Observed issues › harvest/cce-c-implicit-struct-map-partly-present
craycc stops with "overlaps present region but is not contained" when a target region implicitly maps a partly mapped structure
Legality: contested. An independent adjudicating reading finds the result not required: The restriction on partially present storage covers only list items with an explicit data-mapping attribute, so it does not apply to the implicit map of s, and no passage requires success when a larger overlapping object is implicitly mapped. How this is checked
Affected variants
| Compiler | MI300A gfx942 unified memory (XNACK 1) | MI300A gfx942 copy (XNACK 0) | MI250X gfx90a copy (XNACK 0) | MI300X gfx942 copy (XNACK 0) | MI350X gfx950 copy (XNACK 0) | A100 sm_80 separate memory | A100 sm_80 managed memory |
|---|---|---|---|---|---|---|---|
| AFAR 24.3.0 | correct | correct | unknown not run | correct | correct | n/a | n/a |
| CCE 19.0.0 | behaves incorrectly | behaves incorrectly | unknown not run | unknown not run | unknown not run | n/a | n/a |
| CCE 20.0.2 | behaves incorrectly | behaves incorrectly | unknown not run | unknown not run | unknown not run | n/a | n/a |
| CCE 21.0.3 | behaves incorrectly | behaves incorrectly | unknown not run | unknown not run | unknown not run | n/a | n/a |
| CCE 21.0.4 | behaves incorrectly | behaves incorrectly | unknown not run | unknown not run | unknown not run | n/a | n/a |
| NVHPC 25.11 | n/a | n/a | n/a | n/a | n/a | fails another way | fails another way |
| NVHPC 26.9 | n/a | n/a | n/a | n/a | n/a | fails another way | fails another way |
| GCC 16.2.0 | fails another way | fails another way | fails another way | fails another way | fails another way | fails another way | n/a |
"behaves incorrectly" / "not supported": this issue, confirmed in up to four environments (another build directory, another node, a repeat). "fails another way": a different failure, not counted as this issue. "unknown": not run there, or the runs disagree.
Messages
| Result | Compiler and GPU | Message |
|---|---|---|
| behaves incorrectly | CCE 19.0.0, MI300A unified CCE 19.0.0, MI300A copy | showACC: libcrayacc/acc_present.c:762 CRAY_ACC_ERROR - Host region (7ffffffc7bd0 to 7fffffff8920) overlaps present region (7fffffff8910 to 7fffffff8920 index 0) but is not contained for 'unknown' from sit |
| behaves incorrectly | CCE 20.0.2, MI300A unified CCE 20.0.2, MI300A copy | showACC: libcrayacc/acc_present.c:762 CRAY_ACC_ERROR - Host region (7ffffffc7db0 to 7fffffff8b00) overlaps present region (7fffffff8af0 to 7fffffff8b00 index 0) but is not contained for 's' from site-harv |
| behaves incorrectly | CCE 21.0.3, MI300A unified CCE 21.0.3, MI300A copy | showACC: libcrayacc/acc_present.c:762 CRAY_ACC_ERROR - Host region (7ffffffc7d40 to 7fffffff8a90) overlaps present region (7fffffff8a80 to 7fffffff8a90 index 0) but is not contained for 's' from site-harv |
| behaves incorrectly | CCE 21.0.4, MI300A unified CCE 21.0.4, MI300A copy | showACC: libcrayacc/acc_present.c:762 CRAY_ACC_ERROR - Host region (7ffffffc7d30 to 7fffffff8a80) overlaps present region (7fffffff8a70 to 7fffffff8a80 index 0) but is not contained for 's' from site-harv |
| fails another way | NVHPC 25.11, A100 separate | showPresent table dump for device[1]: NVIDIA Tesla GPU 0, compute capability 8.0, threadid=1 Hint: specify 0x800 bit in NV_ACC_DEBUG for verbose info. host:0x294cc100 device:0x71f6272fa000 size:400 presentcount:0+2 line:64 name:s.p[:100] host:0x7fffa39d1b88 device:0x71f6272fa200 size:4 presentcount:0+1 line:64 name:s.a host:0x7fffa39d1b8c device:0x71f6272fa400 size:4 presentcount:0+1 line:64 name:s.b allocated block device:0x71f6272fa000 size:512 thread:1 allocated block device:0x71f6272fa200 size:512 thread:1 allocated block device:0x71f6272fa400 size:512 thread:1 |
