From f28aff6e7256840b1f172e8890ccba9e5b618b1a Mon Sep 17 00:00:00 2001 From: Yashar Alan Zenker Date: Sat, 4 Jul 2026 19:26:25 +0200 Subject: [PATCH] Added: Working non standard threaded SIMD implementation untested. --- Implementierung/main.c | 2 +- Implementierung/tricorn_V4.c | 182 +++++++++++++++++++++++++++++------ Implementierung/tricorn_V4.h | 8 ++ 3 files changed, 162 insertions(+), 30 deletions(-) diff --git a/Implementierung/main.c b/Implementierung/main.c index 431fd96..421788c 100644 --- a/Implementierung/main.c +++ b/Implementierung/main.c @@ -223,7 +223,7 @@ int main(int argc, char** argv) tricorn_V3(complexA,width,height,resolution,n,colored,tempArr); break; case 4: - // Uses non SIMD threaded version + // Uses SIMD threaded version tricorn_V4(complexA,width,height,resolution,n,colored,tempArr); break; default: diff --git a/Implementierung/tricorn_V4.c b/Implementierung/tricorn_V4.c index 8295593..26cc286 100644 --- a/Implementierung/tricorn_V4.c +++ b/Implementierung/tricorn_V4.c @@ -1,8 +1,22 @@ -#include "tricorn_V2.h" +#include "tricorn_V4.h" +static void compute(void * args); static unsigned converge(float c_real, float c_imag, unsigned n); +static size_t get_thread_count(); +typedef struct { + unsigned n; + _Bool color; + _Bool flipped; + float start_r; + float start_i; + float res; + size_t start_row; + size_t stop_row; + size_t width; + unsigned char* img; +} compute_args; /** * @brief Generates a Tricorn fractal image using an optimized SISD implementation. * @param start The starting complex coordinate (bottom-left or top-left depending on height sign). @@ -14,51 +28,152 @@ static unsigned converge(float c_real, float c_imag, unsigned n); * @param img Pointer to the pre-allocated memory buffer where the bitmap data will be stored. */ void tricorn_V4(float complex start, size_t width, ssize_t height, float res, unsigned n, _Bool color, unsigned char* img){ - int bytesPerColor = color ? 3 : 1; - + + const ssize_t min_pixels_per_thread = 50; + _Bool flipped = height < 0; + height = flipped ? -height : height; + + + size_t thread_count = get_thread_count(); + + if(height >= min_pixels_per_thread) + while((height/thread_count) < min_pixels_per_thread) thread_count--; + else thread_count = 1; + printf("%lu \n",thread_count); + const size_t y_step = (height+thread_count-1)/thread_count; + + const float start_r = crealf(start); + const float start_i = cimagf(start); + + + float directional_res = (flipped ? -1.0f: 1.0f)*res; + pthread_t threads[thread_count]; + for(size_t i = 0; i < thread_count; i++){ + compute_args* args = malloc(sizeof(*args)); + args->start_r =start_r; + args->start_i =start_i-i*directional_res*y_step; + args->start_row = i*y_step; + args->stop_row = height < (args->start_row+y_step) ? height: (args->start_row+y_step); + args->width = width; + args-> flipped = flipped; + args->color = color; + args->res = res; + args->img = img; + args->n = n; + if(pthread_create(&threads[i],NULL,compute,args)){ + compute(args); + free(args); + args = NULL; + } + } + for(int i = 0; i < thread_count; i++){ + pthread_join(threads[i],NULL); + } + + return; +} + +static void compute(void * vargs){ + + compute_args *args = vargs; + vargs = NULL; + + + int bytes_per_color = args->color ? 3 : 1; // Absolute height for iteration purposes - ssize_t actual_height = height < 0 ? (ssize_t)(-height) : height; - ssize_t imag_step = height < 0 ? -1:1; + size_t simd_width = (args->width / 4) * 4; - float c_start_real = creal(start); - float c_start_imag = cimag(start); + float c_start_real = args->start_r; + __m128 four = _mm_set1_ps(4.0f); + __m128 minus_two = _mm_set1_ps(-2.0f); - float c_imag = c_start_imag; - float vertical_step = imag_step * res; - size_t position = 0; + float c_imag = args->start_i; + __m128 c_imag_vector = _mm_set1_ps(c_imag); + float four_res = 4 * args->res; + float vertical_step = (args->flipped? -1.0f: 1.0f) * args->res; + __m128 vertical_step_vector = _mm_set1_ps(vertical_step); - // Simple iteration over every pixel in the picture - for(ssize_t y = 0; y < actual_height; y++){ + for(size_t y = args->start_row; y < args->stop_row;y++) { + size_t x = 0; + __m128 c_real_vector = _mm_setr_ps(c_start_real,c_start_real + 1.0f*args->res,c_start_real + 2.0f*args->res, c_start_real + 3.0f*args->res); + for(; x < simd_width;x += 4) { //SIMD Loop + __m128 real = _mm_setzero_ps(); + __m128 imag = _mm_setzero_ps(); + __m128i iterations = _mm_setzero_si128(); + __m128i active_mask = _mm_set1_epi32(-1); - // Calculate imaginary part on each row as it stays the same + for(unsigned i = 0; i < args->n;i++) { + __m128 real2 = _mm_mul_ps(real,real); + __m128 imag2 = _mm_mul_ps(imag,imag); - float c_real = c_start_real; - for(size_t x = 0; x < width; x++){ - unsigned value = converge(c_real,c_imag,n); + __m128 unbounded = _mm_cmple_ps(_mm_add_ps(real2,imag2),four); + __m128i unbounded_i = _mm_castps_si128(unbounded); - // Pixel value position in the output array + active_mask = _mm_and_si128(active_mask,unbounded_i); - if(!color){ // Grayscale colors as specified by the assignment r = g = b - img[position] = value; + if(_mm_test_all_zeros(active_mask,active_mask)) { + break; + } + + iterations = _mm_sub_epi32(iterations,active_mask); + + imag = _mm_add_ps(c_imag_vector,_mm_mul_ps(minus_two, _mm_mul_ps(real,imag))); + real = _mm_add_ps(_mm_sub_ps(real2,imag2),c_real_vector); } - else{ // Temporary colors with random multipliers - img[position] = ((value*3) % 255); - img[position + 1] = ((value*2) % 255) ; - img[position + 2] = ((value*5) % 255); + + + __m128i mod64 = _mm_and_si128(iterations, _mm_set1_epi32(63)); + __m128i times4 = _mm_slli_epi32(mod64, 2); + __m128i values = _mm_sub_epi32(_mm_set1_epi32(255), times4); + + __m128i max_iter = _mm_set1_epi32(args->n); + __m128i is_max = _mm_cmpeq_epi32(iterations, max_iter); + values = _mm_andnot_si128(is_max, values); + + unsigned int value_array[4]; + _mm_storeu_si128((__m128i*)value_array,values); + + for(unsigned i = 0 ; i < 4;i++) { + unsigned int value = value_array[i]; + size_t position = (y*args->width + (x+i)) * bytes_per_color; + + if(!args->color) { + args->img[position] = value; + } + else { + args->img[position] = ((value*3) % 255); + args->img[position + 1] = ((value*2) % 255); + args->img[position + 2] = ((value*5) % 255); + } } + c_real_vector = _mm_add_ps(c_real_vector,_mm_set1_ps(four_res)); + } - position += bytesPerColor; - c_real += res; + for(;x < args->width; x++) { + float c_real = c_start_real + (x*args->res); + unsigned int value = converge(c_real,c_imag,args->n); + size_t position = (y*args->width + x) * bytes_per_color; + + if(!args->color) { + args->img[position] = value; + } + else { + args->img[position] = ((value*3) % 255); + args->img[position + 1] = ((value*2) % 255); + args->img[position + 2] = ((value*5) % 255); + } } c_imag -= vertical_step; + c_imag_vector = _mm_sub_ps(c_imag_vector,vertical_step_vector); } + free(args); + args = NULL; } // Iterative application of the equation given by the assignment -static inline unsigned converge(float c_real, float c_imag, unsigned n) { - const float convergenceLimit = 4.0f; +static unsigned converge(float c_real, float c_imag, unsigned n) { float real = 0.0f; float imag = 0.0f; @@ -66,12 +181,21 @@ static inline unsigned converge(float c_real, float c_imag, unsigned n) { float real2 = real * real; float imag2 = imag * imag; - if (real2 + imag2 > convergenceLimit) { + if (real2 + imag2 > 4.0f) { return 255 - 4 * (i % 64); } imag = -2.0f * real * imag + c_imag; real = real2 - imag2 + c_real; } - return 0; + return 0; +} +static inline size_t get_thread_count(){ + #ifdef WIN32 + SYSTEM_INFO sysinfo; + GetSystemInfo(&sysinfo); + return sysinfo.dwNumberOfProcessors; + #else + return sysconf(_SC_NPROCESSORS_ONLN); + #endif } \ No newline at end of file diff --git a/Implementierung/tricorn_V4.h b/Implementierung/tricorn_V4.h index 528f773..25ad3bd 100644 --- a/Implementierung/tricorn_V4.h +++ b/Implementierung/tricorn_V4.h @@ -2,4 +2,12 @@ #include #include #include +#include +#include +#ifdef WIN32 +#include +#else +#include + +#endif void tricorn_V4(float complex start, size_t width, ssize_t height, float res, unsigned n, _Bool color, unsigned char* img); \ No newline at end of file