Skip to content

GitLab

Commit 83181423 authored by hubertlu-tw's avatar hubertlu-tw
Browse files

Hipify self_multihead_attn 

Enable HIP floa to hald conversion
parent 61416180
Hide whitespace changes
Inline Side-by-side
Showing with 346 additions and 51 deletions
apex/contrib/csrc/multihead_attn/encdec_multihead_attn_cuda.cu
View file @ 83181423
#include <vector> #include <vector>
#include <iostream> #include <iostream>
//below lines enable hip float to half conversion which are disabled by default in hip_fp16.h
#undef __HIP_NO_HALF_OPERATORS__
#undef __HIP_NO_HALF_CONVERSIONS__
//#endif
#include <ATen/ATen.h> #include <ATen/ATen.h>
#include <ATen/cuda/CUDAContext.h> #include <ATen/cuda/CUDAContext.h>
#include <cuda.h> #include <cuda.h>
......
apex/contrib/csrc/multihead_attn/self_multihead_attn_bias_additive_mask_cuda.cu
View file @ 83181423
#include <vector> #include <vector>
#include <math.h> #include <math.h>
#include <iostream> #include <iostream>
//below lines enable hip float to half conversion which are disabled by default in hip_fp16.h
#undef __HIP_NO_HALF_OPERATORS__
#undef __HIP_NO_HALF_CONVERSIONS__
//#endif
#include <cuda.h> #include <cuda.h>
#include <cuda_runtime.h> #include <cuda_runtime.h>
#include <cuda_fp16.h> #include <cuda_fp16.h>
......
apex/contrib/csrc/multihead_attn/self_multihead_attn_bias_cuda.cu
View file @ 83181423
#include <vector> #include <vector>
#include <iostream> #include <iostream>
//below lines enable hip float to half conversion which are disabled by default in hip_fp16.h
#undef __HIP_NO_HALF_OPERATORS__
#undef __HIP_NO_HALF_CONVERSIONS__
//#endif
#include <ATen/ATen.h> #include <ATen/ATen.h>
#include <cuda.h> #include <cuda.h>
#include <cuda_runtime.h> #include <cuda_runtime.h>
......
apex/contrib/csrc/multihead_attn/self_multihead_attn_cpp.cpp
View file @ 83181423
...@@ -3,7 +3,7 @@ ...@@ -3,7 +3,7 @@
namespace multihead_attn { namespace multihead_attn {
namespace self { namespace self {
namespace cublas_gemmex { namespace rocblas_gemm_ex {
std::vector<torch::Tensor> fwd_cuda( std::vector<torch::Tensor> fwd_cuda(
bool use_time_mask, bool use_time_mask,
...@@ -121,12 +121,12 @@ std::vector<torch::Tensor> bwd( ...@@ -121,12 +121,12 @@ std::vector<torch::Tensor> bwd(
); );
} }
} // end namespace cublas_gemmex } // end namespace rocblas_gemm_ex
} // end namespace self } // end namespace self
} // end namespace multihead_attn } // end namespace multihead_attn
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) { PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("forward", &multihead_attn::self::cublas_gemmex::fwd, "Self Multihead Attention Forward."); m.def("forward", &multihead_attn::self::rocblas_gemm_ex::fwd, "Self Multihead Attention Forward.");
m.def("backward", &multihead_attn::self::cublas_gemmex::bwd, "Self Multihead Attention Backward."); m.def("backward", &multihead_attn::self::rocblas_gemm_ex::bwd, "Self Multihead Attention Backward.");
} }
apex/contrib/csrc/multihead_attn/self_multihead_attn_cuda.cu
View file @ 83181423
#include <vector> #include <vector>
#include <iostream> #include <iostream>
//below lines enable hip float to half conversion which are disabled by default in hip_fp16.h
#undef __HIP_NO_HALF_OPERATORS__
#undef __HIP_NO_HALF_CONVERSIONS__
//#endif
#include <ATen/ATen.h> #include <ATen/ATen.h>
