diff --git a/llvm/test/VISC/parboil/benchmarks/pipeline/Makefile b/llvm/test/VISC/parboil/benchmarks/pipeline/Makefile
index e57042e682df48f745aff09a782bb7e0f8abbb47..3a265f737f93433d5a83a5a6a503a6c879a09d8a 100644
--- a/llvm/test/VISC/parboil/benchmarks/pipeline/Makefile
+++ b/llvm/test/VISC/parboil/benchmarks/pipeline/Makefile
@@ -16,6 +16,16 @@ ifeq ($(PLATFORM),)
PLATFORM=default
endif
+ifeq ($(POS),)
+ POS = left
+endif
+
+ifeq ($(POS),middle)
+ APP_CXXFLAGS+=-DMIDDLE
+else ifeq ($(POS),right)
+ APP_CXXFLAGS+=-DRIGHT
+endif
+
BIN = $(addsuffix -$(VERSION), $(APP))
SRCDIR = src/$(VERSION)
@@ -37,7 +47,7 @@ OUTPUT = $(RUNDIR)/output.txt
#ARGS = -i $(VIDEO1) -o $(OUTPUT)
#ARGS = -i $(VIDEO)
#ARGS = -i $(VIDEO3)
-ARGS = -i $(VIDEO4)
+ARGS = -i $(VIDEO4)
#ARGS = -i $(MATRIX) -o $(OUTPUT)
#TOOL = tools/compare-output
TOOL=echo
diff --git a/llvm/test/VISC/parboil/benchmarks/pipeline/src/visc_parallel/main.cc b/llvm/test/VISC/parboil/benchmarks/pipeline/src/visc_parallel/main.cc
index 3efbcc1ebeff30be0351dd82dbb767ef7963c04e..838d461e247df0f58e9c7d701e9714f4065048dd 100755
--- a/llvm/test/VISC/parboil/benchmarks/pipeline/src/visc_parallel/main.cc
+++ b/llvm/test/VISC/parboil/benchmarks/pipeline/src/visc_parallel/main.cc
@@ -23,6 +23,39 @@
#include <parboil.h>
#include <visc.h>
+
+#define NUM_RUNS 5
+#define DEPTH 3
+#define HEIGHT 640
+#define WIDTH 480
+std::string input_window = "GPU Pipeline - Input Video";
+std::string output_window = "GPU Pipeline - Edge Mapping";
+
+
+#ifdef MIDDLE
+ #define POSX_IN 640
+ #define POSY_IN 0
+ #define POSX_OUT 640
+ #define POSY_OUT 540
+
+#elif RIGHT
+ #define POSX_IN 1280
+ #define POSY_IN 0
+ #define POSX_OUT 1280
+ #define POSY_OUT 540
+
+#else // LEFT
+ #define POSX_IN 0
+ #define POSY_IN 0
+ #define POSX_OUT 0
+ #define POSY_OUT 540
+#endif
+
+
+//#define NUM_FRAMES 20
+
+
+
// Definitions of sizes for edge detection kernels
#define MIN_BR 0.0f
@@ -205,7 +238,6 @@ void laplacianEstimate(float *Is, size_t bytesIs,
int m, int n) {
__visc__hint(visc::DEVICE);
- //__visc__hint(visc::GPU_TARGET);
__visc__attributes(2, Is, B, 1, L);
// 3x3 image area
float imageArea[SZB][SZB];
@@ -315,7 +347,7 @@ void computeZeroCrossings(float *L, size_t bytesL,
float *S, size_t bytesS,
int m, int n) {
__visc__hint(visc::DEVICE);
- //__visc__hint(visc::SPIR_TARGET);
+ //__visc__hint(visc::CPU_TARGET);
__visc__attributes(2, L, B, 1, S);
// 3x3 image area
@@ -525,6 +557,7 @@ void computeMaxGradientLeaf(float *G, size_t bytesG,
int m, int n) {
__visc__hint(visc::DEVICE);
+ //__visc__hint(visc::CPU_TARGET);
__visc__attributes(1, G, 1, maxG);
void* thisNode = __visc__getNode();
@@ -600,6 +633,7 @@ void computeMaxGradientTB(float *G, size_t bytesG,
int m, int n,
int block_x) {
__visc__hint(visc::DEVICE);
+ //__visc__hint(visc::CPU_TARGET);
__visc__attributes(2, G, maxG, 1, maxG);
void* CMGLeafNode = __visc__createNode1D(computeMaxGradientLeaf, block_x);
__visc__bindIn(CMGLeafNode, 0, 0, 0); // Bind G
@@ -646,6 +680,7 @@ void rejectZeroCrossings(float *S, size_t bytesS,
float *E, size_t bytesE,
int m, int n) {
