[OpenMP] Add tests for mapping of chained 'containing' structs #156703
+351
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-offload Author: Julian Brown (jtb20) ChangesThis PR adds several new tests for mapping of chained structures, i.e. those resembling: These are currently XFAILed, although the first two tests actually work with unified memory -- I'm not sure if it's possible to easily improve the condition on the XFAILs in question to make them more accurate. These cases are all fixed by the WIP PR #153683. Full diff: https://github.com/llvm/llvm-project/pull/156703.diff 3 Files Affected:
diff --git a/offload/test/mapping/chained_containing_structs_1.cc b/offload/test/mapping/chained_containing_structs_1.cc
new file mode 100644
index 0000000000000..9ca148a7823b0
--- /dev/null
+++ b/offload/test/mapping/chained_containing_structs_1.cc
@@ -0,0 +1,58 @@
+// RUN: %libomptarget-compilexx-and-run-generic
+// XFAIL: *
+
+#include <cstdlib>
+#include <cstdio>
+#include <cassert>
+
+struct S {
+ int a;
+ int b;
+ int c;
+};
+
+struct T {
+ S *s0;
+ S *s1;
+ S *s2;
+};
+
+int main() {
+ T *v = (T *) malloc (sizeof(T));
+ v->s0 = (S *) malloc (sizeof(S));
+ v->s1 = (S *) malloc (sizeof(S));
+ v->s2 = (S *) malloc (sizeof(S));
+ v->s0->a = 10;
+ v->s0->b = 10;
+ v->s0->c = 10;
+ v->s1->a = 20;
+ v->s1->b = 20;
+ v->s1->c = 20;
+ v->s2->a = 30;
+ v->s2->b = 30;
+ v->s2->c = 30;
+
+#pragma omp target map(to: v[:1]) map(tofrom: v->s1->b, v->s1->c, v->s2->b)
+ {
+ v->s1->b += 3;
+ v->s1->c += 5;
+ v->s2->b += 7;
+ }
+
+ assert (v->s0->a == 10);
+ assert (v->s0->b == 10);
+ assert (v->s0->c == 10);
+ assert (v->s1->a == 20);
+ assert (v->s1->b == 23);
+ assert (v->s1->c == 25);
+ assert (v->s2->a == 30);
+ assert (v->s2->b == 37);
+ assert (v->s2->c == 30);
+
+ free(v->s0);
+ free(v->s1);
+ free(v->s2);
+ free(v);
+
+ return 0;
+}
diff --git a/offload/test/mapping/chained_containing_structs_2.cc b/offload/test/mapping/chained_containing_structs_2.cc
new file mode 100644
index 0000000000000..22660a4fb8c3f
--- /dev/null
+++ b/offload/test/mapping/chained_containing_structs_2.cc
@@ -0,0 +1,76 @@
+// RUN: %libomptarget-compilexx-and-run-generic
+// XFAIL: *
+
+#include <cstdlib>
+#include <cstdio>
+#include <cassert>
+
+struct R {
+ int d;
+ int e;
+ int f;
+};
+
+struct S {
+ R *r0;
+ R *r1;
+ R *r2;
+};
+
+struct T {
+ S *s0;
+ S *s1;
+ S *s2;
+};
+
+int main() {
+ T *v = (T *) malloc (sizeof(T));
+
+ v->s0 = (S *) malloc (sizeof(S));
+ v->s1 = (S *) malloc (sizeof(S));
+ v->s2 = (S *) malloc (sizeof(S));
+
+ v->s0->r0 = (R *) calloc (1, sizeof(R));
+ v->s0->r1 = (R *) calloc (1, sizeof(R));
+ v->s0->r2 = (R *) calloc (1, sizeof(R));
+
+ v->s1->r0 = (R *) calloc (1, sizeof(R));
+ v->s1->r1 = (R *) calloc (1, sizeof(R));
+ v->s1->r2 = (R *) calloc (1, sizeof(R));
+
+ v->s2->r0 = (R *) calloc (1, sizeof(R));
+ v->s2->r1 = (R *) calloc (1, sizeof(R));
+ v->s2->r2 = (R *) calloc (1, sizeof(R));
+
+ #pragma omp target map(to: v->s1, v->s2, *v->s1, v->s1->r1, *v->s2, v->s2->r0) \
+ map(tofrom: v->s1->r1->d, v->s1->r1->e, v->s1->r2->d, v->s1->r2->f, v->s2->r0->e)
+ {
+ v->s1->r1->d += 3;
+ v->s1->r1->e += 5;
+ v->s1->r2->d += 7;
+ v->s1->r2->f += 9;
+ v->s2->r0->e += 11;
+ }
+
+ assert(v->s1->r1->d == 3);
+ assert(v->s1->r1->e == 5);
+ assert(v->s1->r2->d == 7);
+ assert(v->s1->r2->f == 9);
+ assert(v->s2->r0->e == 11);
+
+ free(v->s0->r0);
