Merge branch 'cupbop:master' into master

README.md

@@ -27,10 +27,10 @@ Currently, CuPBoP support serveral CPU backends, including x86, AArch64, and RIS
    export CuPBoP_PATH=`pwd`
    export LD_LIBRARY_PATH=$CuPBoP_PATH/build/runtime:$CuPBoP_PATH/build/runtime/threadPool:$LD_LIBRARY_PATH
    ```
-If you are using boson, you can pre-installed llvm 10.0.0 
 
- LLVM_PATH=/opt/llvm-10.0.0
+   If you are using boson, you can pre-installed llvm 10.0.0\
- export PATH=$LLVM_PATH/bin:$PATH
+   `LLVM_PATH=/opt/llvm-10.0.0`\
+   `export PATH=$LLVM_PATH/bin:$PATH`
 
 2. As CuPBoP relies on CUDA structures, we need to download the CUDA header file
 

compilation/KernelTranslation/lib/insert_warp_loop.cpp

@@ -272,7 +272,6 @@ void AddContextSaveRestore(llvm::Instruction *instruction,
   std::vector<Instruction *> uses;
   Function *f2 = instruction->getParent()->getParent();
 
-
   for (Instruction::use_iterator ui = instruction->use_begin(),
                                  ue = instruction->use_end();
        ui != ue; ++ui) {

compilation/KernelTranslation/lib/memory_hierarchy.cpp

@@ -89,11 +89,12 @@ void mem_share2global(llvm::Module *M) {
           } else if (element_type->isStructTy()) {
             auto undef = llvm::UndefValue::get(element_type);
             llvm::GlobalVariable *global_memory = new llvm::GlobalVariable(
-                *M, element_type, false, llvm::GlobalValue::ExternalLinkage, undef,
+                *M, element_type, false, llvm::GlobalValue::ExternalLinkage,
-                new_name, NULL, llvm::GlobalValue::GeneralDynamicTLSModel, 0,
+                undef, new_name, NULL,
-                false);
+                llvm::GlobalValue::GeneralDynamicTLSModel, 0, false);
             global_memory->setDSOLocal(true);
-            Comdat * comdat = M->getOrInsertComdat(StringRef(share_memory->getName()));
+            Comdat *comdat =
+                M->getOrInsertComdat(StringRef(share_memory->getName()));
             comdat->setSelectionKind(Comdat::SelectionKind::Any);
             global_memory->setComdat(comdat);
             global_memory->setLinkage(llvm::GlobalValue::LinkOnceODRLinkage);
@@ -103,7 +104,6 @@ void mem_share2global(llvm::Module *M) {
                 std::pair<GlobalVariable *, GlobalVariable *>(share_memory,
                                                               global_memory));
 
-
           } else {
             assert(0 && "The required Share Memory Type is not supported\n");
           }

examples/dwt2d/common.h

@@ -45,18 +45,20 @@
 // divide and round up macro
 #define DIVANDRND(a, b) ((((a) % (b)) != 0) ? ((a) / (b) + 1) : ((a) / (b)))
 
-#  define cudaCheckError( msg ) {                                            \
+#define cudaCheckError(msg)                                                    \
+  {                                                                            \
     cudaError_t err = cudaGetLastError();                                      \
     if (cudaSuccess != err) {                                                  \
-        fprintf(stderr, "%s: %i: %s: %s.\n",                                 \
+      fprintf(stderr, "%s: %i: %s: %s.\n", __FILE__, __LINE__, msg,            \
-                __FILE__, __LINE__, msg, cudaGetErrorString( err) );         \
+              cudaGetErrorString(err));                                        \
       exit(-1);                                                                \
-    } }
+    }                                                                          \
+  }
 
-#  define cudaCheckAsyncError( msg ) {                                       \
+#define cudaCheckAsyncError(msg)                                               \
+  {                                                                            \
     cudaThreadSynchronize();                                                   \
     cudaCheckError(msg);                                                       \
   }
 
-
 #endif

examples/dwt2d/components.h

@@ -29,7 +29,8 @@
 
 /* Separate compoents of source 8bit RGB image */
 template <typename T>
-void rgbToComponents(T *d_r, T *d_g, T *d_b, unsigned char * src, int width, int height);
+void rgbToComponents(T *d_r, T *d_g, T *d_b, unsigned char *src, int width,
+                     int height);
 
