Vector evaluations

benchmarks/vector/saxpy/Makefile

@@ -1,11 +1,11 @@
 LIB_PATH = ../../../runtime
 
-COMP     = /nethome/ekim79/riscv-gnu-toolchain/drops/bin/riscv32-unknown-elf-gcc
+COMP     = /home/priya/dev/riscv_vec/riscv-gnu/bin/riscv32-unknown-elf-gcc
 
 CC_FLAGS = -ffreestanding -O0 -Wl,--gc-sections -nostartfiles -nostdlib -nostartfiles -nodefaultlibs -Wl,-Bstatic,-T,$(LIB_PATH)/mains/vortex_link.ld -march=rv32imv -mabi=ilp32
 
-DMP  = /nethome/ekim79/riscv-gnu-toolchain/drops/bin/riscv32-unknown-elf-objdump
-CPY  = /nethome/ekim79/riscv-gnu-toolchain/drops/bin/riscv32-unknown-elf-objcopy
+DMP  = /home/priya/dev/riscv_vec/riscv-gnu/bin/riscv32-unknown-elf-objdump
+CPY  = /home/priya/dev/riscv_vec/riscv-gnu/bin/riscv32-unknown-elf-objcopy
 
 # VX_STR  = ../../startup/vx_start.s
 
@@ -17,7 +17,7 @@ VX_API  = $(LIB_PATH)/vx_api/vx_api.c
 VX_TEST = $(LIB_PATH)/tests/tests.c
 VX_FIO  = $(LIB_PATH)/fileio/fileio.s
 VX_VEC  = vx_vec_saxpy.s       #float --> int
-LIBS    = /nethome/ekim79/riscv-gnu-toolchain/drops/riscv32-unknown-elf/lib/libc.a /nethome/ekim79/riscv-gnu-toolchain/drops/riscv32-unknown-elf/lib/libstdc++.a -static-libgcc -lgcc
+LIBS    = /home/priya/dev/riscv_vec/riscv-gnu/riscv32-unknown-elf/lib/libc.a /home/priya/dev/riscv_vec/riscv-gnu/riscv32-unknown-elf/lib/libstdc++.a -static-libgcc -lgcc
 
 VX_MAIN = vx_vec_saxpy
 

benchmarks/vector/saxpy/vx_vec_saxpy.c

@@ -16,7 +16,7 @@ int main()
 {
     vx_tmc(1);
 
-    int n = 4; //#define NUM_DATA 65536
+    int n = 64; //#define NUM_DATA 65536
 
     int *a = (int*)malloc(sizeof(int) * n); 
     int *b = (int*)malloc(sizeof(int) * n); 
@@ -39,9 +39,18 @@ int main()
 //    for(int i = 0; i < n; ++i) printf("%d \n", b[i]);
 #endif
 
+    int startCycles = vx_getCycles();
+    int startInst = vx_getInst();
     vx_vec_saxpy(n, factor, a, b);
+    int endCycles = vx_getCycles();
+    int endInst = vx_getInst();
 
-#if 1
+    int totalInst = (endInst - startInst);
+    int totalCycles = (endCycles - startCycles);
+
+    printf("\nCycles = %d, Instructions = %d", totalCycles, totalInst);
+
+#if 0
     printf("\nsaxpy\na[%d]: ", n);
     for(int i = 0; i < n; ++i) printf("%d ", a[i]);
     printf("\n\nb[%d]: ", n);

benchmarks/vector/saxpy/vx_vec_saxpy.s

@@ -10,8 +10,8 @@
 #     a2      a
 #     a3      b
 vx_vec_saxpy:
-    vsetvli a4, a0, e32
 loop:
+    vsetvli a4, a0, e32
     vlw.v v0, (a2)
     sub a0, a0, a4
     slli a4, a4, 2

benchmarks/vector/sfilter/vx_vec_sfilter.c

@@ -35,12 +35,21 @@ int main()
         c[i] = 0;
     }
 
-    int N = 32;
+    int N = 4;
+    int startCycles = vx_getCycles();
+    int startInst = vx_getInst();
     for(int y = 1; y < (NUM_DATA-1); y++){
         for(int x = 1; x < (NUM_DATA-1); x = x+N) {
             vx_vec_sfilter(a, b, ldc, m, x, y, N);
         }
     }
+    int endCycles = vx_getCycles();
+    int endInst = vx_getInst();
+
+    int totalInst = (endInst - startInst);
+    int totalCycles = (endCycles - startCycles);
+
+    printf("\nCycles = %d, Instructions = %d", totalCycles, totalInst);
     
 
     for(int y = 1; y < (NUM_DATA-1); ++y) 

benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.c

@@ -16,76 +16,89 @@ int main()
 {
     vx_tmc(1);
 
