static int md5_final(struct shash_desc *desc, u8 *out)
{
struct md5_state *mctx = shash_desc_ctx(desc);
const unsigned int offset = mctx->byte_count & 0x3f;
char *p = (char *)mctx->block + offset;
int padding = 56 - (offset + 1);
*p++ = 0x80;
if (padding < 0) {
memset(p, 0x00, padding + sizeof (u64));
md5_transform_helper(mctx);
p = (char *)mctx->block;
padding = 56;
}
memset(p, 0, padding);
mctx->block[14] = mctx->byte_count << 3;
mctx->block[15] = mctx->byte_count >> 29;
le32_to_cpu_array(mctx->block, (sizeof(mctx->block) -
sizeof(u64)) / sizeof(u32));
md5_transform(mctx->hash, mctx->block);
cpu_to_le32_array(mctx->hash, sizeof(mctx->hash) / sizeof(u32));
memcpy(out, mctx->hash, sizeof(mctx->hash));
memset(mctx, 0, sizeof(*mctx));
return 0;
}
void md5_finish(struct md_context *context) {
int buflen = context->count & 63;
uint64_t bitcount;
context->buffer[buflen++] = 0x80;
memset(&context->buffer[buflen], 0, 64-buflen);
if(buflen > 56) {
md5_transform(context);
memset(context->buffer, 0, 64);
}
// *(uint64_t *)(&context->buffer[56]) = context->count * 8;
bitcount = context->count * 8;
memcpy(&context->buffer[56], &bitcount, sizeof(uint64_t));
md5_transform(context);
memcpy(context->digest, context->h, 16);
return;
}
static inline void md5_transform_helper(struct md5_ctx *ctx)
{
#ifdef DEBUG_MD5
printk("md5_transform_helper running\n");
#endif //DEBUG_MD5
//le32_to_cpu_array(ctx->block, sizeof(ctx->block) / sizeof(u32));
md5_transform(ctx->hash, ctx->block);
}
__u32 secure_ipv6_id(const __be32 daddr[4])
{
__u32 hash[4];
memcpy(hash, daddr, 16);
md5_transform(hash, net_secret);
return hash[0];
}
void md5_hash (unsigned char * sum, unsigned char * data, int data_len)
{
struct md5_context context;
int i;
context.A = md5_A;
context.B = md5_B;
context.C = md5_C;
context.D = md5_D;
data_len %= 55;
// turned out faster than memset
for (i = 0; i < 16; i++)
context.M[i] ^= context.M[i];
// faster than memcpy
for (i = 0; i < data_len / 4; i++)
context.M[i] = ((unsigned int *) data)[i];
i <<= 2;
switch (data_len & 3)
{
case 0 :
((unsigned char *) context.M)[i] = 0x80;
break;
case 1 :
((unsigned char *) context.M)[i] = data[i];
i++;
((unsigned char *) context.M)[i] = 0x80;
break;
case 2 :
((unsigned char *) context.M)[i] = data[i];
i++;
((unsigned char *) context.M)[i] = data[i];
i++;
((unsigned char *) context.M)[i] = 0x80;
break;
case 3 :
((unsigned char *) context.M)[i] = data[i];
i++;
((unsigned char *) context.M)[i] = data[i];
i++;
((unsigned char *) context.M)[i] = data[i];
i++;
((unsigned char *) context.M)[i] = 0x80;
break;
}
context.M[14] = data_len << 3; // * 8
md5_transform(&context);
*((uint32_t *) &(sum[ 0])) = context.A;
*((uint32_t *) &(sum[ 4])) = context.B;
*((uint32_t *) &(sum[ 8])) = context.C;
*((uint32_t *) &(sum[12])) = context.D;
}
static unsigned int get_random_int(void)
{
__u32 hash, random;
unsigned int ret;
hash = get_cycles();
md5_transform(&hash, &random);
ret = hash;
return ret;
}
void md5_update(struct md_context *context, uint8_t *chunk, uint64_t chunk_size) {
int buflen = context->count & 63;
context->count += chunk_size;
if((buflen + chunk_size) < 64) {
memcpy(&context->buffer[buflen], chunk, chunk_size);
return;
}
memcpy(&context->buffer[buflen], chunk, 64 - buflen);
md5_transform(context);
chunk_size -= (64 - buflen);
