mirror of
https://github.com/HDFGroup/hdf5.git
synced 2024-11-27 02:10:55 +08:00
2072 lines
73 KiB
C
2072 lines
73 KiB
C
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
|
||
* Copyright by the Board of Trustees of the University of Illinois. *
|
||
* All rights reserved. *
|
||
* *
|
||
* This file is part of HDF5. The full HDF5 copyright notice, including *
|
||
* terms governing use, modification, and redistribution, is contained in *
|
||
* the COPYING file, which can be found at the root of the source code *
|
||
* distribution tree, or in https://support.hdfgroup.org/ftp/HDF5/releases. *
|
||
* If you do not have access to either file, you may request a copy from *
|
||
* help@hdfgroup.org. *
|
||
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
|
||
|
||
/* Programmer: Mike McGreevy
|
||
* October 7, 2010
|
||
*/
|
||
#include "h5test.h"
|
||
|
||
#define H5F_FRIEND /*suppress error about including H5Fpkg */
|
||
#define H5FD_FRIEND /*suppress error about including H5FDpkg */
|
||
#define H5FD_TESTING
|
||
#include "H5Fpkg.h"
|
||
#include "H5FDpkg.h"
|
||
|
||
#include "H5CXprivate.h" /* API Contexts */
|
||
#include "H5Iprivate.h"
|
||
#include "H5VLprivate.h" /* Virtual Object Layer */
|
||
|
||
/* Filename */
|
||
/* (The file names are the same as the define in accum_swmr_reader.c) */
|
||
const char *FILENAME[] = {
|
||
"accum",
|
||
"accum_swmr_big",
|
||
NULL
|
||
};
|
||
|
||
/* The reader forked by test_swmr_write_big() */
|
||
#define SWMR_READER "accum_swmr_reader"
|
||
|
||
/* "big" I/O test values */
|
||
#define BIG_BUF_SIZE (6 * 1024 * 1024)
|
||
|
||
/* Random I/O test values */
|
||
#define RANDOM_BUF_SIZE (1 * 1024 * 1024)
|
||
#define MAX_RANDOM_SEGMENTS (5 * 1024)
|
||
#define RAND_SEG_LEN (1024)
|
||
#define RANDOM_BASE_OFF (1024 * 1024)
|
||
|
||
/* Function Prototypes */
|
||
unsigned test_write_read(H5F_t *f);
|
||
unsigned test_write_read_nonacc_front(H5F_t *f);
|
||
unsigned test_write_read_nonacc_end(H5F_t *f);
|
||
unsigned test_accum_overlap(H5F_t *f);
|
||
unsigned test_accum_overlap_clean(H5F_t *f);
|
||
unsigned test_accum_overlap_size(H5F_t *f);
|
||
unsigned test_accum_non_overlap_size(H5F_t *f);
|
||
unsigned test_accum_adjust(H5F_t *f);
|
||
unsigned test_read_after(H5F_t *f);
|
||
unsigned test_free(H5F_t *f);
|
||
unsigned test_big(H5F_t *f);
|
||
unsigned test_random_write(H5F_t *f);
|
||
unsigned test_swmr_write_big(hbool_t newest_format);
|
||
|
||
/* Helper Function Prototypes */
|
||
void accum_printf(const H5F_t *f);
|
||
|
||
/* Private Test H5Faccum Function Wrappers */
|
||
#define accum_write(a,s,b) H5F_block_write(f, H5FD_MEM_DEFAULT, (haddr_t)(a), (size_t)(s), (b))
|
||
#define accum_read(a,s,b) H5F_block_read(f, H5FD_MEM_DEFAULT, (haddr_t)(a), (size_t)(s), (b))
|
||
#define accum_free(f,a,s) H5F__accum_free(f, H5FD_MEM_DEFAULT, (haddr_t)(a), (hsize_t)(s))
|
||
#define accum_flush(f) H5F__accum_flush(f)
|
||
#define accum_reset(f) H5F__accum_reset(f, TRUE)
|
||
|
||
/* ================= */
|
||
/* Main Test Routine */
|
||
/* ================= */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: main
|
||
*
|
||
* Purpose: Test the metadata accumulator code
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Mike McGreevy
|
||
* October 7, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
int
|
||
main(void)
|
||
{
|
||
unsigned nerrors = 0; /* track errors */
|
||
hbool_t api_ctx_pushed = FALSE; /* Whether API context pushed */
|
||
hid_t fid = -1;
|
||
hid_t fapl = -1; /* File access property list */
|
||
char filename[1024];
|
||
H5F_t * f = NULL; /* File for all tests */
|
||
|
||
|
||
/* Test Setup */
|
||
HDputs("Testing the metadata accumulator");
|
||
|
||
/* File access property list */
|
||
h5_reset();
|
||
if((fapl = h5_fileaccess()) < 0)
|
||
FAIL_STACK_ERROR
|
||
h5_fixname(FILENAME[0], fapl, filename, sizeof filename);
|
||
|
||
/* Create a test file */
|
||
if((fid = H5Fcreate(filename, H5F_ACC_TRUNC, H5P_DEFAULT, fapl)) < 0) FAIL_STACK_ERROR
|
||
|
||
/* Push API context */
|
||
if(H5CX_push() < 0) FAIL_STACK_ERROR
|
||
api_ctx_pushed = TRUE;
|
||
|
||
/* Get H5F_t * to internal file structure */
|
||
if(NULL == (f = (H5F_t *)H5VL_object(fid))) FAIL_STACK_ERROR
|
||
|
||
/* We'll be writing lots of garbage data, so extend the
|
||
file a ways. 10MB should do. */
|
||
if(H5FD_set_eoa(f->shared->lf, H5FD_MEM_DEFAULT, (haddr_t)(1024*1024*10)) < 0) FAIL_STACK_ERROR
|
||
|
||
/* Reset metadata accumulator for the file */
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR
|
||
|
||
/* Test Functions */
|
||
nerrors += test_write_read(f);
|
||
nerrors += test_write_read_nonacc_front(f);
|
||
nerrors += test_write_read_nonacc_end(f);
|
||
nerrors += test_accum_overlap(f);
|
||
nerrors += test_accum_overlap_clean(f);
|
||
nerrors += test_accum_overlap_size(f);
|
||
nerrors += test_accum_non_overlap_size(f);
|
||
nerrors += test_accum_adjust(f);
|
||
nerrors += test_read_after(f);
|
||
nerrors += test_free(f);
|
||
nerrors += test_big(f);
|
||
nerrors += test_random_write(f);
|
||
|
||
/* Pop API context */
|
||
if(api_ctx_pushed && H5CX_pop() < 0) FAIL_STACK_ERROR
|
||
api_ctx_pushed = FALSE;
|
||
|
||
/* End of test code, close and delete file */
|
||
if(H5Fclose(fid) < 0) TEST_ERROR
|
||
|
||
/* This test uses a different file */
|
||
nerrors += test_swmr_write_big(TRUE);
|
||
nerrors += test_swmr_write_big(FALSE);
|
||
|
||
if(nerrors)
|
||
goto error;
|
||
HDputs("All metadata accumulator tests passed.");
|
||
h5_cleanup(FILENAME, fapl);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
if(api_ctx_pushed) H5CX_pop();
|
||
|
||
HDputs("*** TESTS FAILED ***");
|
||
return 1;
|
||
} /* end main() */
|
||
|
||
/* ============================= */
|
||
/* Individual Unit Test Routines */
|
||
/* ============================= */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_write_read
|
||
*
|
||
* Purpose: Simple test to write to then read from metadata accumulator.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Mike McGreevy
|
||
* October 7, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_write_read(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int *write_buf, *read_buf;
|
||
|
||
TESTING("simple write/read to/from metadata accumulator");
|
||
|
||
/* Allocate buffers */
|
||
write_buf = (int *)HDmalloc(1024 * sizeof(int));
|
||
HDassert(write_buf);
|
||
read_buf = (int *)HDcalloc((size_t)1024, sizeof(int));
|
||
HDassert(read_buf);
|
||
|
||
/* Fill buffer with data, zero out read buffer */
|
||
for(i = 0; i < 1024; i++)
|
||
write_buf[i] = i + 1;
|
||
|
||
/* Do a simple write/read/verify of data */
|
||
/* Write 1KB at Address 0 */
|
||
if(accum_write(0, 1024, write_buf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(0, 1024, read_buf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(write_buf, read_buf, (size_t)1024) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(write_buf);
|
||
HDfree(read_buf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(write_buf);
|
||
HDfree(read_buf);
|
||
|
||
return 1;
|
||
} /* test_write_read */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_write_read_nonacc_front
|
||
*
|
||
* Purpose: Simple test to write to then read from before metadata accumulator.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Allen Byrne
|
||
* October 8, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_write_read_nonacc_front(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int *write_buf, *read_buf;
|
||
|
||
TESTING("simple write/read to/from before metadata accumulator");
|
||
|
||
/* Allocate buffers */
|
||
write_buf = (int *)HDmalloc(2048 * sizeof(int));
|
||
HDassert(write_buf);
|
||
read_buf = (int *)HDcalloc((size_t)2048, sizeof(int));
|
||
HDassert(read_buf);
|
||
|
||
/* Fill buffer with data, zero out read buffer */
|
||
for(i = 0; i < 2048; i++)
|
||
write_buf[i] = i + 1;
|
||
|
||
/* Do a simple write/read/verify of data */
|
||
/* Write 1KB at Address 0 */
|
||
if(accum_write(0, 1024, write_buf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(1024, 1024, write_buf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(0, 1024, read_buf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(write_buf, read_buf, (size_t)1024) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(write_buf);
|
||
HDfree(read_buf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(write_buf);
|
||
HDfree(read_buf);
|
||
|
||
return 1;
|
||
} /* test_write_read */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_write_read_nonacc_end
|
||
*
|
||
* Purpose: Simple test to write to then read from after metadata accumulator.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Allen Byrne
|
||
* October 8, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_write_read_nonacc_end(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int *write_buf, *read_buf;
|
||
|
||
