Clean up canonical mt benchmarks and reorganize extra versions in /mt. All versions support support at least 1/2/4 threads.

10 files changed, 92 insertions, 150 deletions

diff --git a/benchmarks/Makefile b/benchmarks/Makefile
index f8db5b9..3346c7c 100644
--- a/benchmarks/Makefile
+++ b/benchmarks/Makefile
@@ -26,8 +26,8 @@ bmarks = \
 	dhrystone \
 	spmv \
 	mt-vvadd \
+	mt-matmul \
 	#vec-fft \
-	#mt-matmul \
 	#vec-vvadd \
 	#vec-cmplxmult \
 	#vec-matmul \
@@ -38,7 +38,6 @@ bmarks_host = \
 	towers \
 	vvadd \
 	multiply \
-	mm \
 	spmv \
 	vec-vvadd \
 	vec-cmplxmult \
diff --git a/benchmarks/common/util.h b/benchmarks/common/util.h
index 6c4f963..638f024 100644
--- a/benchmarks/common/util.h
+++ b/benchmarks/common/util.h
@@ -113,4 +113,15 @@ static void __attribute__((noinline)) barrier(int ncores)
 #include "encoding.h"
 #endif
 
+#define stringify_1(s) #s
+#define stringify(s) stringify_1(s)
+#define stats(code, iter) do { \
+    unsigned long _c = -rdcycle(), _i = -rdinstret(); \
+    code; \
+    _c += rdcycle(), _i += rdinstret(); \
+    if (cid == 0) \
+      printf("\n%s: %ld cycles, %ld.%ld cycles/iter, %ld.%ld CPI\n", \
+             stringify(code), _c, _c/iter, 10*_c/iter%10, _c/_i, 10*_c/_i%10); \
+  } while(0)
+
 #endif //__UTIL_H
diff --git a/benchmarks/mt-matmul/bmark.mk b/benchmarks/mt-matmul/bmark.mk
index 4b7fcb7..6a7140f 100644
--- a/benchmarks/mt-matmul/bmark.mk
+++ b/benchmarks/mt-matmul/bmark.mk
@@ -10,6 +10,7 @@
 
 mt_matmul_c_src = \
 	mt-matmul.c \
+	matmul.c \
 	syscalls.c \
 
 mt_matmul_riscv_src = \
diff --git a/benchmarks/mt-matmul/dataset.h b/benchmarks/mt-matmul/dataset.h
index dde3ee4..2c59a33 100644
--- a/benchmarks/mt-matmul/dataset.h
+++ b/benchmarks/mt-matmul/dataset.h
@@ -1,9 +1,12 @@
+#ifndef __DATASET_H
+#define __DATASET_H
 
 #define ARRAY_SIZE 1024 
 
-
 #define DIM_SIZE 32 
 
+typedef double data_t;
+
 static data_t input1_data[ARRAY_SIZE] = 
 {
     0,   3,   2,   0,   3,   1,   0,   3,   2,   3,   2,   0,   3,   3,   1,   2,   3,   0,   0,   1, 
@@ -172,3 +175,5 @@ static data_t verify_data[ARRAY_SIZE] =
    69,  73,  94,  89
 };
 
+
+#endif //__DATASET_H
diff --git a/benchmarks/mt-matmul/matmul.c b/benchmarks/mt-matmul/matmul.c
new file mode 100644
index 0000000..95fbe03
--- /dev/null
+++ b/benchmarks/mt-matmul/matmul.c
@@ -0,0 +1,20 @@
+#include "dataset.h"
+
+//--------------------------------------------------------------------------
+// single-thread, naive version
+//
+void __attribute__((noinline)) matmul(const int coreid, const int ncores, const int lda,  const data_t A[], const data_t B[], data_t C[] )
+{
+   int i, j, k;
+  
+   for ( i = 0; i < lda; i++ )
+   {
+      for ( j = 0; j < lda; j++ )  
+      {
+         for ( k = coreid; k < lda; k+=ncores ) 
+         {
+            C[i + j*lda] += A[j*lda + k] * B[k*lda + i];
+         }
+      }
+   }
+}
diff --git a/benchmarks/mt-matmul/mt-matmul.c b/benchmarks/mt-matmul/mt-matmul.c
index 1584a5d..2353962 100644
--- a/benchmarks/mt-matmul/mt-matmul.c
+++ b/benchmarks/mt-matmul/mt-matmul.c
@@ -25,62 +25,20 @@
 //--------------------------------------------------------------------------
 // Input/Reference Data
 
