Merge branch 'hpvm-release/hotfix' into hpvm-release-exp

hpvm/include/hpvm.h

@@ -6,6 +6,8 @@
  *cr
  ***************************************************************************/
 
+// This changes hpvm::DEVICE (backward-compatible hint name) to hpvm::CUDNN_TARGET.
+// hpvm::DEVICE is deprecated; do not use.
 #ifndef DEVICE
 #define DEVICE CUDNN_TARGET
 #endif

hpvm/test/benchmarks/CMakeLists.txt

@@ -34,5 +34,22 @@ function(add_hpvm_benchmark
   add_custom_target(${target_name} DEPENDS ${output_bin_path})
 endfunction(add_hpvm_benchmark)
 
+function(add_hpvm_cpu_gpu_benchmark
+  target_prefix common_flags language_mode main_src util_bitcodes
+)
+  set(all_flags_cpu ${common_flags} "-DDEVICE=CPU_TARGET")
+  add_hpvm_benchmark(
+    "${target_prefix}_cpu" "${target_prefix}-cpu" "${all_flags_cpu}" ${language_mode}
+    ${main_src} "${util_bitcodes}"
+  )
+  if(HPVM_USE_OPENCL)
+    set(all_flags_gpu ${common_flags} "--opencl" "-DDEVICE=GPU_TARGET")
+    add_hpvm_benchmark(
+      "${target_prefix}_gpu" "${target_prefix}-gpu" "${all_flags_gpu}" ${language_mode}
+      ${main_src} "${util_bitcodes}"
+    )
+  endif()
+endfunction()
+
 add_subdirectory(hpvm-cava)
 add_subdirectory(pipeline)

hpvm/test/benchmarks/hpvm-cava/CMakeLists.txt

@@ -8,17 +8,4 @@ set(
 
 # Sets ${util_bitcodes}
 hpvm_compile_util_sources("${compiler_flags}" "${other_srcs}" "c")
-
-set(all_flags_cpu ${compiler_flags} "-DDEVICE=CPU_TARGET" "-lpthread")
-add_hpvm_benchmark(
-  "hpvm_cava_cpu" "hpvm-cava-cpu" "${all_flags_cpu}" "c"
-  src/main.c "${util_bitcodes}"
-)
-
-if(HPVM_USE_OPENCL)
-  set(all_flags_gpu ${compiler_flags} "-DDEVICE=GPU_TARGET" "--opencl" "-lpthread")
-  add_hpvm_benchmark(
-    "hpvm_cava_gpu" "hpvm-cava-gpu" "${all_flags_gpu}" "c"
-    src/main.c "${util_bitcodes}"
-  )
-endif()
+add_hpvm_cpu_gpu_benchmark(hpvm_cava "${compiler_flags}" c src/main.c "${util_bitcodes}")

hpvm/test/benchmarks/hpvm-cava/src/main.c

@@ -12,6 +12,9 @@
 
 #include "hpvm.h"
 
+// Max file extension size
+#define MAX_EXT_SIZE 20
+
 int NUM_TEST_CASES;
 int NUM_CLASSES;
 int INPUT_DIM;
@@ -875,29 +878,33 @@ int main(int argc, char *argv[]) {
 
