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

[selection/codegen] bugfix in distributed tile indices initialization

Browse Source
This commit is contained in:
Philippe Tillet
2019-06-11 10:45:19 -07:00
parent 06b5992509
commit 7d50b87681
2 changed files with 7 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
lib/codegen/selection.cpp
View File

@@ -29,7 +29,8 @@ void distributed_tile::init_indices() {
indices_t current; indices_t current;
for(size_t d = 0; d < id.size(); d++) for(size_t d = 0; d < id.size(); d++)
current.push_back(axes_[d].values[id[d]]); current.push_back(axes_[d].values[id[d]]);
indices_[current] = indices_.size(); size_t sz = indices_.size();
indices_[current] = sz;
values_[current] = UndefValue::get(ty_); values_[current] = UndefValue::get(ty_);
ordered_indices_.push_back(current); ordered_indices_.push_back(current);
id[0]++; id[0]++;
@@ -840,10 +841,6 @@ void selection::lower_tile_instruction(ir::instruction *ins, llvm::IRBuilder<> &
unsigned id = linear / vector_size; unsigned id = linear / vector_size;
if(linear % vector_size == 0) if(linear % vector_size == 0)
packets[id] = result->get_value(idx); packets[id] = result->get_value(idx);
});
in->for_each([&](indices_t idx){
unsigned linear = in->get_linear_index(idx);
unsigned id = linear / vector_size;
packets[id] = builder.CreateInsertElement(packets.at(id), in->get_value(idx), linear % vector_size); packets[id] = builder.CreateInsertElement(packets.at(id), in->get_value(idx), linear % vector_size);
}); });
result->for_each([&](indices_t idx){ result->for_each([&](indices_t idx){

8
lib/driver/module.cpp
View File

@@ -243,10 +243,12 @@ std::string cu_module::compile_llvm_module(llvm::Module* module) {
layout += "-i64:64-i128:128-v16:16-v32:32-n16:32:64"; layout += "-i64:64-i128:128-v16:16-v32:32-n16:32:64";
// create // create
llvm::SmallVector<char, 0> buffer; llvm::SmallVector<char, 0> buffer;
module::compile_llvm_module(module, "nvptx64-nvidia-cuda", "sm_75", layout, buffer, "", Assembly); module::compile_llvm_module(module, "nvptx64-nvidia-cuda", "sm_70", layout, buffer, "", Assembly);
std::string result(buffer.begin(), buffer.end()); std::string result(buffer.begin(), buffer.end());
std::string to_replace = ".version 6.3"; size_t start_replace = result.find(".version");
result.replace(result.find(to_replace), to_replace.size(), ".version 6.4"); size_t end_replace = result.find('\n', start_replace);
assert(start_replace != std::string::npos);
result.replace(start_replace, end_replace - start_replace, ".version 6.4");
return result; return result;
} }