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

Re-sync with upstream.

Browse Source 
[ROCm/clr commit: 602280eb8b]
This commit is contained in:
Alex Voicu
2018-10-18 12:27:03 +01:00
parent 2577b11365 cf95358003
commit c505c2be38
103 changed files with 1478 additions and 1308 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/hipamd/tests/src/kernel/hipDynamicShared.cpp
+2 -2
View File
@@ -30,7 +30,7 @@ THE SOFTWARE.
#include "test_common.h"
template <typename T>
__global__ void testExternSharedKernel(hipLaunchParm lp, const T* A_d, const T* B_d, T* C_d,
__global__ void testExternSharedKernel(const T* A_d, const T* B_d, T* C_d,
size_t numElements, size_t groupElements) {
// declare dynamic shared memory
#if defined(__HIP_PLATFORM_HCC__)
@@ -114,7 +114,7 @@ void testExternShared(size_t N, size_t groupElements) {
size_t groupMemBytes = groupElements * sizeof(T);
// launch kernel with dynamic shared memory
hipLaunchKernel(HIP_KERNEL_NAME(testExternSharedKernel<T>), dim3(blocks), dim3(threadsPerBlock),
hipLaunchKernelGGL(HIP_KERNEL_NAME(testExternSharedKernel<T>), dim3(blocks), dim3(threadsPerBlock),
groupMemBytes, 0, A_d, B_d, C_d, N, groupElements);
HIPCHECK(hipDeviceSynchronize());
projects/clr/hipamd/tests/src/kernel/hipDynamicShared2.cpp
+2 -2
View File
@@ -32,7 +32,7 @@ THE SOFTWARE.
#define LEN 16 * 1024
#define SIZE LEN * 4
__global__ void vectorAdd(hipLaunchParm lp, float* Ad, float* Bd) {
__global__ void vectorAdd(float* Ad, float* Bd) {
HIP_DYNAMIC_SHARED(float, sBd);
int tx = threadIdx.x;
for (int i = 0; i < LEN / 64; i++) {
@@ -53,7 +53,7 @@ int main() {
hipMalloc(&Bd, SIZE);
hipMemcpy(Ad, A, SIZE, hipMemcpyHostToDevice);
hipMemcpy(Bd, B, SIZE, hipMemcpyHostToDevice);
hipLaunchKernel(vectorAdd, dim3(1, 1, 1), dim3(64, 1, 1), SIZE, 0, Ad, Bd);
hipLaunchKernelGGL(vectorAdd, dim3(1, 1, 1), dim3(64, 1, 1), SIZE, 0, Ad, Bd);
hipMemcpy(B, Bd, SIZE, hipMemcpyDeviceToHost);
for (int i = 0; i < LEN; i++) {
assert(B[i] > 1.0f && B[i] < 3.0f);
projects/clr/hipamd/tests/src/kernel/hipEmptyKernel.cpp
+2 -2
View File
@@ -25,10 +25,10 @@ THE SOFTWARE.
#include "test_common.h"
__global__ void Empty(hipLaunchParm lp, int param) {}
__global__ void Empty(int param) {}
int main() {
hipLaunchKernel(HIP_KERNEL_NAME(Empty), dim3(1), dim3(1), 0, 0, 0);
hipLaunchKernelGGL(HIP_KERNEL_NAME(Empty), dim3(1), dim3(1), 0, 0, 0);
hipDeviceSynchronize();
passed();
}
projects/clr/hipamd/tests/src/kernel/hipGridLaunch.cpp
+2 -2
View File
@@ -37,7 +37,7 @@ __device__ int foo(int i) { return i + 1; }
//---
// Syntax we would like to support with GRID_LAUNCH enabled:
template <typename T>
__global__ void vectorADD2(hipLaunchParm lp, T* A_d, T* B_d, T* C_d, size_t N) {
__global__ void vectorADD2(T* A_d, T* B_d, T* C_d, size_t N) {
size_t offset = (blockIdx.x * blockDim.x + threadIdx.x);
size_t stride = blockDim.x * gridDim.x;
@@ -63,7 +63,7 @@ int test_gl2(size_t N) {
HIPCHECK(hipMemcpy(A_d, A_h, Nbytes, hipMemcpyHostToDevice));
HIPCHECK(hipMemcpy(B_d, B_h, Nbytes, hipMemcpyHostToDevice));
hipLaunchKernel(vectorADD2, dim3(blocks), dim3(threadsPerBlock), 0, 0, A_d, B_d, C_d, N);
hipLaunchKernelGGL(vectorADD2, dim3(blocks), dim3(threadsPerBlock), 0, 0, A_d, B_d, C_d, N);
HIPCHECK(hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
projects/clr/hipamd/tests/src/kernel/hipLanguageExtensions.cpp
+1 -1
View File
@@ -32,7 +32,7 @@ THE SOFTWARE.
