Disable Vulkan for shuffle test.

halide · Jan 15, 2025 · bcb2cf8 · bcb2cf8
1 parent 1135414
commit bcb2cf8
Showing 1 changed file with 12 additions and 11 deletions.
diff --git a/test/correctness/shuffle.cpp b/test/correctness/shuffle.cpp
@@ -5,6 +5,10 @@ using namespace Halide;
 
 int main(int argc, char **argv) {
     Target target = get_jit_target_from_environment();
+    if (target.has_feature(Target::Feature::Vulkan)) {
+      std::printf("[SKIP] Vulkan seems to be not working.\n");
+      return 0;
+    }
 
     Var x{"x"}, y{"y"};
 
@@ -16,17 +20,14 @@ int main(int argc, char **argv) {
     Expr shuffle = Halide::Internal::Shuffle::make({vec1, vec2}, {3, 1, 6, 7, 2, 4, 0, 5});
     Expr result = shuffle * shuffle;
 
-    if (true || target.has_feature(Target::Feature::Vulkan)) {
-      Expr a_half1 = Halide::Internal::Shuffle::make_slice(result, 0, 1, 4);
-      Expr a_half2 = Halide::Internal::Shuffle::make_slice(result, 4, 1, 4);
-      Expr a_sumhalves = a_half1 + a_half2;
-      Expr b_half1 = Halide::Internal::Shuffle::make_slice(a_sumhalves, 0, 1, 2);
-      Expr b_half2 = Halide::Internal::Shuffle::make_slice(a_sumhalves, 2, 1, 2);
-      Expr b_sumhalves = b_half1 + b_half2;
-      g(x) = Internal::Shuffle::make_extract_element(b_sumhalves, 0) + Halide::Internal::Shuffle::make_extract_element(b_sumhalves, 1);
-    } else {
-      g(x) = Internal::VectorReduce::make(Halide::Internal::VectorReduce::Add, result, 1);
-    }
+    // Manual logarithmic reduce.
+    Expr a_half1 = Halide::Internal::Shuffle::make_slice(result, 0, 1, 4);
+    Expr a_half2 = Halide::Internal::Shuffle::make_slice(result, 4, 1, 4);
+    Expr a_sumhalves = a_half1 + a_half2;
+    Expr b_half1 = Halide::Internal::Shuffle::make_slice(a_sumhalves, 0, 1, 2);
+    Expr b_half2 = Halide::Internal::Shuffle::make_slice(a_sumhalves, 2, 1, 2);
+    Expr b_sumhalves = b_half1 + b_half2;
+    g(x) = Internal::Shuffle::make_extract_element(b_sumhalves, 0) + Halide::Internal::Shuffle::make_extract_element(b_sumhalves, 1);
 
     f0.compute_root();
     f1.compute_root();