feat: part2 and zip

.gitignore

@@ -1,3 +1,5 @@
 myexp
+test_auto_vectorize
+**/*.s
 **/*.o
 **/*.swp

part1/vectorOP.cpp

@@ -64,7 +64,7 @@ void clampedExpVector(float *values, int *exponents, float *output, int N)
       float *tmp = new float[VECTOR_WIDTH];
       for (int j = 0; i + j < N; j++)
         tmp[j] = values[i+j];
-      for (int j = N; j < i + VECTOR_WIDTH; j++)
+      for (int j = N % VECTOR_WIDTH; j < VECTOR_WIDTH; j++)
         tmp[j] = 0;
       _pp_vload_float(x, tmp, maskAll);
       delete []tmp;
@@ -76,7 +76,7 @@ void clampedExpVector(float *values, int *exponents, float *output, int N)
       int *tmp = new int[VECTOR_WIDTH];
       for (int j = 0; i + j < N; j++)
         tmp[j] = exponents[i+j];
-      for (int j = N; j < i + VECTOR_WIDTH; j++)
+      for (int j = N % VECTOR_WIDTH; j < VECTOR_WIDTH; j++)
         tmp[j] = 0;
       _pp_vload_int(y, tmp, maskAll);
       delete []tmp;
@@ -99,7 +99,7 @@ void clampedExpVector(float *values, int *exponents, float *output, int N)
 
     if (i + VECTOR_WIDTH > N) {
       float *tmp = new float[VECTOR_WIDTH];
-      _pp_vstore_float(output + i, result, maskAll);
+      _pp_vstore_float(tmp, result, maskAll);
       for (int j = 0; i + j < N; j++)
         output[i+j] = tmp[j];
       delete []tmp;

part2/test.sh

@@ -0,0 +1,9 @@
+RUNS=${1:-5}
+median() { sort -n | awk '{a[NR]=$1} END{print (NR%2)?a[(NR+1)/2]:(a[NR/2]+a[NR/2+1])/2}'; }
+for cfg in "novec:" "vec:VECTORIZE=1" "avx2:VECTORIZE=1 AVX2=1"; do
+  name=${cfg%%:*}; flags=${cfg#*:}
+  make -s clean; make -s $flags >/dev/null 2>&1 || { echo "build $name failed"; exit 1; }
+  t=$(for r in $(seq $RUNS); do ./test_auto_vectorize -t 1 | grep -oE '^[0-9.]+'; done | median)
+  echo "$name $t"; eval "T_$name=$t"
+done
+awk -v n=$T_novec -v v=$T_vec -v a=$T_avx2 'BEGIN{printf "vec/novec  speedup: %.2fx\navx2/vec   speedup: %.2fx\navx2/novec speedup: %.2fx\n", n/v, v/a, n/a}'

part2/test1.c

@@ -1,7 +1,10 @@
 #include "test.h"
 
-void test1(float *a, float *b, float *c, int N) {
+void test1(float *__restrict a, float *__restrict b, float *__restrict c, int N) {
   __builtin_assume(N == 1024);
+  a = (float *)__builtin_assume_aligned(a, 32);
+  b = (float *)__builtin_assume_aligned(b, 32);
+  c = (float *)__builtin_assume_aligned(c, 32);
 
   for (int i=0; i<I; i++) {
     for (int j=0; j<N; j++) {

part2/test2.c

@@ -11,9 +11,8 @@ void test2(float *__restrict a, float *__restrict b, float *__restrict c, int N)
     for (int j = 0; j < N; j++)
     {
       /* max() */
-      c[j] = a[j];
-      if (b[j] > a[j])
-        c[j] = b[j];
+      if (b[j] > a[j]) c[j] = b[j];
+      else c[j] = a[j];
     }
   }
 }

part2/test2.c.orig

@@ -0,0 +1,19 @@
+#include "test.h"
+
+void test2(float *__restrict a, float *__restrict b, float *__restrict c, int N)
+{
+  __builtin_assume(N == 1024);
+  a = (float *)__builtin_assume_aligned(a, 16);
+  b = (float *)__builtin_assume_aligned(b, 16);
+
+  for (int i = 0; i < I; i++)
+  {
+    for (int j = 0; j < N; j++)
+    {
+      /* max() */
+      c[j] = a[j];
+      if (b[j] > a[j])
+        c[j] = b[j];
+    }
+  }
+}