FastSort/src/sort/sam_io.cpp

233 lines
8.9 KiB
C++
Raw Normal View History

2025-04-17 16:17:44 +08:00
#include "sam_io.h"
#include <htslib/sam.h>
#include <libdeflate.h>
#include <stdint.h>
#include <stdio.h>
#include <zlib.h>
#include <spdlog/spdlog.h>
#include "const_val.h"
2025-04-17 16:17:44 +08:00
namespace nsgv {
extern bool gIsBigEndian;
};
static const uint8_t g_magic[19] = "\037\213\010\4\0\0\0\0\0\377\6\0\102\103\2\0\0\0";
size_t fileSize(FILE *fp) {
spdlog::info("test file start");
size_t bufSize = 64L * 1024 * 1024;
uint8_t *testBuf = (uint8_t *)malloc(bufSize);
size_t readState = 0;
size_t fileSize = 0;
while ((readState = fread(testBuf, 1, bufSize, fp)) > 0) {
fileSize += readState;
}
spdlog::info("test file size: {}", fileSize);
return fileSize;
}
static inline void packInt16(uint8_t* buffer, uint16_t value) {
buffer[0] = value;
buffer[1] = value >> 8;
}
static inline void packInt32(uint8_t* buffer, uint32_t value) {
buffer[0] = value;
buffer[1] = value >> 8;
buffer[2] = value >> 16;
buffer[3] = value >> 24;
}
// dlen本身长度一定得够存放压缩后的数据否则会返回错误
int bgzfCompress(void* _dst, size_t* dlen, const void* src, size_t slen, int level) {
// fprintf(stderr, "bgzfCompress, slen: %zu\n", slen);
if (slen == 0) {
// EOF block
if (*dlen < 28)
return -1;
memcpy(_dst, "\037\213\010\4\0\0\0\0\0\377\6\0\102\103\2\0\033\0\3\0\0\0\0\0\0\0\0\0", 28);
*dlen = 28;
return 0;
}
uint8_t* dst = (uint8_t*)_dst;
if (level == 0) {
// Uncompressed data
if (*dlen < slen + 5 + BLOCK_HEADER_LENGTH + BLOCK_FOOTER_LENGTH)
return -1;
dst[BLOCK_HEADER_LENGTH] = 1; // BFINAL=1, BTYPE=00; see RFC1951
u16_to_le(slen, &dst[BLOCK_HEADER_LENGTH + 1]); // length
u16_to_le(~slen, &dst[BLOCK_HEADER_LENGTH + 3]); // ones-complement length
memcpy(dst + BLOCK_HEADER_LENGTH + 5, src, slen);
*dlen = slen + 5 + BLOCK_HEADER_LENGTH + BLOCK_FOOTER_LENGTH;
} else {
level = level > 0 ? level : 6; // libdeflate doesn't honour -1 as default
// NB levels go up to 12 here.
int lvl_map[] = {0, 1, 2, 3, 5, 6, 7, 8, 10, 12};
level = lvl_map[level > 9 ? 9 : level];
struct libdeflate_compressor* z = libdeflate_alloc_compressor(level);
if (!z)
return -1;
// Raw deflate
size_t clen = libdeflate_deflate_compress(z, src, slen, dst + BLOCK_HEADER_LENGTH, *dlen - BLOCK_HEADER_LENGTH - BLOCK_FOOTER_LENGTH);
if (clen <= 0) {
hts_log_error("Call to libdeflate_deflate_compress failed");
libdeflate_free_compressor(z);
return -1;
}
*dlen = clen + BLOCK_HEADER_LENGTH + BLOCK_FOOTER_LENGTH;
libdeflate_free_compressor(z);
}
// write the header
memcpy(dst, g_magic, BLOCK_HEADER_LENGTH); // the last two bytes are a place holder for the length of the block
packInt16(&dst[16], *dlen - 1); // write the compressed length; -1 to fit 2 bytes
// write the footer
uint32_t crc = libdeflate_crc32(0, src, slen);
packInt32((uint8_t*)&dst[*dlen - 8], crc);
packInt32((uint8_t*)&dst[*dlen - 4], slen);
return 0;
}
2025-04-17 16:17:44 +08:00
int bgzfUncompress(uint8_t *dst, size_t *dlen, const uint8_t *src, size_t slen, uint32_t expected_crc) {
struct libdeflate_decompressor *z = libdeflate_alloc_decompressor();
if (!z) {
hts_log_error("Call to libdeflate_alloc_decompressor failed");
return -1;
}
int ret = libdeflate_deflate_decompress(z, src, slen, dst, *dlen, dlen);
libdeflate_free_decompressor(z);
if (ret != LIBDEFLATE_SUCCESS) {
hts_log_error("Inflate operation failed: %d", ret);
return -1;
}
uint32_t crc = libdeflate_crc32(0, (unsigned char *)dst, *dlen);
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
// Pretend the CRC was OK so the fuzzer doesn't have to get it right
crc = expected_crc;
#endif
if (crc != expected_crc) {
hts_log_error("CRC32 checksum mismatch");
return -2;
}
return 0;
}
// 读取并解压一个gz block
size_t readUncompressOneBlock(FILE *fpr, uint8_t *fBuf, DataBuffer *uDataPtr) {
DataBuffer &uData = *uDataPtr;
size_t readState = 0;
int blockLen = 0; // 每个gz block的大小
size_t dlen = SINGLE_BLOCK_SIZE; // gz block解压之后的真正大小
readState = fread(fBuf, 1, BLOCK_HEADER_LENGTH, fpr); // 先读取一个gz block的头部
blockLen = unpackInt16(&fBuf[16]) + 1; // block的字节数