#include <cuda.h> #include <cuda.h>
#include <cuda_runtime.h> #include <cuda_runtime.h>
...@@ -77,10 +80,11 @@ std::vector<torch::Tensor> fwd_cuda( ...@@ -77,10 +80,11 @@ std::vector<torch::Tensor> fwd_cuda(
char a_layout_t{'t'}; char a_layout_t{'t'};
char a_layout_n{'n'}; char a_layout_n{'n'};
char b_layout_n{'n'}; char b_layout_n{'n'};
// TODO (OK)
THCublasCheck(cublasSetMathMode(handle, CUBLAS_TENSOR_OP_MATH)); // THCublasCheck(cublasSetMathMode(handle, CUBLAS_TENSOR_OP_MATH));
// Input Linear Fwd // Input Linear Fwd
THCublasCheck(cublasGemmEx(handle, // TODO (OK)
THCublasCheck(rocblas_gemm_ex(handle,
CUBLAS_OP_T, CUBLAS_OP_T,
CUBLAS_OP_N, CUBLAS_OP_N,
output_lin_dim, output_lin_dim,
...@@ -88,19 +92,44 @@ std::vector<torch::Tensor> fwd_cuda( ...@@ -88,19 +92,44 @@ std::vector<torch::Tensor> fwd_cuda(
embed_dim, embed_dim,
static_cast<const void*>(&alpha), static_cast<const void*>(&alpha),
static_cast<const void*>(input_weights.data_ptr()), static_cast<const void*>(input_weights.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(inputs.data_ptr()), static_cast<const void*>(inputs.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(&beta), static_cast<const void*>(&beta),
q_lin_results_ptr, q_lin_results_ptr,
CUDA_R_16F, rocblas_datatype_f16_r,
output_lin_dim, output_lin_dim,
CUDA_R_32F, q_lin_results_ptr,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); rocblas_datatype_f16_r,
output_lin_dim,
rocblas_datatype_f32_r,
algo,
solution_index,
flags));
// THCublasCheck(cublasGemmEx(handle,
// CUBLAS_OP_T,
// CUBLAS_OP_N,
// output_lin_dim,
// batches,
// embed_dim,
// static_cast<const void*>(&alpha),
// static_cast<const void*>(input_weights.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(inputs.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(&beta),
// q_lin_results_ptr,
// CUDA_R_16F,
// output_lin_dim,
// CUDA_R_32F,
// CUBLAS_GEMM_DEFAULT_TENSOR_OP));
// MatMul1 of Dot-Product Attention Plus scaling by 1/Sqrt(head size) // MatMul1 of Dot-Product Attention Plus scaling by 1/Sqrt(head size)
// TODO (OK)
gemm_switch_fp32accum( state, gemm_switch_fp32accum( state,
a_layout_t, a_layout_t,
b_layout_n, b_layout_n,
...@@ -118,7 +147,28 @@ std::vector<torch::Tensor> fwd_cuda( ...@@ -118,7 +147,28 @@ std::vector<torch::Tensor> fwd_cuda(
static_cast<half*>(softmax_results_ptr), static_cast<half*>(softmax_results_ptr),
k_seq_len, k_seq_len,
k_seq_len*q_seq_len, k_seq_len*q_seq_len,
static_cast<half*>(softmax_results_ptr),
k_seq_len,
k_seq_len*q_seq_len,
attn_batches); attn_batches);
// gemm_switch_fp32accum( state,
// a_layout_t,
// b_layout_n,
// k_seq_len,
// q_seq_len,
// head_dim,
// scale,
// static_cast<const half*>(k_lin_results_ptr),
// lead_dim,
// batch_stride,
// static_cast<const half*>(q_lin_results_ptr),
// lead_dim,
// batch_stride,
// beta,
// static_cast<half*>(softmax_results_ptr),
// k_seq_len,
// k_seq_len*q_seq_len,
// attn_batches);
// Padded Softmax // Padded Softmax
bool softmax_success = false; bool softmax_success = false;
...@@ -162,6 +212,7 @@ std::vector<torch::Tensor> fwd_cuda( ...@@ -162,6 +212,7 @@ std::vector<torch::Tensor> fwd_cuda(
} }
// Matmul2 // Matmul2
// TODO (OK)
gemm_switch_fp32accum( state, gemm_switch_fp32accum( state,
a_layout_n, a_layout_n,
b_layout_n, b_layout_n,