 /* Copy a 8bit source image data into a color compoment of type T */
 template <typename T>

examples/dwt2d/dwt.h

@@ -28,13 +28,14 @@
 #define _DWT_H
 
 template <typename T>
-int nStage2dDWT(T *in, T *out, T * backup, int pixWidth, int pixHeight, int stages, bool forward);
+int nStage2dDWT(T *in, T *out, T *backup, int pixWidth, int pixHeight,
+                int stages, bool forward);
 
 template <typename T>
-int writeNStage2DDWT(T *component_cuda, int width, int height, 
+int writeNStage2DDWT(T *component_cuda, int width, int height, int stages,
-                     int stages, const char * filename, const char * suffix);
-template<typename T>
-int writeLinear(T *component_cuda, int width, int height, 
                      const char *filename, const char *suffix);
+template <typename T>
+int writeLinear(T *component_cuda, int width, int height, const char *filename,
+                const char *suffix);
 
 #endif

examples/dwt2d/dwt_cuda/common.h

@@ -29,29 +29,22 @@
 /// POSSIBILITY OF SUCH DAMAGE.
 ///
 
-
 #ifndef DWT_COMMON_H
 #define DWT_COMMON_H
 
-
-#include <cstdio>
 #include <algorithm>
+#include <cstdio>
 #include <vector>
 
-
-
 // compile time minimum macro
 #define CTMIN(a, b) (((a) < (b)) ? (a) : (b))
 
-
-
 // performance testing macros
 #if defined(GPU_DWT_TESTING)
 #define PERF_BEGIN                                                             \
   {                                                                            \
     dwt_cuda::CudaDWTTester PERF_TESTER;                                       \
-    for(int PERF_N = PERF_TESTER.getNumIterations(); PERF_N--; ) \
+    for (int PERF_N = PERF_TESTER.getNumIterations(); PERF_N--;) {             \
-    { \
       PERF_TESTER.beginTestIteration();
 
 #define PERF_END(PERF_NAME, PERF_W, PERF_H)                                    \
@@ -64,25 +57,20 @@
 #define PERF_END(PERF_NAME, PERF_W, PERF_H)
 #endif // GPU_DWT_TESTING
 
-
-
 namespace dwt_cuda {
 
-  
 /// Divide and round up.
 template <typename T>
 __device__ __host__ inline T divRndUp(const T &n, const T &d) {
   return (n / d) + ((n % d) ? 1 : 0);
 }
 
-  
 // 9/7 forward DWT lifting schema coefficients
 const float f97Predict1 = -1.586134342;  ///< forward 9/7 predict 1
 const float f97Update1 = -0.05298011854; ///< forward 9/7 update 1
 const float f97Predict2 = 0.8829110762;  ///< forward 9/7 predict 2
 const float f97Update2 = 0.4435068522;   ///< forward 9/7 update 2
 
-
 // 9/7 reverse DWT lifting schema coefficients
 const float r97update2 = -f97Update2;   ///< undo 9/7 update 2
 const float r97predict2 = -f97Predict2; ///< undo 9/7 predict 2
@@ -93,7 +81,6 @@ namespace dwt_cuda {
 const float scale97Mul = 1.23017410491400f;
 const float scale97Div = 1.0 / scale97Mul;
 
-  
 // 5/3 forward DWT lifting schema coefficients
 const float forward53Predict = -0.5f; /// forward 5/3 predict
 const float forward53Update = 0.25f;  /// forward 5/3 update
@@ -102,8 +89,6 @@ namespace dwt_cuda {
 const float reverse53Update = -forward53Update;   /// undo 5/3 update
 const float reverse53Predict = -forward53Predict; /// undo 5/3 predict
 
-  
-  
 /// Functor which adds scaled sum of neighbors to given central pixel.
 struct AddScaledSum {
   const float scale; // scale of neighbors
@@ -112,7 +97,8 @@ namespace dwt_cuda {
 
     // if(threadIdx.x == 0) {
 
-      //   printf("scale  %f, p %f c %f n %f , result: %f\n", scale, p, c, n, scale * (p + n) );
+    //   printf("scale  %f, p %f c %f n %f , result: %f\n", scale, p, c, n,
+    //   scale * (p + n) );
 
     // }
 
@@ -120,8 +106,6 @@ namespace dwt_cuda {
   }
 };
 