readState = fread(&fBuf[BLOCK_HEADER_LENGTH], 1, blockLen - BLOCK_HEADER_LENGTH, fpr);
// 解压gz block
uint32_t crc = le_to_u32(fBuf + blockLen - 8);
size_t newDataSize = uData.maxLen;
while (uData.curLen + SINGLE_BLOCK_SIZE > newDataSize) newDataSize *= 2;
uData.ReAllocMem(newDataSize); // 需要重新开辟空间
int ret = bgzfUncompress(&uData.data[uData.curLen], &dlen, (Bytef *)fBuf + BLOCK_HEADER_LENGTH,
blockLen - BLOCK_HEADER_LENGTH, crc);
if (ret < 0) {
spdlog::error("Error decompressing gz block");
exit(0);
}
uData.curLen += dlen;
spdlog::info("gz block size: {}, uncompressed size: {}", blockLen, dlen);
return readState;
}
// 解析sam头部信息
void parseSamHeader(FILE *fpr, HeaderBuf &hdrBuf) {
const int kMaxBlockSize = SINGLE_BLOCK_SIZE;
2025-04-17 16:17:44 +08:00
sam_hdr_t *header = NULL;
uint8_t *fBuf = (uint8_t*)malloc(kMaxBlockSize);
DataBuffer uData; // 用来放解压缩后的数据,待解析
2025-04-17 16:17:44 +08:00
int magicLen;
int32_t i, nameLen, numNames = 0;
header = sam_hdr_init(); // 初始化header
uData.AllocMem(kMaxBlockSize); // 初始化解压数据的buffer
readUncompressOneBlock(fpr, fBuf, &uData); // 读取第一个gz block
2025-04-17 16:17:44 +08:00
// 解析header
magicLen = 4;
if (memcmp(uData.data, "BAM\1", magicLen)) {
spdlog::error("Invalid BAM binary header");
return;
}
uData.readPos += magicLen; // 在解压数据中读取了magicLen个字节
header->l_text = le_to_u32(uData.data + uData.readPos); uData.readPos += 4; // 解析l_text的长度
2025-04-17 16:17:44 +08:00
header->text = (char *)malloc(header->l_text + 1);
header->text[header->l_text] = 0; // 在text末尾添加'\0'
while (uData.readPos + header->l_text > uData.curLen) { // header内容超过了一个block继续读
readUncompressOneBlock(fpr, fBuf, &uData); // 读取包含header内容的gz block
}
2025-04-17 16:17:44 +08:00
memcpy(header->text, &uData.data[uData.readPos], header->l_text);
uData.readPos += header->l_text;
memcpy(&header->n_targets, &uData.data[uData.readPos], 4); // n_targets
uData.readPos += 4;
spdlog::info("num target: {}", header->n_targets);
if (nsgv::gIsBigEndian) ed_swap_4p(&header->n_targets);
if (header->n_targets > 0) {
header->target_name = (char **)calloc(header->n_targets, sizeof(char *));
header->target_len = (uint32_t *)calloc(header->n_targets, sizeof(uint32_t));
} else {
header->target_name = NULL;
header->target_len = NULL;
}
for (int i = 0; i != header->n_targets; ++i) {
while (uData.readPos + 4 > uData.curLen) readUncompressOneBlock(fpr, fBuf, &uData);
2025-04-17 16:17:44 +08:00
memcpy(&nameLen, &uData.data[uData.readPos], 4);
uData.readPos += 4;
if (nsgv::gIsBigEndian) ed_swap_4p(&nameLen);
header->target_name[i] = (char *)malloc(nameLen);
++numNames;
while (uData.readPos + nameLen > uData.curLen) readUncompressOneBlock(fpr, fBuf, &uData);
2025-04-17 16:17:44 +08:00
memcpy(header->target_name[i], &uData.data[uData.readPos], nameLen);
uData.readPos += nameLen;
if (header->target_name[i][nameLen - 1] != '\0') {
/* Fix missing NUL-termination. Is this being too nice?
We could alternatively bail out with an error. */
char *newName = (char*)realloc(header->target_name[i], nameLen + 1);
header->target_name[i] = newName;
header->target_name[i][nameLen] = '\0';
}
while (uData.readPos + 4 > uData.curLen) readUncompressOneBlock(fpr, fBuf, &uData);
2025-04-17 16:17:44 +08:00
memcpy(&header->target_len[i], &uData.data[uData.readPos], 4);
uData.readPos += 4;
if (nsgv::gIsBigEndian) ed_swap_4p(&header->target_len[i]);
// spdlog::info("nameLen {}, {}, {}", nameLen, header->target_len[i], header->target_name[i]);
}
// spdlog::info("test res: {} {} {} {}", header->l_text, dlen, header->n_targets, header->text);
spdlog::info("udata: readPos: {}, curLen: {}, maxLen: {}", uData.readPos, uData.curLen, uData.maxLen);
2025-04-17 16:17:44 +08:00
hdrBuf.header = header;
if (uData.readPos < uData.curLen) { // 如果还有数据则将其保存在hdrBuf中用以后续的bam解析
hdrBuf.data = (uint8_t *)malloc(uData.curLen - uData.readPos);
memcpy(hdrBuf.data, &uData.data[uData.readPos], uData.curLen - uData.readPos);
hdrBuf.dataLen = uData.curLen - uData.readPos;
}
2025-04-17 16:17:44 +08:00
free(fBuf);
2025-04-17 16:17:44 +08:00
}