Merge branch 'master' of github.com:halide/Halide

jrprice · May 24, 2013 · 3b54fcd · 3b54fcd
2 parents 2526faa + c8f4c65
commit 3b54fcd
Show file tree

Hide file tree

Showing 22 changed files with 458 additions and 236 deletions.
diff --git a/cpp/Makefile b/cpp/Makefile
@@ -121,7 +121,10 @@ TESTS = $(shell ls test/*.cpp)
 ERROR_TESTS = $(shell ls test/error/*.cpp)
 
 # TODO: move this implementation into Makefile.tests which contains a .NOTPARALLEL rule?
-tests: $(TESTS:test/%.cpp=test_%) $(ERROR_TESTS:test/error/%.cpp=error_%)
+tests: build_tests run_tests
+
+run_tests: $(TESTS:test/%.cpp=test_%) $(ERROR_TESTS:test/error/%.cpp=error_%)
+build_tests: $(TESTS:test/%.cpp=$(BIN_DIR)/test_%) $(ERROR_TESTS:test/error/%.cpp=$(BIN_DIR)/error_%)
 
 $(BIN_DIR)/test_internal: test/internal.cpp $(BIN_DIR)/libHalide.so
 	$(CXX) $(CXX_FLAGS)  $< -Isrc -L$(BIN_DIR) -lHalide -lpthread -ldl -o $@	

diff --git a/cpp/src/Argument.h b/cpp/src/Argument.h
@@ -29,7 +29,7 @@ struct Argument {
     /** If this is a scalar parameter, then this is its type */
     Type type;
 
-    Argument() {}
+    Argument() : is_buffer(false) {}
     Argument(const std::string &_name, bool _is_buffer, Type _type) : 
         name(_name), is_buffer(_is_buffer), type(_type) {}
 };

diff --git a/cpp/src/Bounds.cpp b/cpp/src/Bounds.cpp
@@ -512,7 +512,7 @@ map<string, Region> compute_regions_touched(Stmt s, bool consider_calls, bool co
     s.accept(&r);
     map<string, Region> regions;
     for (map<string, vector<Interval> >::iterator iter = r.regions.begin(); 
-         iter != r.regions.end(); iter++) {
+         iter != r.regions.end(); ++iter) {
         Region region;
         const vector<Interval> &box = iter->second;
         for (size_t i = 0; i < box.size(); i++) {

diff --git a/cpp/src/CodeGen.cpp b/cpp/src/CodeGen.cpp
@@ -48,9 +48,13 @@ using std::stack;
 #define InitializeNVPTXTarget() InitializeTarget(NVPTX)
 
 CodeGen::CodeGen() : 
-    module(NULL), function(NULL), context(NULL), 
+    module(NULL), owns_module(false), 
+    function(NULL), context(NULL), 
     builder(NULL), 
-    value(NULL), buffer_t(NULL) {
+    value(NULL), 
+    void_t(NULL), i1(NULL), i8(NULL), i16(NULL), i32(NULL), i64(NULL),
+    f16(NULL), f32(NULL), f64(NULL),
+    buffer_t(NULL) {
 
     // Initialize the targets we want to generate code for which are enabled
     // in llvm configuration
@@ -1024,7 +1028,7 @@ void CodeGen::visit(const Call *op) {
     if (op->name == "debug to file") {
         assert(op->args.size() == 9);
         const Call *filename = op->args[0].as<Call>();
-        const Call *func = op->args[1].as<Call>();
+        const Load *func = op->args[1].as<Load>();
         assert(func && filename && "Malformed debug_to_file node");
         // Grab the function from the initial module
         llvm::Function *debug_to_file = module->getFunction("halide_debug_to_file");
@@ -1343,7 +1347,7 @@ void CodeGen::visit(const For *op) {
         do_par_for->setDoesNotAlias(4);
         //do_par_for->setDoesNotCapture(4);
         ptr = builder->CreatePointerCast(ptr, i8->getPointerTo());
-        vector<Value *> args = vec((Value *)function, min, extent, ptr);
+        vector<Value *> args = vec<Value *>(function, min, extent, ptr);
         log(4) << "Creating call to do_par_for\n";
         builder->CreateCall(do_par_for, args);
 