-    int w = 4;
-    int h = 4;
-    int d = 4;
+    int m = 64;
+    int k = 64;
+    int n = 64;
 
-    int* a1 = (int*)malloc(sizeof(int) * w * h);
-    int* b1 = (int*)malloc(sizeof(int) * h * d);
-    int* c1 = (int*)malloc(sizeof(int) * w * d);
-    int* d1 = (int*)malloc(sizeof(int) * w * d); //verfication
+    int* a1 = (int*)malloc(sizeof(int) * m * k);
+    int* b1 = (int*)malloc(sizeof(int) * k * n);
+    int* c1 = (int*)malloc(sizeof(int) * m * n);
+    int* d1 = (int*)malloc(sizeof(int) * m * n); //verfication
 
-    for (int i = 0; i < (w * h); ++i) a1[i] = i;
-    for (int i = 0; i < (h * d); ++i) b1[i] = 1;
-    for (int i = 0; i < (w * d); ++i) c1[i] = 0;
-    for (int i = 0; i < (w * d); ++i) d1[i] = 0;
+    for (int i = 0; i < (m * k); ++i) a1[i] = i;
+    for (int i = 0; i < (k * n); ++i) b1[i] = 1;
+    for (int i = 0; i < (m * n); ++i) c1[i] = 0;
+    for (int i = 0; i < (m * n); ++i) d1[i] = 0;
 
 
 #if 0
-    printf("sgemm_nn\na[%d]:", w*h);
-    for (int i = 0; i < w*h; ++i) {
-        if(!(i % h)) printf("\n");
+    printf("sgemm_nn\na[%d]:", m*k);
+    for (int i = 0; i < m*k; ++i) {
+        if(!(i % k)) printf("\n");
         printf("%d ", a1[i]);
     }
-    printf("\n\nb[%d]:", h*d);
-    for (int i = 0; i < h*d; ++i) {
-        if (!(i % d)) printf("\n");
+    printf("\n\nb[%d]:", k*n);
+    for (int i = 0; i < k*n; ++i) {
+        if (!(i % n)) printf("\n");
         printf("%d ", b1[i]);
     }
 #endif
 
     int lda = 4;
     int ldb = 4;
-    int ldc = 4; //64;
-    int vsize = 4;
+    int ldc = 64; //64;
+    int vsize = 32;
 
-  for (int n = 0; n < h; n++) {
-       for (int i = 0; i < w; i=+4) {
-           for (int m = 0; m < d; m++) {
-                 vx_vec_sgemm_nn(i, m, n, a1, b1, c1, ldc, vsize);
+
+  int startCycles = vx_getCycles();
+  int startInst = vx_getInst();
+  for (int r = 0; r < m; r++) {
+       for (int c = 0; c < n; c++) {
+           for (int i = 0; i < k;) {
+//               d1[r*k+i] += a1[r*k+c]*b1[i*n+c];
+                 vx_vec_sgemm_nn(i, r, c, a1, b1, c1, ldc, vsize);
                  i = i + vsize;
            }
        }
     }
+    int endCycles = vx_getCycles();
+    int endInst = vx_getInst();
 
-#if 1 
-    printf("\n\nc[%d]:", d*h);
-    for (int i = 0; i < d*h; ++i) {
-        if (!(i % h)) printf("\n");
+    int totalInst = (endInst - startInst);
+    int totalCycles = (endCycles - startCycles);
+    printf("\nCycles = %d, Instructions = %d", totalCycles, totalInst);
+//    vx_vec_sgemm_nn(n, a1, b1, c1);
+
+#if 0
+    printf("\n\nc[%d]:", m*n);
+    for (int i = 0; i < m*n; ++i) {
+        if (!(i % n)) printf("\n");
         printf("%d ", c1[i]);
     }
 #endif
 
