Skip to content

[metax] modify kernels #74

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
metax666 merged 31 commits into from
Sep 29, 2025
Merged

[metax] modify kernels #74

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
31 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
93f75b7
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 24, 2025
e7278d0
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 26, 2025
36aab9d
modify kernels
jxwangmetax Sep 26, 2025
01586f9
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 28, 2025
3531ed0
modify kernels
jxwangmetax Sep 29, 2025
f0562bb
Merge branch 'develop' into develop
jxwangmetax Sep 29, 2025
f9ab930
Merge branch 'metax666:develop' into develop
jxwangmetax Sep 29, 2025
5ccb3ae
modify kernels
jxwangmetax Sep 29, 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
37 changes: 16 additions & 21 deletions backends/metax_gpu/kernels/gpudnn/conv_grad_kernel_register.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -437,26 +437,22 @@ void ConvCudnnGradKernel(const Context& dev_ctx,
dev_ctx.template Alloc<T>(filter_grad);
}

// bool has_use_addto = dev_ctx.HasDnnAttr("use_addto");
bool has_use_addto = "true";
bool has_use_addto = dev_ctx.HasDnnAttr("use_addto");
VLOG(4) << "GPUContext contains `use_addto`: " << has_use_addto;
// bool use_addto = has_use_addto
// ? PADDLE_GET_CONST(bool, "true")
// : false;
bool use_addto = "true";
bool use_addto = has_use_addto
? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("use_addto"))
: false;
std::vector<int> dilations = dilations_t;
std::vector<int> strides = strides_t;
std::vector<int> paddings = paddings_t;

// bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
bool has_exhaustive_search = "true";
bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
VLOG(4) << "GPUContext contains `exhaustive_search`: "
<< has_exhaustive_search;
// bool exhaustive_search_attr =
// has_exhaustive_search
// ? PADDLE_GET_CONST(bool, "true")
// : false;
bool exhaustive_search_attr = "true";
bool exhaustive_search_attr =
has_exhaustive_search
? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("exhaustive_search"))
: false;
bool exhaustive_search =
FLAGS_cudnn_exhaustive_search || exhaustive_search_attr;
bool deterministic = FLAGS_cudnn_deterministic;
Expand Down Expand Up @@ -835,14 +831,13 @@ void ConvCudnnGradGradKernel(
T* transformed_dx = nullptr;
std::vector<int> dilations = dilations_t;

// bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
// VLOG(4) << "GPUContext contains `exhaustive_search`: "
// << has_exhaustive_search;
// bool exhaustive_search_attr =
// has_exhaustive_search
// ? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("exhaustive_search"))
// : false;
bool exhaustive_search_attr = "true";
bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
VLOG(4) << "GPUContext contains `exhaustive_search`: "
<< has_exhaustive_search;
bool exhaustive_search_attr =
has_exhaustive_search
? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("exhaustive_search"))
: false;
bool exhaustive_search =
FLAGS_cudnn_exhaustive_search || exhaustive_search_attr;
bool deterministic = FLAGS_cudnn_deterministic;
Expand Down
19 changes: 10 additions & 9 deletions backends/metax_gpu/kernels/gpudnn/conv_kernel_register.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -228,15 +228,16 @@ void ConvCudnnKernel(const Context& dev_ctx,
std::vector<int> paddings = paddings_t;
std::vector<int> dilations = dilations_t;

// bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
// VLOG(4) << "GPUContext contains `exhaustive_search`: "
// << has_exhaustive_search;
// bool exhaustive_search_attr =
// has_exhaustive_search
// ? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("exhaustive_search"))
// : false;

bool exhaustive_search = FLAGS_cudnn_exhaustive_search;
bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
VLOG(4) << "GPUContext contains `exhaustive_search`: "
<< has_exhaustive_search;
bool exhaustive_search_attr =
has_exhaustive_search
? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("exhaustive_search"))
: false;

bool exhaustive_search =
FLAGS_cudnn_exhaustive_search || exhaustive_search_attr;
bool deterministic = FLAGS_cudnn_deterministic;

PADDLE_ENFORCE_EQ(exhaustive_search && deterministic,
Expand Down
15 changes: 7 additions & 8 deletions backends/metax_gpu/kernels/gpudnn/conv_transpose_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -260,14 +260,13 @@ void ConvTransposeRawGPUDNNKernel(const Context& dev_ctx,
return;
}

// bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
// bool exhaustive_search_attr =
// has_exhaustive_search
// ? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("exhaustive_search"))
// : false;
// bool exhaustive_search =
// FLAGS_cudnn_exhaustive_search || exhaustive_search_attr;
bool exhaustive_search = FLAGS_cudnn_exhaustive_search;
bool has_exhaustive_search = dev_ctx.HasDnnAttr("exhaustive_search");
bool exhaustive_search_attr =
has_exhaustive_search
? PADDLE_GET_CONST(bool, dev_ctx.GetDnnAttr("exhaustive_search"))
: false;
bool exhaustive_search =
FLAGS_cudnn_exhaustive_search || exhaustive_search_attr;

bool deterministic = FLAGS_cudnn_deterministic;
PADDLE_ENFORCE_EQ(exhaustive_search && deterministic,
Expand Down
14 changes: 6 additions & 8 deletions backends/metax_gpu/kernels/metax_kernel/depthwise_conv_grad_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,14 +54,12 @@ void DepthwiseConvGradKernel(const Context& dev_ctx,
return;
}

// bool has_fuse_relu = dev_ctx.HasDnnAttr("fuse_relu_before_depthwise_conv");
// bool fuse_relu =
// has_fuse_relu
// ? PADDLE_GET_CONST(
// bool, dev_ctx.GetDnnAttr("fuse_relu_before_depthwise_conv"))
// : false;
bool has_fuse_relu = false;
bool fuse_relu = false;
bool has_fuse_relu = dev_ctx.HasDnnAttr("fuse_relu_before_depthwise_conv");
bool fuse_relu =
has_fuse_relu
? PADDLE_GET_CONST(
bool, dev_ctx.GetDnnAttr("fuse_relu_before_depthwise_conv"))
: false;

std::vector<int> strides = strides_t;
std::vector<int> paddings = paddings_t;
Expand Down
14 changes: 6 additions & 8 deletions backends/metax_gpu/kernels/metax_kernel/depthwise_conv_kernel.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,14 +48,12 @@ void DepthwiseConvKernel(const Context& dev_ctx,

const bool channel_last = (data_format == "NHWC" || data_format == "NDHWC");

// bool has_fuse_relu = dev_ctx.HasDnnAttr("fuse_relu_before_depthwise_conv");
// bool fuse_relu =
// has_fuse_relu
// ? PADDLE_GET_CONST(
// bool, dev_ctx.GetDnnAttr("fuse_relu_before_depthwise_conv"))
// : false;
bool has_fuse_relu = false;
bool fuse_relu = false;
bool has_fuse_relu = dev_ctx.HasDnnAttr("fuse_relu_before_depthwise_conv");
bool fuse_relu =
has_fuse_relu
? PADDLE_GET_CONST(
bool, dev_ctx.GetDnnAttr("fuse_relu_before_depthwise_conv"))
: false;

if (channel_last) {
PADDLE_ENFORCE_EQ(
Expand Down
Loading