Merge branch 'master' into nvptx_merge

LICENSE.txt

@@ -1,4 +1,4 @@
-Copyright (c) 2010-2013, Intel Corporation
+Copyright (c) 2010-2014, Intel Corporation
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -77,7 +77,7 @@ covered by the following license:
 University of Illinois/NCSA
 Open Source License
 
-Copyright (c) 2003-2013 University of Illinois at Urbana-Champaign.
+Copyright (c) 2003-2014 University of Illinois at Urbana-Champaign.
 All rights reserved.
 
 Developed by:

Makefile

@@ -34,11 +34,41 @@
 # ispc Makefile
 #
 
+define newline
+
+
+endef
+
+define WARNING_BODY
+ ============================== !!! WARNING !!! =============================== \n
+Location of LLVM files in your PATH is different than path in LLVM_HOME \n
+variable (or LLVM_HOME is not set). The most likely this means that you are \n
+using default LLVM installation on your system, which is very bad sign. \n
+Note, that ISPC uses LLVM optimizer and is highly dependent on it. We recommend \n
+using *patched* version of LLVM 3.3 or 3.4. Patches are availible in \n
+llvm_patches folder. You can build LLVM manually, or run our scripts, which \n
+will do all the work for you. Do the following: \n
+1. Create a folder, where LLVM will reside and set LLVM_HOME variable to its \n
+  path. \n
+2. Set ISPC_HOME variable to your ISPC location (probably current folder).
+3. Run alloy.py tool to checkout and build LLVM: \n
+  alloy.py -b --version=3.4 \n
+4. Add $$LLVM_HOME/bin-3.4/bin path to your PATH. \n
+==============================================================================
+endef
+
 # If you have your own special version of llvm and/or clang, change
 # these variables to match.
 LLVM_CONFIG=$(shell which llvm-config)
 CLANG_INCLUDE=$(shell $(LLVM_CONFIG) --includedir)
 
+RIGHT_LLVM = $(WARNING_BODY)
+ifdef LLVM_HOME
+	ifeq ($(findstring $(LLVM_HOME), $(LLVM_CONFIG)), $(LLVM_HOME))
+		RIGHT_LLVM = LLVM from $$LLVM_HOME is used.
+	endif
+endif
+
 # Enable ARM by request
 # To enable: make ARM_ENABLED=1
 ARM_ENABLED=0
@@ -56,7 +86,7 @@ endif
 ARCH_TYPE = $(shell arch)
 
 LLVM_CXXFLAGS=$(shell $(LLVM_CONFIG) --cppflags)
-LLVM_VERSION=LLVM_$(shell $(LLVM_CONFIG) --version | sed -e s/\\./_/ -e s/svn// -e s/\.0//)
+LLVM_VERSION=LLVM_$(shell $(LLVM_CONFIG) --version | sed -e 's/svn//' -e 's/\./_/' -e 's/\..*//')
 LLVM_VERSION_DEF=-D$(LLVM_VERSION)
 
 LLVM_COMPONENTS = engine ipo bitreader bitwriter instrumentation linker nvptx
@@ -84,7 +114,7 @@ ifeq ($(LLVM_VERSION),LLVM_3_4)
 endif
 
 ifeq ($(LLVM_VERSION),LLVM_3_5)
-    ISPC_LIBS += -lcurses
+    ISPC_LIBS += -lcurses -lz
 endif
 
 ifeq ($(ARCH_OS),Linux)
@@ -188,6 +218,7 @@ llvm_check:
 	 echo "ERROR: llvm-config not found in your PATH";  \
 	 echo "******************************************"; \
 	 echo; exit 1)
+	@echo -e '$(subst $(newline), ,$(RIGHT_LLVM))'
 
 print_llvm_src: llvm_check
 	@echo Using LLVM `llvm-config --version` from `llvm-config --libdir`

alloy.py

@@ -84,6 +84,10 @@ def build_LLVM(version_LLVM, revision, folder, tarball, debug, selfbuild, extra,
     # Here we understand what and where do we want to build
     current_path = os.getcwd()
     llvm_home = os.environ["LLVM_HOME"]
+    
+
+    make_sure_dir_exists(llvm_home)
+    
     os.chdir(llvm_home)
     FOLDER_NAME=version_LLVM
     if  version_LLVM == "trunk":
@@ -292,6 +296,8 @@ def build_ispc(version_LLVM, make):
         p_temp = os.getenv("LLVM_INSTALL_DIR")
         v_temp = os.getenv("LLVM_VERSION")
         os.environ["LLVM_INSTALL_DIR"] = os.environ["LLVM_HOME"] + "\\bin-" + version_LLVM
+        if version_LLVM == "3.2":
+            temp = "3_2"
         if version_LLVM == "3.3":
             temp = "3_3"
         if version_LLVM == "3.4":
@@ -467,7 +473,10 @@ def validation_run(only, only_targets, reference_branch, number, notify, update,
                     for i2 in range(0,len(opts)):
                         stability.arch = arch[i1]
                         stability.no_opt = opts[i2]
-                        execute_stability(stability, R, print_version)
+                        try:
+                            execute_stability(stability, R, print_version)
+                        except:
+                            print_debug("Exception in execute_stability - maybe some test subprocess terminated before it should have\n", False, stability_log)
                         print_version = 0
             for j in range(0,len(sde_targets)):
                 stability.target = sde_targets[j][1]
@@ -574,6 +583,8 @@ def validation_run(only, only_targets, reference_branch, number, notify, update,
         f_lines = fp.readlines()
         fp.close()
         line = ""
+        if  not sys.exc_info()[0] == None:
+            line = line + "Last exception: " + str(sys.exc_info()) + '\n'
         for i in range(0,len(f_lines)):
             line = line + f_lines[i][:-1]
             line = line + '   \n'
@@ -581,7 +592,7 @@ def validation_run(only, only_targets, reference_branch, number, notify, update,
         msg.attach(text)
         attach_mail_file(msg, alloy_build, "alloy_build.log")
         s = smtplib.SMTP(smtp_server)
-        s.sendmail('ISPC_test_system', options.notify, msg.as_string())
+        s.sendmail('ISPC_test_system', options.notify.split(" "), msg.as_string())
         s.quit()
 
 def Main():
@@ -628,7 +639,8 @@ def Main():
     current_path = os.getcwd()
     make = "make -j" + options.speed
     if os.environ["ISPC_HOME"] != os.getcwd():
-        error("you ISPC_HOME and your current path are different!\n", 2)
+        error("you ISPC_HOME and your current path are different! (" + os.environ["ISPC_HOME"] + " is not equal to " + os.getcwd() +
+        ")\n", 2)
     if options.perf_llvm == True:
         if options.branch == "master":
             options.branch = "trunk"
@@ -657,6 +669,7 @@ from optparse import OptionParser
 from optparse import OptionGroup
 import sys
 import os
+import errno
 import operator
 import time
 import glob
@@ -677,6 +690,7 @@ import common
 error = common.error
 take_lines = common.take_lines
 print_debug = common.print_debug
+make_sure_dir_exists = common.make_sure_dir_exists
 if __name__ == '__main__':
     # parsing options
     class MyParser(OptionParser):

ast.cpp

@@ -389,6 +389,11 @@ lCheckAllOffSafety(ASTNode *node, void *data) {
         return false;
     }
 
