TupleDescdesc;

intn_attrs;

inttile_pos;

uint64filter_mask;

vops_tile_hdr**tiles;

}vops_unnest_context;

func_ctx->user_fctx=user_ctx;

user_ctx->n_attrs=n_attrs;

user_ctx->tile_pos=0;

user_ctx->filter_mask=filter_mask;

filter_mask= ~0;

for (i=0;i<n_attrs;i++) {

Form_pg_attributeattr=src_desc->attrs[i];

vops_typetid=vops_get_type(attr->atttypid);

n_attrs=user_ctx->n_attrs;

for (j=user_ctx->tile_pos;j<TILE_SIZE;j++) {

if (filter_mask& ((uint64)1 <<j)) {

if (user_ctx->filter_mask& ((uint64)1 <<j))

{

for (i=0;i<n_attrs;i++) {

if (user_ctx->types[i]!=VOPS_LAST) {

vops_tile_hdr*tile=user_ctx->tiles[i];

Now most interesting thing: compare performance results on original table and using vector operations on VOPS projection.

All measurements were performed at desktop with 16Gb of RAM and quad-core i7-4770 CPU @ 3.40GHz processor with enabled hyper-threading.

Data set for benchmark was generated by dbgen utility included in TPC-H benchmark.

Scale factor is100 which corresponds to about 8Gb database. It can completely fit in memory,

Scale factor is10 which corresponds to about 8Gb database. It can completely fit in memory,

so we are measuring best query execution time for<i>warm</i> data.

Postgres was configured with shared buffer size equal to 8Gb.

For each query we measured time of sequential and parallel execution with 8 parallel workers.