...@@ -179,10 +230,32 @@ std::vector<torch::Tensor> fwd_cuda( ...@@ -179,10 +230,32 @@ std::vector<torch::Tensor> fwd_cuda(
static_cast<half*>(matmul2_results.data_ptr()), static_cast<half*>(matmul2_results.data_ptr()),
head_dim*attn_batches, head_dim*attn_batches,
head_dim, head_dim,
static_cast<half*>(matmul2_results.data_ptr()),
head_dim*attn_batches,
head_dim,
attn_batches); attn_batches);
// gemm_switch_fp32accum( state,
// a_layout_n,
// b_layout_n,
// head_dim,
// q_seq_len,
// k_seq_len,
// alpha,
// static_cast<const half*>(v_lin_results_ptr),
// lead_dim,
// batch_stride,
// (is_training) ? static_cast<const half*>(dropout_results.data_ptr()) : static_cast<const half*>(softmax_results.data_ptr()) ,
// k_seq_len,
// k_seq_len*q_seq_len,
// beta,
// static_cast<half*>(matmul2_results.data_ptr()),
// head_dim*attn_batches,
// head_dim,
// attn_batches);
// Output Linear // Output Linear
THCublasCheck(cublasGemmEx(handle, // TODO
THCublasCheck(rocblas_gemm_ex(handle,
CUBLAS_OP_T, CUBLAS_OP_T,
CUBLAS_OP_N, CUBLAS_OP_N,
embed_dim, embed_dim,
...@@ -190,19 +263,43 @@ std::vector<torch::Tensor> fwd_cuda( ...@@ -190,19 +263,43 @@ std::vector<torch::Tensor> fwd_cuda(
embed_dim, embed_dim,
static_cast<const void*>(&alpha), static_cast<const void*>(&alpha),
static_cast<const void*>(output_weights.data_ptr()), static_cast<const void*>(output_weights.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(matmul2_results.data_ptr()), static_cast<const void*>(matmul2_results.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(&beta), static_cast<const void*>(&beta),
static_cast<void*>(outputs.data_ptr()), static_cast<void*>(outputs.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
CUDA_R_32F, static_cast<void*>(outputs.data_ptr()),
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); rocblas_datatype_f16_r,
embed_dim,
THCublasCheck(cublasSetMathMode(handle, CUBLAS_DEFAULT_MATH)); rocblas_datatype_f32_r,
algo,
solution_index,
flags));
// THCublasCheck(cublasGemmEx(handle,
// CUBLAS_OP_T,
// CUBLAS_OP_N,
// embed_dim,
// batches,
// embed_dim,
// static_cast<const void*>(&alpha),
// static_cast<const void*>(output_weights.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(matmul2_results.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(&beta),
// static_cast<void*>(outputs.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// CUDA_R_32F,
// CUBLAS_GEMM_DEFAULT_TENSOR_OP));
// TODO (OK)
// THCublasCheck(cublasSetMathMode(handle, CUBLAS_DEFAULT_MATH));
return { return {
input_lin_results, input_lin_results,
...@@ -270,11 +367,12 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -270,11 +367,12 @@ std::vector<torch::Tensor> bwd_cuda(
char a_layout_t{'t'}; char a_layout_t{'t'};
char b_layout_n{'n'}; char b_layout_n{'n'};
char b_layout_t{'t'}; char b_layout_t{'t'};
// TODO (OK)
THCublasCheck(cublasSetMathMode(handle, CUBLAS_TENSOR_OP_MATH)); // THCublasCheck(cublasSetMathMode(handle, CUBLAS_TENSOR_OP_MATH));
// Output Linear Dgrad // Output Linear Dgrad
THCublasCheck(cublasGemmEx(handle, // TODO (OK)
THCublasCheck(rocblas_gemm_ex(handle,
CUBLAS_OP_N, CUBLAS_OP_N,
CUBLAS_OP_N, CUBLAS_OP_N,
embed_dim, embed_dim,
...@@ -282,20 +380,45 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -282,20 +380,45 @@ std::vector<torch::Tensor> bwd_cuda(
embed_dim, embed_dim,
