Haftung; Geschäftsführer: Thomas
Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht
München, HRB 106955
commit 9044b7efb3518de180a5b3168615b7e12d93eea8
Author: Marcel Vollweiler
Date: Tue Dec 13 12:04:48 2022 +
OpenMP, libgomp: Handle unified shared memory
Hi Jakub,
> * testsuite/libgomp.c-c++-common/icv-4.c: Bugfix.
Better say what exactly you changed in words.
Changed.
> --- a/gcc/gimplify.cc
> +++ b/gcc/gimplify.cc
> @@ -14153,7 +14153,7 @@ optimize_target_teams (tree target, gimple_seq
*pre_p)
>struct gimplify_omp_ctx *target_ctx
Hi Jakub,
The last version of this patch was slightly adapted to the latest changes of the
device-specific environment variable syntax
(https://gcc.gnu.org/pipermail/gcc-patches/2022-September/601145.html), also
considering the latest related bug fixes (commits 994ea892bd02d and
7d37c7f67c1bb).
Hi Jakub,
Am 22.08.2022 um 17:35 schrieb Jakub Jelinek:
+/* Default values of ICVs according to the OpenMP standard. */
+struct gomp_default_icv_t gomp_default_icv_values = {
+ .run_sched_var = GFS_DYNAMIC,
+ .run_sched_chunk_size = 1,
+ .max_active_levels_var = 1,
+ .bind_var =
Hi Jakub,
This patch was reduced a bit and most of your comments were considered in the
last submission of the environment variable syntax extension patch
(https://gcc.gnu.org/pipermail/gcc-patches/2022-August/599175.html). This patch
also builds on that envvar patch version.
The nteams-var
Hi Jakub,
I'm not sure we can rely on execv on all targets that do support libgomp.
Any reason why you actually need this, rather than using
dg-set-target-env-var directive(s) and perhaps return 0; if getenv doesn't
return the expected values?
Interesting topic. After some (internal)
Hi,
I’d like to ping the patch for the OpenMP runtime routines omp_get_max_teams,
omp_set_num_teams, and omp_{gs}et_teams_thread_limit on offload devices:
https://gcc.gnu.org/pipermail/gcc-patches/2022-April/593260.html
This patch builds on the following patch which is currently in
Hi Jakub,
Am 17.05.2022 um 20:08 schrieb Jakub Jelinek:
On Tue, May 17, 2022 at 11:57:02AM +0200, Marcel Vollweiler wrote:
More importantly, I have no idea how this can work when you pass arg_size 0
and arg_align 0. The s variable is in the current function frame, with
arg_size 0 nothing
Hi Jakub,
--- a/libgomp/libgomp.map
+++ b/libgomp/libgomp.map
@@ -224,6 +224,8 @@ OMP_5.1 {
omp_set_teams_thread_limit_8_;
omp_get_teams_thread_limit;
omp_get_teams_thread_limit_;
+omp_target_memcpy_async;
+omp_target_memcpy_rect_async;
} OMP_5.0.2;
These should be
Hi Jakub,
diff --git a/gcc/cp/semantics.cc b/gcc/cp/semantics.cc
index 0cb17a6..452ecfd 100644
--- a/gcc/cp/semantics.cc
+++ b/gcc/cp/semantics.cc
@@ -8534,11 +8534,14 @@ finish_omp_clauses (tree clauses, enum
c_omp_region_type ort)
{
if (handle_omp_array_sections (c, ort))
Hi,
This is a follow up patch of the patch that adds the OpenMP runtime routine
omp_target_is_accessible:
https://gcc.gnu.org/pipermail/gcc-patches/2022-March/591601.html
It considers now also unified shared memory (usm) that was submitted recently
(but not yet approved/committed):
Hi Jakub,
Am 05.05.2022 um 11:33 schrieb Jakub Jelinek:
On Mon, Mar 14, 2022 at 04:42:14PM +0100, Marcel Vollweiler wrote:
--- a/libgomp/libgomp.map
+++ b/libgomp/libgomp.map
@@ -226,6 +226,11 @@ OMP_5.1 {
omp_get_teams_thread_limit_;
} OMP_5.0.2;
+OMP_5.1.1 {
+ global
Hi,
This patch adds support for omp_get_max_teams, omp_set_num_teams, and
omp_{gs}et_teams_thread_limit on offload devices.
