[PING^5] nvptx: Allow '--with-arch' to override the default '-misa' (was: nvptx multilib setup)

Message ID 87iln518mx.fsf@dem-tschwing-1.ger.mentorg.com
State New, archived
Headers
Series [PING^5] nvptx: Allow '--with-arch' to override the default '-misa' (was: nvptx multilib setup) |

Commit Message

Thomas Schwinge Aug. 6, 2022, 7:20 p.m. UTC
  Hi Tom!

Ping.


Grüße
 Thomas


On 2022-07-27T17:48:58+0200, I wrote:
> Hi Tom!
>
> Ping.
>
>
> Grüße
>  Thomas
>
>
> On 2022-07-20T14:46:03+0200, I wrote:
>> Hi Tom!
>>
>> Ping.
>>
>>
>> Grüße
>>  Thomas
>>
>>
>> On 2022-07-13T10:42:44+0200, I wrote:
>>> Hi Tom!
>>>
>>> Ping.
>>>
>>>
>>> Grüße
>>>  Thomas
>>>
>>>
>>> On 2022-07-05T16:59:23+0200, I wrote:
>>>> Hi Tom!
>>>>
>>>> Ping.
>>>>
>>>>
>>>> Grüße
>>>>  Thomas
>>>>
>>>>
>>>> On 2022-06-15T23:18:10+0200, I wrote:
>>>>> Hi Tom!
>>>>>
>>>>> On 2022-05-13T16:20:14+0200, I wrote:
>>>>>> On 2022-02-04T13:09:29+0100, Tom de Vries via Gcc <gcc@gcc.gnu.org> wrote:
>>>>>>> On 2/4/22 08:21, Thomas Schwinge wrote:
>>>>>>>> On 2022-02-03T13:35:55+0000, "vries at gcc dot gnu.org via Gcc-bugs" <gcc-bugs@gcc.gnu.org> wrote:
>>>>>>>>> I've tested this using (recommended) driver 470.94 on boards:
>>>>>>
>>>>>>>>> while iterating over dimensions { -mptx=3.1 , -mptx=6.3 } x { GOMP_NVPTX_JIT=-O0, <default> }.
>>>>>>>>
>>>>>>>> Do you use separate (nvptx-none offload target only?) builds for
>>>>>>>> different '-mptx' variants (likewise: '-misa'), or have you hacked up the
>>>>>>>> multilib configuration?
>>>>>>>
>>>>>>> Neither, I'm using --target_board=unix/foffload= for that.
>>>>>>
>>>>>> ACK, I see.  So these flags then only affect GCC/nvptx code generation
>>>>>> for the actual user code (here: GCC libgomp test cases), but for the
>>>>>> GCC/nvptx target libraries (such as: libc, libm, libgfortran, libgomp --
>>>>>> the latter especially relevant for OpenMP), it uses PTX code from one of
>>>>>> the two "pre-compiled" GCC/nvptx multilibs: default or '-mptx=3.1'.
>>>>>>
>>>>>> Meaning, one can't just use such a flag for "completely building code"
>>>>>> for a specific configuration.  Random example,
>>>>>> '-foffload-options=nvptx-none=-march=sm_75': as GCC/nvptx target
>>>>>> libraries aren't being built for '-march=sm_75' multilib,
>>>>>> '-foffload-options=nvptx-none=-march=sm_75' uses the default multilib,
>>>>>> which isn't '-march=sm_75'.
>>>>>>
>>>>>>
>>>>>>>   ('gcc/config/nvptx/t-nvptx:MULTILIB_OPTIONS'
>>>>>>>> etc., I suppose?)  Should we add a few representative configurations to
>>>>>>>> be built by default?  And/or, should we have a way to 'configure' per
>>>>>>>> user needs (I suppose: '--with-multilib-list=[...]', as supported for a
>>>>>>>> few other targets?)?  (I see there's also a new
>>>>>>>> '--with-multilib-generator=[...]', haven't looked in detail.)  No matter
>>>>>>>> which way: again, combinatorial explosion is a problem, of course...
>>>>>>>
>>>>>>> As far as I know, the gcc build doesn't finish when switching default to
>>>>>>> higher than sm_35, so there's little point to go to a multilib setup at
>>>>>>> this point.  But once we fix that, we could reconsider, otherwise,
>>>>>>> things are likely to regress again.
>>>>>>
>>>>>> As far as I remember, several issues have been fixed.  Still waiting for
>>>>>> Roger's "middle-end: Support ABIs that pass FP values as wider integers"
>>>>>> or something similar, but that PR104489 issue is being worked around by
>>>>>> "Limit HFmode support to mexperimental", if I got that right.
>>>>>>
>>>>>> Now I'm not suggesting we should now enable all or any random GCC/nvptx
>>>>>> multilibs, to get all these variants of GCC/nvptx target libraries built;
>>>>>> especially also given that GCC/nvptx code generation currently doesn't
>>>>>> make too much use of the new capabilities.
>>>>>>
>>>>>> However, we do have a specific request that a customer would like to be
>>>>>> able to change at GCC 'configure' time the GCC/nvptx default multilib
>>>>>> (including that being used for building corresponding GCC/nvptx target
>>>>>> libraries).
>>>>>>
>>>>>> Per 'gcc/doc/install.texi', I do see that some GCC targets allow for
>>>>>> GCC 'configure'-time '--with-multilib-list=[...]', or
>>>>>> '--with-multilib-generator=[...]', and I suppose we could be doing
>>>>>> something similar?  But before starting implementing, I'd like your
>>>>>> input, as you'll be the one to approve in the end.  And/or, maybe you've
>>>>>> already made up your own ideas about that?
>>>>>
>>>>> So, instead of "random GCC/nvptx multilib configuration" (last
>>>>> paragraph), I've come up with a way to implement our customer's request
>>>>> (second last paragraph): 'configure' GCC/nvptx '--with-arch=sm_70'.
>>>>>
>>>>> I think I've implemented this in a way so that "random GCC/nvptx multilib
>>>>> configuration" may eventually be implemented on top of that.  For easy
>>>>> review/testing I've split my changes into three commits, see attached
>>>>> "nvptx: Make default '-misa=sm_30' explicit",
>>>>> "nvptx: Introduce dummy multilib option for default '-misa=sm_30'",
>>>>> "nvptx: Allow '--with-arch' to override the default '-misa'".
>>>>>
>>>>> To the best of my knowledge, the first two patches do not change any
>>>>> user-visible behavior (I generally 'diff'ed target libraries, and
>>>>> compared a good number of 'gcc -print-multi-directory [flags]'), and
>>>>> likewise with the third patch, given implicit (default) or explicit
>>>>> '--with-arch=sm_30', and that with '--with-arch=sm_70', for example, the
>>>>> '-misa=sm_70' multilib variants are used for implicit (default) or
>>>>> explicit '-misa=sm_70' or higher, and the '-misa=sm_30' multilib variants
>>>>> are used for explicit lower '-misa'.
>>>>>
>>>>> What do you think, OK to push to master branch?
>>>>>
>>>>>
>>>>> Grüße
>>>>>  Thomas