TESTING("simple write/read to/from after metadata accumulator");
|
||
|
||
/* Allocate buffers */
|
||
write_buf = (int *)HDmalloc(2048 * sizeof(int));
|
||
HDassert(write_buf);
|
||
read_buf = (int *)HDcalloc((size_t)2048, sizeof(int));
|
||
HDassert(read_buf);
|
||
|
||
/* Fill buffer with data, zero out read buffer */
|
||
for(i = 0; i < 2048; i++)
|
||
write_buf[i] = i + 1;
|
||
|
||
/* Do a simple write/read/verify of data */
|
||
/* Write 1KB at Address 0 */
|
||
if(accum_write(1024, 1024, write_buf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(0, 1024, write_buf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(1024, 1024, read_buf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(write_buf, read_buf, (size_t)1024) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(write_buf);
|
||
HDfree(read_buf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(write_buf);
|
||
HDfree(read_buf);
|
||
|
||
return 1;
|
||
} /* test_write_read */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_free
|
||
*
|
||
* Purpose: Simple test to free metadata accumulator.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Raymond Lu
|
||
* October 8, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_free(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int32_t *wbuf = NULL;
|
||
int32_t *rbuf = NULL;
|
||
int32_t *expect = NULL;
|
||
|
||
TESTING("simple freeing metadata accumulator");
|
||
|
||
/* Write and free the whole accumulator. */
|
||
wbuf = (int32_t *)HDmalloc(256 * sizeof(int32_t));
|
||
HDassert(wbuf);
|
||
rbuf = (int32_t *)HDmalloc(256 * sizeof(int32_t));
|
||
HDassert(rbuf);
|
||
expect = (int32_t *)HDmalloc(256 * sizeof(int32_t));
|
||
HDassert(expect);
|
||
|
||
/* Fill buffer with data */
|
||
for(i = 0; i < 256; i++)
|
||
wbuf[i] = (int32_t)(i + 1);
|
||
|
||
if(accum_write(0, 256 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
if(accum_free(f, 0, 256 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Free an empty accumulator */
|
||
if(accum_free(f, 0, 256 * 1024 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write second quarter of the accumulator */
|
||
if(accum_write(64 * sizeof(int32_t), 64 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Free the second quarter of the accumulator, the requested area
|
||
* is bigger than the data region on the right side. */
|
||
if(accum_free(f, 64 * sizeof(int32_t), 65 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write half of the accumulator. */
|
||
if(accum_write(0, 128 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Free the first block of 4B */
|
||
if(accum_free(f, 0, sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(1 * sizeof(int32_t), 127 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf + 1, rbuf, 127 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Free the block of 4B at 127*4B */
|
||
if(accum_free(f, 127 * sizeof(int32_t), sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(1 * sizeof(int32_t), 126 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf + 1, rbuf, 126 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Free the block of 4B at 2*4B */
|
||
if(accum_free(f, 2 * sizeof(int32_t), sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(1 * sizeof(int32_t), 1 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf + 1, rbuf, 1 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
if(accum_read(3 * sizeof(int32_t), 124 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf + 3, rbuf, 124 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Test freeing section that overlaps the start of the accumulator and is
|
||
* entirely before dirty section */
|
||
if(accum_write(64 * sizeof(int32_t), 128 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 64, wbuf, 128 * sizeof(int32_t));
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(68 * sizeof(int32_t), 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 68, wbuf, 4 * sizeof(int32_t));
|
||
if(accum_free(f, 62 * sizeof(int32_t), 4 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(66 * sizeof(int32_t), 126 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(expect + 66, rbuf, 126 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Test freeing section that overlaps the start of the accumulator and
|
||
* completely contains dirty section */
|
||
if(accum_write(64 * sizeof(int32_t), 128 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 64, wbuf, 128 * sizeof(int32_t));
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(68 * sizeof(int32_t), 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 68, wbuf, 4 * sizeof(int32_t));
|
||
if(accum_free(f, 62 * sizeof(int32_t), 16 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(78 * sizeof(int32_t), 114 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(expect + 78, rbuf, 114 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Test freeing section completely contained in accumulator and is entirely
|
||
* before dirty section */
|
||
if(accum_write(64 * sizeof(int32_t), 128 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 64, wbuf, 128 * sizeof(int32_t));
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(72 * sizeof(int32_t), 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 72, wbuf, 4 * sizeof(int32_t));
|
||
if(accum_free(f, 66 * sizeof(int32_t), 4 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(70 * sizeof(int32_t), 122 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(expect + 70, rbuf, 122 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Test freeing section completely contained in accumulator, starts before
|
||
* dirty section, and ends in dirty section */
|
||
if(accum_write(64 * sizeof(int32_t), 128 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 64, wbuf, 128 * sizeof(int32_t));
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(72 * sizeof(int32_t), 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 72, wbuf, 4 * sizeof(int32_t));
|
||
if(accum_free(f, 70 * sizeof(int32_t), 4 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(74 * sizeof(int32_t), 118 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(expect + 74, rbuf, 118 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Test freeing section completely contained in accumulator and completely
|
||
* contains dirty section */
|
||
if(accum_write(64 * sizeof(int32_t), 128 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 64, wbuf, 128 * sizeof(int32_t));
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(72 * sizeof(int32_t), 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 72, wbuf, 4 * sizeof(int32_t));
|
||
if(accum_free(f, 70 * sizeof(int32_t), 8 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(78 * sizeof(int32_t), 114 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(expect + 78, rbuf, 114 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Test freeing section completely contained in accumulator, starts at start
|
||
* of dirty section, and ends in dirty section */
|
||
if(accum_write(64 * sizeof(int32_t), 128 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 64, wbuf, 128 * sizeof(int32_t));
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
if(accum_write(72 * sizeof(int32_t), 8 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemcpy(expect + 72, wbuf, 8 * sizeof(int32_t));
|
||
if(accum_free(f, 72 * sizeof(int32_t), 4 * sizeof(int32_t)) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Check that the accumulator still contains the correct data */
|
||
if(accum_read(76 * sizeof(int32_t), 116 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(expect + 76, rbuf, 116 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
HDfree(wbuf);
|
||
wbuf = NULL;
|
||
HDfree(rbuf);
|
||
rbuf = NULL;
|
||
HDfree(expect);
|
||
expect = NULL;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
return 0;
|
||
|
||
error:
|
||
if(wbuf)
|
||
HDfree(wbuf);
|
||
if(rbuf)
|
||
HDfree(rbuf);
|
||
if(expect)
|
||
HDfree(expect);
|
||
|
||
return 1;
|
||
} /* test_free */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_accum_overlap
|
||
*
|
||
* Purpose: This test will write a series of pieces of data
|
||
* to the accumulator with the goal of overlapping
|
||
* the writes in various different ways.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Mike McGreevy
|
||
* October 7, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_accum_overlap(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int32_t *wbuf, *rbuf;
|
||
|
||
TESTING("overlapping write to metadata accumulator");
|
||
|
||
/* Allocate buffers */
|
||
wbuf = (int32_t *)HDmalloc(4096 * sizeof(int32_t));
|
||
HDassert(wbuf);
|
||
rbuf = (int32_t *)HDcalloc((size_t)4096, sizeof(int32_t));
|
||
HDassert(rbuf);
|
||
|
||
/* Case 1: No metadata in accumulator */
|
||
/* Write 10 1's at address 40 */
|
||
/* @0:| 1111111111| */
|
||
/* Put some data in the accumulator initially */
|
||
for(i = 0; i < 10; i++)