-typedef double data_t;
 #include "dataset.h"
  
-  
+
 //--------------------------------------------------------------------------
 // Basic Utilities and Multi-thread Support
 
-__thread unsigned long coreid;
-
 #include "util.h"
+
    
-#define stringify_1(s) #s
-#define stringify(s) stringify_1(s)
-#define stats(code) do { \
-    unsigned long _c = -rdcycle(), _i = -rdinstret(); \
-    code; \
-    _c += rdcycle(), _i += rdinstret(); \
-    if (coreid == 0) \
-      printf("%s: %ld cycles, %ld.%ld cycles/iter, %ld.%ld CPI\n", \
-             stringify(code), _c, _c/DIM_SIZE/DIM_SIZE/DIM_SIZE, 10*_c/DIM_SIZE/DIM_SIZE/DIM_SIZE%10, _c/_i, 10*_c/_i%10); \
-  } while(0)
- 
 //--------------------------------------------------------------------------
 // matmul function
  
-// single-thread, naive version
-void __attribute__((noinline)) matmul_naive(const int lda,  const data_t A[], const data_t B[], data_t C[] )
-{
-   int i, j, k;
-
-   if (coreid > 0)
-      return;
-  
-   for ( i = 0; i < lda; i++ )
-      for ( j = 0; j < lda; j++ )  
-      {
-         for ( k = 0; k < lda; k++ ) 
-         {
-            C[i + j*lda] += A[j*lda + k] * B[k*lda + i];
-         }
-      }
-
-}
- 
-
+ extern void __attribute__((noinline)) matmul(const int coreid, const int ncores, const int lda,  const data_t A[], const data_t B[], data_t C[] );
 
-void __attribute__((noinline)) matmul(const int lda,  const data_t A[], const data_t B[], data_t C[] )
-{
-   
-   // ***************************** //
-   // **** ADD YOUR CODE HERE ***** //
-   // ***************************** //
-   //
-   // feel free to make a separate function for MI and MSI versions.
- 
-}
 
 //--------------------------------------------------------------------------
 // Main
@@ -90,46 +48,16 @@ void __attribute__((noinline)) matmul(const int lda,  const data_t A[], const da
   
 void thread_entry(int cid, int nc)
 {
-   coreid = cid;
-
-   // static allocates data in the binary, which is visible to both threads
    static data_t results_data[ARRAY_SIZE];
 
-
-   // Execute the provided, naive matmul
-   barrier(nc);
-   stats(matmul_naive(DIM_SIZE, input1_data, input2_data, results_data); barrier(nc));
+   stats(matmul(cid, nc, DIM_SIZE, input1_data, input2_data, results_data); barrier(nc), DIM_SIZE/DIM_SIZE/DIM_SIZE);
  
-   
-   // verify
    int res = verifyDouble(ARRAY_SIZE, results_data, verify_data);
-   if (res)
-      exit(res);
-
-#if 0
-   // clear results from the first trial
-   size_t i;
-   if (coreid == 0) 
-      for (i=0; i < ARRAY_SIZE; i++)
-         results_data[i] = 0;
-   barrier(nc);
 
