Skip to content

Commit

Permalink
add ref trace src
Browse files Browse the repository at this point in the history
  • Loading branch information
@dongchen-coder
dongchen-coder committed Jul 8, 2019
1 parent 510c2bd commit f1de749
Show file tree
Hide file tree
Showing 36 changed files with 2,754 additions and 0 deletions.
6 changes: 6 additions & 0 deletions test_facility/Makefile
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,9 @@ BIN_DIR=$(TEST_DIR)/bin
TRACE_SRC_DIR=$(TEST_DIR)/trace_src
TRACE_RESULT_DIR=$(TEST_DIR)/trace_result

TRACE_REF_SRC_DIR=$(TEST_DIR)/ref_trace_src
TRACE_REF_RESULT_DIR=$(TEST_DIR)/ref_trace_result

OPT= $(LLVM_BIN_DIR)/opt
LLC= $(LLVM_BIN_DIR)/llc
CC= $(CLANG_BIN_DIR)/clang
Expand Down Expand Up @@ -105,6 +108,9 @@ trace_run:
$(foreach name, $(all_bench), g++ -std=c++11 $(TRACE_SRC_DIR)/$(name).cpp -O3 -o $(BIN_DIR)/$(name)_trace ;)
$(foreach name, $(all_bench), $(BIN_DIR)/$(name)_trace > $(TRACE_RESULT_DIR)/$(name)_trace_result.txt ;)

trace_ref_run:
$(foreach name, $(all_bench), g++ -std=c++11 $(TRACE_REF_SRC_DIR)/$(name)_ref_arr.cpp -O3 -o $(BIN_DIR)/$(name)_ref_arr_trace ;)
$(foreach name, $(all_bench), $(BIN_DIR)/$(name)_ref_arr_trace > $(TRACE_REF_RESULT_DIR)/$(name)_ref_arr_trace_result.txt ;)

clean:
rm $(BC_DIR)/*.bc
Expand Down
72 changes: 72 additions & 0 deletions test_facility/ref_trace_src/2mm_ref_arr.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,72 @@

#include "../utility/rt.h"
#include "../utility/data_size.h"

#ifdef ORG
#define NI 256
#define NJ 256
#define NK 256
#define NL 256
#elif defined (TX)
#elif defined (FX)
#elif defined (EX)
#endif

#define TMP_OFFSET 0
#define A_OFFSET NI * NJ
#define B_OFFSET NI * NJ + NI * NK
#define D_OFFSET NI * NJ + NI * NK + NJ * NK
#define C_OFFSET NI * NJ + NI * NK + NJ * NK + NI * NL



void mm2_trace(double* tmp, double* A, double* B, double* C, double* D, double alpha, double beta) {

int i, j, k;

for (i = 0; i < NI; i++) {
for (j = 0; j < NJ; j++) {
tmp[i * NJ + j] = 0.0;
rtTmpAccess(TMP_OFFSET + i * NJ + j, 0, 0);
for (k = 0; k < NK; ++k) {
tmp[i * NJ + j] += alpha * A[i * NK + k] * B[k * NJ + j];
rtTmpAccess(A_OFFSET + i * NK + k, 1, 1);
rtTmpAccess(B_OFFSET + k * NJ + j, 2, 2);
rtTmpAccess(TMP_OFFSET + i * NJ + j, 3, 0);
rtTmpAccess(TMP_OFFSET + i * NJ + j, 4, 0);
}
}
}
for (i = 0; i < NI; i++) {
for (j = 0; j < NL; j++) {
D[i * NL + j] *= beta;
rtTmpAccess(D_OFFSET + i * NL + j, 5, 3);
for (k = 0; k < NJ; ++k) {
D[i * NL + j] += tmp[i * NJ + k] * C[k * NL + j];
rtTmpAccess(TMP_OFFSET + i * NJ + k, 6, 0);
rtTmpAccess(C_OFFSET + k * NL + j, 7, 4);
rtTmpAccess(D_OFFSET + i * NL + j, 8, 3);
rtTmpAccess(D_OFFSET + i * NL + j, 9, 3);
}
}
}
}


int main() {

double* tmp = (double*)malloc( NI * NJ * sizeof(double));
double* A = (double*)malloc( NI * NK * sizeof(double));
double* B = (double*)malloc( NK * NJ * sizeof(double));
double* C = (double*)malloc( NJ * NL * sizeof(double));
double* D = (double*)malloc( NI * NL * sizeof(double));
double alpha = 0.1;
double beta = 0.5;

mm2_trace(tmp, A, B, C, D, alpha, beta);

dumpSetSize();

return 0;
}

139 changes: 139 additions & 0 deletions test_facility/ref_trace_src/3mm_ref_arr.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,139 @@

#include "../utility/rt.h"
#include "../utility/data_size.h"

