gcc/11: apply upstream fix 103910 so openjdk builds

Our openjdk derivations fail to build with `-march=opteron` (via
either `NIX_CFLAGS_COMPILE` or `hostPlatform.gcc.arch`).  This was
fixed upstream in gcc12 but not carried back to gcc11:

  https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=d243f4009d8071b734df16cd70f4c5d09a373769

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103910

Since gcc11 is still nixpkgs' primary compiler, this PR cherry-picks
the fix out of gcc12.

Unfortunately since `-march=` can be added after the `gcc11`
derivation is built we can't apply this patch conditionally (which
we don't usually do anyways).
This commit is contained in:
Adam Joseph 2023-01-23 20:35:10 -08:00
parent d381e51fb5
commit b83f032ffc
2 changed files with 45 additions and 1 deletions

View File

@ -82,7 +82,10 @@ let majorVersion = "11";
++ optional (stdenv.isDarwin && targetPlatform.isAvr) ./avr-gcc-11.3-darwin.patch
# Obtain latest patch with ../update-mcfgthread-patches.sh
++ optional (!crossStageStatic && targetPlatform.isMinGW && threadsCross.model == "mcf") ./Added-mcf-thread-model-support-from-mcfgthread.patch;
++ optional (!crossStageStatic && targetPlatform.isMinGW && threadsCross.model == "mcf") ./Added-mcf-thread-model-support-from-mcfgthread.patch
# openjdk build fails without this on -march=opteron; is upstream in gcc12
++ [ ./gcc-issue-103910.patch ];
/* Cross-gcc settings (build == host != target) */
crossMingw = targetPlatform != hostPlatform && targetPlatform.libc == "msvcrt";

View File

@ -0,0 +1,41 @@
From d243f4009d8071b734df16cd70f4c5d09a373769 Mon Sep 17 00:00:00 2001
From: Andrew Pinski <apinski@marvell.com>
Date: Wed, 5 Jan 2022 22:00:07 +0000
Subject: [PATCH] Fix target/103910: missing GTY on x86_mfence causing PCH
usage to ICE
With -O3 -march=opteron, a mfence builtin is added after the loop
to say the nontemporal stores are no longer needed. This all good
without precompiled headers as the function decl that is referneced
by x86_mfence is referenced in another variable but with precompiled
headers, x86_mfence is all messed up and the decl was GC'ed away.
This fixes the problem by marking x86_mfence as GTY to save/restore
during precompiled headers just like most other variables in
the header file.
Committed as obvious after a bootstrap/test on x86_64-linux-gnu.
gcc/ChangeLog:
PR target/103910
* config/i386/i386.h (x86_mfence): Mark with GTY.
---
gcc/config/i386/i386.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h
index f027608eefa..3ac0f698ae2 100644
--- a/gcc/config/i386/i386.h
+++ b/gcc/config/i386/i386.h
@@ -486,7 +486,7 @@ extern unsigned char ix86_prefetch_sse;
/* Fence to use after loop using storent. */
-extern tree x86_mfence;
+extern GTY(()) tree x86_mfence;
#define FENCE_FOLLOWING_MOVNT x86_mfence
/* Once GDB has been enhanced to deal with functions without frame
--
2.31.1