vector benchmark

benchmarks/vector/sgemm_nn/Makefile

@@ -0,0 +1,33 @@
+LIB_PATH = ../../../runtime
+
+COMP     = /nethome/ekim79/riscv-gnu-toolchain/drops/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
+
+# VX_STR  = ../../startup/vx_start.s
+
+NEWLIB  = $(LIB_PATH)/newlib/newlib.c
+VX_STR  = $(LIB_PATH)/startup/vx_start.s
+VX_INT  = $(LIB_PATH)/intrinsics/vx_intrinsics.s
+VX_IO   = $(LIB_PATH)/io/vx_io.s $(LIB_PATH)/io/vx_io.c
+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_sgemm_nn.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
+
+VX_MAIN = vx_vec_sgemm_nn
+
+all: HEX DUMP ELF
+
+DUMP: ELF
+	$(DMP) -D $(VX_MAIN).elf > $(VX_MAIN).dump
+
+HEX: ELF
+	$(CPY) -O ihex $(VX_MAIN).elf $(VX_MAIN).hex
+
+ELF:
+	$(COMP) $(CC_FLAGS) $(VX_STR) $(VX_VEC) $(VX_FIO) $(NEWLIB) $(VX_INT) $(VX_IO) $(VX_API) $(VX_TEST) $(VX_MAIN).c $(LIBS) -Iinclude  -o $(VX_MAIN).elf

benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.c

@@ -0,0 +1,95 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include "../../../runtime/intrinsics/vx_intrinsics.h"
+#include "vx_vec_sgemm_nn.h"
+
+//---------------------------------------------------------------
+/* # void sgemm_nn(size_t n, size_t m, size_t k, 
+#          int *a,   // m * k matri size_t lda, 
+#          int *b,   // k * n matrix  size_t ldb, 
+#          int *c,   // m * n matrix  size_t ldc)
+#  c += a*b (alpha=1, no transpose on input matrices)
+#  matrices stored in C row-major order */
+//---------------------------------------------------------------
+
+int main()
+{
+    vx_tmc(1);
+
+    int m = 3;
+    int k = 3;
+    int n = 3;
+
+    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 < (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 1
+    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]:", k*n);
+    for (int i = 0; i < k*n; ++i) {
+        if (!(i % n)) printf("\n");
+        printf("%d ", b1[i]);
+    }
+#endif
+
+    vx_vec_sgemm_nn(n, m, k, a1, b1, c1);
+//    vx_vec_sgemm_nn(n, a1, b1, c1);
+
+#if 1 
+    printf("\n\nc[%d]:\n", 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 < k; r++) {
+       for (int c = 0; c < m; c++) {
+           for (int i = 0; i < n; i++) {
+               d1[r*k+i] += a1[r*k+c]*b1[i*n+c];
+           }
+       }
+    }
+
+#if 1
+   printf("\n\nc[%d]:\n", m*n);
+   for(int i = 0; i < m; ++i) {
+      for(int j = 0; j < n; ++j) {
+          printf("%d ", c1[i*m+j]);
+      }
+      printf("\n");
+    }
+#endif
+
+
+    for(int i = 0; i < m*n; ++i)
+    {
+        if(c1[i] != d1[i])
+        {
+           printf("\n<sgemm_nn> FAILED at <index: %d>! \n", i);
+           return 1;
+        }
+    }
+
+    printf("\nPASS.......................... <sgemm_nn> \n");
+
+
+    free(a1); free(b1); free(c1);
+
+    vx_tmc(0);
+
+    return 0;
+
+}

benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.h

@@ -0,0 +1,13 @@
+#pragma once
+
+
+#ifdef __cplusplus
+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 n, int m, int k, int* a1, int* b1, int* c1);
+//void vx_vec_sgemm_nn(int n, int* a1, int* b1, int* c1);
+#ifdef __cplusplus
+}
+#endif                   

benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.s

@@ -0,0 +1,61 @@
+.type vx_vec_sgemm_nn, @function
+.global vx_vec_sgemm_nn
+# RV64IDV system
+#
+# void
+# sgemm_nn(size_t n, size_t m, size_t k,
+#          int *a,   // m * k matrix
+#          int *b,   // k * n matrix
+#          int *c)   // m * n matrix
+#
+#  c += a*b (alpha=1, no transpose on input matrices)
+#  matrices stored in C row-major order
+#
+# for (int r = 0; r < k; r++) {
+#    for (int c = 0; c < m; c++) {
+#        for (int i = 0; i < n; i++) {
+#            c[r*k+i] += a[r*k+c]*b[i*n+c];
+#        }
+#    }
+# }
+# a0 = n, a1 = m, a2 = k 
+# a3 = a, a4 = b, a5 = c
+# v0 = a, v2 = b, v2 = c
+# x0 = i, x1 = c, x2 = r 
+#
+vx_vec_sgemm_nn:
+        vsetvli t0, a2, e32, m8 # k
+loop_row:                   # a[m][k]
+    vlw.v v0, (a3)
+        sub a2, a2, t0
+        slli t0, t0, 2
+        add a3, a3, t0
+
+        vsetvli t1, a1, e32, m8 # m
+loop_col:                   # b[k][n]
+    vlw.v v1, (a4)
+        sub a1, a1, t1
+        slli t1, t1, 2
+        add a4, a4, t1
+
+        vsetvli t2, a0, e32, m8 # n
+loop_iner:
+    vlw.v v2, (a5)         # c[][]
+        sub a0, a0, t2
+        slli t2, t2, 2
+        add a5, a5, t2
+
+    bnez t2, loop_iner
+
+    bnez t1, loop_col
+
+
+ #   vadd.vv v0, v0, v0
+ #   vsw.v v0, (a5)
+ #   add a5, a5, t0
+
+    bnez t0, loop_row
+    ret
+
+
+