#ifdef ORG
#define NI 256
#define NJ 256
#define NL 256
#define NK 256
#define NM 256
#elif defined (TX)
#define NI 512
#define NJ 256
#define NL 512
#define NK 256
#define NM 256
#elif defined (FX)
#define NI 1024
#define NJ 256
#define NL 1024
#define NK 256
#define NM 256
#elif defined (EX)
#define NI 2048
#define NJ 256
#define NL 2048
#define NK 256
#define NM 256
#endif

#define A_OFFSET 0
#define B_OFFSET NK * NI
#define C_OFFSET NK * NI + NK * NJ
#define D_OFFSET NK * NI + NK * NJ + NJ * NM
#define E_OFFSET NK * NI + NK * NJ + NJ * NM + NM * NL
#define F_OFFSET NK * NI + NK * NJ + NJ * NM + NM * NL + NI * NJ
#define G_OFFSET NK * NI + NK * NJ + NJ * NM + NM * NL + NI * NJ + NJ * NL


void mm3_cpu_trace(int *A, int *B, int *C, int *D, int *E, int *F, int *G) {

int i, j, k;

/* E := A*B */
for (i = 0; i < NI; i++)
{
for (j = 0; j < NJ; j++)
{
E[i * NJ + j] = 0;

rtTmpAccess(E_OFFSET + i * NJ + j, 0, 0);

for (k = 0; k < NK; ++k)
{
E[i * NJ + j] += A[i * NK + k] * B[k * NJ + j];
rtTmpAccess(A_OFFSET + i * NK + k, 1, 1);
rtTmpAccess(B_OFFSET + k * NJ + j, 2, 2);
rtTmpAccess(E_OFFSET + i * NJ + j, 3, 0);
rtTmpAccess(E_OFFSET + i * NJ + j, 4, 0);
}
}
}

/* F := C*D */
for (i = 0; i < NJ; i++)
{
for (j = 0; j < NL; j++)
{
F[i * NL + j] = 0;

rtTmpAccess(F_OFFSET + i * NL + j, 5, 3);

for (k = 0; k < NM; ++k)
{
F[i * NL + j] += C[i * NM + k] * D[k * NL + j];
rtTmpAccess(C_OFFSET + i * NM + k, 6, 4);
rtTmpAccess(D_OFFSET + k * NL + j, 7, 5);
rtTmpAccess(F_OFFSET + i * NL + j, 8, 3);
rtTmpAccess(F_OFFSET + i * NL + j, 9, 3);
}
}
}

/* G := E*F */
for (i = 0; i < NI; i++)
{
for (j = 0; j < NL; j++)
{
G[i * NL + j] = 0;

rtTmpAccess(G_OFFSET + i * NL + j, 10, 6);

for (k = 0; k < NJ; ++k)
{
G[i * NL + j] += E[i * NJ + k] * F[k * NL + j];
rtTmpAccess(E_OFFSET + i * NJ + k, 11, 0);
rtTmpAccess(F_OFFSET + k * NL + j, 12, 3);
rtTmpAccess(G_OFFSET + i * NL + j, 13, 6);
rtTmpAccess(G_OFFSET + i * NL + j, 14, 6);
}
}
}

return;
}



int main() {
int* a = (int*)malloc( NK * NI * sizeof(int) );
int* b = (int*)malloc( NK * NJ * sizeof(int) );
int* c = (int*)malloc( NJ * NM * sizeof(int) );
int* d = (int*)malloc( NM * NL * sizeof(int) );
int* e = (int*)malloc( NI * NJ * sizeof(int) );
int* f = (int*)malloc( NJ * NL * sizeof(int) );
int* g = (int*)malloc( NI * NL * sizeof(int) );

for (int i = 0; i < NK * NI; i++) {
a[i] = i % 256;
}

for (int i = 0; i < NK * NJ; i++) {
b[i] = i % 256;
}

for (int i = 0; i < NJ * NM; i++) {
c[i] = i % 256;
}

for (int i = 0; i < NM * NL; i++) {
d[i] = i % 256;
}

mm3_cpu_trace(a, b, c, d, e, f, g);

dumpSetSize();

return 0;
}
Binary file added test_facility/ref_trace_src/a.out
View file
Binary file not shown.
119 changes: 119 additions & 0 deletions test_facility/ref_trace_src/adi_ref_arr.cpp
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,119 @@

#include "../utility/rt.h"
#include "../utility/data_size.h"

#ifdef ORG
#define N 1024
#define TSTEPS 10
#elif defined (TX)
#elif defined (FX)
#elif defined (EX)
#endif

#define P_OFFSET 0
#define Q_OFFSET 0 + N * N
#define V_OFFSET 0 + N * N + N * N
#define U_OFFSET 0 + N * N + N * N + N * N

void adi_trace(double* p, double* q, double* v, double* u) {

int t, i, j;
double DX, DY, DT, B1, B2, mul1, mul2, a, b, c, d, e, f;