|
||
wbuf[i] = 1;
|
||
if(accum_write(40, 10 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(40, 10 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 10 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 2: End of new piece aligns with start of accumulated data */
|
||
/* Write 5 2's at address 20 */
|
||
/* @0:| 222221111111111| */
|
||
for(i = 0; i < 5; i++)
|
||
wbuf[i] = 2;
|
||
if(accum_write(20, 5 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(20, 5 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 5 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 3: Start of new piece aligns with start of accumulated data */
|
||
/* Write 3 3's at address 20 */
|
||
/* @0:| 333221111111111| */
|
||
for(i = 0; i < 3; i++)
|
||
wbuf[i] = 3;
|
||
if(accum_write(20, 3 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(20, 3 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 3 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 4: New piece overlaps start of accumulated data */
|
||
/* Write 5 4's at address 8 */
|
||
/* @0:| 444443221111111111| */
|
||
for(i = 0; i < 5; i++)
|
||
wbuf[i] = 4;
|
||
if(accum_write(8, 5 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(8, 5 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 5 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 5: New piece completely within accumulated data */
|
||
/* Write 4 5's at address 48 */
|
||
/* @0:| 444443221155551111| */
|
||
for(i = 0; i < 4; i++)
|
||
wbuf[i] = 5;
|
||
if(accum_write(48, 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(48, 4 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 4 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 6: End of new piece aligns with end of accumulated data */
|
||
/* Write 3 6's at address 68 */
|
||
/* @0:| 444443221155551666| */
|
||
for(i = 0; i < 3; i++)
|
||
wbuf[i] = 6;
|
||
if(accum_write(68, 3 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(68, 3 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 3 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 7: New piece overlaps end of accumulated data */
|
||
/* Write 5 7's at address 76 */
|
||
/* @0:| 4444432211555516677777| */
|
||
for(i = 0; i < 5; i++)
|
||
wbuf[i] = 7;
|
||
if(accum_write(76, 5 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(76, 5 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 5 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 8: Start of new piece aligns with end of accumulated data */
|
||
/* Write 3 8's at address 96 */
|
||
/* @0:| 4444432211555516677777888| */
|
||
for(i = 0; i < 3; i++)
|
||
wbuf[i] = 8;
|
||
if(accum_write(96, 3 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(96, 3 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 3 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Set up expected data buffer and verify contents of
|
||
accumulator as constructed by cases 1-8, above */
|
||
for(i = 0; i < 5; i++)
|
||
wbuf[i] = 4;
|
||
for(i = 5; i < 6; i++)
|
||
wbuf[i] = 3;
|
||
for(i = 6; i < 8; i++)
|
||
wbuf[i] = 2;
|
||
for(i = 8; i < 10; i++)
|
||
wbuf[i] = 1;
|
||
for(i = 10; i < 14; i++)
|
||
wbuf[i] = 5;
|
||
for(i = 14; i < 15; i++)
|
||
wbuf[i] = 1;
|
||
for(i = 15; i < 17; i++)
|
||
wbuf[i] = 6;
|
||
for(i = 17; i < 22; i++)
|
||
wbuf[i] = 7;
|
||
for(i = 22; i < 25; i++)
|
||
wbuf[i] = 8;
|
||
if(accum_read(8, 25 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 25 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 9: New piece completely before accumulated data */
|
||
/* Write 1 9 at address 0 */
|
||
/* @0:|9 4444432211555516677777888| */
|
||
for(i = 0; i < 1; i++)
|
||
wbuf[i] = 9;
|
||
if(accum_write(0, 1 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(0, 1 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 1 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 10: New piece completely after accumulated data */
|
||
/* Write 4 3's at address 116 */
|
||
/* @0:|9 4444432211555516677777888 3333| */
|
||
for(i = 0; i < 4; i++)
|
||
wbuf[i] = 3;
|
||
if(accum_write(116, 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(116, 4 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 4 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 11: New piece completely overlaps accumulated data */
|
||
/* Write 6 4's at address 112 */
|
||
/* @0:|9 4444432211555516677777888 444444| */
|
||
for(i = 0; i < 6; i++)
|
||
wbuf[i] = 4;
|
||
if(accum_write(112, 6 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(112, 6 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 6 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 1;
|
||
} /* test_accum_overlap */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_accum_overlap_clean
|
||
*
|
||
* Purpose: This test will write a series of pieces of data
|
||
* to the accumulator with the goal of overlapping
|
||
* the writes in various different ways, with clean
|
||
* areas in the accumulator.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Neil Fortner
|
||
* October 8, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_accum_overlap_clean(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int32_t *wbuf, *rbuf;
|
||
|
||
TESTING("overlapping write to partially clean metadata accumulator");
|
||
|
||
/* Allocate buffers */
|
||
wbuf = (int32_t *)HDmalloc(4096 * sizeof(int32_t));
|
||
HDassert(wbuf);
|
||
rbuf = (int32_t *)HDcalloc((size_t)4096, sizeof(int32_t));
|
||
HDassert(rbuf);
|
||
|
||
/* Case 1: No metadata in accumulator */
|
||
/* Write 10 1's at address 40 */
|
||
/* @0:| 1111111111| */
|
||
/* Put some data in the accumulator initially */
|
||
for(i = 0; i < 10; i++)
|
||
wbuf[i] = 1;
|
||
if(accum_write(40, 10 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(40, 10 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 10 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 2: End of new piece aligns with start of clean accumulated data */
|
||
/* Write 5 2's at address 20 */
|
||
/* @0:| 222221111111111| */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
for(i = 0; i < 5; i++)
|
||
wbuf[i] = 2;
|
||
if(accum_write(20, 5 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(20, 5 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 5 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 3: Start of new piece aligns with start of accumulated data,
|
||
* completely encloses dirty section of accumulator */
|
||
/* Write 6 3's at address 20 */
|
||
/* @0:| 333333111111111| */
|
||
for(i = 0; i < 6; i++)
|
||
wbuf[i] = 3;
|
||
if(accum_write(20, 6 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(20, 6 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 6 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 4: New piece completely within accumulated data, overlaps
|
||
* end of dirty section of accumulator */
|
||
/* Write 2 4's at address 40 */
|
||
/* @0:| 333334411111111| */
|
||
for(i = 0; i < 2; i++)
|
||
wbuf[i] = 4;
|
||
if(accum_write(40, 2 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(40, 2 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 2 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 5: New piece completely within accumulated data, completely
|
||
* after dirty section of accumulator */
|
||
/* Write 2 5's at address 52 */
|
||
/* @0:| 333334415511111| */
|
||
for(i = 0; i < 2; i++)
|
||
wbuf[i] = 5;
|
||
if(accum_write(52, 2 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(52, 2 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 2 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 6: New piece completely within clean accumulated data */
|
||
/* Write 3 6's at address 44 */
|
||
/* @0:| 333334666511111| */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
for(i = 0; i < 3; i++)
|
||
wbuf[i] = 6;
|
||
if(accum_write(44, 3 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(44, 3 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 3 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 7: New piece overlaps start of clean accumulated data */
|
||
/* Write 2 7's at address 16 */
|
||
/* @0:| 7733334666511111| */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
for(i = 0; i < 2; i++)
|
||
wbuf[i] = 7;
|
||
if(accum_write(16, 2 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(16, 2 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 2 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 8: New piece overlaps start of accumulated data, completely
|
||
* encloses dirty section of accumulator */
|
||
/* Write 4 8's at address 12 */
|
||
/* @0:| 88883334666511111| */
|
||
for(i = 0; i < 4; i++)
|
||
wbuf[i] = 8;
|
||
if(accum_write(12, 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(12, 4 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 4 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 9: Start of new piece aligns with end of clean accumulated data */
|
||
/* Write 3 9's at address 80 */
|
||
/* @0:| 88883334666511111999| */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
for(i = 0; i < 3; i++)
|
||
wbuf[i] = 9;
|
||
if(accum_write(80, 3 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(80, 3 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 3 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 10: New piece overlaps end of clean accumulated data */
|
||
/* Write 3 2's at address 88 */
|
||
/* @0:| 888833346665111119922| */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR;
|
||
for(i = 0; i < 2; i++)
|
||
wbuf[i] = 2;
|
||
if(accum_write(88, 2 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(88, 2 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 2 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 11: New piece overlaps end of accumulated data, completely encloses
|
||
* dirty section of accumulator */
|
||
/* Write 4 7's at address 84 */
|
||
/* @0:| 8888333466651111197777| */
|
||
for(i = 0; i < 4; i++)
|
||
wbuf[i] = 7;
|
||
if(accum_write(84, 4 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(84, 4 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 4 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Set up expected data buffer and verify contents of
|
||
accumulator as constructed by cases 1-11, above */
|
||
for(i = 0; i < 4; i++)
|
||
wbuf[i] = 8;
|
||
for(i = 4; i < 7; i++)
|
||
wbuf[i] = 3;
|
||
for(i = 7; i < 8; i++)
|
||
wbuf[i] = 4;
|
||
for(i = 8; i < 11; i++)
|
||
wbuf[i] = 6;
|
||
for(i = 11; i < 12; i++)
|
||
wbuf[i] = 5;
|
||
for(i = 12; i < 17; i++)
|
||
wbuf[i] = 1;
|
||
for(i = 17; i < 18; i++)
|
||
wbuf[i] = 9;
|
||
for(i = 18; i < 22; i++)
|
||
wbuf[i] = 7;
|
||
if(accum_read(12, 22 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 22 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 1;
|
||
} /* test_accum_overlap_clean */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_accum_non_overlap_size
|
||
*
|
||
* Purpose: This test will write a series of pieces of data
|
||
* to the accumulator with the goal of not overlapping
|
||
* the writes with a data size larger then the accum size.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Allen Byrne
|
||
* October 8, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_accum_non_overlap_size(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int32_t *wbuf, *rbuf;
|
||
|
||
TESTING("non-overlapping write to accumulator larger then accum_size");
|
||
|
||
/* Allocate buffers */
|
||
wbuf = (int *)HDmalloc(4096 * sizeof(int32_t));
|
||
HDassert(wbuf);
|
||
rbuf = (int *)HDcalloc((size_t)4096, sizeof(int32_t));
|
||
HDassert(rbuf);
|
||
|
||
/* Case 1: No metadata in accumulator */
|
||
/* Write 10 1's at address 140 */
|
||
/* @0:| 1111111111| */
|
||
/* Put some data in the accumulator initially */
|
||
for(i = 0; i < 10; i++)
|
||
wbuf[i] = 1;
|
||
if(accum_write(140, 10 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(140, 10 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 10 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 9: New piece completely before accumulated data */
|
||
/* Write 20 9 at address 0 */
|
||
/* @0:|9 1111111111| */
|
||
for(i = 0; i < 20; i++)
|
||
wbuf[i] = 9;
|
||
if(accum_write(0, 20 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(0, 20 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 20 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 1;
|
||
} /* test_accum_non_overlap_size */
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_accum_overlap_size
|
||
*
|
||
* Purpose: This test will write a series of pieces of data
|
||
* to the accumulator with the goal of overlapping
|
||
* the writes with a data size completely overlapping
|
||
* the accumulator at both ends.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Allen Byrne
|
||
* October 8, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_accum_overlap_size(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int32_t *wbuf, *rbuf;
|
||
|
||
TESTING("overlapping write to accumulator larger then accum_size");
|
||
|
||
/* Allocate buffers */
|
||
wbuf = (int32_t *)HDmalloc(4096 * sizeof(int32_t));
|
||
HDassert(wbuf);
|
||
rbuf = (int32_t *)HDcalloc((size_t)4096, sizeof(int32_t));
|
||
HDassert(rbuf);
|
||
|
||
/* Case 1: No metadata in accumulator */
|
||
/* Write 10 1's at address 64 */
|
||
/* @0:| 1111111111| */
|
||
/* Put some data in the accumulator initially */
|
||
for(i = 0; i < 10; i++)
|
||
wbuf[i] = 1;
|
||
if(accum_write(64, 10 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(64, 10 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 10 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
/* Case 9: New piece completely before accumulated data */
|
||
/* Write 72 9 at address 60 */
|
||
/* @0:|9 1111111111| */
|
||
for(i = 0; i < 72; i++)
|
||
wbuf[i] = 9;
|
||
if(accum_write(60, 72 * sizeof(int32_t), wbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(60, 72 * sizeof(int32_t), rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, 72 * sizeof(int32_t)) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 1;
|
||
} /* test_accum_overlap_size */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_accum_adjust
|
||
*
|
||
* Purpose: This test examines the various ways the accumulator might
|
||
* adjust itself as a result of data appending or prepending
|
||
* to it.
|
||
*
|
||
* This test program covers all the code in H5F_accum_adjust,
|
||
* but NOT all possible paths through said code. It only covers
|
||
* six potential paths through the function. (Again, though, each
|
||
* piece of code within an if/else statement in H5F_accum_adjust is
|
||
* covered by one of the paths in this test function). Since there
|
||
* are a ridiculous number of total possible paths through this
|
||
* function due to its large number of embedded if/else statements,
|
||
* that's certainly a lot of different test cases to write by hand.
|
||
* (Though if someone comes across this code and has some free
|
||
* time, go for it).
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Mike McGreevy
|
||
* October 11, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_accum_adjust(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int s = 1048576; /* size of buffer */
|
||
int32_t *wbuf, *rbuf;
|
||
|
||
TESTING("accumulator adjustments after append/prepend of data");
|
||
|
||
/* Allocate buffers */
|
||
wbuf = (int32_t *)HDmalloc((size_t)s * sizeof(int32_t));
|
||
HDassert(wbuf);
|
||
rbuf = (int32_t *)HDcalloc((size_t)s, sizeof(int32_t));
|
||
HDassert(rbuf);
|
||
|
||
/* Fill up write buffer */
|
||
for(i = 0; i < s; i++)
|
||
wbuf[i] = i + 1;
|
||
|
||
/* ================================================================ */
|
||
/* CASE 1: Prepending small block to large, fully dirty accumulator */
|
||
/* ================================================================ */
|
||
|
||
/* Write data to the accumulator to fill it just under 1MB (max size),
|
||
* but not quite full. This will force the accumulator to, on subsequent
|
||
* writes, a) have to adjust since it's nearly full, and b) prevent
|
||
* an increase in size because it's already at it's maximum size */
|
||
if(accum_write((1024 * 1024), (1024 * 1024) - 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a small (1KB) block that prepends to the front of the accumulator. */
|
||
/* ==> Accumulator will need more buffer space */
|
||
/* ==> Accumulator will try to resize, but see that it's getting too big */
|
||
/* ==> Size of new block is less than half maximum size of accumulator */
|
||
/* ==> New block is being prepended to accumulator */
|
||
/* ==> Accumulator is dirty, it will be flushed. */
|
||
/* ==> Dirty region overlaps region to eliminate from accumulator */
|
||
if(accum_write((1024 * 1024) - 1024, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read back and verify first write */
|
||
if(accum_read((1024 * 1024), (1024 * 1024) - 1, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)((1024 * 1024) - 1)) != 0) TEST_ERROR;
|
||
|
||
/* Read back and verify second write */
|
||
if(accum_read((1024 * 1024) - 1024, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)1024) != 0) TEST_ERROR;
|
||
|
||
/* Reset accumulator for next case */
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* ================================================================ */
|
||
/* Case 2: Prepending large block to large, fully dirty accumulator */
|
||
/* ================================================================ */
|
||
|
||
/* Write data to the accumulator to fill it just under 1MB (max size),
|
||
* but not quite full. This will force the accumulator to, on subsequent
|
||
* writes, a) have to adjust since it's nearly full, and b) prevent
|
||
* an increase in size because it's already at it's maximum size */
|
||