+ free(v->s0->r1);
+ free(v->s0->r2);
+ free(v->s1->r0);
+ free(v->s1->r1);
+ free(v->s1->r2);
+ free(v->s2->r0);
+ free(v->s2->r1);
+ free(v->s2->r2);
+ free(v->s0);
+ free(v->s1);
+ free(v->s2);
+ free(v);
+
+ return 0;
+}
diff --git a/offload/test/mapping/chained_containing_structs_3.cc b/offload/test/mapping/chained_containing_structs_3.cc
new file mode 100644
index 0000000000000..406da074d9361
--- /dev/null
+++ b/offload/test/mapping/chained_containing_structs_3.cc
@@ -0,0 +1,222 @@
+// RUN: %libomptarget-compilexx-and-run-generic
+// XFAIL: *
+
+#include <cstdlib>
+#include <cstdio>
+#include <cassert>
+#include <cstring>
+
+#include <omp.h>
+
+struct R {
+ int d;
+ int e;
+ int f;
+};
+
+struct S {
+ int a;
+ int b;
+ struct {
+ int c;
+ R r;
+ R *rp;
+ } sub;
+ int g;
+};
+
+struct T {
+ int a;
+ int *ptr;
+ int b;
+};
+
+int main() {
+ R r;
+ R *rp = new R;
+ S s;
+ S *sp = new S;
+ T t;
+ T *tp = new T;
+
+ memset(&r, 0, sizeof(R));
+ memset(rp, 0, sizeof(R));
+ memset(&s, 0, sizeof(S));
+ memset(sp, 0, sizeof(S));
+ memset(&t, 0, sizeof(T));
+ memset(tp, 0, sizeof(T));
+
+ s.sub.rp = new R;
+ sp->sub.rp = new R;
+
+ memset(s.sub.rp, 0, sizeof(R));
+ memset(sp->sub.rp, 0, sizeof(R));
+
+ t.ptr = new int[10];
+ tp->ptr = new int[10];
+
+ memset(t.ptr, 0, sizeof(int)*10);
+ memset(tp->ptr, 0, sizeof(int)*10);
+
+#pragma omp target map(tofrom: r) map(tofrom: r.e)
+{
+ r.d++;
+ r.e += 2;
+ r.f += 3;
+}
+ assert(r.d == 1);
+ assert(r.e == 2);
+ assert(r.f == 3);
+
+#pragma omp target map(tofrom: rp[:1]) map(tofrom: rp->e)
+{
+ rp->d++;
+ rp->e += 2;
+ rp->f += 3;
+}
+
+ assert(rp->d == 1);
+ assert(rp->e == 2);
+ assert(rp->f == 3);
+
+ int v;
+ int *orig_addr_v = &v;
+ bool separate_memory_space;
+
+#pragma omp target data use_device_addr(v)
+ {
+ void *mapped_ptr_v =
+ omp_get_mapped_ptr(orig_addr_v, omp_get_default_device());
+ separate_memory_space = mapped_ptr_v != orig_addr_v;
+ }
+
+#pragma omp target map(to: s) map(tofrom: s.sub.r.e)
+{
+ s.b++;
+ s.sub.r.d+=2;
+ s.sub.r.e+=3;
+ s.sub.r.f+=4;
+}
+
+ if (separate_memory_space) {
+ assert(s.b == 0);
+ assert(s.sub.r.d == 0);
+ assert(s.sub.r.e == 3);
+ assert(s.sub.r.f == 0);
+ } else {
+ assert(s.b == 1);
+ assert(s.sub.r.d == 2);
+ assert(s.sub.r.e == 3);
+ assert(s.sub.r.f == 4);
+ }
+
+#pragma omp target map(to: s, s.b) map(to: s.sub.rp[:1]) map(tofrom: s.sub.rp->e)
+{
+ s.b++;
+ s.sub.rp->d+=2;
+ s.sub.rp->e+=3;
+ s.sub.rp->f+=4;
+}
+
+ if (separate_memory_space) {
+ assert(s.b == 0);
+ assert(s.sub.rp->d == 0);
+ assert(s.sub.rp->e == 3);
+ assert(s.sub.rp->f == 0);
+ } else {
+ assert(s.b == 2);
+ assert(s.sub.rp->d == 2);
+ assert(s.sub.rp->e == 3);
+ assert(s.sub.rp->f == 4);
+ }
+
+#pragma omp target map(to: sp[:1]) map(tofrom: sp->sub.r.e)
+{
+ sp->b++;
+ sp->sub.r.d+=2;
+ sp->sub.r.e+=3;
+ sp->sub.r.f+=4;
+}
+
+ if (separate_memory_space) {
+ assert(sp->b == 0);
+ assert(sp->sub.r.d == 0);
+ assert(sp->sub.r.e == 3);
+ assert(sp->sub.r.f == 0);
+ } else {
+ assert(sp->b == 1);
+ assert(sp->sub.r.d == 2);
+ assert(sp->sub.r.e == 3);
+ assert(sp->sub.r.f == 4);
+ }
+
+#pragma omp target map(to: sp[:1]) map(to: sp->sub.rp[:1]) map(tofrom: sp->sub.rp->e)
+{
+ sp->b++;
+ sp->sub.rp->d+=2;
+ sp->sub.rp->e+=3;
+ sp->sub.rp->f+=4;
+}
+
+ if (separate_memory_space) {
+ assert(sp->b == 0);
+ assert(sp->sub.rp->d == 0);
+ assert(sp->sub.rp->e == 3);
+ assert(sp->sub.rp->f == 0);
