (regno))
+ stack_pointer_really_used = true;
+
+ cfun->machine->inhibit_logues_a1_adjusts = ! (stack_pointer_really_used
+ || frame_pointer_needed
+ || crtl->calls_eh_return);
/* -128 is a limit of single addi instruction. */
if (IN_RANGE (total_size, 1, 128))
{
- insn = emit_insn (gen_addsi3 (stack_pointer_rtx, stack_pointer_rtx,
- GEN_INT (-total_size)));
- RTX_FRAME_RELATED_P (insn) = 1;
- note_rtx = gen_rtx_SET (stack_pointer_rtx,
- plus_constant (Pmode, stack_pointer_rtx,
- -total_size));
- add_reg_note (insn, REG_FRAME_RELATED_EXPR, note_rtx);
+ xtensa_emit_adjust_stack_ptr (-total_size,
+ ADJUST_SP_NEED_NOTE);
offset = total_size - UNITS_PER_WORD;
}
else if (callee_save_size)
@@ -3194,33 +3249,14 @@ xtensa_expand_prologue (void)
* move it to its final location. */
if (total_size > 1024)
{
- insn = emit_insn (gen_addsi3 (stack_pointer_rtx, stack_pointer_rtx,
- GEN_INT (-callee_save_size)));
- RTX_FRAME_RELATED_P (insn) = 1;
- note_rtx = gen_rtx_SET (stack_pointer_rtx,
- plus_constant (Pmode, stack_pointer_rtx,
- -callee_save_size));
- add_reg_note (insn, REG_FRAME_RELATED_EXPR, note_rtx);
+ xtensa_emit_adjust_stack_ptr (-callee_save_size,
+ ADJUST_SP_NEED_NOTE);
offset = callee_save_size - UNITS_PER_WORD;
}
else
{
- if (xtensa_simm8x256 (-total_size))
- insn = emit_insn (gen_addsi3 (stack_pointer_rtx,
- stack_pointer_rtx,
- GEN_INT (-total_size)));
- else
- {
- rtx tmp_reg = gen_rtx_REG (Pmode, A9_REG);
- emit_move_insn (tmp_reg, GEN_INT (total_size));
- insn = emit_insn (gen_subsi3 (stack_pointer_rtx,
- stack_pointer_rtx, tmp_reg));
- }
- RTX_FRAME_RELATED_P (insn) = 1;
- note_rtx = gen_rtx_SET (stack_pointer_rtx,
- plus_constant (Pmode, stack_pointer_rtx,
- -total_size));
- add_reg_note (insn, REG_FRAME_RELATED_EXPR, note_rtx);
+ xtensa_emit_adjust_stack_ptr (-total_size,
+ ADJUST_SP_NEED_NOTE);
offset = total_size - UNITS_PER_WORD;
}
}
@@ -3242,27 +3278,8 @@ xtensa_expand_prologue (void)
}
if (total_size > 1024
|| (!callee_save_size && total_size > 128))
- {
- if (xtensa_simm8x256 (callee_save_size - total_size))
- insn = emit_insn (gen_addsi3 (stack_pointer_rtx,
- stack_pointer_rtx,
- GEN_INT (callee_save_size -
- total_size)));
- else
- {
- rtx tmp_reg = gen_rtx_REG (Pmode, A9_REG);
- emit_move_insn (tmp_reg, GEN_INT (total_size -
- callee_save_size));
- insn = emit_insn (gen_subsi3 (stack_pointer_rtx,
- stack_pointer_rtx, tmp_reg));
- }
- RTX_FRAME_RELATED_P (insn) = 1;
- note_rtx = gen_rtx_SET (stack_pointer_rtx,
- plus_constant (Pmode, stack_pointer_rtx,
- callee_save_size -
- total_size));
- add_reg_note (insn, REG_FRAME_RELATED_EXPR, note_rtx);
- }
+ xtensa_emit_adjust_stack_ptr (callee_save_size - total_size,
+ ADJUST_SP_NEED_NOTE);
}
if (frame_pointer_needed)
@@ -3329,21 +3346,11 @@ xtensa_expand_epilogue (bool sibcall_p)
if (cfun->machine->current_frame_size > (frame_pointer_needed ? 127 : 1024))
{
- if (xtensa_simm8x256 (cfun->machine->current_frame_size -
- cfun->machine->callee_save_size))
- emit_insn (gen_addsi3 (stack_pointer_rtx, frame_pointer_needed ?
- hard_frame_pointer_rtx : stack_pointer_rtx,
- GEN_INT (cfun->machine->current_frame_size -
- cfun->machine->callee_save_size)));
- else
- {
- rtx tmp_reg = gen_rtx_REG (Pmode, A9_REG);
- emit_move_insn (tmp_reg, GEN_INT (cfun->machine->current_frame_size -
- cfun->machine->callee_save_size));
- emit_insn (gen_addsi3 (stack_pointer_rtx, frame_pointer_needed ?
- hard_frame_pointer_rtx : stack_pointer_rtx,
- tmp_reg));
- }
+ xtensa_emit_adjust_stack_ptr (cfun->machine->current_frame_size -
+ cfun->machine->callee_save_size,
+ frame_pointer_needed
+ ? ADJUST_SP_FRAME_PTR
+ : ADJUST_SP_NONE);
offset = cfun->machine->callee_save_size - UNITS_PER_WORD;
}
else
@@ -3384,24 +3391,11 @@ xtensa_expand_epilogue (bool sibcall_p)
else
offset = cfun->machine->callee_save_size;
if (offset)
- emit_insn (gen_addsi3 (stack_pointer_rtx,
- stack_pointer_rtx,
- GEN_INT (offset)));
+ xtensa_emit_adjust_stack_ptr (offset, ADJUST_SP_NONE);
}
else
- {
- if (xtensa_simm8x256 (cfun->machine->current_frame_size))
- emit_insn (gen_addsi3 (stack_pointer_rtx, stack_pointer_rtx,
- GEN_INT (cfun->machine->current_frame_size)));
- else
- {
- rtx tmp_reg = gen_rtx_REG (Pmode, A9_REG);
- emit_move_insn (tmp_reg,
- GEN_INT (cfun->machine->current_frame_size));
- emit_insn (gen_addsi3 (stack_pointer_rtx, stack_pointer_rtx,
- tmp_reg));
- }
- }
+ xtensa_emit_adjust_stack_ptr (cfun->machine->current_frame_size,
+ ADJUST_SP_NONE);
}
if (crtl->calls_eh_return)