| fails another way | NVHPC 25.11, A100 managed | showPresent table dump for device[1]: NVIDIA Tesla GPU 0, compute capability 8.0, threadid=1 Hint: specify 0x800 bit in NV_ACC_DEBUG for verbose info. host:0x7fffaa284fa8 device:0x7544652fa000 size:4 presentcount:0+1 line:64 name:s.a host:0x7fffaa284fac device:0x7544652fa200 size:4 presentcount:0+1 line:64 name:s.b allocated block device:0x7544652fa000 size:512 thread:1 allocated block device:0x7544652fa200 size:512 thread:1 Present table errors: s lives at 0x7fffaa254268 size 200016 partially present in |
| fails another way | NVHPC 26.9, A100 separate | showPresent table dump for device[1]: NVIDIA Tesla GPU 0, compute capability 8.0, threadid=1 Hint: specify 0x800 bit in NV_ACC_DEBUG for verbose info. host:0xeedc110 device:0x7c70ad2fa000 size:400 presentcount:0+2 line:64 name:s.p[:100] host:0x7ffd32713478 device:0x7c70ad2fa200 size:4 presentcount:0+1 line:64 name:s.a host:0x7ffd3271347c device:0x7c70ad2fa400 size:4 presentcount:0+1 line:64 name:s.b allocated block device:0x7c70ad2fa000 size:512 thread:1 allocated block device:0x7c70ad2fa200 size:512 thread:1 allocated block device:0x7c70ad2fa400 size:512 thread:1 |
| fails another way | NVHPC 26.9, A100 managed | showPresent table dump for device[1]: NVIDIA Tesla GPU 0, compute capability 8.0, threadid=1 Hint: specify 0x800 bit in NV_ACC_DEBUG for verbose info. host:0x7fff77bc5cb8 device:0x7bd47d2fa000 size:4 presentcount:0+1 line:64 name:s.a host:0x7fff77bc5cbc device:0x7bd47d2fa200 size:4 presentcount:0+1 line:64 name:s.b allocated block device:0x7bd47d2fa000 size:512 thread:1 allocated block device:0x7bd47d2fa200 size:512 thread:1 Present table errors: s lives at 0x7fff77b94f78 size 200016 partially present in |
| fails another way | GCC 16.2.0, MI300A unified GCC 16.2.0, MI300A copy | showsite-harvest--cce-c-implicit-struct-map-partly-present.c: In function ‘main’:
site-harvest--cce-c-implicit-struct-map-partly-present.c:89:9: internal compiler error: Segmentation fault
89 | #pragma omp target data map(s.p[:N], s.p, s.a, s.b)
| ^~~
0x1fb617d internal_error(char const*, ...)
<dir>:787
0xe5ac4f crash_signal
<dir>:325 |
| fails another way | GCC 16.2.0, MI250X | showsite-harvest--cce-c-implicit-struct-map-partly-present.c: In function ‘main’:
site-harvest--cce-c-implicit-struct-map-partly-present.c:62:9: internal compiler error: Segmentation fault
62 | #pragma omp target data map(s.p[:N], s.p, s.a, s.b)
| ^~~
0x1fb617d internal_error(char const*, ...)
<dir><host>-gcc-src/gcc-16.2.0/gcc/diagnostic-global-context.cc:787
0xe5ac4f crash_signal
<dir><host>-gcc-src/gcc-16.2.0/gcc/toplev.cc:325 |
| fails another way | GCC 16.2.0, MI300X GCC 16.2.0, MI350X | showsite-harvest--cce-c-implicit-struct-map-partly-present.c: In function ‘main’:
site-harvest--cce-c-implicit-struct-map-partly-present.c:89:9: internal compiler error: Segmentation fault
89 | #pragma omp target data map(s.p[:N], s.p, s.a, s.b)
| ^~~
0x1fc4d9d internal_error(char const*, ...)
<home>/<user>/gcc-work/gcc-16.2.0/gcc/diagnostic-global-context.cc:787
0xe4bebf crash_signal
<home>/<user>/gcc-work/gcc-16.2.0/gcc/toplev.cc:325 |
| fails another way | GCC 16.2.0, A100 separate | showsite-harvest--cce-c-implicit-struct-map-partly-present.c: In function ‘main’:
site-harvest--cce-c-implicit-struct-map-partly-present.c:62:9: internal compiler error: Segmentation fault
62 | #pragma omp target data map(s.p[:N], s.p, s.a, s.b)
| ^~~
0x208693b internal_error(char const*, ...)