The patch builds on the following patches which are submitted, but not yet
approved/committed:
- [PATCH] OpenMP, libgomp: Environment variable syntax extension.
Hi Jakub,
! { dg-do run }
! { dg-additional-options "-fdefault-integer-8" }
program set_num_teams_8
use omp_lib
omp_set_num_teams (42)
if (omp_get_num_teams () .ne. 42) stop 1
end program
I modified your suggested test case a bit:
program set_num_teams_8
use omp_lib
use,
Hi,
This patch fixes a small bug for omp_set_num_teams in fortran.c.
Marcel
-
Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634
München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas
Heurung, Frank Thürauf; Sitz der Gesellschaft:
Hi Tobias,
Minor remark to the test:
On 11.03.22 13:30, Marcel Vollweiler wrote:
+ int d = omp_get_default_device ();
...
+ int shared_mem = 0;
+ #pragma omp target map (alloc: shared_mem) device (d)
+shared_mem = 1;
+ if (omp_target_is_accessible (p, sizeof (int), d) != shared_mem
Hi,
This patch adds the OpenMP runtime routine "omp_target_is_accessible" which was
introduced in OpenMP 5.1 (specification section 3.8.4):
"The omp_target_is_accessible routine tests whether host memory is accessible
from a given device."
"This routine returns true if the storage of size
Hi Jakub,
This is an update to the patch from Tue Mar 8:
https://gcc.gnu.org/pipermail/gcc-patches/2022-March/591343.html
I just added "get_mapped_ptr" to the "omp_runtime_apis" array in omp-low.cc and
replaced "omp_get_num_devices" by "gomp_get_num_devices" in target.c.
Marcel
Hi Jakub,
diff --git a/libgomp/libgomp.map b/libgomp/libgomp.map
index 2ac5809..00a4858 100644
--- a/libgomp/libgomp.map
+++ b/libgomp/libgomp.map
@@ -224,6 +224,7 @@ OMP_5.1 {
omp_set_teams_thread_limit_8_;
omp_get_teams_thread_limit;
omp_get_teams_thread_limit_;
+
Hi,
This patch adds the OpenMP runtime routine "omp_get_mapped_ptr" which was
introduced in OpenMP 5.1 (specification section 3.8.11):
"The omp_get_mapped_ptr routine returns the device pointer that is associated
with a host pointer for a given device."
"The device_num argument must be greater
Hi,
The patch for adding the has_device_addr clause on the target construct was
recently committed (bbb7f8604e1dfc08f44354cfd93d2287f2fdd489).
Additionally, this patch adds support for list items in the has_device_addr
clause which type is given by C++ template parameters.
Marcel
Hi,
This patch adds two new OpenMP runtime routines: omp_target_memcpy_async and
omp_target_memcpy_rect_async. Both functions are introduced in OpenMP 5.1 as
asynchronous variants of omp_target_memcpy and omp_target_memcpy_rect.
In contrast to the synchronous variants, the asynchronous
Hi Jakub,
+case OMP_CLAUSE_HAS_DEVICE_ADDR:
+ t = OMP_CLAUSE_DECL (c);
+ if (TREE_CODE (t) == TREE_LIST)
+{
+ if (handle_omp_array_sections (c, ort))
+remove = true;
+ else
+{
+ t = OMP_CLAUSE_DECL (c);
+
Hi,
This patch considers the environment variable syntax extension for
device-specific variants of environment variables from OpenMP 5.1 (see
OpenMP 5.1 specification, p. 75 and p. 639). An environment variable
(e.g. OMP_NUM_TEAMS) can have different suffixes:
_DEV (e.g. OMP_NUM_TEAMS_DEV):
Hi,
This patch considers the environment variable syntax extension for
device-specific variants of environment variables from OpenMP 5.1 (see
OpenMP 5.1 specification, p. 75 and p. 639). An environment variable
(e.g. OMP_NUM_TEAMS) can have different suffixes:
_DEV (e.g. OMP_NUM_TEAMS_DEV):
Hi Thomas,
Am 18.01.2022 um 13:25 schrieb Thomas Schwinge:
Hi!
