Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
23 commits
Select commit Hold shift + click to select a range
51da6b7
modify cmake for warpctc and warprnnt
jxwangmetax Sep 16, 2025
b60cba2
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 16, 2025
1abea54
modify conv for tf32 and fp32
jxwangmetax Sep 16, 2025
fc63171
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 16, 2025
f26987f
modify conv kernel
jxwangmetax Sep 16, 2025
a0cb0a7
modify library to static library
jxwangmetax Sep 16, 2025
d6579b8
Merge branch 'develop' into develop
jxwangmetax Sep 16, 2025
9a9372b
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 16, 2025
9cd1f2e
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 16, 2025
fb902f4
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 17, 2025
7b018df
modify kernel
jxwangmetax Sep 17, 2025
e61cf0d
modify fused_bias_dropout_residual_layer_norm
jxwangmetax Sep 17, 2025
0e5be1a
Merge branch 'develop' into develop
jxwangmetax Sep 17, 2025
2757fb7
modify compile
jxwangmetax Sep 18, 2025
74a263a
Merge branch 'develop' of https://github.com/jxwangmetax/PaddleCustom…
jxwangmetax Sep 18, 2025
88d5eae
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 19, 2025
b2b41c2
modify blas
jxwangmetax Sep 19, 2025
6ea2fab
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 22, 2025
6556cce
modify blas
jxwangmetax Sep 22, 2025
1cbe0d8
modify blas
jxwangmetax Sep 22, 2025
554b3cb
modify blas
jxwangmetax Sep 22, 2025
dfac884
modify context
jxwangmetax Sep 22, 2025
5845b41
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 22, 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
18 changes: 18 additions & 0 deletions backends/metax_gpu/kernels/metax_kernel/metax_context.cc
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,24 @@
#include "kernels/metax_kernel/metax_context.h"

namespace phi {
const bool allow_tf32_cublas = []() -> bool {
const char* v = std::getenv("ALLOW_TF32_CUBLAS");
if (v) {
return std::atoi(v);
}
return true;
}();

const bool allow_tf32_cudnn = []() -> bool {
const char* v = std::getenv("ALLOW_TF32_CUDNN");
if (v) {
return std::atoi(v);
}
return false;
}();

bool AllowTF32Cublas() { return allow_tf32_cublas; }
bool AllowTF32Cudnn() { return allow_tf32_cudnn; }
void DnnWorkspaceHandle::RunFuncSync(
const std::function<void(void*)>& cudnn_func,
size_t required_workspace_bytes,
Expand Down
2 changes: 2 additions & 0 deletions backends/metax_gpu/kernels/metax_kernel/metax_context.h
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,8 @@
cublasLtHandle_t GetBlasLtHandle();

namespace phi {
bool AllowTF32Cublas();
bool AllowTF32Cudnn();
class DnnWorkspaceHandle {
public:
inline DnnWorkspaceHandle(Allocator* allocator, gpuStream_t stream)
Expand Down
Loading