-   
-   // Execute your faster matmul
-   barrier(nc);
-   stats(matmul(DIM_SIZE, input1_data, input2_data, results_data); barrier(nc));
- 
 #ifdef DEBUG
    printArray("results:", ARRAY_SIZE, results_data);
    printArray("verify :", ARRAY_SIZE, verify_data);
 #endif
-   
-   // verify
-   res = verify(ARRAY_SIZE, results_data, verify_data);
-   if (res)
-      exit(res);
-   barrier(nc);
-#endif
 
-   exit(0);
+   exit(res);
 }
diff --git a/benchmarks/mt-vvadd/bmark.mk b/benchmarks/mt-vvadd/bmark.mk
index 72b2d34..ff969c1 100644
--- a/benchmarks/mt-vvadd/bmark.mk
+++ b/benchmarks/mt-vvadd/bmark.mk
@@ -10,6 +10,7 @@
 
 mt_vvadd_c_src = \
 	mt-vvadd.c \
+	vvadd.c \
 	syscalls.c \
 
 mt_vvadd_riscv_src = \
diff --git a/benchmarks/mt-vvadd/dataset.h b/benchmarks/mt-vvadd/dataset.h
index ce9f936..51f25df 100644
--- a/benchmarks/mt-vvadd/dataset.h
+++ b/benchmarks/mt-vvadd/dataset.h
@@ -1,6 +1,10 @@
+#ifndef __DATASET_H
+#define __DATASET_H
 
 #define DATA_SIZE 1000 
 
+typedef double data_t;
+
 static data_t input1_data[DATA_SIZE] = 
 {
   0.00, 15.00, 10.00, 3.00, 14.00, 6.00, 2.00, 18.00, 11.00, 15.00, 11.00, 0.00, 17.00, 16.00, 7.00, 13.00, 18.00, 2.00, 2.00, 5.00, 
@@ -163,3 +167,5 @@ static data_t verify_data[DATA_SIZE] =
   32.00, 21.00, 26.00, 13.00, 6.00, 24.00, 22.00, 27.00, 17.00, 26.00, 19.00, 21.00, 19.00, 9.00, 21.00, 25.00, 17.00, 27.00, 15.00, 18.00
 };
 
+
+#endif //__DATASET_H
diff --git a/benchmarks/mt-vvadd/mt-vvadd.c b/benchmarks/mt-vvadd/mt-vvadd.c
index 2116115..48eae6a 100644
--- a/benchmarks/mt-vvadd/mt-vvadd.c
+++ b/benchmarks/mt-vvadd/mt-vvadd.c
@@ -24,49 +24,20 @@
 //--------------------------------------------------------------------------
 // Input/Reference Data
 
-typedef double data_t;
 #include "dataset.h"
  
   
 //--------------------------------------------------------------------------
 // Basic Utilities and Multi-thread Support
 
-__thread unsigned long coreid;
-
 #include "util.h"
    
-#define stringify_1(s) #s
-#define stringify(s) stringify_1(s)
-#define stats(code) do { \
-    unsigned long _c = -rdcycle(), _i = -rdinstret(); \
-    code; \
-    _c += rdcycle(), _i += rdinstret(); \
-    if (coreid == 0) \
-      printf("%s: %ld cycles, %ld.%ld cycles/iter, %ld.%ld CPI\n", \
-             stringify(code), _c, _c/DATA_SIZE, 10*_c/DATA_SIZE%10, _c/_i, 10*_c/_i%10); \
-  } while(0)
  
 //--------------------------------------------------------------------------
 // vvadd function
 
-//perform in-place vvadd
-void __attribute__((noinline)) vvadd(int ncores, size_t n, data_t* __restrict__ x, const data_t* __restrict__ y)
-{
-   size_t i;
-
-   // interleave accesses
-   for (i = coreid; i < n; i+=ncores)
-   {
-      x[i] = x[i] + y[i];
-   }
-}
+extern void __attribute__((noinline)) vvadd(int coreid, int ncores, size_t n, const data_t* x, const data_t* y, data_t* z);
 
