blurring stuff should work perfectly fine now

blur.c

@@ -32,7 +32,6 @@ blur_image_surface (cairo_surface_t *surface, int radius)
 {
     cairo_surface_t *tmp;
     int width, height;
-//    int src_stride, dst_stride;
     uint32_t *src, *dst;
 
     if (cairo_surface_status (surface))
@@ -64,15 +63,18 @@ blur_image_surface (cairo_surface_t *surface, int radius)
     return;
 
     src = (uint32_t*)cairo_image_surface_get_data (surface);
-//    src_stride = cairo_image_surface_get_stride (surface);
 
     dst = (uint32_t*)cairo_image_surface_get_data (tmp);
-//    dst_stride = cairo_image_surface_get_stride (tmp);
-
-    //blur_impl_naive(src, dst, width, height, src_stride, dst_stride, 10000);
-    //blur_impl_sse2(src, dst, width, height, 4.5);
+    
+#ifdef __SSE4_1__
     blur_impl_ssse3(src, dst, width, height, 4.5);
-
+#elif __SSE2__
+    blur_impl_sse2(src, dst, width, height, 4.5);
+#else
+    int src_stride = cairo_image_surface_get_stride (surface);
+    int dst_stride = cairo_image_surface_get_stride (tmp);
+    blur_impl_naive(src, dst, width, height, src_stride, dst_stride, 10000);
+#endif
     cairo_surface_destroy (tmp);
     cairo_surface_flush (surface);
     cairo_surface_mark_dirty (surface);

blur_simd.c

@@ -51,6 +51,7 @@ void blur_impl_sse2(uint32_t *src, uint32_t *dst, int width, int height, float s
 }
 
 void blur_impl_horizontal_pass_sse2(uint32_t *src, uint32_t *dst, float *kernel, int width, int height) {
+    uint32_t* o_src = src;
     for (int row = 0; row < height; row++) {
         for (int column = 0; column < width; column++, src++) {
             __m128i rgbaIn[REGISTERS_CNT];
@@ -59,7 +60,7 @@ void blur_impl_horizontal_pass_sse2(uint32_t *src, uint32_t *dst, float *kernel,
             int leftBorder = column < HALF_KERNEL;
             int rightBorder = column > width - HALF_KERNEL;
             if (leftBorder || rightBorder) {
-                uint32_t _rgbaIn[KERNEL_SIZE] ALIGN16;
+                uint32_t _rgbaIn[KERNEL_SIZE + 1] ALIGN16;
                 int i = 0;
                 if (leftBorder) {
                     // for kernel size 7x7 and column == 0, we have:
@@ -80,8 +81,22 @@ void blur_impl_horizontal_pass_sse2(uint32_t *src, uint32_t *dst, float *kernel,
                 for (int k = 0; k < REGISTERS_CNT; k++)
                     rgbaIn[k] = _mm_load_si128((__m128i*)(_rgbaIn + 4*k));
             } else {
-                for (int k = 0; k < REGISTERS_CNT; k++)
+                for (int k = 0; k < REGISTERS_CNT; k++) {
+#if 0
+                    printf("%p -> %p (%ld) || %p->%p\n", 
+                        o_src,
+                        o_src + (height * width),
+                        o_src + (height * width) - src,
+                        src + 4*k - HALF_KERNEL, 
+                        ((__m128i*)src + 4*k - HALF_KERNEL) + 1
+                    );
+#endif
+                    // if this copy would go out of bounds, break
+                    if ((long long) (((__m128i*) src + 4*k - HALF_KERNEL) + 1) 
+                            > (long long) (o_src + (height * width)))
+                        break;
                     rgbaIn[k] = _mm_loadu_si128((__m128i*)(src + 4*k - HALF_KERNEL));
+                }
             }
 
             // unpack each pixel, convert to float,
@@ -198,7 +213,9 @@ void blur_impl_horizontal_pass_ssse3(uint32_t *src, uint32_t *dst, int8_t *kerne
                 }
             } else {
                 for (int k = 0; k < REGISTERS_CNT; k++) {
-                    if ((long long)(((__m128i*) src + 4*k - HALF_KERNEL) + 1) > (long long)((o_src + (width * height)))) break;
+                    if ((long long) (((__m128i*) src + 4*k - HALF_KERNEL) + 1) 
+                            > (long long) (o_src + (height * width)))
+                        break;
 #if 0
                     printf("K: %d; p: %p -> %p\n", k, src+4*k - HALF_KERNEL, ((__m128i*) (src +4*k - HALF_KERNEL)) + 1);
                     printf("%p->%p, %p->%p (%ld)\n", (__m128i*) src + 4*k - HALF_KERNEL, ((__m128i*) src + 4*k - HALF_KERNEL) + 1, o_src, o_src + (width * height), o_src + (width * height) - src);

configure.ac

@@ -101,8 +101,8 @@ AC_PROG_LN_S
 AM_PROG_AR
 
 AX_FLAGS_WARN_ALL
-AX_APPEND_FLAG([-msse4.1], [AM_CFLAGS])
-AX_APPEND_FLAG([-O2], [AM_CFLAGS])
+AX_APPEND_FLAG([-march=native], [AM_CFLAGS])
+ AX_APPEND_FLAG([-O2], [AM_CFLAGS])
 AX_APPEND_FLAG([-funroll-loops], [AM_CFLAGS])
 AX_APPEND_FLAG([-std=gnu99], [AM_CFLAGS])
 AX_CHECK_COMPILE_FLAG([-Wunused-value], [AX_APPEND_FLAG([-Wunused-value], [AM_CFLAGS])])

m4/ax_check_enable_debug.m4

@@ -76,11 +76,11 @@ AC_DEFUN([AX_CHECK_ENABLE_DEBUG],[
     AS_CASE([$enable_debug],
       [yes],[
 	AC_MSG_RESULT(yes)
-	CFLAGS="${CFLAGS} -g -O2"
-	CXXFLAGS="${CXXFLAGS} -g -O2"
-	FFLAGS="${FFLAGS} -g -O2"
-	FCFLAGS="${FCFLAGS} -g -O2"
-	OBJCFLAGS="${OBJCFLAGS} -g -O2"
+	CFLAGS="${CFLAGS} -g -O0"
+	CXXFLAGS="${CXXFLAGS} -g -O0"
+	FFLAGS="${FFLAGS} -g -O0"
+	FCFLAGS="${FCFLAGS} -g -O0"
+	OBJCFLAGS="${OBJCFLAGS} -g -O0"
       ],
       [info],[
 	AC_MSG_RESULT(info)

m4/ax_code_coverage.m4

@@ -141,8 +141,8 @@ AC_DEFUN([AX_CODE_COVERAGE],[
 
 		dnl Build the code coverage flags
 		CODE_COVERAGE_CPPFLAGS="-DNDEBUG"
-		CODE_COVERAGE_CFLAGS="-O2 -g -fprofile-arcs -ftest-coverage"
-		CODE_COVERAGE_CXXFLAGS="-O2 -g -fprofile-arcs -ftest-coverage"
+		CODE_COVERAGE_CFLAGS="-O0 -g -fprofile-arcs -ftest-coverage"
+		CODE_COVERAGE_CXXFLAGS="-O0 -g -fprofile-arcs -ftest-coverage"
 		CODE_COVERAGE_LDFLAGS="-lgcov"
 
 		AC_SUBST([CODE_COVERAGE_CPPFLAGS])