static_cast<const void*>(&alpha), static_cast<const void*>(&alpha),
static_cast<const void*>(output_weights.data_ptr()), static_cast<const void*>(output_weights.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(output_grads.data_ptr()), static_cast<const void*>(output_grads.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(&beta), static_cast<const void*>(&beta),
static_cast<void*>(output_lin_grads.data_ptr()), static_cast<void*>(output_lin_grads.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim,
static_cast<void*>(output_lin_grads.data_ptr()),
rocblas_datatype_f16_r,
embed_dim, embed_dim,
CUDA_R_32F, rocblas_datatype_f32_r,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); algo,
solution_index,
flags));
// THCublasCheck(cublasGemmEx(handle,
// CUBLAS_OP_N,
// CUBLAS_OP_N,
// embed_dim,
// batches,
// embed_dim,
// static_cast<const void*>(&alpha),
// static_cast<const void*>(output_weights.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(output_grads.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(&beta),
// static_cast<void*>(output_lin_grads.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// CUDA_R_32F,
// CUBLAS_GEMM_DEFAULT_TENSOR_OP));
// Output Linear Wgrad // Output Linear Wgrad
THCublasCheck(cublasGemmEx(handle, // TODO (OOK)
THCublasCheck(rocblas_gemm_ex(handle,
CUBLAS_OP_N, CUBLAS_OP_N,
CUBLAS_OP_T, CUBLAS_OP_T,
embed_dim, embed_dim,
...@@ -303,19 +426,44 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -303,19 +426,44 @@ std::vector<torch::Tensor> bwd_cuda(
batches, batches,
static_cast<const void*>(&alpha), static_cast<const void*>(&alpha),
static_cast<const void*>(matmul2_results.data_ptr()), static_cast<const void*>(matmul2_results.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(output_grads.data_ptr()), static_cast<const void*>(output_grads.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(&beta), static_cast<const void*>(&beta),
static_cast<void*>(output_weight_grads.data_ptr()), static_cast<void*>(output_weight_grads.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim,
static_cast<void*>(output_weight_grads.data_ptr()),
rocblas_datatype_f16_r,
embed_dim, embed_dim,
CUDA_R_32F, rocblas_datatype_f32_r,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); algo,
solution_index,
flags));
// THCublasCheck(cublasGemmEx(handle,
// CUBLAS_OP_N,
// CUBLAS_OP_T,
// embed_dim,
// embed_dim,
// batches,
// static_cast<const void*>(&alpha),
// static_cast<const void*>(matmul2_results.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(output_grads.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(&beta),
// static_cast<void*>(output_weight_grads.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// CUDA_R_32F,
// CUBLAS_GEMM_DEFAULT_TENSOR_OP));
// MatMul2 Dgrad1 // MatMul2 Dgrad1
// TODO (OK)
gemm_switch_fp32accum( state, gemm_switch_fp32accum( state,
a_layout_t, a_layout_t,
b_layout_n, b_layout_n,
...@@ -333,9 +481,31 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -333,9 +481,31 @@ std::vector<torch::Tensor> bwd_cuda(
static_cast<half*>(matmul2_grads.data_ptr()), static_cast<half*>(matmul2_grads.data_ptr()),
k_seq_len, k_seq_len,
k_seq_len*q_seq_len, k_seq_len*q_seq_len,
static_cast<half*>(matmul2_grads.data_ptr()),
k_seq_len,
k_seq_len*q_seq_len,
attn_batches); attn_batches);