+    if (dynamic_cast<PrintStmt *>(node) != NULL) {
+        *okPtr = false;
+        return false;
+    }
+
     if (dynamic_cast<NewExpr *>(node) != NULL ||
         dynamic_cast<DeleteStmt *>(node) != NULL) {
         // We definitely don't want to run the uniform variants of these if

builtins.cpp

@@ -765,7 +765,7 @@ AddBitcodeToModule(const unsigned char *bitcode, int length,
     llvm::MemoryBuffer *bcBuf = llvm::MemoryBuffer::getMemBuffer(sb);
 #if defined(LLVM_3_5)
     llvm::ErrorOr<llvm::Module *> ModuleOrErr = llvm::parseBitcodeFile(bcBuf, *g->ctx);
-    if (llvm::error_code EC = ModuleOrErr.getError())
+    if (std::error_code EC = ModuleOrErr.getError())
         Error(SourcePos(), "Error parsing stdlib bitcode: %s", EC.message().c_str());
     else {
         llvm::Module *bcModule = ModuleOrErr.get();

builtins/util.m4

@@ -1498,7 +1498,8 @@ define <$1 x $2> @__atomic_compare_exchange_$3_global($2* %ptr, <$1 x $2> %cmp,
    %cmp_LANE_ID = extractelement <$1 x $2> %cmp, i32 LANE
    %val_LANE_ID = extractelement <$1 x $2> %val, i32 LANE
   ifelse(LLVM_VERSION,LLVM_3_5,`
-    %r_LANE_ID = cmpxchg $2 * %ptr, $2 %cmp_LANE_ID, $2 %val_LANE_ID seq_cst seq_cst
+    %r_LANE_ID_t = cmpxchg $2 * %ptr, $2 %cmp_LANE_ID, $2 %val_LANE_ID seq_cst seq_cst
+    %r_LANE_ID = extractvalue { $2, i1 } %r_LANE_ID_t, 0
   ',`
     %r_LANE_ID = cmpxchg $2 * %ptr, $2 %cmp_LANE_ID, $2 %val_LANE_ID seq_cst
   ')
@@ -1513,7 +1514,8 @@ define <$1 x $2> @__atomic_compare_exchange_$3_global($2* %ptr, <$1 x $2> %cmp,
 define $2 @__atomic_compare_exchange_uniform_$3_global($2* %ptr, $2 %cmp,
                                                        $2 %val) nounwind alwaysinline {
   ifelse(LLVM_VERSION,LLVM_3_5,`
-   %r = cmpxchg $2 * %ptr, $2 %cmp, $2 %val seq_cst seq_cst
+   %r_t = cmpxchg $2 * %ptr, $2 %cmp, $2 %val seq_cst seq_cst
+   %r = extractvalue { $2, i1 } %r_t, 0
   ',`
    %r = cmpxchg $2 * %ptr, $2 %cmp, $2 %val seq_cst
   ')

cbackend.cpp

@@ -69,6 +69,7 @@
     #include "llvm/IR/CallSite.h"
     #include "llvm/IR/CFG.h"
     #include "llvm/IR/GetElementPtrTypeIterator.h"
+    #include "llvm/Support/FileSystem.h"
 #else
     #include "llvm/Analysis/Verifier.h"
     #include <llvm/Assembly/PrintModulePass.h>
@@ -1769,7 +1770,11 @@ std::string CWriter::GetValueName(const llvm::Value *Operand) {
 
   // Resolve potential alias.
   if (const llvm::GlobalAlias *GA = llvm::dyn_cast<llvm::GlobalAlias>(Operand)) {
+#if defined(LLVM_3_5)
+    if (const llvm::Value *V = GA->getAliasee())
+#else
     if (const llvm::Value *V = GA->resolveAliasedGlobal(false))
+#endif
       Operand = V;
   }
 
@@ -2158,7 +2163,13 @@ static SpecialGlobalClass getGlobalVariableClass(const llvm::GlobalVariable *GV)
 
   // Otherwise, if it is other metadata, don't print it.  This catches things
   // like debug information.
+#if defined(LLVM_3_5)
+  // Here we compare char *
+  if (!strcmp(GV->getSection(), "llvm.metadata"))
+#else
+  // Here we compare strings
   if (GV->getSection() == "llvm.metadata")
+#endif
     return NotPrinted;
 
   return NotSpecial;
@@ -3282,10 +3293,16 @@ void CWriter::visitBinaryOperator(llvm::Instruction &I) {
       if ((I.getOpcode() == llvm::Instruction::Shl ||
            I.getOpcode() == llvm::Instruction::LShr ||
            I.getOpcode() == llvm::Instruction::AShr)) {
-          if (LLVMVectorValuesAllEqual(I.getOperand(1))) {
+          llvm::Value *splat = NULL;
-              Out << "__extract_element(";
+          if (LLVMVectorValuesAllEqual(I.getOperand(1), &splat)) {
-              writeOperand(I.getOperand(1));
+              if (splat) {
-              Out << ", 0) ";
+                  // Avoid __extract_element(splat(value), 0), if possible.
+                  writeOperand(splat);
+              } else {
+                  Out << "__extract_element(";
+                  writeOperand(I.getOperand(1));
+                  Out << ", 0) ";
+              }
           }
           else
               writeOperand(I.getOperand(1));

common.py

@@ -34,6 +34,7 @@
 # // Author: Filippov Ilia
 import sys
 import os
+import errno
 import shutil
 
 def write_to_file(filename, line):
@@ -49,6 +50,14 @@ def remove_if_exists(filename):
         else:
             os.remove(filename)
 
