add 8

04/00/main.cpp

@@ -1,9 +1,11 @@
-struct A {
-    static inline int other(int a) {
-        return a;
-    }
-};
+#include <cmath>
 
-int func() {
-    return A().other(233);
+void func(float *a, float *b, float *c) {
+    for (int i = 0; i < 1024; i++) {
+        float tmp = 0;
+        for (int j = 0; j < 1024; j++) {
+            tmp += a[i] * b[j];
+        }
+        c[i] += tmp;
+    }
 }

04/00/run.sh

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

04/4_simd/09_vectorize_reduction/main.cpp

@@ -0,0 +1,7 @@
+float func(float *a) {
+    float ret = 0;
+    for (int i = 0; i < 1024; i++) {
+        ret += a[i];
+    }
+    return ret;
+}

04/4_simd/09_vectorize_reduction/pseudo.cpp

@@ -0,0 +1,12 @@
+#include <x86intrin.h>
+
+float func(float *a) {
+    __m128 ret = _mm_setzero_ps();
+    for (int i = 0; i < 1024; i += 4) {
+        __m128 a_i = _mm_loadu_ps(&a[i]);
+        ret = _mm_add_ps(ret, a_i);
+    }
+    float r[4];
+    _mm_storeu_ps(r, ret);
+    return r[0] + r[1] + r[2] + r[3];
+}

04/4_simd/09_vectorize_reduction/run.sh

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

04/5_loops/06_loop_with_external_call/main.cpp

@@ -0,0 +1,10 @@
+void other();
+
+float func(float *a) {
+    float ret = 0;
+    for (int i = 0; i < 1024; i++) {
+        ret += a[i];
+        other();
+    }
+    return ret;
+}

04/5_loops/06_loop_with_external_call/run.sh

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

04/5_loops/07_loop_with_inlineable_call/main.cpp

@@ -0,0 +1,11 @@
+void other() {
+}
+
+float func(float *a) {
+    float ret = 0;
+    for (int i = 0; i < 1024; i++) {
+        ret += a[i];
+        other();
+    }
+    return ret;
+}

04/5_loops/07_loop_with_inlineable_call/run.sh

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

04/5_loops/08_loop_with_random_access/.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_with_random_access/main.cpp

@@ -0,0 +1,5 @@
+void func(float *a, int *b) {
+    for (int i = 0; i < 1024; i++) {
+        a[b[i]] += 1;
+    }
+}

04/5_loops/08_loop_with_random_access/run.sh

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

04/5_loops/09_loop_with_skip_access/.vim_localrc

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

04/5_loops/09_loop_with_skip_access/main.cpp

@@ -0,0 +1,5 @@
+void func(float *a) {
+    for (int i = 0; i < 1024; i++) {
+        a[i * 2] += 1;
+    }
+}

04/5_loops/09_loop_with_skip_access/run.sh

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

04/5_loops/10_loop_with_ordered_access/.vim_localrc

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

04/5_loops/10_loop_with_ordered_access/main.cpp

@@ -0,0 +1,5 @@
+void func(float *a) {
+    for (int i = 0; i < 1024; i++) {
+        a[i] += 1;
+    }
+}

04/5_loops/10_loop_with_ordered_access/run.sh

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

04/5_loops/11_loop_invariant_motion/.vim_localrc

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

04/5_loops/12_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/13_loop_unrolling/.vim_localrc

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

04/7_stl/04_std_vector_aos/.vim_localrc

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

04/7_stl/04_std_vector_aos/main.cpp

@@ -0,0 +1,15 @@
+#include <vector>
+
+struct MyVec {
+    float x;
+    float y;
+    float z;
+};
+
+std::vector<MyVec> a;
+
+void func() {
+    for (std::size_t i = 0; i < a.size(); i++) {
+        a[i].x *= a[i].y;
+    }
+}