-  
-  
 /// Returns index ranging from 0 to num threads, such that first half
 /// of threads get even indices and others get odd indices. Each thread
 /// gets different index.
@@ -129,13 +113,10 @@ namespace dwt_cuda {
 ///                              parityIdx:   0  2  4  6  1  3  5  7
 /// @tparam THREADS  total count of participating threads
 /// @return parity-separated index of thread
-  template <int THREADS>
+template <int THREADS> __device__ inline int parityIdx() {
-  __device__ inline int parityIdx() {
   return (threadIdx.x * 2) - (THREADS - 1) * (threadIdx.x / (THREADS / 2));
 }
 
-          
-  
 /// size of shared memory
 #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 200)
 const int SHM_SIZE = 48 * 1024;
@@ -143,8 +124,6 @@ namespace dwt_cuda {
 const int SHM_SIZE = 16 * 1024;
 #endif
 
-  
-  
 /// Perrformance and return code tester.
 class CudaDWTTester {
 private:
@@ -185,9 +164,7 @@ namespace dwt_cuda {
   CudaDWTTester() : disabled(testRunning) {}
 
   /// Gets rpefered number of iterations
-    int getNumIterations() {
+  int getNumIterations() { return disabled ? 1 : 31; }
-      return disabled ? 1 : 31;
-    }
 
   /// Starts one test iteration.
   void beginTestIteration() {
@@ -225,25 +202,24 @@ namespace dwt_cuda {
       for (int i = times.size(); i--;) {
         sum += times[i];
       }
-        const double median = (times[times.size() / 2]
+      const double median =
-                             + times[(times.size() - 1) / 2]) * 0.5f;
+          (times[times.size() / 2] + times[(times.size() - 1) / 2]) * 0.5f;
       printf("  %s:   %7.3f ms (mean)   %7.3f ms (median)   %7.3f ms (max)  "
-               "(%d x %d)\n", name, (sum / times.size()), median, 
+             "(%d x %d)\n",
-               times[times.size() - 1], sizeX, sizeY);
+             name, (sum / times.size()), median, times[times.size() - 1], sizeX,
+             sizeY);
     }
   }
 };
 
-  
-  
 /// Simple cudaMemcpy wrapped in performance tester.
 /// @param dest  destination bufer
 /// @param src   source buffer
 /// @param sx    width of copied image
 /// @param sy    height of copied image
 template <typename T>
-  inline void memCopy(T * const dest, const T * const src,
+inline void memCopy(T *const dest, const T *const src, const size_t sx,
-                      const size_t sx, const size_t sy) {
+                    const size_t sy) {
   cudaError_t status;
   PERF_BEGIN
   status = cudaMemcpy(dest, src, sx * sy * sizeof(T), cudaMemcpyDeviceToDevice);
@@ -251,11 +227,6 @@ namespace dwt_cuda {
   CudaDWTTester::check(status, "memcpy device > device");
 }
 
-  
-  
 } // end of namespace dwt_cuda
 
-
-
 #endif // DWT_COMMON_CUDA_H
-

examples/dwt2d/dwt_cuda/dwt.h

@@ -58,10 +58,8 @@
 #ifndef DWT_CUDA_H
 #define DWT_CUDA_H
 
-
 namespace dwt_cuda {
 
-  
 /// Forward 5/3 2D DWT. See common rules (above) for more details.
 /// @param in      Expected to be normalized into range [-128, 127].
 ///                Will not be preserved (will be overwritten).
@@ -71,7 +69,6 @@ namespace dwt_cuda {
 /// @param levels  number of recursive DWT levels
 void fdwt53(int *in, int *out, int sizeX, int sizeY, int levels);
 
-  
 /// Reverse 5/3 2D DWT. See common rules (above) for more details.
 /// @param in      Input DWT coefficients. Format described in common rules.
 ///                Will not be preserved (will be overwritten).
@@ -82,7 +79,6 @@ namespace dwt_cuda {
 /// @param levels  number of recursive DWT levels
 void rdwt53(int *in, int *out, int sizeX, int sizeY, int levels);
 
-  
 /// Forward 9/7 2D DWT. See common rules (above) for more details.
 /// @param in      Input DWT coefficients. Should be normalized (in range
 ///                [-0.5, 0.5]). Will not be preserved (will be overwritten).
@@ -92,7 +88,6 @@ namespace dwt_cuda {
 /// @param levels  number of recursive DWT levels
 void fdwt97(float *in, float *out, int sizeX, int sizeY, int levels);
 
