Commit b2d4d6f2 authored by Andreas Marek's avatar Andreas Marek

Start to unify 128bit,256bit, and 512bit real kernels

parent 86f84746
......@@ -914,7 +914,6 @@ EXCLUDE = @top_srcdir@/src/GPU/check_for_gpu.F90 \
@top_srcdir@/src/elpa2/kernels/real_sse_6hv_double_precision.c \
@top_srcdir@/src/elpa2/kernels/complex_avx512_1hv_single_precision.c \
@top_srcdir@/src/elpa2/kernels/asm_x86_64_single_precision.s \
@top_srcdir@/src/elpa2/kernels/real_avx512_6hv_template.c \
@top_srcdir@/src/elpa2/kernels/real_avx-avx2_2hv_single_precision.c \
@top_srcdir@/src/elpa2/kernels/complex_simple.F90 \
@top_srcdir@/src/elpa2/kernels/complex_avx-avx2_1hv_single_precision.c \
......
......@@ -775,7 +775,6 @@ EXTRA_DIST = \
src/elpa2/kernels/complex_sse_1hv_template.c \
src/elpa2/kernels/complex_sse_2hv_template.c \
src/elpa2/kernels/complex_template.F90 \
src/elpa2/kernels/real_avx512_2hv_template.c \
src/elpa2/kernels/real_avx512_4hv_template.c \
src/elpa2/kernels/real_avx512_6hv_template.c \
src/elpa2/kernels/real_vsx_2hv_template.c \
......
......@@ -48,8 +48,12 @@
#define REALCASE 1
#define DOUBLE_PRECISION 1
#define BLOCK2 1
#define VEC_SET 512
#include "../../general/precision_macros.h"
#include "real_avx512_2hv_template.c"
#include "real_182bit_BLOCK_template.c"
#undef BLOCK2
#undef VEC_SET
#undef REALCASE
#undef DOUBLE_PRECISION
......@@ -48,8 +48,12 @@
#define REALCASE 1
#define SINGLE_PRECISION 1
#define BLOCK2 1
#define VEC_SET 512
#include "../../general/precision_macros.h"
#include "real_avx512_2hv_template.c"
#include "real_128bit_BLOCK_template.c"
#undef BLOCK2
#undef VEC_SET
#undef REALCASE
#undef SINGLE_PRECISION
This diff is collapsed.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment