Commit efa05704 authored by Paul's avatar Paul
Browse files

Remove comments

parent 7eb45111
...@@ -22,12 +22,6 @@ ...@@ -22,12 +22,6 @@
* THE SOFTWARE. * THE SOFTWARE.
*/ */
/**
* Contains a templated struct implementation that wraps several rocBLAS API calls
* used by the GEMM operator. These are accessed by methods declared in gemm_impl.hpp
*
*/
#include <rocblas/rocblas.h> #include <rocblas/rocblas.h>
#include <migraphx/gpu/gemm_impl.hpp> #include <migraphx/gpu/gemm_impl.hpp>
#include <migraphx/reduce_dims.hpp> #include <migraphx/reduce_dims.hpp>
...@@ -446,9 +440,6 @@ struct gemm_impl ...@@ -446,9 +440,6 @@ struct gemm_impl
ctx.finish(); ctx.finish();
}); });
// todo: Measured time dropped from 20 us to about 6.7 us when I raised hot_calls from
// 1 to 11. The higher the hot_calls value, the faster per-call time up to at least 25,
// and increasing cold_calls makes little or no difference. Why?
host_time /= hot_calls; host_time /= hot_calls;
// dev/evaluation only: track time for first solution. // dev/evaluation only: track time for first solution.
...@@ -579,9 +570,6 @@ int32_t gemm_finalize(context& ctx, ...@@ -579,9 +570,6 @@ int32_t gemm_finalize(context& ctx,
int32_t solution_idx) int32_t solution_idx)
{ {
#ifdef MIGRAPHX_USE_ROCBLAS_TUNING_API #ifdef MIGRAPHX_USE_ROCBLAS_TUNING_API
// This code should be called only if either the environment var.
// MIGRAPHX_ENABLE_GEMM_TUNING, or option --exhaustive-tune, is set
if(solution_idx == 0) if(solution_idx == 0)
{ {
auto gemm_item = gemm_impl<int32_t>(output_shape, input_shapes, alpha, beta, compute_fp32); auto gemm_item = gemm_impl<int32_t>(output_shape, input_shapes, alpha, beta, compute_fp32);
......
...@@ -25,7 +25,6 @@ ...@@ -25,7 +25,6 @@
#define MIGRAPHX_GUARD_MIGRAPHLIB_ROCBLAS_HPP #define MIGRAPHX_GUARD_MIGRAPHLIB_ROCBLAS_HPP
#include <migraphx/manage_ptr.hpp> #include <migraphx/manage_ptr.hpp>
#include <migraphx/gpu/config.hpp> #include <migraphx/gpu/config.hpp>
// ROCBLAS_BETA_FEATURES_API is defined by CMake, if available.
#include <rocblas/rocblas.h> #include <rocblas/rocblas.h>
namespace migraphx { namespace migraphx {
......
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