-  
 /// Reverse 9/7 2D DWT. See common rules (above) for more details.
 /// @param in      Input DWT coefficients. Format described in common rules.
 ///                Will not be preserved (will be overwritten).
@@ -103,10 +98,6 @@ namespace dwt_cuda {
 /// @param levels  number of recursive DWT levels
 void rdwt97(float *in, float *out, int sizeX, int sizeY, int levels);
 
-  
 } // namespace dwt_cuda
 
-
-
 #endif // DWT_CUDA_H
-

examples/dwt2d/dwt_cuda/io.h

@@ -30,16 +30,13 @@
 /// POSSIBILITY OF SUCH DAMAGE.
 ///
 
-
 #ifndef IO_H
 #define IO_H
 
-
 #include "common.h"
 
 namespace dwt_cuda {
 
-  
 /// Base for all IO classes - manages mirroring.
 class DWTIO {
 protected:
@@ -80,13 +77,11 @@ namespace dwt_cuda {
   }
 };
 
-
 /// Base class for pixel loader and writer - manages computing start index,
 /// stride and end of image for loading column of pixels.
 /// @tparam T        type of image pixels
 /// @tparam CHECKED  true = be prepared to image boundary, false = don't care
-  template <typename T, bool CHECKED>
+template <typename T, bool CHECKED> class VerticalDWTPixelIO : protected DWTIO {
-  class VerticalDWTPixelIO : protected DWTIO {
 protected:
   int end;    ///< index of bottom neightbor of last pixel of column
   int stride; ///< increment of pointer to get to next pixel
@@ -97,8 +92,8 @@ namespace dwt_cuda {
   /// @param firstX  x-coordinate of first pixel to use
   /// @param firstY  y-coordinate of first pixel to use
   /// @return index of pixel at position [x, y] in the image
-    __device__ int initialize(const int sizeX, const int sizeY,
+  __device__ int initialize(const int sizeX, const int sizeY, int firstX,
-                              int firstX, int firstY) {
+                            int firstY) {
     // initialize all pointers and stride
     end = CHECKED ? (sizeY * sizeX + firstX) : 0;
     stride = sizeX;
@@ -106,8 +101,6 @@ namespace dwt_cuda {
   }
 };
 
-
-
 /// Writes reverse transformed pixels directly into output image.
 /// @tparam T        type of output pixels
 /// @tparam CHECKED  true = be prepared to image boundary, false = don't care
@@ -122,8 +115,8 @@ namespace dwt_cuda {
   /// @param sizeY   height of the image
   /// @param firstX  x-coordinate of first pixel to write into
   /// @param firstY  y-coordinate of first pixel to write into
-    __device__ void init(const int sizeX, const int sizeY, 
+  __device__ void init(const int sizeX, const int sizeY, int firstX,
-                         int firstX, int firstY) {
+                       int firstY) {
     if (firstX < sizeX) {
       next = this->initialize(sizeX, sizeY, firstX, firstY);
     } else {
@@ -145,43 +138,28 @@ namespace dwt_cuda {
   }
 };
 
-
-  
 /// Loads pixels from input image.
 /// @tparam T        type of image input pixels
 /// @tparam CHECKED  true = be prepared to image boundary, false = don't care
 template <typename T, bool CHECKED>
-  class VerticalDWTPixelLoader
+class VerticalDWTPixelLoader : protected VerticalDWTPixelIO<const T, CHECKED> {
-          : protected VerticalDWTPixelIO<const T, CHECKED> {
 private:
   int last; ///< index of last loaded pixel
 public:
-  
-
   //******************* FOR TEST **********************
-  __device__ int getlast(){
+  __device__ int getlast() { return last; }
-		return last;
+  __device__ int getend() { return this->end; }
-	}
+  __device__ int getstride() { return this->stride; }
-  __device__ int getend(){
+  __device__ void setend(int a) { this->end = a; }
-		return this->end;
-	}
-  __device__ int getstride(){
-		return this->stride;
-	}
-  __device__ void setend(int a){
-      this->end=a;
-	}
   //******************* FOR TEST **********************
 