DX = 1.0 / N;
DY = 1.0 / N;
DT = 1.0 / TSTEPS;
B1 = 2.0;
B2 = 1.0;
mul1 = B1 * DT / (DX * DX);
mul2 = B2 * DT / (DY * DY);

a = -mul1 / 2.0;
b = 1.0 + mul1;
c = a;
d = -mul2 / 2.0;
e = 1.0 + mul2;
f = d;

for (t=1; t <= TSTEPS; t++) {
//Column Sweep
for (i=1; i< N-1; i++) {
v[0 * N + i] = 1.0;
p[i * N + 0] = 0.0;
q[i * N + 0] = v[0 * N + i];
rtTmpAccess(V_OFFSET + 0 * N + i, 0, 0);
rtTmpAccess(P_OFFSET + i * N + 0, 1, 1);
rtTmpAccess(V_OFFSET + 0 * N + i, 2, 0);
rtTmpAccess(Q_OFFSET + i * N + 0, 3, 2);
for (j=1; j< N-1; j++) {
p[i * N + j] = -c / (a*p[i * N + j-1]+b);
q[i * N + j] = (-d*u[j * N + i-1]+(1.0 + 2.0 * d)*u[j * N + i] - f*u[j * N + i+1]-a*q[i * N + j-1])/(a*p[i * N + j-1]+b);
rtTmpAccess(P_OFFSET + i * N + j-1, 4, 1);
rtTmpAccess(P_OFFSET + i * N + j, 5, 1);
rtTmpAccess(U_OFFSET + j * N + i-1, 6, 3);
rtTmpAccess(U_OFFSET + j * N + i, 7, 3);
rtTmpAccess(U_OFFSET + j * N + i+1, 8, 3);
rtTmpAccess(Q_OFFSET + i * N + j-1, 9, 2);
rtTmpAccess(P_OFFSET + i * N + j-1, 10, 1);
rtTmpAccess(Q_OFFSET + i * N + j, 11, 2);
}
v[(N - 1) * N + i] = 1.0;
rtTmpAccess(V_OFFSET + (N - 1) * N + i, 12, 0);
for (j= N-2; j>=1; j--) {
v[j * N + i] = p[i * N + j] * v[(j+1) * N + i] + q[i * N + j];
rtTmpAccess(P_OFFSET + i * N + j, 13, 1);
rtTmpAccess(V_OFFSET + (j+1) * N + i, 14, 0);
rtTmpAccess(Q_OFFSET + i * N + j, 15, 2);
rtTmpAccess(V_OFFSET + j * N + i, 16, 0);
}
}
//Row Sweep
for (i=1; i < N - 1; i++) {
u[i * N + 0] = 1.0;
p[i * N + 0] = 0.0;
q[i * N + 0] = u[i * N + 0];
rtTmpAccess(U_OFFSET + i * N + 0, 17, 3);
rtTmpAccess(P_OFFSET + i * N + 0, 18, 1);
rtTmpAccess(U_OFFSET + i * N + 0, 19, 3);
rtTmpAccess(Q_OFFSET + i * N + 0, 20, 2);
for (j=1; j< N - 1; j++) {
p[i * N + j] = -f / (d*p[i * N + j-1]+e);
q[i * N + j] = (-a*v[(i-1) * N + j]+(1.0 + 2.0 * a)*v[i * N + j] - c*v[(i+1) * N + j]-d*q[i * N + j-1])/(d*p[i * N + j-1]+e);
rtTmpAccess(P_OFFSET + i * N + j-1, 21, 1);
rtTmpAccess(P_OFFSET + i * N + j, 22, 1);
rtTmpAccess(V_OFFSET + (i-1) * N + j, 23, 0);
rtTmpAccess(V_OFFSET + i * N + j, 24, 0);
rtTmpAccess(V_OFFSET + (i+1) * N + j, 25, 0);
rtTmpAccess(Q_OFFSET + i * N + j-1, 26, 2);
rtTmpAccess(P_OFFSET + i * N + j-1, 27, 1);
rtTmpAccess(Q_OFFSET + i * N + j, 28, 2);
}
u[i * N + N - 1 ] = 1.0;
rtTmpAccess(U_OFFSET + i * N + N - 1, 29, 3);
for (j= N - 2; j>=1; j--) {
u[i * N + j] = p[i * N + j] * u[i * N + j+1] + q[i * N + j];
rtTmpAccess(P_OFFSET + i * N + j, 30, 1);
rtTmpAccess(U_OFFSET + i * N + j+1, 31, 3);
rtTmpAccess(Q_OFFSET + i * N + j, 32, 2);
rtTmpAccess(U_OFFSET + i * N + j, 33, 3);
}
}
}

}

int main() {

double * p = (double *) malloc(N * N * sizeof(double));
double * q = (double *) malloc(N * N * sizeof(double));
double * v = (double *) malloc(N * N * sizeof(double));
double * u = (double *) malloc(N * N * sizeof(double));

adi_trace(p, q, v, u);

dumpSetSize();

return 0;
}


Loading

0 comments on commit f1de749

Please sign in to comment.