void ec_encode_data_sse(int len, int k, int rows, unsigned char *g_tbls, unsigned char **data,
unsigned char **coding)
{
if (len < 16) {
ec_encode_data_base(len, k, rows, g_tbls, data, coding);
return;
}
while (rows >= 4) {
gf_4vect_dot_prod_sse(len, k, g_tbls, data, coding);
g_tbls += 4 * k * 32;
coding += 4;
rows -= 4;
}
switch (rows) {
case 3:
gf_3vect_dot_prod_sse(len, k, g_tbls, data, coding);
break;
case 2:
gf_2vect_dot_prod_sse(len, k, g_tbls, data, coding);
break;
case 1:
gf_vect_dot_prod_sse(len, k, g_tbls, data, *coding);
break;
case 0:
break;
}
}
int main(int argc, char *argv[])
{
int i,j, rtest, srcs;
void *buf;
u8 g1[TEST_SOURCES], g2[TEST_SOURCES], g_tbls[2*TEST_SOURCES*32];
u8 *dest1, *dest2, *dest_ref1, *dest_ref2, *dest_ptrs[2];
u8 *buffs[TEST_SOURCES];
int align, size;
unsigned char *efence_buffs[TEST_SOURCES];
unsigned int offset;
u8 *ubuffs[TEST_SOURCES];
u8 *udest_ptrs[2];
printf("gf_2vect_dot_prod_sse: %dx%d ", TEST_SOURCES, TEST_LEN);
mk_gf_field();
// Allocate the arrays
for(i=0; i<TEST_SOURCES; i++){
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest2 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref2 = buf;
dest_ptrs[0] = dest1;
dest_ptrs[1] = dest2;
// Test of all zeros
for(i=0; i<TEST_SOURCES; i++)
memset(buffs[i], 0, TEST_LEN);
memset(dest1, 0, TEST_LEN);
memset(dest2, 0, TEST_LEN);
memset(dest_ref1, 0, TEST_LEN);
memset(dest_ref2, 0, TEST_LEN);
memset(g1, 2, TEST_SOURCES);
memset(g2, 1, TEST_SOURCES);
for(i=0; i<TEST_SOURCES; i++){
gf_vect_mul_init(g1[i], &g_tbls[i*32]);
gf_vect_mul_init(g2[i], &g_tbls[32*TEST_SOURCES + i*32]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32*TEST_SOURCES], buffs, dest_ref2);
gf_2vect_dot_prod_sse(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)){
printf("Fail zero vect_dot_prod_sse test1\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_sse:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)){
printf("Fail zero vect_dot_prod_sse test2\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_sse:");
dump(dest2, 25);
return -1;
}
putchar('.');
// Rand data test
for(rtest=0; rtest<RANDOMS; rtest++){
for(i=0; i<TEST_SOURCES; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
for (i=0; i<TEST_SOURCES; i++){
g1[i] = rand();
g2[i] = rand();
}
for(i=0; i<TEST_SOURCES; i++){
gf_vect_mul_init(g1[i], &g_tbls[i*32]);
gf_vect_mul_init(g2[i], &g_tbls[(32*TEST_SOURCES) + (i*32)]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32*TEST_SOURCES], buffs, dest_ref2);
gf_2vect_dot_prod_sse(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)){
printf("Fail rand 2vect_dot_prod_sse test1 %d\n", rtest);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_sse:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)){
printf("Fail rand 2vect_dot_prod_sse test2 %d\n", rtest);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_sse:");
dump(dest2, 25);
return -1;
}
putchar('.');
}
// Rand data test with varied parameters
for(rtest=0; rtest<RANDOMS; rtest++){
for (srcs = TEST_SOURCES; srcs > 0; srcs--){
for(i=0; i<srcs; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
for (i=0; i<srcs; i++){
g1[i] = rand();
g2[i] = rand();
}
for(i=0; i<srcs; i++){
gf_vect_mul_init(g1[i], &g_tbls[i*32]);
gf_vect_mul_init(g2[i], &g_tbls[(32*srcs) + (i*32)]);
}
gf_vect_dot_prod_base(TEST_LEN, srcs, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, srcs, &g_tbls[32*srcs], buffs, dest_ref2);
gf_2vect_dot_prod_sse(TEST_LEN, srcs, g_tbls, buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)){
