Re: [AD] DO_PARALLELOGRAM_MAP again

[Peter Wang <novalazy@xxxxxxxxxx>]

On 2010-09-11, Pavel Sountsov <slabode@xxxxxxxxxx> wrote:
> 
> Anyway, it's not like the drawer from the primitives addon is fatally 
> slower... it is slow for two reasons: it uses function pointers instead 
> of macro inlining; it uses al_put_pixel calls instead of direct memory 
> writes. Both can be fixed in short order, but as you can imagine they 
> make the development of those functions harder... (which is the very 
> problem of parallelogram_map) so I did not do them. If that drawer 
> becomes used for all bitmap drawing, it won't be a problem to speed it 
> up via some macro work.

Yes, I see.  With this quick hack `ex_blend_bench 3' is nearly as fast
as `ex_blend_bench 2' on my machine.  Restoring the commented out lines
in fix_var makes it about twice as slow.

Peter


diff --git a/addons/primitives/tri_soft.c b/addons/primitives/tri_soft.c
index 2c17f3f..5ba84dd 100644
--- a/addons/primitives/tri_soft.c
+++ b/addons/primitives/tri_soft.c
@@ -245,11 +245,12 @@ static void shader_grad_any_step(uintptr_t state, int minor)
 static int fix_var(float var, int max_var)
 {
    const int ivar = (int)floorf(var);
-   const int ret = ivar % max_var;
-   if(ret >= 0)
-      return ret;
-   else
-      return ret + max_var;
+//    const int ret = ivar % max_var;
+//    if(ret >= 0)
+//       return ret;
+//    else
+//       return ret + max_var;
+   return ivar;
 }
 
 #define SHADE_COLORS(A, B) \
@@ -277,6 +278,9 @@ typedef struct {
    int w, h;
 } state_texture_solid_any_2d;
 
+#include <allegro5/internal/aintern_bitmap.h>
+#include <allegro5/internal/aintern_pixels.h>
+
 static void shader_texture_solid_any_draw_shade(uintptr_t state, int x1, int y, int x2)
 {
    state_texture_solid_any_2d* s = (state_texture_solid_any_2d*)state;
@@ -303,9 +307,39 @@ static void shader_texture_solid_any_draw_shade_white(uintptr_t state, int x1, i
    int x;
    ALLEGRO_COLOR color;
    
-   for (x = x1; x <= x2; x++) {
-      color = al_get_pixel(s->texture, fix_var(u, s->w), fix_var(v, s->h));
-      al_put_blended_pixel(x, y - 1, color);
+   ALLEGRO_BITMAP *target = al_get_target_bitmap();
+
+   if (y < target->lock_y || y >= target->lock_y + target->lock_h)
+      return;
+
+   for (x = x1; x <= target->lock_x; x++) {
+      u += s->du_dx;
+      v += s->dv_dx;
+   }
+
+   const ALLEGRO_LOCKED_REGION *lr = &target->locked_region;
+   const int format = lr->format;
+   unsigned char *data = lr->data;
+   data += (y-1 - target->lock_y) * lr->pitch;
+   data += x * al_get_pixel_size(lr->format);
+
+   const int src_format = s->texture->locked_region.format;
+   const int src_size = al_get_pixel_size(src_format);
+
+   x -= target->lock_x;
+   x2 -= target->lock_x;
+   if (x2 > target->lock_w)
+      x2 = target->lock_w;
+
+   for (; x <= x2; x++) {
+      //color = al_get_pixel(s->texture, fix_var(u, s->w), fix_var(v, s->h));
+      unsigned char *src_data = s->texture->locked_region.data;
+      src_data += s->texture->locked_region.pitch * fix_var(v, s->h);
+      src_data += src_size * fix_var(u, s->w);
+      _AL_INLINE_GET_PIXEL(src_format, src_data, color, false);
+
+      // al_put_blended_pixel(x, y - 1, color);
+      _AL_INLINE_PUT_PIXEL(format, data, color, true);
       
       u += s->du_dx;
       v += s->dv_dx;