https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88764
Bug ID: 88764
Summary: [nvptx, libgomp, testsuite] Update testsuite for
default vector length larger than 32
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: vries at gcc dot gnu.org
Target Milestone: ---
With the nvptx vector-length patches arriving on trunk, it starts to become
important to run the libgomp openacc testsuite with a vector-length default of
larger than 32, at the most extreme: 1024, f.i. by setting:
...
-#define PTX_DEFAULT_VECTOR_LENGTH PTX_WARP_SIZE
+#define PTX_DEFAULT_VECTOR_LENGTH PTX_CTA_SIZE
...
or by using -fopenacc-dim=::1024 as default in the libgomp testsuite.
There are however test-cases that assume default vector_length of 32, so we
need some way to indicate this, f.i. (or to start with) an effective target
openacc_default_vector_length_32, and update those test-cases.
Focusing on the execution failures, with '#define PTX_DEFAULT_VECTOR_LENGTH
PTX_CTA_SIZE' I see these tests failing with insufficient resources:
...
FAIL: libgomp.oacc-fortran/kernels-loop-2.f95 -DACC_DEVICE_TYPE_nvidia=1
-DACC_MEM_SHARED=0 -O2 execution test
FAIL: libgomp.oacc-fortran/kernels-loop-2.f95 -DACC_DEVICE_TYPE_nvidia=1
-DACC_MEM_SHARED=0 -O3 -fomit-frame-pointer -funroll-loops -fpeel-loops
-ftracer -finline-functions execution test
FAIL: libgomp.oacc-fortran/kernels-loop-2.f95 -DACC_DEVICE_TYPE_nvidia=1
-DACC_MEM_SHARED=0 -O3 -g execution test
FAIL: libgomp.oacc-c/../libgomp.oacc-c-c++-common/kernels-loop-2.c
-DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -O2 execution test
FAIL: libgomp.oacc-c++/../libgomp.oacc-c-c++-common/kernels-loop-2.c
-DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -O2 execution test
...
and these tests failing due to a hard-coded check verifying default vector
length is 32:
...
FAIL: libgomp.oacc-c++/../libgomp.oacc-c-c++-common/parallel-dims.c
-DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -O0 execution test
FAIL: libgomp.oacc-c++/../libgomp.oacc-c-c++-common/parallel-dims.c
-DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -O2 execution test
FAIL: libgomp.oacc-c/../libgomp.oacc-c-c++-common/parallel-dims.c
-DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -O0 execution test
FAIL: libgomp.oacc-c/../libgomp.oacc-c-c++-common/parallel-dims.c
-DACC_DEVICE_TYPE_nvidia=1 -DACC_MEM_SHARED=0 -O2 execution test
...