chunk += (64 - buflen);
while(chunk_size >= 64) {
memcpy(context->buffer, chunk, 64);
md5_transform(context);
chunk_size -= 64;
chunk += 64;
}
memcpy(context->buffer, chunk, chunk_size);
return;
}
static unsigned char *
md5_final()
{
int i, buflen = length & 63;
buffer[buflen++] = 0x80;
memset (buffer+buflen, 0, 64 - buflen);
if (buflen > 56)
{
md5_transform (buffer);
memset (buffer, 0, 64);
buflen = 0;
}
*(UINT4 *) (buffer + 56) = cpu_to_le32 (8 * length);
*(UINT4 *) (buffer + 60) = 0;
md5_transform (buffer);
for (i = 0; i < 4; i++)
state[i] = cpu_to_le32 (state[i]);
return (unsigned char *) state;
}
u32 secure_ipv4_port_ephemeral(__be32 saddr, __be32 daddr, __be16 dport)
{
u32 hash[MD5_DIGEST_WORDS];
hash[0] = (__force u32)saddr;
hash[1] = (__force u32)daddr;
hash[2] = (__force u32)dport ^ net_secret[14];
hash[3] = net_secret[15];
md5_transform(hash, net_secret);
return hash[0];
}
__u32 secure_ip_id(__be32 daddr)
{
u32 hash[MD5_DIGEST_WORDS];
hash[0] = (__force __u32) daddr;
hash[1] = net_secret[13];
hash[2] = net_secret[14];
hash[3] = net_secret[15];
md5_transform(hash, net_secret);
return hash[0];
}
/* exact copy of tcp_sequence number function from net/core/secure.c */
__u32 secure_tcp_sequence_number(u32 saddr, u32 daddr,
u16 sport, u16 dport)
{
u32 hash[MD5_DIGEST_WORDS];
hash[0] = saddr;
hash[1] = daddr;
hash[2] = (sport << 16) + dport;
hash[3] = net_secret[15];
md5_transform(hash, net_secret);
return seq_scale(hash[0]);
}
void md5_final (MD5Schedule_ref ctx, char *digest /* 16 chars */)
{
register unsigned int count = (ctx->bits[0] >> 3) & 0x3F ;
register unsigned char *p = ctx->in + count ;
*p++ = 0x80;
count = 63 - count ;
if (count < 8)
{
byte_zero(p, count) ;
uint32_little_endian(ctx->in, 16) ;
md5_transform(ctx->buf, (uint32 *)ctx->in) ;
byte_zero(ctx->in, 56) ;
}
else byte_zero(p, count - 8) ;
uint32_little_endian(ctx->in, 14) ;
byte_copy(ctx->in + 56, 4, (char *)&ctx->bits[0]) ;
byte_copy(ctx->in + 60, 4, (char *)&ctx->bits[1]) ;
md5_transform(ctx->buf, (uint32 *)ctx->in) ;
uint32_little_endian((char *)ctx->buf, 4) ;
byte_copy(digest, 16, (char *)ctx->buf) ;
}
__u32 secure_tcp_sequence_number(__be32 saddr, __be32 daddr,
__be16 sport, __be16 dport)
{
u32 hash[MD5_DIGEST_WORDS];
hash[0] = (__force u32)saddr;
hash[1] = (__force u32)daddr;
hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
hash[3] = net_secret[15];
md5_transform(hash, net_secret);
return seq_scale(hash[0]);
}
/* Generate HMAC-MD5 intermediate Hash */
static void sa_hmac_md5_get_pad(const u8 *key, u16 key_sz, u32 *ipad, u32 *opad)
{
u8 k_ipad[MD5_MESSAGE_BYTES];
u8 k_opad[MD5_MESSAGE_BYTES];
int i;
for (i = 0; i < key_sz; i++) {
k_ipad[i] = key[i] ^ 0x36;
k_opad[i] = key[i] ^ 0x5c;
}
/* Instead of XOR with 0 */
for (; i < SHA_MESSAGE_BYTES; i++) {
k_ipad[i] = 0x36;
k_opad[i] = 0x5c;
}
/* SHA-1 on k_ipad */
md5_init(ipad);
md5_transform(ipad, (u32 *)k_ipad);
/* SHA-1 on k_opad */
md5_init(opad);
md5_transform(ipad, (u32 *)k_opad);
}
/*
* Final wrapup - pad to 64-byte boundary with the bit pattern
* 1 0* (64-bit count of bits processed, MSB-first)
*/
static inline void
md5_final (md5_ctx_t * ctx, u8 out[16])
{
/* Number of bytes in ctx->in */
const int offset = ctx->byte_count & 0x3f;
/* p points after last content byte in ctx->in */
u8 *p = (u8 *) ctx->in + offset;