printf("Fail rand 2vect_dot_prod_sse test1 srcs=%d\n", srcs);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_sse:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)){
printf("Fail rand 2vect_dot_prod_sse test2 srcs=%d\n", srcs);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_sse:");
dump(dest2, 25);
return -1;
}
putchar('.');
}
}
// Run tests at end of buffer for Electric Fence
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
for(size=EFENCE_TEST_MIN_SIZE; size<=TEST_SIZE; size+=align){
for(i=0; i<TEST_SOURCES; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
for(i=0; i<TEST_SOURCES; i++) // Line up TEST_SIZE from end
efence_buffs[i] = buffs[i] + TEST_LEN - size;
for (i=0; i<TEST_SOURCES; i++){
g1[i] = rand();
g2[i] = rand();
}
for(i=0; i<TEST_SOURCES; i++){
gf_vect_mul_init(g1[i], &g_tbls[i*32]);
gf_vect_mul_init(g2[i], &g_tbls[(32*TEST_SOURCES) + (i*32)]);
}
gf_vect_dot_prod_base(size, TEST_SOURCES, &g_tbls[0], efence_buffs, dest_ref1);
gf_vect_dot_prod_base(size, TEST_SOURCES, &g_tbls[32*TEST_SOURCES], efence_buffs, dest_ref2);
gf_2vect_dot_prod_sse(size, TEST_SOURCES, g_tbls, efence_buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, size)){
printf("Fail rand 2vect_dot_prod_sse test1 %d\n", rtest);
dump_matrix(efence_buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, align);
printf("dprod_sse:");
dump(dest1, align);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, size)){
printf("Fail rand 2vect_dot_prod_sse test2 %d\n", rtest);
dump_matrix(efence_buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, align);
printf("dprod_sse:");
dump(dest2, align);
return -1;
}
putchar('.');
}
// Test rand ptr alignment if available
for(rtest=0; rtest<RANDOMS; rtest++){
size = (TEST_LEN - PTR_ALIGN_CHK_B) & ~15;
srcs = rand() % TEST_SOURCES;
if (srcs == 0)
continue;
offset = (PTR_ALIGN_CHK_B != 0) ? 1 : PTR_ALIGN_CHK_B;
// Add random offsets
for(i=0; i<srcs; i++)
ubuffs[i] = buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
udest_ptrs[0] = dest1 + (rand() & (PTR_ALIGN_CHK_B - offset));
udest_ptrs[1] = dest2 + (rand() & (PTR_ALIGN_CHK_B - offset));
memset(dest1, 0, TEST_LEN); // zero pad to check write-over
memset(dest2, 0, TEST_LEN);
for(i=0; i<srcs; i++)
for(j=0; j<size; j++)
ubuffs[i][j] = rand();
for (i=0; i<srcs; i++){
g1[i] = rand();
g2[i] = rand();
}
for(i=0; i<srcs; i++){
gf_vect_mul_init(g1[i], &g_tbls[i*32]);
gf_vect_mul_init(g2[i], &g_tbls[(32*srcs) + (i*32)]);
}
gf_vect_dot_prod_base(size, srcs, &g_tbls[0], ubuffs, dest_ref1);
gf_vect_dot_prod_base(size, srcs, &g_tbls[32*srcs], ubuffs, dest_ref2);
gf_2vect_dot_prod_sse(size, srcs, g_tbls, ubuffs, udest_ptrs);
if (memcmp(dest_ref1, udest_ptrs[0], size)){
printf("Fail rand 2vect_dot_prod_sse test ualign srcs=%d\n", srcs);
dump_matrix(ubuffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_sse:");
dump(udest_ptrs[0], 25);
return -1;
}
if (memcmp(dest_ref2, udest_ptrs[1], size)){
printf("Fail rand 2vect_dot_prod_sse test ualign srcs=%d\n", srcs);
dump_matrix(ubuffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_sse:");
dump(udest_ptrs[1], 25);
return -1;
}
// Confirm that padding around dests is unchanged
memset(dest_ref1, 0, PTR_ALIGN_CHK_B); // Make reference zero buff
offset = udest_ptrs[0] - dest1;
if (memcmp(dest1, dest_ref1, offset)){
printf("Fail rand ualign pad1 start\n");
return -1;
}
if (memcmp(dest1 + offset + size, dest_ref1, PTR_ALIGN_CHK_B - offset)){
printf("Fail rand ualign pad1 end\n");
return -1;
}
offset = udest_ptrs[1] - dest2;
if (memcmp(dest2, dest_ref1, offset)){