if(accum_write((1024 * 1024), (1024 * 1024) - 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a large (just under 1MB) block to the front of the accumulator. */
|
||
/* ==> Accumulator will need more buffer space */
|
||
/* ==> Accumulator will try to resize, but see that it's getting too big */
|
||
/* ==> Size of new block is larger than half maximum size of accumulator */
|
||
/* ==> New block is being prepended to accumulator */
|
||
/* ==> Accumulator is dirty, it will be flushed. */
|
||
/* ==> Dirty region overlaps region to eliminate from accumulator */
|
||
if(accum_write(5, (1024 * 1024) - 5, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read back and verify both pieces of data */
|
||
if(accum_read(1048576, 1048575, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)1048576) != 0) TEST_ERROR;
|
||
|
||
if(accum_read(5, 1048571, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)1048571) != 0) TEST_ERROR;
|
||
|
||
/* Reset accumulator for next case */
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* ========================================================= */
|
||
/* Case 3: Appending small block to large, clean accumulator */
|
||
/* ========================================================= */
|
||
|
||
/* Write data to the accumulator to fill it just under 1MB (max size),
|
||
* but not quite full. This will force the accumulator to, on subsequent
|
||
* writes, a) have to adjust since it's nearly full, and b) prevent
|
||
* an increase in size because it's already at it's maximum size */
|
||
if(accum_write(0, (1024 * 1024) - 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Flush the accumulator -- we want to test the case when
|
||
accumulator contains clean data */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR
|
||
|
||
/* Write a small (1KB) block to the end of the accumulator */
|
||
/* ==> Accumulator will need more buffer space */
|
||
/* ==> Accumulator will try to resize, but see that it's getting too big */
|
||
/* ==> Size of new block is larger than half maximum size of accumulator */
|
||
/* ==> New block being appended to accumulator */
|
||
/* ==> Accumulator is NOT dirty */
|
||
/* ==> Since we're appending, need to adjust location of accumulator */
|
||
if(accum_write((1024 * 1024) - 1, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a piece of metadata outside current accumulator to force write
|
||
to disk */
|
||
if(accum_write(0, 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read in the piece we wrote to disk above, and then verify that
|
||
the data is as expected */
|
||
if(accum_read((1024 * 1024) - 1, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)1024) != 0) TEST_ERROR;
|
||
|
||
/* Reset accumulator for next case */
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* ==================================================================== */
|
||
/* Case 4: Appending small block to large, partially dirty accumulator, */
|
||
/* with existing dirty region NOT aligning with the new block */
|
||
/* ==================================================================== */
|
||
|
||
/* Write data to the accumulator to fill it just under 1MB (max size),
|
||
* but not quite full. This will force the accumulator to, on subsequent
|
||
* writes, a) have to adjust since it's nearly full, and b) prevent
|
||
* an increase in size because it's already at it's maximum size */
|
||
if(accum_write(0, (1024 * 1024) - 5, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Flush the accumulator to clean it */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR
|
||
|
||
/* write to part of the accumulator so just the start of it is dirty */
|
||
if(accum_write(0, 5, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a small (~340KB) piece of data to the other end of the accumulator */
|
||
/* ==> Accumulator will need more buffer space */
|
||
/* ==> Accumulator will try to resize, but see that it's getting too big */
|
||
/* ==> Size of new block is less than than half maximum size of accumulator */
|
||
/* ==> New block being appended to accumulator */
|
||
/* ==> We can slide the dirty region down, to accomodate the request */
|
||
/* ==> Max Buffer Size - (dirty offset + adjust size) >= 2 * size) */
|
||
/* ==> Need to adjust location of accumulator while appending */
|
||
/* ==> Accumulator will need to be reallocated */
|
||
if(accum_write(1048571, 349523, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a piece of metadata outside current accumulator to force write
|
||
to disk */
|
||
if(accum_write(1398900, 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read in the piece we wrote to disk above, and then verify that
|
||
the data is as expected */
|
||
if(accum_read(1048571, 349523, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)349523) != 0) TEST_ERROR;
|
||
|
||
/* Reset accumulator for next case */
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* ==================================================================== */
|
||
/* Case 5: Appending small block to large, partially dirty accumulator, */
|
||
/* with existing dirty region aligning with new block */
|
||
/* ==================================================================== */
|
||
|
||
/* Write data to the accumulator to fill it just under max size (but not full) */
|
||
if(accum_write(0, (1024 * 1024) - 5, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Flush the accumulator to clean it */
|
||
if(accum_flush(f) < 0) FAIL_STACK_ERROR
|
||
|
||
/* write to part of the accumulator so it's dirty, but not entirely dirty */
|
||
/* (just the begging few bytes will be clean) */
|
||
if(accum_write(10, (1024 * 1024) - 15, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a small piece of data to the dirty end of the accumulator */
|
||
/* ==> Accumulator will need more buffer space */
|
||
/* ==> Accumulator will try to resize, but see that it's getting too big */
|
||
/* ==> Size of new block is less than than half maximum size of accumulator */
|
||
/* ==> New block being appended to accumulator */
|
||
/* ==> We can slide the dirty region down, to accomodate the request */
|
||
/* ==> Max Buffer Size - (dirty offset + adjust size) < 2 * size) */
|
||
/* ==> Need to adjust location of accumulator while appending */
|
||
if(accum_write((1024 * 1024) - 5, 10, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a piece of metadata outside current accumulator to force write
|
||
to disk */
|
||
if(accum_write(0, 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read in the piece we wrote to disk above, and then verify that
|
||
the data is as expected */
|
||
if(accum_read((1024 * 1024) - 5, 10, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)10) != 0) TEST_ERROR;
|
||
|
||
/* Reset accumulator for next case */
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* ================================================================= */
|
||
/* Case 6: Appending small block to large, fully dirty accumulator */
|
||
/* ================================================================= */
|
||
|
||
/* Write data to the accumulator to fill it just under 1MB (max size),
|
||
* but not quite full. This will force the accumulator to, on subsequent
|
||
* writes, a) have to adjust since it's nearly full, and b) prevent
|
||
* an increase in size because it's already at it's maximum size */
|
||
if(accum_write(0, (1024 * 1024) - 5, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a small (~340KB) piece of data to the end of the accumulator */
|
||
/* ==> Accumulator will need more buffer space */
|
||
/* ==> Accumulator will try to resize, but see that it's getting too big */
|
||
/* ==> Size of new block is less than than half maximum size of accumulator */
|
||
/* ==> New block being appended to accumulator */
|
||
/* ==> We cannot slide dirty region down, it's all dirty */
|
||
/* ==> Dirty region overlaps region to eliminate from accumulator */
|
||
/* ==> Need to adjust location of accumulator while appending */
|
||
if(accum_write(1048571, 349523, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a piece of metadata outside current accumulator to force write
|
||
to disk */
|
||
if(accum_write(1398900, 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read in the piece we wrote to disk above, and then verify that
|
||
the data is as expected */
|
||
if(accum_read(1048571, 349523, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)349523) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 1;
|
||
} /* test_accum_adjust */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_read_after
|
||
*
|
||
* Purpose: This test will verify the case when metadata is read partly
|
||
* from the accumulator and partly from disk. The test will
|
||
* write a block of data at address 512, force the data to be
|
||
* written to disk, write new data partially overlapping the
|
||
* original block from below, then read data at address 512.
