[rocm-libraries] ROCm/rocm-libraries#4872 (commit ca623f7)

[CK] Small improvements for grouped conv backward weight
 (#4872)

## Motivation

Improvements for CK Tile convolution builder run function and atol/rtol
calculations.

## Technical Details

- Add preprocessing function for wrw when k_batch is larger than 1 for
builder run function
- Divide num acums by number of groups to get real number of accums

## Test Plan

CI wrw tests

## Test Result

pending

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

AICK-783

example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc

@@ -134,7 +134,7 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
         float max_accumulated_value =
             *std::max_element(wei_host_result.mData.begin(), wei_host_result.mData.end());
 
-        const ck::index_t num_accums         = out.GetElementSize() / conv_param.K_;
+        const ck::index_t num_accums = out.GetElementSize() / (conv_param.K_ * conv_param.G_);
         const ck::index_t num_accums_split_k = split_k;
         double rtol = ck::utils::get_relative_threshold<InDataType, WeiDataType, AccDataType>(
             num_accums / num_accums_split_k);