Makefile

@@ -10,6 +10,9 @@ static:
 shared:
 	$(MAKE) -C src shared
 
+dylib:
+	$(MAKE) -C src dylib
+
 sample:
 	$(MAKE) -C example
 

src/Makefile

@@ -5,6 +5,7 @@ AR ?= ar
 
 TARGET = libeclfft.a
 TARGET2 = libeclfft.so
+TARGET3 = libeclfft.dylib
 
 ifndef OPENCL_INCLUDE
 ifdef CUDA_INSTALL_PATH
@@ -35,6 +36,8 @@ static: $(TARGET)
 
 shared: $(TARGET2)
 
+dylib: $(TARGET3)
+
 $(TARGET): $(OBJECTS)
 	$(AR) rcs $(TARGET) $(OBJECTS)
 	mkdir -p ../lib
@@ -43,6 +46,9 @@ $(TARGET): $(OBJECTS)
 $(TARGET2): $(OBJECTS)
 	$(CXX) $(LDFLAGS) -shared -o $(TARGET2) $(OBJECTS)
 
+$(TARGET3): $(OBJECTS)
+	$(CXX) $(LDFLAGS) -dynamiclib -o $(TARGET3) $(OBJECTS)
+
 fft_setup.o: fft_setup.cpp fft_internal.h fft_base_kernels.h
 	$(CXX) $(CXXFLAGS) -c fft_setup.cpp
 
@@ -58,4 +64,4 @@ install: ../include/clFFT.h libeclfft.*
 	cp  libeclfft.* $(PREFIX)/lib/
 
 clean:
-	rm -f *.o *.a *.so
+	rm -f *.o *.a *.so *.dylib

src/fft_execute.cpp

@@ -348,7 +348,7 @@ clFFT_ExecutePlannar( cl_command_queue queue, clFFT_Plan Plan, cl_int batchSize,
 
 cl_int
 clFFT_1DTwistInterleaved(clFFT_Plan Plan, cl_command_queue queue, cl_mem array,
-                         size_t numRows, size_t numCols, size_t startRow, size_t rowsToProcess, clFFT_Direction dir)
+                         unsigned numRows, unsigned numCols, unsigned startRow, unsigned rowsToProcess, clFFT_Direction dir)
 {
     cl_fft_plan *plan = (cl_fft_plan *) Plan;
 
@@ -387,7 +387,7 @@ clFFT_1DTwistInterleaved(clFFT_Plan Plan, cl_command_queue queue, cl_mem array,
 
 cl_int
 clFFT_1DTwistPlannar(clFFT_Plan Plan, cl_command_queue queue, cl_mem array_real, cl_mem array_imag,
-                     size_t numRows, size_t numCols, size_t startRow, size_t rowsToProcess, clFFT_Direction dir)
+                     unsigned numRows, unsigned numCols, unsigned startRow, unsigned rowsToProcess, clFFT_Direction dir)
 {
     cl_fft_plan *plan = (cl_fft_plan *) Plan;
 

src/fft_internal.h

@@ -81,10 +81,10 @@ typedef struct kernel_info_t
 {
     cl_kernel kernel;
     char *kernel_name;
-    size_t lmem_size;
-    size_t num_workgroups;
-    size_t num_xforms_per_workgroup;
-    size_t num_workitems_per_workgroup;
+    unsigned lmem_size;
+    unsigned num_workgroups;
+    unsigned num_xforms_per_workgroup;
+    unsigned num_workitems_per_workgroup;
     cl_fft_kernel_dir dir;
     int in_place_possible;
     kernel_info_t *next;
@@ -138,7 +138,7 @@ typedef struct
     // batch size different than the first call. last_batch_size caches the last
     // batch size with which this plan is used so that we dont keep allocating/deallocating
     // temp buffer if same batch size is used again and again.
-    size_t                  last_batch_size;
+    unsigned                  last_batch_size;
 