|
||
* The data read should be partly new and partly original.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Larry Knox
|
||
* October 8, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_read_after(H5F_t *f)
|
||
{
|
||
int i = 0;
|
||
int s = 128; /* size of buffer */
|
||
int32_t *wbuf, *rbuf;
|
||
|
||
TESTING("reading data from both accumulator and disk");
|
||
|
||
/* Allocate buffers */
|
||
wbuf = (int32_t *)HDmalloc((size_t)s * sizeof(int32_t));
|
||
HDassert(wbuf);
|
||
rbuf = (int32_t *)HDcalloc((size_t)s, sizeof(int32_t));
|
||
HDassert(rbuf);
|
||
|
||
/* Fill up write buffer with 1s */
|
||
for(i = 0; i < s; i++)
|
||
wbuf[i] = 1;
|
||
|
||
/* Write data to the accumulator to fill it. */
|
||
if(accum_write(512, 512, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write a piece of metadata outside current accumulator to force write
|
||
to disk */
|
||
if(accum_write(0, 1, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Fill up write buffer with 2s */
|
||
for(i = 0; i < s; i++)
|
||
wbuf[i] = 2;
|
||
|
||
/* Write a block of 2s of the original size that will overlap the lower half
|
||
of the original block */
|
||
if(accum_write(256, 512, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read 128 bytes at the original address, and then */
|
||
if(accum_read(512, 512, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Set the second half of wbuf back to 1s */
|
||
for(i = 64; i < s; i++)
|
||
wbuf[i] = 1;
|
||
|
||
/* Read in the piece we wrote to disk above, and then verify that
|
||
the data is as expected */
|
||
if(accum_read(512, 512, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)128) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
|
||
return 1;
|
||
} /* end test_read_after */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_big
|
||
*
|
||
* Purpose: This test exercises writing large pieces of metadata to the
|
||
* file.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Quincey Koziol
|
||
* October 12, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_big(H5F_t *f)
|
||
{
|
||
uint8_t *wbuf, *wbuf2, *rbuf, *zbuf; /* Buffers for reading & writing, etc */
|
||
unsigned u; /* Local index variable */
|
||
|
||
/* Allocate space for the write & read buffers */
|
||
wbuf = (uint8_t *)HDmalloc((size_t)BIG_BUF_SIZE);
|
||
HDassert(wbuf);
|
||
wbuf2 = (uint8_t *)HDmalloc((size_t)BIG_BUF_SIZE);
|
||
HDassert(wbuf2);
|
||
rbuf = (uint8_t *)HDcalloc((size_t)(BIG_BUF_SIZE + 1536), (size_t)1);
|
||
HDassert(rbuf);
|
||
zbuf = (uint8_t *)HDcalloc((size_t)(BIG_BUF_SIZE + 1536), (size_t)1);
|
||
HDassert(zbuf);
|
||
|
||
/* Initialize write buffers */
|
||
for(u = 0; u < BIG_BUF_SIZE; u++) {
|
||
wbuf[u] = (uint8_t)u;
|
||
wbuf2[u] = (uint8_t)(u + 1);
|
||
} /* end for */
|
||
|
||
TESTING("large metadata I/O operations");
|
||
|
||
/* Write large data segment to file */
|
||
if(accum_write(0, BIG_BUF_SIZE, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read entire segment back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(0, BIG_BUF_SIZE, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)BIG_BUF_SIZE);
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section to middle of accumulator */
|
||
if(accum_write(1024, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read entire segment back from file */
|
||
/* (Read covers entire dirty region) */
|
||
if(accum_read(0, BIG_BUF_SIZE, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(zbuf, rbuf, (size_t)1024) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf + 1024, (size_t)1024) != 0) TEST_ERROR;
|
||
if(HDmemcmp(zbuf, rbuf + 2048, (size_t)(BIG_BUF_SIZE - 2048)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(1024, 1024, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)BIG_BUF_SIZE);
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section to overlap with end of "big" region */
|
||
if(accum_write(BIG_BUF_SIZE - 512, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read entire segment back from file */
|
||
/* (Read covers bottom half of dirty region) */
|
||
if(accum_read(0, BIG_BUF_SIZE, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(zbuf, rbuf, (size_t)(BIG_BUF_SIZE - 512)) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf + (BIG_BUF_SIZE - 512), (size_t)512) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(BIG_BUF_SIZE - 512, 1024, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)BIG_BUF_SIZE);
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section to overlap with beginning of "big" region */
|
||
if(accum_write(0, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read entire segment back from file */
|
||
/* (Read covers bottom half of dirty region) */
|
||
if(accum_read(512, BIG_BUF_SIZE, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(wbuf + 512, rbuf, (size_t)512) != 0) TEST_ERROR;
|
||
if(HDmemcmp(zbuf, rbuf + 512, (size_t)(BIG_BUF_SIZE - 512)) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(0, 1024, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)BIG_BUF_SIZE);
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section to middle of accumulator */
|
||
/* (With write buffer #1) */
|
||
if(accum_write(1024, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write entire segment to from file */
|
||
/* (With write buffer #2) */
|
||
/* (Write covers entire dirty region) */
|
||
if(accum_write(0, BIG_BUF_SIZE, wbuf2) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read entire segment back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(wbuf2, rbuf, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(0, BIG_BUF_SIZE, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)BIG_BUF_SIZE);
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section to overlap with end of "big" region */
|
||
/* (With write buffer #1) */
|
||
if(accum_write(BIG_BUF_SIZE - 512, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write entire segment to from file */
|
||
/* (With write buffer #2) */
|
||
/* (Read covers bottom half of dirty region) */
|
||
if(accum_write(0, BIG_BUF_SIZE, wbuf2) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read both segments back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE + 512, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(wbuf2, rbuf, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf + 512, rbuf + BIG_BUF_SIZE, (size_t)512) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(0, BIG_BUF_SIZE + 512, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)(BIG_BUF_SIZE + 512));
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section to be past "big" region */
|
||
/* (With write buffer #1) */
|
||
if(accum_write(BIG_BUF_SIZE + 512, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read section before "big" region */
|
||
/* (To enlarge accumulator, to it will intersect with big write) */
|
||
if(accum_read(BIG_BUF_SIZE - 512, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write entire segment to from file */
|
||
/* (With write buffer #2) */
|
||
/* (Doesn't overlap with small section) */
|
||
if(accum_write(0, BIG_BUF_SIZE, wbuf2) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read both segments & gap back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE + 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(wbuf2, rbuf, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
if(HDmemcmp(zbuf, rbuf + BIG_BUF_SIZE, (size_t)512) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf, rbuf + BIG_BUF_SIZE + 512, (size_t)512) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(0, BIG_BUF_SIZE + 1536, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)(BIG_BUF_SIZE + 1024));
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section to be past "big" region */
|
||
/* (With write buffer #1) */
|
||
if(accum_write(BIG_BUF_SIZE + 512, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read section before "big" region */
|
||
/* (To enlarge accumulator, so it will intersect with big write) */
|
||
if(accum_read(BIG_BUF_SIZE - 512, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(accum_read(BIG_BUF_SIZE + 1536, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write entire segment to from file */
|
||
/* (With write buffer #2) */
|
||
/* (Overwriting dirty region, but not invalidating entire accumulator) */
|
||
if(accum_write(1536, BIG_BUF_SIZE, wbuf2) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read both segments & gap back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE + 1536, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(zbuf, rbuf, (size_t)1536) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf2, rbuf + 1536, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(1536, BIG_BUF_SIZE, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)(BIG_BUF_SIZE + 1536));
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section before "big" region */
|
||
/* (With write buffer #1) */
|
||
if(accum_write(1024, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read section before "big" region */
|
||
/* (To enlarge accumulator, so it will intersect with big write) */
|
||
if(accum_read(0, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write entire segment to from file */
|
||
/* (With write buffer #2) */
|
||
/* (Overwriting dirty region, but not invalidating entire accumulator) */
|
||
if(accum_write(512, BIG_BUF_SIZE, wbuf2) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read both segments & gap back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE + 512, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(zbuf, rbuf, (size_t)512) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf2, rbuf + 512, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(512, BIG_BUF_SIZE, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)(BIG_BUF_SIZE + 512));
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section before "big" region */
|
||
/* (With write buffer #1) */
|
||
if(accum_write(0, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read section before "big" region */
|
||
/* (To enlarge accumulator, so it will intersect with big write) */
|
||
if(accum_read(1024, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write entire segment to from file */
|
||
/* (With write buffer #2) */
|
||
/* (Avoiding dirty region, and not invalidating entire accumulator) */
|
||
if(accum_write(1536, BIG_BUF_SIZE, wbuf2) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read both segments & gap back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE + 1536, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)1024) != 0) TEST_ERROR;
|
||
if(HDmemcmp(zbuf, rbuf + 1024, (size_t)512) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf2, rbuf + 1536, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
|
||
|
||
/* Reset data in file back to zeros & reset the read buffer */
|
||
if(accum_write(0, BIG_BUF_SIZE + 1536, zbuf) < 0) FAIL_STACK_ERROR;
|
||
HDmemset(rbuf, 0, (size_t)(BIG_BUF_SIZE + 1536));
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
|
||
/* Write small section before "big" region */
|
||
/* (With write buffer #1) */
|
||
if(accum_write(0, 1024, wbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read section before "big" region */
|
||
/* (To enlarge accumulator, so it will intersect with big write) */
|
||
if(accum_read(1024, 1024, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Write entire segment to from file */
|
||
/* (With write buffer #2) */
|
||
/* (Partially overwriting dirty region, and not invalidating entire accumulator) */
|
||
if(accum_write(512, BIG_BUF_SIZE, wbuf2) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Read both segments back from file */
|
||
if(accum_read(0, BIG_BUF_SIZE + 512, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read */
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)512) != 0) TEST_ERROR;
|
||
if(HDmemcmp(wbuf2, rbuf + 512, (size_t)BIG_BUF_SIZE) != 0) TEST_ERROR;
|
||
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(wbuf2);
|
||
HDfree(rbuf);
|
||
HDfree(zbuf);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
HDfree(wbuf);
|
||
HDfree(wbuf2);
|
||
HDfree(rbuf);
|
||
HDfree(zbuf);
|
||
|
||
return 1;
|
||
} /* end test_big() */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_random_write
|
||
*
|
||
* Purpose: This test writes random pieces of data to the file and
|
||
* then reads it all back.
