ae99966a27
This patch enables applications that did not request OpenMP unified_shared_memory to run with the same zero-copy behavior, where mapped memory does not result in extra memory allocations and memory copies, but CPU-allocated memory is accessed from the device. The name for this behavior is "automatic zero-copy" and it relies on detecting: that the runtime is running on a MI300A, that the user did not select unified_shared_memory in their program, and that XNACK (unified memory support) is enabled in the current GPU configuration. If all these conditions are met, then automatic zero-copy is triggered. This patch also introduces an environment variable OMPX_APU_MAPS that, if set, triggers automatic zero-copy also on non APU GPUs (e.g., on discrete GPUs). This patch is still missing support for global variables, which will be provided in a subsequent patch. Co-authored-by: Thorsten Blass <thorsten.blass@amd.com>
58 lines
1.9 KiB
C++
58 lines
1.9 KiB
C++
// clang-format off
|
|
// RUN: %libomptarget-compilexx-generic
|
|
// RUN: env OMPX_APU_MAPS=1 HSA_XNACK=1 LIBOMPTARGET_INFO=30 %libomptarget-run-generic 2>&1 \
|
|
// RUN: | %fcheck-generic -check-prefix=INFO_ZERO -check-prefix=CHECK
|
|
|
|
// RUN: %libomptarget-compilexx-generic
|
|
// RUN: env HSA_XNACK=0 LIBOMPTARGET_INFO=30 %libomptarget-run-generic 2>&1 \
|
|
// RUN: | %fcheck-generic -check-prefix=INFO_COPY -check-prefix=CHECK
|
|
|
|
// UNSUPPORTED: aarch64-unknown-linux-gnu
|
|
// UNSUPPORTED: aarch64-unknown-linux-gnu-LTO
|
|
// UNSUPPORTED: nvptx64-nvidia-cuda
|
|
// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
|
|
// UNSUPPORTED: x86_64-pc-linux-gnu
|
|
// UNSUPPORTED: x86_64-pc-linux-gnu-LTO
|
|
|
|
// REQUIRES: unified_shared_memory
|
|
|
|
// clang-format on
|
|
|
|
#include <cstdio>
|
|
|
|
int main() {
|
|
int n = 1024;
|
|
|
|
// test various mapping types
|
|
int *a = new int[n];
|
|
int k = 3;
|
|
int b[n];
|
|
|
|
for (int i = 0; i < n; i++)
|
|
b[i] = i;
|
|
|
|
// clang-format off
|
|
// INFO_ZERO: Return HstPtrBegin 0x{{.*}} Size=4096 for unified shared memory
|
|
// INFO_ZERO: Return HstPtrBegin 0x{{.*}} Size=4096 for unified shared memory
|
|
|
|
// INFO_COPY: Creating new map entry with HstPtrBase=0x{{.*}}, HstPtrBegin=0x{{.*}}, TgtAllocBegin=0x{{.*}}, TgtPtrBegin=0x{{.*}}, Size=4096,
|
|
// INFO_COPY: Creating new map entry with HstPtrBase=0x{{.*}}, HstPtrBegin=0x{{.*}}, TgtAllocBegin=0x{{.*}}, TgtPtrBegin=0x{{.*}}, Size=4096,
|
|
// INFO_COPY: Mapping exists with HstPtrBegin=0x{{.*}}, TgtPtrBegin=0x{{.*}}, Size=4096, DynRefCount=1 (update suppressed)
|
|
// INFO_COPY: Mapping exists with HstPtrBegin=0x{{.*}}, TgtPtrBegin=0x{{.*}}, Size=4096, DynRefCount=1 (update suppressed)
|
|
// clang-format on
|
|
#pragma omp target teams distribute parallel for map(tofrom : a[ : n]) \
|
|
map(to : b[ : n])
|
|
for (int i = 0; i < n; i++)
|
|
a[i] = i + b[i] + k;
|
|
|
|
int err = 0;
|
|
for (int i = 0; i < n; i++)
|
|
if (a[i] != i + b[i] + k)
|
|
err++;
|
|
|
|
// CHECK: PASS
|
|
if (err == 0)
|
|
printf("PASS\n");
|
|
return err;
|
|
}
|