Skip to content

[AMDGPU] Keep fdiv correctly rounded under fast_math (fixes modulo bug #749) #773

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

Open
ZJLi2013 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[AMDGPU] Keep fdiv correctly rounded under fast_math (fixes modulo bug #749) #773

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 10 additions & 0 deletions quadrants/codegen/llvm/codegen_llvm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -572,6 +572,16 @@ void TaskCodeGenLLVM::visit(BinaryOpStmt *stmt) {
}
} else if (op == BinaryOpType::div) {
if (is_real(stmt->ret_type.get_element_type())) {
// Under fast_math the builder carries `afn` (ApproxFunc). On AMDGPU that
// makes fdiv lower to an approximate v_rcp_f32 (~2.5 ULP), so exact cases
// like 180.0/180.0 return 0.99999994 and silently break floor(a/b) and the
// `a - floor(a/b)*b` modulo expansion (quadrants#749). Emit the division
// with afn cleared so it stays correctly rounded; `afn` is meant for
// transcendentals, not basic arithmetic. No-op on CPU/CUDA.
llvm::IRBuilderBase::FastMathFlagGuard fmf_guard(*builder);
auto fmf = builder->getFastMathFlags();
fmf.setApproxFunc(false);
builder->setFastMathFlags(fmf);
llvm_val[stmt] = builder->CreateFDiv(llvm_val[stmt->lhs], llvm_val[stmt->rhs]);
} else if (is_signed(stmt->ret_type.get_element_type())) {
llvm_val[stmt] = builder->CreateSDiv(llvm_val[stmt->lhs], llvm_val[stmt->rhs]);
Expand Down