ops/declarable compiler warnings (deeplearning4j#10183)

Author: agibsonccc

libnd4j/include/memory/impl/MemoryTracker.cpp

@@ -106,7 +106,7 @@ void MemoryTracker::countIn(MemoryType type, Pointer ptr, LongType numBytes) {
 
     std::string stack("");
     messages = backtrace_symbols(array, size);
-    for (int i = 1; i < size && messages != NULL; ++i) {
+    for (size_t i = 1; i < size && messages != NULL; ++i) {
       stack += demangle(messages[i]) + "\n";
     }
 

libnd4j/include/ops/declarable/generic/bitwise/toggle_bits.cpp

@@ -30,7 +30,7 @@
 namespace sd {
 namespace ops {
 OP_IMPL(toggle_bits, -1, -1, true) {
-  for (int i = 0; i < block.width(); i++) {
+  for (size_t i = 0; i < block.width(); i++) {
     auto x = INPUT_VARIABLE(i);
     auto z = OUTPUT_VARIABLE(i);
 

libnd4j/include/ops/declarable/generic/blas/batched_gemm.cpp

@@ -139,7 +139,7 @@ CUSTOM_OP_IMPL(batched_gemm, -1, -1, false, 0, 9) {
     }
   }
 
-  REQUIRE_TRUE(vA.size() == vB.size() && vA.size() == vC.size() && vA.size() == batchSize, 0,
+  REQUIRE_TRUE(vA.size() == vB.size() && vA.size() == vC.size() && vA.size() == static_cast<size_t>(batchSize), 0,
                "BatchedGemm: mismatched numbers of A, B, C for unknown reason");
 
   helpers::bgemm(vA,
@@ -175,7 +175,7 @@ DECLARE_SHAPE_FN(batched_gemm) {
   auto firstInput = inputShape->at(2);
   auto secondInput =   inputShape->at(batchSize + 2);
   auto firstType = ArrayOptions::dataType(inputShape->at(0));
-  for (int e = 1; e < block.width(); e++) {
+  for (size_t e = 1; e < block.width(); e++) {
     REQUIRE_TRUE(firstType == ArrayOptions::dataType(inputShape->at(1)), 0,
                  "BatchedGemm: all inputs must have same data type");
   }

libnd4j/include/ops/declarable/generic/blas/tensormmul.cpp

@@ -110,7 +110,7 @@ struct IndexComparator
 std::vector<LongType> argsort(const std::vector<LongType>& array)
 {
   std::vector<LongType> indices(array.size());
-  for (LongType i = 0; i < array.size(); ++i) indices[i] = i;
+  for (size_t i = 0; i < array.size(); ++i) indices[i] = i;
 
   std::sort(indices.begin(), indices.end(), IndexComparator(array));
 
@@ -155,7 +155,7 @@ CUSTOM_OP_IMPL(tensormmul_bp, 4, 2, false, 0, -1) {
   for (LongType i = 0; i < Arank; ++i)
     axes_a_grad.push_back(i);
 
-  for (LongType i = 0; i < axes0Sum.size(); ++i)
+  for (size_t i = 0; i < axes0Sum.size(); ++i)
     axes_a_grad.erase(std::remove(axes_a_grad.begin(), axes_a_grad.end(), axes0Sum[i]), axes_a_grad.end());
 
 
@@ -165,7 +165,7 @@ CUSTOM_OP_IMPL(tensormmul_bp, 4, 2, false, 0, -1) {
   for (LongType i = 0; i < Brank; ++i)
     axes_b_grad.push_back(i);
 
-  for (LongType i = 0; i < axes1Sum.size(); ++i)
+  for (size_t i = 0; i < axes1Sum.size(); ++i)
     axes_b_grad.erase(std::remove(axes_b_grad.begin(), axes_b_grad.end(), axes1Sum[i]), axes_b_grad.end());
 
   //used for post result permute to reshape result to be expected output
@@ -185,12 +185,12 @@ CUSTOM_OP_IMPL(tensormmul_bp, 4, 2, false, 0, -1) {
   }
 
   std::vector<LongType> axes_b_gradA;
-  for (LongType i = 0; i < axes_b_grad.size(); i++) {
+  for (size_t i = 0; i < axes_b_grad.size(); i++) {
     axes_b_gradA.push_back(i);
   }
 
   std::vector<LongType> axes_a_gradB;
-  for (LongType i = 0; i < axes_a_grad.size(); i++) {
+  for (size_t i = 0; i < axes_a_grad.size(); i++) {
     axes_a_gradB.push_back(i);
   }
 

libnd4j/include/ops/declarable/generic/boolean/choose.cpp

@@ -59,26 +59,16 @@ DECLARE_TYPES(choose) {
 }
 
 DECLARE_SHAPE_FN(choose) {
-  LongType const* shape;
-  int rank;
   int mode = INT_ARG(0);
   auto numResults = NDArrayFactory::create<LongType>(0L);
   if (block.width() > 1) {
     auto first = INPUT_VARIABLE(0);
     auto second = INPUT_VARIABLE(1);
-    if (first->lengthOf() > second->lengthOf()) {
-      shape = first->shapeInfo();
-      rank = first->rankOf();
-    } else {
-      shape = second->shapeInfo();
-      rank = second->rankOf();
-    }
+
 
     helpers::chooseFunctorArray(block.launchContext(), first, second, mode, nullptr, &numResults);
   } else {
     auto first = INPUT_VARIABLE(0);
-    shape = first->shapeInfo();
-    rank = first->rankOf();
     double scalar = T_ARG(0);
 
     helpers::chooseFunctorScalar(block.launchContext(), first, scalar, mode, nullptr, &numResults);

libnd4j/include/ops/declarable/generic/boolean/where.cpp

@@ -108,7 +108,7 @@ DECLARE_SHAPE_FN(Where) {
     for (LongType i = 0; i < condition->lengthOf(); i++)
       if (condition->e<bool>(i)) numOfTrue++;
 
