From Mike Frysinger and/or Gentoo:

- libsdl-PIC-load-mmx-masks-from-stack.patch
this one may be a little controversial ... the fix here is again that you cant
reference the memory addresses like this to load into a mmx register, so the
way to do it is to push two 32bit words onto the stack, load the 64bit value
off of the stack into the mmx register, and then adjust the stack so that
it's back to normal.

--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%401233

src/hermes/mmxp2_32.asm

@@ -29,31 +29,43 @@ GLOBAL _ConvertMMXpII32_16BGR555
 
 EXTERN _mmxreturn
  
-SECTION .data
-	
-ALIGN 8
+;; Macros for conversion routines
+
+%macro _push_immq_mask 1
+	push dword %1
+	push dword %1
+%endmacro
 
-;; Constants for conversion routines
+%macro load_immq 2
+	_push_immq_mask %2
+	movq %1, [esp]
+%endmacro
 
-mmx32_rgb888_mask dd 00ffffffh,00ffffffh
+%macro pand_immq 2
+	_push_immq_mask %2
+	pand %1, [esp]
+%endmacro
 
-mmx32_rgb565_b dd 000000f8h, 000000f8h
-mmx32_rgb565_g dd 0000fc00h, 0000fc00h
-mmx32_rgb565_r dd 00f80000h, 00f80000h
+%define CLEANUP_IMMQ_LOADS(num) \
+	add esp, byte 8 * num
 
-mmx32_rgb555_rb dd 00f800f8h,00f800f8h
-mmx32_rgb555_g dd 0000f800h,0000f800h
-mmx32_rgb555_mul dd 20000008h,20000008h
-mmx32_bgr555_mul dd 00082000h,00082000h
+%define mmx32_rgb888_mask 00ffffffh
+%define mmx32_rgb565_b 000000f8h
+%define mmx32_rgb565_g 0000fc00h
+%define mmx32_rgb565_r 00f80000h
 
+%define mmx32_rgb555_rb 00f800f8h
+%define mmx32_rgb555_g 0000f800h
+%define mmx32_rgb555_mul 20000008h
+%define mmx32_bgr555_mul 00082000h
 
-			
 SECTION .text
 
 _ConvertMMXpII32_24RGB888:
 
         ; set up mm6 as the mask, mm7 as zero
-        movq mm6, qword [mmx32_rgb888_mask]
+        load_immq mm6, mmx32_rgb888_mask
+        CLEANUP_IMMQ_LOADS(1)
         pxor mm7, mm7
 
         mov edx, ecx                    ; save ecx
@@ -115,9 +127,10 @@ _ConvertMMXpII32_24RGB888:
 _ConvertMMXpII32_16RGB565:
 
         ; set up masks
-        movq mm5, [mmx32_rgb565_b]
-        movq mm6, [mmx32_rgb565_g]
-        movq mm7, [mmx32_rgb565_r]
+        load_immq mm5, mmx32_rgb565_b
+        load_immq mm6, mmx32_rgb565_g
+        load_immq mm7, mmx32_rgb565_r
+        CLEANUP_IMMQ_LOADS(3)
 
         mov edx, ecx
         shr ecx, 2
@@ -181,9 +194,10 @@ _ConvertMMXpII32_16RGB565:
 	
 _ConvertMMXpII32_16BGR565:
 
-        movq mm5, [mmx32_rgb565_r]
-        movq mm6, [mmx32_rgb565_g]
-        movq mm7, [mmx32_rgb565_b]
+        load_immq mm5, mmx32_rgb565_r
+        load_immq mm6, mmx32_rgb565_g
+        load_immq mm7, mmx32_rgb565_b
+        CLEANUP_IMMQ_LOADS(3)
 
         mov edx, ecx
         shr ecx, 2
@@ -253,7 +267,7 @@ _ConvertMMXpII32_16BGR555:
         ; except it uses a different multiplier for the pmaddwd
         ; instruction.  cool huh.
 
-        movq mm7, qword [mmx32_bgr555_mul]
+        load_immq mm7, mmx32_bgr555_mul
         jmp _convert_bgr555_cheat
 
 ; This is the same as the Intel version.. they obviously went to
@@ -263,9 +277,10 @@ _ConvertMMXpII32_16BGR555:
 ; (I think) a more accurate name..
 _ConvertMMXpII32_16RGB555:
 
-        movq mm7,qword [mmx32_rgb555_mul]
+	load_immq mm7, mmx32_rgb555_mul
 _convert_bgr555_cheat:
-        movq mm6,qword [mmx32_rgb555_g]
+	load_immq mm6, mmx32_rgb555_g
+	CLEANUP_IMMQ_LOADS(2)
         
 	mov edx,ecx		           ; Save ecx 
 
@@ -280,12 +295,14 @@ _convert_bgr555_cheat:
 	movq mm0,[esi]
 	movq mm3,mm2
 
-	pand mm3,qword [mmx32_rgb555_rb]
+	pand_immq mm3, mmx32_rgb555_rb
 	movq mm1,mm0
 
-	pand mm1,qword [mmx32_rgb555_rb]
+	pand_immq mm1, mmx32_rgb555_rb
 	pmaddwd mm3,mm7
 
+	CLEANUP_IMMQ_LOADS(2)
+
 	pmaddwd mm1,mm7
 	pand mm2,mm6
 
@@ -302,13 +319,13 @@ _convert_bgr555_cheat:
 	movq mm0,mm4
 	psrld mm1,6
 
-	pand mm0,qword [mmx32_rgb555_rb]
+	pand_immq mm0, mmx32_rgb555_rb
 	packssdw mm1,mm3
 
 	movq mm3,mm5
 	pmaddwd mm0,mm7
 
-	pand mm3,qword [mmx32_rgb555_rb]
+	pand_immq mm3, mmx32_rgb555_rb
 	pand mm4,mm6
 
 	movq [edi],mm1			
@@ -329,12 +346,14 @@ _convert_bgr555_cheat:
 	movq mm3,mm2
 	movq mm1,mm0
 
-	pand mm3,qword [mmx32_rgb555_rb]
+	pand_immq mm3, mmx32_rgb555_rb
 	packssdw mm5,mm4
 
-	pand mm1,qword [mmx32_rgb555_rb]
+	pand_immq mm1, mmx32_rgb555_rb
 	pand mm2,mm6
 
+	CLEANUP_IMMQ_LOADS(4)
+
 	movq [edi+8],mm5
 	pmaddwd mm3,mm7