public inbox for glibc-cvs@sourceware.org
help / color / mirror / Atom feed
From: Noah Goldstein <nwg@sourceware.org>
To: glibc-cvs@sourceware.org
Subject: [glibc] x86/fpu: Factor out shared avx2/avx512 code in svml_{s|d}_wrapper_impl.h
Date: Mon, 28 Nov 2022 04:41:32 +0000 (GMT) [thread overview]
Message-ID: <20221128044132.E1A7D3858298@sourceware.org> (raw)
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=f704192911c6c7b65a54beab3ab369fca7609a5d
commit f704192911c6c7b65a54beab3ab369fca7609a5d
Author: Noah Goldstein <goldstein.w.n@gmail.com>
Date: Fri Nov 18 16:13:32 2022 -0800
x86/fpu: Factor out shared avx2/avx512 code in svml_{s|d}_wrapper_impl.h
Code is exactly the same for the two so better to only maintain one
version.
All math and mathvec tests pass on x86.
Diff:
---
sysdeps/x86_64/fpu/svml_d_wrapper_impl.h | 172 +--------------------------
sysdeps/x86_64/fpu/svml_s_wrapper_impl.h | 172 +--------------------------
sysdeps/x86_64/fpu/svml_sd_wrapper_impl.h | 190 ++++++++++++++++++++++++++++++
3 files changed, 192 insertions(+), 342 deletions(-)
diff --git a/sysdeps/x86_64/fpu/svml_d_wrapper_impl.h b/sysdeps/x86_64/fpu/svml_d_wrapper_impl.h
index 9900f85a55..f63b49f4b8 100644
--- a/sysdeps/x86_64/fpu/svml_d_wrapper_impl.h
+++ b/sysdeps/x86_64/fpu/svml_d_wrapper_impl.h
@@ -82,174 +82,4 @@
ret
.endm
-/* AVX/AVX2 ISA version as wrapper to SSE ISA version. */
-.macro WRAPPER_IMPL_AVX callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-32, %rsp
- subq $32, %rsp
- vmovaps %ymm0, (%rsp)
- vzeroupper
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps %xmm0, (%rsp)
- vmovaps 16(%rsp), %xmm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine xmm0 (return of second call) with result of first
- call (saved on stack). Might be worth exploring logic that
- uses `vpblend` and reads in ymm1 using -16(rsp). */
- vmovaps (%rsp), %xmm1
- vinsertf128 $1, %xmm0, %ymm1, %ymm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 2 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
-.macro WRAPPER_IMPL_AVX_ff callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-32, %rsp
- subq $64, %rsp
- vmovaps %ymm0, (%rsp)
- vmovaps %ymm1, 32(%rsp)
- vzeroupper
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps 48(%rsp), %xmm1
- vmovaps %xmm0, (%rsp)
- vmovaps 16(%rsp), %xmm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine xmm0 (return of second call) with result of first
- call (saved on stack). Might be worth exploring logic that
- uses `vpblend` and reads in ymm1 using -16(rsp). */
- vmovaps (%rsp), %xmm1
- vinsertf128 $1, %xmm0, %ymm1, %ymm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 3 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
-.macro WRAPPER_IMPL_AVX_fFF callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- andq $-32, %rsp
- subq $32, %rsp
- vmovaps %ymm0, (%rsp)
- pushq %rbx
- pushq %r14
- movq %rdi, %rbx
- movq %rsi, %r14
- vzeroupper
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps 32(%rsp), %xmm0
- leaq 16(%rbx), %rdi
- leaq 16(%r14), %rsi
- call HIDDEN_JUMPTARGET(\callee)
- popq %r14
- popq %rbx
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* AVX512 ISA version as wrapper to AVX2 ISA version. */
-.macro WRAPPER_IMPL_AVX512 callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-64, %rsp
- subq $64, %rsp
- vmovups %zmm0, (%rsp)
- call HIDDEN_JUMPTARGET(\callee)
- vmovupd %ymm0, (%rsp)
