From 0e45a531466fecf31d17134f3a2d9d343d2a23cf Mon Sep 17 00:00:00 2001
From: Strahinja Stamenkovic <sstamenk@amd.com>
Date: Fri, 15 Nov 2024 09:58:28 -0800
Subject: [PATCH] Added maximum gridDim.y overflow heck before calling
 transposeNoOverlap kernel so that TransposeBigMLFloat16 test passes

---
 onnxruntime/core/providers/rocm/fpgeneric.cu  | 19 +++++++++++++++++++
 .../providers/rocm/shared_inc/fpgeneric.h     |  2 +-
 2 files changed, 20 insertions(+), 1 deletion(-)

diff --git a/onnxruntime/core/providers/rocm/fpgeneric.cu b/onnxruntime/core/providers/rocm/fpgeneric.cu
index d130758bec084..3237ce57e67e6 100644
--- a/onnxruntime/core/providers/rocm/fpgeneric.cu
+++ b/onnxruntime/core/providers/rocm/fpgeneric.cu
@@ -53,6 +53,25 @@ __global__ void CopyVectorBFloat16(const onnxruntime::BFloat16* x, int incx, onn
 
 }  // namespace
 
+dim3 rocblasTransposeHelperDimGrid(int m, int n) {
+  return dim3((n + TRANS_TILE_DIM - 1) / TRANS_TILE_DIM, (m + TRANS_TILE_DIM - 1) / TRANS_TILE_DIM, 1);
+}
+
+// rocblasTransposeHelper can only be used if it won't overflow the maxGridSize y dimension size
+__host__ bool CanUse_rocblasTransposeHelper_MLFloat16(int m, int n) {
+  dim3 dimGrid = rocblasTransposeHelperDimGrid(m, n);
+
+  int deviceId;
+  hipError_t hipError = hipGetDevice(&deviceId);
+  if(hipError != 0) return false;
+
+  hipDeviceProp_t deviceProp;
+  hipError = hipGetDeviceProperties(&deviceProp, deviceId);
+  if(hipError != 0) return false;
+
+  return dimGrid.y < deviceProp.maxGridSize[1];
+}
+
 rocblas_status rocblasTransposeHelper(hipStream_t stream, rocblas_handle, rocblas_operation , rocblas_operation , int m, int n, const half*, const half* A, int, const half*, const half*, int, half* C, int) {
   if (C != A) {
     dim3 dimGrid((n + TRANS_TILE_DIM - 1) / TRANS_TILE_DIM, (m + TRANS_TILE_DIM - 1) / TRANS_TILE_DIM, 1);
diff --git a/onnxruntime/core/providers/rocm/shared_inc/fpgeneric.h b/onnxruntime/core/providers/rocm/shared_inc/fpgeneric.h
index d93f70785c093..c165158f7e461 100644
--- a/onnxruntime/core/providers/rocm/shared_inc/fpgeneric.h
+++ b/onnxruntime/core/providers/rocm/shared_inc/fpgeneric.h
@@ -470,7 +470,7 @@ inline rocblas_status rocblasTransposeHelper(hipStream_t /*stream*/, rocblas_han
   return rocblas_dgeam(handle, transa, transb, m, n, alpha, A, lda, beta, B, ldb, C, ldc);
 }
 
-inline bool CanUse_rocblasTransposeHelper_MLFloat16(int /*m*/, int /*n*/) { return true; }  // CUDA has a limited grid size of 65536, ROCm has higher limits.
+bool CanUse_rocblasTransposeHelper_MLFloat16(int m, int n);
 rocblas_status rocblasTransposeHelper(hipStream_t stream, rocblas_handle, rocblas_operation, rocblas_operation, int m, int n, const half*, const half* A, int, const half*, const half*, int, half* C, int);
 
 // copy