printf("Fail rand ualign pad2 start\n");
return -1;
}
if (memcmp(dest2 + offset + size, dest_ref1, PTR_ALIGN_CHK_B - offset)){
printf("Fail rand ualign pad2 end\n");
return -1;
}
putchar('.');
}
// Test all size alignment
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
for(size=TEST_LEN; size>15; size-=align){
srcs = TEST_SOURCES;
for(i=0; i<srcs; i++)
for(j=0; j<size; j++)
buffs[i][j] = rand();
for (i=0; i<srcs; i++){
g1[i] = rand();
g2[i] = rand();
}
for(i=0; i<srcs; i++){
gf_vect_mul_init(g1[i], &g_tbls[i*32]);
gf_vect_mul_init(g2[i], &g_tbls[(32*srcs) + (i*32)]);
}
gf_vect_dot_prod_base(size, srcs, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(size, srcs, &g_tbls[32*srcs], buffs, dest_ref2);
gf_2vect_dot_prod_sse(size, srcs, g_tbls, buffs, dest_ptrs);
if (memcmp(dest_ref1, dest_ptrs[0], size)){
printf("Fail rand 2vect_dot_prod_sse test ualign len=%d\n", size);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_sse:");
dump(dest_ptrs[0], 25);
return -1;
}
if (memcmp(dest_ref2, dest_ptrs[1], size)){
printf("Fail rand 2vect_dot_prod_sse test ualign len=%d\n", size);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_sse:");
dump(dest_ptrs[1], 25);
return -1;
}
}
printf("Pass\n");
return 0;
}
int main(int argc, char *argv[])
{
int i,j;
void *buf;
u8 g1[TEST_SOURCES], g2[TEST_SOURCES], g_tbls[2*TEST_SOURCES*32];
u8 *dest1, *dest2, *dest_ref1, *dest_ref2, *dest_ptrs[2];
u8 *buffs[TEST_SOURCES];
struct perf start, stop;
printf("gf_2vect_dot_prod_sse: %dx%d\n", TEST_SOURCES, TEST_LEN);
mk_gf_field();
// Allocate the arrays
for(i=0; i<TEST_SOURCES; i++){
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest2 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref2 = buf;
dest_ptrs[0] = dest1;
dest_ptrs[1] = dest2;
// Performance test
for(i=0; i<TEST_SOURCES; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
memset(dest1, 0, TEST_LEN);
memset(dest2, 0, TEST_LEN);
memset(dest_ref1, 0, TEST_LEN);
memset(dest_ref2, 0, TEST_LEN);
for (i=0; i<TEST_SOURCES; i++){
g1[i] = rand();
g2[i] = rand();
}
for(j=0; j<TEST_SOURCES; j++){
gf_vect_mul_init(g1[j], &g_tbls[j*32]);
gf_vect_mul_init(g2[j], &g_tbls[(32*TEST_SOURCES) + (j*32)]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32*TEST_SOURCES], buffs, dest_ref2);
#ifdef DO_REF_PERF
perf_start(&start);
for (i=0; i<TEST_LOOPS/100; i++){
for(j=0; j<TEST_SOURCES; j++){
gf_vect_mul_init(g1[j], &g_tbls[j*32]);
gf_vect_mul_init(g2[j], &g_tbls[(32*TEST_SOURCES) + (j*32)]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32*TEST_SOURCES], buffs, dest_ref2);
}
perf_stop(&stop);
printf("gf_2vect_dot_prod_base" TEST_TYPE_STR ": ");
perf_print(stop,start, (long long)TEST_LEN*(TEST_SOURCES+2)*i);
#endif
gf_2vect_dot_prod_sse(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
perf_start(&start);
for (i=0; i<TEST_LOOPS; i++) {
for (j=0; j<TEST_SOURCES; j++){
gf_vect_mul_init(g1[j], &g_tbls[j*32]);
gf_vect_mul_init(g2[j], &g_tbls[(32*TEST_SOURCES) + (j*32)]);
}
gf_2vect_dot_prod_sse(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
}
perf_stop(&stop);
printf("gf_2vect_dot_prod_sse" TEST_TYPE_STR ": ");
perf_print(stop,start, (long long)TEST_LEN*(TEST_SOURCES+2)*i);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)){
printf("Fail perf vect_dot_prod_sse test1\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_sse:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)){
printf("Fail perf vect_dot_prod_sse test2\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_sse:");
dump(dest2, 25);
return -1;
}
printf("pass perf check\n");
return 0;
}