04/7_stl/04_std_vector_aos/run.sh

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

04/7_stl/05_std_vector_soa/.vim_localrc

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

04/7_stl/05_std_vector_soa/main.cpp

@@ -0,0 +1,15 @@
+#include <vector>
+
+struct MyVec {
+    std::vector<float> x;
+    std::vector<float> y;
+    std::vector<float> z;
+};
+
+MyVec a;
+
+void func() {
+    for (std::size_t i = 0; i < a.x.size(); i++) {
+        a.x[i] *= a.y[i];
+    }
+}

04/7_stl/05_std_vector_soa/run.sh

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

04/8_math/01_div_becomes_mul/.vim_localrc

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

04/8_math/01_div_becomes_mul/main.cpp

@@ -0,0 +1,3 @@
+float func(float a) {
+    return a / 2;
+}

04/8_math/01_div_becomes_mul/run.sh

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

04/8_math/02_common_multiple_div/.vim_localrc

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

04/8_math/02_common_multiple_div/main.cpp

@@ -0,0 +1,5 @@
+void func(float *a, float b) {
+    for (int i = 0; i < 1024; i++) {
+        a[i] /= b;
+    }
+}

04/8_math/02_common_multiple_div/run.sh

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

04/8_math/03_no_multiple_div/.vim_localrc

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

04/8_math/03_no_multiple_div/main.cpp

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

04/8_math/03_no_multiple_div/run.sh

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

04/8_math/04_no_multiple_div_fast_math/.vim_localrc

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

04/8_math/04_no_multiple_div_fast_math/main.cpp

@@ -0,0 +1,5 @@
+void func(float *a, float b) {
+    for (int i = 0; i < 1024; i++) {
+        a[i] /= b;
+    }
+}

04/8_math/04_no_multiple_div_fast_math/run.sh

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

04/8_math/05_std_sqrt_overloads/.vim_localrc

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

04/8_math/05_std_sqrt_overloads/main.cpp

@@ -0,0 +1,13 @@
+#include <cmath>
+#include <type_traits>
+
+void func(float *a) {
+    for (int i = 0; i < 1024; i++) {
+        auto x = sqrt(a[i]);      // wrong and bad
+        static_assert(std::is_same_v<decltype(x), double>);
+        auto y = sqrtf(a[i]);     // correct but bad
+        static_assert(std::is_same_v<decltype(y), float>);
+        auto z = std::sqrt(a[i]); // correct and good
+        static_assert(std::is_same_v<decltype(z), float>);
+    }
+}

04/8_math/05_std_sqrt_overloads/run.sh

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

04/8_math/06_vectorize_sqrt/.vim_localrc

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

04/8_math/06_vectorize_sqrt/main.cpp

@@ -0,0 +1,7 @@
+#include <cmath>
+
+void func(float *a) {
+    for (int i = 0; i < 1024; i++) {
+        a[i] = std::sqrt(a[i]);
+    }
+}

04/8_math/06_vectorize_sqrt/run.sh

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

04/8_math/07_nested_loop_add/.vim_localrc

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

04/8_math/07_nested_loop_add/main.cpp

@@ -0,0 +1,9 @@
+#include <cmath>
+
+void func(float *a, float *b, float *c) {
+    for (int i = 0; i < 1024; i++) {
+        for (int j = 0; j < 1024; j++) {
+            c[i] += a[i] * b[j];
+        }
+    }
+}

04/8_math/07_nested_loop_add/run.sh

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

04/8_math/08_nested_loop_add_vectorize/.vim_localrc

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

04/8_math/08_nested_loop_add_vectorize/main.cpp

@@ -0,0 +1,11 @@
+#include <cmath>
+
+void func(float *a, float *b, float *c) {
+    for (int i = 0; i < 1024; i++) {
+        float tmp = c[i];
+        for (int j = 0; j < 1024; j++) {
+            tmp += a[i] * b[j];
+        }
+        c[i] = tmp;
+    }
+}

04/8_math/08_nested_loop_add_vectorize/run.sh

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