+def make_sure_dir_exists(path):
+    try:
+        os.makedirs(path)
+    except OSError as exception:
+        if exception.errno != errno.EEXIST:
+            raise
+
+
 # detect version which is printed after command
 def take_lines(command, which):
     os.system(command + " > " + "temp_detect_version")

docs/ReleaseNotes.txt

@@ -1,3 +1,47 @@
+=== v1.7.0 === (18 April 2014)
+
+A major new version of ISPC with several language and library extensions and
+fixes in debug info support. Binaries for all platforms are based on patched
+version on LLVM 3.4. There also performance improvements beyond switchover to
+LLVM 3.4.
+
+The list of language and library changes:
+
+* Support for varying types in exported functions was added. See documentation
+  for more details.
+
+* get_programCount() function was moved from stdlib.ispc to
+  examples/util/util.isph, which needs to be included somewhere in your
+  project, if you want to use it.
+
+* Library functions for saturated arithmetic were added. add/sub/mul/div
+  operations are supported for signed and unsigned 8/16/32/64 integer types
+  (both uniform and varying).
+
+* The algorithm for selecting overloaded function was extended to cover more
+  types of overloading. Handling of reference types in overloaded functions was
+  fixed. The rules for selecting the best match were changed to match C++,
+  which requires the function to be the best match for all parameters. In
+  ambiguous cases, a warning is issued, but it will be converted to an error
+  in the next release.
+
+* Explicit typecasts between any two reference types were allowed.
+
+* Implicit cast of pointer to const type to void* was disallowed.
+
+The list of other notable changes is:
+
+* Number of fixes for better debug info support.
+
+* Memory corruption bug was fixed, which caused rare but not reproducible
+  compile time fails.
+
+* Alias analysis was enabled (more aggressive optimizations are expected).
+
+* A bug involving inaccurate handling of "const" qualifier was fixed. As a
+  result, more "const" qualifiers may appear in .h files, which may cause
+  compilation errors.
+
 === v1.6.0 === (19 December 2013)
 
 A major new version of ISPC with major improvements in performance and

docs/ispc.rst

@@ -50,6 +50,7 @@ Contents:
   + `Updating ISPC Programs For Changes In ISPC 1.3`_
   + `Updating ISPC Programs For Changes In ISPC 1.5.0`_
   + `Updating ISPC Programs For Changes In ISPC 1.6.0`_
+  + `Updating ISPC Programs For Changes In ISPC 1.7.0`_
 
 * `Getting Started with ISPC`_
 
@@ -299,6 +300,32 @@ becomes a keyword and it potentially creates a conflict with existing user
 function. Also a new library function packed_store_active2() was introduced,
 which also may create a conflict with existing user functions.
 
+Updating ISPC Programs For Changes In ISPC 1.7.0
+------------------------------------------------
+
+This release contains several changes that may affect compatibility with
+older versions:
+
+* The algorithm for selecting overloaded functions was extended to cover more
+  types of overloading, and handling of reference types was fixed. At the same
+  time the old scheme, which blindly used the function with "the best score"
+  summed for all arguments, was switched to the C++ approach, which requires
+  "the best score" for each argument. If the best function doesn't exist, a
+  warning is issued in this version. It will be turned into an error in the
+  next version. A simple example: Suppose we have two functions: max(int, int)
+  and max(unsigned int, unsigned int). The new rules lead to an error when
+  calling max(int, unsigned int), as the best choice is ambiguous.
+
+* Implicit cast of pointer to const type to void* was disallowed. Use explicit
+  cast if needed.
+
+* A bug which prevented "const" qualifiers from appearing in emitted .h files
+  was fixed. Consequently, "const" qualifiers now properly appearing in emitted
+  .h files may cause compile errors in pre-existing codes.
+
+* get_ProgramCount() was moved from stdlib to examples/util/util.isph file. You
+  need to include this file to be able to use this function.
+
 
 Getting Started with ISPC
 =========================

docs/news.rst

@@ -2,6 +2,14 @@
 ispc News
 =========
 
+ispc 1.7.0 is Released
+----------------------
+
+A major new version of ISPC with several language and library extensions and
+fixes in debug info support. Binaries for all platforms are based on patched
+version on LLVM 3.4. There also performance improvements beyond switchover to
+LLVM 3.4.
+
 ispc 1.6.0 is Released
 ----------------------
 
@@ -11,7 +19,6 @@ a number of language and library extensions. Released binaries are based on
 patched LLVM 3.3 on Linux and MacOS and LLVM 3.4rc3 on Windows. Please refer
 to Release Notes for complete set of changes.
 
-
 ispc 1.5.0 is Released
 ----------------------
 

docs/template-news.txt

@@ -57,7 +57,7 @@
 %(body)s
 </div>
     <div class="clearfix"></div>
-    <div id="footer"> &copy; 2011-2013 <strong>Intel Corporation</strong> | Valid <a href="http://validator.w3.org/check?uri=referer">XHTML</a> | <a href="http://jigsaw.w3.org/css-validator/check/referer">CSS</a> | ClearBlue  by: <a href="http://www.themebin.com/">ThemeBin</a>
+    <div id="footer"> &copy; 2011-2014 <strong>Intel Corporation</strong> | Valid <a href="http://validator.w3.org/check?uri=referer">XHTML</a> | <a href="http://jigsaw.w3.org/css-validator/check/referer">CSS</a> | ClearBlue  by: <a href="http://www.themebin.com/">ThemeBin</a>
       <!-- Please Do Not remove this link, thank u -->
       </div>
       </div>

docs/template-perf.txt

@@ -57,7 +57,7 @@
 %(body)s
 </div>
     <div class="clearfix"></div>
-    <div id="footer"> &copy; 2011-2013 <strong>Intel Corporation</strong> | Valid <a href="http://validator.w3.org/check?uri=referer">XHTML</a> | <a href="http://jigsaw.w3.org/css-validator/check/referer">CSS</a> | ClearBlue  by: <a href="http://www.themebin.com/">ThemeBin</a>
+    <div id="footer"> &copy; 2011-2014 <strong>Intel Corporation</strong> | Valid <a href="http://validator.w3.org/check?uri=referer">XHTML</a> | <a href="http://jigsaw.w3.org/css-validator/check/referer">CSS</a> | ClearBlue  by: <a href="http://www.themebin.com/">ThemeBin</a>
       <!-- Please Do Not remove this link, thank u -->
       </div>
       </div>