-    LongType const* theNewShape;
+    LongType * theNewShape;
     if (numOfTrue > 0) {
       LongType* newShape;
       ALLOCATE(newShape, block.getWorkspace(), shape::shapeInfoLength(2), sd::LongType);

libnd4j/include/ops/declarable/generic/broadcastable/multiply.cpp

@@ -36,7 +36,7 @@ BROADCASTABLE_OP_IMPL(multiply, 0, 0) {
 
   BROADCAST_CHECK_EMPTY(x, y, z);
 
-  const LongType* zShapeInfo = nullptr;
+   LongType* zShapeInfo = nullptr;
   const bool areShapesBroadcastable =
       ShapeUtils::evalBroadcastShapeInfo(x->shapeInfo(), y->shapeInfo(), true, zShapeInfo, block.getWorkspace());
   REQUIRE_TRUE(areShapesBroadcastable, 0, "MULTIPLY OP: the shapes of x %s and y %s are not suitable for broadcast !",
@@ -72,7 +72,7 @@ CUSTOM_OP_IMPL(multiply_bp, 3, 2, false, 0, 0) {
   auto dLdx = OUTPUT_VARIABLE(0);
   auto dLdy = OUTPUT_VARIABLE(1);
 
-  const LongType* dLdzShapeInfo = nullptr;
+   LongType* dLdzShapeInfo = nullptr;
   const bool areShapesBroadcastable =
       ShapeUtils::evalBroadcastShapeInfo(x->shapeInfo(), y->shapeInfo(), true, dLdzShapeInfo, block.getWorkspace());
   REQUIRE_TRUE(areShapesBroadcastable, 0,

libnd4j/include/ops/declarable/generic/broadcastable/pow.cpp

@@ -63,7 +63,7 @@ CUSTOM_OP_IMPL(Pow_bp, 3, 2, false, 0, 0) {
   auto dLdx = OUTPUT_VARIABLE(0);
   auto dLdy = OUTPUT_VARIABLE(1);
 
-  const LongType* dLdzShapeInfo = nullptr;
+   LongType* dLdzShapeInfo = nullptr;
   const bool areShapesBroadcastable =
       ShapeUtils::evalBroadcastShapeInfo(x->shapeInfo(), y->shapeInfo(), true, dLdzShapeInfo, block.getWorkspace());
   REQUIRE_TRUE(areShapesBroadcastable, 0,

libnd4j/include/ops/declarable/generic/datatypes/bitcast.cpp

@@ -39,7 +39,7 @@ CUSTOM_OP_IMPL(bitcast, 1, 1, false, 0, 1) {
   auto outputSize = DataTypeUtils::sizeOf(newType);
   auto lastSize = outputSize / inputSize;
   if (inputSize < outputSize) {
-    REQUIRE_TRUE(input->sizeAt(-1) == lastSize, 0,
+    REQUIRE_TRUE(static_cast<size_t>(input->sizeAt(-1)) == lastSize, 0,
                  "BITCAST: %llu > %llu. So last dimension should be %i, but %i given.", inputSize, outputSize, lastSize,
                  input->sizeAt(-1));
   }
@@ -84,12 +84,13 @@ DECLARE_SHAPE_FN(bitcast) {
     auto outputShape = ConstantShapeHelper::getInstance().bufferForShapeInfo(newType, shape::order(inShape), shapeOf)->primary();
     return SHAPELIST(outputShape);
   }
-  REQUIRE_TRUE(shape::sizeAt(inShape, static_cast<sd::LongType>(-1)) == outputSize / inputSize, 0,
+  REQUIRE_TRUE(shape::sizeAt(inShape, static_cast<size_t>(static_cast<sd::LongType>(-1))) ==
+                   static_cast<sd::LongType>(outputSize / inputSize), 0,
                "BITCAST: %llu > %llu. So last dimension should be %i, but %i given.", inputSize, outputSize,
                outputSize / inputSize, shape::sizeAt(inShape, static_cast<sd::LongType>(-1)));
   std::vector<sd::LongType> shapeOf(inputRank - 1);
 
-  for (auto i = 0; i < shapeOf.size(); ++i) {
+  for (size_t i = 0; i < shapeOf.size(); ++i) {
     shapeOf[i] = inShape[i + 1];
   }
 

libnd4j/include/ops/declarable/generic/images/adjust_contrast.cpp

@@ -53,7 +53,7 @@ CONFIGURABLE_OP_IMPL(adjust_contrast, 1, 1, true, 0, 0) {
 
   // fill up axes vector first
   std::vector<LongType> axes(input->rankOf() - 1);
-  for (auto i = 0; i < axes.size(); ++i) axes[i] = i;
+  for (size_t i = 0; i < axes.size(); ++i) axes[i] = i;
   // mean as reduction for last dimension set
   auto mean = input->reduceAlongDimension(reduce::Mean, &axes);
   auto part1 = (*input - mean);