// gemm_switch_fp32accum( state,
// a_layout_t,
// b_layout_n,
// k_seq_len,
// q_seq_len,
// head_dim,
// alpha,
// static_cast<const half*>(v_lin_results_ptr),
// lead_dim,
// batch_stride,
// static_cast<const half*>(output_lin_grads.data_ptr()),
// head_dim*attn_batches,
// head_dim,
// beta,
// static_cast<half*>(matmul2_grads.data_ptr()),
// k_seq_len,
// k_seq_len*q_seq_len,
// attn_batches);
// Matmul2 Dgrad2 // Matmul2 Dgrad2
// TODO (OK)
gemm_switch_fp32accum( state, gemm_switch_fp32accum( state,
a_layout_n, a_layout_n,
b_layout_t, b_layout_t,
...@@ -353,7 +523,28 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -353,7 +523,28 @@ std::vector<torch::Tensor> bwd_cuda(
v_lin_grads_ptr, v_lin_grads_ptr,
lead_dim, lead_dim,
batch_stride, batch_stride,
v_lin_grads_ptr,
lead_dim,
batch_stride,
attn_batches); attn_batches);
// gemm_switch_fp32accum( state,
// a_layout_n,
// b_layout_t,
// head_dim,
// k_seq_len,
// q_seq_len,
// alpha,
// static_cast<const half*>(output_lin_grads.data_ptr()),
// head_dim*attn_batches,
// head_dim,
// static_cast<const half*>(dropout_results.data_ptr()),
// k_seq_len,
// k_seq_len*q_seq_len,
// beta,
// v_lin_grads_ptr,
// lead_dim,
// batch_stride,
// attn_batches);
// Apply Dropout Mask and Scale by Dropout Probability // Apply Dropout Mask and Scale by Dropout Probability
apex_masked_scale_cuda<at::Half,float,uint32_t>( apex_masked_scale_cuda<at::Half,float,uint32_t>(
...@@ -375,6 +566,7 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -375,6 +566,7 @@ std::vector<torch::Tensor> bwd_cuda(
assert(softmax_success); assert(softmax_success);
// Matmul1 Dgrad1 // Matmul1 Dgrad1
// TODO (OK)
gemm_switch_fp32accum( state, gemm_switch_fp32accum( state,
a_layout_n, a_layout_n,
b_layout_n, b_layout_n,
...@@ -392,9 +584,31 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -392,9 +584,31 @@ std::vector<torch::Tensor> bwd_cuda(
q_lin_grads_ptr, q_lin_grads_ptr,
lead_dim, lead_dim,
batch_stride, batch_stride,
q_lin_grads_ptr,
lead_dim,
batch_stride,
attn_batches); attn_batches);
// gemm_switch_fp32accum( state,
// a_layout_n,
// b_layout_n,
// head_dim,
// q_seq_len,
// k_seq_len,
// scale,
// k_lin_results_ptr,
// lead_dim,
// batch_stride,
// static_cast<half*>(matmul2_grads.data_ptr()),
// k_seq_len,
// k_seq_len*q_seq_len,
// beta,
// q_lin_grads_ptr,
// lead_dim,
// batch_stride,
// attn_batches);
// Matmul1 Dgrad2 // Matmul1 Dgrad2
// TODO (OK)
gemm_switch_fp32accum( state, gemm_switch_fp32accum( state,
a_layout_n, a_layout_n,
b_layout_t, b_layout_t,
...@@ -411,11 +625,33 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -411,11 +625,33 @@ std::vector<torch::Tensor> bwd_cuda(
beta, beta,
k_lin_grads_ptr, k_lin_grads_ptr,
lead_dim, lead_dim,
batch_stride,
k_lin_grads_ptr,
lead_dim,
batch_stride, batch_stride,
attn_batches); attn_batches);
// gemm_switch_fp32accum( state,
// a_layout_n,
// b_layout_t,
// head_dim,
// k_seq_len,
// q_seq_len,
// scale,
// q_lin_results_ptr,
// lead_dim,
// batch_stride,
// static_cast<half*>(matmul2_grads.data_ptr()),
// k_seq_len,
// k_seq_len*q_seq_len,
// beta,
// k_lin_grads_ptr,
// lead_dim,
// batch_stride,
// attn_batches);
// Input Linear Dgrad // Input Linear Dgrad
THCublasCheck(cublasGemmEx(handle, // TODO (OK)
THCublasCheck(rocblas_gemm_ex(handle,
CUBLAS_OP_N, CUBLAS_OP_N,