/* Bytes of padding needed to make 56 bytes (-8..55) */
int padding = 56 - (offset + 1);
/* Set the first byte of padding to 0x80. There is always room. */
*p++ = 0x80;
if (padding < 0) { /* Padding forces an extra block */
memset (p, 0x00, padding + sizeof (u64));
md5_transform_helper (ctx);
p = (u8 *) ctx->in;
padding = 56;
}
/* pad remaining bytes w/ 0x00 */
memset (p, 0x00, padding);
/* Append length in bits and transform */
ctx->in[14] = ctx->byte_count << 3; /* low order word first */
ctx->in[15] = ctx->byte_count >> 29;
/* keep the appended bit-count words in host order! */
le32_to_cpu_array (ctx->in,
(sizeof (ctx->in) - sizeof (u64)) / sizeof (u32));
md5_transform (ctx->hash, ctx->in);
/* convert digest buf from host to LE byteorder */
cpu_to_le32_array (ctx->hash, sizeof (ctx->hash) / sizeof (u32));
/* copy to output buffer */
memcpy (out, ctx->hash, sizeof (ctx->hash));
/* wipe context */
memset (ctx, 0, sizeof (ctx));
}
u32 secure_ipv6_port_ephemeral(const __be32 *saddr, const __be32 *daddr,
__be16 dport)
{
u32 secret[MD5_MESSAGE_BYTES / 4];
u32 hash[MD5_DIGEST_WORDS];
u32 i;
memcpy(hash, saddr, 16);
for (i = 0; i < 4; i++)
secret[i] = net_secret[i] + (__force u32) daddr[i];
secret[4] = net_secret[4] + (__force u32)dport;
for (i = 5; i < MD5_MESSAGE_BYTES / 4; i++)
secret[i] = net_secret[i];
md5_transform(hash, secret);
return hash[0];
}
__u32 secure_tcpv6_sequence_number(__be32 *saddr, __be32 *daddr,
__be16 sport, __be16 dport)
{
u32 secret[MD5_MESSAGE_BYTES / 4];
u32 hash[MD5_DIGEST_WORDS];
u32 i;
memcpy(hash, saddr, 16);
for (i = 0; i < 4; i++)
secret[i] = net_secret[i] + daddr[i];
secret[4] = net_secret[4] +
(((__force u16)sport << 16) + (__force u16)dport);
for (i = 5; i < MD5_MESSAGE_BYTES / 4; i++)
secret[i] = net_secret[i];
md5_transform(hash, secret);
return seq_scale(hash[0]);
}
u64 secure_dccp_sequence_number(__be32 saddr, __be32 daddr,
__be16 sport, __be16 dport)
{
u32 hash[MD5_DIGEST_WORDS];
u64 seq;
hash[0] = (__force u32)saddr;
hash[1] = (__force u32)daddr;
hash[2] = ((__force u16)sport << 16) + (__force u16)dport;
hash[3] = net_secret[15];
md5_transform(hash, net_secret);
seq = hash[0] | (((u64)hash[1]) << 32);
seq += ktime_to_ns(ktime_get_real());
seq &= (1ull << 48) - 1;
return seq;
}
/*! \fn static void md5_final(struct crypto_tfm *tfm, u8 *out)
* \ingroup IFX_MD5_FUNCTIONS
* \brief compute final md5 value
* \param tfm linux crypto algo transform
* \param out final md5 output value
*/
static int md5_final(struct shash_desc *desc, u8 *out)
{
struct md5_ctx *mctx = shash_desc_ctx(desc);
const unsigned int offset = mctx->byte_count & 0x3f;
char *p = (char *)mctx->block + offset;
int padding = 56 - (offset + 1);
volatile struct deu_hash_t *hashs = (struct deu_hash_t *) HASH_START;
unsigned long flag;
*p++ = 0x80;
if (padding < 0) {
memset(p, 0x00, padding + sizeof (u64));
md5_transform_helper(mctx);
p = (char *)mctx->block;
padding = 56;
}
memset(p, 0, padding);
mctx->block[14] = endian_swap(mctx->byte_count << 3);
mctx->block[15] = endian_swap(mctx->byte_count >> 29);
#if 0
le32_to_cpu_array(mctx->block, (sizeof(mctx->block) -
sizeof(u64)) / sizeof(u32));
#endif
md5_transform(mctx, mctx->hash, mctx->block);
CRTCL_SECT_START;
*((u32 *) out + 0) = endian_swap (hashs->D1R);
*((u32 *) out + 1) = endian_swap (hashs->D2R);
