Re: [x86] Fix PR target/99264

[Eric Botcazou]

> LGTM.

Thanks.  However, I overlooked an issue with pathologically large frames 
(larger than SEH_MAX_FRAME_SIZE, i.e. 2GB and for which we cannot emit CFI 
directives) that is visible in the gnat.dg testsuite under the form of an ICE.

Tested on x86_64-w64-mingw32, applied on mainline/10/9 branches as obvious.


2021-03-03 Eric Botcazou  <ebotca...@adacore.com>

        PR target/99234
        * config/i386/i386.c (ix86_compute_frame_layout): For a SEH target,
        point back the hard frame pointer to its default location when the
        frame is larger than SEH_MAX_FRAME_SIZE.

-- 
Eric Botcazou
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index e85b06b6824..6672da597ba 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -6660,7 +6660,8 @@ ix86_compute_frame_layout (void)
       frame->hard_frame_pointer_offset = frame->sse_reg_save_offset;
 
       /* If we can leave the frame pointer where it is, do so.  Also, return
-	 the establisher frame for __builtin_frame_address (0).  */
+	 the establisher frame for __builtin_frame_address (0) or else if the
+	 frame overflows the SEH maximum frame size.  */
       const HOST_WIDE_INT diff
 	= frame->stack_pointer_offset - frame->hard_frame_pointer_offset;
       if (diff <= 255)
@@ -6678,6 +6679,8 @@ ix86_compute_frame_layout (void)
 	     frame that is addressable with 8-bit offsets.  */
 	  frame->hard_frame_pointer_offset = frame->stack_pointer_offset - 128;
 	}
+      else
+	frame->hard_frame_pointer_offset = frame->hfp_save_offset;
     }
 }