Maybe improved SIMD

Implementierung/tricorn_V0.c

@@ -1,7 +1,7 @@
 #include "tricorn_V0.h"
 #include <immintrin.h>
 
-static unsigned converge(float c_rea, float c_imag, unsigned n);
+static unsigned converge(float c_real, float c_imag, unsigned n);
 
 
 /**
@@ -28,15 +28,17 @@ void tricorn_V0(float complex start, size_t width, ssize_t height, float res, un
     __m128 four = _mm_set1_ps(4.0f);
     __m128 minus_two = _mm_set1_ps(-2.0f);
 
+    float c_imag = c_start_imag;
+    __m128 c_imag_vector = _mm_set1_ps(c_imag);
+
+    float four_res = 4 * res;
+    float vertical_step = imag_step * res;
+    __m128 vertical_step_vector = _mm_set1_ps(vertical_step);
+
     for(ssize_t y = 0; y < actual_height;y++) {
-        float c_imag = c_start_imag - (y*imag_step) * res;
-        __m128 c_imag_vector = _mm_set1_ps(c_imag);
-
         size_t x = 0;
+        __m128 c_real_vector = _mm_setr_ps(c_start_real,c_start_real + 1.0f*res,c_start_real + 2.0f*res, c_start_real + 3.0f*res);
         for(; x < simd_width;x += 4) { //SIMD Loop
-            float c_real = c_start_real + (x*res);
-            __m128 c_real_vector = _mm_setr_ps(c_real,c_real+1.0f*res,c_real+2.0f*res,c_real+3.0f*res);
-
             __m128 real = _mm_setzero_ps();
             __m128 imag = _mm_setzero_ps();
             __m128i iterations = _mm_setzero_si128();
@@ -86,6 +88,7 @@ void tricorn_V0(float complex start, size_t width, ssize_t height, float res, un
                     img[position + 2] = ((value*5) % 255);
                 }
             }
+            c_real_vector = _mm_add_ps(c_real_vector,_mm_set1_ps(four_res));
         }
 
         for(;x < width; x++) {
@@ -102,8 +105,10 @@ void tricorn_V0(float complex start, size_t width, ssize_t height, float res, un
                 img[position + 1] = ((value*2) % 255);
                 img[position + 2] = ((value*5) % 255);
             }
+            c_real += res;
         }
-
+        c_imag -= vertical_step;
+        c_imag_vector = _mm_sub_ps(c_imag_vector,vertical_step_vector);
     }
 }
 

Implementierung/tricorn_V1.c

@@ -1,7 +1,6 @@
 #include "tricorn_V1.h"
 
-static const float convergenceLimit = 2.0f; // Mandelbrot set limit
-static unsigned converge(float complex position,unsigned n);
+static unsigned converge(float c_real, float c_imag, unsigned n);
 
 
 /**
@@ -15,20 +14,30 @@ static unsigned converge(float complex position,unsigned n);
  * @param img Pointer to the pre-allocated memory buffer where the bitmap data will be stored.
  */
 void tricorn_V1(float complex start, size_t width, ssize_t height, float res, unsigned n, _Bool color, unsigned char* img){
+    int bytesPerColor = color ? 3 : 1;
     
     // Absolute height for iteration purposes
-    height = height < 0 ?  height*-1 : height;
-    unsigned bytesPerColor = color ? 3 : 1;
+    ssize_t actual_height = height < 0 ?  (ssize_t)(-height) : height;
+    ssize_t imag_step = height < 0 ? -1:1;
+
+    float c_start_real = creal(start);
+    float c_start_imag = cimag(start);
+
+    float c_imag = c_start_imag;
+    size_t position = 0;
+
 
     // Simple iteration over every pixel in the picture
-    for(size_t i = 0; i < width; i++){
-        for(ssize_t j = 0; j < height; j++){
+    for(ssize_t y = 0; y < actual_height; y++){
 
-            unsigned value = converge(start + (i*res - I*j*res),n+1);
+        // Calculate imaginary part on each row as it stays the same
+
+        float c_real = c_start_real;
+
+        for(size_t x = 0; x < width; x++){
+            unsigned value = converge(c_real,c_imag,n);
 
             // Pixel value position in the output array
-            size_t position = (j*width +i) * bytesPerColor;
-
 
             if(!color){ // Grayscale colors as specified by the assignment r = g = b
                 img[position] =  value;
@@ -39,24 +48,29 @@ void tricorn_V1(float complex start, size_t width, ssize_t height, float res, un
                 img[position + 2] = ((value*5) % 255);
             }
 
+            position += bytesPerColor;
+            c_real += res;
+
         }
+        c_imag -= imag_step * res;
     }
 }
 
 // Iterative application of the equation given by the assignment
-static unsigned converge(float complex position,unsigned n){
+static inline unsigned converge(float c_real, float c_imag, unsigned n) {
+    float real = 0.0f;
+    float imag = 0.0f;
 
-    complex float result  = 0;
-    for(unsigned i = 0; i < n; i++){
-         // Detection of unbounded behaviour
-        // If the magnitude of the result vector is greater than the convergence limit, we know that the function will grow unbounded
-        // Current convergence limit taken from mandelbrot set
-        if(sqrtf(powf (crealf(result),2)+powf(cimagf(result),2)) > convergenceLimit)
-            return 255-4*(i % 64);
+    for (unsigned i = 0; i < n; i++) {
+        float real2 = real * real;
+        float imag2 = imag * imag;
 
-        // Application of the given complex equation with m = 2
-        result = cpowf(conjf(result),2) + position;
+        if (real2 + imag2 > 4.0f) {
+            return 255 - 4 * (i % 64);
+        }
 
+        imag = -2.0f * real * imag + c_imag;
+        real = real2 - imag2 + c_real;
     }
     return 0;
 }

Implementierung/tricorn_V2.c

@@ -23,7 +23,7 @@ void tricorn_V2(float complex start, size_t width, ssize_t height, float res, un
     for(size_t i = 0; i < width; i++){
         for(ssize_t j = 0; j < height; j++){
 
-            unsigned value = converge(start + (i*res - I*j*res),n+1);
+            unsigned value = converge(start + (i*res - I*j*res),n);
 
             // Pixel value position in the output array
             size_t position = (j*width +i) * bytesPerColor;