*((u32 *) out + 2) = endian_swap (hashs->D3R);
*((u32 *) out + 3) = endian_swap (hashs->D4R);
CRTCL_SECT_END;
// Wipe context
memset(mctx, 0, sizeof(*mctx));
return 0;
}
u64 secure_dccpv6_sequence_number(__be32 *saddr, __be32 *daddr,
__be16 sport, __be16 dport)
{
u32 secret[MD5_MESSAGE_BYTES / 4];
u32 hash[MD5_DIGEST_WORDS];
u64 seq;
u32 i;
memcpy(hash, saddr, 16);
for (i = 0; i < 4; i++)
secret[i] = net_secret[i] + daddr[i];
secret[4] = net_secret[4] +
(((__force u16)sport << 16) + (__force u16)dport);
for (i = 5; i < MD5_MESSAGE_BYTES / 4; i++)
secret[i] = net_secret[i];
md5_transform(hash, secret);
seq = hash[0] | (((u64)hash[1]) << 32);
seq += ktime_to_ns(ktime_get_real());
seq &= (1ull << 48) - 1;
return seq;
}
static inline void md5_transform_helper(struct md5_ctx *ctx)
{
le32_to_cpu_array(ctx->block, sizeof(ctx->block) / sizeof(u32));
md5_transform(ctx->hash, ctx->block);
}
static inline void
md5_transform_helper (md5_ctx_t * ctx)
{
le32_to_cpu_array (ctx->in, sizeof (ctx->in) / sizeof (u32));
md5_transform (ctx->hash, ctx->in);
}
static void md5_final(void *ctx, u8 *out)
{
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
struct md5_ctx *mctx = ctx;
const unsigned int offset = mctx->byte_count & 0x3f;
char *p = (char *) mctx->block + offset;
int padding = 56 - (offset + 1);
unsigned long flag;
#ifndef ONLY_IN_MEM
volatile struct deu_hash_t *hashs = (struct hash_t *) HASH_START;
#else
unsigned char *prin = NULL;
struct deu_hash_t *hashs = NULL;
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
hashs = (struct deu_hash_t *) kmalloc(sizeof(*hashs), GFP_KERNEL);
memset(hashs, 0, sizeof(*hashs));
prin = (unsigned char *) hashs;
#endif
#ifdef DEBUG_MD5
printk("md5_final running\n");
//hexdump(mctx,sizeof(*mctx));
printk("block before and after transform\n");
hexdump(mctx->block, sizeof(mctx->block));
#endif //DEBUG_MD5
*p++ = 0x80;
if(padding < 0)
{
memset(p, 0x00, padding + sizeof(u64));
md5_transform_helper(mctx);
p = (char *) mctx->block;
padding = 56;
}
memset(p, 0, padding);
#ifdef DEBUG_MD5
hexdump(&mctx->byte_count, sizeof(mctx->byte_count));
#endif //DEBUG_MD5
mctx->block[14] = cpu_to_le32(mctx->byte_count << 3);
mctx->block[15] = cpu_to_le32(mctx->byte_count >> 29);
#ifdef DEBUG_MD5
hexdump(mctx->block, sizeof(mctx->block));
#endif //DEBUG_MD5
//le32_to_cpu_array(mctx->block, (sizeof(mctx->block) -
// sizeof(u64)) / sizeof(u32));
md5_transform(mctx->hash, mctx->block);
local_irq_save(flag);
#ifdef CONFIG_CRYPTO_DEV_DANUBE_DMA
//wait for processing
while(hashs->controlr.BSY)
{
// this will not take long
}
#endif
*((u32 *) out + 0) = le32_to_cpu(hashs->D1R);
*((u32 *) out + 1) = le32_to_cpu(hashs->D2R);
*((u32 *) out + 2) = le32_to_cpu(hashs->D3R);
*((u32 *) out + 3) = le32_to_cpu(hashs->D4R);
hashs->controlr.SM = 0; // switch off again for next dma transfer
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
#ifdef CONFIG_CRYPTO_DEV_DANUBE_DMA
struct dma_device_info *dma_device;
_ifx_deu_device *pDev = ifx_deu;
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
dma_device = pDev->dma_device;
//if(dma_device) dma_device_release(dma_device);
if(dma_device)
{
dma_device_unregister(pDev->dma_device);
dma_device_release(pDev->dma_device);
}
#endif
//cpu_to_le32_array(mctx->hash, sizeof(mctx->hash) / sizeof(u32));