-----------------
Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955
  

Comments

Tom de Vries Sept. 18, 2022, 8:53 a.m. UTC | #1
On 8/6/22 21:20, Thomas Schwinge wrote:
> Hi Tom!
> 

Hi Thomas,

thanks for doing this.

Series approved.

As I mentioned, I'm not completely happy with the multilib name, but I 
don't think it makes sense to post-pone approval for this.

Thanks,
- Tom

> Ping.
> 
> 
> Grüße
>   Thomas
> 
> 
> On 2022-07-27T17:48:58+0200, I wrote:
>> Hi Tom!
>>
>> Ping.
>>
>>
>> Grüße
>>   Thomas
>>
>>
>> On 2022-07-20T14:46:03+0200, I wrote:
>>> Hi Tom!
>>>
>>> Ping.
>>>
>>>
>>> Grüße
>>>   Thomas
>>>
>>>
>>> On 2022-07-13T10:42:44+0200, I wrote:
>>>> Hi Tom!
>>>>
>>>> Ping.
>>>>
>>>>
>>>> Grüße
>>>>   Thomas
>>>>
>>>>
>>>> On 2022-07-05T16:59:23+0200, I wrote:
>>>>> Hi Tom!
>>>>>
>>>>> Ping.
>>>>>
>>>>>
>>>>> Grüße
>>>>>   Thomas
>>>>>
>>>>>
>>>>> On 2022-06-15T23:18:10+0200, I wrote:
>>>>>> Hi Tom!
>>>>>>
>>>>>> On 2022-05-13T16:20:14+0200, I wrote:
>>>>>>> On 2022-02-04T13:09:29+0100, Tom de Vries via Gcc <gcc@gcc.gnu.org> wrote:
>>>>>>>> On 2/4/22 08:21, Thomas Schwinge wrote:
>>>>>>>>> On 2022-02-03T13:35:55+0000, "vries at gcc dot gnu.org via Gcc-bugs" <gcc-bugs@gcc.gnu.org> wrote:
>>>>>>>>>> I've tested this using (recommended) driver 470.94 on boards:
>>>>>>>
>>>>>>>>>> while iterating over dimensions { -mptx=3.1 , -mptx=6.3 } x { GOMP_NVPTX_JIT=-O0, <default> }.
>>>>>>>>>
>>>>>>>>> Do you use separate (nvptx-none offload target only?) builds for
>>>>>>>>> different '-mptx' variants (likewise: '-misa'), or have you hacked up the
>>>>>>>>> multilib configuration?
>>>>>>>>
>>>>>>>> Neither, I'm using --target_board=unix/foffload= for that.
>>>>>>>
>>>>>>> ACK, I see.  So these flags then only affect GCC/nvptx code generation
>>>>>>> for the actual user code (here: GCC libgomp test cases), but for the
>>>>>>> GCC/nvptx target libraries (such as: libc, libm, libgfortran, libgomp --
>>>>>>> the latter especially relevant for OpenMP), it uses PTX code from one of
>>>>>>> the two "pre-compiled" GCC/nvptx multilibs: default or '-mptx=3.1'.
>>>>>>>
>>>>>>> Meaning, one can't just use such a flag for "completely building code"
>>>>>>> for a specific configuration.  Random example,
>>>>>>> '-foffload-options=nvptx-none=-march=sm_75': as GCC/nvptx target
>>>>>>> libraries aren't being built for '-march=sm_75' multilib,
>>>>>>> '-foffload-options=nvptx-none=-march=sm_75' uses the default multilib,
>>>>>>> which isn't '-march=sm_75'.
>>>>>>>
>>>>>>>
>>>>>>>>    ('gcc/config/nvptx/t-nvptx:MULTILIB_OPTIONS'
>>>>>>>>> etc., I suppose?)  Should we add a few representative configurations to
>>>>>>>>> be built by default?  And/or, should we have a way to 'configure' per
>>>>>>>>> user needs (I suppose: '--with-multilib-list=[...]', as supported for a
>>>>>>>>> few other targets?)?  (I see there's also a new
>>>>>>>>> '--with-multilib-generator=[...]', haven't looked in detail.)  No matter
>>>>>>>>> which way: again, combinatorial explosion is a problem, of course...
>>>>>>>>
