Skip to content

[SYCL][E2E][Matrix] Use integer inputs for bfloat16 accumulator tests #23609

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
dkhaldi wants to merge 2 commits into
base: sycl
Choose a base branch
from
+19 −6
Open

[SYCL][E2E][Matrix] Use integer inputs for bfloat16 accumulator tests #23609

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3bfacb4
[SYCL][E2E][Matrix] Use integer inputs for bfloat16 accumulator tests…
dkhaldi Oct 23, 2026
a71ef05
Retrigger CI
dkhaldi Oct 29, 2026
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
22 changes: 19 additions & 3 deletions sycl/test-e2e/Matrix/Inputs/joint_matrix_16bit_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,8 +100,23 @@ void test() {
TResult D[MATRIX_M][MATRIX_N];
TResult DRef[MATRIX_M][MATRIX_N];

matrix_rand<Tab>(MATRIX_M, MATRIX_K, (Tab *)A, Tab(1));
matrix_rand<Tab>(MATRIX_K / VF, MATRIX_N * VF, (Tab *)B, Tab(1));
// A bfloat16 C or D is rounded after every DPAS while the reference rounds
// once, and with only 8 significant bits the difference can exceed the
// tolerance. Integer inputs keep every partial sum exactly representable, so
// the result does not depend on where the rounding happens.
constexpr bool bf16_acc =
std::is_same_v<TAcc, bfloat16> || std::is_same_v<TResult, bfloat16>;
if constexpr (!bf16_acc) {
matrix_rand<Tab>(MATRIX_M, MATRIX_K, (Tab *)A, Tab(1));
matrix_rand<Tab>(MATRIX_K / VF, MATRIX_N * VF, (Tab *)B, Tab(1));
} else {
std::mt19937 gen(0);
std::uniform_int_distribution<int> dist(-1, 1);
matrix_fill(MATRIX_M, MATRIX_K, (Tab *)A,
[&](int, int) { return Tab(dist(gen)); });
matrix_fill(MATRIX_K / VF, MATRIX_N * VF, (Tab *)B,
[&](int, int) { return Tab(dist(gen)); });
}

matrix_fill(MATRIX_M, MATRIX_N, (TAcc *)C, TAcc(1));
matrix_fill(MATRIX_M, MATRIX_N, (TResult *)D, TResult(1));
Expand All @@ -116,7 +131,8 @@ void test() {
B_layout, VF>(MD, MC, MA, MB);
matrix_multiply_ref<Tab, Tab, TResult, VF>(
(Tab *)A, (Tab *)B, (TResult *)DRef, MATRIX_M, MATRIX_N, MATRIX_K / VF);
assert(matrix_compare(MATRIX_M, MATRIX_N, (TResult *)D, (TResult *)DRef));
assert((matrix_compare<TResult, TResult, bf16_acc>(
MATRIX_M, MATRIX_N, (TResult *)D, (TResult *)DRef)));
}

template <typename TLow, typename THigh, size_t TM, size_t TN, size_t TK,
Expand Down
3 changes: 0 additions & 3 deletions sycl/test-e2e/Matrix/joint_matrix_bfloat16_accumulator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,9 +20,6 @@
// XFAIL: windows && arch-intel_gpu_lnl_m && O0
// XFAIL-TRACKER: CMPLRLLVM-72111

// UNSUPPORTED: linux && (arch-intel_gpu_pvc || arch-intel_gpu_bmg_g21 || arch-intel_gpu_ptl_u || arch-intel_gpu_ptl_h)
// UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/21096

// RUN: %{build} -o %t.out
// RUN: %{run} %t.out
// RUN: %if gpu %{ env IGC_JointMatrixLoadStoreOpt=2 %{run} %t.out %}
Expand Down
Loading