//memcpy(out, mctx->hash, sizeof(mctx->hash));
local_irq_restore(flag);
// Wipe context
memset(mctx, 0, sizeof(*mctx));
}
int main (int argc, char *argv[])
{
uint64_t skip = 0;
uint64_t left = -1;
if (argc >= 2) skip = atoll (argv[1]);
if (argc >= 3) left = atoll (argv[2]);
printf ("Loading Kernel...\n");
const char *filename = KERNEL_SRC;
struct stat s;
if (stat (filename, &s) == -1)
{
fprintf (stderr, "%s: %s in line %d\n", filename, strerror (errno), __LINE__);
return (-1);
}
FILE *fp = fopen (filename, "rb");
if (fp == NULL)
{
fprintf (stderr, "%s: %s in line %d\n", filename, strerror (errno), __LINE__);
return (-1);
}
char *source_buf = (char *) malloc (s.st_size + 1);
if (!fread (source_buf, sizeof (char), s.st_size, fp))
{
fprintf (stderr, "%s: %s in line %d\n", filename, strerror (errno), __LINE__);
return (-1);
}
source_buf[s.st_size] = 0;
fclose (fp);
const char *sourceBuf[] = { source_buf };
const size_t sourceLen[] = { s.st_size + 1 };
printf ("Initializing OpenCL...\n");
cl_platform_id platform;
cl_uint num_devices = 0;
cl_device_id devices[MAX_PLATFORM];
gc_clGetPlatformIDs (1, &platform, NULL);
gc_clGetDeviceIDs (platform, DEV_TYPE, MAX_PLATFORM, devices, &num_devices);
gpu_ctx_t gpu_ctxs[MAX_GPU];
memset (gpu_ctxs, 0, sizeof (gpu_ctxs));
for (cl_uint device_id = 0; device_id < num_devices; device_id++)
{
cl_device_id device = devices[device_id];
cl_context context = gc_clCreateContext (NULL, 1, &device, NULL, NULL);
cl_program program = gc_clCreateProgramWithSource (context, 1, sourceBuf, sourceLen);
gc_clBuildProgram (program, 1, &device, BUILD_OPTS, NULL, NULL);
cl_kernel kernel = gc_clCreateKernel (program, KERNEL_NAME);
cl_command_queue command_queue = gc_clCreateCommandQueue (context, device, 0);
cl_uint max_compute_units;
gc_clGetDeviceInfo (device, CL_DEVICE_MAX_COMPUTE_UNITS, sizeof (max_compute_units), &max_compute_units, NULL);
char device_name[BUFSIZ];
memset (device_name, 0, sizeof (device_name));
gc_clGetDeviceInfo (device, CL_DEVICE_NAME, sizeof (device_name), &device_name, NULL);
printf ("Found new device #%2d: %s, %u compute units\n", device_id, device_name, max_compute_units);
const int num_threads = GPU_THREADS;
const int num_elements = max_compute_units * num_threads * GPU_ACCEL;
/**
* GPU memory
*/
const size_t size_block = num_elements * sizeof (block_t);
const size_t size_results = num_threads * sizeof (uint32_t);
cl_mem d_block = gc_clCreateBuffer (context, CL_MEM_READ_ONLY, size_block, NULL);
cl_mem d_results = gc_clCreateBuffer (context, CL_MEM_WRITE_ONLY, size_results, NULL);
gc_clSetKernelArg (kernel, 0, sizeof (cl_mem), (void *) &d_block);
gc_clSetKernelArg (kernel, 1, sizeof (cl_mem), (void *) &d_results);
/**
* Host memory
*/
block_t *h_block = (block_t *) malloc (size_block);
uint32_t *h_results = (uint32_t *) malloc (size_results);
memset (h_results, 0xff, size_results);
gc_clEnqueueWriteBuffer (command_queue, d_results, CL_TRUE, 0, size_results, h_results, 0, NULL, NULL);
/**
* Buffers for candidates
*/
uint8_t **plains_buf = (uint8_t **) calloc (num_elements * VECT_SIZE, sizeof (uint8_t *));
for (int i = 0; i < num_elements * VECT_SIZE; i++)
{
/* Agreed, this is not nice. But who cares nowadays? */
plains_buf[i] = (uint8_t *) malloc (MAX_LINELEN);
}