>>>>>>>> As far as I know, the gcc build doesn't finish when switching default to
>>>>>>>> higher than sm_35, so there's little point to go to a multilib setup at
>>>>>>>> this point.  But once we fix that, we could reconsider, otherwise,
>>>>>>>> things are likely to regress again.
>>>>>>>
>>>>>>> As far as I remember, several issues have been fixed.  Still waiting for
>>>>>>> Roger's "middle-end: Support ABIs that pass FP values as wider integers"
>>>>>>> or something similar, but that PR104489 issue is being worked around by
>>>>>>> "Limit HFmode support to mexperimental", if I got that right.
>>>>>>>
>>>>>>> Now I'm not suggesting we should now enable all or any random GCC/nvptx
>>>>>>> multilibs, to get all these variants of GCC/nvptx target libraries built;
>>>>>>> especially also given that GCC/nvptx code generation currently doesn't
>>>>>>> make too much use of the new capabilities.
>>>>>>>
>>>>>>> However, we do have a specific request that a customer would like to be
>>>>>>> able to change at GCC 'configure' time the GCC/nvptx default multilib
>>>>>>> (including that being used for building corresponding GCC/nvptx target
>>>>>>> libraries).
>>>>>>>
>>>>>>> Per 'gcc/doc/install.texi', I do see that some GCC targets allow for
>>>>>>> GCC 'configure'-time '--with-multilib-list=[...]', or
>>>>>>> '--with-multilib-generator=[...]', and I suppose we could be doing
>>>>>>> something similar?  But before starting implementing, I'd like your
>>>>>>> input, as you'll be the one to approve in the end.  And/or, maybe you've
>>>>>>> already made up your own ideas about that?
>>>>>>
>>>>>> So, instead of "random GCC/nvptx multilib configuration" (last
>>>>>> paragraph), I've come up with a way to implement our customer's request
>>>>>> (second last paragraph): 'configure' GCC/nvptx '--with-arch=sm_70'.
>>>>>>
>>>>>> I think I've implemented this in a way so that "random GCC/nvptx multilib
>>>>>> configuration" may eventually be implemented on top of that.  For easy
>>>>>> review/testing I've split my changes into three commits, see attached
>>>>>> "nvptx: Make default '-misa=sm_30' explicit",
>>>>>> "nvptx: Introduce dummy multilib option for default '-misa=sm_30'",
>>>>>> "nvptx: Allow '--with-arch' to override the default '-misa'".
>>>>>>
>>>>>> To the best of my knowledge, the first two patches do not change any
>>>>>> user-visible behavior (I generally 'diff'ed target libraries, and
>>>>>> compared a good number of 'gcc -print-multi-directory [flags]'), and
>>>>>> likewise with the third patch, given implicit (default) or explicit
>>>>>> '--with-arch=sm_30', and that with '--with-arch=sm_70', for example, the
>>>>>> '-misa=sm_70' multilib variants are used for implicit (default) or
>>>>>> explicit '-misa=sm_70' or higher, and the '-misa=sm_30' multilib variants
>>>>>> are used for explicit lower '-misa'.
>>>>>>
>>>>>> What do you think, OK to push to master branch?
>>>>>>
>>>>>>
>>>>>> Grüße
>>>>>>   Thomas
> 
> 
> -----------------
> Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955
  