#include <test_common.h>
#ifdef __HCC__
#include <hc.hpp>
#include <hc.hpp>
#endif
// cudaA
projects/clr/hipamd/tests/src/kernel/hipLaunchParm.cpp
+2 -2
View File
@@ -916,7 +916,7 @@ int main() {
hipLaunchKernelGGL(HIP_KERNEL_NAME(vAdd), dim3(1024), 1, 0, 0, Ad);
hipLaunchKernelGGL(HIP_KERNEL_NAME(vAdd), dim3(1024), dim3(1), 0, 0, Ad);
// Test: Passing hipLaunchKernel inside another macro:
// Test: Passing hipLaunchKernelGGL inside another macro:
float e0;
GPU_PRINT_TIME(hipLaunchKernelGGL(vAdd, dim3(1024),
dim3(1), 0, 0, Ad), e0, j);
@@ -924,7 +924,7 @@ int main() {
dim3(1), 0, 0, Ad)), e0, j);
#ifdef EXTRA_PARENS_1
// Don't wrap hipLaunchKernel in extra set of parens:
// Don't wrap hipLaunchKernelGGL in extra set of parens:
GPU_PRINT_TIME((hipLaunchKernelGGL(vAdd, dim3(1024),
dim3(1), 0, 0, Ad)), e0, j);
#endif
projects/clr/hipamd/tests/src/kernel/hipPrintfKernel.cpp
+2 -2
View File
@@ -27,10 +27,10 @@ THE SOFTWARE.
#include "test_common.h"
__global__ void run_printf(hipLaunchParm lp) { printf("Hello World\n"); }
__global__ void run_printf() { printf("Hello World\n"); }
int main() {
hipLaunchKernel(HIP_KERNEL_NAME(run_printf), dim3(1), dim3(1), 0, 0);
hipLaunchKernelGGL(HIP_KERNEL_NAME(run_printf), dim3(1), dim3(1), 0, 0);
hipDeviceSynchronize();
passed();
}
projects/clr/hipamd/tests/src/kernel/hipTestConstant.cpp
+2 -2
View File
@@ -35,7 +35,7 @@ THE SOFTWARE.
__constant__ int Value[LEN];
__global__ void Get(hipLaunchParm lp, int* Ad) {
__global__ void Get(int* Ad) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
Ad[tid] = Value[tid];
}
@@ -52,7 +52,7 @@ int main() {
HIP_ASSERT(hipMalloc((void**)&Ad, SIZE));
HIP_ASSERT(hipMemcpyToSymbol(HIP_SYMBOL(Value), A, SIZE, 0, hipMemcpyHostToDevice));
hipLaunchKernel(Get, dim3(1, 1, 1), dim3(LEN, 1, 1), 0, 0, Ad);
hipLaunchKernelGGL(Get, dim3(1, 1, 1), dim3(LEN, 1, 1), 0, 0, Ad);
HIP_ASSERT(hipMemcpy(B, Ad, SIZE, hipMemcpyDeviceToHost));
for (unsigned i = 0; i < LEN; i++) {
projects/clr/hipamd/tests/src/kernel/hipTestMallocKernel.cpp
+4 -4
View File
@@ -32,12 +32,12 @@ THE SOFTWARE.