docs/template.txt

@@ -57,7 +57,7 @@
 %(body)s
 </div>
     <div class="clearfix"></div>
-    <div id="footer"> &copy; 2011-2013 <strong>Intel Corporation</strong> | Valid <a href="http://validator.w3.org/check?uri=referer">XHTML</a> | <a href="http://jigsaw.w3.org/css-validator/check/referer">CSS</a> | ClearBlue  by: <a href="http://www.themebin.com/">ThemeBin</a>
+    <div id="footer"> &copy; 2011-2014 <strong>Intel Corporation</strong> | Valid <a href="http://validator.w3.org/check?uri=referer">XHTML</a> | <a href="http://jigsaw.w3.org/css-validator/check/referer">CSS</a> | ClearBlue  by: <a href="http://www.themebin.com/">ThemeBin</a>
       <!-- Please Do Not remove this link, thank u -->
       </div>
       </div>

doxygen.cfg

@@ -31,7 +31,7 @@ PROJECT_NAME           = "Intel SPMD Program Compiler"
 # This could be handy for archiving the generated documentation or
 # if some version control system is used.
 
-PROJECT_NUMBER         = 1.6.1dev
+PROJECT_NUMBER         = 1.7.1dev
 
 # The OUTPUT_DIRECTORY tag is used to specify the (relative or absolute)
 # base path where the generated documentation will be put.

examples/tasksys.cpp

@@ -183,7 +183,7 @@ struct TaskInfo {
     void *data;
     int taskIndex;
     int taskCount3d[3];
-#if defined(ISPC_IS_WINDOWS)
+#if defined(  ISPC_USE_CONCRT)
     event taskEvent;
 #endif
     int taskCount() const { return taskCount3d[0]*taskCount3d[1]*taskCount3d[2]; }
@@ -1003,7 +1003,7 @@ TaskGroup::Launch(int baseIndex, int count) {
         // Actually run the task. 
         // TBB does not expose the task -> thread mapping so we pretend it's 1:1
         int threadIndex = ti->taskIndex;
-        int threadCount = ti->taskCount;
+        int threadCount = ti->taskCount();
 
         ti->func(ti->data, threadIndex, threadCount, ti->taskIndex, ti->taskCount(),
             ti->taskIndex0(), ti->taskIndex1(), ti->taskIndex2(),
@@ -1033,7 +1033,7 @@ TaskGroup::Launch(int baseIndex, int count) {
 
             // TBB does not expose the task -> thread mapping so we pretend it's 1:1
             int threadIndex = ti->taskIndex;
-            int threadCount = ti->taskCount;
+            int threadCount = ti->taskCount();
             ti->func(ti->data, threadIndex, threadCount, ti->taskIndex, ti->taskCount(),
             ti->taskIndex0(), ti->taskIndex1(), ti->taskIndex2(),
             ti->taskCount0(), ti->taskCount1(), ti->taskCount2());

examples/util/util.isph

@@ -1,5 +1,42 @@
+/*
+  Copyright (c) 2014, Intel Corporation
+  All rights reserved.
+
+  Redistribution and use in source and binary forms, with or without
+  modification, are permitted provided that the following conditions are
+  met:
+
+    * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its
+      contributors may be used to endorse or promote products derived from
+      this software without specific prior written permission.
+
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
+   IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+   TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
+   PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef UTIL_ISPH
+#define UTIL_ISPH
 
 // utility function to read the value of programCount from C/C++
 export uniform int32 get_programCount() {
   return programCount;
 }
+
+#endif // UTIL_ISPH

expr.cpp

@@ -5143,9 +5143,18 @@ MemberExpr::create(Expr *e, const char *id, SourcePos p, SourcePos idpos,
                   exprType->GetString().c_str());
         return NULL;
     }
-
+    if (CastType<StructType>(exprType) != NULL) {
-    if (CastType<StructType>(exprType) != NULL)
+      const StructType *st = CastType<StructType>(exprType);
+      if (st->IsDefined()) { 
         return new StructMemberExpr(e, id, p, idpos, derefLValue);
+      }
+      else {
+        Error(p, "Member operator \"%s\" can't be applied to declared "
+              "struct \"%s\" containing an undefined struct type.", derefLValue ? "->" : ".",
+              exprType->GetString().c_str());
+        return NULL;
+      }
+    }
     else if (CastType<VectorType>(exprType) != NULL)
         return new VectorMemberExpr(e, id, p, idpos, derefLValue);
     else if (CastType<UndefinedStructType>(exprType)) {
@@ -8719,6 +8728,12 @@ NewExpr::TypeCheck() {
               "but not defined type \"%s\".", allocType->GetString().c_str());
         return NULL;
     }
+    const StructType *st = CastType<StructType>(allocType);
+    if (st != NULL && !st->IsDefined()) {
+        Error(pos, "Can't dynamically allocate storage for declared "
+              "type \"%s\" containing undefined member type.", allocType->GetString().c_str());
+        return NULL;
+    }
 
     // Otherwise we only need to make sure that if we have an expression
     // giving a number of elements to allocate that it can be converted to

ispc.cpp

@@ -241,40 +241,6 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         }
     }
 
-#if defined(ISPC_ARM_ENABLED) && !defined(__arm__)
-    if (cpu == NULL && !strncmp(isa, "neon", 4))
-        // If we're compiling NEON on an x86 host and the CPU wasn't
-        // supplied, don't go and set the CPU based on the host...
-        cpu = "cortex-a9";
-#endif
-
-    if (cpu == NULL) {
-        std::string hostCPU = llvm::sys::getHostCPUName();
-        if (hostCPU.size() > 0)
-            cpu = strdup(hostCPU.c_str());
-        else {
-            Warning(SourcePos(), "Unable to determine host CPU!\n");
-            cpu = "generic";
-        }
-    }
-    else {
-        bool foundCPU = false;
-        for (int i = 0; i < int(sizeof(supportedCPUs) / sizeof(supportedCPUs[0]));
-             ++i) {
-            if (!strcmp(cpu, supportedCPUs[i])) {
-                foundCPU = true;
-                break;
-            }
-        }
-        if (foundCPU == false) {
-            Error(SourcePos(), "Error: CPU type \"%s\" unknown. Supported CPUs: "
-                    "%s.", cpu, SupportedCPUs().c_str());
-            return;
-        }
-    }
-
-    this->m_cpu = cpu;
-
     if (arch == NULL) {
 #ifdef ISPC_ARM_ENABLED
         if (!strncmp(isa, "neon", 4))
@@ -311,6 +277,8 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_arch = arch;
     }
 
