AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

merge from master

Sfoglia il codice sorgente
This commit is contained in:
Anusha Godavarthy Surya
2019-10-25 15:52:09 +05:30
parent 259d8b4cdf 70f2cd1317
commit 5f47e99ffe
40 ha cambiato i file con 1341 aggiunte e 690 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
tests/src/deviceLib/hipMathFunctions.cpp
+6 -1
Vedi File
@@ -29,7 +29,12 @@ THE SOFTWARE.
#include "hip/hip_runtime.h"
#include "test_common.h"
#if __HIP_ARCH_GFX803__ || __HIP_ARCH_GFX900__ || __HIP_ARCH_GFX906__ || __HIP_ARCH_GFX908__
#if __HIP_ARCH_GFX803__ || \
__HIP_ARCH_GFX900__ || \
__HIP_ARCH_GFX906__ || \
__HIP_ARCH_GFX908__ || \
__HIP_ARCH_GFX1010__ || \
__HIP_ARCH_GFX1012__
__global__ void kernel_abs_int64(long long *input, long long *output) {
int tx = threadIdx.x;
tests/src/deviceLib/hipTestHalf.cpp
+6 -1
Vedi File
@@ -28,7 +28,12 @@ THE SOFTWARE.
#include "test_common.h"
#if __HIP_ARCH_GFX803__ || __HIP_ARCH_GFX900__ || __HIP_ARCH_GFX906__ || __HIP_ARCH_GFX908__
#if __HIP_ARCH_GFX803__ || \
__HIP_ARCH_GFX900__ || \
__HIP_ARCH_GFX906__ || \
__HIP_ARCH_GFX908__ || \
__HIP_ARCH_GFX1010__ || \
__HIP_ARCH_GFX1012__
__device__ void test_convert() {
__half x;
tests/src/deviceLib/hipTestNativeHalf.cpp
+6 -1
Vedi File
@@ -32,7 +32,12 @@ THE SOFTWARE.
using namespace std;
#if __HIP_ARCH_GFX803__ || __HIP_ARCH_GFX900__ || __HIP_ARCH_GFX906__ || __HIP_ARCH_GFX908__
#if __HIP_ARCH_GFX803__ || \
__HIP_ARCH_GFX900__ || \
__HIP_ARCH_GFX906__ || \
__HIP_ARCH_GFX908__ || \
__HIP_ARCH_GFX1010__ || \
__HIP_ARCH_GFX1012__
__global__
void __halfTest(bool* result, __half a) {
tests/src/runtimeApi/memory/hipMemcpy_simple.cpp
+2 -2
Vedi File
@@ -120,8 +120,8 @@ void simpleTest2(size_t numElements, bool usePinnedHost) {
HIPCHECK(hipHostFree(A_h1));
HIPCHECK(hipHostFree(A_h2));
} else {
free(A_h1);
free(A_h2);
aligned_free(A_h1);
aligned_free(A_h2);
}
}
tests/src/runtimeApi/memory/hipMemset2D.cpp
+7 -6
Vedi File
@@ -45,7 +45,7 @@ bool testhipMemset2D(int memsetval,int p_gpuDevice)
char *A_d;
char *A_h;
bool testResult = true;
HIPCHECK ( hipMemAllocPitch((hipDeviceptr_t*)&A_d, &pitch_A, width , numH,16) );
HIPCHECK(hipMemAllocPitch((hipDeviceptr_t*)&A_d, &pitch_A, width , numH,16));
A_h = (char*)malloc(sizeElements);
HIPASSERT(A_h != NULL);
for (size_t i=0; i<elements; i++) {
@@ -89,8 +89,9 @@ bool testhipMemset2DAsync(int memsetval,int p_gpuDevice)
}
hipStream_t stream;
HIPCHECK(hipStreamCreate(&stream));
HIPCHECK ( hipMemset2DAsync(A_d, pitch_A, memsetval, numW, numH, stream) );
HIPCHECK ( hipMemcpy2D(A_h, width, A_d, pitch_A, numW, numH, hipMemcpyDeviceToHost));
HIPCHECK(hipMemset2DAsync(A_d, pitch_A, memsetval, numW, numH, stream) );
HIPCHECK(hipStreamSynchronize(stream));
HIPCHECK(hipMemcpy2D(A_h, width, A_d, pitch_A, numW, numH, hipMemcpyDeviceToHost));
for (int i=0; i<elements; i++) {
if (A_h[i] != memsetval) {
@@ -112,9 +113,9 @@ int main(int argc, char *argv[])
hipCtx_t context;
hipCtxCreate(&context, 0, p_gpuDevice);
bool testResult = false;
testResult = testhipMemset2D(memsetval, p_gpuDevice);
testResult = testhipMemset2DAsync(memsetval, p_gpuDevice);
bool testResult = true;
testResult &= testhipMemset2D(memsetval, p_gpuDevice);
testResult &= testhipMemset2DAsync(memsetval, p_gpuDevice);
hipCtxDestroy(context);
if(testResult){
passed();
tests/src/runtimeApi/memory/hipMemset3D.cpp
+67 -15
Vedi File
@@ -41,27 +41,27 @@ bool testhipMemset3D(int memsetval,int p_gpuDevice)
size_t elements = numW* numH* depth;
printf ("testhipMemset3D memsetval=%2x device=%d\n", memsetval, p_gpuDevice);
printf ("testhipMemset3D memsetval=%2x device=%d\n", memsetval, p_gpuDevice);
char *A_h;
bool testResult = true;
hipExtent extent = make_hipExtent(width, numH, depth);
hipPitchedPtr devPitchedPtr;
HIPCHECK(hipMalloc3D(&devPitchedPtr, extent));
