Skip to content

amdgpu: accept raw-HSACO kernels with a partial implicit kernarg suffix #59

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

Draft
zjgarvey wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

amdgpu: accept raw-HSACO kernels with a partial implicit kernarg suffix #59

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
614feef
amdgpu: accept raw-HSACO kernels with a partial implicit kernarg suffix
zjgarvey Jun 8, 2026
979fac0
test: partial implicit kernarg suffix is accepted, not rejected
zjgarvey Jun 10, 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
14 changes: 10 additions & 4 deletions runtime/src/iree/hal/drivers/amdgpu/executable.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2138,13 +2138,19 @@ iree_status_t iree_hal_amdgpu_executable_raw_hsaco_custom_kernarg_layout(
if (!iree_host_size_checked_add(
(iree_host_size_t)implicit_args_offset,
(iree_host_size_t)IREE_AMDGPU_KERNEL_IMPLICIT_ARGS_SIZE,
&implicit_args_end) ||
kernel_args->kernarg_size < implicit_args_end) {
&implicit_args_end)) {
return iree_make_status(
IREE_STATUS_INVALID_ARGUMENT,
"AMDGPU kernel `%.*s` has truncated implicit kernarg suffix",
IREE_STATUS_OUT_OF_RANGE,
"AMDGPU kernel `%.*s` implicit kernarg suffix offset overflows",
(int)symbol_name.size, symbol_name.data);
}
// NOTE: kernel_args->kernarg_size may be smaller than implicit_args_end for
// hand-written assembly kernels (e.g. MIOpen miopenSp3AsmConv*) that
// declare a partial hidden suffix. That is valid: the custom-direct
// dispatch path reserves max(kernarg_size, explicit, implicit_offset +
// IMPLICIT_SIZE) and zero-fills the remainder, so accept it rather than
// rejecting the load.
(void)implicit_args_end;
*out_implicit_args_offset = implicit_args_offset;
}
return iree_ok_status();
Expand Down
17 changes: 10 additions & 7 deletions runtime/src/iree/hal/drivers/amdgpu/executable_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -366,8 +366,12 @@ TEST(ExecutableTest, RawHsacoCustomKernargLayoutAcceptsFullImplicitSuffix) {
EXPECT_EQ(implicit_args_offset, 16u);
}

TEST(ExecutableTest,
RawHsacoCustomKernargLayoutRejectsTruncatedImplicitSuffix) {
// Hand-written assembly kernels may declare only a partial hidden suffix
// (kernarg_size < implicit offset + IREE_AMDGPU_KERNEL_IMPLICIT_ARGS_SIZE).
// The layout must accept this: dispatch reserves
// max(kernarg_size, implicit offset + implicit size) and zero-fills, so a
// short declared size is harmless.
TEST(ExecutableTest, RawHsacoCustomKernargLayoutAcceptsPartialImplicitSuffix) {
const iree_hal_amdgpu_hsaco_metadata_arg_t args[] = {
MakeArg(IREE_HAL_AMDGPU_HSACO_METADATA_ARG_KIND_GLOBAL_BUFFER,
/*offset=*/0, /*size=*/8),
Expand All @@ -381,11 +385,10 @@ TEST(ExecutableTest,

iree_host_size_t explicit_kernarg_size = 0;
uint16_t implicit_args_offset = 0;
IREE_EXPECT_STATUS_IS(
IREE_STATUS_INVALID_ARGUMENT,
iree_hal_amdgpu_executable_raw_hsaco_custom_kernarg_layout(
&kernel, &kernel_args, &explicit_kernarg_size,
&implicit_args_offset));
IREE_ASSERT_OK(iree_hal_amdgpu_executable_raw_hsaco_custom_kernarg_layout(
&kernel, &kernel_args, &explicit_kernarg_size, &implicit_args_offset));
EXPECT_EQ(explicit_kernarg_size, 8u);
EXPECT_EQ(implicit_args_offset, 16u);
}

} // namespace
Expand Down
Loading