+    const char * cpuFromIsa;
+
     // Check default LLVM generated targets
     if (!strcasecmp(isa, "sse2") ||
         !strcasecmp(isa, "sse2-i32x4")) {
@@ -319,15 +287,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 16;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 4;
-        this->m_attributes = "+sse,+sse2,-sse3,-sse4a,-ssse3,-popcnt"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",-sse4.1,-sse4.2"
-#else
-        ",-sse41,-sse42"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
+        cpuFromIsa = "core2";
     }
     else if (!strcasecmp(isa, "sse2-x2") ||
              !strcasecmp(isa, "sse2-i32x8")) {
@@ -336,15 +298,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 16;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 8;
-        this->m_attributes = "+sse,+sse2,-sse3,-sse4a,-ssse3,-popcnt"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",-sse4.1,-sse4.2"
-#else
-        ",-sse41,-sse42"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
+        cpuFromIsa = "core2";
     }
     else if (!strcasecmp(isa, "sse4") ||
              !strcasecmp(isa, "sse4-i32x4")) {
@@ -353,16 +309,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 16;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 4;
-        // TODO: why not sse42 and popcnt?
-        this->m_attributes = "+sse,+sse2,+sse3,-sse4a,+ssse3,-popcnt,+cmov"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+sse4.1,-sse4.2"
-#else
-        ",+sse41,-sse42"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
+        cpuFromIsa = "corei7";
     }
     else if (!strcasecmp(isa, "sse4x2") ||
              !strcasecmp(isa, "sse4-x2") ||
@@ -372,15 +321,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 16;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 8;
-        this->m_attributes = "+sse,+sse2,+sse3,-sse4a,+ssse3,-popcnt,+cmov"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+sse4.1,-sse4.2"
-#else
-        ",+sse41,-sse42"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
+        cpuFromIsa = "corei7";
     }
     else if (!strcasecmp(isa, "sse4-i8x16")) {
         this->m_isa = Target::SSE4;
@@ -388,15 +331,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 16;
         this->m_dataTypeWidth = 8;
         this->m_vectorWidth = 16;
-        this->m_attributes = "+sse,+sse2,+sse3,-sse4a,+ssse3,-popcnt,+cmov"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+sse4.1,-sse4.2"
-#else
-        ",+sse41,-sse42"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 8;
+        cpuFromIsa = "corei7";
     }
     else if (!strcasecmp(isa, "sse4-i16x8")) {
         this->m_isa = Target::SSE4;
@@ -404,15 +341,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 16;
         this->m_dataTypeWidth = 16;
         this->m_vectorWidth = 8;
-        this->m_attributes = "+sse,+sse2,+sse3,-sse4a,+ssse3,-popcnt,+cmov"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+sse4.1,-sse4.2"
-#else
-        ",+sse41,-sse42"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 16;
+        cpuFromIsa = "corei7";
     }
     else if (!strcasecmp(isa, "generic-4") ||
              !strcasecmp(isa, "generic-x4")) {
@@ -452,7 +383,10 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_maskBitCount = 1;
         this->m_hasHalf = true;
         this->m_hasTranscendentals = true;
-        this->m_hasTrigonometry = true;
+        // It's set to false, because stdlib implementation of math functions
+        // is faster on MIC, than "native" implementation profided by the
+        // icc compiler.
+        this->m_hasTrigonometry = false;
         this->m_hasGather = this->m_hasScatter = true;
         this->m_hasRsqrtd = this->m_hasRcpd = true;
     }
@@ -499,9 +433,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 4;
-        this->m_attributes = "+avx,+popcnt,+cmov";
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
+        cpuFromIsa = "corei7-avx";
     }
     else if (!strcasecmp(isa, "avx") ||
              !strcasecmp(isa, "avx1") ||
@@ -511,9 +445,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 8;
-        this->m_attributes = "+avx,+popcnt,+cmov";
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
+        cpuFromIsa = "corei7-avx";
     }
     else if (!strcasecmp(isa, "avx-i64x4") ||
              !strcasecmp(isa, "avx1-i64x4")) {
@@ -522,9 +456,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 64;
         this->m_vectorWidth = 4;
-        this->m_attributes = "+avx,+popcnt,+cmov";
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 64;
+        cpuFromIsa = "corei7-avx";
     }
     else if (!strcasecmp(isa, "avx-x2") ||
              !strcasecmp(isa, "avx1-x2") ||
@@ -534,9 +468,9 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 16;
-        this->m_attributes = "+avx,+popcnt,+cmov";
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
+        cpuFromIsa = "corei7-avx";
     }
     else if (!strcasecmp(isa, "avx1.1") ||
              !strcasecmp(isa, "avx1.1-i32x8")) {
@@ -545,20 +479,11 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 8;
-        this->m_attributes = "+avx,+popcnt,+cmov,+f16c"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+rdrnd"
-#else
-        ",+rdrand"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
         this->m_hasHalf = true;
-#if !defined(LLVM_3_1)
-        // LLVM 3.2+ only
         this->m_hasRand = true;
-#endif
+        cpuFromIsa = "core-avx-i";
     }
     else if (!strcasecmp(isa, "avx1.1-x2") ||
              !strcasecmp(isa, "avx1.1-i32x16")) {
@@ -567,20 +492,11 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 16;
-        this->m_attributes = "+avx,+popcnt,+cmov,+f16c"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+rdrnd"
-#else
-        ",+rdrand"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
         this->m_hasHalf = true;
-#if !defined(LLVM_3_1)
-        // LLVM 3.2+ only
         this->m_hasRand = true;
-#endif
+        cpuFromIsa = "core-avx-i";
     }
     else if (!strcasecmp(isa, "avx1.1-i64x4")) {
         this->m_isa = Target::AVX11;
@@ -588,20 +504,11 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 64;
         this->m_vectorWidth = 4;
-        this->m_attributes = "+avx,+popcnt,+cmov,+f16c"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+rdrnd"
-#else
-        ",+rdrand"
-#endif
-        ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 64;
         this->m_hasHalf = true;
-#if !defined(LLVM_3_1)
-        // LLVM 3.2+ only
         this->m_hasRand = true;
-#endif
+        cpuFromIsa = "core-avx-i";
     }
     else if (!strcasecmp(isa, "avx2") ||
              !strcasecmp(isa, "avx2-i32x8")) {
@@ -610,24 +517,12 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 8;
-        this->m_attributes = "+avx2,+popcnt,+cmov,+f16c"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+rdrnd"
-#else
-        ",+rdrand"
-#endif
-#ifndef LLVM_3_1
-            ",+fma"
-#endif // !LLVM_3_1
-            ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
         this->m_hasHalf = true;
-#if !defined(LLVM_3_1)
-        // LLVM 3.2+ only
         this->m_hasRand = true;
         this->m_hasGather = true;
-#endif
+        cpuFromIsa = "core-avx2";
     }
     else if (!strcasecmp(isa, "avx2-x2") ||
              !strcasecmp(isa, "avx2-i32x16")) {
@@ -636,24 +531,12 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 32;
         this->m_vectorWidth = 16;
-        this->m_attributes = "+avx2,+popcnt,+cmov,+f16c"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+rdrnd"
-#else
-        ",+rdrand"
-#endif
-#ifndef LLVM_3_1
-            ",+fma"
-#endif // !LLVM_3_1
-            ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 32;
         this->m_hasHalf = true;
-#if !defined(LLVM_3_1)
-        // LLVM 3.2+ only
         this->m_hasRand = true;
         this->m_hasGather = true;
-#endif
+        cpuFromIsa = "core-avx2";
     }
     else if (!strcasecmp(isa, "avx2-i64x4")) {
         this->m_isa = Target::AVX2;
@@ -661,24 +544,12 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         this->m_nativeVectorAlignment = 32;
         this->m_dataTypeWidth = 64;
         this->m_vectorWidth = 4;
-        this->m_attributes = "+avx2,+popcnt,+cmov,+f16c"
-#if defined(LLVM_3_4) || defined(LLVM_3_5)
-        ",+rdrnd"
-#else
-        ",+rdrand"
-#endif
-#ifndef LLVM_3_1
-            ",+fma"
-#endif // !LLVM_3_1
-            ;
         this->m_maskingIsFree = false;
         this->m_maskBitCount = 64;
         this->m_hasHalf = true;
-#if !defined(LLVM_3_1)
-        // LLVM 3.2+ only
         this->m_hasRand = true;
         this->m_hasGather = true;
-#endif
+        cpuFromIsa = "core-avx2";
     }
 #ifdef ISPC_ARM_ENABLED
     else if (!strcasecmp(isa, "neon-i8x16")) {
@@ -736,6 +607,47 @@ Target::Target(const char *arch, const char *cpu, const char *isa, bool pic) :
         error = true;
     }
 