Maybe I'm just totally confused -- as so often ;-) -- but things seem
strange here:
On 2022-01-12T10:43:05+0100, Marcel Vollweiler wrote:
Currently omp_get_device_num does not work on gcn targets with more than
one offload
Hi,
Currently omp_get_device_num does not work on gcn targets with more than
one offload device. The reason is that GOMP_DEVICE_NUM_VAR is static in
icv-device.c and thus "__gomp_device_num" is not visible in the offload
image.
This patch removes "static" such that "__gomp_device_num" is now
Hi,
I'd like to ping the patch for the OpenMP 'has_device_addr' clause on
the target construct:
https://gcc.gnu.org/pipermail/gcc-patches/2021-November/585361.html
Thanks
Marcel
-
Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634
München;
Hi Jakub,
this is again a new version of the 'has_device_addr' patch. It includes
further minor changes in the C/C++ part and in addition the Fortran
implementation.
Tested on x86_64-linux with nvptx offloading with no regressions.
Marcel
-
Siemens Electronic Design Automation
Hi Jakub,
Am 20.10.2021 um 14:38 schrieb Jakub Jelinek:
On Mon, Oct 18, 2021 at 06:17:20PM +0200, Marcel Vollweiler wrote:
@@ -14255,6 +14257,16 @@ c_parser_omp_clause_use_device_addr (c_parser *parser,
tree list)
list);
}
+/* OpenMP 5.1
Hi,
This patch adds the 'has_device_addr' clause to the OpenMP 'target'
construct which was introduced in OpenMP 5.1:
"The has_device_addr clause was added to the target construct to allow
access to variables or array sections that already have a device
address" (OpenMP 5.1 Specification, p.
Hi Jakub,
Am 11.10.2021 um 11:49 schrieb Jakub Jelinek:
On Mon, Oct 11, 2021 at 11:40:54AM +0200, Marcel Vollweiler wrote:
libgomp: Add tests for omp_atv_serialized and deprecate omp_atv_sequential.
The variable omp_atv_sequential was replaced by omp_atv_serialized in OpenMP
5.1
Hi,
The variable omp_atv_sequential was replaced by omp_atv_serialized in
OpenMP 5.1. This was already implemented by Jakub (C/C++, commit
ea82325afec) and Tobias (Fortran, commit fff15bad1ab).
This patch adds two tests to check if omp_atv_serialized is available
(one test for C/C++ and one for
Hi,
this patch adds support for the 'seq_cst' memory order clause on the
'flush' directive which was introduced in OpenMP 5.1 (p.275ff of the
OpenMP 5.1 Specification):
"If neither memory-order-clause nor a list appears on the flush
construct then the behavior is as if memory-order-clause is
Am 01.09.2021 um 11:02 schrieb Jakub Jelinek:
On Wed, Sep 01, 2021 at 09:06:31AM +0200, Christophe Lyon wrote:
* gfortran.dg/gomp/target-device-ancestor-4.f90: New test.
The last new test fails on aarch64:
/gcc/testsuite/gfortran.dg/gomp/target-device-ancestor-4.f90:7:15: Error:
:32PM +0200, Marcel Vollweiler wrote:
--- a/gcc/c/c-parser.c
+++ b/gcc/c/c-parser.c
@@ -15864,37 +15864,81 @@ c_parser_omp_clause_map (c_parser *parser, tree list)
}
/* OpenMP 4.0:
- device ( expression
Hi Jakub,
this is the second version of the patch for the device-modifiers for
'omp target device'.
