diff --git a/Makefile b/Makefile
index 3e5114f..4c4a095 100644
--- a/Makefile
+++ b/Makefile
@@ -79,7 +79,7 @@ build/app.exe: $(OBJS) MakeFile
 	$(LINK) @build/link.lnk
 
 clean:
-	rm -rf build
+	rm -rf build/*.obj build/*.exe
 	# Nettoyage des logs d'erreurs
 	rm -f *.err src/**/*.err
 
diff --git a/src/part3D/defines.h b/src/part3D/defines.h
index c4e7cf9..5434220 100644
--- a/src/part3D/defines.h
+++ b/src/part3D/defines.h
@@ -50,8 +50,28 @@ typedef int32_t fixed;
 #define f_to_int(a)    ((a) >> 16)
 #define int_to_f(a)    ((a) << 16)
 
-static __inline fixed f_mul(fixed a, fixed b) { return (fixed)(((int64_t)a * b) >> 16); }
-static __inline fixed f_div(fixed a, fixed b) { if (b == 0) return 0; return (fixed)(((int64_t)a << 16) / b); }
+//static __inline fixed f_mul(fixed a, fixed b) { return (fixed)(((int64_t)a * b) >> 16); }
+//static __inline fixed f_div(fixed a, fixed b) { if (b == 0) return 0; return (fixed)(((int64_t)a << 16) / b); }
+/* Optimisation OpenWatcom Inline Assembly pour mathématiques 16.16 */
+fixed f_mul(fixed a, fixed b);
+#pragma aux f_mul = \
+    "imul   edx" \
+    "shrd   eax, edx, 16" \
+    parm [eax] [edx] value [eax] modify [edx];
+
+fixed f_div(fixed a, fixed b);
+#pragma aux f_div = \
+    "test   ebx, ebx" \
+    "jz     div_zero" \
+    "cdq" \
+    "shld   edx, eax, 16" \
+    "sal    eax, 16" \
+    "idiv   ebx" \
+    "jmp    end_div" \
+"div_zero:" \
+    "xor    eax, eax" \
+"end_div:" \
+    parm [eax] [ebx] value [eax] modify [edx];
 
 typedef struct { fixed m[3][3]; } Matrix3;
 typedef struct { fixed x, y, z; } Vector3;
diff --git a/src/part3D/graph.c b/src/part3D/graph.c
index 1eb7f50..dfffda0 100644
--- a/src/part3D/graph.c
+++ b/src/part3D/graph.c
@@ -26,13 +26,30 @@ uint8_t *vga = (uint8_t *)0xA0000;
 void set_vga_mode(int mode);
 #pragma aux set_vga_mode = "int 0x10" parm [ax];
 
+void fast_clear_dwords(void *dest, uint32_t val, int count);
+#pragma aux fast_clear_dwords = \
+    "cld" \
+    "rep 	stosd" \
+    parm [edi] [eax] [ecx] modify [edi ecx];
+
+void fast_copy_dwords(void *dest, void *src, int count);
+#pragma aux fast_copy_dwords = \
+    "cld" \
+    "rep	movsd" \
+    parm [edi] [esi] [ecx] modify [edi esi ecx];
+
 void clear_buffers(uint8_t color) {
-	memset(backbuffer, color, SCREEN_PIXELS);
-	memset(zbuffer, 0xFF, SCREEN_PIXELS * sizeof(uint16_t));
+	//memset(backbuffer, color, SCREEN_PIXELS);
+	uint32_t c32 = color | (color << 8) | (color << 16) | (color << 24);
+	fast_clear_dwords(backbuffer, c32, SCREEN_PIXELS / 4);
+
+	//memset(zbuffer, 0xFF, SCREEN_PIXELS * sizeof(uint16_t));
+	fast_clear_dwords(zbuffer, 0xFFFFFFFFUL, SCREEN_PIXELS / 2); /* 2 pixels (16 bits) par itération 32 bits */
 }
 
 void clear_zbuffer(void) {
-	memset(zbuffer, 0xFF, SCREEN_PIXELS * sizeof(uint16_t));
+	//memset(zbuffer, 0xFF, SCREEN_PIXELS * sizeof(uint16_t));
+	fast_clear_dwords(zbuffer, 0xFFFFFFFFUL, SCREEN_PIXELS / 2);
 }
 
 void blit_image(Image *img, int x, int y) {
@@ -106,6 +123,7 @@ void apply_256_palette(RGB *pal) {
 }
 
 void flip(void) {
+	//fast_copy_dwords(vga, backbuffer, SCREEN_PIXELS / 4);
 	memcpy(vga, backbuffer, SCREEN_PIXELS);
 }