-  
-  
   /// Initializes loader - sets input size and a position of first pixel.
   /// @param sizeX   width of the image
   /// @param sizeY   height of the image
   /// @param firstX  x-coordinate of first pixel to load
   /// @param firstY  y-coordinate of first pixel to load
-    __device__ void init(const int sizeX, const int sizeY,
+  __device__ void init(const int sizeX, const int sizeY, int firstX,
-                         int firstX, int firstY) {
+                       int firstY) {
     // correctly mirror x coordinate
     this->mirror(firstX, sizeX);
 
@@ -208,7 +186,8 @@ namespace dwt_cuda {
       this->stride = -this->stride; // reverse loader's direction
     }
     // avoid reading from negative indices if loader is checked
-      // return (CHECKED && (last < 0)) ? 0 : input[last];  // TODO: use this checked variant later
+    // return (CHECKED && (last < 0)) ? 0 : input[last];  // TODO: use this
+    // checked variant later
     if (last < 0) {
       return 0;
     }
@@ -220,14 +199,11 @@ namespace dwt_cuda {
   }
 };
 
-
-
 /// Base for band write and loader. Manages computing strides and pointers
 /// to first and last pixels in a linearly-stored-bands correct way.
 /// @tparam T        type of band coefficients
 /// @tparam CHECKED  true = be prepared to image boundary, false = don't care
-  template <typename T, bool CHECKED>
+template <typename T, bool CHECKED> class VerticalDWTBandIO : protected DWTIO {
-  class VerticalDWTBandIO : protected DWTIO {
 protected:
   /// index of bottom neighbor of last pixel of loaded column
   int end;
@@ -279,12 +255,10 @@ namespace dwt_cuda {
       end = 0;
     }
 
-
     //***********for test**************
     //	end = CHECKED;
     //***********for test**************
 
-	
     // finally, return index of the first item
     return columnOffset                      // right column
            + (firstY / 2) * verticalStride   // right row
@@ -292,9 +266,6 @@ namespace dwt_cuda {
   }
 };
 
-
-
-
 /// Directly loads coefficients from four consecutively stored transformed
 /// bands.
 /// @tparam T        type of input band coefficients
@@ -324,11 +295,12 @@ namespace dwt_cuda {
       return 0;
     }
     // avoid reading from negative indices if loader is checked
-      // return (CHECKED && (last < 0)) ? 0 : input[last];  // TODO: use this checked variant later
+    // return (CHECKED && (last < 0)) ? 0 : input[last];  // TODO: use this
+    // checked variant later
     return input[last];
   }
-  public:
 
+public:
   /// Initializes loader - sets input size and a position of first pixel.
   /// @param imageSizeX   width of the image
   /// @param imageSizeY   height of the image
@@ -336,8 +308,8 @@ namespace dwt_cuda {
   ///                     (Parity determines vertically low or high band.)
   /// @param firstY       y-coordinate of first pixel to load
   ///                     (Parity determines horizontally low or high band.)
-    __device__ void init(const int imageSizeX, const int imageSizeY,
+  __device__ void init(const int imageSizeX, const int imageSizeY, int firstX,
-                         int firstX, const int firstY) {
+                       const int firstY) {
     this->mirror(firstX, imageSizeX);
     last = this->initialize(imageSizeX, imageSizeY, firstX, firstY);
 
@@ -371,12 +343,8 @@ namespace dwt_cuda {
   __device__ T loadHighFrom(const T *const input) {
     return updateAndLoad(input, this->strideLowToHigh);
   }
-
 };
 
-
-
-
 /// Directly saves coefficients into four transformed bands.
 /// @tparam T        type of output band coefficients
 /// @tparam CHECKED  true = be prepared to image boundary, false = don't care
@@ -392,7 +360,8 @@ namespace dwt_cuda {
   /// @param stride  increment of the pointer to get to next output index
   __device__ int saveAndUpdate(T *const output, const T &item,
                                const int &stride) {
-//	if(blockIdx.x == 0 && blockIdx.y == 11 && threadIdx.x == 0){		//test, Mar 20					  
+    //	if(blockIdx.x == 0 && blockIdx.y == 11 && threadIdx.x == 0){
+    ////test, Mar 20
     if ((!CHECKED) || (next != this->end)) {
       // if(next == 4) {
       //   printf(" next: %d  stride: %d val: %f \n", next, stride, item );
@@ -407,8 +376,8 @@ namespace dwt_cuda {
     // }
     return next;
   }
-  public:
 
