gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[CODEGEN] Add cache modifier to tl.load (#351)

Browse Source 
* Add cache modifier to tl.load
* Add comment to cache_modifier
* Remove force_nc_cache
* Update test
This commit is contained in:
daadaada
2021-10-18 13:14:04 +08:00
committed by GitHub
parent 90ded16c32
commit 858dec8372
16 changed files with 119 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
lib/codegen/transform/peephole.cc
View File

@@ -116,7 +116,7 @@ bool peephole::rewrite_load_to_shared(ir::instruction *value, ir::builder& build
int nts = layout->nts(layout->get_order()[0]);
int dtsize = value->get_type()->get_scalar_ty()->get_primitive_size_in_bits() / 8;
if(nts*dtsize >= 4){
ir::value* new_load = builder.create_masked_load_async(ptr, msk, val);
ir::value* new_load = builder.create_masked_load_async(ptr, msk, val, ld->get_cache_modifier());
copy_to_shared->replace_all_uses_with(new_load);
return true;
}
@@ -206,7 +206,8 @@ bool peephole::rewrite_select_masked_load(ir::instruction *value, ir::builder& b
builder.set_insert_point(select);
ir::value* new_load = builder.create_masked_load(if_value->get_pointer_operand(),
if_value->get_mask_operand(),
select->get_else_value_op());
select->get_else_value_op(),
if_value->get_cache_modifier());
select->replace_all_uses_with(new_load);
return true;
}

12
lib/codegen/transform/pipeline.cc
View File

@@ -111,6 +111,8 @@ struct pipeline_info_t {
};
void pipeline::run(ir::module &mod) {
if (num_stages_ <= 1)
return;
// *Very* conservative heuristics for pre-fetching.
// A load instruction can be pipelined if:
// - the pointer is a phi node that references a value
@@ -176,7 +178,7 @@ void pipeline::run(ir::module &mod) {
false_value = remat_false_value;
} else
false_value = builder.create_splat(ir::undef_value::get(ty->get_scalar_ty()), ty->get_block_shapes());
first_loads[0] = builder.create_masked_load(first_ptrs[0], first_masks[0], false_value);
first_loads[0] = builder.create_masked_load(first_ptrs[0], first_masks[0], false_value, load->get_cache_modifier());
for (int stage = 1; stage < num_stages-1; ++stage) {
// mask is the loop condition of the previous iteration
@@ -191,7 +193,7 @@ void pipeline::run(ir::module &mod) {
first_masks[stage] = builder.create_and(first_masks[stage], remat_mask);
false_value = remat_false_value;
}
first_loads[stage] = builder.create_masked_load(first_ptrs[stage], first_masks[stage], false_value);
first_loads[stage] = builder.create_masked_load(first_ptrs[stage], first_masks[stage], false_value, load->get_cache_modifier());
}
// create new phis for induction variables
@@ -220,7 +222,7 @@ void pipeline::run(ir::module &mod) {
next_mask = builder.create_and(next_mask, remat_mask);
false_value = remat_false_value;
}
ir::value* next_load = builder.create_masked_load(next_ptr, next_mask, false_value);
ir::value* next_load = builder.create_masked_load(next_ptr, next_mask, false_value, load->get_cache_modifier());
// phi node
@@ -255,7 +257,7 @@ void pipeline::run(ir::module &mod) {
}
else
false_value = builder.create_splat(ir::undef_value::get(ty->get_scalar_ty()), ty->get_block_shapes());
ir::value* first_load = builder.create_masked_load(first_ptr, first_mask, false_value);
ir::value* first_load = builder.create_masked_load(first_ptr, first_mask, false_value, load->get_cache_modifier());
// pre-fetch next iteration
builder.set_insert_point(block->get_inst_list().back());
ir::value* next_ptr = ptr->get_value_for_block(block);
@@ -266,7 +268,7 @@ void pipeline::run(ir::module &mod) {
next_mask = builder.create_and(next_mask, remat_mask);
false_value = remat_false_value;
}
ir::value* next_load = builder.create_masked_load(next_ptr, next_mask, false_value);
ir::value* next_load = builder.create_masked_load(next_ptr, next_mask, false_value, load->get_cache_modifier());
// phi node
builder.set_insert_point(block->get_first_non_phi());
ir::phi_node* new_load = builder.create_phi(ty, 2);