floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Allow optimization for small m, large n only if it can be made threadsafe 

otherwise the introduction of a static array in 8e5a108 to improve #532 breaks concurrent calls from multiple threads as seen in #1844
This commit is contained in:
Martin Kroeker 2018-11-10 15:45:54 +01:00 committed by GitHub
parent 4f43668eec
commit 0427277cef
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 33 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
driver/level2/gemv_thread.c
View File

@ -62,9 +62,36 @@
#endif #endif
#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 #define Y_DUMMY_NUM 1024
#if defined(USE_OPENMP)
static FLOAT y_dummy[Y_DUMMY_NUM]; static FLOAT y_dummy[Y_DUMMY_NUM];
#pragma omp threadprivate(y_dummy)
# else
static thread_local FLOAT y_dummy[Y_DUMMY_NUM];
# endif
#endif #endif
static int gemv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *dummy1, FLOAT *buffer, BLASLONG pos){ static int gemv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *dummy1, FLOAT *buffer, BLASLONG pos){
@ -105,10 +132,12 @@ static int gemv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
#ifdef TRANSA #ifdef TRANSA
y += n_from * incy * COMPSIZE; y += n_from * incy * COMPSIZE;
#else #else
# ifndef UNSAFE
//for split matrix row (n) direction and vector x of gemv_n //for split matrix row (n) direction and vector x of gemv_n
x += n_from * incx * COMPSIZE; x += n_from * incx * COMPSIZE;
//store partial result for every thread //store partial result for every thread
y += (m_to - m_from) * 1 * COMPSIZE * pos; y += (m_to - m_from) * 1 * COMPSIZE * pos;
# endif
#endif #endif
} }
@ -136,7 +165,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
BLASLONG width, i, num_cpu; BLASLONG width, i, num_cpu;
#ifndef TRANSA #if !defined(TRANSA) && !defined(iUNSAFE)
int split_x=0; int split_x=0;
#endif #endif
@ -212,7 +241,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
i -= width; i -= width;
} }
#ifndef TRANSA #if !defined(TRANSA) && !defined(UNSAFE)
//try to split matrix on row direction and x. //try to split matrix on row direction and x.
//Then, reduction. //Then, reduction.
if (num_cpu < nthreads) { if (num_cpu < nthreads) {
@ -272,7 +301,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
exec_blas(num_cpu, queue); exec_blas(num_cpu, queue);
} }
#ifndef TRANSA #if !defined(TRANSA) && !defined(UNSAFE)
if(split_x==1){ if(split_x==1){
//reduction //reduction
for(i=0; i<num_cpu; i++){ for(i=0; i<num_cpu; i++){