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Quincey Koziol
|
||
* October 11, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_random_write(H5F_t *f)
|
||
{
|
||
uint8_t *wbuf, *rbuf; /* Buffers for reading & writing */
|
||
unsigned seed = 0; /* Random # seed */
|
||
size_t *off; /* Offset of buffer segments to write */
|
||
size_t *len; /* Size of buffer segments to write */
|
||
size_t cur_off; /* Current offset */
|
||
size_t nsegments; /* Number of segments to write */
|
||
size_t swap; /* Position to swap with */
|
||
unsigned u; /* Local index variable */
|
||
|
||
/* Allocate space for the write & read buffers */
|
||
wbuf = (uint8_t *)HDmalloc((size_t)RANDOM_BUF_SIZE);
|
||
HDassert(wbuf);
|
||
rbuf = (uint8_t *)HDcalloc((size_t)RANDOM_BUF_SIZE, (size_t)1);
|
||
HDassert(rbuf);
|
||
|
||
/* Initialize write buffer */
|
||
for(u = 0; u < RANDOM_BUF_SIZE; u++)
|
||
wbuf[u] = (uint8_t)u;
|
||
|
||
TESTING("random writes to accumulator");
|
||
|
||
/* Choose random # seed */
|
||
seed = (unsigned)HDtime(NULL);
|
||
#ifdef QAK
|
||
/* seed = (unsigned)1155438845; */
|
||
HDfprintf(stderr, "Random # seed was: %u\n", seed);
|
||
#endif /* QAK */
|
||
HDsrandom(seed);
|
||
|
||
/* Allocate space for the segment length buffer */
|
||
off = (size_t *)HDmalloc(MAX_RANDOM_SEGMENTS * sizeof(size_t));
|
||
HDassert(off);
|
||
len = (size_t *)HDmalloc(MAX_RANDOM_SEGMENTS * sizeof(size_t));
|
||
HDassert(len);
|
||
|
||
/* Randomly choose lengths of segments */
|
||
cur_off = 0;
|
||
for(u = 0; u < MAX_RANDOM_SEGMENTS; ) {
|
||
size_t length = 0; /* Length of current segment */
|
||
|
||
/* Choose random length of segment, allowing for variance */
|
||
do {
|
||
length += (size_t)(HDrandom() % RAND_SEG_LEN) + 1;
|
||
} while((HDrandom() & 256) >= 128); /* end while */
|
||
|
||
/* Check for going off end of buffer */
|
||
if((cur_off + length) > RANDOM_BUF_SIZE)
|
||
length = RANDOM_BUF_SIZE - cur_off;
|
||
|
||
/* Set offset & length of segment */
|
||
off[u] = cur_off;
|
||
len[u] = length;
|
||
|
||
/* Advance array offset */
|
||
u++;
|
||
|
||
/* Advance current offset */
|
||
cur_off += length;
|
||
|
||
/* If we've used up entire buffer before hitting limit of segments, get out */
|
||
if(cur_off >= RANDOM_BUF_SIZE)
|
||
break;
|
||
} /* end for */
|
||
nsegments = u;
|
||
|
||
/* Increase length of last segment, if it doesn't reach end of buffer */
|
||
if(nsegments < MAX_RANDOM_SEGMENTS)
|
||
len[nsegments - 1] = RANDOM_BUF_SIZE - off[nsegments - 1];
|
||
|
||
/* Shuffle order of segments, to randomize positions to write */
|
||
for(u = 0; u < nsegments; u++) {
|
||
size_t tmp; /* Temporary holder for offset & length values */
|
||
|
||
/* Choose value within next few elements to to swap with */
|
||
swap = ((size_t)HDrandom() % 8) + u;
|
||
if(swap >= nsegments)
|
||
swap = nsegments - 1;
|
||
|
||
/* Swap values */
|
||
tmp = off[u]; off[u] = off[swap]; off[swap] = tmp;
|
||
tmp = len[u]; len[u] = len[swap]; len[swap] = tmp;
|
||
} /* end for */
|
||
|
||
/* Write data segments to file */
|
||
for(u = 0; u < nsegments; u++) {
|
||
if(accum_write(RANDOM_BASE_OFF + off[u], len[u], wbuf + off[u]) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify individual reads */
|
||
if(accum_read(RANDOM_BASE_OFF + off[u], len[u], rbuf) < 0) FAIL_STACK_ERROR;
|
||
if(HDmemcmp(wbuf + off[u], rbuf, len[u]) != 0) TEST_ERROR;
|
||
} /* end for */
|
||
|
||
/* Read entire region back from file */
|
||
if(accum_read(RANDOM_BASE_OFF, RANDOM_BUF_SIZE, rbuf) < 0) FAIL_STACK_ERROR;
|
||
|
||
/* Verify data read back in */
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)RANDOM_BUF_SIZE) != 0) TEST_ERROR;
|
||
|
||
if(accum_reset(f) < 0) FAIL_STACK_ERROR;
|
||
|
||
PASSED();
|
||
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
HDfree(off);
|
||
HDfree(len);
|
||
|
||
return 0;
|
||
|
||
error:
|
||
/* Release memory */
|
||
HDfree(wbuf);
|
||
HDfree(rbuf);
|
||
HDfree(off);
|
||
HDfree(len);
|
||
|
||
HDfprintf(stderr, "Random # seed was: %u\n", seed);
|
||
return 1;
|
||
} /* end test_random_write() */
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: test_swmr_write_big
|
||
*
|
||
* Purpose: A SWMR test: verifies that writing "large" metadata to a file
|
||
* opened with SWMR_WRITE will flush the existing metadata in the
|
||
* accumulator to disk first before writing the "large" metadata
|
||
* to disk.
|
||
* This test will fork and exec a reader "accum_swmr_reader" which
|
||
* opens the same file with SWMR_READ and verifies that the correct
|
||
* metadata is read from disk.
|
||
*
|
||
* Return: Success: 0
|
||
* Failure: 1
|
||
*
|
||
* Programmer: Vailin Choi; April 2013
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
unsigned
|
||
test_swmr_write_big(hbool_t newest_format)
|
||
{
|
||
hid_t fid = -1; /* File ID */
|
||
hid_t fapl = -1; /* File access property list */
|
||
H5F_t *rf = NULL; /* File pointer */
|
||
char filename[1024];
|
||
uint8_t *wbuf2 = NULL, *rbuf = NULL; /* Buffers for reading & writing */
|
||
uint8_t wbuf[1024]; /* Buffer for reading & writing */
|
||
unsigned u; /* Local index variable */
|
||
#ifdef H5_HAVE_UNISTD_H
|
||
pid_t pid; /* Process ID */
|
||
#endif /* H5_HAVE_UNISTD_H */
|
||
int status; /* Status returned from child process */
|
||
char *new_argv[] = {NULL};
|
||
char *driver = NULL; /* VFD string (from env variable) */
|
||
hbool_t api_ctx_pushed = FALSE; /* Whether API context pushed */
|
||
|
||
if(newest_format)
|
||
TESTING("SWMR write of large metadata: with latest format")
|
||
else
|
||
TESTING("SWMR write of large metadata: with non-latest-format")
|
||
|
||
#if !(defined(H5_HAVE_FORK) && defined(H5_HAVE_WAITPID))
|
||
|
||
SKIPPED();
|
||
HDputs(" Test skipped due to fork or waitpid not defined.");
|
||
return 0;
|
||
|
||
#else /* defined(H5_HAVE_FORK && defined(H5_HAVE_WAITPID) */
|
||
|
||
/* Skip this test if SWMR I/O is not supported for the VFD specified
|
||
* by the environment variable.
