Skip to content

[CPU] add support for mamba causal conv1d for qwen3-next #12309

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
FlamingoPg merged 10 commits into from
Dec 4, 2025
Merged

[CPU] add support for mamba causal conv1d for qwen3-next #12309

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
401ca72
add casaul_conv1d support on CPU
mingfeima Oct 15, 2025
6ff4329
add support for casaul_conv1d_fn with varlen suport (contiguous batch…
mingfeima Oct 20, 2025
8baa066
add causal_conv1d_update support on CPU
mingfeima Oct 21, 2025
3fcaed1
add conv indices support for conv1d_update kernel
mingfeima Oct 22, 2025
b88b571
add support for varlen kernel with conv_states
mingfeima Oct 28, 2025
913a54e
add test cases for causal conv1d on CPU
mingfeima Nov 6, 2025
ffe3e3b
add test_causal_conv1d.py to run suite
mingfeima Nov 6, 2025
e4a88a5
change test file order according to alphabetical order
mingfeima Nov 6, 2025
41cd3ef
update format
mingfeima Nov 6, 2025
0abfa78
fix: keep sgl_kernel import for C++ ops registration
mingfeima Dec 3, 2025
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
31 changes: 31 additions & 0 deletions sgl-kernel/csrc/cpu/common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,29 @@ namespace {
} \
}()

#define AT_DISPATCH_BOOL2(BOOL_V1, BOOL_NAME1, BOOL_V2, BOOL_NAME2, ...) \
[&] { \
if (BOOL_V1) { \
constexpr bool BOOL_NAME1 = true; \
if (BOOL_V2) { \
constexpr bool BOOL_NAME2 = true; \
return __VA_ARGS__(); \
} else { \
constexpr bool BOOL_NAME2 = false; \
return __VA_ARGS__(); \
} \
} else { \
constexpr bool BOOL_NAME1 = false; \
if (BOOL_V2) { \
constexpr bool BOOL_NAME2 = true; \
return __VA_ARGS__(); \
} else { \
constexpr bool BOOL_NAME2 = false; \
return __VA_ARGS__(); \
} \
} \
}()

// dispatch: bfloat16, float16, int8_t, fp8_e4m3
#define CPU_DISPATCH_PACKED_TYPES(TYPE, ...) \
[&] { \
Expand Down Expand Up @@ -105,6 +128,8 @@ namespace {

#define CHECK_EQ(a, b) TORCH_CHECK((a) == (b), "CHECK_EQ(" #a ", " #b ") failed. ", a, " vs ", b)

#define CHECK_GE(a, b) TORCH_CHECK((a) >= (b), "CHECK_GE(" #a ", " #b ") failed. ", a, " vs ", b)

// [NB] Parallel Routines
//
// * at::parallel_for - applies for most of generic use cases, this will be compiled
Expand Down Expand Up @@ -321,4 +346,10 @@ struct Unroll<1> {
}
};

// conditional data ptr for optional tensor
template <typename T>
inline T* conditional_data_ptr(const std::optional<at::Tensor>& opt) {
return opt.has_value() ? opt.value().data_ptr<T>() : nullptr;
}

} // anonymous namespace
Loading
Loading