================
@@ -0,0 +1,72 @@
+// This test checks that #pragma omp target update to(data[0:4:3]) correctly
+// updates every third element (stride 3) from the host to the device,
partially
+// across the array
+
+// RUN: %libomptarget-compile-run-and-check-generic
+#include <omp.h>
+#include <stdio.h>
+
+int main() {
+ int len = 11;
+ double data[len];
+
+ // Initialize on host
+ for (int i = 0; i < len; i++)
+ data[i] = i;
+
+ // Initial values
+ printf("original host array values:\n");
+ for (int i = 0; i < len; i++)
+ printf("%f\n", data[i]);
+ printf("\n");
+
+ // CHECK: 0.000000
+ // CHECK: 1.000000
+ // CHECK: 2.000000
+ // CHECK: 3.000000
+ // CHECK: 4.000000
+ // CHECK: 5.000000
+ // CHECK: 6.000000
+ // CHECK: 7.000000
+ // CHECK: 8.000000
+ // CHECK: 9.000000
+ // CHECK: 10.000000
+
+#pragma omp target data map(tofrom : data[0 : len])
+ {
+ // Initialize device array to 20
+#pragma omp target
+ for (int i = 0; i < len; i++)
+ data[i] = 20.0;
+
+ // Modify host data for strided elements
+ data[0] = 10.0;
+ data[3] = 10.0;
+ data[6] = 10.0;
+ data[9] = 10.0;
+
+#pragma omp target update to(data[0 : 4 : 3]) // indices 0,3,6,9
+
+ // Verify on device by adding 5
+#pragma omp target
+ for (int i = 0; i < len; i++)
+ data[i] += 5.0;
+ }
+
+ printf("device array values after update to:\n");
+ for (int i = 0; i < len; i++)
+ printf("%f\n", data[i]);
+ printf("\n");
+
+ // CHECK: 15.000000
+ // CHECK: 25.000000
+ // CHECK: 25.000000
+ // CHECK: 15.000000
+ // CHECK: 25.000000
+ // CHECK: 25.000000
+ // CHECK: 15.000000
+ // CHECK: 25.000000
+ // CHECK: 25.000000
+ // CHECK: 15.000000
+ // CHECK: 25.000000
----------------
amitamd7 wrote:
I agree with the stride 2 point but stride 3 tests a less obvious pattern
(non-power-of-2, doesn't evenly divide array length). Stride 2 exists Also
consistent with the already existent "partial" tests.
They affect different code paths in Clang - single vs. partial vs. multiple
strides affect AST handling, descriptor construction, and mapping components
differently. Separate approach also helps tracking issues & debugging.
https://github.com/llvm/llvm-project/pull/169623
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