__visc__hint(visc::DEVICE);
+ //__visc__hint(visc::CPU_TARGET);
__visc__attributes(3, S, G, maxG, 1, E);
void* thisNode = __visc__getNode();
@@ -786,9 +821,7 @@ void edgeDetection(float *I, size_t bytesI, // 0
}
}
-#define NUM_RUNS 1
-#define DEPTH 3
-//#define NUM_FRAMES 20
+
using namespace cv;
void getNextFrame(VideoCapture& VC, Mat& F) {
@@ -801,6 +834,15 @@ void getNextFrame(VideoCapture& VC, Mat& F) {
}
+//void showInOut(Mat& Input, Mat& Output) {
+ //Mat in, out;
+ //resize(Input, in, Size(512, 768));
+ //resize(Output, out, Size(512, 768));
+ //imshow(input_window, in);
+ //imshow(output_window, out);
+//}
+
+
int main (int argc, char *argv[]) {
struct pb_Parameters *params;
@@ -808,8 +850,6 @@ int main (int argc, char *argv[]) {
size_t I_sz;
int block_x, grid_x;
- std::string input_window = "Input Frame";
- std::string output_window = "Edge Map";
std::cout << "Using OpenCV" << CV_VERSION << "\n";
@@ -835,6 +875,8 @@ int main (int argc, char *argv[]) {
std::cout << "Number of frames = " << NUM_FRAMES << "\n";
namedWindow(input_window, CV_WINDOW_AUTOSIZE);
namedWindow(output_window, CV_WINDOW_AUTOSIZE);
+ moveWindow(input_window, POSX_IN, POSY_IN);
+ moveWindow(output_window, POSX_OUT, POSY_OUT);
Mat src, Is, L, S, G, E;
@@ -858,7 +900,7 @@ int main (int argc, char *argv[]) {
pb_InitializeTimerSet(&timers);
__visc__init();
- pb_SwitchToTimer( &timers, pb_TimerID_COMPUTE );
+ //pb_SwitchToTimer( &timers, pb_TimerID_COMPUTE );
// copy A to device memory
I_sz = src.size[0]*src.size[1]*sizeof(float);
@@ -895,14 +937,19 @@ int main (int argc, char *argv[]) {
grid_x = 1;
// Copy A and B^T into device memory
- pb_SwitchToTimer( &timers, pb_TimerID_COMPUTE );
+ //pb_SwitchToTimer( &timers, pb_TimerID_COMPUTE );
+
+ //showInOut(src, E);
+ Mat in, out;
+ resize(src, in, Size(HEIGHT, WIDTH));
+ resize(E, out, Size(HEIGHT, WIDTH));
+ imshow(input_window, in);
+ imshow(output_window, out);
+ waitKey(0);
+ //NUM_FRAMES = 20;
pb_SwitchToTimer( &timers, visc_TimerID_COMPUTATION );
struct InStruct* args = (struct InStruct*)malloc (sizeof(InStruct));
-
- imshow(input_window, src);
- imshow(output_window, Is);
- waitKey(0);
packData(args, (float*)src.data, I_sz,
(float*)Is.data, I_sz,
(float*)L.data, I_sz,
@@ -920,16 +967,18 @@ int main (int argc, char *argv[]) {
// Check if the total elements is a multiple of block size
assert(src.size[0]*src.size[1] % block_x == 0);
- imshow(input_window, src);
- imshow(output_window, E);
- waitKey(0);
+ //imshow(input_window, src);
+ //imshow(output_window, E);
+ //waitKey(0);
for(unsigned j=0; j<NUM_RUNS; j++) {
std::cout << "Run: " << j << "\n";
void* DFG = __visc__launch(1, edgeDetection, (void*)args);
cap = VideoCapture(params->inpFiles[0]);
getNextFrame(cap, src);
+
//packData(args, A.data, BlockSize, &matB[i], BlockSize, &matC[i], BlockSize, BlockElements);
+
if(NUM_FRAMES >=2) {
//__visc__push(DFG, args);
//__visc__push(DFG, args);
@@ -938,6 +987,7 @@ int main (int argc, char *argv[]) {
args->I = (float*) src.data;
*maxG = 0.0;
+
llvm_visc_track_mem(src.data, I_sz);
llvm_visc_track_mem(Is.data, I_sz);
