add ivdep examp

04/5_loops/04_no_pointer_aliasing_ivdep/main.cpp

@@ -0,0 +1,6 @@
+void func(float *a, float *b) {
+#pragma GCC ivdep
+    for (int i = 0; i < 1024; i++) {
+        a[i] = b[i] + 1;
+    }
+}

04/5_loops/05_loop_invariant_motion/main.cpp → 04/5_loops/06_loop_invariant_motion/main.cpp

@@ -1,5 +1,3 @@
-#include <cmath>
-
 void func(float *__restrict a, float *__restrict b, float dt) {
     for (int i = 0; i < 1024; i++) {
         a[i] = a[i] + b[i] * (dt * dt);

04/5_loops/05_loop_invariant_motion/pseudo.cpp → 04/5_loops/06_loop_invariant_motion/pseudo.cpp

@@ -1,5 +1,3 @@
-#include <cmath>
-
 void func(float *__restrict a, float *__restrict b, float dt) {
     float dt2 = dt * dt;
     for (int i = 0; i < 1024; i++) {

04/5_loops/07_loop_invariant_failed/.vim_localrc

@@ -0,0 +1,2 @@
+" vim: ft=vim
+nnoremap <silent> <F7> :wa<CR>:!make\|\|(echo -n .;read -n1)<CR><CR>

04/5_loops/06_loop_invariant_failed/main.cpp → 04/5_loops/07_loop_invariant_failed/main.cpp

@@ -1,5 +1,3 @@
-#include <cmath>
-
 void func(float *__restrict a, float *__restrict b, float dt) {
     for (int i = 0; i < 1024; i++) {
         a[i] = a[i] + b[i] * dt * dt;

04/5_loops/07_loop_invariant_failed/run.sh

@@ -0,0 +1,4 @@
+set -e
+
+gcc -fopenmp -O3 -fomit-frame-pointer -fverbose-asm -S main.cpp -o /tmp/main.S
+vim /tmp/main.S

04/5_loops/08_loop_unrolling/.vim_localrc

@@ -0,0 +1,2 @@
+" vim: ft=vim
+nnoremap <silent> <F7> :wa<CR>:!make\|\|(echo -n .;read -n1)<CR><CR>

04/5_loops/08_loop_unrolling/main.cpp

@@ -0,0 +1,6 @@
+void func(float *a) {
+#pragma GCC unroll 4
+    for (int i = 0; i < 1024; i++) {
+        a[i] = 1;
+    }
+}

04/5_loops/08_loop_unrolling/pseudo.cpp

@@ -0,0 +1,8 @@
+void func(float *a) {
+    for (int i = 0; i < 1024; i += 4) {
+        a[i + 0] = 1;
+        a[i + 1] = 1;
+        a[i + 2] = 1;
+        a[i + 3] = 1;
+    }
+}

04/5_loops/08_loop_unrolling/run.sh

@@ -0,0 +1,4 @@
+set -e
+
+gcc -fopenmp -O3 -fomit-frame-pointer -fverbose-asm -S main.cpp -o /tmp/main.S
+vim /tmp/main.S