FreeChainXenon/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

[libgomp, nvptx] Remove hard-coded const in nvptx_open_device

Browse source 
CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR is defined in cuda driver
api version 6.0 and higher.

Currently nvptx_open_device uses a hard-coded constant instead.

This patch fixes that by:
- defining CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR to the hardcoded
  constant at toplevel, if not present in cuda.h, and
- using CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR in nvptx_open_device

Build on x86_64 with nvptx accelerator and reg-tested libgomp.

2018-08-08  Tom de Vries  <tdevries@suse.de>

	* plugin/plugin-nvptx.c
	(CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR): Define.
	(nvptx_open_device): Use
	CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR.

From-SVN: r263406
This commit is contained in:
Tom de Vries 2018-08-08 14:26:19 +00:00 • committed by Tom de Vries
parent 94767dacea
commit b113af959c
2 changed files with 12 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
libgomp/ChangeLog
View file

@ -1,3 +1,10 @@
2018-08-08 Tom de Vries <tdevries@suse.de>
* plugin/plugin-nvptx.c
(CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR): Define.
(nvptx_open_device): Use
CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR.
2018-08-08 Tom de Vries <tdevries@suse.de>
* plugin/plugin-nvptx.c (cuda_error): Move declaration of cuGetErrorString ...

7
libgomp/plugin/plugin-nvptx.c
View file

@ -51,6 +51,7 @@
#if CUDA_VERSION < 6000
extern CUresult cuGetErrorString (CUresult, const char **);
#define CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR 82
#endif
#define DO_PRAGMA(x) _Pragma (#x)
@ -741,9 +742,11 @@ nvptx_open_device (int n)
&pi, CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK, dev);
ptx_dev->regs_per_block = pi;
/* CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR = 82 is defined only
/* CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR is defined only
in CUDA 6.0 and newer. */
r = CUDA_CALL_NOCHECK (cuDeviceGetAttribute, &pi, 82, dev);
r = CUDA_CALL_NOCHECK (cuDeviceGetAttribute, &pi,
CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR,
dev);
/* Fallback: use limit of registers per block, which is usually equal. */
if (r == CUDA_ERROR_INVALID_VALUE)
pi = ptx_dev->regs_per_block;