Skip to content

Enable fp8 on sm89 #3624

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
jjsjann123 merged 4 commits into from
Dec 21, 2024
Merged

Enable fp8 on sm89 #3624

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
ee3060b
smoke test
jjsjann123 Dec 19, 2024
39b7773
fixing check and tests
jjsjann123 Dec 20, 2024
44e949b
why did we print that we skip earlier?
jjsjann123 Dec 20, 2024
48e620f
Merge branch 'main' into fp8_enable_on_sm89
jjsjann123 Dec 20, 2024
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
17 changes: 16 additions & 1 deletion csrc/device_lower/analysis/device_version.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,8 @@
* SPDX-License-Identifier: BSD-3-Clause
*/
// clang-format on
#include <cuda.h>

#include <device_lower/analysis/device_version.h>
#include <device_lower/lower2device.h>
#include <mma_type.h>
Expand All @@ -19,9 +21,22 @@ void MinimumDeviceVersion::dispatch(Val* val) {
}
if (val->dtype() == DataType::Float8_e4m3fn ||
val->dtype() == DataType::Float8_e5m2) {
// See release note
// https://docs.nvidia.com/cuda/archive/12.1.0/parallel-thread-execution/index.html#ptx-isa-version-8-1
#if (CUDA_VERSION >= 12010)
ensureVersion(
{9, 0},
{8, 9},
"Fusion contains Float8_xxx values which was introduced in Ada (8.9)");
// See release note
// https://docs.nvidia.com/cuda/archive/11.8.0/parallel-thread-execution/index.html#ptx-isa-version-7-8
#elif (CUDA_VERSION >= 11080)
ensureVersion(
{8, 9},
"Fusion contains Float8_xxx values which was introduced in Hopper (9.0)");
#else
NVF_ERROR(
"Fusion contains Float8_xxx values which was not supported in given CUDA version");
#endif // (CUDA_VERSION >= 12010)
}
IterVisitor::dispatch(val);
}
Expand Down
10 changes: 7 additions & 3 deletions tests/cpp/test_gpu1.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2711,13 +2711,17 @@ TEST_F(NVFuserTest, FusionFp8CastOps_CUDA) {
std::vector<c10::IValue> inputs = {input1};

KernelExecutor ke;

#if (CUDA_VERSION >= 12010)
if (!deviceMajorMinorCheck(8, 9)) {
#elif (CUDA_VERSION >= 11080)
if (!deviceMajorMinorCheck(9)) {
#else
if (true) {
#endif
ASSERT_THAT(
[&]() { ke.compile(&fusion, inputs); },
testing::ThrowsMessage<nvfuser::nvfError>(testing::HasSubstr(
"Reason: Fusion contains Float8_xxx values which was introduced in Hopper (9.0)")));
GTEST_SKIP() << "skipping tests on pre-HOPPER GPUs";
"Reason: Fusion contains Float8_xxx values")));
} else {
ke.compile(&fusion, inputs);
auto outputs = ke.run(inputs);
Expand Down