Enable variable-steps tests

Author: kurapov-peter

omniscidb/QueryEngine/Compiler/genx.cpp

@@ -187,8 +187,8 @@ int32_t atomicSum32SkipVal(GENERIC_ADDR_SPACE int32_t* addr,
 int64_t atomicSum64SkipVal(GENERIC_ADDR_SPACE int64_t* addr,
                            const int64_t val,
                            const int64_t skip_val) {
-  int32_t old = atomic_xchg_int_64(addr, 0);
-  int32_t old2 = agg_sum_shared(addr, old == skip_val ? val : (val + old));
+  int64_t old = atomic_xchg_int_64(addr, 0);
+  int64_t old2 = agg_sum_shared(addr, old == skip_val ? val : (val + old));
   return old == skip_val ? old2 : (old2 + old);
 }
 
@@ -212,6 +212,16 @@ int64_t agg_sum_int64_skip_val_shared(GENERIC_ADDR_SPACE int64_t* agg,
   return 0;
 }
 
+int64_t agg_sum_skip_val_shared(GENERIC_ADDR_SPACE int64_t* agg,
+                                const int64_t val,
+                                const int64_t skip_val) {
+  if (val != skip_val) {
+    const int64_t old = atomicSum64SkipVal(agg, val, skip_val);
+    return old;
+  }
+  return 0;
+}
+
 void agg_max_int32_skip_val_shared(GENERIC_ADDR_SPACE int32_t* agg,
                                    const int32_t val,
                                    const int32_t skip_val) {

omniscidb/QueryEngine/Compiler/genx.ll

@@ -167,20 +167,6 @@ define void @agg_sum_double_skip_val_shared(i64 addrspace(4)* %agg, double nound
     ret void
 }
 
-define i64 @agg_sum_skip_val_shared(i64 addrspace(4)* %agg, i64 noundef %val, i64 noundef %skip_val) {
-    %no_skip = icmp ne i64 %val, %skip_val
-    br i1 %no_skip, label %.noskip, label %.skip
-.noskip:
-    %old = atomicrmw xchg i64 addrspace(4)* %agg, i64 0 monotonic
-    %isempty = icmp eq i64 %old, -9223372036854775808
-    %sel = select i1 %isempty, i64 0, i64 %old
-    %new_val = add nsw i64 %val, %sel
-    %old2 = atomicrmw add i64 addrspace(4)* %agg, i64 %new_val monotonic
-    ret i64 %old2
-.skip:
-    ret i64 0
-}
-
 define void @atomic_or(i32 addrspace(4)* %addr, i32 noundef %val) {
 .entry:
     %orig = load atomic i32, i32 addrspace(4)* %addr unordered, align 8

omniscidb/Tests/L0SharedMemoryTest.cpp

@@ -368,13 +368,12 @@ TEST(SingleColumn, VariableSteps_FixedEntries_2) {
 }
 
 TEST(SingleColumn, VariableSteps_FixedEntries_3) {
-  GTEST_SKIP();
   TestInputData input;
   input.setDeviceId(0)
       .setNumInputBuffers(4)
       .setAggWidth(8)
       .setMinEntry(0)
-      .setMaxEntry(367)
+      .setMaxEntry(204)
       .setKeylessHash(true)
       .setTargetIndexForKey(0)
       .setTargetInfos(generate_custom_agg_target_infos(
@@ -394,13 +393,12 @@ TEST(SingleColumn, VariableSteps_FixedEntries_3) {
 }
 
 TEST(SingleColumn, VariableSteps_FixedEntries_4) {
-  GTEST_SKIP();
   TestInputData input;
   input.setDeviceId(0)
       .setNumInputBuffers(4)
       .setAggWidth(8)
       .setMinEntry(0)
-      .setMaxEntry(517)
+      .setMaxEntry(255)
       .setKeylessHash(true)
       .setTargetIndexForKey(0)
       .setTargetInfos(generate_custom_agg_target_infos(