Skip to content

GitLab

Commit adcf522c authored by Khalique Ahmed's avatar Khalique Ahmed
Browse files

change type for alpha beta

parent 9f755219
Hide whitespace changes
Inline Side-by-side
Showing with 56 additions and 60 deletions
src/targets/gpu/gemm_impl.cpp
View file @ adcf522c
...@@ -83,13 +83,9 @@ void gemm_impl(context& ctx, ...@@ -83,13 +83,9 @@ void gemm_impl(context& ctx,
auto a_lens = args[0].get_shape().lens(); auto a_lens = args[0].get_shape().lens();
auto b_lens = args[1].get_shape().lens(); auto b_lens = args[1].get_shape().lens();
output_shape.visit_type([&](auto as) { output_shape.visit_type([&](auto as) {
auto alpha_r = as(alpha); auto alpha_r = compute_fp32 ? alpha : as(alpha);
auto beta_r = as(beta); auto beta_r = compute_fp32 ? beta : as(beta);
if(compute_fp32)
{
alpha_r = alpha;
beta_r = beta;
}
auto out_lens = output_shape.lens(); auto out_lens = output_shape.lens();
rocblas_int m = out_lens[dim_0]; rocblas_int m = out_lens[dim_0];
rocblas_int n = out_lens[dim_1]; rocblas_int n = out_lens[dim_1];
...@@ -109,63 +105,63 @@ void gemm_impl(context& ctx, ...@@ -109,63 +105,63 @@ void gemm_impl(context& ctx,
// C^T = (B^T) * (A^T). That is the reason we input args[1] as // C^T = (B^T) * (A^T). That is the reason we input args[1] as
// A and args[0] as B in calling the rocblas_gemm. // A and args[0] as B in calling the rocblas_gemm.
rocblas_invoke(&rocblas_gemm_ex, rocblas_invoke(&rocblas_gemm_ex,
ctx.get_stream().get_rocblas(), ctx.get_stream().get_rocblas(),
transb ? rocblas_operation_transpose : rocblas_operation_none, transb ? rocblas_operation_transpose : rocblas_operation_none,
transa ? rocblas_operation_transpose : rocblas_operation_none, transa ? rocblas_operation_transpose : rocblas_operation_none,
n, n,
m, m,
k, k,
&alpha_r, &alpha_r,
to_pointer(args.at(1)), to_pointer(args.at(1)),
arg_type, arg_type,
ldb, ldb,
to_pointer(args.at(0)), to_pointer(args.at(0)),
arg_type, arg_type,
lda, lda,
&beta_r, &beta_r,
to_pointer(args[2]), to_pointer(args[2]),
output_type, output_type,
ldc, ldc,
is_3inputs ? to_pointer(args[3]) : to_pointer(args[2]), is_3inputs ? to_pointer(args[3]) : to_pointer(args[2]),
output_type, output_type,
ldc, ldc,
compute_type, compute_type,
rocblas_gemm_algo_standard, rocblas_gemm_algo_standard,
0, 0,
flag); flag);
} }
else else
{ {
rocblas_invoke(&rocblas_gemm_strided_batched_ex, rocblas_invoke(&rocblas_gemm_strided_batched_ex,
ctx.get_stream().get_rocblas(), ctx.get_stream().get_rocblas(),
transb ? rocblas_operation_transpose : rocblas_operation_none, transb ? rocblas_operation_transpose : rocblas_operation_none,
transa ? rocblas_operation_transpose : rocblas_operation_none, transa ? rocblas_operation_transpose : rocblas_operation_none,
n, n,
m, m,
k, k,
&alpha_r, &alpha_r,
to_pointer(args.at(1)), to_pointer(args.at(1)),
arg_type, arg_type,
ldb, ldb,
k * n, k * n,
to_pointer(args.at(0)), to_pointer(args.at(0)),
arg_type, arg_type,
lda, lda,
m * k, m * k,
&beta_r, &beta_r,
to_pointer(args[2]), to_pointer(args[2]),
output_type, output_type,
ldc, ldc,
m * n, m * n,
is_3inputs ? to_pointer(args[3]) : to_pointer(args[2]), is_3inputs ? to_pointer(args[3]) : to_pointer(args[2]),
output_type, output_type,
ldc, ldc,
m * n, m * n,
num_matrices, num_matrices,
compute_type, compute_type,
rocblas_gemm_algo_standard, rocblas_gemm_algo_standard,
0, 0,
flag); flag);
} }
}); });
} }
......
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