have activelaneid_u32 call ockl_activelane_u32 (ROCm#1018)

include/hc.hpp

@@ -2817,7 +2817,10 @@ extern "C" inline __attribute((always_inline)) std::uint64_t __cycle_u64() __HC_
  *
  * @return The result will be in the range 0 to WAVESIZE - 1.
  */
-extern "C" unsigned int __activelaneid_u32() __HC__;
+extern "C" unsigned int __ockl_activelane_u32(void);
+extern "C" inline unsigned int __activelaneid_u32() __HC__ {
+	return __ockl_activelane_u32();
+}
 
 /**
  * Return a bit mask shows which active work-items in the

tests/Unit/AMDGPU/activelaneid.cpp

@@ -0,0 +1,107 @@
+
+// RUN: %hc %s -o %t.out && %t.out
+
+#include <hc.hpp>
+
+#include <iostream>
+#include <random>
+#include <vector>
+
+#define WAVEFRONT_SIZE (64) // as of now, all HSA agents have wavefront size of 64
+
+#define GRID_SIZE (WAVEFRONT_SIZE * WAVEFRONT_SIZE)
+
+#define TEST_DEBUG (0)
+
+// A test case to verify builtin function
+// - __activelaneid_u32
+
+// test __activelaneid_u32
+bool test() {
+  using namespace hc;
+  bool ret = true;
+
+  // initialize test data
+  // test is a table of size WAVEFRONT_SIZE * WAVEFRONT_SIZE
+  std::vector<uint32_t> test(GRID_SIZE);
+
+  std::random_device rd;
+  std::uniform_int_distribution<int> int_dist(0, WAVEFRONT_SIZE - 1);
+
+  // for each block of WAVEFRONT_SIZE, we randomly set 1s inside the block
+  // the number of 1s in the block equals to the index of the block
+  // (the 1st block of WAVEFRONT_SIZE has 0 1s, the 2nd block of WAVEFRONT_SIZE has 1 1, and so on)
+  for (int i = 0; i < WAVEFRONT_SIZE; ++i) {
+    for (int j = 0; j < WAVEFRONT_SIZE; ++j) {
+      if (j < i) {
+        test[i * WAVEFRONT_SIZE + j] = 1;
+      } else {
+        test[i * WAVEFRONT_SIZE + j] = 0;
+      }
+    }
+
+
+    for (int j = 0; j < WAVEFRONT_SIZE * 10; ++j) {
+      int k1 = int_dist(rd);
+      int k2 = int_dist(rd); 
+      if (k1 != k2) {
+        test[i * WAVEFRONT_SIZE + k1] ^= test[i * WAVEFRONT_SIZE + k2] ^= test[i * WAVEFRONT_SIZE + k1] ^= test[i * WAVEFRONT_SIZE + k2];     }
+    }
+
+  }
+
+#if TEST_DEBUG
+  for (int i = 0; i < WAVEFRONT_SIZE; ++i) {
+    for (int j = 0; j < WAVEFRONT_SIZE; ++j) {
+      std::cout << test[i * WAVEFRONT_SIZE + j] << " ";
+    }
+    std::cout << "\n";
+  }
+#endif
+
+  array<uint32_t, 1> test_GPU(GRID_SIZE);
+  copy(test.begin(), test_GPU);
+
+  array<uint32_t, 1> output_GPU(GRID_SIZE);
+  extent<1> ex(GRID_SIZE);
+  parallel_for_each(ex, [&](index<1>& idx) [[hc]] {
+    if (test_GPU[idx] == 1)
+      output_GPU(idx) = __activelaneid_u32();
+    else
+      output_GPU(idx) = 99;
+  }).wait();
+
+  // verify result
+  std::vector<uint32_t> output = output_GPU;
+  for (int i = 0; i < WAVEFRONT_SIZE; ++i) {
+    int activeLaneID = 0;
+    for (int j = 0; j < WAVEFRONT_SIZE; ++j) {
+       if (test[i * WAVEFRONT_SIZE +j] == 1) 
+         ret &= (output[i * WAVEFRONT_SIZE + j] == activeLaneID++);
+       else
+	 ret &= (output[i * WAVEFRONT_SIZE +j] == 99);
+
+#if TEST_DEBUG
+      if (!ret) {
+        std::cout << "FAILED: laneid " << activeLaneID << " "; 
+	ret = true;
+      }
+      std::cout << "i: " << i << " j: " << j << " input = " << test[i * WAVEFRONT_SIZE +j] << " j = " << j << " and output = " << output[i * WAVEFRONT_SIZE + j] << "\n";
+#endif
+    }
+#if TEST_DEBUG
+    std::cout << "\n";
+#endif
+  }
+
+  return ret;
+}
+
+int main() {
+  bool ret = true;
+
+  ret &= test();
+
+  return !(ret == true);
+}
+