HISQ mixed precision deflation

For this case study we are using a 48x48x48x12 configuration as provided by the HotQCD collaboration.

Parameter	Value
Volume	48x48x48x12
Gauge action	Improved Symanzik
beta	6.794
Fermion action	HISQ fermions
light quark mass	0.00167
strange quark mass	0.0450

Here were are going to examine the quark mass dependence of the solve time as we scale from the light to strange quark masses and progressively optimize the solver with mixed precision and deflation. For this study we are using the staggered_invert_test example code that is included with QUDA and run on a workstation using 2x Quadro GV100 GPUs. All of these runs are done using a launch syntax of the form

export QUDA_RESOURCE_PATH.
ARGS="--dim 48 48 24 12 --gridsize 1 1 2 1 --load-gauge /scratch/mathias/l4812f21b6794m00167m0450c_130.quda --compute-fat-long true --test 1"

mpirun -np 2 tests/staggered_invert_test $ARGS $RECON $PREC $SOLVER $EIG --mass 0.00167 --verbosity verbose

where we will adjust the variables SOLVER, PREC, and RECON according to the solver parameters as desired.

Starting point - Pure Double CG solver

Our initial starting point is using a standard double precision CG solver. This uses the parameters:

SOLVER="--inv-type cg --tol 1e-10 --reliable-delta 0.001 --niter 10000"
PREC="--prec double"

which means we run a CG solver to a relative residual tolerance of 1e-10, reliably updating the true residual every time the iterated residual drops by 3 orders of magnitude, with a maximum iteration count of 10000 using double precision only.

Mass	Iterations	Time	GFLOPS
0.00167	6178	20.8658	468
0.00334	4551	15.4005	467
0.00668	2621	8.88522	467
0.01336	1431	4.8658	467
0.02672	759	2.59715	467
0.05344	400	1.38505	467

With HISQ fermions we can also use compression on the long-link field to reduce the memory traffic. We do so with these parameters

RECON="--recon 13 --recon-sloppy 9"

where we only use the maximal reconstruct-9 compression on the sloppy updates to ensure stability. In doing so we that the iteration count is constant, and we improve the overall solve time by around 1.2x. From now on, we will assume that gauge compression is always used.

Mass	Iterations	Time	GFLOPS
0.00167	6178	17.5381	556
0.00334	4551	12.9228	557
0.00668	2621	7.45696	556
0.01336	1431	4.08501	556
0.02672	759	2.18148	556
0.05344	400	1.16466	556

Mixed-Precision CG

The first significant performance boost is to enable mixed-precision CG. To do so, we simple set the sloppy precision to a lower precision than the outer solver precision. Valid values are single, half and quarter, with the latter two formats being QUDA's custom block-fixed-point formats. E.g.,

SOLVER="--inv-type cg --tol 1e-10 --reliable-delta 0.1 --niter 10000"
PREC="--prec double --prec-sloppy single"

Note we also change the reliable-delta parameter such that the true residual is recomputed every time the residual drops by an order of magnitude. This is to minimize the divergence of the solver due to the reduced precision.

	double-single			double-half			double-quarter
Mass	Iterations	Time	GFLOPS	Iterations	Time	GFLOPS	Iterations	Time	GFLOPS
0.00167	6179	8.73093	1120	7710	6.65217	1830	14973	8.58425	2760
0.00334	4552	6.43182	1120	5118	4.4089	1830	7786	4.51143	2730
0.00668	2621	3.71859	1120	2884	2.5061	1820	3790	2.23699	2690
0.01336	1431	2.04482	1110	1456	1.28795	1800	1869	1.13276	2640
0.02672	759	1.09941	1100	759	0.677756	1790	900	0.570441	2540
0.05344	400	0.593991	1090	400	0.374092	1730	487	0.324537	2440

These results demonstrate that more GFLOPS doesn't mean reduced time to solution. In particular we see that the double-quarter solver is not numerically stable as we reduce the quark mass. At the light quark mass we more than double the iteration count for a solve time similar to the double-single solver. On the other hand the double-single solver has almost identical iteration count to the pure double solver, and as a result has the expected 2x speedup. While the double-half shows increased iteration count at small quark mass, overall it is stable and represents the sweet spot with a 2.6x speedup over the pure double solver at the light quark mass and 3.1x speedup at the heavy quark mass.