Add additional samples based on CUDA toolkit version 11.7 and cuFile …

…library 1.3.0.44

gds/samples/Makefile

@@ -67,7 +67,7 @@ release_samples:
 	@echo "Release samples Built"
 
 
-build: cufile_sample_001 cufile_sample_002 cufile_sample_003 cufile_sample_004 cufile_sample_005 cufile_sample_006 cufile_sample_007 cufile_sample_008 cufile_sample_009 cufile_sample_010 cufile_sample_011 cufile_sample_012 cufile_sample_013 cufile_sample_014 cufile_sample_015 cufile_sample_016 cufile_sample_017 cufile_sample_018 cufile_sample_001_static cufile_sample_002_static cufile_sample_003_static cufile_sample_004_static cufile_sample_005_static cufile_sample_006_static cufile_sample_007_static cufile_sample_008_static cufile_sample_009_static cufile_sample_010_static cufile_sample_011_static cufile_sample_012_static cufile_sample_013_static cufile_sample_014_static cufile_sample_015_static cufile_sample_016_static cufile_sample_017_static cufile_sample_018_static
+build: cufile_sample_001 cufile_sample_002 cufile_sample_003 cufile_sample_004 cufile_sample_005 cufile_sample_006 cufile_sample_007 cufile_sample_008 cufile_sample_009 cufile_sample_010 cufile_sample_011 cufile_sample_012 cufile_sample_013 cufile_sample_014 cufile_sample_015 cufile_sample_016 cufile_sample_017 cufile_sample_018 cufile_sample_019 cufile_sample_001_static cufile_sample_002_static cufile_sample_003_static cufile_sample_004_static cufile_sample_005_static cufile_sample_006_static cufile_sample_007_static cufile_sample_008_static cufile_sample_009_static cufile_sample_010_static cufile_sample_011_static cufile_sample_012_static cufile_sample_013_static cufile_sample_014_static cufile_sample_015_static cufile_sample_016_static cufile_sample_017_static cufile_sample_018_static cufile_sample_019_static cufile_sample_020_static cufile_sample_020 cufile_sample_021 cufile_sample_022
 
 %: %.cc $(CUFILE_INCLUDE_PATH)/cufile.h
 	$(CC) $(CXXFLAGS) $< -o $@ $(LDFLAGS)
@@ -78,6 +78,6 @@ install:
 	cp cufile_sample_*.cc cufile_sample_*.h Makefile README $(INSTALL_GDSSAMPLES_PREFIX)/
 
 clean:
-	rm -f cufile_sample_001 cufile_sample_001.o cufile_sample_002 cufile_sample_002.o cufile_sample_003 cufile_sample_003.o cufile_sample_004 cufile_sample_004.o cufile_sample_005 cufile_sample_005.o cufile_sample_006 cufile_sample_006.o cufile_sample_007 cufile_sample_007.o cufile_sample_008 cufile_sample_008.o cufile_sample_009 cufile_sample_009.o cufile_sample_010 cufile_sample_010.o cufile_sample_011 cufile_sample_011.o cufile_sample_012 cufile_sample_012.o cufile_sample_013 cufile_sample_013.o cufile_sample_014 cufile_sample_014.o cufile_sample_015 cufile_sample_015.o cufile_sample_016.o cufile_sample_016 cufile_sample_017.o cufile_sample_017 cufile_sample_018.o cufile_sample_018 cufile_sample_001_static cufile_sample_002_static cufile_sample_003_static cufile_sample_004_static cufile_sample_005_static cufile_sample_006_static cufile_sample_007_static cufile_sample_008_static cufile_sample_009_static cufile_sample_010_static cufile_sample_011_static cufile_sample_012_static cufile_sample_013_static cufile_sample_014_static cufile_sample_015_static cufile_sample_016_static cufile_sample_017_static cufile_sample_018_static
+	rm -f cufile_sample_001 cufile_sample_001.o cufile_sample_002 cufile_sample_002.o cufile_sample_003 cufile_sample_003.o cufile_sample_004 cufile_sample_004.o cufile_sample_005 cufile_sample_005.o cufile_sample_006 cufile_sample_006.o cufile_sample_007 cufile_sample_007.o cufile_sample_008 cufile_sample_008.o cufile_sample_009 cufile_sample_009.o cufile_sample_010 cufile_sample_010.o cufile_sample_011 cufile_sample_011.o cufile_sample_012 cufile_sample_012.o cufile_sample_013 cufile_sample_013.o cufile_sample_014 cufile_sample_014.o cufile_sample_015 cufile_sample_015.o cufile_sample_016.o cufile_sample_016 cufile_sample_017.o cufile_sample_017 cufile_sample_018.o cufile_sample_019.o cufile_sample_018 cufile_sample_001_static cufile_sample_002_static cufile_sample_003_static cufile_sample_004_static cufile_sample_005_static cufile_sample_006_static cufile_sample_007_static cufile_sample_008_static cufile_sample_009_static cufile_sample_010_static cufile_sample_011_static cufile_sample_012_static cufile_sample_013_static cufile_sample_014_static cufile_sample_015_static cufile_sample_016_static cufile_sample_017_static cufile_sample_018_static cufile_sample_019_static cufile_sample_019 cufile_sample_020_static cufile_sample_020 cufile_sample_021 cufile_sample_022
 