-void __attribute__((noinline)) vvadd_opt(size_t n, data_t* __restrict__ x, const data_t* __restrict__ y)
-{
-   // ***************************** //
-   // **** ADD YOUR CODE HERE ***** //
-   // ***************************** //
-}
 
 //--------------------------------------------------------------------------
 // Main
@@ -76,57 +47,41 @@ void __attribute__((noinline)) vvadd_opt(size_t n, data_t* __restrict__ x, const
   
 void thread_entry(int cid, int nc)
 {
-   coreid = cid;
-
    // static allocates data in the binary, which is visible to both threads
    static data_t results_data[DATA_SIZE];
    
-   // because we're going to perform an in-place vvadd (and we're going to run
-   // it a couple of times) let's copy the input data to a temporary results
-   // array
-   
-   size_t i;
-   if (coreid == 0)
-   {
-      for (i = 0; i < DATA_SIZE; i++)
-         results_data[i] = input1_data[i];
-   }
-
-
-   // Execute the provided, terrible vvadd
+   // First do out-of-place vvadd
    barrier(nc);
-   stats(vvadd(nc, DATA_SIZE, results_data, input2_data); barrier(nc));
+   stats(vvadd(cid, nc, DATA_SIZE, input1_data, input2_data, results_data); barrier(nc), DATA_SIZE);
  
-   
-   // verify
-   int res = verifyDouble(DATA_SIZE, results_data, verify_data);
-   if (res)
-      exit(res);
-
-#if 0
-   // reset results from the first trial
-   if (coreid == 0) 
-   {
-      for (i=0; i < DATA_SIZE; i++)
-         results_data[i] = input1_data[i];
+   if(cid == 0) {
+//#ifdef DEBUG
+     printDoubleArray("out-of-place results: ", DATA_SIZE, results_data);
+     printDoubleArray("out-of-place verify : ", DATA_SIZE, verify_data);
+//#endif
+     int res = verifyDouble(DATA_SIZE, results_data, verify_data);
+     if(res) exit(res);
    }
-   barrier(nc);
 
-   // Execute your faster vvadd
+   // Second do in-place vvadd
+   // Copying input
+   size_t i;
+   if(cid == 0) {
+     for (i = 0; i < DATA_SIZE; i++)
+           results_data[i] = input1_data[i];
+   }
    barrier(nc);
-   stats(vvadd_opt(DATA_SIZE, results_data, input2_data); barrier(nc));
-
+   stats(vvadd(cid, nc, DATA_SIZE, results_data, input2_data, results_data); barrier(nc), DATA_SIZE);
+ 
+   if(cid == 0) {
 #ifdef DEBUG
-   printDoubleArray("results: ", DATA_SIZE, results_data);
-   printDoubleArray("verify : ", DATA_SIZE, verify_data);
+     printDoubleArray("in-place results: ", DATA_SIZE, results_data);
+     printDoubleArray("in-place verify : ", DATA_SIZE, verify_data);
 #endif
+     int res = verifyDouble(DATA_SIZE, results_data, verify_data);
+     if(res) exit(res);
+   }
    
-   // verify
-   res = verifyDouble(DATA_SIZE, results_data, verify_data);
-   if (res)
-      exit(res);
    barrier(nc);
-#endif
-
    exit(0);
 }
diff --git a/benchmarks/mt-vvadd/vvadd.c b/benchmarks/mt-vvadd/vvadd.c
new file mode 100644
index 0000000..8f4d43f
--- /dev/null
+++ b/benchmarks/mt-vvadd/vvadd.c
@@ -0,0 +1,16 @@
+#include "stdlib.h"
+#include "dataset.h"
+
+//--------------------------------------------------------------------------
+// vvadd function
+
+void __attribute__((noinline)) vvadd(int coreid, int ncores, size_t n, const data_t* x, const data_t* y, data_t* z)
+{
+   size_t i;
+
+   // interleave accesses
+   for (i = coreid; i < n; i+=ncores)
+   {
+      z[i] = x[i] + y[i];
+   }
+}