Patch

From 8343b7d0ed96cf5d45c12b36ea34a156229b8745 Mon Sep 17 00:00:00 2001
From: Thomas Schwinge <thomas@codesourcery.com>
Date: Sat, 11 Jun 2022 19:37:10 +0200
Subject: [PATCH 3/3] nvptx: Allow '--with-arch' to override the default
 '-misa'

---
 gcc/config.gcc                           |  5 ++
 gcc/config/nvptx/gen-multilib-matches.sh | 60 ++++++++++++++++++++++++
 gcc/config/nvptx/t-nvptx                 | 21 +++++++--
 gcc/doc/install.texi                     |  9 ++++
 gcc/doc/invoke.texi                      |  4 +-
 5 files changed, 93 insertions(+), 6 deletions(-)
 create mode 100755 gcc/config/nvptx/gen-multilib-matches.sh

diff --git a/gcc/config.gcc b/gcc/config.gcc
index d492b7769fe..8f51d333668 100644
--- a/gcc/config.gcc
+++ b/gcc/config.gcc
@@ -5442,10 +5442,15 @@  case "${target}" in
 	nvptx-*)
 		supported_defaults=arch
 		TM_MULTILIB_CONFIG=$with_arch
+		#TODO 'sm_[...]' list per 'nvptx-sm.def'.
 		case $with_arch in
 			sm_30 )
 				# OK; default.
 				;;
+			sm_35 | sm_53 | sm_70 | sm_75 | sm_80 )
+				# OK, but we'd like 'sm_30', too.
+				TM_MULTILIB_CONFIG="$TM_MULTILIB_CONFIG sm_30"
+				;;
 			* )
 				echo "Unknown arch used in --with-arch=$with_arch" 1>&2
 				exit 1
diff --git a/gcc/config/nvptx/gen-multilib-matches.sh b/gcc/config/nvptx/gen-multilib-matches.sh
new file mode 100755
index 00000000000..9a5878e855b
--- /dev/null
+++ b/gcc/config/nvptx/gen-multilib-matches.sh
@@ -0,0 +1,60 @@ 
+#!/bin/sh
+
+# Print nvptx 'MULTILIB_MATCHES'
+
+# Copyright (C) 2022 Free Software Foundation, Inc.
+#
+# This file is part of GCC.
+#
+# GCC is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3, or (at your option)
+# any later version.
+#
+# GCC 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 General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with GCC; see the file COPYING3.  If not see
+# <http://www.gnu.org/licenses/>.
+
+set -e
+
+nvptx_sm_def="$1/nvptx-sm.def"
+multilib_options_isa_default=$2
+multilib_options_isa_list=$3
+
+sms=$(grep ^NVPTX_SM $nvptx_sm_def | sed 's/.*(//;s/,.*//')
+
+# Every variant in 'sms' has to either be remapped to the default variant
+# ('.', which is always built), or does get built as non-default variant
+# ('misa=sm_SM'; thus not remapped), or has to be remapped to the "next lower"
+# variant that does get built.
+
+# The "lowest" variant has to be built.
+sm_next_lower=INVALID
+
+for sm in $sms; do
+    if [ x"sm_$sm" = x"$multilib_options_isa_default" ]; then
+	sm_map=.
+    elif expr " $multilib_options_isa_list " : ".* sm_$sm " > /dev/null; then
+	sm_map=
+    else
+	sm_map=$sm_next_lower
+    fi
+
+    if [ x"$sm_map" = x ]; then
+	sm_next_lower=$sm
+    else
+	# Output format as required for 'MULTILIB_MATCHES'.
+	if [ x"$sm_map" = x. ]; then
+	    echo ".=misa?sm_$sm"
+	else
+	    echo "misa?sm_$sm_map=misa?sm_$sm"
+	fi
+
+	sm_next_lower=$sm_map
+    fi
+done
diff --git a/gcc/config/nvptx/t-nvptx b/gcc/config/nvptx/t-nvptx
index c797d57690f..9c5cbda0070 100644
--- a/gcc/config/nvptx/t-nvptx
+++ b/gcc/config/nvptx/t-nvptx
@@ -42,11 +42,22 @@  MULTILIB_OPTIONS += mgomp
 