llvm_visc_track_mem(L.data, I_sz);
@@ -949,7 +999,7 @@ int main (int argc, char *argv[]) {
llvm_visc_track_mem(B, bytesB);
llvm_visc_track_mem(Sx, bytesSx);
llvm_visc_track_mem(Sy, bytesSy);
-
+
__visc__push(DFG, args);
__visc__pop(DFG);
@@ -964,12 +1014,38 @@ int main (int argc, char *argv[]) {
//llvm_visc_request_mem(G.data, I_sz);
llvm_visc_request_mem(maxG, bytesMaxG);
llvm_visc_request_mem(E.data, I_sz);
+ //std::cout << "src.data = " << (float*)src.data << "\n";
+ //std::cout << "Is.data = " << (float*)Is.data << "\n";
+ //std::cout << "L.data = " << (float*)L.data << "\n";
+ //std::cout << "S.data = " << (float*)S.data << "\n";
+ //std::cout << "G.data = " << (float*)G.data << "\n";
+ //std::cout << "E.data = " << (float*)E.data << "\n";
//std::cout << "Max G = " << *maxG << "\n";
-
- imshow(output_window, E);
- imshow(input_window, src);
+
+ Mat in, out;
+ resize(src, in, Size(HEIGHT, WIDTH));
+ //std::cout << "Show E\n";
+ resize(E, out, Size(HEIGHT, WIDTH));
+ imshow(output_window, out);
+ imshow(input_window, in);
waitKey(1);
- getNextFrame(cap, src);
+ //waitKey(0);
+ //std::cout << "Show Is\n";
+ //resize(Is, out, Size(HEIGHT, WIDTH));
+ //imshow(output_window, out);
+ //waitKey(0);
+ //std::cout << "Show L\n";
+ //resize(L, out, Size(HEIGHT, WIDTH));
+ //imshow(output_window, out);
+ //waitKey(0);
+ //std::cout << "Show S\n";
+ //resize(S, out, Size(HEIGHT, WIDTH));
+ //imshow(output_window, out);
+ //waitKey(0);
+ //std::cout << "Show G\n";
+ //resize(G, out, Size(HEIGHT, WIDTH));
+ //imshow(output_window, out);
+ //waitKey(0);
llvm_visc_untrack_mem(src.data);
llvm_visc_untrack_mem(Is.data);
@@ -983,7 +1059,7 @@ int main (int argc, char *argv[]) {
llvm_visc_untrack_mem(Sx);
llvm_visc_untrack_mem(Sy);
-
+ getNextFrame(cap, src);
}
//__visc__pop(DFG);
@@ -993,7 +1069,8 @@ int main (int argc, char *argv[]) {
__visc__push(DFG, args);
__visc__pop(DFG);
}
-
+
+
__visc__wait(DFG);
}
@@ -1011,26 +1088,26 @@ int main (int argc, char *argv[]) {
//src.size[0], src.size[1], matE);
//}
- std::cout << "Show Is" << "\n";
- Mat output(src.size[0], src.size[1], CV_32F);
- imshow(output_window, Is);
- waitKey(0);
-
- std::cout << "Show G" << "\n";
- imshow(output_window, L);
- waitKey(0);
-
- std::cout << "Show L" << "\n";
- imshow(output_window, S);
- waitKey(0);
+ //std::cout << "Show Is" << "\n";
+ //Mat output(src.size[0], src.size[1], CV_32F);
+ //imshow(output_window, Is);
+ //waitKey(0);
+
+ //std::cout << "Show G" << "\n";
+ //imshow(output_window, L);
+ //waitKey(0);
+
+ //std::cout << "Show L" << "\n";
+ //imshow(output_window, S);
+ //waitKey(0);
- std::cout << "Show S" << "\n";
- imshow(output_window, G);
- waitKey(0);
+ //std::cout << "Show S" << "\n";
+ //imshow(output_window, G);
+ //waitKey(0);
- std::cout << "Show E" << "\n";
- imshow(output_window, E);
- waitKey(0);
+ //std::cout << "Show E" << "\n";
+ //imshow(output_window, E);
+ //waitKey(0);
//double GPUtime = pb_GetElapsedTime(&(timers.timers[pb_TimerID_KERNEL]));
//std::cout<< "GFLOPs = " << 2.* src.size[0] * src.size[1] * src.size[1]/GPUtime/1e9 << std::endl;