Am 20.07.2021 um 15:30 schrieb Jakub Jelinek:
On Wed, Jul 07, 2021 at 07:59:58PM +0200, Marcel Vollweiler wrote:
OpenMP: Add support for device-modifiers for 'omp target device'
gcc/c
This patch adds device-modifiers to the device clause:
#pragma omp target device ([ device-modifier :] integer-expression)
where device-modifier is either 'ancestor' or 'device_num'.
The 'device_num' case
#pragma omp target device (device_num : integer-expression)
is treated in the
Am 16.06.2021 um 19:19 schrieb Joseph Myers:
On Wed, 16 Jun 2021, Julian Brown wrote:
+if test x$gcc_cv_as_gcn_global_load_fixed = xyes; then
+ AC_DEFINE(HAVE_GCN_ASM_GLOBAL_LOAD_FIXED, 1, [Define if your
assembler has fixed global_load functions.])
+else
+
Changed the variable "gcc_cv_as_global_load_fixed" into
"gcc_cv_as_gcn_global_load_fixed" in order to have the "gcn" substring
also in the config.patch file.
Am 09.06.2021 um 16:47 schrieb Marcel Vollweiler:
This patch fixes an issue with global_load assembler f
This patch fixes an issue with global_load assembler functions leading
to a "invalid operand for instruction" error since in different LLVM
versions those functions use either one or two registers.
In this patch a compatibility check is added to the configure.ac.
Marcel
-
Mentor
Hi Jakub,
Am 20.05.2021 um 10:57 schrieb Jakub Jelinek:
On Thu, May 20, 2021 at 10:47:52AM +0200, Marcel Vollweiler wrote:
--- a/gcc/fortran/openmp.c
+++ b/gcc/fortran/openmp.c
@@ -1710,10 +1710,21 @@ gfc_match_omp_clauses (gfc_omp_clauses **cp, const
omp_mask mask,
&& g
Hi,
This patch adds handling for the map-type-modifier 'close' in the map
clause in the Fortran parser (for C and C++ parsers the changes were
already committed).
'close' was introduced with OpenMP 5.0: "The close map-type-modifier is
a hint to the runtime to allocate memory close to the target
Am 12.05.2021 um 19:37 schrieb Marcel Vollweiler:
-
Mentor Graphics (Deutschland) GmbH, Arnulfstrasse 201, 80634 München
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung,
Frank Thürauf
-
Mentor Graphics (Deutschland) GmbH, Arnulfstrasse 201
-
Mentor Graphics (Deutschland) GmbH, Arnulfstrasse 201, 80634 München
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Frank
Thürauf
commit 8982a5354d2697eeb12a77d49b6730db90053618
Author: Marcel Vollweiler
Date: Wed May 12 10:14:41 2021 -0700
Am 11.05.2021 um 17:20 schrieb Jakub Jelinek:
On Tue, May 11, 2021 at 04:27:55PM +0200, Marcel Vollweiler wrote:
The usual wording would be
"too many % modifiers"
Changed for 'always' and 'close' for C and C++.
One extra thing, sorry, forgot to mention, for the translator
Am 10.05.2021 um 20:34 schrieb Jakub Jelinek:
On Mon, May 10, 2021 at 04:11:39PM +0200, Marcel Vollweiler wrote:
@@ -15660,37 +15665,54 @@ c_parser_omp_clause_map (c_parser *parser, tree list)
if (!parens.require_open (parser))
return list;
- if (c_parser_next_token_is (parser
Hi,
This patch adds handling for the map-type-modifier 'close' in the map
clause that was introduced with OpenMP 5.0: "The close map-type-modifier
is a hint to the runtime to allocate memory close to the target device."
In OpenMP 5.0 'close' can be used beside/together with 'always' in a
list
48 matches
Mail list logo