-   for (int r = 0; r < h; r++) {
-       for (int c = 0; c < w; c++) {
-           for (int i = 0; i < d; i++) {
-               d1[r*h+i] += a1[r*h+c]*b1[i*d+c];
+   for (int r = 0; r < m; r++) {
+       for (int c = 0; c < n; c++) {
+           for (int i = 0; i < k; i++) {
+               d1[c*ldc+i] += a1[c*ldc+r]*b1[i + (r*ldc)];
+               //printf("d[%d] += a[%d]*b[%d]\n", c*ldc+i, c*ldc+r , i + (r*ldc));
+               //printf("%d %d %d\n", d1[c*ldc+i] , a1[c*ldc+r] , b1[i + (r*ldc)]);
            }
        }
     }
 
-#if 1
-   printf("\n\nc[%d]:\n", w*d);
-   for(int i = 0; i < w; ++i) {
-      for(int j = 0; j < d; ++j) {
-          printf("%d ", d1[i*w+j]);
+#if 0
+   printf("\n\nc[%d]:\n", m*n);
+   for(int i = 0; i < m; ++i) {
+      for(int j = 0; j < n; ++j) {
+          printf("%d ", d1[i*m+j]);
       }
       printf("\n");
     }
 #endif
 
 
-    for(int i = 0; i < w*d; ++i)
+    for(int i = 0; i < m*n; ++i)
     {
         if(c1[i] != d1[i])
         {

benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.h

@@ -6,7 +6,7 @@ extern "C" {
 #endif
 
 //void vx_vec_sgemm_nn(int n, int m, int k, int* a1, int lda, int* b1, int ldb, int* c1, int ldc);
-void vx_vec_sgemm_nn(int i, int m, int n, int* a1, int* b1, int* c1, int ldc, int vsize);
+void vx_vec_sgemm_nn(int n, int m, int k, int* a1, int* b1, int* c1, int ldc, int vsize);
 //void vx_vec_sgemm_nn(int n, int* a1, int* b1, int* c1);
 #ifdef __cplusplus
 }

benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.s

@@ -10,33 +10,33 @@
 #           }
 #       }
 #    } 
-# a0 = i, a1 = m, a2 = n,  a3 = a, a4 = b, a5 = c, a6 = ldc, a7 = vsize
-#
+# a3 = a, a4 = b, a5 = c
+# a0 = i, a1 = m, a2 = n
+# a6 = ldc
 vx_vec_sgemm_nn:
-    vsetvli t0, a7, e32 # <--- vsize 
-    mul x11, a6, a2  # n*ldc
-    add x12, x11, a1  # i + (n*ldc)
-    slli x12, x12, 2 
-    add a3, x12, a3  # a[i+ n*ldc]
-    lw x13, (a3)
+    vsetvli t0, a7, e32
+     mul t1, a6, a2  # n*ldc
+     add t2, t1, a1  # i + (n*ldc)
+     slli t2, t2, 2
+     add a3, t2, a3  # a[i+ n*ldc]
+     lw t3, (a3)
 
-    mul x14, a1, a6  # m*ldc
-    add x15, a0, x14  # i + m*ldc
-    slli x15, x15, 2
-    add a4, x15, a4  # b[i + m*ldc]
-    vlw.v v0, (a4)
-    vmul.vx v2, v1, x13
-##   lw x6, (a4)
-##   lw x10, (a4)  # b
-##   mul x11, x3, x10
+     mul t4, a1, a6  # m*ldc
+     add t5, a0, t4  # i + m*ldc
+     slli t5, t5, 2
+     add a4, t5, a4  # b[i + m*ldc]
+ #   lw x6, (a4)
+
+     vlw.v v0, (a4)
+     vmul.vx v1, v0, t3
+ 
+     mul t6, a2, a6  # n*ldc
+     add t0, a0, t6  # i + n*ldc
+     slli t0, t0, 2
+     add a5, t0, a5  # c[i + m*ldc]
+
+     vlw.v v2, (a5) #c
+     vadd.vv v2, v2, v1
+     vsw.v v2, (a5)
 
-    mul x6, a2, a6  # n*ldc
-    add x7, a0, x6  # i + n*ldc
-    add a5, x7, a5  # c[i + m*ldc]
-    vlw.v v3, (a5) # c
-    vadd.vv v3, v3, v2
-    vsw.v v3, (a5)
-##   lw x12, (a5)
-##   add x12, x12, x11
-##   sw x12, (a5)    
     ret