From 54a56ee8a087201efc20f69721725e92f8c332c5 Mon Sep 17 00:00:00 2001
From: ytshih <ytshih@it.cs.nycu.edu.tw>
Date: Mon, 5 Oct 2026 09:13:57 +0800
Subject: [PATCH] feat: part2 and zip

---
 .gitignore         |   2 ++
 HW1_111550013.zip  | Bin 0 -> 1668 bytes
 part1/vectorOP.cpp |   5 +++++
 part2/test.sh      |   9 +++++++++
 part2/test1.c      |   5 ++++-
 part2/test2.c      |   5 ++---
 part2/test2.c.orig |  19 +++++++++++++++++++
 7 files changed, 41 insertions(+), 4 deletions(-)
 create mode 100644 HW1_111550013.zip
 create mode 100644 part2/test.sh
 create mode 100644 part2/test2.c.orig

diff --git a/.gitignore b/.gitignore
index c2769f1..732b31c 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,5 @@
 myexp
+test_auto_vectorize
+**/*.s
 **/*.o
 **/*.swp
diff --git a/HW1_111550013.zip b/HW1_111550013.zip
new file mode 100644
index 0000000000000000000000000000000000000000..00ccbb33f3f8679c6c84046cc7848563bd47fe53
GIT binary patch
literal 1668
zcmWIWW@Zs#U|`^2*zVyPJF9<!t}H79!&3nU1|9|(hO*S;lKdk70KMdbg3u672Ig1a
z4`sz&Jd{;h!Og(P(&^8@z>>wnz_2$wH~+Soz`u9#g~f&rt89<o+<wfnJ&|`CL$k0>
zT(>R<kDJwPAC*hOn(lcM|L=R;o#Gi5oUSV-vC;J&_xqi@+n0yOxOmDbe_6k7#|268
z#i9GZuZpovJsFf1v;NZdt*&}g&rP@(_5KA%#V$Vs*M&2m1vSJl><+#@qi`C_LxFci
zymRch%p@|6z6mCu5Eb{{v5V<>@JElYW|NM)ecda{E!LSL6_+0|X|YsO{gj<?&#N-N
zBuIU;`^xa_^VcfBZO{Fd%y_dtb;T!_@|>C@K8GZ4xVGL}%l7!9(wdgwWpW4p{I>LL
z`*yr_)}u*HcN4zql<qWnIZ<gL)0Q@MgJgaV9;ImymHZVZequbaK2A{X%Z6nRNje-o
zf(Op%Bpu|Lch5tGGe1f+=&(_`q_7mHo$kZHH2x1eoEQ&^>}&u2eE%7}|1;`$RG9zP
zZFzq#!RC{i-I{I3n&Kk2d$<Sw5Pqwh!q(V0QJCQ|XYJfWS890A&AR6CZ8GPnS;_(q
zTp{0XE`G%Lr0Mm}o7{&yo$od53BIv#`*QPjn*a17-)JA3GDVv?BE?8Gpyk==m}gu;
zKB?PS#g0!&dfK15*|W2!)~Y*kr5%Hvj=!v5&+I0?hi3Ud-k$qynx)OI)0HANJ4yA-
zb;)BkNmAK)r)Fh}vss!JZol)9m4%hjmM?q8JaLJqvp(spSo`Tj*~Dyt>XrrP#cMl+
zrfKt(&kL=49=rRM-e1wTN~QA{V_pZ$5c$M^RPP%1#)CeI)|Mjgg;>)>^u8Dd-q<em
z)#lx*d&@7aIlnTjd7ZFL<Hk>|JDA^R>EDU6J6&<xGk4Ak?N>=m{!_p6yJWwZ>3C4K
zTh8aix&>Jp!G9Nif9+#g|5S3-n@OjF``1K^Exx=wyXSm&srk&A+p2qy&RP}sBK_ps
z7bmtF20s1z=6w0{%QsEW=-rQ<cfi$i_lBvJVd~+N{f-~;D=(ilX=-AtWM9^^|3+$4
ze2QkXY@JjoWZh+Ga`4Et#@Ly2E=Ap0(7EvGir+Dr$GyJVcKmERK4n$r%TJN#%pT_R
zCl@N;dR7p%*=N&o`Hf%guJ4!JyRp8z;JiWV^4#w~c2*~99lalV`oEX>l4EB&?ENR}
zKTZsu(dLq|oAYDzDvAC4skSe#u;_fasC(y%XrB7goR|OieX(CK{~OQuX2%yx8BW|d
z`!}~&kNx{yvs>3M9+{QBwj@e!`Zp_yIW4p9ru|FunK0|evuR~_qg;*Vo0hx?O<4Pt
zkL8Zp7rwlzKW+)(Dj)41=62Qm{+dy3eD$UEgL`I+vz6ooMGw8y{1U8J7NmZ`ynB6S
zVPD^$rne`a<#7BvzQB@mqD7a*%q?p+Ejp~F9hbAUEHI=eD315|Vpr+IQr;yS8wD$4
zKC&npR8PE69<zGEGSQ%$y@%Bv{OE7pebC|d-j|27-^^c7m_9wPCgq;<EcLe!PQ72`
zSev+b!I_pv+>Lf3DbwEs^BzvEo$-lfSF!7tyM51%PlV68x8RY{ihS?+)xk}<bJu-*
zp!zr7PnY@j@^y*v*V^<YV|>K^{BY`>x$vCv!Z)XzP6hvVa^F5ZpV#%tv&>D;+1R~$
z7?yd6bauZ8ZEn<*@O#{OaqZ%33oo8%3XMDMzQy=|zwLC6X{kBYbC=Jbzx2(9FIO)I
zJw4IuS?2NU(Saa&X1*P7)^3|NOOmtd{G30_IUiieJ}X+28h;@2b(}!RkK^B7zm=3R
z?@V>g54Ha7x_J9ZsWi?_yzjh!^x5s{+!ehd<N6Bc+y}Q3-;~}yn=p4bR~eV^s#o=l
zpdyxw0R#j+Tw~j6w_Vd_U|;}Yb_N-S(xM!_l8O>=>ALCLp)3$CMJ-)3N=gcft@QOX
z5|guYQ}i<P^&PVF^O7?%^Yonzt3r%@a`Jss^;1(aOSl5O8JXmmaTVwiAV)ATGB7YK
zX#_D*i+5H?@s4IW#6(=~M>a8;l>y-!Rt9D)J_+z<Wdm8v#K6QbpMinlJS&I?02RsV
A#{d8T