+public:
   /// Initializes writer - sets output size and a position of first pixel.
   /// @param output       output image
   /// @param imageSizeX   width of the image
@@ -455,29 +424,17 @@ namespace dwt_cuda {
   }
 
   //*******Add three functions to get private values*******
-	__device__ int getnext(){
+  __device__ int getnext() { return next; }
-		return next;
-	}
 
-	__device__ int getend(){
+  __device__ int getend() { return this->end; }
-		return this->end;
-	}
 
-	__device__ int getstrideHighToLow(){
+  __device__ int getstrideHighToLow() { return this->strideHighToLow; }
-		return this->strideHighToLow;
-	}
 
-	__device__ int getstrideLowToHigh(){
+  __device__ int getstrideLowToHigh() { return this->strideLowToHigh; }
-		return this->strideLowToHigh;
-	}
 
   //*******Add three functions to get private values*******
 };
 
-  
-  
 } // namespace dwt_cuda
 
-
 #endif // IO_H
-

examples/dwt2d/dwt_cuda/transform_buffer.h

@@ -30,14 +30,11 @@
 /// POSSIBILITY OF SUCH DAMAGE.
 ///
 
-
 #ifndef TRANSFORM_BUFFER_H
 #define TRANSFORM_BUFFER_H
 
-
 namespace dwt_cuda {
 
-  
 /// Buffer (in shared memory of GPU) where block of input image is stored,
 /// but odd and even lines are separated. (Generates less bank conflicts when
 /// using lifting schema.) All operations expect SIZE_X threads.
@@ -79,8 +76,6 @@ namespace dwt_cuda {
   /// buffer for both even and odd columns
   T data[2 * BUFFER_SIZE + PADDING];
 
-    
-    
   /// Applies specified function to all central elements while also passing
   /// previous and next elements as parameters.
   /// @param count         count of central elements to apply function to
@@ -123,16 +118,13 @@ namespace dwt_cuda {
   }
 
 public:
-
   __device__ void getPrintData() {
     //
     for (int i = 0; i < 2 * BUFFER_SIZE + PADDING; i++) {
       printf(" index: %d  data: %f \n ", i, data[i]);
     }
-
   }
 
-    
   /// Gets offset of the column with given index. Central columns have
   /// indices from 0 to NUM_LINES - 1, left boundary columns have negative
   /// indices and right boundary columns indices start with NUM_LINES.
@@ -144,14 +136,10 @@ namespace dwt_cuda {
            + (columnIndex & 1) * ODD_OFFSET; // select odd or even buffer
   }
 
-    
   /// Provides access to data of the transform buffer.
   /// @param index  index of the item to work with
   /// @return reference to item at given index
-    __device__ T & operator[] (const int index) {
+  __device__ T &operator[](const int index) { return data[index]; }
-      return data[index];
-    }
-    
 
   /// Applies specified function to all horizontally even elements in
   /// specified lines. (Including even elements in boundaries except
@@ -163,8 +151,7 @@ namespace dwt_cuda {
   ///                   parameters: previous (odd) element, the even
   ///                   element itself and finally next (odd) element
   template <typename FUNC>
-    __device__ void forEachHorizontalEven(const int firstLine,
+  __device__ void forEachHorizontalEven(const int firstLine, const int numLines,
-                                          const int numLines,
                                         const FUNC &func) {
     // number of even elemens to apply function to
     const int count = numLines * VERTICAL_STRIDE - 1;
@@ -177,14 +164,14 @@ namespace dwt_cuda {
 
     // if(threadIdx.x == 0) {
 
-      //   printf("forEachHorizontalEven count  %d, centerOffset %d prevOffset %d nextOffset %d \n", count, centerOffset, prevOffset, nextOffset);
+    //   printf("forEachHorizontalEven count  %d, centerOffset %d prevOffset %d
+    //   nextOffset %d \n", count, centerOffset, prevOffset, nextOffset);
     // }
 
     // call generic horizontal step function
     horizontalStep(count, prevOffset, centerOffset, nextOffset, func);
   }
 