- vmovupd 32(%rsp), %ymm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine ymm0 (return of second call) with result of first
- call (saved on stack). */
- vmovaps (%rsp), %ymm1
- vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 2 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
-.macro WRAPPER_IMPL_AVX512_ff callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-64, %rsp
- addq $-128, %rsp
- vmovups %zmm0, (%rsp)
- vmovups %zmm1, 64(%rsp)
- /* ymm0 and ymm1 are already set. */
- call HIDDEN_JUMPTARGET(\callee)
- vmovups 96(%rsp), %ymm1
- vmovaps %ymm0, (%rsp)
- vmovups 32(%rsp), %ymm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine ymm0 (return of second call) with result of first
- call (saved on stack). */
- vmovaps (%rsp), %ymm1
- vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 3 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
-.macro WRAPPER_IMPL_AVX512_fFF callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-64, %rsp
- subq $64, %rsp
- vmovaps %zmm0, (%rsp)
- pushq %rbx
- pushq %r14
- movq %rdi, %rbx
- movq %rsi, %r14
- /* ymm0 is already set. */
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps 48(%rsp), %ymm0
- leaq 32(%rbx), %rdi
- leaq 32(%r14), %rsi
- call HIDDEN_JUMPTARGET(\callee)
- popq %r14
- popq %rbx
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
+#include "svml_sd_wrapper_impl.h"
diff --git a/sysdeps/x86_64/fpu/svml_s_wrapper_impl.h b/sysdeps/x86_64/fpu/svml_s_wrapper_impl.h
index fd9b363045..8d8e5ef7ec 100644
--- a/sysdeps/x86_64/fpu/svml_s_wrapper_impl.h
+++ b/sysdeps/x86_64/fpu/svml_s_wrapper_impl.h
@@ -118,174 +118,4 @@
ret
.endm
-/* AVX/AVX2 ISA version as wrapper to SSE ISA version. */
-.macro WRAPPER_IMPL_AVX callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-32, %rsp
- subq $32, %rsp
- vmovaps %ymm0, (%rsp)
- vzeroupper
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps %xmm0, (%rsp)
- vmovaps 16(%rsp), %xmm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine xmm0 (return of second call) with result of first
- call (saved on stack). Might be worth exploring logic that
- uses `vpblend` and reads in ymm1 using -16(rsp). */
- vmovaps (%rsp), %xmm1
- vinsertf128 $1, %xmm0, %ymm1, %ymm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 2 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
-.macro WRAPPER_IMPL_AVX_ff callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-32, %rsp
- subq $64, %rsp
- vmovaps %ymm0, (%rsp)
- vmovaps %ymm1, 32(%rsp)
- vzeroupper
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps 48(%rsp), %xmm1
- vmovaps %xmm0, (%rsp)
- vmovaps 16(%rsp), %xmm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine xmm0 (return of second call) with result of first
- call (saved on stack). Might be worth exploring logic that
- uses `vpblend` and reads in ymm1 using -16(rsp). */
- vmovaps (%rsp), %xmm1
- vinsertf128 $1, %xmm0, %ymm1, %ymm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 3 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
-.macro WRAPPER_IMPL_AVX_fFF callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- andq $-32, %rsp
- subq $32, %rsp
- vmovaps %ymm0, (%rsp)
- pushq %rbx
- pushq %r14
- movq %rdi, %rbx
- movq %rsi, %r14
- vzeroupper
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps 32(%rsp), %xmm0
- leaq 16(%rbx), %rdi
- leaq 16(%r14), %rsi
- call HIDDEN_JUMPTARGET(\callee)
- popq %r14
- popq %rbx