 .PHONY : build install clean

gds/samples/README.md

@@ -128,3 +128,24 @@ buffer offsets of a memory allocated using single allocation and single buffer r
 ``` bash
 ./cufile_sample_018 <file-path>
 ```
+
+**Note**: Following samples need cuFile library version 11.6 and above 
+**cufile_sample_019**: This sample shows the usage of cuFile Batch API for writes.
+``` bash
+./cufile_sample_019 <file-path> <gpuid> <num batch entries>
+```
+
+**cufile_sample_020**: This sample shows the usage of cuFile Batch API for reads.
+``` bash
+./cufile_sample_020 <file-path>  <gpuid> <num batch entries>
+```
+
+**cufile_sample_021**: This sample shows the usage of cuFile Batch API to cancel I/O after submitting a batch read.
+``` bash
+./cufile_sample_021 <file-path>  <gpuid> <num batch entries>
+```
+
+**cufile_sample_022**: This sample shows the usage of cuFile Batch API to perform cuFileBatchIOGetStatus after submitting a batch read.
+``` bash
+./cufile_sample_022 <file-path>  <gpuid> <num batch entries>
+```

gds/samples/cufile_sample_001.cc

@@ -87,6 +87,7 @@ int main(int argc, char *argv[]) {
 	check_cudaruntimecall(cudaMalloc(&devPtr, size));
 	// filler
 	check_cudaruntimecall(cudaMemset((void*)(devPtr), 0xab, size));
+	check_cudaruntimecall(cudaStreamSynchronize(0));
 
 	std::cout << "registering device memory of size :" << size << std::endl;
 	// registers device memory

gds/samples/cufile_sample_002.cc

@@ -83,7 +83,7 @@ int main(int argc, char *argv[]) {
 
 	// filler for device memory
 	check_cudaruntimecall(cudaMemset(devPtr, 0xab, size));
-
+	check_cudaruntimecall(cudaStreamSynchronize(0));
 	check_cudaruntimecall(cudaGetDevice(&idx));
 
 	std::cout << "writing from gpuid: " << idx << std::endl;

gds/samples/cufile_sample_003.cc

@@ -125,7 +125,7 @@ int main(int argc, char *argv[]) {
 	check_cudaruntimecall(cudaMalloc(&devPtr, size));
 	// special case for holes
 	check_cudaruntimecall(cudaMemset(devPtr, 0, size));
-
+	check_cudaruntimecall(cudaStreamSynchronize(0));
 	std::cout << "reading file to device memory :" << TEST_READWRITEFILE
 				<< std::endl;
 

gds/samples/cufile_sample_004.cc

@@ -140,6 +140,7 @@ int main(int argc, char *argv[]) {
 	check_cudadrivercall(cuCtxCreate(&cuCtx, 0, cudev));
 	check_cudadrivercall(cuMemAlloc(&cudevPtr, size));
 	check_cudadrivercall(cuMemsetD8(cudevPtr, 0x0, size));
+	check_cudadrivercall(cuStreamSynchronize(0));
 
 	std::cout << "reading file to device memory :"
 		<< TEST_READWRITEFILE << std::endl;

gds/samples/cufile_sample_006.cc

@@ -89,7 +89,7 @@ int main(int argc, char *argv[]) {
 
 	check_cudaruntimecall(cudaMalloc(&devPtr, size));
 	check_cudaruntimecall(cudaMemset(devPtr, 0x00, size));
-
+	check_cudaruntimecall(cudaStreamSynchronize(0));
 	std::cout << "reading file sequentially :" << TESTFILE
 			  << " chunk size : " << CHUNK_SIZE <<  std::endl;
 	do {

gds/samples/cufile_sample_014.cc

@@ -86,7 +86,7 @@ int main(int argc, char *argv[]) {
 	size = std::min(size, MAX_BUFFER_SIZE);
 	check_cudaruntimecall(cudaMalloc(&devPtr, size));
 	check_cudaruntimecall(cudaMemset((void*)(devPtr), 0x00, size));
-
+	check_cudaruntimecall(cudaStreamSynchronize(0));
 	std::cout << "registering device memory of size :" << size << std::endl;
 	status = cuFileBufRegister(devPtr, size, 0);
 	if (status.err != CU_FILE_SUCCESS) {

gds/samples/cufile_sample_015.cc

@@ -164,6 +164,7 @@ int main(int argc, char *argv[]) {
     assert((devPtr = alloc_memory(size, mem_type)) != nullptr);
     // special case for holes
     assert(cudaMemset(devPtr, 0, size) == cudaSuccess);
+    check_cudaruntimecall(cudaStreamSynchronize(0));
 
     std::cout << "reading file to device memory :" << TEST_READWRITEFILE
         << std::endl;

gds/samples/cufile_sample_018.cc

@@ -16,7 +16,6 @@
 #include <unistd.h>
 #include <string.h>
 #include <sys/types.h>
-#include <assert.h>
 
