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

[codegen/selection] tensor cores now used for transposed layotus

Browse Source
This commit is contained in:
Philippe Tillet
2019-07-17 17:20:38 -07:00
parent bfa39b8992
commit 2f0817b2cd
5 changed files with 57 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
lib/codegen/shmem_allocation.cpp
View File

@@ -15,9 +15,22 @@ unsigned shmem_allocation::is_ld_padded(ir::value *x) {
if(dynamic_cast<ir::trans_inst*>(x))
return 4;
for(ir::user* user: x->get_users())
if(dynamic_cast<ir::dot_inst*>(user))
if(params_->get_fragment(user, 0) == tune::HMMA_FRAGMENT_C){
return 16;
if(auto dot = dynamic_cast<ir::dot_inst*>(user)){
bool is_hmma = params_->get_fragment(user, 0) == tune::HMMA_FRAGMENT_C;
bool is_op_0 = x == dot->get_operand(0);
bool is_op_1 = x == dot->get_operand(1);
if(is_hmma && is_op_0){
if(dot->is_a_trans())
return 20;
else
return 16;
}
if(is_hmma && is_op_1){
if(!dot->is_b_trans())
return 20;
else
return 16;
}
}
if(auto* phi = dynamic_cast<ir::phi_node*>(x)) {
unsigned result = 0;