#define NUM 1024
#define SIZE NUM * 8
__global__ void Alloc(hipLaunchParm lp, uint64_t* Ptr) {
__global__ void Alloc(uint64_t* Ptr) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
Ptr[tid] = (uint64_t)malloc(128);
}
__global__ void Free(hipLaunchParm lp, uint64_t* Ptr) {
__global__ void Free(uint64_t* Ptr) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
free((void*)Ptr[tid]);
}
@@ -54,10 +54,10 @@ int main() {
HIP_ASSERT(hipSetDevice(i));
HIP_ASSERT(hipMalloc((void**)&dPtr, SIZE));
HIP_ASSERT(hipMemcpy(dPtr, hPtr, SIZE, hipMemcpyHostToDevice));
hipLaunchKernel(Alloc, dim3(1, 1, 1), dim3(NUM, 1, 1), 0, 0, dPtr);
hipLaunchKernelGGL(Alloc, dim3(1, 1, 1), dim3(NUM, 1, 1), 0, 0, dPtr);
HIP_ASSERT(hipMemcpy(hPtr, dPtr, SIZE, hipMemcpyDeviceToHost));
assert(hPtr[0] != 0);
hipLaunchKernel(Free, dim3(1, 1, 1), dim3(NUM, 1, 1), 0, 0, dPtr);
hipLaunchKernelGGL(Free, dim3(1, 1, 1), dim3(NUM, 1, 1), 0, 0, dPtr);
HIP_ASSERT(hipFree(dPtr));
for (uint32_t i = 1; i < NUM; i++) {
assert(hPtr[i] == hPtr[i - 1] + 4096);
projects/clr/hipamd/tests/src/kernel/hipTestMemKernel.cpp
+20 -20
View File
@@ -34,52 +34,52 @@ THE SOFTWARE.
#define LEN11 11 * 4
#define LEN12 12 * 4
__global__ void MemCpy8(hipLaunchParm lp, uint8_t* In, uint8_t* Out) {
__global__ void MemCpy8(uint8_t* In, uint8_t* Out) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memcpy(Out + tid * 8, In + tid * 8, 8);
}
__global__ void MemCpy9(hipLaunchParm lp, uint8_t* In, uint8_t* Out) {
__global__ void MemCpy9(uint8_t* In, uint8_t* Out) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memcpy(Out + tid * 9, In + tid * 9, 9);
}
__global__ void MemCpy10(hipLaunchParm lp, uint8_t* In, uint8_t* Out) {
__global__ void MemCpy10(uint8_t* In, uint8_t* Out) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memcpy(Out + tid * 10, In + tid * 10, 10);
}
__global__ void MemCpy11(hipLaunchParm lp, uint8_t* In, uint8_t* Out) {
__global__ void MemCpy11(uint8_t* In, uint8_t* Out) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memcpy(Out + tid * 11, In + tid * 11, 11);
}
__global__ void MemCpy12(hipLaunchParm lp, uint8_t* In, uint8_t* Out) {
__global__ void MemCpy12(uint8_t* In, uint8_t* Out) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memcpy(Out + tid * 12, In + tid * 12, 12);
}
__global__ void MemSet8(hipLaunchParm lp, uint8_t* In) {
__global__ void MemSet8(uint8_t* In) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memset(In + tid * 8, 1, 8);
}
__global__ void MemSet9(hipLaunchParm lp, uint8_t* In) {
__global__ void MemSet9(uint8_t* In) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memset(In + tid * 9, 1, 9);
}
__global__ void MemSet10(hipLaunchParm lp, uint8_t* In) {
__global__ void MemSet10(uint8_t* In) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memset(In + tid * 10, 1, 10);
}
__global__ void MemSet11(hipLaunchParm lp, uint8_t* In) {
__global__ void MemSet11(uint8_t* In) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memset(In + tid * 11, 1, 11);
}
__global__ void MemSet12(hipLaunchParm lp, uint8_t* In) {
__global__ void MemSet12(uint8_t* In) {
int tid = threadIdx.x + blockIdx.x * blockDim.x;
memset(In + tid * 12, 1, 12);
}
@@ -98,8 +98,8 @@ int main() {
hipMalloc((void**)&Bd, LEN8);
hipMalloc((void**)&Cd, LEN8);
hipMemcpy(Ad, A, LEN8, hipMemcpyHostToDevice);
