Skip to content

Optimize gemv for small M, large N only if it can be done in a threadsafe manner #1865

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Nov 12, 2018
Merged

Optimize gemv for small M, large N only if it can be done in a threadsafe manner #1865

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4f43668
Merge pull request #2 from xianyi/develop
martin-frbg Nov 10, 2018
0427277
Allow optimization for small m, large n only if it can be made thread…
martin-frbg Nov 10, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 33 additions & 4 deletions driver/level2/gemv_thread.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,9 +62,36 @@
#endif
#endif

#ifndef TRANSA
#ifndef thread_local
# if __STDC_VERSION__ >= 201112 && !defined __STDC_NO_THREADS__
# define thread_local _Thread_local
# elif defined _WIN32 && ( \
defined _MSC_VER || \
defined __ICL || \
defined __DMC__ || \
defined __BORLANDC__ )
# define thread_local __declspec(thread)
/* note that ICC (linux) and Clang are covered by __GNUC__ */
# elif defined __GNUC__ || \
defined __SUNPRO_C || \
defined __xlC__
# define thread_local __thread
# else
# define UNSAFE
#endif
#endif
#if defined USE_OPENMP
#undef UNSAFE
#endif

#if !defined(TRANSA) && !defined(UNSAFE)
#define Y_DUMMY_NUM 1024
#if defined(USE_OPENMP)
static FLOAT y_dummy[Y_DUMMY_NUM];
#pragma omp threadprivate(y_dummy)
# else
static thread_local FLOAT y_dummy[Y_DUMMY_NUM];
# endif
#endif

static int gemv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *dummy1, FLOAT *buffer, BLASLONG pos){
Expand Down Expand Up @@ -105,10 +132,12 @@ static int gemv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
#ifdef TRANSA
y += n_from * incy * COMPSIZE;
#else
# ifndef UNSAFE
//for split matrix row (n) direction and vector x of gemv_n
x += n_from * incx * COMPSIZE;
//store partial result for every thread
y += (m_to - m_from) * 1 * COMPSIZE * pos;
# endif
#endif
}

Expand Down Expand Up @@ -136,7 +165,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x

BLASLONG width, i, num_cpu;

#ifndef TRANSA
#if !defined(TRANSA) && !defined(iUNSAFE)
int split_x=0;
#endif

Expand Down Expand Up @@ -212,7 +241,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
i -= width;
}

#ifndef TRANSA
#if !defined(TRANSA) && !defined(UNSAFE)
//try to split matrix on row direction and x.
//Then, reduction.
if (num_cpu < nthreads) {
Expand Down Expand Up @@ -272,7 +301,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
exec_blas(num_cpu, queue);
}

#ifndef TRANSA
#if !defined(TRANSA) && !defined(UNSAFE)
if(split_x==1){
//reduction
for(i=0; i<num_cpu; i++){
Expand Down