KJ16609/gcc-reflection
1
0
Fork 0
mirror of https://forge.sourceware.org/marek/gcc.git synced 2026-02-22 03:47:02 -05:00
Code Issues Packages Projects Releases Wiki Activity

Simplify OpenACC 'no_create' clause implementation

Browse Source 
For 'OFFSET_INLINED', 'gomp_map_val' does the right thing, and we may then
simplify the device plugins accordingly.

This is a follow-up to
Subversion r279551 (Git commit a6163563f2)
"Add OpenACC 2.6's no_create",
Subversion r279622 (Git commit 5bcd470bf0)
"Use gomp_map_val for OpenACC host-to-device address translation".

	libgomp/
	* target.c (gomp_map_vars_internal): Use 'OFFSET_INLINED' for
	'GOMP_MAP_IF_PRESENT'.
	* plugin/plugin-gcn.c (gcn_exec, GOMP_OFFLOAD_openacc_exec)
	(GOMP_OFFLOAD_openacc_async_exec): Adjust.
	* plugin/plugin-nvptx.c (nvptx_exec, GOMP_OFFLOAD_openacc_exec)
	(GOMP_OFFLOAD_openacc_async_exec): Likewise.
	* testsuite/libgomp.oacc-c-c++-common/no_create-1.c: Add 'async'
	testing.
	* testsuite/libgomp.oacc-c-c++-common/no_create-2.c: Likewise.
This commit is contained in:
Thomas Schwinge
2023-02-27 12:02:02 +01:00
parent b5037d4a07
commit 199867d07b
5 changed files with 54 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
libgomp/plugin/plugin-gcn.c
View File

@@ -3064,7 +3064,7 @@ wait_queue (struct goacc_asyncqueue *aq)
/* Execute an OpenACC kernel, synchronously or asynchronously. */
static void
gcn_exec (struct kernel_info *kernel, size_t mapnum, void **hostaddrs,
gcn_exec (struct kernel_info *kernel, size_t mapnum,
void **devaddrs, unsigned *dims, void *targ_mem_desc, bool async,
struct goacc_asyncqueue *aq)
{
@@ -3077,9 +3077,7 @@ gcn_exec (struct kernel_info *kernel, size_t mapnum, void **hostaddrs,
/* devaddrs must be double-indirect on the target. */
void **ind_da = alloc_by_agent (kernel->agent, sizeof (void*) * mapnum);
for (size_t i = 0; i < mapnum; i++)
hsa_fns.hsa_memory_copy_fn (&ind_da[i],
devaddrs[i] ? &devaddrs[i] : &hostaddrs[i],
sizeof (void *));
hsa_fns.hsa_memory_copy_fn (&ind_da[i], &devaddrs[i], sizeof (void *));
struct hsa_kernel_description *hsa_kernel_desc = NULL;
for (unsigned i = 0; i < kernel->module->image_desc->kernel_count; i++)
@@ -3887,27 +3885,27 @@ GOMP_OFFLOAD_async_run (int device, void *tgt_fn, void *tgt_vars,
void
GOMP_OFFLOAD_openacc_exec (void (*fn_ptr) (void *), size_t mapnum,
void **hostaddrs, void **devaddrs, unsigned *dims,
void **hostaddrs __attribute__((unused)),
void **devaddrs, unsigned *dims,
void *targ_mem_desc)
{
struct kernel_info *kernel = (struct kernel_info *) fn_ptr;
gcn_exec (kernel, mapnum, hostaddrs, devaddrs, dims, targ_mem_desc, false,
NULL);
gcn_exec (kernel, mapnum, devaddrs, dims, targ_mem_desc, false, NULL);
}
/* Run an asynchronous OpenACC kernel on the specified queue. */
void
GOMP_OFFLOAD_openacc_async_exec (void (*fn_ptr) (void *), size_t mapnum,
void **hostaddrs, void **devaddrs,
void **hostaddrs __attribute__((unused)),
void **devaddrs,
unsigned *dims, void *targ_mem_desc,
struct goacc_asyncqueue *aq)
{
struct kernel_info *kernel = (struct kernel_info *) fn_ptr;
gcn_exec (kernel, mapnum, hostaddrs, devaddrs, dims, targ_mem_desc, true,
aq);
gcn_exec (kernel, mapnum, devaddrs, dims, targ_mem_desc, true, aq);
}
/* Create a new asynchronous thread and queue for running future kernels. */