hipLaunchKernel(MemCpy8, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernel(MemSet8, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipLaunchKernelGGL(MemCpy8, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernelGGL(MemSet8, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipMemcpy(B, Bd, LEN8, hipMemcpyDeviceToHost);
hipMemcpy(C, Cd, LEN8, hipMemcpyDeviceToHost);
for (uint32_t i = 0; i < LEN8; i++) {
@@ -126,8 +126,8 @@ int main() {
hipMalloc((void**)&Bd, LEN9);
hipMalloc((void**)&Cd, LEN9);
hipMemcpy(Ad, A, LEN9, hipMemcpyHostToDevice);
hipLaunchKernel(MemCpy9, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernel(MemSet9, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipLaunchKernelGGL(MemCpy9, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernelGGL(MemSet9, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipMemcpy(B, Bd, LEN9, hipMemcpyDeviceToHost);
hipMemcpy(C, Cd, LEN9, hipMemcpyDeviceToHost);
for (uint32_t i = 0; i < LEN9; i++) {
@@ -154,8 +154,8 @@ int main() {
hipMalloc((void**)&Bd, LEN10);
hipMalloc((void**)&Cd, LEN10);
hipMemcpy(Ad, A, LEN10, hipMemcpyHostToDevice);
hipLaunchKernel(MemCpy10, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernel(MemSet10, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipLaunchKernelGGL(MemCpy10, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernelGGL(MemSet10, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipMemcpy(B, Bd, LEN10, hipMemcpyDeviceToHost);
hipMemcpy(C, Cd, LEN10, hipMemcpyDeviceToHost);
for (uint32_t i = 0; i < LEN10; i++) {
@@ -182,8 +182,8 @@ int main() {
hipMalloc((void**)&Bd, LEN11);
hipMalloc((void**)&Cd, LEN11);
hipMemcpy(Ad, A, LEN11, hipMemcpyHostToDevice);
hipLaunchKernel(MemCpy11, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernel(MemSet11, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipLaunchKernelGGL(MemCpy11, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernelGGL(MemSet11, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipMemcpy(B, Bd, LEN11, hipMemcpyDeviceToHost);
hipMemcpy(C, Cd, LEN11, hipMemcpyDeviceToHost);
for (uint32_t i = 0; i < LEN11; i++) {
@@ -210,8 +210,8 @@ int main() {
hipMalloc((void**)&Bd, LEN12);
hipMalloc((void**)&Cd, LEN12);
hipMemcpy(Ad, A, LEN12, hipMemcpyHostToDevice);
hipLaunchKernel(MemCpy12, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernel(MemSet12, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipLaunchKernelGGL(MemCpy12, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Ad, Bd);
hipLaunchKernelGGL(MemSet12, dim3(2, 1, 1), dim3(2, 1, 1), 0, 0, Cd);
hipMemcpy(B, Bd, LEN12, hipMemcpyDeviceToHost);
hipMemcpy(C, Cd, LEN12, hipMemcpyDeviceToHost);
for (uint32_t i = 0; i < LEN12; i++) {
projects/clr/hipamd/tests/src/kernel/inline_asm_vadd.cpp
+2 -2
View File
@@ -33,7 +33,7 @@ OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWA
// Device (Kernel) function, it must be void
// hipLaunchParm provides the execution configuration
__global__ void vadd_asm(hipLaunchParm lp, float* out, float* in) {
__global__ void vadd_asm(float* out, float* in) {
int i = blockDim.x * blockIdx.x + threadIdx.x;
#ifdef __HIP_PLATFORM_NVCC__
@@ -82,7 +82,7 @@ int main() {
hipMemcpy(gpuResultVector, VectorB, NUM * sizeof(float), hipMemcpyHostToDevice);
// Lauching kernel from host
hipLaunchKernel(vadd_asm, dim3(NUM / THREADS_PER_BLOCK_X), dim3(THREADS_PER_BLOCK_X), 0, 0,
hipLaunchKernelGGL(vadd_asm, dim3(NUM / THREADS_PER_BLOCK_X), dim3(THREADS_PER_BLOCK_X), 0, 0,
gpuResultVector, gpuVector);
// Memory transfer from device to host