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

[FRONTEND] Added volatile flag for load (#407)

Browse Source
This commit is contained in:
Philippe Tillet
2021-12-30 22:33:24 -08:00
committed by GitHub
parent 985798f101
commit 03f1256f60
11 changed files with 51 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
lib/codegen/selection/generator.cc
View File

@@ -629,7 +629,11 @@ void generator::visit_load_inst(ir::load_inst* x){
// -----
std::ostringstream asm_oss;
asm_oss << "@$" << n_words; // predicate
asm_oss << " ld.global";
asm_oss << " ld";
// std::cout << x->get_is_volatile() << std::endl;
if(x->get_is_volatile())
asm_oss << ".volatile";
asm_oss << ".global";
if (x->get_cache_modifier() == ir::load_inst::CA) asm_oss << ".ca";
if (x->get_cache_modifier() == ir::load_inst::CG) asm_oss << ".cg";
if(n_words > 1)

3
lib/codegen/transform/peephole.cc
View File

@@ -207,7 +207,8 @@ bool peephole::rewrite_select_masked_load(ir::instruction *value, ir::builder& b
ir::value* new_load = builder.create_masked_load(if_value->get_pointer_operand(),
if_value->get_mask_operand(),
select->get_else_value_op(),
if_value->get_cache_modifier());
if_value->get_cache_modifier(),
if_value->get_is_volatile());
select->replace_all_uses_with(new_load);
return true;
}

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

@@ -178,7 +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, load->get_cache_modifier());
first_loads[0] = builder.create_masked_load(first_ptrs[0], first_masks[0], false_value, load->get_cache_modifier(), load->get_is_volatile());
for (int stage = 1; stage < num_stages-1; ++stage) {
// mask is the loop condition of the previous iteration
@@ -193,7 +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, load->get_cache_modifier());
first_loads[stage] = builder.create_masked_load(first_ptrs[stage], first_masks[stage], false_value, load->get_cache_modifier(), load->get_is_volatile());
}
// create new phis for induction variables
@@ -222,7 +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, load->get_cache_modifier());
ir::value* next_load = builder.create_masked_load(next_ptr, next_mask, false_value, load->get_cache_modifier(), load->get_is_volatile());
// phi node
@@ -257,7 +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, load->get_cache_modifier());
ir::value* first_load = builder.create_masked_load(first_ptr, first_mask, false_value, load->get_cache_modifier(), load->get_is_volatile());
// pre-fetch next iteration
builder.set_insert_point(block->get_inst_list().back());
ir::value* next_ptr = ptr->get_value_for_block(block);
@@ -268,7 +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, load->get_cache_modifier());
ir::value* next_load = builder.create_masked_load(next_ptr, next_mask, false_value, load->get_cache_modifier(), load->get_is_volatile());
// phi node
builder.set_insert_point(block->get_first_non_phi());
ir::phi_node* new_load = builder.create_phi(ty, 2);
@@ -327,4 +327,4 @@ void pipeline::run(ir::module &mod) {
}
}
}
}