   // Output the image.
   // NOTE: We deliberately perform this file I/O outside of the kernel.
-  char str[50], base_str[50];
+  const int len = strlen(args.args[OUTPUT_IMAGE_BIN]);
+  const char *base_str = args.args[OUTPUT_IMAGE_BIN];
+  char *str = malloc(sizeof(char)*(len + MAX_EXT_SIZE + 1)); // Handles the extensions below
   strcpy(base_str, args.args[OUTPUT_IMAGE_BIN]);
   strcpy(str, base_str);
-  strcat(str, ".bin");
+  strncat(str, ".bin", MAX_EXT_SIZE);
   printf("Writing output image to %s\n", str);
   write_image_to_binary(str, image_out, row_size, col_size);
   strcpy(str, base_str);
-  strcat(str, "_gamut.bin");
+  strncat(str, "_gamut.bin", MAX_EXT_SIZE);
   printf("Writing output image to %s\n", str);
   write_image_to_binary(str, image_out_gamut, row_size, col_size);
   strcpy(str, base_str);
-  strcat(str, "_demosaic.bin");
+  strncat(str, "_demosaic.bin", MAX_EXT_SIZE);
   printf("Writing output image to %s\n", str);
   write_image_to_binary(str, image_out_demosaic, row_size, col_size);
   strcpy(str, base_str);
-  strcat(str, "_denoise.bin");
+  strncat(str, "_denoise.bin", MAX_EXT_SIZE);
   printf("Writing output image to %s\n", str);
   write_image_to_binary(str, image_out_denoise, row_size, col_size);
   strcpy(str, base_str);
-  strcat(str, "_transform.bin");
+  strncat(str, "_transform.bin", MAX_EXT_SIZE);
   printf("Writing output image to %s\n", str);
   write_image_to_binary(str, image_out_transform, row_size, col_size);
 
+  free(str);
+
   __hpvm__cleanup();
 
   return 0;

hpvm/test/benchmarks/pipeline/CMakeLists.txt

 find_package(OpenCV 2)
 if(${OpenCV_FOUND})
-  set(
-    all_flags
-    -O3 -I${OpenCV_INCLUDE_DIRS}
-    -ffast-math -fno-lax-vector-conversions -fno-vectorize -fno-slp-vectorize
-    -lpthread
+  foreach(incl_dir ${OpenCV_INCLUDE_DIRS})
+    list(APPEND all_flags "-I${incl_dir}")
+  endforeach()
+  foreach(link_lib ${OpenCV_LIBS})
+    list(APPEND all_flags "-l${link_lib}")
+  endforeach()
+  list(
+    APPEND all_flags -L "${OpenCV_INSTALL_PATH}/lib"
+    -O3 -ffast-math -fno-lax-vector-conversions -fno-vectorize -fno-slp-vectorize
   )
-  add_hpvm_benchmark("pipeline_cpu" "pipeline-cpu" "${all_flags}" "c++" src/main.cc "")
-  if(HPVM_USE_OPENCL)
-    add_hpvm_benchmark("pipeline_gpu" "pipeline-gpu" "${all_flags};--opencl" "c++" src/main.cc "")
-  endif()
+  add_hpvm_cpu_gpu_benchmark("pipeline" "${all_flags}" c++ src/main.cc "")
 else()
   message(WARNING "opencv-2 not found; not compiling HPVM benchmark 'pipeline'.")
 endif()

hpvm/test/benchmarks/pipeline/src/main.cc

@@ -27,6 +27,10 @@
 #define HEIGHT 640
 #define WIDTH 480
 
+#ifndef DEVICE
+#error "The macro 'DEVICE' must be defined to CPU_TARGET or GPU_TARGET."
+#endif
+
 std::string input_window = "GPU Pipeline - Input Video";
 std::string output_window = "GPU Pipeline - Edge Mapping";
 

hpvm/tools/hpvm-clang/main.py.in

@@ -40,6 +40,9 @@ def compile_hpvm_c(
 ):
     from subprocess import check_output
 
+    # FIXME: Added this because Exceptions are not supported in HPVM.
+    flags = (flags or []) + ["no-exceptions"]
+
     passes = ["LLVMBuildDFG"]
     pass_flags = ["buildDFG"]
     if tensor_target == "tensor":
@@ -229,8 +232,8 @@ See option -b for that."""
         help="[clang emit-llvm] clang++ flags (such as -ffastmath)"
     )
     parser.add_argument(
-        "-O", type=str, default="0", metavar="level",
-        help="[clang emit-llvm] Optimization level"
+        "-O", type=str, default="1", metavar="level",
+        help="[clang emit-llvm] Optimization level. Note that default is -O1."
     )
     parser.add_argument(
         "--std", type=str,