Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Added simple tile size calculation #2

Open
wants to merge 12 commits into
base: impala
Choose a base branch
from
Open

Added simple tile size calculation #2

Changes from 1 commit
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
a8674dc
Added simple tile size calculation
cod3monk Feb 24, 2017
c097a4f
Merge branch 'master' of github.com:cod3monk/stincilla
cod3monk Feb 24, 2017
1751ddd
added debug flag for tiling
cod3monk Feb 27, 2017
48cc401
removed PE on debug output
cod3monk Mar 2, 2017
366e4b1
added support for any stencil shape
cod3monk Mar 2, 2017
290493a
Merge branch 'master' of github.com:AnyDSL/stincilla
cod3monk Mar 10, 2017
cf8a53c
removed for-loops to allow partial evaluation
cod3monk Mar 10, 2017
47465c9
Embarrassing little performance issue fix
cod3monk Mar 16, 2017
8221a71
Partially evaluate the loop for tiling.
richardmembarth Mar 20, 2017
7abdb58
Analyze stencil only once.
richardmembarth Mar 20, 2017
4087dd1
Merge remote-tracking branch 'upstream/master'
richardmembarth Nov 21, 2017
23a6ef4
Porting to new PE syntax.
richardmembarth Nov 21, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
82 changes: 46 additions & 36 deletions mapping_cpu.impala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -83,48 +83,56 @@ struct CacheSizes {
}

fn tiled_loop(xl: i32, xu: i32, yl: i32, yu: i32, mask: Mask, img: Img, body: fn(i32, i32) -> ()) -> () {
let debug_tiling = false;
fn get_tile_dims(mask: Mask, img: Img) -> (i32, i32) {
// TODO make this dimension independent, currently only 2D is supported
// TODO get cache information from machine/compiler/somewhere else

let caches = CacheSizes {sizes: [32*1024, 256*1024, 20*1024*1024], count: 3}; // in bytes
let element_size = sizeof[i32](); // TODO make this dynamic based on used type. templates?

//print_string("img: (");
//print_int(img.width);
//print_string(", ");
//print_int(img.height);
//print_string(" : ");
//print_int(img.stride);
//print_string(")\n");
//
//print_string("mask: (");
//print_int(mask.size_x);
//print_string(", ");
//print_int(mask.size_y);
//print_string(")\n");
if debug_tiling @{
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No @ is required here - the frontend will already discard the conditional code if debug_tiling is false.

print_string("img: (");
print_int(img.width);
print_string(", ");
print_int(img.height);
print_string(" : ");
print_int(img.stride);
print_string(")\n");

print_string("mask: (");
print_int(mask.size_x);
print_string(", ");
print_int(mask.size_y);
print_string(")\n");
}

let mut inner_tile_size = img.width;
for i in range(0, caches.count) {
// This is a worst-case model, correct for box stencils, pessimistic for all others
// TODO take detailed shape into account
let min_loop_length = 200; // contols when loop should be blocked
let cache_blocked = ((caches.sizes(i) / (mask.size_y*element_size) / 100 * 90)>>6)<<6;
if cache_blocked < img.width && cache_blocked > 200 {
if cache_blocked < img.width && cache_blocked > min_loop_length {
inner_tile_size = cache_blocked;
}

//print_int(cache_sizes(i));
//print_string(" <= ");
//print_int(img.width*mask.size_y*element_size);
//if cache_sizes(i) <= img.width*mask.size_y*element_size {
// print_string(" = false; ");
//} else {
// print_string(" = true; ");
//}
if debug_tiling @{
print_int(caches.sizes(i));
print_string(" <= ");
print_int(img.width*mask.size_y*element_size);
if caches.sizes(i) <= img.width*mask.size_y*element_size {
print_string(" = false; ");
} else {
print_string(" = true; ");
}
}
}
if debug_tiling @{
print_string("\nRecommended inner tiling: ");
print_int(inner_tile_size);
print_string("\n");
}
//print_string("\nRecommended inner tiling: ");
//print_int(inner_tile_size);
//print_string("\n");

(inner_tile_size, img.height)
}
Expand All @@ -135,17 +143,19 @@ fn tiled_loop(xl: i32, xu: i32, yl: i32, yu: i32, mask: Mask, img: Img, body: fn
let y_lower = yl;

fn tile(cur_lvl: i32, xl: i32, xu: i32, yl: i32, yu: i32) -> () {
//print_string("tile(");
//print_int(cur_lvl);
//print_string(", ");
//print_int(xl);
//print_string(", ");
//print_int(xu);
//print_string(", ");
//print_int(yl);
//print_string(", ");
//print_int(yu);
//print_string(")\n");
if debug_tiling @{
print_string("tile(");
print_int(cur_lvl);
print_string(", ");
print_int(xl);
print_string(", ");
print_int(xu);
print_string(", ");
print_int(yl);
print_string(", ");
print_int(yu);
print_string(")\n");
}

let (xtile_dim, ytile_dim) = @get_tile_dims(mask, img);
if cur_lvl == 0 {
Expand Down