add Shap Magic to check if use cat

src/predictor/gpu_predictor.cu

@@ -428,6 +428,8 @@ class DeviceModel {
   }
 };
 
+#define ShapSplitMagic 99999
+
 struct ShapSplitCondition {
   ShapSplitCondition() = default;
   XGBOOST_DEVICE
@@ -437,6 +439,7 @@ struct ShapSplitCondition {
         feature_upper_bound(feature_upper_bound),
         is_missing_branch(is_missing_branch), categories{std::move(cats)} {
     assert(feature_lower_bound <= feature_upper_bound);
+    cat_flag = ShapSplitMagic;
   }
 
   /*! Feature values >= lower and < upper flow down this path. */
@@ -444,6 +447,7 @@ struct ShapSplitCondition {
   float feature_upper_bound;
   /*! Feature value set to true flow down this path. */
   common::CatBitField categories;
+  int cat_flag;
   /*! Do missing values flow down this path? */
   bool is_missing_branch;
 
@@ -453,7 +457,7 @@ struct ShapSplitCondition {
     if (isnan(x)) {
       return is_missing_branch;
     }
-    if (categories.Size() != 0) {
+    if (cat_flag == ShapSplitMagic && categories.Size() != 0) {
       auto cat = static_cast<uint32_t>(x);
       return categories.Check(cat);
     } else {
@@ -480,7 +484,7 @@ struct ShapSplitCondition {
   // Combine two split conditions on the same feature
   XGBOOST_DEVICE void Merge(ShapSplitCondition other) {
     // Combine duplicate features
-    if (categories.Size() != 0 || other.categories.Size() != 0) {
+    if (cat_flag == ShapSplitMagic && (categories.Size() != 0 || other.categories.Size() != 0)) {
       categories = Intersect(categories, other.categories);
     } else {
       feature_lower_bound = max(feature_lower_bound, other.feature_lower_bound);

src/tree/gpu_hist/evaluate_splits.cu

@@ -14,6 +14,11 @@
 #if defined(XGBOOST_USE_HIP)
 #include <hip/hip_cooperative_groups.h>
 
+#ifdef __AMDGCN_WAVEFRONT_SIZE
+#undef WAVEFRONT_SIZE
+#define WAVEFRONT_SIZE __AMDGCN_WAVEFRONT_SIZE
+#endif
+
 #define WARP_SIZE WAVEFRONT_SIZE
 #elif defined(XGBOOST_USE_CUDA)
 #define WARP_SIZE 32