From d827beb0950e656e76685b10c2282ad83366c250 Mon Sep 17 00:00:00 2001
From: James Osborn <osborn@alcf.anl.gov>
Date: Mon, 18 Sep 2023 23:05:38 -0500
Subject: [PATCH] restore needsFullBlock for SharedMemoryCache

---
 include/dslash_helper.cuh                         | 6 +++---
 include/targets/generic/special_ops.h             | 6 ++++++
 include/targets/sycl/shared_memory_cache_helper.h | 6 +++++-
 include/targets/sycl/thread_array.h               | 6 +++---
 include/targets/sycl/thread_local_cache.h         | 6 +++---
 5 files changed, 20 insertions(+), 10 deletions(-)

diff --git a/include/dslash_helper.cuh b/include/dslash_helper.cuh
index f6e5776f86..8a6261699d 100644
--- a/include/dslash_helper.cuh
+++ b/include/dslash_helper.cuh
@@ -702,15 +702,15 @@ namespace quda
 
 #ifdef QUDA_DSLASH_FAST_COMPILE
 	if constexpr (allthreads) {
-	  dslash.template operator()<kernel_type == UBER_KERNEL ? INTERIOR_KERNEL : kernel_type, allthreads>(x_cb, s, parity, active);
+	  dslash.template operator()<kernel_type == UBER_KERNEL ? INTERIOR_KERNEL : kernel_type, true>(x_cb, s, parity, active);
 	} else {
 	  dslash.template operator()<kernel_type == UBER_KERNEL ? INTERIOR_KERNEL : kernel_type>(x_cb, s, parity);
 	}
 #else
 	if constexpr (allthreads) {
 	  switch (parity) {
-	  case 0: dslash.template operator()<kernel_type == UBER_KERNEL ? INTERIOR_KERNEL : kernel_type, allthreads>(x_cb, s, 0, active); break;
-	  case 1: dslash.template operator()<kernel_type == UBER_KERNEL ? INTERIOR_KERNEL : kernel_type, allthreads>(x_cb, s, 1, active); break;
+	  case 0: dslash.template operator()<kernel_type == UBER_KERNEL ? INTERIOR_KERNEL : kernel_type, true>(x_cb, s, 0, active); break;
+	  case 1: dslash.template operator()<kernel_type == UBER_KERNEL ? INTERIOR_KERNEL : kernel_type, true>(x_cb, s, 1, active); break;
 	  }
 	} else {
 	  switch (parity) {
diff --git a/include/targets/generic/special_ops.h b/include/targets/generic/special_ops.h
index 363a6c49f9..b5a740c2ab 100644
--- a/include/targets/generic/special_ops.h
+++ b/include/targets/generic/special_ops.h
@@ -108,7 +108,13 @@ namespace quda {
   template <typename T> inline constexpr bool explicitSpecialOps = explicitSpecialOpsS<T>::value;
 
   // hasSpecialOps
+#if 1
   template <typename T> inline constexpr bool hasSpecialOps = !std::is_same_v<getSpecialOps<T>,NoSpecialOps>;
+#else
+  template <typename T> struct hasSpecialOpsImpl { static constexpr bool value = false; };
+  template <typename ...U> struct hasSpecialOpsImpl<SpecialOps<U...>> { static constexpr bool value = true; };
+  template <typename T> inline constexpr bool hasSpecialOps = hasSpecialOpsImpl<T>::value;
+#endif
 
   // combineOps
   template <typename ...T> struct combineOpsS {};
diff --git a/include/targets/sycl/shared_memory_cache_helper.h b/include/targets/sycl/shared_memory_cache_helper.h
index eb30b79416..09131c2c4e 100644
--- a/include/targets/sycl/shared_memory_cache_helper.h
+++ b/include/targets/sycl/shared_memory_cache_helper.h
@@ -4,6 +4,10 @@
 
 #include <../generic/shared_memory_cache_helper.h>
 
+namespace quda {
+  template <typename T, typename D, typename O> static constexpr bool needsFullBlock<SharedMemoryCache<T,D,O>> = true;
+}
+
 #else
 
 #include <target_device.h>
@@ -30,7 +34,7 @@ namespace quda
      This accessor supports both explicit run-time block size and
      compile-time sizing.
 
-     * For run-time block size, the constructor should be initialied
+     * For run-time block size, the constructor should be initialized
        with the desired block size.
 
      * For compile-time block size, no arguments should be passed to
diff --git a/include/targets/sycl/thread_array.h b/include/targets/sycl/thread_array.h
index b2cb58978b..18c6173279 100644
--- a/include/targets/sycl/thread_array.h
+++ b/include/targets/sycl/thread_array.h
@@ -19,6 +19,6 @@ namespace quda
 
 #endif
 
-namespace quda {
-  template <typename T, int n> static constexpr bool needsFullBlock<thread_array<T,n>> = false;
-}
+//namespace quda {
+//  template <typename T, int n> static constexpr bool needsFullBlock<thread_array<T,n>> = false;
+//}
diff --git a/include/targets/sycl/thread_local_cache.h b/include/targets/sycl/thread_local_cache.h
index dac0eb0681..daff28bf06 100644
--- a/include/targets/sycl/thread_local_cache.h
+++ b/include/targets/sycl/thread_local_cache.h
@@ -6,6 +6,6 @@
 #include "../generic/thread_local_cache_noshared.h"
 #endif
 
-namespace quda {
-  template <typename T, int N, typename O> static constexpr bool needsFullBlock<ThreadLocalCache<T,N,O>> = false;
-}
+//namespace quda {
+//  template <typename T, int N, typename O> static constexpr bool needsFullBlock<ThreadLocalCache<T,N,O>> = false;
+//}