Skip to content

Fix performance regression (#5559) #5567

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

Merged
anmyachev merged 1 commit into from
Dec 1, 2025
Merged

Fix performance regression (#5559) #5567

Show file tree
Hide file tree
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
6 changes: 2 additions & 4 deletions python/test/unit/language/test_matmul.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1224,10 +1224,8 @@ def test_mxfp8_mxfp4_matmul(M, N, K, BLOCK_M, BLOCK_N, BLOCK_K, NUM_STAGES, B_TR
pytest.xfail("None scale has not been tested on XPU backend")
if not (A_DATA_TYPE == "float8e5" and B_DATA_TYPE == "float4"):
pytest.xfail(f"(A: {A_DATA_TYPE}, B: {B_DATA_TYPE}) has not been tested on XPU backend")
if (BLOCK_M, BLOCK_N,
BLOCK_K) == (128, 256,
256) and CONST_SCALE and triton.runtime.driver.active.utils.get_device_properties(
triton.runtime.driver.active.get_current_device())["max_shared_mem"] < 196608:
if ((BLOCK_M, BLOCK_N, BLOCK_K) == (128, 256, 256) and triton.runtime.driver.active.utils.get_device_properties(
triton.runtime.driver.active.get_current_device())["max_shared_mem"] < 196608):
pytest.xfail("XPU: Not enough shared memory")
if not PACK_B_ALONG_K and B_DATA_TYPE != "float4":
pytest.xfail("Pack along K can only be False for float4")
Expand Down
6 changes: 5 additions & 1 deletion third_party/intel/lib/TritonIntelGPUTransforms/RemoveLayoutConversions.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1646,7 +1646,11 @@ void LayoutRematerialization::hoistConvertDotOperand() {
void LayoutRematerialization::hoistConvertDotOperand(
ConvertLayoutOp convertOp) {
auto targetType = convertOp.getType();
// The pass is targeted to MMA dot operands

// The pass is targeted to NVidia.
auto dotEnc = dyn_cast<DotOperandEncodingAttr>(targetType.getEncoding());
if (!(dotEnc && isa<NvidiaMmaEncodingAttr>(dotEnc.getParent())))
return;

auto canBePipelined = [&](ConvertLayoutOp convertOp) {
// FIXME: Check that the parent is a for loop
Expand Down
Loading