 // CUDA includes
 #include <cuda.h>
@@ -36,6 +35,7 @@
 
 #define ALIGN_UP(x, align_to)   (((x) + ((align_to)-1)) & ~((align_to)-1))
 #define ALIGN_DOWN(x, a)        ((unsigned long)(x) & ~(((unsigned long)(a)) - 1))
+#define MAX_RETRY 3
 
 //Macro for checking cuda errors following a cuda launch or api call
 #define cudaCheckError() {                                          \
@@ -59,6 +59,7 @@ static void *read_thread_fn(void *data)
 {
 	int ret;
 	thread_data_t *t = (thread_data_t *)data;
+	int cnt;
 
 	cudaSetDevice(0);
 	cudaCheckError();
@@ -72,18 +73,31 @@ static void *read_thread_fn(void *data)
 	fl.l_len = ALIGN_UP(t->size, PAGE_SIZE);
 
 	// Acquire lock at 4K boundary
-	if (fcntl(t->fd, F_SETLKW, &fl) == -1) {
-		printf("Failed to acquire read lock from offset %ld size %ld errno %d\n",
-				(unsigned long) fl.l_start, (unsigned long) fl.l_len, errno);
-		exit(1);
-	}
+        cnt = 0;
+        while (1) {
+                cnt++;
+                if (fcntl(t->fd, F_SETLKW, &fl) == -1) {
+                        printf("Failed to acquire read lock from offset %ld size %ld errno %d\n",
+                                (unsigned long) fl.l_start, (unsigned long) fl.l_len, errno);
+                        if (cnt == MAX_RETRY) {
+                                exit(1);
+                        } else {
+                                printf("Retrying fcntl for read..\n");
+                        }
+                } else {
+                        break;
+                }
+        }
 
         printf("Read lock acquired from offset %ld size %ld. Submit read at offset %ld size %ld\n",
                         (unsigned long) fl.l_start, (unsigned long) fl.l_len,
                         (unsigned long) t->offset, (unsigned long) t->size);
 
 	ret = cuFileRead(t->cfr_handle, t->devPtr, t->size, t->offset, 0);
-	assert(ret > 0);
+	if (ret < 0) {
+		perror("cuFileRead Failed");
+		exit(1);
+	}
 
 	fl.l_type = F_UNLCK;  /* set to unlock same region */
 	if (fcntl(t->fd, F_SETLKW, &fl) == -1) {
@@ -101,6 +115,7 @@ static void *write_thread_fn(void *data)
 {
 	int ret;
 	thread_data_t *t = (thread_data_t *)data;
+	int cnt;
 
 	/*
 	 * We need to set the CUDA device; threads will not inherit main thread's
@@ -120,19 +135,31 @@ static void *write_thread_fn(void *data)
 	// Acquire lock at 4K boundary
 	fl.l_start = ALIGN_DOWN(t->offset, PAGE_SIZE);
 	fl.l_len = ALIGN_UP(t->size, PAGE_SIZE);
+        cnt = 0;
+        while (1) {
+                cnt++;
+                if (fcntl(t->fd, F_SETLKW, &fl) == -1) {
+                        printf("Failed to acquire write lock from offset %ld size %ld errno %d\n",
+                                (unsigned long) fl.l_start, (unsigned long) fl.l_len, errno);
+                        if (cnt == MAX_RETRY) {
+                                exit(1);
+                        } else {
+                                printf("Retrying fcntl for write..\n");
+                        }
+                } else {
+                        break;
+                }
+        }
 
-	if (fcntl(t->fd, F_SETLKW, &fl) == -1) {
-		printf("Failed to acquire write lock from offset %ld size %ld errno %d\n",
-				(unsigned long) fl.l_start, (unsigned long) fl.l_len, errno);
-		exit(1);
-	}
-
 	printf("Write lock acquired from offset %ld size %ld. Submit write at offset %ld size %ld\n",
 			(unsigned long) fl.l_start, (unsigned long) fl.l_len,
 			(unsigned long) t->offset, (unsigned long) t->size);
 
 	ret = cuFileWrite(t->cfr_handle, t->devPtr, t->size, t->offset, 0);
-	assert(ret > 0);
+	if (ret < 0) {
+		perror("cuFileWrite Failed");
+		exit(1);
+	}
 
 	fl.l_type = F_UNLCK;  /* set to unlock same region */
 	if (fcntl(t->fd, F_SETLKW, &fl) == -1) {
@@ -194,6 +221,9 @@ int main(int argc, char **argv) {
 	cudaMemset(devPtr, 0xab, KB(4));
 	cudaCheckError();
 
+	cudaStreamSynchronize(0);
+	cudaCheckError();
+
 	// Thread 0 will write to file from offset 10 - write size 100 bytes
 	// This is an unaligned write as offset is not 4K aligned. GDS will
 	// convert this write to Read-Modify-Write