+ } else {
+ assert(sp->b == 2);
+ assert(sp->sub.rp->d == 2);
+ assert(sp->sub.rp->e == 3);
+ assert(sp->sub.rp->f == 4);
+ }
+
+#pragma omp target map(tofrom: t) map(tofrom: t.ptr[2:1])
+{
+ t.a++;
+ t.ptr[2]+=2;
+ t.b+=3;
+}
+
+ assert(t.a == 1);
+ assert(t.ptr[2] == 2);
+ assert(t.b == 3);
+
+#pragma omp target map(tofrom: t) map(tofrom: t.a)
+{
+ t.b++;
+}
+
+ assert(t.b == 4);
+
+#pragma omp target map(tofrom: t) map(tofrom: t.ptr[2:1], t.a)
+{
+ t.a++;
+ t.ptr[2]+=2;
+ t.b+=3;
+}
+
+ assert(t.a == 2);
+ assert(t.ptr[2] == 4);
+ assert(t.b == 7);
+
+#pragma omp target map(tofrom: t) map(tofrom: t.ptr[2:1], t.a)
+{
+ /* Empty */
+}
+
+ assert(t.a == 2);
+ assert(t.ptr[2] == 4);
+ assert(t.b == 7);
+
+ delete s.sub.rp;
+ delete sp->sub.rp;
+
+ delete[] t.ptr;
+ delete[] tp->ptr;
+
+ delete rp;
+ delete sp;
+ delete tp;
+
+ return 0;
+}
|
abhinavgaba
reviewed
Sep 3, 2025
| v->s2->b += 7; | ||
| } | ||
|
|
||
| assert (v->s0->a == 10); |
There was a problem hiding this comment.
Up to you, but it might be better to do something like:
printf("%d\n", v->s0->b); // CHECK: 10That makes it easier to glance at failing test logs to see the expected/actual value.
There was a problem hiding this comment.
I've done this, after puzzling for a while how to deal with different output for (effectively runtime-dependent) unified vs. separate address spaces for mapping... I couldn't find anything that does that in the existing offload/mapping tests, though I may have missed something. Anyway, this new version seems to work.
| int *orig_addr_v = &v; | ||
| bool separate_memory_space; | ||
|
|
||
| #pragma omp target data use_device_addr(v) |
There was a problem hiding this comment.
Did you intend to have a map(v) here? otherwise the use_device_addr lookup will always fail. Also, v is not used within the scope of this region, so use_device_addr should not be needed.
This PR adds several new tests for mapping of chained structures, i.e. those resembling: #pragma omp target map(tofrom: a->b->c) These are currently XFAILed, although the first two tests actually work with unified memory -- I'm not sure if it's possible to easily improve the condition on the XFAILs in question to make them more accurate. These cases are all fixed by the WIP PR llvm#153683.
6bb40a8 to
9df11a3
Compare
abhinavgaba
reviewed
Sep 4, 2025
| { | ||
| void *mapped_ptr_v = | ||
| omp_get_mapped_ptr(orig_addr_v, omp_get_default_device()); | ||
| separate_memory_space = mapped_ptr_v != (void*) orig_addr_v; |
There was a problem hiding this comment.
I'm not sure if this is a reliable way to check for unified shared memory between device and host. omp_get_default_device may return the same value as omp_get_initial_device, if no device is available, in which case we're using host-fallback.
@dreachem, is that correct? Is there a better way to check for shared memory?
If not, it might be better to create a copy of the test in unified_shared_memory, and add pragma omp requires unified_shared_memory.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds several new tests for mapping of chained structures, i.e. those resembling:
These are currently XFAILed, although the first two tests actually work with unified memory -- I'm not sure if it's possible to easily improve the condition on the XFAILs in question to make them more accurate.
These cases are all fixed by the WIP PR #153683.