AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[dtest] Kernel Execution time fix for hipExtLaunchKernelGGL and hipExtModuleLaunchKernel API's

Browse Source 
Fix for kernel execution time using clock rate

SWDEV-238517 for enhancing hip unit tests

Change-Id: Id06df31387ecab96f219e0c1fbcdda1609f75605


[ROCm/hip commit: 0b871a24f6]
This commit is contained in:
DURGESH KROTTAPALLI
2021-01-11 21:10:55 +05:30
committed by Mohan Kumar Mithur
parent e7073fc17c
commit 76eb4b7b1e
3 changed files with 44 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/hip/tests/src/runtimeApi/module/matmul.cpp
+14 -13
View File
@@ -20,7 +20,8 @@ THE SOFTWARE.
#include"hip/hip_runtime.h"
__device__ int deviceGlobal = 1;
extern "C" __global__ void matmulK(int* A, int* B, int* C, int N) {
extern "C" __global__ void matmulK(int clockrate, int* A, int* B, int* C,
int N) {
int ROW = blockIdx.y*blockDim.y+threadIdx.y;
int COL = blockIdx.x*blockDim.x+threadIdx.x;
int tmpSum = 0;
@@ -48,31 +49,31 @@ extern "C" __global__ void KernelandExtraParams(int* A, int* B, int* C,
D[ROW * N + COL] = tmpSum;
}
extern "C" __global__ void SixteenSecKernel() {
uint64_t wait_t = 32000000000,
start = clock64(), cur;
do { cur = clock64()-start;}while (cur < wait_t);
extern "C" __global__ void SixteenSecKernel(int clockrate) {
uint64_t wait_t = 16000,
start = clock64()/clockrate, cur;
do { cur = clock64()/clockrate-start;}while (cur < wait_t);
}
extern "C" __global__ void TwoSecKernel() {
extern "C" __global__ void TwoSecKernel(int clockrate) {
if (deviceGlobal == 0x2222) {
deviceGlobal = 0x3333;
}
uint64_t wait_t = 4000000000,
start = clock64(), cur;
do { cur = clock64()-start;}while (cur < wait_t);
uint64_t wait_t = 2000,
start = clock64()/clockrate, cur;
do { cur = clock64()/clockrate-start;}while (cur < wait_t);
if (deviceGlobal != 0x3333) {
deviceGlobal = 0x5555;
}
}
extern "C" __global__ void FourSecKernel() {
extern "C" __global__ void FourSecKernel(int clockrate) {
if (deviceGlobal == 1) {
deviceGlobal = 0x2222;
}
uint64_t wait_t = 8000000000,
start = clock64(), cur;
do { cur = clock64()-start;}while (cur < wait_t);
uint64_t wait_t = 4000,
start = clock64()/clockrate, cur;
do { cur = clock64()/clockrate-start;}while (cur < wait_t);
if (deviceGlobal == 0x2222) {
deviceGlobal = 0x4444;
}