<home>/gcc-build/gcc-16.2.0/gcc/diagnostic-global-context.cc:787
0xe91b23 crash_signal
<home>/gcc-build/gcc-16.2.0/gcc/toplev.cc:325 |
Reproduce
CCE 19.0.0, MI300A unified
module load cce/19.0.0 craype-accel-amd-gfx942 rocm/6.4.3 module unload cray-mpich craycc -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx942 -O2 repro.c -o repro HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro
CCE 20.0.2, MI300A unified
module load cce/20.0.2 craype-accel-amd-gfx942 rocm/6.4.3 module unload cray-mpich craycc -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx942 -O2 repro.c -o repro HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro
CCE 21.0.3, MI300A unified
module load cce/21.0.3 craype-accel-amd-gfx942 rocm/7.2.0 module unload cray-mpich craycc -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx942 -O2 repro.c -o repro HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro
CCE 21.0.4, MI300A unified
module load cce/21.0.4 craype-accel-amd-gfx942 rocm/7.2.0 module unload cray-mpich craycc -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx942 -O2 repro.c -o repro HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro
AFAR 24.3.0, MI300A unified (runs correctly)
$AFAR/bin/amdclang -fopenmp --offload-arch=gfx942 -O2 repro.c -o repro HSA_XNACK=1 OMP_TARGET_OFFLOAD=mandatory ./repro
Program (96 lines)
// # Finding: CCE craycc stops with "overlaps present region but is
// # not contained" when a target region implicitly maps a structure some of
// # whose members a target data construct already mapped
//
// - compilers: CCE craycc 19.0.0, 20.0.2, 21.0.3, 21.0.4, -O2: this file on
// MI300A gfx942 (HSA_XNACK 0 and 1); the OpenMP Examples program
// target_struct_map.4 (Case 1 of which this file is) also on MI250X gfx90a.
// - expected: 4 202 dev= 1 (s.p[0] = 0*2+4, s.p[99] = 99*2+4)
// - actual: no output; the runtime stops at the target construct:
// "ACC: libcrayacc/acc_present.c:762 CRAY_ACC_ERROR - Host region (... to
// ...) overlaps present region (... to ... index 0) but is not contained
// for 's'", exit status 1, every run, both memory modes.
// - rule: the target construct references s (the call saxpy(&s)), so s is
// treated as if it appeared in a map clause with map-type tofrom (5.2 §5.8.1
// p.149 l.4-5); s.a, s.b and s.p are already present as one storage block
// (list items with the same containing structure form "a single mappable
// storage block", 5.2 §5.8.3 p.152 l.28-31); and "If a single contiguous
// part of the original storage of a list item with an implicit data-mapping
// attribute has corresponding storage in the device data environment prior
// to a task encountering the construct on which the map clause appears,
// only that part of the original storage will have corresponding storage in
// the device data environment as a result of the map clause" (5.2 §5.8.3
// p.154 l.18-21; the same text in 5.0 §2.19.7.1 p.320 l.18-21). The
// OpenMP Examples document this form as working (target_struct_map.4,
// Case 1: "Explicitly mapping S1.p, S1.a, and S1.b rather than S1 avoids
// mapping the entire structure").
// - same-compiler controls (correct, 4 202 dev= 1 on CCE 19, 20, 21, 21.0.4):
// s.big added to the target data map (the whole structure present); no
// target data and map(s.p[:N], s.a, s.b) on the target construct (the
// Examples' Case 3, 5.2 §5.8.3 p.154 l.22-27). Both are files now
// (controls/cce-c-implicit-struct-map-partly-present-whole-control.c and
// controls/cce-c-implicit-struct-map-partly-present-target-only-control.c);
// re-run 2026-11-01 on MI300A, HSA_XNACK 0 and 1: correct on CCE 19, 20,
// 21, 21.0.4 and on AFAR, while this file fails on CCE 19, 20, 21, 21.0.4
// and is correct on AFAR (results/triage2/cce.jsonl).
// - also: the Examples' Case 2 (target data map(s.p[:N]), then target
// map(s.p[:0], s.a, s.b) with s implicit) stops the same way on every CCE
// cell; libgomp's target-link-2.c (declare target link(arr), enter data
// map(arr[5:10]), then a target region that references arr) stops with the
// same message on CCE 19, 20, 21 (gfx90a): an implicit map (5.2 §5.8.1
// p.148 l.9-11) of a partly present array.