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* AVX512 ISA version as wrapper to AVX2 ISA version. */
-.macro WRAPPER_IMPL_AVX512 callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-64, %rsp
- subq $64, %rsp
- vmovups %zmm0, (%rsp)
- call HIDDEN_JUMPTARGET(\callee)
- vmovupd %ymm0, (%rsp)
- vmovupd 32(%rsp), %ymm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine ymm0 (return of second call) with result of first
- call (saved on stack). */
- vmovaps (%rsp), %ymm1
- vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 2 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
-.macro WRAPPER_IMPL_AVX512_ff callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-64, %rsp
- addq $-128, %rsp
- vmovups %zmm0, (%rsp)
- vmovups %zmm1, 64(%rsp)
- /* ymm0 and ymm1 are already set. */
- call HIDDEN_JUMPTARGET(\callee)
- vmovups 96(%rsp), %ymm1
- vmovaps %ymm0, (%rsp)
- vmovups 32(%rsp), %ymm0
- call HIDDEN_JUMPTARGET(\callee)
- /* combine ymm0 (return of second call) with result of first
- call (saved on stack). */
- vmovaps (%rsp), %ymm1
- vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
-
-/* 3 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
-.macro WRAPPER_IMPL_AVX512_fFF callee
- pushq %rbp
- cfi_adjust_cfa_offset (8)
- cfi_rel_offset (%rbp, 0)
- movq %rsp, %rbp
- cfi_def_cfa_register (%rbp)
- andq $-64, %rsp
- subq $64, %rsp
- vmovaps %zmm0, (%rsp)
- pushq %rbx
- pushq %r14
- movq %rdi, %rbx
- movq %rsi, %r14
- /* ymm0 is already set. */
- call HIDDEN_JUMPTARGET(\callee)
- vmovaps 48(%rsp), %ymm0
- leaq 32(%rbx), %rdi
- leaq 32(%r14), %rsi
- call HIDDEN_JUMPTARGET(\callee)
- popq %r14
- popq %rbx
- movq %rbp, %rsp
- cfi_def_cfa_register (%rsp)
- popq %rbp
- cfi_adjust_cfa_offset (-8)
- cfi_restore (%rbp)
- ret
-.endm
+#include "svml_sd_wrapper_impl.h"
diff --git a/sysdeps/x86_64/fpu/svml_sd_wrapper_impl.h b/sysdeps/x86_64/fpu/svml_sd_wrapper_impl.h
new file mode 100644
index 0000000000..bd934ad578
--- /dev/null
+++ b/sysdeps/x86_64/fpu/svml_sd_wrapper_impl.h
@@ -0,0 +1,190 @@
+/* Common float/double wrapper implementations of vector math
+ functions.
+ Copyright (C) 2022 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+/* AVX/AVX2 ISA version as wrapper to SSE ISA version. */
+.macro WRAPPER_IMPL_AVX callee
+ pushq %rbp
+ cfi_adjust_cfa_offset (8)
+ cfi_rel_offset (%rbp, 0)
+ movq %rsp, %rbp
+ cfi_def_cfa_register (%rbp)
+ andq $-32, %rsp
+ subq $32, %rsp
+ vmovaps %ymm0, (%rsp)
+ vzeroupper
+ call HIDDEN_JUMPTARGET(\callee)
+ vmovaps %xmm0, (%rsp)
+ vmovaps 16(%rsp), %xmm0
+ call HIDDEN_JUMPTARGET(\callee)
+ /* combine xmm0 (return of second call) with result of first
+ call (saved on stack). Might be worth exploring logic that
+ uses `vpblend` and reads in ymm1 using -16(rsp). */
+ vmovaps (%rsp), %xmm1
+ vinsertf128 $1, %xmm0, %ymm1, %ymm0
+ movq %rbp, %rsp
+ cfi_def_cfa_register (%rsp)
+ popq %rbp
+ cfi_adjust_cfa_offset (-8)
+ cfi_restore (%rbp)
+ ret
+.endm
+
+/* 2 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
+.macro WRAPPER_IMPL_AVX_ff callee
+ pushq %rbp
+ cfi_adjust_cfa_offset (8)
+ cfi_rel_offset (%rbp, 0)
+ movq %rsp, %rbp
+ cfi_def_cfa_register (%rbp)
+ andq $-32, %rsp
+ subq $64, %rsp
+ vmovaps %ymm0, (%rsp)
+ vmovaps %ymm1, 32(%rsp)
+ vzeroupper
+ call HIDDEN_JUMPTARGET(\callee)
+ vmovaps 48(%rsp), %xmm1
+ vmovaps %xmm0, (%rsp)
+ vmovaps 16(%rsp), %xmm0
+ call HIDDEN_JUMPTARGET(\callee)