CUBLAS_OP_N, CUBLAS_OP_N,
embed_dim, embed_dim,
...@@ -423,20 +659,45 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -423,20 +659,45 @@ std::vector<torch::Tensor> bwd_cuda(
output_lin_dim, output_lin_dim,
static_cast<const void*>(&alpha), static_cast<const void*>(&alpha),
static_cast<const void*>(input_weights.data_ptr()), static_cast<const void*>(input_weights.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(q_lin_grads_ptr), static_cast<const void*>(q_lin_grads_ptr),
CUDA_R_16F, rocblas_datatype_f16_r,
output_lin_dim, output_lin_dim,
static_cast<const void*>(&beta), static_cast<const void*>(&beta),
static_cast<void*>(input_grads.data_ptr()), static_cast<void*>(input_grads.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
CUDA_R_32F, static_cast<void*>(input_grads.data_ptr()),
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); rocblas_datatype_f16_r,
embed_dim,
rocblas_datatype_f32_r,
algo,
solution_index,
flags));
// THCublasCheck(cublasGemmEx(handle,
// CUBLAS_OP_N,
// CUBLAS_OP_N,
// embed_dim,
// batches,
// output_lin_dim,
// static_cast<const void*>(&alpha),
// static_cast<const void*>(input_weights.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(q_lin_grads_ptr),
// CUDA_R_16F,
// output_lin_dim,
// static_cast<const void*>(&beta),
// static_cast<void*>(input_grads.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// CUDA_R_32F,
// CUBLAS_GEMM_DEFAULT_TENSOR_OP));
// Input Linear Wgrad // Input Linear Wgrad
THCublasCheck(cublasGemmEx(handle, // TODO (OK)
THCublasCheck(rocblas_gemm_ex(handle,
CUBLAS_OP_N, CUBLAS_OP_N,
CUBLAS_OP_T, CUBLAS_OP_T,
embed_dim, embed_dim,
...@@ -444,18 +705,43 @@ std::vector<torch::Tensor> bwd_cuda( ...@@ -444,18 +705,43 @@ std::vector<torch::Tensor> bwd_cuda(
batches, batches,
static_cast<const void*>(&alpha), static_cast<const void*>(&alpha),
static_cast<const void*>(inputs.data_ptr()), static_cast<const void*>(inputs.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim, embed_dim,
static_cast<const void*>(q_lin_grads_ptr), static_cast<const void*>(q_lin_grads_ptr),
CUDA_R_16F, rocblas_datatype_f16_r,
output_lin_dim, output_lin_dim,
static_cast<const void*>(&beta), static_cast<const void*>(&beta),
static_cast<void*>(input_weight_grads.data_ptr()), static_cast<void*>(input_weight_grads.data_ptr()),
CUDA_R_16F, rocblas_datatype_f16_r,
embed_dim,
static_cast<void*>(input_weight_grads.data_ptr()),
rocblas_datatype_f16_r,
embed_dim, embed_dim,
CUDA_R_32F, rocblas_datatype_f32_r,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); algo,
THCublasCheck(cublasSetMathMode(handle, CUBLAS_DEFAULT_MATH)); solution_index,
flags));
// THCublasCheck(cublasGemmEx(handle,
// CUBLAS_OP_N,
// CUBLAS_OP_T,
// embed_dim,
// output_lin_dim,
// batches,
// static_cast<const void*>(&alpha),
// static_cast<const void*>(inputs.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// static_cast<const void*>(q_lin_grads_ptr),
// CUDA_R_16F,
// output_lin_dim,
// static_cast<const void*>(&beta),
// static_cast<void*>(input_weight_grads.data_ptr()),
// CUDA_R_16F,
// embed_dim,
// CUDA_R_32F,
// CUBLAS_GEMM_DEFAULT_TENSOR_OP));
// TODO (OK)
// THCublasCheck(cublasSetMathMode(handle, CUBLAS_DEFAULT_MATH));
return { return {
input_grads, input_grads,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment