diff options
| author | A. Wilcox <AWilcox@Wilcox-Tech.com> | 2024-05-23 23:40:10 -0500 |
|---|---|---|
| committer | A. Wilcox <AWilcox@Wilcox-Tech.com> | 2024-08-07 09:57:51 -0500 |
| commit | 93cf7fc8b05fe3003b6f16b766622544cdf3830b (patch) | |
| tree | 594b1c2e06144b9e0b0dae1310399b5935ddc5ff /system/gcc/341-gccgo-libucontext-stack.patch | |
| parent | d15fa95546e4f4200700750986fb5e249d9bfbe1 (diff) | |
| download | packages-93cf7fc8b05fe3003b6f16b766622544cdf3830b.tar.gz packages-93cf7fc8b05fe3003b6f16b766622544cdf3830b.tar.bz2 packages-93cf7fc8b05fe3003b6f16b766622544cdf3830b.tar.xz packages-93cf7fc8b05fe3003b6f16b766622544cdf3830b.zip | |
system/gcc: Update to 13.3.0
0012-static-pie and 201-ada were forward-ported by my own paws.
Three digit 0xx underscore patches are from Gentoo. This is also the
source of the insn-split and match-split patches, which significantly
increase performance on build. They are also the source of the new
configure option --with-matchpd-partitions. On systems with very many
cores/threads, a number higher than 32 would probably perform even
better, but 32 should give decent perf on 8+ threads, and runs quite
nicely on the 64-thread Talos II I ran the testbuild on.
202 is from Void. The 300s are originally written by me.
The Ada patch now includes an improvement to use posix_openpt instead
of getpt, which fixes link errors later on.
We also now use STAGE1_CFLAGS and BOOT_CFLAGS to improve build times.
Improvement was clocked from 40m to 32m (-8m, or 20%) on gwyn.
The patch for Go name mangling is needed to build packages that use
reflect2, which includes user/gitlab-runner.
Diffstat (limited to 'system/gcc/341-gccgo-libucontext-stack.patch')
| -rw-r--r-- | system/gcc/341-gccgo-libucontext-stack.patch | 38 |
1 files changed, 31 insertions, 7 deletions
diff --git a/system/gcc/341-gccgo-libucontext-stack.patch b/system/gcc/341-gccgo-libucontext-stack.patch index 2befc5d32..cc88bda5a 100644 --- a/system/gcc/341-gccgo-libucontext-stack.patch +++ b/system/gcc/341-gccgo-libucontext-stack.patch @@ -4,15 +4,39 @@ This causes the Go frontend to sometimes go below the stack. On 32-bit Arm systems, this resulted in #832. It is likely, though not known, that this could have caused other issues we've seen in gccgo as well. ---- gcc-8.5.0/libgo/runtime/go-callers.c.old 2021-05-14 03:42:11.000000000 -0500 -+++ gcc-8.5.0/libgo/runtime/go-callers.c 2022-12-25 01:46:15.834110528 -0600 -@@ -117,7 +117,8 @@ - Solaris (http://gcc.gnu.org/PR52583 comment #21). */ +Additionally, since we strip libgo.so, filename is always NULL. This caused +the unwinder to continue through runtime_mstart which caused the same issue. +--- gcc-8.5.0/libgo/runtime/go-callers.c.old 2023-01-01 17:56:27.302982459 -0600 ++++ gcc-8.5.0/libgo/runtime/go-callers.c 2023-01-01 17:58:00.612903483 -0600 +@@ -165,26 +165,11 @@ if (function != NULL) { - if (__builtin_strcmp (function, "makecontext") == 0) + if (__builtin_strcmp (function, "makecontext") == 0 -+ || __builtin_strcmp (function, "libucontext_makecontext") == 0) ++ || __builtin_strcmp (function, "libucontext_makecontext") == 0 ++ || __builtin_strcmp (function, "runtime_mstart") == 0 ++ || __builtin_strcmp (function, "runtime.kickoff") == 0 ++ || __builtin_strcmp (function, "runtime.main") == 0) return 1; - if (filename != NULL) - { +- if (filename != NULL) +- { +- const char *p; +- +- p = strrchr (filename, '/'); +- if (p == NULL) +- p = filename; +- if (__builtin_strcmp (p, "/proc.c") == 0) +- { +- if (__builtin_strcmp (function, "runtime_mstart") == 0) +- return 1; +- } +- else if (__builtin_strcmp (p, "/proc.go") == 0) +- { +- if (__builtin_strcmp (function, "runtime.kickoff") == 0 +- || __builtin_strcmp (function, "runtime.main") == 0) +- return 1; +- } +- } + } + + return arg->index >= arg->max; |