+1

examples_ptx/mandelbrot_tasks/Makefile

@@ -1,8 +1,63 @@
-
-EXAMPLE=mandelbrot_tasks
-CPP_SRC=mandelbrot_tasks.cpp mandelbrot_tasks_serial.cpp
+PROG=mandelbrot_tasks
 ISPC_SRC=mandelbrot_tasks.ispc
-ISPC_IA_TARGETS=sse2-i32x4,sse4-i32x8,avx1-i32x16,avx2-i32x16
-ISPC_ARM_TARGETS=neon
+CXX_SRC=mandelbrot_tasks.cpp  mandelbrot_tasks_serial.cpp
+NVCC_SRC=../nvcc_helpers.cu
+
+CXX=g++
+CXXFLAGS=-O3 -I$(CUDATK)/include
+
+NVCC=nvcc
+NVCC_FLAGS=-O3 -arch=sm_35
+
+LD=nvcc
+LDFLAGS=-lcudart -lcudadevrt
+
+PTXCC=ptxcc
+PTXCC_FLAGS = -maxrregcount=32 -Xptxas=-v
+
+ISPC=ispc
+ISPC_FLAGS=-O3 --math-lib=default --target=nvptx64 --opt=fast-math
+
+# PTXGEN = $(HOME)/ptxgen
+# PTXGEN += -opt=3
+# PTXGEN += -ftz=1 -prec-div=0 -prec-sqrt=0 -fma=1
+
+###.SUFFIXES: .bc .o .cu 
+
+
+ISPC_OBJS=objs/$(ISPC_SRC:.ispc=_ispc.o)
+ISPC_BCS=objs/$(ISPC_SRC:.ispc=_ispc.bc)
+ISPC_HEADERS=objs/$(ISPC_SRC:.ispc=_ispc.h)
+CXX_OBJS=objs/$(CXX_SRC:.cpp=_gcc.o)
+NVCC_OBJS=objs/$(NVCC_SRC:.cu=_nvcc.o)
+
+OBJS=$(ISPC_OBJS) $(CXX_OBJS) $(NVCC_OBJS)
+
+all: $(PROG)
+
+dirs:
+	/bin/mkdir -p objs/
+
+objs/%.cpp objs/%.o objs/%.h: dirs
+
+clean: 
+	/bin/rm -rf $(PROG) objs
+
+$(PROG): $(OBJS)
+	$(LD) -o $@ $^ $(LDFLAGS)
+
+objs/%_gcc.o: %.cpp dirs $(ISPC_HEADERS)
+	$(CXX) $(CXXFLAGS)  -o $@ -c $<
+
+objs/%_cu.o: %.cu dirs
+	$(NVCC) $(NVCC_FLAGS) -o $@ -c $<
+
+objs/%_ispc.h objs/%_ispc.bc: %.ispc dirs
+	$(ISPC) $(ISPC_FLAGS) --emit-llvm -h objs/$*_ispc.h -o objs/$*_ispc.bc $<
+
+objs/%_ispc.o: objs/%_ispc.bc 
+	$(PTXCC) $(PTXCC_FLAGS) $< -o $@
+
+
+	 
 
-include ../common.mk

examples_ptx/mandelbrot_tasks/mandelbrot_tasks.ispc

@@ -33,21 +33,21 @@
 
 static inline int
 mandel(float c_re, float c_im, int count) {
-    float z_re = c_re, z_im = c_im;
-    int i;
-    for (i = 0; i < count; ++i) {
-        if (z_re * z_re + z_im * z_im > 4.)
-            break;
+  float z_re = c_re, z_im = c_im;
+  int i;
+  for (i = 0; i < count; ++i) {
+    if (z_re * z_re + z_im * z_im > 4.0f)
+      break;
 
-        float new_re = z_re*z_re - z_im*z_im;
-        float new_im = 2.f * z_re * z_im;
-        unmasked {
-            z_re = c_re + new_re;
-            z_im = c_im + new_im;
-        }
+    float new_re = z_re*z_re - z_im*z_im;
+    float new_im = 2.0f * z_re * z_im;
+    unmasked {
+      z_re = c_re + new_re;
+      z_im = c_im + new_im;
     }
+  }
 
-    return i;
+  return i;
 }