ST_SquareGrid — Returns a set of grid squares and cell indices that completely cover the bounds of the geometry argument.
setof record ST_SquareGrid(float8 size, geometry bounds
Starts with the concept of a square tiling of the plane. For a given planar SRS, and a given edge size, starting at the origin of the SRS, there is one unique square tiling of the plane, Tiling(SRS, Size). This function answers the question: what grids in a given Tiling(SRS, Size) overlap with a given bounds.
The SRS for the output squares is the SRS provided by the bounds geometry.
Doubling or edge size of the square generates a new parent tiling that perfectly fits with the original tiling. Standard web map tilings in mercator are just powers-of-two square grids in the mercator plane.
To do a point summary against a square tiling, generate a square grid using the extent of the points as the bounds, then spatially join to that grid. Note the estimated extent might be off from actual extent, so be cautious and at very least make sure you've analyzed your table.
SELECT COUNT(*), squares.geom FROM pointtable AS pts INNER JOIN ST_SquareGrid( 1000, ST_SetSRID(ST_EstimatedExtent('pointtable', 'geom'), 3857) ) AS squares ON ST_Intersects(pts.geom, squares.geom) GROUP BY squares.geom
This yields the same result as the first example but will be slower for a large number of points
SELECT COUNT(*), squares.geom FROM pointtable AS pts INNER JOIN ST_SquareGrid( 1000, pts.geom ) AS squares ON ST_Intersects(pts.geom, squares.geom) GROUP BY squares.geom