ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-14 02:02:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] Fix Batched GEMM Example GPU verification (#2800)

Browse Source 
Added more batched GEMM test cases

[ROCm/composable_kernel commit: e82ccbdaf7]
This commit is contained in:
aledudek
2025-09-09 09:30:57 +02:00
committed by GitHub
parent f8c8263798
commit 9cc281007e
4 changed files with 79 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/ck_tile/16_batched_gemm/batched_gemm.hpp
View File

@@ -71,7 +71,7 @@ auto create_args(int argc, char* argv[])
.insert("batch_stride_b", "2097152", "Batch B stride")
.insert("batch_stride_c", "524288", "Batch C stride")
.insert("batch_count", "8", "Batch count")
.insert("v", "2", "0. No validation, 1. Validation on CPU, 2. Validation on GPU")
.insert("v", "1", "0. No validation, 1. Validation on CPU, 2. Validation on GPU")
.insert("prec", "fp16", "data type. fp16/bf16/fp8/bf8")
.insert("warmup", "50", "number of iterations before benchmark the kernel")
.insert("repeat", "100", "number of iterations to benchmark the kernel")

29
example/ck_tile/16_batched_gemm/run_batched_gemm_example.inc
View File

@@ -246,23 +246,9 @@ int run_batched_gemm_example_with_layouts(int argc,
c_m_n_gpu_ref.SetZero();
c_m_n_gpu_buf_ref.SetZero();
ADataType* d_A;
BDataType* d_B;
CDataType* d_C;
ck_tile::hip_check_error(hipMalloc(&d_A, batch_count * M * K * sizeof(ADataType)));
ck_tile::hip_check_error(hipMalloc(&d_B, batch_count * N * K * sizeof(BDataType)));
ck_tile::hip_check_error(hipMalloc(&d_C, batch_count * M * N * sizeof(CDataType)));
ck_tile::hip_check_error(hipMemcpy(d_A,
a_m_k_dev_buf.GetDeviceBuffer(),
batch_count * M * K * sizeof(ADataType),
hipMemcpyHostToDevice));
ck_tile::hip_check_error(hipMemcpy(d_B,
b_k_n_dev_buf.GetDeviceBuffer(),
batch_count * N * K * sizeof(BDataType),
hipMemcpyHostToDevice));
ADataType* d_A = static_cast<ADataType*>(a_m_k_dev_buf.GetDeviceBuffer());
BDataType* d_B = static_cast<BDataType*>(b_k_n_dev_buf.GetDeviceBuffer());
CDataType* d_C = static_cast<CDataType*>(c_m_n_gpu_buf_ref.GetDeviceBuffer());
ck_tile::reference_batched_gemm_gpu<ADataType,
BDataType,
@@ -284,15 +270,6 @@ int run_batched_gemm_example_with_layouts(int argc,
batch_stride_C,
batch_count);
ck_tile::hip_check_error(hipMemcpy(c_m_n_gpu_buf_ref.GetDeviceBuffer(),
d_C,
batch_count * M * N * sizeof(CDataType),
hipMemcpyDeviceToHost));
ck_tile::hip_check_error(hipFree(d_A));
ck_tile::hip_check_error(hipFree(d_B));
ck_tile::hip_check_error(hipFree(d_C));
c_m_n_gpu_buf_ref.FromDevice(c_m_n_gpu_ref.data());
const float max_accumulated_value =
*std::max_element(c_m_n_gpu_ref.mData.begin(), c_m_n_gpu_ref.mData.end());