scnc
/
qmcpack
mirror of https://github.com/QMCPACK/qmcpack.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Bugfix with sse/vgh and update edison. 

git-svn-id: https://subversion.assembla.com/svn/qmcdev/trunk@6013 e5b18d87-469d-4833-9cc0-8cdfa06e9491
This commit is contained in:
Jeongnim Kim 2013-10-22 21:00:49 +00:00
parent 1d9c0e7c1a
commit 0792719e7c
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
config/XC30Intel.cmake
View File

@ -36,8 +36,8 @@
SET(CMAKE_SYSTEM_PROCESSOR "XC30")
set(CMAKE_CXX_COMPILER /opt/cray/craype/1.01/bin/CC)
set(CMAKE_C_COMPILER /opt/cray/craype/1.01/bin/cc)
set(CMAKE_CXX_COMPILER /usr/common/usg/darshan/craype/1.06/bin/CC)
set(CMAKE_C_COMPILER /usr/common/usg/darshan/craype/1.06/bin/cc)
set(GNU_OPTS "-DADD_ -DINLINE_ALL=inline -DDISABLE_TIMER -DUSE_REAL_STRUCT_FACTOR")
set(INTEL_OPTS " -static -g -restrict -unroll -O3 -ip -xAVX -openmp -Wno-deprecated")
set(XT_FLAGS "-D_CRAYMPI")# -DHAVE_FMA4=1")

8
src/einspline/multi_bspline_eval_sse_s_impl.h
View File

@ -1603,13 +1603,13 @@ eval_multi_UBspline_3d_s_vghgh (const multi_UBspline_3d_s *spline,
// y-dependent vectors
_MM_MATVEC4_PS (A_s[ 0], A_s[ 1], A_s[ 2], A_s[ 3], tpy, b4);
_MM_MATVEC4_PS (A_s[ 4], A_s[ 5], A_s[ 6], A_s[ 7], tpy, db4);
_MM_MATVEC4_PS (A_s[ 8], A_s[ 9], A_s[10], A_s[11], tpy, d2a4);
_MM_MATVEC4_PS (A_s[12], A_s[13], A_s[14], A_s[15], tpy, d3a4);
_MM_MATVEC4_PS (A_s[ 8], A_s[ 9], A_s[10], A_s[11], tpy, d2b4);
_MM_MATVEC4_PS (A_s[12], A_s[13], A_s[14], A_s[15], tpy, d3b4);
// z-dependent vectors
_MM_MATVEC4_PS (A_s[ 0], A_s[ 1], A_s[ 2], A_s[ 3], tpz, c4);
_MM_MATVEC4_PS (A_s[ 4], A_s[ 5], A_s[ 6], A_s[ 7], tpz, dc4);
_MM_MATVEC4_PS (A_s[ 8], A_s[ 9], A_s[10], A_s[11], tpz, d2a4);
_MM_MATVEC4_PS (A_s[12], A_s[13], A_s[14], A_s[15], tpz, d3a4);
_MM_MATVEC4_PS (A_s[ 8], A_s[ 9], A_s[10], A_s[11], tpz, d2c4);
_MM_MATVEC4_PS (A_s[12], A_s[13], A_s[14], A_s[15], tpz, d3c4);
__m128 a[4], b[4], c[4], da[4], db[4], dc[4], d2a[4], d2b[4], d2c[4], d3a[4], d3b[4], d3c[4];
__m128 tmp;
// Unpack a values