A_h = (char*)malloc(sizeElements);
HIPASSERT(A_h != NULL);
for (size_t i=0; i<elements; i++) {
A_h = (char*)malloc(sizeElements);
HIPASSERT(A_h != NULL);
for (size_t i=0; i<elements; i++) {
A_h[i] = 1;
}
HIPCHECK ( hipMemset3D( devPitchedPtr, memsetval, extent) );
hipMemcpy3DParms myparms = {0};
myparms.srcPos = make_hipPos(0,0,0);
myparms.dstPos = make_hipPos(0,0,0);
myparms.dstPtr = make_hipPitchedPtr(A_h, width , numW, numH);
myparms.srcPtr = devPitchedPtr;
myparms.extent = extent;
HIPCHECK(hipMemset3D( devPitchedPtr, memsetval, extent));
hipMemcpy3DParms myparms = {0};
myparms.srcPos = make_hipPos(0,0,0);
myparms.dstPos = make_hipPos(0,0,0);
myparms.dstPtr = make_hipPitchedPtr(A_h, width , numW, numH);
myparms.srcPtr = devPitchedPtr;
myparms.extent = extent;
#ifdef __HIP_PLATFORM_NVCC__
myparms.kind = hipMemcpyKindToCudaMemcpyKind(hipMemcpyDeviceToHost);
myparms.kind = hipMemcpyKindToCudaMemcpyKind(hipMemcpyDeviceToHost);
#else
myparms.kind = hipMemcpyDeviceToHost;
#endif
@@ -69,7 +69,58 @@ bool testhipMemset3D(int memsetval,int p_gpuDevice)
for (int i=0; i<elements; i++) {
if (A_h[i] != memsetval) {
testResult = false;
testResult = false;
printf("mismatch at index:%d computed:%02x, memsetval:%02x\n", i, (int)A_h[i], (int)memsetval);
break;
}
}
HIPCHECK(hipFree(devPitchedPtr.ptr));
free(A_h);
return testResult;
}
bool testhipMemset3DAsync(int memsetval,int p_gpuDevice)
{
size_t numH = 256;
size_t numW = 256;
size_t depth = 10;
size_t width = numW * sizeof(char);
size_t sizeElements = width * numH * depth;
size_t elements = numW* numH* depth;
printf ("testhipMemset3D memsetval=%2x device=%d\n", memsetval, p_gpuDevice);
char *A_h;
bool testResult = true;
hipExtent extent = make_hipExtent(width, numH, depth);
hipPitchedPtr devPitchedPtr;
HIPCHECK(hipMalloc3D(&devPitchedPtr, extent));
A_h = (char*)malloc(sizeElements);
HIPASSERT(A_h != NULL);
for (size_t i=0; i<elements; i++) {
A_h[i] = 1;
}
hipStream_t stream;
HIPCHECK(hipStreamCreate(&stream));
HIPCHECK(hipMemset3DAsync(devPitchedPtr, memsetval, extent, stream));
HIPCHECK(hipStreamSynchronize(stream));
hipMemcpy3DParms myparms = {0};
myparms.srcPos = make_hipPos(0,0,0);
myparms.dstPos = make_hipPos(0,0,0);
myparms.dstPtr = make_hipPitchedPtr(A_h, width , numW, numH);
myparms.srcPtr = devPitchedPtr;
myparms.extent = extent;
#ifdef __HIP_PLATFORM_NVCC__
myparms.kind = hipMemcpyKindToCudaMemcpyKind(hipMemcpyDeviceToHost);
#else
myparms.kind = hipMemcpyDeviceToHost;
#endif
HIPCHECK(hipMemcpy3D(&myparms));
for (int i=0; i<elements; i++) {
if (A_h[i] != memsetval) {
testResult = false;
printf("mismatch at index:%d computed:%02x, memsetval:%02x\n", i, (int)A_h[i], (int)memsetval);
break;
}
@@ -82,9 +133,10 @@ bool testhipMemset3D(int memsetval,int p_gpuDevice)
int main(int argc, char *argv[])
{
HipTest::parseStandardArguments(argc, argv, true);
bool testResult = false;
HIPCHECK(hipSetDevice(p_gpuDevice));
testResult = testhipMemset3D(memsetval, p_gpuDevice);
bool testResult = true;
testResult &= testhipMemset3D(memsetval, p_gpuDevice);
testResult &= testhipMemset3DAsync(memsetval, p_gpuDevice);
if (testResult) {
passed();
} else {
tests/src/runtimeApi/module/hipLaunchCooperativeKernel.cpp
+1 -1
Vedi File
@@ -116,7 +116,7 @@ int main() {
dimBlock.x = workgroups[i];
// Calculate the device occupancy to know how many blocks can be run concurrently
hipOccupancyMaxActiveBlocksPerMultiprocessor(reinterpret_cast<uint32_t*>(&numBlocks),
hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlocks,
test_gws, dimBlock.x * dimBlock.y * dimBlock.z, dimBlock.x * sizeof(long));
dimGrid.x = deviceProp.multiProcessorCount * std::min(numBlocks, 32);
tests/src/runtimeApi/module/hipOccupancyMaxActiveBlocksPerMultiprocessor.cpp
+3 -18
Vedi File
@@ -30,10 +30,6 @@ THE SOFTWARE.