|
||
*/
|
||
driver = HDgetenv("HDF5_DRIVER");
|
||
if(!H5FD__supports_swmr_test(driver)) {
|
||
SKIPPED();
|
||
HDputs(" Test skipped due to VFD not supporting SWMR I/O.");
|
||
return 0;
|
||
}
|
||
|
||
/* File access property list */
|
||
if((fapl = h5_fileaccess()) < 0)
|
||
FAIL_STACK_ERROR
|
||
h5_fixname(FILENAME[1], fapl, filename, sizeof filename);
|
||
|
||
/* Both cases will result in v3 superblock and version 2 object header for SWMR */
|
||
if(newest_format) { /* latest format */
|
||
if(H5Pset_libver_bounds(fapl, H5F_LIBVER_LATEST, H5F_LIBVER_LATEST) < 0)
|
||
FAIL_STACK_ERROR
|
||
|
||
if((fid = H5Fcreate(filename, H5F_ACC_TRUNC, H5P_DEFAULT, fapl)) < 0)
|
||
FAIL_STACK_ERROR
|
||
}
|
||
else { /* non-latest-format */
|
||
if((fid = H5Fcreate(filename, H5F_ACC_TRUNC|H5F_ACC_SWMR_WRITE, H5P_DEFAULT, fapl)) < 0)
|
||
FAIL_STACK_ERROR
|
||
} /* end if */
|
||
|
||
/* Close the file */
|
||
if(H5Fclose(fid) < 0)
|
||
FAIL_STACK_ERROR
|
||
|
||
/* Open the file with SWMR_WRITE */
|
||
if((fid = H5Fopen(filename, H5F_ACC_RDWR | H5F_ACC_SWMR_WRITE, fapl)) < 0)
|
||
FAIL_STACK_ERROR
|
||
|
||
/* Push API context */
|
||
if(H5CX_push() < 0) FAIL_STACK_ERROR
|
||
api_ctx_pushed = TRUE;
|
||
|
||
/* Get H5F_t * to internal file structure */
|
||
if(NULL == (rf = (H5F_t *)H5VL_object(fid))) FAIL_STACK_ERROR
|
||
|
||
/* We'll be writing lots of garbage data, so extend the
|
||
file a ways. 10MB should do. */
|
||
if(H5FD_set_eoa(rf->shared->lf, H5FD_MEM_DEFAULT, (haddr_t)(1024*1024*10)) < 0)
|
||
FAIL_STACK_ERROR
|
||
|
||
if(H5Fflush(fid, H5F_SCOPE_GLOBAL) < 0)
|
||
FAIL_STACK_ERROR;
|
||
|
||
/* Reset metadata accumulator for the file */
|
||
if(accum_reset(rf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
|
||
/* Allocate space for the write & read buffers */
|
||
if((wbuf2 = (uint8_t *)HDmalloc((size_t)BIG_BUF_SIZE)) == NULL)
|
||
FAIL_STACK_ERROR;
|
||
if((rbuf = (uint8_t *)HDmalloc((size_t)BIG_BUF_SIZE)) == NULL)
|
||
FAIL_STACK_ERROR;
|
||
|
||
/* Initialize wbuf with "0, 1, 2...1024"*/
|
||
for(u = 0; u < 1024; u++)
|
||
wbuf[u] = (uint8_t)u;
|
||
|
||
/* Write [1024, 1024] bytes with wbuf */
|
||
if(H5F_block_write(rf, H5FD_MEM_DEFAULT, (haddr_t)1024, (size_t)1024, wbuf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
/* Read the data */
|
||
if(H5F_block_read(rf, H5FD_MEM_DEFAULT, (haddr_t)1024, (size_t)1024, rbuf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
/* Verify the data read is correct */
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)1024) != 0)
|
||
TEST_ERROR;
|
||
/* Flush the data to disk */
|
||
if(accum_reset(rf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
|
||
/* Initialize wbuf with all 1s */
|
||
for(u = 0; u < 1024; u++)
|
||
wbuf[u] = (uint8_t)1;
|
||
|
||
/* Initialize wbuf2 */
|
||
for(u = 0; u < BIG_BUF_SIZE; u++)
|
||
wbuf2[u] = (uint8_t)(u + 1);
|
||
|
||
/* Write [1024,1024] with wbuf--all 1s */
|
||
if(H5F_block_write(rf, H5FD_MEM_DEFAULT, (haddr_t)1024, (size_t)1024, wbuf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
/* Read the data */
|
||
if(H5F_block_read(rf, H5FD_MEM_DEFAULT, (haddr_t)1024, (size_t)1024, rbuf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
/* Verify the data read is correct */
|
||
if(HDmemcmp(wbuf, rbuf, (size_t)1024) != 0)
|
||
TEST_ERROR;
|
||
/* The data stays in the accumulator */
|
||
|
||
/* Write a large piece of metadata [2048, BIG_BUF_SIZE] with wbuf2 */
|
||
if(H5F_block_write(rf, H5FD_MEM_DEFAULT, (haddr_t)2048, (size_t)BIG_BUF_SIZE, wbuf2) < 0)
|
||
FAIL_STACK_ERROR;
|
||
/* Read the data */
|
||
if(H5F_block_read(rf, H5FD_MEM_DEFAULT, (haddr_t)2048, (size_t)BIG_BUF_SIZE, rbuf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
/* Verify the data read is correct */
|
||
if(HDmemcmp(wbuf2, rbuf, (size_t)BIG_BUF_SIZE) != 0)
|
||
TEST_ERROR;
|
||
|
||
/* Fork child process to verify that the data at [1024, 2014] does get written to disk */
|
||
if((pid = HDfork()) < 0) {
|
||
HDperror("fork");
|
||
FAIL_STACK_ERROR;
|
||
}
|
||
else if(0 == pid) { /* Child process */
|
||
/* Run the reader */
|
||
status = HDexecv(SWMR_READER, new_argv);
|
||
HDprintf("errno from execv = %s\n", strerror(errno));
|
||
FAIL_STACK_ERROR;
|
||
} /* end if */
|
||
|
||
/* Parent process -- wait for the child process to complete */
|
||
while(pid != HDwaitpid(pid, &status, 0))
|
||
/*void*/;
|
||
|
||
/* Check if child process terminates normally and its return value */
|
||
if(WIFEXITED(status) && !WEXITSTATUS(status)) {
|
||
/* Flush the accumulator */
|
||
if(accum_reset(rf) < 0)
|
||
FAIL_STACK_ERROR;
|
||
|
||
/* Close and remove the file */
|
||
if(H5Fclose(fid) < 0)
|
||
FAIL_STACK_ERROR;
|
||
|
||
/* Close the property list */
|
||
if(H5Pclose(fapl) < 0)
|
||
FAIL_STACK_ERROR;
|
||
|
||
/* Pop API context */
|
||
if(api_ctx_pushed && H5CX_pop() < 0) FAIL_STACK_ERROR
|
||
api_ctx_pushed = FALSE;
|
||
|
||
/* Release memory */
|
||
if(wbuf2)
|
||
HDfree(wbuf2);
|
||
if(rbuf)
|
||
HDfree(rbuf);
|
||
PASSED();
|
||
return 0;
|
||
} /* end if */
|
||
|
||
error:
|
||
/* Closing and remove the file */
|
||
H5Fclose(fid);
|
||
|
||
if(api_ctx_pushed) H5CX_pop();
|
||
|
||
H5Pclose(fapl);
|
||
|
||
/* Release memory */
|
||
if(wbuf2)
|
||
HDfree(wbuf2);
|
||
if(rbuf)
|
||
HDfree(rbuf);
|
||
|
||
return 1;
|
||
|
||
#endif
|
||
|
||
} /* end test_swmr_write_big() */
|
||
|
||
|
||
/*-------------------------------------------------------------------------
|
||
* Function: accum_printf
|
||
*
|
||
* Purpose: Debug function to print some stats about the accumulator
|
||
*
|
||
* Return: Success: SUCCEED
|
||
* Failure: FAIL
|
||
*
|
||
* Programmer: Mike McGreevy
|
||
* October 7, 2010
|
||
*
|
||
*-------------------------------------------------------------------------
|
||
*/
|
||
void
|
||
accum_printf(const H5F_t *f)
|
||
{
|
||
H5F_meta_accum_t * accum = &f->shared->accum;
|
||
|
||
HDprintf("\n");
|
||
HDprintf("Current contents of accumulator:\n");
|
||
if(accum->alloc_size == 0) {
|
||
HDprintf("=====================================================\n");
|
||
HDprintf(" No accumulator allocated.\n");
|
||
HDprintf("=====================================================\n");
|
||
}
|
||
else {
|
||
HDprintf("=====================================================\n");
|
||
HDprintf(" accumulator allocated size == %zu\n", accum->alloc_size);
|
||
HDprintf(" accumulated data size == %zu\n", accum->size);
|
||
HDfprintf(stdout, " accumulator dirty? == %t\n", accum->dirty);
|
||
HDprintf("=====================================================\n");
|
||
HDfprintf(stdout, " start of accumulated data, loc = %a\n", accum->loc);
|
||
if(accum->dirty) {
|
||
HDfprintf(stdout, " start of dirty region, loc = %a\n", (haddr_t)(accum->loc + accum->dirty_off));
|
||
HDfprintf(stdout, " end of dirty region, loc = %a\n", (haddr_t)(accum->loc + accum->dirty_off + accum->dirty_len));
|
||
} /* end if */
|
||
HDfprintf(stdout, " end of accumulated data, loc = %a\n", (haddr_t)(accum->loc + accum->size));
|
||
HDfprintf(stdout, " end of accumulator allocation, loc = %a\n", (haddr_t)(accum->loc + accum->alloc_size));
|
||
HDprintf("=====================================================\n");
|
||
}
|
||
HDprintf("\n\n");
|
||
} /* accum_printf() */
|
||
|