literal 0
HcmV?d00001

diff --git a/part1/vectorOP.cpp b/part1/vectorOP.cpp
index e4e5e63..8b47e3f 100644
--- a/part1/vectorOP.cpp
+++ b/part1/vectorOP.cpp
@@ -1,4 +1,6 @@
 #include "PPintrin.h"
+#include <algorithm>
+#include <iostream>
 
 // implementation of absSerial(), but it is vectorized using PP intrinsics
 void absVector(float *values, float *output, int N)
@@ -84,6 +86,9 @@ void clampedExpVector(float *values, int *exponents, float *output, int N)
       _pp_vload_int(y, exponents + i, maskAll); // y = exponents[i];
     }
 
+    cout << "min exponents: " << *min_element(y.value, y.value+VECTOR_WIDTH) << endl;
+    cout << "max exponents: " << *max_element(y.value, y.value+VECTOR_WIDTH) << endl;
+
     _pp_vset_float(result, 1.f, maskAll);
     maskNonZero = _pp_init_ones();
     while (_pp_cntbits(maskNonZero) > 0) {
diff --git a/part2/test.sh b/part2/test.sh
new file mode 100644
index 0000000..46ab149
--- /dev/null
+++ b/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}'
diff --git a/part2/test1.c b/part2/test1.c
index 686f5f7..479719b 100644
--- a/part2/test1.c
+++ b/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++) {
diff --git a/part2/test2.c b/part2/test2.c
index eac500e..7c87170 100644
--- a/part2/test2.c
+++ b/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];
     }
   }
 }
diff --git a/part2/test2.c.orig b/part2/test2.c.orig
new file mode 100644
index 0000000..eac500e
--- /dev/null
+++ b/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];
+    }
+  }
+}