     // temporary buffer for interleaved plan
     cl_mem                  tempmemobj;
@@ -163,25 +163,25 @@ typedef struct
     // Maximum size of signal for which local memory transposed based
     // fft is sufficient i.e. no global mem transpose (communication)
     // is needed
-    size_t                  max_localmem_fft_size;
+    unsigned                    max_localmem_fft_size;
 
     // Maximum work items per work group allowed. This, along with max_radix below controls
     // maximum local memory being used by fft kernels of this plan. Set to 256 by default
-    size_t                  max_work_item_per_workgroup;
+    unsigned                  max_work_item_per_workgroup;
 
     // Maximum base radix for local memory fft ... this controls the maximum register
     // space used by work items. Currently defaults to 16
-    size_t                  max_radix;
+    unsigned                  max_radix;
 
     // Device depended parameter that tells how many work-items need to be read consecutive
     // values to make sure global memory access by work-items of a work-group result in
     // coalesced memory access to utilize full bandwidth e.g. on NVidia tesla, this is 16
-    size_t                  min_mem_coalesce_width;
+    unsigned                  min_mem_coalesce_width;
 
     // Number of local memory banks. This is used to geneate kernel with local memory
     // transposes with appropriate padding to avoid bank conflicts to local memory
     // e.g. on NVidia it is 16.
-    size_t                  num_local_mem_banks;
+    unsigned                  num_local_mem_banks;
 }cl_fft_plan;
 
 void FFT1D(cl_fft_plan *plan, cl_fft_kernel_dir dir);

src/fft_kernelstring.cpp

@@ -605,7 +605,7 @@ static void
 insertTwiddleKernel(string &kernelString, int Nr, int numIter, int Nprev, int len, int numWorkItemsPerXForm)
 {
     int z, k;
-    int logNPrev = log2(Nprev);
+    int logNPrev = (int)log2(Nprev);
 
     for(z = 0; z < numIter; z++)
     {
@@ -716,8 +716,8 @@ static void
 insertLocalLoadIndexArithmatic(string &kernelString, int Nprev, int Nr, int numWorkItemsReq, int numWorkItemsPerXForm, int numXFormsPerWG, int offset, int midPad)
 {
     int Ncurr = Nprev * Nr;
-    int logNcurr = log2(Ncurr);
-    int logNprev = log2(Nprev);
+    int logNcurr = (int)log2(Ncurr);
+    int logNprev = (int)log2(Nprev);
     int incr = (numWorkItemsReq + offset) * Nr + midPad;
 
     if(Ncurr < numWorkItemsPerXForm)
@@ -1280,7 +1280,7 @@ createGlobalFFTKernelString(cl_fft_plan *plan, int n, int BS, cl_fft_kernel_dir
         int gInInc = threadsPerBlock / batchSize;
 
 
-        int lgStrideO = log2(strideO);
+        int lgStrideO = (int)log2(strideO);
         int numBlocksPerXForm = strideI / batchSize;
         int numBlocks = numBlocksPerXForm;
         if(!vertical)
@@ -1333,7 +1333,7 @@ createGlobalFFTKernelString(cl_fft_plan *plan, int n, int BS, cl_fft_kernel_dir
         }
         else
         {
-            int lgNumBlocksPerXForm = log2(numBlocksPerXForm);
+            int lgNumBlocksPerXForm = (int)log2(numBlocksPerXForm);
             localString += string("bNum = groupId & ") + num2str(numBlocksPerXForm - 1) + string(";\n");
             localString += string("xNum = groupId >> ") + num2str(lgNumBlocksPerXForm) + string(";\n");
             localString += string("indexIn = mul24(bNum, ") + num2str(batchSize) + string(");\n");
@@ -1349,7 +1349,7 @@ createGlobalFFTKernelString(cl_fft_plan *plan, int n, int BS, cl_fft_kernel_dir
         }
 