// - other families: AFAR amdclang prints 4 202 dev= 1 (this file and the
// three Examples cases, gfx942 X0 and X1, gfx90a). NVHPC nvc 25.11 and 26.9
// (A100, separate and managed memory) fail this file at run time, exit
// status 1, with a different message: "FATAL ERROR: variable in data clause
// is partially present on the device: name=s" at the target construct (the
// present table lists s.a, s.b and, in separate memory, s.p[:100]). That is
// the OpenACC runtime's partial-presence error, and it reaches further than
// CCE's: NVHPC also fails the whole-structure control (s.big added, so every
// member is present, each as its own entry) that CCE and AFAR run correctly,
// and runs the target-only control correctly. All four gate environments
// agree on both releases and both memory modes (site matrix rows,
// 2026-11-01, c on another GPU; a rerun of the file and both controls, three
// runs each, in a, b, d on GPU 0 and c on GPU 1:
// results/triage2/wingtip/nvhpc/nv.[abcd].jsonl). The Examples program's
// SIGSEGV on NVHPC is the 8 MB stack (its three structures hold 12 MB): with
// a 64 MB stack (ulimit -s 65536) it stops at Case 1 with the same
// partial-presence error (name=S1; nv3.jsonl). The failure is NVHPC's own
// and not CCE's, so this finding stays CCE's and the site shows NVHPC as
// failing another way; an NVHPC finding (the whole-structure control is the
// cleaner reproducer) is not made here. GCC 16.2 does not compile this file:
// the target data map list is the trigger of an internal compiler error
// on both targets (gcc-c-target-data-struct-pointer-section-ice.c).
// - harvest: dev programs h2-ompex-target-struct-map-4-c and
// h2-gomp-ccc-target-link-2 (data/bench/harvest2), wrong_answer on every CCE
// cell, gate confirmed on gfx90a.
// - legality: offload.engine.prerun.check accepts it (the rules with the race
// rule, and the AFAR clang front end), and accepts both controls (re-checked
// 2026-11-01, Tuolumne pbatch; results/triage2/legal2.json). The rule it
// relies on is the one quoted above (5.2 §5.8.3 p.154 l.18-21: the part of
// an implicitly mapped list item that is present is the only part mapped);
// AFAR runs the file correctly, and the OpenMP Examples document the form.
// CCE: craycc -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa
// -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx942 -fopenmp-version=52 -O2
// Run: OMP_TARGET_OFFLOAD=mandatory ./a.out
#include <stdio.h>
#include <stdlib.h>
#include <omp.h>
#define N 100
struct foo { char big[200000]; float a, b; float *p; };
#pragma omp begin declare target
void saxpy(struct foo *s) { for (int i = 0; i < N; i++) s->p[i] = s->p[i] * s->a + s->b; }
#pragma omp end declare target
int main(void) {
struct foo s;
int dev = 0;
s.a = 2; s.b = 4; s.p = malloc(sizeof(float) * N);
for (int i = 0; i < N; i++) s.p[i] = i;
#pragma omp target data map(s.p[:N], s.p, s.a, s.b)
#pragma omp target map(from: dev)
{ saxpy(&s); dev = !omp_is_initial_device(); }
printf("%.0f %.0f dev= %d\n", s.p[0], s.p[N - 1], dev);
int ok = s.p[0] == 4 && s.p[N - 1] == 202;
printf("RESULT ran_on_host=%d check=%s\n", !dev, ok ? "pass" : "fail");
return 0;
}
Something to add? Dispute its legality · Report it fixed in a newer release · Report a mistake on this page
Expected and observed
Expected
4 202 dev= 1
Observed where it behaves incorrectly
ACC: libcrayacc/acc_present.c:762 CRAY_ACC_ERROR - Host region (7ffffffc7bd0 to 7fffffff8920) overlaps present region (7fffffff8910 to 7fffffff8920 index 0) but is not contained for 'unknown' from sit
What the specification requires
The specification text this program relies on and the compiler is expected to support.
OpenMP 5.2 §5.8.1 p.149 l.4-5
OpenMP 5.2 §5.8.3 p.152 l.28-31
a single mappable storage block
OpenMP 5.2 §5.8.3 p.154 l.18-21
If a single contiguous part of the original storage of a list item with an implicit data-mapping attribute has corresponding storage in the device data environment prior to a task encountering the construct on which the map clause appears, only that part of the original storage will have corresponding storage in the device data environment as a result of the map clause
OpenMP 5.0 §2.19.7.1 p.320 l.18-21
OpenMP 5.2 §5.8.3 p.154 l.22-27
OpenMP 5.2 §5.8.1 p.148 l.9-11
OpenMP 5.2 §5.8.3 p.154 l.18-21
Evidence
Other compilers run it correctly
AFAR 24.3.0