+#if defined(ISPC_ARM_ENABLED) && !defined(__arm__)
+    if (cpu == NULL && !strncmp(isa, "neon", 4))
+
+        cpu = "cortex-a9";
+#endif
+
+    if (cpu == NULL) {
+#ifndef ISPC_ARM_ENABLED
+        if (isa == NULL) {
+#endif
+            std::string hostCPU = llvm::sys::getHostCPUName();
+            if (hostCPU.size() > 0)
+                cpu = strdup(hostCPU.c_str());
+            else {
+                Warning(SourcePos(), "Unable to determine host CPU!\n");
+                cpu = "generic";
+            }
+#ifndef ISPC_ARM_ENABLED
+        }
+        else {
+            cpu = cpuFromIsa;
+        }
+#endif
+    }
+    else {
+        bool foundCPU = false;
+        for (int i = 0; i < int(sizeof(supportedCPUs) / sizeof(supportedCPUs[0]));
+             ++i) {
+            if (!strcmp(cpu, supportedCPUs[i])) {
+                foundCPU = true;
+                break;
+            }
+        }
+        if (foundCPU == false) {
+            Error(SourcePos(), "Error: CPU type \"%s\" unknown. Supported CPUs: "
+                    "%s.", cpu, SupportedCPUs().c_str());
+            return;
+        }
+    }
+    this->m_cpu = cpu;
+
     if (!error) {
         // Create TargetMachine
         std::string triple = GetTripleString();

ispc.h

@@ -38,7 +38,7 @@
 #ifndef ISPC_H
 #define ISPC_H
 
-#define ISPC_VERSION "1.6.1dev"
+#define ISPC_VERSION "1.7.1dev"
 
 #if !defined(LLVM_3_1) && !defined(LLVM_3_2) && !defined(LLVM_3_3) && !defined(LLVM_3_4) && !defined(LLVM_3_5)
 #error "Only LLVM 3.1, 3.2, 3.3, 3.4 and the 3.5 development branch are supported"

ispc.vcxproj

@@ -105,7 +105,6 @@
     <ClInclude Include="sym.h" />
     <ClInclude Include="type.h" />
     <ClInclude Include="util.h" />
-    <ClInclude Include="winstuff\unistd.h" />
   </ItemGroup>
   <ItemGroup>
     <CustomBuild Include="stdlib.ispc">
@@ -396,7 +395,7 @@
       <WarningLevel>Level3</WarningLevel>
       <Optimization>Disabled</Optimization>
       <PreprocessorDefinitions>NOMINMAX;%LLVM_VERSION%</PreprocessorDefinitions>
-      <AdditionalIncludeDirectories>$(LLVM_INSTALL_DIR)\include;.;.\winstuff;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>$(LLVM_INSTALL_DIR)\include;.;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
       <DisableSpecificWarnings>4146;4800;4996;4355;4624;4244</DisableSpecificWarnings>
     </ClCompile>
     <Link>
@@ -415,7 +414,7 @@
       <FunctionLevelLinking>true</FunctionLevelLinking>
       <IntrinsicFunctions>true</IntrinsicFunctions>
       <PreprocessorDefinitions>NOMINMAX;%LLVM_VERSION%</PreprocessorDefinitions>
-      <AdditionalIncludeDirectories>$(LLVM_INSTALL_DIR)\include;.;.\winstuff;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>$(LLVM_INSTALL_DIR)\include;.;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
       <DisableSpecificWarnings>4146;4800;4996;4355;4624;4244</DisableSpecificWarnings>
     </ClCompile>
     <Link>

llvmutil.cpp

@@ -818,7 +818,8 @@ LLVMExtractVectorInts(llvm::Value *v, int64_t ret[], int *nElts) {
 
 static bool
 lVectorValuesAllEqual(llvm::Value *v, int vectorLength,
-                      std::vector<llvm::PHINode *> &seenPhis);
+                      std::vector<llvm::PHINode *> &seenPhis,
+                      llvm::Value **splatValue = NULL);
 