 multilib_options_isa_list := $(TM_MULTILIB_CONFIG)
 multilib_options_isa_default := $(word 1,$(multilib_options_isa_list))
-# Add the default '-misa' as a multilib option:
-MULTILIB_OPTIONS += misa=$(multilib_options_isa_default)
-# ..., but don't handle it specially (remap to default):
-MULTILIB_MATCHES += .=misa?$(multilib_options_isa_default)
-# ..., and don't actually build it:
+multilib_options_misa_list := $(addprefix misa=,$(multilib_options_isa_list))
+# Add the requested '-misa' variants as a multilib option ('misa=VAR1/misa=VAR2/misa=VAR3' etc.):
+empty :=
+space := $(empty) $(empty)
+MULTILIB_OPTIONS += $(subst $(space),/,$(multilib_options_misa_list))
+# ..., and remap '-misa' variants as appropriate:
+multilib_matches := $(shell $(srcdir)/config/nvptx/gen-multilib-matches.sh $(srcdir)/config/nvptx $(multilib_options_isa_default) "$(multilib_options_isa_list)")
+MULTILIB_MATCHES += $(multilib_matches)
+# ..., and don't actually build what's the default '-misa':
 MULTILIB_EXCEPTIONS += *misa=$(multilib_options_isa_default)*
 
 MULTILIB_OPTIONS += mptx=3.1
+# Filter out invalid '-misa'/'-mptx=3.1' combinations; per 'nvptx-sm.def',
+# 'nvptx.opt:ptx_version', 'nvptx.cc:first_ptx_version_supporting_sm'
+# (that is, '-mptx=3.1' only for sm_30, sm_35 variants):
+MULTILIB_EXCEPTIONS += $(foreach misa,$(filter-out %=sm_30 %=sm_35,$(multilib_options_misa_list)),*$(misa)/mptx=3.1)
+# ..., and special care has to be taken if '-mptx=3.1' is invalid for the
+# default variant:
+MULTILIB_EXCEPTIONS += $(if $(filter-out sm_30 sm_35,$(multilib_options_isa_default)),mgomp/mptx=3.1 mptx=3.1)
diff --git a/gcc/doc/install.texi b/gcc/doc/install.texi
index 45ef39da04e..7b4073bdafc 100644
--- a/gcc/doc/install.texi
+++ b/gcc/doc/install.texi
@@ -4617,6 +4617,15 @@  the GCC sources.
 Use the @option{--disable-sjlj-exceptions} and
 @option{--enable-newlib-io-long-long} options when configuring.
 
+The @option{--with-arch} option may be specified to override the
+default value for the @option{-march} option, and to also build
+corresponding target libraries.
+The default is @option{--with-arch=sm_30}.
+
+For example, if @option{--with-arch=sm_70} is specified,
+@option{-march=sm_30} and @option{-march=sm_70} target libraries are
+built, and code generation defaults to @option{-march=sm_70}.
+
 @html
 <hr />
 @end html
diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
index e11b5bdb347..d96152b32a8 100644
--- a/gcc/doc/invoke.texi
+++ b/gcc/doc/invoke.texi
@@ -27549,7 +27549,9 @@  supported.
 Generate code for the specified PTX ISA target architecture
 (e.g.@: @samp{sm_35}).  Valid architecture strings are @samp{sm_30},
 @samp{sm_35}, @samp{sm_53}, @samp{sm_70}, @samp{sm_75} and
-@samp{sm_80}.  The default target architecture is sm_30.
+@samp{sm_80}.
+The default depends on how the compiler has been configured, see
+@option{--with-arch}.
 
 This option sets the value of the preprocessor macro
 @code{__PTX_SM__}; for instance, for @samp{sm_35}, it has the value
-- 
2.25.1