portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-18 07:47:06 +01:00
Code Releases Activity

Precision change improvements

Browse Source 
Added a new, much faster implementation of precision change that uses (optionally) a precomputed workspace containing pointer offsets that is device resident, such that all lattice copying occurs only on the device and no host<->device transfer is required, other than the pointer table. It also avoids the need to unpack and repack the fields using explicit lane copying. When this new precisionChange is called without a workspace, one will be computed on-the-fly; however it is still considerably faster than the original implementation.

In the special case of using double2 and when the Grids are the same, calls to the new precisionChange will automatically use precisionChangeFast, such that there is a single API call for all precision changes.

Reliable update and mixed-prec multishift have been modified to precompute precision change workspaces

Renamed the original precisionChange as precisionChangeOrig

Fixed incorrect pointer offset bug in copyLane

Added a test and a benchmark for precisionChange

Added a test for reliable update CG
This commit is contained in:
Christopher Kelly
2023-02-21 10:52:42 -05:00
parent 472ed2dd5c
commit 1db58a8acc
7 changed files with 616 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
Grid/algorithms/iterative/ConjugateGradientMultiShiftMixedPrec.h
View File

@ -130,6 +130,9 @@ public:
GRID_TRACE("ConjugateGradientMultiShiftMixedPrec");
GridBase *DoublePrecGrid = src_d.Grid();
precisionChangeWorkspace pc_wk_s_to_d(DoublePrecGrid,SinglePrecGrid);
precisionChangeWorkspace pc_wk_d_to_s(SinglePrecGrid,DoublePrecGrid);
////////////////////////////////////////////////////////////////////////
// Convenience references to the info stored in "MultiShiftFunction"
////////////////////////////////////////////////////////////////////////
@ -200,10 +203,10 @@ public:
r_d = p_d;
//MdagM+m[0]
precisionChangeFast(p_f,p_d);
precisionChange(p_f, p_d, pc_wk_d_to_s);
Linop_f.HermOpAndNorm(p_f,mmp_f,d,qq); // mmp = MdagM p d=real(dot(p, mmp)), qq=norm2(mmp)
precisionChangeFast(tmp_d,mmp_f);
precisionChange(tmp_d, mmp_f, pc_wk_s_to_d);
Linop_d.HermOpAndNorm(p_d,mmp_d,d,qq); // mmp = MdagM p d=real(dot(p, mmp)), qq=norm2(mmp)
tmp_d = tmp_d - mmp_d;
std::cout << " Testing operators match "<<norm2(mmp_d)<<" f "<<norm2(mmp_f)<<" diff "<< norm2(tmp_d)<<std::endl;
@ -263,7 +266,7 @@ public:
AXPYTimer.Stop();
PrecChangeTimer.Start();
precisionChangeFast(p_f, p_d); //get back single prec search direction for linop
precisionChange(p_f, p_d, pc_wk_d_to_s); //get back single prec search direction for linop
PrecChangeTimer.Stop();
cp=c;
@ -272,7 +275,7 @@ public:
MatrixTimer.Stop();
PrecChangeTimer.Start();
precisionChangeFast(mmp_d, mmp_f); // From Float to Double
precisionChange(mmp_d, mmp_f, pc_wk_s_to_d); // From Float to Double
PrecChangeTimer.Stop();
AXPYTimer.Start();