 
 /** This function checks to see if the given (scalar or vector) value is an
@@ -1068,20 +1069,37 @@ lVectorShiftRightAllEqual(llvm::Value *val, llvm::Value *shift,
 
 static bool
 lVectorValuesAllEqual(llvm::Value *v, int vectorLength,
-                      std::vector<llvm::PHINode *> &seenPhis) {
+                      std::vector<llvm::PHINode *> &seenPhis,
+                      llvm::Value **splatValue) {
     if (vectorLength == 1)
         return true;
 
-    if (llvm::isa<llvm::ConstantAggregateZero>(v))
+    if (llvm::isa<llvm::ConstantAggregateZero>(v)) {
+        if (splatValue) {
+            llvm::ConstantAggregateZero *caz =
+                llvm::dyn_cast<llvm::ConstantAggregateZero>(v);
+            *splatValue = caz->getSequentialElement();
+        }
         return true;
+    }
 
     llvm::ConstantVector *cv = llvm::dyn_cast<llvm::ConstantVector>(v);
-    if (cv != NULL)
+    if (cv != NULL) {
-        return (cv->getSplatValue() != NULL);
+        llvm::Value* splat = cv->getSplatValue();
+        if (splat != NULL && splatValue) {
+            *splatValue = splat;
+        }
+        return (splat != NULL);
+    }
 
     llvm::ConstantDataVector *cdv = llvm::dyn_cast<llvm::ConstantDataVector>(v);
-    if (cdv != NULL)
+    if (cdv != NULL) {
-        return (cdv->getSplatValue() != NULL);
+        llvm::Value* splat = cdv->getSplatValue();
+        if (splat != NULL && splatValue) {
+            *splatValue = splat;
+        }
+        return (splat != NULL);
+    }
 
     llvm::BinaryOperator *bop = llvm::dyn_cast<llvm::BinaryOperator>(v);
     if (bop != NULL) {
@@ -1178,14 +1196,14 @@ lVectorValuesAllEqual(llvm::Value *v, int vectorLength,
     where the values are actually all equal.
 */
 bool
-LLVMVectorValuesAllEqual(llvm::Value *v) {
+LLVMVectorValuesAllEqual(llvm::Value *v, llvm::Value **splat) {
     llvm::VectorType *vt =
         llvm::dyn_cast<llvm::VectorType>(v->getType());
     Assert(vt != NULL);
     int vectorLength = vt->getNumElements();
 
     std::vector<llvm::PHINode *> seenPhis;
-    bool equal = lVectorValuesAllEqual(v, vectorLength, seenPhis);
+    bool equal = lVectorValuesAllEqual(v, vectorLength, seenPhis, splat);
 
     Debug(SourcePos(), "LLVMVectorValuesAllEqual(%s) -> %s.",
           v->getName().str().c_str(), equal ? "true" : "false");
@@ -1551,6 +1569,8 @@ lExtractFirstVectorElement(llvm::Value *v,
                                                      phiMap);
         llvm::Value *v1 = lExtractFirstVectorElement(bop->getOperand(1),
                                                      phiMap);
+        Assert(v0 != NULL);
+        Assert(v1 != NULL);
         // Note that the new binary operator is inserted immediately before
         // the previous vector one
         return llvm::BinaryOperator::Create(bop->getOpcode(), v0, v1,
@@ -1597,10 +1617,22 @@ lExtractFirstVectorElement(llvm::Value *v,
         return scalarPhi;
     }
 
+    // We should consider "shuffle" case and "insertElement" case separately.
+    // For example we can have shuffle(mul, undef, zero) but function
+    // "LLVMFlattenInsertChain" can handle only case shuffle(insertElement, undef, zero).
+    // Also if we have insertElement under shuffle we will handle it the next call of
+    // "lExtractFirstVectorElement" function.
+    if (llvm::isa<llvm::ShuffleVectorInst>(v)) {
+        llvm::ShuffleVectorInst *shuf = llvm::dyn_cast<llvm::ShuffleVectorInst>(v);
+        llvm::Value *indices = shuf->getOperand(2);
+        if (llvm::isa<llvm::ConstantAggregateZero>(indices)) {
+            return lExtractFirstVectorElement(shuf->getOperand(0), phiMap);
+        }
+    }
+
     // If we have a chain of insertelement instructions, then we can just
     // flatten them out and grab the value for the first one.
-    if (llvm::isa<llvm::InsertElementInst>(v) ||
+    if (llvm::isa<llvm::InsertElementInst>(v)) {
-        llvm::isa<llvm::ShuffleVectorInst>(v)) {
         return LLVMFlattenInsertChain(v, vt->getNumElements(), false);
     }
 

llvmutil.h

@@ -228,7 +228,8 @@ extern llvm::Constant *LLVMMaskAllOff;
 /** Tests to see if all of the elements of the vector in the 'v' parameter
     are equal.  Like lValuesAreEqual(), this is a conservative test and may
     return false for arrays where the values are actually all equal.  */
-extern bool LLVMVectorValuesAllEqual(llvm::Value *v);
+extern bool LLVMVectorValuesAllEqual(llvm::Value *v,
+                                     llvm::Value **splat = NULL);
 
 /** Given vector of integer-typed values, this function returns true if it
     can determine that the elements of the vector have a step of 'stride'

opt.cpp

@@ -541,8 +541,6 @@ Optimize(llvm::Module *module, int optLevel) {
         optPM.add(llvm::createScalarReplAggregatesPass());
         optPM.add(llvm::createEarlyCSEPass());
         optPM.add(llvm::createLowerExpectIntrinsicPass());
-        optPM.add(llvm::createTypeBasedAliasAnalysisPass());
-        optPM.add(llvm::createBasicAliasAnalysisPass());
 
         // Early optimizations to try to reduce the total amount of code to
         // work with if we can

run_tests.py

@@ -60,9 +60,14 @@ def run_command(cmd):
     lexer.escape = ''
     arg_list = list(lexer)
     