diff --git a/cpp/src/CodeGen_ARM.cpp b/cpp/src/CodeGen_ARM.cpp
@@ -600,13 +600,13 @@ void CodeGen_ARM::visit(const Div *op) {
         // Do the shift (add 8 or 16 to narrow back down)
         if (op->type == Int(32, 2) && shift == 0) {
             Constant *shift_amount = ConstantInt::get(wider, -32);
-            val = call_intrin(narrower, "vshiftn.v2i32", vec(wide_val, (Value *)shift_amount));
+            val = call_intrin(narrower, "vshiftn.v2i32", vec<Value *>(wide_val, shift_amount));
         } else if (op->type == Int(16, 4) && shift == 0) {
             Constant *shift_amount = ConstantInt::get(wider, -16);
-            val = call_intrin(narrower, "vshiftn.v4i16", vec(wide_val, (Value *)shift_amount));
+            val = call_intrin(narrower, "vshiftn.v4i16", vec<Value *>(wide_val, shift_amount));
         } else if (op->type == Int(8, 8) && shift == 0) {
             Constant *shift_amount = ConstantInt::get(wider, -8);
-            val = call_intrin(narrower, "vshiftn.v8i8", vec(wide_val, (Value *)shift_amount));
+            val = call_intrin(narrower, "vshiftn.v8i8", vec<Value *>(wide_val, shift_amount));
         } else {
             Constant *shift_amount = ConstantInt::get(wider, (shift + op->type.bits));
             val = builder->CreateLShr(wide_val, shift_amount);
@@ -649,13 +649,13 @@ void CodeGen_ARM::visit(const Div *op) {
         // Narrow 
         if (op->type == UInt(32, 2) && shift == 0) {
             Constant *shift_amount = ConstantInt::get(wider, -32);
-            val = call_intrin(narrower, "vshiftn.v2i32", vec(val, (Value *)shift_amount));
+            val = call_intrin(narrower, "vshiftn.v2i32", vec<Value *>(val, shift_amount));
         } else if (op->type == UInt(16, 4) && shift == 0) {
             Constant *shift_amount = ConstantInt::get(wider, -16);
-            val = call_intrin(narrower, "vshiftn.v4i16", vec(val, (Value *)shift_amount));
+            val = call_intrin(narrower, "vshiftn.v4i16", vec<Value *>(val, shift_amount));
         } else if (op->type == UInt(8, 8) && shift == 0) {
             Constant *shift_amount = ConstantInt::get(wider, -8);
-            val = call_intrin(narrower, "vshiftn.v8i8", vec(val, (Value *)shift_amount));
+            val = call_intrin(narrower, "vshiftn.v8i8", vec<Value *>(val, shift_amount));
         } else {
             int shift_bits = op->type.bits;
             // For methods 0 and 1, we can do the final shift here too
@@ -939,7 +939,7 @@ void CodeGen_ARM::visit(const Store *op) {
         return;
     }
 
-    if (ramp && is_one(ramp->stride) && 
+    if (is_one(ramp->stride) && 
         call && call->name == "interleave vectors") {
         assert(call->args.size() == 2 && "Wrong number of args to interleave vectors");
         vector<Value *> args(call->args.size() + 2);

diff --git a/cpp/src/CodeGen_C.cpp b/cpp/src/CodeGen_C.cpp
@@ -549,12 +549,10 @@ void CodeGen_C::visit(const Allocate *op) {
     do_indent();
     stream << print_type(op->type) << ' ';
 
-    // For sizes less than 32k, do a stack allocation
+    // For sizes less than 8k, do a stack allocation
     bool on_stack = false;
-    int stack_size = 0;
     if (const IntImm *sz = op->size.as<IntImm>()) {
-        stack_size = sz->value;
-        on_stack = stack_size <= 32*1024;
+        on_stack = sz->value <= 8*1024;
     }
 
     if (on_stack) {

diff --git a/cpp/src/CodeGen_PTX_Dev.cpp b/cpp/src/CodeGen_PTX_Dev.cpp
@@ -103,9 +103,9 @@ void CodeGen_PTX_Dev::compile(Stmt stmt, std::string name, const std::vector<Arg
     builder->CreateBr(body_block);
 
     // Add the nvvm annotation that it is a kernel function. 
-    MDNode *mdNode = MDNode::get(*context, vec((Value*)function,
-                                              (Value*)MDString::get(*context, "kernel"),
-                                              (Value*)ConstantInt::get(i32, 1)));
+    MDNode *mdNode = MDNode::get(*context, vec<Value *>(function,
+                                                        MDString::get(*context, "kernel"),
+                                                        ConstantInt::get(i32, 1)));
     module->getOrInsertNamedMetadata("nvvm.annotations")->addOperand(mdNode);