         // Load Data
-        int lgBatchSize = log2(batchSize);
+        int lgBatchSize = (int)log2(batchSize);
         localString += string("tid = lId;\n");
         localString += string("i = tid & ") + num2str(batchSize - 1) + string(";\n");
         localString += string("j = tid >> ") + num2str(lgBatchSize) + string(";\n");
@@ -1442,40 +1442,40 @@ createGlobalFFTKernelString(cl_fft_plan *plan, int n, int BS, cl_fft_kernel_dir
             localString += string("lMemStore = sMem + mad24(i, ") + num2str(radix + 1) + string(", j << ") + num2str((int)log2(R1/R2)) + string(");\n");
             localString += string("lMemLoad = sMem + mad24(tid >> ") + num2str((int)log2(radix)) + string(", ") + num2str(radix+1) + string(", tid & ") + num2str(radix-1) + string(");\n");
 
-            for(int i = 0; i < R1/R2; i++)
-                for(int j = 0; j < R2; j++)
+            for(i = 0; i < R1/R2; i++)
+                for(j = 0; j < R2; j++)
                     localString += string("lMemStore[ ") + num2str(i + j*R1) + string("] = a[") + num2str(i*R2+j) + string("].x;\n");
             localString += string("barrier(CLK_LOCAL_MEM_FENCE);\n");
             if(threadsPerBlock >= radix)
             {
-                for(int i = 0; i < R1; i++)
+                for(i = 0; i < R1; i++)
                 localString += string("a[") + num2str(i) + string("].x = lMemLoad[") + num2str(i*(radix+1)*(threadsPerBlock/radix)) + string("];\n");
             }
             else
             {
                 int innerIter = radix/threadsPerBlock;
                 int outerIter = R1/innerIter;
-                for(int i = 0; i < outerIter; i++)
-                    for(int j = 0; j < innerIter; j++)
+                for(i = 0; i < outerIter; i++)
+                    for(j = 0; j < innerIter; j++)
                         localString += string("a[") + num2str(i*innerIter+j) + string("].x = lMemLoad[") + num2str(j*threadsPerBlock + i*(radix+1)) + string("];\n");
             }
             localString += string("barrier(CLK_LOCAL_MEM_FENCE);\n");
 
-            for(int i = 0; i < R1/R2; i++)
-                for(int j = 0; j < R2; j++)
+            for(i = 0; i < R1/R2; i++)
+                for(j = 0; j < R2; j++)
                     localString += string("lMemStore[ ") + num2str(i + j*R1) + string("] = a[") + num2str(i*R2+j) + string("].y;\n");
             localString += string("barrier(CLK_LOCAL_MEM_FENCE);\n");
             if(threadsPerBlock >= radix)
             {
-                for(int i = 0; i < R1; i++)
+                for(i = 0; i < R1; i++)
                     localString += string("a[") + num2str(i) + string("].y = lMemLoad[") + num2str(i*(radix+1)*(threadsPerBlock/radix)) + string("];\n");
             }
             else
             {
                 int innerIter = radix/threadsPerBlock;
                 int outerIter = R1/innerIter;
-                for(int i = 0; i < outerIter; i++)
-                    for(int j = 0; j < innerIter; j++)
+                for(i = 0; i < outerIter; i++)
+                    for(j = 0; j < innerIter; j++)
                         localString += string("a[") + num2str(i*innerIter+j) + string("].y = lMemLoad[") + num2str(j*threadsPerBlock + i*(radix+1)) + string("];\n");
             }
             localString += string("barrier(CLK_LOCAL_MEM_FENCE);\n");

src/fft_setup.cpp

@@ -219,7 +219,7 @@ int getMaxKernelWorkGroupSize(cl_fft_plan *plan, unsigned int *max_wg_size, unsi
     int reg_needed = 0;
     *max_wg_size = INT_MAX;
     int err;
-    size_t wg_size;
+    unsigned wg_size;
 
     unsigned int i;
     for(i = 0; i < num_devices; i++)
@@ -427,7 +427,7 @@ patch_kernel_source:
     err = clGetContextInfo(context, CL_CONTEXT_DEVICES, sizeof(devices), devices, &ret_size);
     ERR_MACRO(err);
 
-    num_devices = ret_size / sizeof(cl_device_id);
+    num_devices = (int)(ret_size / sizeof(cl_device_id));
 
     for(i = 0; i < num_devices; i++)
     {