-    
   /// Applies given function to all horizontally odd elements in specified
   /// lines. (Including odd elements in boundaries except last odd element
   /// in last right boundary.) SIZE_X threads participate and synchronization
@@ -195,8 +182,7 @@ namespace dwt_cuda {
   ///                   parameters: previous (even) element, the odd
   ///                   element itself and finally next (even) element
   template <typename FUNC>
-    __device__ void forEachHorizontalOdd(const int firstLine,
+  __device__ void forEachHorizontalOdd(const int firstLine, const int numLines,
-                                         const int numLines,
                                        const FUNC &func) {
     // numbet of odd elements to apply function to
     const int count = numLines * VERTICAL_STRIDE - 1;
@@ -208,15 +194,14 @@ namespace dwt_cuda {
     const int nextOffset = prevOffset + 1;
 
     //  if(threadIdx.x == 0) {
-      //   printf("forEachHorizontalOdd count  %d, centerOffset %d prevOffset %d nextOffset %d \n", count, centerOffset, prevOffset, nextOffset);
+    //   printf("forEachHorizontalOdd count  %d, centerOffset %d prevOffset %d
+    //   nextOffset %d \n", count, centerOffset, prevOffset, nextOffset);
     // }
 
-      
     // call generic horizontal step function
     horizontalStep(count, prevOffset, centerOffset, nextOffset, func);
   }
 
-    
   /// Applies specified function to all even elements (except element #0)
   /// of given column. Each thread takes care of one column, so there's
   /// no need for synchronization.
@@ -238,17 +223,15 @@ namespace dwt_cuda {
         /*		__syncthreads();
                         if ((blockIdx.x * blockDim.x + threadIdx.x) == 0){
                                 diffOut[2500]++;
-			diffOut[diffOut[2500]] = 2;//data[columnOffset + row * VERTICAL_STRIDE];
+                                diffOut[diffOut[2500]] = 2;//data[columnOffset +
+           row * VERTICAL_STRIDE];
                         }
                         __syncthreads();
         */		  //--------------- FOR TEST -----------------
-		  
-		  
       }
     }
   }
 
-    
   /// Applies specified function to all odd elements of given column.
   /// Each thread takes care of one column, so there's no need for
   /// synchronization.
@@ -266,12 +249,12 @@ namespace dwt_cuda {
 
       f(prev, data[columnOffset + row * VERTICAL_STRIDE], next);
 
-		
       //--------------- FOR TEST -----------------
       /*		__syncthreads();
                       if ((blockIdx.x * blockDim.x + threadIdx.x) == 0){
                               diffOut[2500]++;
-			diffOut[diffOut[2500]] = 1; //data[columnOffset + row * VERTICAL_STRIDE];
+                              diffOut[diffOut[2500]] = 1; //data[columnOffset +
+         row * VERTICAL_STRIDE];
                       }
 
                       __syncthreads();
@@ -279,8 +262,6 @@ namespace dwt_cuda {
     }
   }
 
-    
-    
   /// Scales elements at specified lines.
   /// @param evenScale  scaling factor for horizontally even elements
   /// @param oddScale   scaling factor for horizontally odd elements
@@ -294,7 +275,9 @@ namespace dwt_cuda {
     const int finalCount = count % SIZE_X;
     const int finalOffset = count - finalCount;
 
-      // printf("scaleHorizontal sizeX: %d  offset %d, count, %d, steps, %d, finalCount %d, finalOffset %d \n", SIZE_X, offset, count, steps, finalCount, finalOffset);
+    // printf("scaleHorizontal sizeX: %d  offset %d, count, %d, steps, %d,
+    // finalCount %d, finalOffset %d \n", SIZE_X, offset, count, steps,
+    // finalCount, finalOffset);
 
     // run iterations, whete all threads participate
     for (int i = 0; i < steps; i++) {
@@ -319,10 +302,8 @@ namespace dwt_cuda {
       // }
       data[threadIdx.x + finalOffset + offset + ODD_OFFSET] *= oddScale;
     }
-
   }
 
-    
   /// Scales elements in specified column.
   /// @param evenScale     scaling factor for vertically even elements
   /// @param oddScale      scaling factor for vertically odd elements
@@ -341,33 +322,17 @@ namespace dwt_cuda {
     }
   }
 
+  //****************For Test(Feb23), test inter parameters*************
+  __device__ int getVERTICAL_STRIDE() { return VERTICAL_STRIDE; }
+  __device__ int getSHM_BANKS() { return SHM_BANKS; }
+  __device__ int getBuffersize() { return BUFFER_SIZE; }
+  __device__ int getPADDING() { return PADDING; }
+  __device__ int getODD_OFFSET() { return ODD_OFFSET; }
 
