Revert "Added support for work group collective functions in xasum routine."

This reverts commit 350e57a.

Author: EJainDev

src/kernels/level1/xasum.opencl

@@ -14,10 +14,6 @@
 // literal). Comment-out this line for syntax-highlighting when developing.
 R"(
 
-#ifdef cl_khr_work_group_uniform_arithmetic
-#pragma OPENCL EXTENSION cl_khr_work_group_uniform_arithmetic : enable
-#endif
-
 // Parameters set by the tuner or by the database. Here they are given a basic default value in case
 // this kernel file is used outside of the CLBlast library.
 #ifndef WGS1
@@ -59,25 +55,18 @@ void Xasum(const int n,
   lm[lid] = acc;
   barrier(CLK_LOCAL_MEM_FENCE);
 
-  // Performs reduction in local memory and stores the per work group result
-  #if defined(cl_khr_work_group_uniform_arithmetic) || defined(__opencl_c_work_group_collective_functions)
-    real result = work_group_reduce_add(lm[lid])
-
-    if (lid == 0) {
-      output[wgid] = result;
-    }
-  #else
-    for (int s=WGS1/2; s>0; s=s>>1) {
-      if (lid < s) {
-        Add(lm[lid], lm[lid], lm[lid + s]);
-      }
-      barrier(CLK_LOCAL_MEM_FENCE);
+  // Performs reduction in local memory
+  for (int s=WGS1/2; s>0; s=s>>1) {
+    if (lid < s) {
+      Add(lm[lid], lm[lid], lm[lid + s]);
     }
+    barrier(CLK_LOCAL_MEM_FENCE);
+  }
 
-    if (lid == 0) {
-      output[wgid] = lm[0];
-    }
-  #endif
+  // Stores the per-workgroup result
+  if (lid == 0) {
+    output[wgid] = lm[0];
+  }
 }
 
 // =================================================================================================
@@ -98,35 +87,22 @@ void XasumEpilogue(const __global real* restrict input,
   Add(lm[lid], input[lid], input[lid + WGS2]);
   barrier(CLK_LOCAL_MEM_FENCE);
 
-  // Performs reduction in local memory and stores the final result of the absolute value
-  #if defined(cl_khr_work_group_uniform_arithmetic) || defined(__opencl_c_work_group_collective_functions)
-    real result = work_group_reduce_add(lm[lid])
-
-    if (lid == 0) {
-      #if (PRECISION == 3232 || PRECISION == 6464) && defined(ROUTINE_ASUM)
-        asum[asum_offset].x = real.x + real.y; // the result is a non-complex number
-      #else
-        asum[asum_offset] = real;
-      #endif
+  // Performs reduction in local memory
+  for (int s=WGS2/2; s>0; s=s>>1) {
+    if (lid < s) {
+      Add(lm[lid], lm[lid], lm[lid + s]);
     }
-  #else
-    for (int s=WGS1/2; s>0; s=s>>1) {
-      if (lid < s) {
-        Add(lm[lid], lm[lid], lm[lid + s]);
-      }
-      barrier(CLK_LOCAL_MEM_FENCE);
-    }
-
-    if (lid == 0) {
-      #if (PRECISION == 3232 || PRECISION == 6464) && defined(ROUTINE_ASUM)
-        asum[asum_offset].x = lm[0].x + lm[0].y; // the result is a non-complex number
-      #else
-        asum[asum_offset] = lm[0];
-      #endif
-    }
-  #endif
+    barrier(CLK_LOCAL_MEM_FENCE);
+  }
 
   // Computes the absolute value and stores the final result
+  if (lid == 0) {
+    #if (PRECISION == 3232 || PRECISION == 6464) && defined(ROUTINE_ASUM)
+      asum[asum_offset].x = lm[0].x + lm[0].y; // the result is a non-complex number
+    #else
+      asum[asum_offset] = lm[0];
+    #endif
+  }
 }
 
 // =================================================================================================