#include "hip/hip_runtime.h"
#include "test_common.h"
#define fileName "vcpy_kernel.code"
#define kernel_name "hello_world"
__global__ void f1(float *a) { *a = 1.0; }
template <typename T>
@@ -44,16 +40,15 @@ __global__ void f2(T *a) { *a = 1; }
int main(int argc, char* argv[]) {
// test case for using kernel function pointer
uint32_t gridSize = 0;
uint32_t blockSize = 0;
int gridSize = 0;
int blockSize = 0;
hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0);
assert(gridSize != 0 && blockSize != 0);
uint32_t numBlock = 0;
int numBlock = 0;
hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, f1, blockSize, 0);
assert(numBlock != 0);
// test case for using kernel function pointer with template
gridSize = 0;
blockSize = 0;
@@ -64,15 +59,5 @@ int main(int argc, char* argv[]) {
hipOccupancyMaxActiveBlocksPerMultiprocessor<void(*)(int *)>(&numBlock, f2, blockSize, 0);
assert(numBlock != 0);
// test case for using kernel with hipFunction_t type
numBlock = 0;
hipModule_t Module;
hipFunction_t Function;
HIPCHECK(hipModuleLoad(&Module, fileName));
HIPCHECK(hipModuleGetFunction(&Function, Module, kernel_name));
HIPCHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, Function, blockSize, 0));
assert(numBlock != 0);
passed();
}
tests/src/runtimeApi/module/hipOccupancyMaxPotentialBlockSize.cpp
+2 -18
Vedi File
@@ -30,22 +30,16 @@ THE SOFTWARE.
#include "hip/hip_runtime.h"
#include "test_common.h"
#define fileName "vcpy_kernel.code"
#define kernel_name "hello_world"
__global__ void f1(float *a) { *a = 1.0; }
template <typename T>
__global__ void f2(T *a) { *a = 1; }
int main(int argc, char* argv[]) {
// test case for using kernel function pointer
uint32_t gridSize = 0;
uint32_t blockSize = 0;
int gridSize = 0;
int blockSize = 0;
hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, f1, 0, 0);
assert(gridSize != 0 && blockSize != 0);
@@ -55,15 +49,5 @@ int main(int argc, char* argv[]) {
hipOccupancyMaxPotentialBlockSize<void(*)(int *)>(&gridSize, &blockSize, f2, 0, 0);
assert(gridSize != 0 && blockSize != 0);
// test case for using kernel with hipFunction_t type
gridSize = 0;
blockSize = 0;
hipModule_t Module;
hipFunction_t Function;
HIPCHECK(hipModuleLoad(&Module, fileName));
HIPCHECK(hipModuleGetFunction(&Function, Module, kernel_name));
HIPCHECK(hipOccupancyMaxPotentialBlockSize(&gridSize, &blockSize, Function, 0, 0));
assert(gridSize != 0 && blockSize != 0);
passed();
}
tests/src/test_common.h
+4 -1
Vedi File
@@ -99,10 +99,13 @@ THE SOFTWARE.
#ifdef _WIN64
#include <tchar.h>
#define aligned_alloc _aligned_malloc
#define aligned_alloc(x,y) _aligned_malloc(y,x)
#define aligned_free(x) _aligned_free(x)
#define popen(x,y) _popen(x,y)
#define pclose(x) _pclose(x)
#define setenv(x,y,z) _putenv_s(x,y)
#else
#define aligned_free(x) free(x)
#endif
// standard command-line variables: