[Re-send, since I never saw the original arrive on the list.]
After the first version passed Kai's testing, we chatted on IRC about
various other bits that might be relevant to the winx64 abi issues.
The conditional call to ix86_eax_live_at_start_p seemed to be harmless
either way, but confusing as a conditional. It seems least confusing
to make the call unconditional.
Finally, adjust the comment for ACCUMULATE_OUTGOING_ARGS to match reality.
r~
PR target/59927
* calls.c (expand_call): Don't double-push for reg_parm_stack_space.
* config/i386/i386.c (init_cumulative_args): Remove sorry for 64-bit
ms-abi vs -mno-accumulate-outgoing-args.
(ix86_expand_prologue): Unconditionally call ix86_eax_live_at_start_p.
* config/i386/i386.h (ACCUMULATE_OUTGOING_ARGS): Fix comment with
respect to ms-abi.
diff --git a/gcc/calls.c b/gcc/calls.c
index d574a95..f392319 100644
--- a/gcc/calls.c
+++ b/gcc/calls.c
@@ -2955,6 +2955,7 @@ expand_call (tree exp, rtx target, int ignore)
/* If we push args individually in reverse order, perform stack alignment
before the first push (the last arg). */
if (PUSH_ARGS_REVERSED && argblock == 0
+ && adjusted_args_size.constant > reg_parm_stack_space
&& adjusted_args_size.constant != unadjusted_args_size)
{
/* When the stack adjustment is pending, we get better code
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 1b7bb3e..0a15e44 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -6110,10 +6110,6 @@ init_cumulative_args (CUMULATIVE_ARGS *cum, /* Argument
info to initialize */
cum->caller = caller;
/* Set up the number of registers to use for passing arguments. */
-
- if (TARGET_64BIT && cum->call_abi == MS_ABI && !ACCUMULATE_OUTGOING_ARGS)
- sorry ("ms_abi attribute requires -maccumulate-outgoing-args "
- "or subtarget optimization implying it");
cum->nregs = ix86_regparm;
if (TARGET_64BIT)
{
@@ -11032,15 +11028,14 @@ ix86_expand_prologue (void)
if (TARGET_64BIT)
r10_live = (DECL_STATIC_CHAIN (current_function_decl) != 0);
- if (!TARGET_64BIT_MS_ABI)
- eax_live = ix86_eax_live_at_start_p ();
- /* Note that SEH directives need to continue tracking the stack
- pointer even after the frame pointer has been set up. */
+ eax_live = ix86_eax_live_at_start_p ();
if (eax_live)
{
insn = emit_insn (gen_push (eax));
allocate -= UNITS_PER_WORD;
+ /* Note that SEH directives need to continue tracking the stack
+ pointer even after the frame pointer has been set up. */
if (sp_is_cfa_reg || TARGET_SEH)
{
if (sp_is_cfa_reg)
diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h
index 0e757c9c..b605ae2 100644
--- a/gcc/config/i386/i386.h
+++ b/gcc/config/i386/i386.h
@@ -1537,12 +1537,12 @@ enum reg_class
mode the difference is less drastic but visible.
FIXME: Unlike earlier implementations, the size of unwind info seems to
- actually grouw with accumulation. Is that because accumulated args
+ actually grow with accumulation. Is that because accumulated args
unwind info became unnecesarily bloated?
-
- 64-bit MS ABI seem to require 16 byte alignment everywhere except for
- function prologue and epilogue. This is not possible without
- ACCUMULATE_OUTGOING_ARGS.
+
+ With the 64-bit MS ABI, we can generate correct code with or without
+ accumulated args, but because of OUTGOING_REG_PARM_STACK_SPACE the code
+ generated without accumulated args is terrible.
If stack probes are required, the space used for large function
arguments on the stack must also be probed, so enable