   //****************For Test(Feb23), test inter parameters*************
-	__device__ int getVERTICAL_STRIDE(){
-		return VERTICAL_STRIDE;
-	}
-	__device__ int getSHM_BANKS(){
-		return SHM_BANKS;
-	}
-	__device__ int  getBuffersize(){		
-		return BUFFER_SIZE;
-	}
-	__device__ int getPADDING(){
-		return PADDING;
-	}
-	__device__ int getODD_OFFSET(){
-		return ODD_OFFSET;
-	}
-
-
-    //****************For Test(Feb23), test inter parameters*************
-	
 
 }; // end of class TransformBuffer
 
-
 } // namespace dwt_cuda
 
-
 #endif // TRANSFORM_BUFFER_H
-

examples/dwt2d/run_cpu.sh

@@ -5,4 +5,3 @@
 ./dwt2d 4.bmp  -d 4x4 -r -5 -l 3
 # ./dwt2d 4.bmp  -d 4x4 -r -9 -l 3
 # ./dwt2d 8.bmp  -d 8x8 -f -9 -l 3
-

examples/dwt2d/test_compile_nvcc.sh

@@ -7,12 +7,3 @@
 /usr/local/cuda/bin/nvcc -arch sm_50 -I. -I/include  -O2  --compiler-options -fno-strict-aliasing -c dwt_cuda/rdwt97.cu -o dwt_cuda/rdwt97.cu.o
 /usr/local/cuda/bin/nvcc -arch sm_50 -I. -I/include  -O2  --compiler-options -fno-strict-aliasing -c dwt_cuda/rdwt53.cu -o dwt_cuda/rdwt53.cu.o
 g++ -fPIC -o nvcc_dwt2d main.cu.o dwt.cu.o components.cu.o dwt_cuda/fdwt53.cu.o dwt_cuda/fdwt97.cu.o dwt_cuda/common.cu.o dwt_cuda/rdwt97.cu.o dwt_cuda/rdwt53.cu.o -L/usr/local/cuda/lib64 -lcudart
-
-
-
-
-
-
-
-
-

examples/microbench/cudamemcpy_test.cc

@@ -1,14 +1,12 @@
 #include <stdio.h>
 
-__global__
+__global__ void saxpy(int n, float a, float *x, float *y) {
-void saxpy(int n, float a, float *x, float *y)
-{
   int i = blockIdx.x * blockDim.x + threadIdx.x;
-  if (i < n) y[i] = a*x[i] + y[i];
+  if (i < n)
+    y[i] = a * x[i] + y[i];
 }
 
-int main(void)
+int main(void) {
-{
   int N = 1 << 20;
   float *x, *y, *d_x, *d_y;
   x = (float *)malloc(N * sizeof(float));

examples/microbench/dummy_kernel.cc

@@ -1,14 +1,11 @@
 #include <stdio.h>
 
-__global__
+__global__ void saxpy(void) {
-void saxpy(void)
-{
   int i = blockIdx.x * blockDim.x + threadIdx.x;
   printf("block_id:%d thread_id:%d \n", i)
 }
 
-int main(void)
+int main(void) {
-{
   int N = 1 << 20;
   float *x, *y, *d_x, *d_y;
   x = (float *)malloc(N * sizeof(float));

examples/microbench/kerne_arg.cc

@@ -1,13 +1,8 @@
 #include <stdio.h>
 
-__global__
+__global__ void saxpy(int N) { printf("hello!: %d\n", N); }
-void saxpy(int N)
-{
-printf("hello!: %d\n", N);
-}
 
-int main(void)
+int main(void) {
-{
   int N = 1 << 20;
   float *x, *y, *d_x, *d_y;
   x = (float *)malloc(N * sizeof(float));

examples/microbench/one_thread_kernel.cc

@@ -1,13 +1,8 @@
 #include <stdio.h>
 
-__global__
+__global__ void saxpy(void) { printf("hello!\n"); }
-void saxpy(void)
-{
-printf("hello!\n");
-}
 
-int main(void)
+int main(void) {
-{
   int N = 1 << 20;
   float *x, *y, *d_x, *d_y;
   x = (float *)malloc(N * sizeof(float));