19
libgomp/plugin/plugin-nvptx.c
View File

@@ -742,8 +742,7 @@ link_ptx (CUmodule *module, const struct targ_ptx_obj *ptx_objs,
}
static void
nvptx_exec (void (*fn), size_t mapnum, void **hostaddrs, void **devaddrs,
unsigned *dims, void *targ_mem_desc,
nvptx_exec (void (*fn), size_t mapnum, unsigned *dims, void *targ_mem_desc,
CUdeviceptr dp, CUstream stream)
{
struct targ_fn_descriptor *targ_fn = (struct targ_fn_descriptor *) fn;
@@ -1530,7 +1529,8 @@ GOMP_OFFLOAD_free (int ord, void *ptr)
void
GOMP_OFFLOAD_openacc_exec (void (*fn) (void *), size_t mapnum,
void **hostaddrs, void **devaddrs,
void **hostaddrs __attribute__((unused)),
void **devaddrs,
unsigned *dims, void *targ_mem_desc)
{
GOMP_PLUGIN_debug (0, " %s: prepare mappings\n", __FUNCTION__);
@@ -1549,7 +1549,7 @@ GOMP_OFFLOAD_openacc_exec (void (*fn) (void *), size_t mapnum,
size_t s = mapnum * sizeof (void *);
hp = alloca (s);
for (int i = 0; i < mapnum; i++)
hp[i] = (devaddrs[i] ? devaddrs[i] : hostaddrs[i]);
hp[i] = devaddrs[i];
CUDA_CALL_ASSERT (cuMemAlloc, &dp, s);
if (profiling_p)
goacc_profiling_acc_ev_alloc (thr, (void *) dp, s);
@@ -1591,8 +1591,7 @@ GOMP_OFFLOAD_openacc_exec (void (*fn) (void *), size_t mapnum,
}
}
nvptx_exec (fn, mapnum, hostaddrs, devaddrs, dims, targ_mem_desc,
dp, NULL);
nvptx_exec (fn, mapnum, dims, targ_mem_desc, dp, NULL);
CUresult r = CUDA_CALL_NOCHECK (cuStreamSynchronize, NULL);
const char *maybe_abort_msg = "(perhaps abort was called)";
@@ -1617,7 +1616,8 @@ cuda_free_argmem (void *ptr)
void
GOMP_OFFLOAD_openacc_async_exec (void (*fn) (void *), size_t mapnum,
void **hostaddrs, void **devaddrs,
void **hostaddrs __attribute__((unused)),
void **devaddrs,
unsigned *dims, void *targ_mem_desc,
struct goacc_asyncqueue *aq)
{
@@ -1639,7 +1639,7 @@ GOMP_OFFLOAD_openacc_async_exec (void (*fn) (void *), size_t mapnum,
block = (void **) GOMP_PLUGIN_malloc (2 * sizeof (void *) + s);
hp = block + 2;
for (int i = 0; i < mapnum; i++)
hp[i] = (devaddrs[i] ? devaddrs[i] : hostaddrs[i]);
hp[i] = devaddrs[i];
CUDA_CALL_ASSERT (cuMemAlloc, &dp, s);
if (profiling_p)
goacc_profiling_acc_ev_alloc (thr, (void *) dp, s);
@@ -1688,8 +1688,7 @@ GOMP_OFFLOAD_openacc_async_exec (void (*fn) (void *), size_t mapnum,
}
}
nvptx_exec (fn, mapnum, hostaddrs, devaddrs, dims, targ_mem_desc,
dp, aq->cuda_stream);
nvptx_exec (fn, mapnum, dims, targ_mem_desc, dp, aq->cuda_stream);
if (mapnum > 0)
GOMP_OFFLOAD_openacc_async_queue_callback (aq, cuda_free_argmem, block);

2
libgomp/target.c
View File

@@ -1207,7 +1207,7 @@ gomp_map_vars_internal (struct gomp_device_descr *devicep,
{
/* Not present, hence, skip entry - including its MAP_POINTER,
when existing. */
tgt->list[i].offset = OFFSET_POINTER;
tgt->list[i].offset = OFFSET_INLINED;
if (i + 1 < mapnum
&& ((typemask & get_kind (short_mapkind, kinds, i + 1))
== GOMP_MAP_POINTER))

30
libgomp/testsuite/libgomp.oacc-c-c++-common/no_create-1.c
View File

@@ -22,15 +22,10 @@ main (int argc, char *argv[])
devptr[0] = &var;
devptr[1] = &arr[2];
}
if (acc_hostptr (devptr[0]) != (void *) &var)
__builtin_abort ();
if (acc_hostptr (devptr[1]) != (void *) &arr[2])
__builtin_abort ();
acc_delete (&var, sizeof (var));
acc_delete (arr, N * sizeof (*arr));
#if ACC_MEM_SHARED
if (devptr[0] != &var)
__builtin_abort ();
@@ -43,6 +38,31 @@ main (int argc, char *argv[])
__builtin_abort ();
#endif
#pragma acc parallel no_create(var, arr[0:N]) copyout(devptr) async
{
devptr[0] = &arr[N - 2];
devptr[1] = &var;
}
#pragma acc wait
if (acc_hostptr (devptr[0]) != (void *) &arr[N - 2])
__builtin_abort ();
if (acc_hostptr (devptr[1]) != (void *) &var)
__builtin_abort ();
#if ACC_MEM_SHARED
if (devptr[0] != &arr[N - 2])
__builtin_abort ();
if (devptr[1] != &var)
__builtin_abort ();
#else
if (devptr[0] == &arr[N - 2])
__builtin_abort ();
if (devptr[1] == &var)
__builtin_abort ();
#endif
acc_delete (&var, sizeof (var));
acc_delete (arr, N * sizeof (*arr));
free (arr);
return 0;

12
libgomp/testsuite/libgomp.oacc-c-c++-common/no_create-2.c
View File

@@ -18,12 +18,22 @@ main (int argc, char *argv[])
devptr[0] = &var;
devptr[1] = &arr[2];
}
if (devptr[0] != &var)
__builtin_abort ();
if (devptr[1] != &arr[2])
__builtin_abort ();
#pragma acc parallel no_create(var, arr[0:N]) copyout(devptr) async
{
devptr[0] = &arr[N - 2];
devptr[1] = &var;
}
#pragma acc wait
if (devptr[0] != &arr[N - 2])
__builtin_abort ();
if (devptr[1] != &var)
__builtin_abort ();
free (arr);
return 0;