size_t *plains_len = (size_t *) calloc (num_elements * VECT_SIZE, sizeof (size_t));
gpu_ctx_t *gpu_ctx = &gpu_ctxs[device_id];
gpu_ctx->context = context;
gpu_ctx->program = program;
gpu_ctx->kernel = kernel;
gpu_ctx->command_queue = command_queue;
gpu_ctx->max_compute_units = max_compute_units;
gpu_ctx->d_block = d_block;
gpu_ctx->d_results = d_results;
gpu_ctx->h_block = h_block;
gpu_ctx->h_results = h_results;
gpu_ctx->num_threads = num_threads;
gpu_ctx->num_elements = num_elements;
gpu_ctx->plains_buf = plains_buf;
gpu_ctx->plains_len = plains_len;
}
/* static salt */
const uint8_t salt_buf[16] =
{
0x97, 0x48, 0x6C, 0xAA,
0x22, 0x5F, 0xE8, 0x77,
0xC0, 0x35, 0xCC, 0x03,
0x73, 0x23, 0x6D, 0x51
};
const size_t salt_len = sizeof (salt_buf);
/* main loop */
printf ("Initialization done, accepting candidates from stdin...\n\n");
cl_uint cur_device_id = 0;
while (!feof (stdin))
{
/* Get new password candidate from stdin */
uint8_t line_buf[MAX_LINELEN];
int cur_c = 0;
int prev_c = 0;
size_t line_len = 0;
for (size_t i = 0; i < MAX_LINELEN - 100; i++) // - 100 = we need some space for salt and padding
{
cur_c = getchar ();
if (cur_c == EOF) break;
if ((prev_c == '\n') && (cur_c == '\0'))
{
line_len--;
break;
}
line_buf[line_len] = cur_c;
line_len++;
prev_c = cur_c;
}
/* chop \r if it exists for some reason (in case user used a dictionary) */
if (line_len >= 2)
{
if ((prev_c == '\r') && (cur_c == '\0')) line_len -= 2;
}
/* skip empty lines */
if (line_len == 0) continue;
/* The following enables distributed computing / resume work */
if (skip)
{
skip--;
continue;
}
if (left)
{
left--;
}
else
{
break;
}
/* Append constant salt */
memcpy (line_buf + line_len, salt_buf, salt_len);
line_len += salt_len;
/* Generate digest out of it */
uint32_t digest[4];
md5_transform ((uint32_t *) line_buf, (uint32_t) line_len, digest);
/* Next garanteed free GPU */
gpu_ctx_t *gpu_ctx = &gpu_ctxs[cur_device_id];
/* Save original buffer in case it cracks it */
memcpy (gpu_ctx->plains_buf[gpu_ctx->num_cached], line_buf, line_len - salt_len);
gpu_ctx->plains_len[gpu_ctx->num_cached] = line_len - salt_len;
/* Next garanteed free memory element on that GPU */
const uint32_t element_div = gpu_ctx->num_cached / 4;
const uint32_t element_mod = gpu_ctx->num_cached % 4;
/* Copy new digest */
gpu_ctx->h_block[element_div].A[element_mod] = digest[0];
gpu_ctx->h_block[element_div].B[element_mod] = digest[1];
gpu_ctx->h_block[element_div].C[element_mod] = digest[2];
gpu_ctx->h_block[element_div].D[element_mod] = digest[3];
gpu_ctx->num_cached++;
/* If memory elements on that GPU are full, switch to the next GPU */
if ((gpu_ctx->num_cached / VECT_SIZE) < gpu_ctx->num_elements) continue;
cur_device_id++;
/* If there is no more GPU left, run the calculation */
if (cur_device_id < num_devices) continue;
/* Fire! */
calc_work (num_devices, gpu_ctxs);
launch_kernel (num_devices, gpu_ctxs);
/* Collecting data has a blocking effect */
check_results (num_devices, gpu_ctxs);
/* Reset buffer state */
for (cl_uint device_id = 0; device_id < num_devices; device_id++)
{
gpu_ctx_t *gpu_ctx = &gpu_ctxs[device_id];
gpu_ctx->num_cached = 0;
}
cur_device_id = 0;
}
/* Final calculation of leftovers */
calc_work (num_devices, gpu_ctxs);
launch_kernel (num_devices, gpu_ctxs);
check_results (num_devices, gpu_ctxs);
return -1;
}
