Fixed pre-processor warnings related to the subgroup shuffling

2 files changed, 30 insertions, 3 deletions

diff --git a/src/kernel_preprocessor.cpp b/src/kernel_preprocessor.cpp
index 1c422d33..bc2ab540 100644
--- a/src/kernel_preprocessor.cpp
+++ b/src/kernel_preprocessor.cpp
@@ -182,6 +182,31 @@ bool EvaluateCondition(std::string condition,
     const auto right = condition.substr(equal_pos + 4);
     return (left == right);
   }
+
+  // Process the not equal sign
+  const auto not_equal_pos = condition.find(" != ");
+  if (not_equal_pos != std::string::npos) {
+    const auto left = condition.substr(0, not_equal_pos);
+    const auto right = condition.substr(not_equal_pos + 4);
+    return (left != right);
+  }
+
+  // Process the smaller than sign
+  const auto smaller_than_pos = condition.find(" < ");
+  if (smaller_than_pos != std::string::npos) {
+    const auto left = condition.substr(0, smaller_than_pos);
+    const auto right = condition.substr(smaller_than_pos + 3);
+    return (left < right);
+  }
+
+  // Process the larger than sign
+  const auto larger_than_pos = condition.find(" > ");
+  if (larger_than_pos != std::string::npos) {
+    const auto left = condition.substr(0, larger_than_pos);
+    const auto right = condition.substr(larger_than_pos + 3);
+    return (left > right);
+  }
+
   printf("Warning unknown condition: %s\n", condition.c_str());
   return false; // unknown error
 }
diff --git a/src/kernels/level3/xgemm_part1.opencl b/src/kernels/level3/xgemm_part1.opencl
index 80a60107..cc03696e 100644
--- a/src/kernels/level3/xgemm_part1.opencl
+++ b/src/kernels/level3/xgemm_part1.opencl
@@ -126,13 +126,15 @@ R"(
 #endif
 
 // Intel subgroups (https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_subgroups.txt)
-#if USE_SUBGROUP_SHUFFLING == 1 && SUBGROUP_SHUFFLING_INTEL
+#if USE_SUBGROUP_SHUFFLING == 1 && SUBGROUP_SHUFFLING_INTEL == 1
   #define SUBGROUP_SIZE 8              // Assumes subgroup size is always 8 on Intel GPUs
 #endif
 
 // NVIDIA warps as subgroups using inline PTX (https://docs.nvidia.com/cuda/inline-ptx-assembly/index.html)
-#if USE_SUBGROUP_SHUFFLING == 1 && (SUBGROUP_SHUFFLING_NVIDIA_PRE_VOLTA || SUBGROUP_SHUFFLING_NVIDIA_POST_VOLTA)
-  #define SUBGROUP_SIZE 32              // Assumes subgroup size is always 32 on NVIDIA GPUs
+#if USE_SUBGROUP_SHUFFLING == 1
+  #if SUBGROUP_SHUFFLING_NVIDIA_PRE_VOLTA == 1 || SUBGROUP_SHUFFLING_NVIDIA_POST_VOLTA == 1
+    #define SUBGROUP_SIZE 32            // Assumes subgroup size is always 32 on NVIDIA GPUs
+  #endif
 #endif
 
 #if NWI != SUBGROUP_SIZE || MDIMC < SUBGROUP_SIZE