-    sp = subprocess.Popen(arg_list, stdin=None,
+    try:
+        sp = subprocess.Popen(arg_list, stdin=None,
                           stdout=subprocess.PIPE,
                           stderr=subprocess.PIPE)
+    except:
+        print_debug("ERROR: The child (%s) raised an esception: %s\n" % (cmd, sys.exc_info()[1]), s, run_tests_log)
+        raise
+
     out = sp.communicate()
     output = ""
     output += out[0].decode("utf-8")
@@ -213,7 +218,7 @@ def run_test(testname):
                     obj_name = "%s.obj" % os.path.basename(filename)
                 exe_name = "%s.exe" % os.path.basename(filename)
 
-                cc_cmd = "%s /I. /I../winstuff /Zi /nologo /DTEST_SIG=%d %s %s /Fe%s" % \
+                cc_cmd = "%s /I. /Zi /nologo /DTEST_SIG=%d %s %s /Fe%s" % \
                          (options.compiler_exe, match, add_prefix("test_static.cpp"), obj_name, exe_name)
                 if should_fail:
                     cc_cmd += " /DEXPECT_FAILURE"
@@ -355,7 +360,11 @@ def run_tasks_from_queue(queue, queue_ret, queue_skip, total_tests_arg, max_test
             sys.exit(0)
 
         if check_test(filename):
-            (compile_error, run_error) = run_test(filename)
+            try:
+                (compile_error, run_error) = run_test(filename)
+            except:
+                sys.exit(-1) # This is in case the child has unexpectedly died
+            
             if compile_error != 0:
                 compile_error_files += [ filename ]
             if run_error != 0:
@@ -680,8 +689,9 @@ def run_tests(options1, args, print_version):
     task_threads = [0] * nthreads
     for x in range(nthreads):
         task_threads[x] = multiprocessing.Process(target=run_tasks_from_queue, args=(q, qret, qskip, total_tests,
-            max_test_length, finished_tests_counter, finished_tests_counter_lock, glob_var))
+                max_test_length, finished_tests_counter, finished_tests_counter_lock, glob_var))
         task_threads[x].start()
+
     # wait for them to all finish and then return the number that failed
     # (i.e. return 0 if all is ok)
     for t in task_threads:
@@ -689,6 +699,12 @@ def run_tests(options1, args, print_version):
     if options.non_interactive == False:
         print_debug("\n", s, run_tests_log)
 
+    
+    for jb in task_threads:
+        if not jb.exitcode == 0:
+            raise OSError(2, 'Some test subprocess has thrown an exception', '')
+
+
     temp_time = (time.time() - start_time)
     elapsed_time = time.strftime('%Hh%Mm%Ssec.', time.gmtime(temp_time))
 

type.cpp

@@ -826,7 +826,7 @@ EnumType::GetDIType(llvm::DIDescriptor scope) const {
 
     llvm::DIFile diFile = pos.GetDIFile();
     llvm::DIType diType =
-        m->diBuilder->createEnumerationType(scope, name, diFile, pos.first_line,
+        m->diBuilder->createEnumerationType(diFile, name, diFile, pos.first_line,
                                             32 /* size in bits */,
                                             32 /* align in bits */,
                                             elementArray
@@ -1956,6 +1956,25 @@ StructType::IsConstType() const {
 }
 
 
+bool
+StructType::IsDefined() const {
+  for (int i = 0; i < GetElementCount(); i++) {
+    const Type *t = GetElementType(i);
+    const UndefinedStructType *ust = CastType<UndefinedStructType>(t);
+    if (ust != NULL) {
+      return false;
+    }
+    const StructType *st = CastType<StructType>(t);
+    if (st != NULL) {
+      if (!st->IsDefined()) {
+        return false;
+      }
+    }
+  }
+  return true;
+}
+
+
 const Type *
 StructType::GetBaseType() const {
     return this;
@@ -2179,7 +2198,7 @@ StructType::GetDIType(llvm::DIDescriptor scope) const {
     llvm::DIArray elements = m->diBuilder->getOrCreateArray(elementLLVMTypes);
     llvm::DIFile diFile = pos.GetDIFile();
     return m->diBuilder->createStructType(
-        scope,
+        diFile,
         name,
         diFile,
         pos.first_line, // Line number
@@ -2422,7 +2441,7 @@ UndefinedStructType::GetDIType(llvm::DIDescriptor scope) const {
     llvm::DIFile diFile = pos.GetDIFile();
     llvm::DIArray elements;
     return m->diBuilder->createStructType(
-        scope,
+        diFile,
         name,
         diFile,
         pos.first_line, // Line number

type.h

@@ -81,15 +81,15 @@ struct Variability {
 /** Enumerant that records each of the types that inherit from the Type
     baseclass. */
 enum TypeId {
-    ATOMIC_TYPE,
+  ATOMIC_TYPE,           // 0
-    ENUM_TYPE,
+  ENUM_TYPE,             // 1
-    POINTER_TYPE,
+  POINTER_TYPE,          // 2
-    ARRAY_TYPE,
+  ARRAY_TYPE,            // 3
-    VECTOR_TYPE,
+  VECTOR_TYPE,           // 4
-    STRUCT_TYPE,
+  STRUCT_TYPE,           // 5
-    UNDEFINED_STRUCT_TYPE,
+  UNDEFINED_STRUCT_TYPE, // 6
-    REFERENCE_TYPE,
+  REFERENCE_TYPE,        // 7
-    FUNCTION_TYPE
+  FUNCTION_TYPE          // 8
 };
 
 
@@ -675,6 +675,7 @@ public:
     bool IsIntType() const;
     bool IsUnsignedType() const;
     bool IsConstType() const;
+    bool IsDefined() const;
 
     const Type *GetBaseType() const;
     const StructType *GetAsVaryingType() const;

winstuff/stdint.h

@@ -1,14 +0,0 @@
-#ifndef MY_STDINT_H
-#define MY_STDINT_H 1
-
-typedef signed char int8_t;
-typedef signed __int16 int16_t;
-typedef signed __int32 int32_t;
-typedef signed __int64 int64_t;
-typedef unsigned char uint8_t;
-
-typedef unsigned __int16 uint16_t;
-typedef unsigned __int32 uint32_t;
-typedef unsigned __int64 uint64_t;
-
-#endif // MY_STDINT_H

winstuff/unistd.h

@@ -1,6 +0,0 @@
-#ifndef MY_UNISTD_H
-#define MY_UNISTD_H 1
-
-inline bool isatty(int) { return false; }
-
-#endif // MY_UNISTD_H