+ /* combine xmm0 (return of second call) with result of first
+ call (saved on stack). Might be worth exploring logic that
+ uses `vpblend` and reads in ymm1 using -16(rsp). */
+ vmovaps (%rsp), %xmm1
+ vinsertf128 $1, %xmm0, %ymm1, %ymm0
+ movq %rbp, %rsp
+ cfi_def_cfa_register (%rsp)
+ popq %rbp
+ cfi_adjust_cfa_offset (-8)
+ cfi_restore (%rbp)
+ ret
+.endm
+
+/* 3 argument AVX/AVX2 ISA version as wrapper to SSE ISA version. */
+.macro WRAPPER_IMPL_AVX_fFF callee
+ pushq %rbp
+ cfi_adjust_cfa_offset (8)
+ cfi_rel_offset (%rbp, 0)
+ movq %rsp, %rbp
+ andq $-32, %rsp
+ subq $32, %rsp
+ vmovaps %ymm0, (%rsp)
+ pushq %rbx
+ pushq %r14
+ movq %rdi, %rbx
+ movq %rsi, %r14
+ vzeroupper
+ call HIDDEN_JUMPTARGET(\callee)
+ vmovaps 32(%rsp), %xmm0
+ leaq 16(%rbx), %rdi
+ leaq 16(%r14), %rsi
+ call HIDDEN_JUMPTARGET(\callee)
+ popq %r14
+ popq %rbx
+ movq %rbp, %rsp
+ cfi_def_cfa_register (%rsp)
+ popq %rbp
+ cfi_adjust_cfa_offset (-8)
+ cfi_restore (%rbp)
+ ret
+.endm
+
+/* AVX512 ISA version as wrapper to AVX2 ISA version. */
+.macro WRAPPER_IMPL_AVX512 callee
+ pushq %rbp
+ cfi_adjust_cfa_offset (8)
+ cfi_rel_offset (%rbp, 0)
+ movq %rsp, %rbp
+ cfi_def_cfa_register (%rbp)
+ andq $-64, %rsp
+ subq $64, %rsp
+ vmovups %zmm0, (%rsp)
+ call HIDDEN_JUMPTARGET(\callee)
+ vmovupd %ymm0, (%rsp)
+ vmovupd 32(%rsp), %ymm0
+ call HIDDEN_JUMPTARGET(\callee)
+ /* combine ymm0 (return of second call) with result of first
+ call (saved on stack). */
+ vmovaps (%rsp), %ymm1
+ vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
+ movq %rbp, %rsp
+ cfi_def_cfa_register (%rsp)
+ popq %rbp
+ cfi_adjust_cfa_offset (-8)
+ cfi_restore (%rbp)
+ ret
+.endm
+
+/* 2 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
+.macro WRAPPER_IMPL_AVX512_ff callee
+ pushq %rbp
+ cfi_adjust_cfa_offset (8)
+ cfi_rel_offset (%rbp, 0)
+ movq %rsp, %rbp
+ cfi_def_cfa_register (%rbp)
+ andq $-64, %rsp
+ addq $-128, %rsp
+ vmovups %zmm0, (%rsp)
+ vmovups %zmm1, 64(%rsp)
+ /* ymm0 and ymm1 are already set. */
+ call HIDDEN_JUMPTARGET(\callee)
+ vmovups 96(%rsp), %ymm1
+ vmovaps %ymm0, (%rsp)
+ vmovups 32(%rsp), %ymm0
+ call HIDDEN_JUMPTARGET(\callee)
+ /* combine ymm0 (return of second call) with result of first
+ call (saved on stack). */
+ vmovaps (%rsp), %ymm1
+ vinserti64x4 $0x1, %ymm0, %zmm1, %zmm0
+ movq %rbp, %rsp
+ cfi_def_cfa_register (%rsp)
+ popq %rbp
+ cfi_adjust_cfa_offset (-8)
+ cfi_restore (%rbp)
+ ret
+.endm
+
+/* 3 argument AVX512 ISA version as wrapper to AVX2 ISA version. */
+.macro WRAPPER_IMPL_AVX512_fFF callee
+ pushq %rbp
+ cfi_adjust_cfa_offset (8)
+ cfi_rel_offset (%rbp, 0)
+ movq %rsp, %rbp
+ cfi_def_cfa_register (%rbp)
+ andq $-64, %rsp
+ subq $64, %rsp
+ vmovaps %zmm0, (%rsp)
+ pushq %rbx
+ pushq %r14
+ movq %rdi, %rbx
+ movq %rsi, %r14
+ /* ymm0 is already set. */
+ call HIDDEN_JUMPTARGET(\callee)
+ vmovaps 48(%rsp), %ymm0
+ leaq 32(%rbx), %rdi
+ leaq 32(%r14), %rsi
+ call HIDDEN_JUMPTARGET(\callee)
+ popq %r14
+ popq %rbx
+ movq %rbp, %rsp
+ cfi_def_cfa_register (%rsp)
+ popq %rbp
+ cfi_adjust_cfa_offset (-8)
+ cfi_restore (%rbp)
+ ret
+.endm
reply other threads:[~2022-11-28 4:41 UTC|newest]
Thread overview: [no followups] expand[flat|nested] mbox.gz Atom feed
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20221128044132.E1A7D3858298@sourceware.org \
--to=nwg@sourceware.org \
--cc=glibc-cvs@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).