Revert "let gcc choose how to reference memory addresses in i420_rgx mmx asm"
authorRafaël Carré <funman@videolan.org>
Wed, 26 Mar 2008 07:05:01 +0000 (08:05 +0100)
committerRafaël Carré <funman@videolan.org>
Wed, 26 Mar 2008 07:05:01 +0000 (08:05 +0100)
This reverts commit 6c50d21cebc1c6c809e56400f4cc5ed7aa6856a4.
Breaks compilation on OSX

modules/video_chroma/i420_rgb_mmx.h

index e9c32fb..cb7a9d8 100644 (file)
 #define USED_U64(foo) \
     static const uint64_t foo __asm__ (#foo) __attribute__((unused))
 #endif
-USED_U64(mmx_80w)     = 0x0080008000800080ULL; /* Will be referenced as %4
-                                                * in inline asm */
-USED_U64(mmx_10w)     = 0x1010101010101010ULL; /* -- as %5 */
-USED_U64(mmx_00ffw)   = 0x00ff00ff00ff00ffULL; /* -- as %6 */
-USED_U64(mmx_Y_coeff) = 0x253f253f253f253fULL; /* -- as %7 */
-
-USED_U64(mmx_U_green) = 0xf37df37df37df37dULL; /* -- as %8 */
-USED_U64(mmx_U_blue)  = 0x4093409340934093ULL; /* -- as %9 */
-USED_U64(mmx_V_red)   = 0x3312331233123312ULL; /* -- as %10 */
-USED_U64(mmx_V_green) = 0xe5fce5fce5fce5fcULL; /* -- as %11 */
-
-USED_U64(mmx_mask_f8) = 0xf8f8f8f8f8f8f8f8ULL; /* -- as %12 */
-USED_U64(mmx_mask_fc) = 0xfcfcfcfcfcfcfcfcULL; /* -- as %13 */
+USED_U64(mmx_80w)     = 0x0080008000800080ULL;
+USED_U64(mmx_10w)     = 0x1010101010101010ULL;
+USED_U64(mmx_00ffw)   = 0x00ff00ff00ff00ffULL;
+USED_U64(mmx_Y_coeff) = 0x253f253f253f253fULL;
+
+USED_U64(mmx_U_green) = 0xf37df37df37df37dULL;
+USED_U64(mmx_U_blue)  = 0x4093409340934093ULL;
+USED_U64(mmx_V_red)   = 0x3312331233123312ULL;
+USED_U64(mmx_V_green) = 0xe5fce5fce5fce5fcULL;
+
+USED_U64(mmx_mask_f8) = 0xf8f8f8f8f8f8f8f8ULL;
+USED_U64(mmx_mask_fc) = 0xfcfcfcfcfcfcfcfcULL;
 #undef USED_U64
 
 #if defined(CAN_COMPILE_MMX)
@@ -60,16 +59,18 @@ USED_U64(mmx_mask_fc) = 0xfcfcfcfcfcfcfcfcULL; /* -- as %13 */
         MMX_INSTRUCTIONS                \
         :                               \
         : "r" (p_y), "r" (p_u),         \
-          "r" (p_v), "r" (p_buffer),    \
-          "m" (mmx_80w), "m" (mmx_10w), \
-          "m" (mmx_00ffw), "m" (mmx_Y_coeff), \
-          "m" (mmx_U_green), "m" (mmx_U_blue), \
-          "m" (mmx_V_red), "m" (mmx_V_green), \
-          "m" (mmx_mask_f8), "m" (mmx_mask_fc) );  \
+          "r" (p_v), "r" (p_buffer) );  \
     } while(0)
 
 #define MMX_END __asm__ __volatile__ ( "emms" )
 
+/* Use RIP-relative code in PIC mode on amd64 */
+#if defined(__x86_64__) && defined(__PIC__)
+#   define G "(%%rip)"
+#else
+#   define G
+#endif
+
 #define MMX_INIT_16 "                                                       \n\
 movd       (%1), %%mm0      # Load 4 Cb       00 00 00 00 u3 u2 u1 u0       \n\
 movd       (%2), %%mm1      # Load 4 Cr       00 00 00 00 v3 v2 v1 v0       \n\
@@ -102,27 +103,27 @@ movq      (%0), %%mm6       # Load 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0       \n\
 # convert the chroma part                                                   \n\
 punpcklbw %%mm4, %%mm0          # scatter 4 Cb    00 u3 00 u2 00 u1 00 u0   \n\
 punpcklbw %%mm4, %%mm1          # scatter 4 Cr    00 v3 00 v2 00 v1 00 v0   \n\
-psubsw    %4, %%mm0             # Cb -= 128                                 \n\
-psubsw    %4, %%mm1             # Cr -= 128                                 \n\
+psubsw    mmx_80w"G", %%mm0     # Cb -= 128                                 \n\
+psubsw    mmx_80w"G", %%mm1     # Cr -= 128                                 \n\
 psllw     $3, %%mm0             # Promote precision                         \n\
 psllw     $3, %%mm1             # Promote precision                         \n\
 movq      %%mm0, %%mm2          # Copy 4 Cb       00 u3 00 u2 00 u1 00 u0   \n\
 movq      %%mm1, %%mm3          # Copy 4 Cr       00 v3 00 v2 00 v1 00 v0   \n\
-pmulhw    %8, %%mm2             # Mul Cb with green coeff -> Cb green       \n\
-pmulhw    %11, %%mm3            # Mul Cr with green coeff -> Cr green       \n\
-pmulhw    %9, %%mm0             # Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0   \n\
-pmulhw    %10, %%mm1            # Mul Cr -> Cred  00 r3 00 r2 00 r1 00 r0   \n\
+pmulhw    mmx_U_green"G", %%mm2 # Mul Cb with green coeff -> Cb green       \n\
+pmulhw    mmx_V_green"G", %%mm3 # Mul Cr with green coeff -> Cr green       \n\
+pmulhw    mmx_U_blue"G", %%mm0  # Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0   \n\
+pmulhw    mmx_V_red"G", %%mm1   # Mul Cr -> Cred  00 r3 00 r2 00 r1 00 r0   \n\
 paddsw    %%mm3, %%mm2          # Cb green + Cr green -> Cgreen             \n\
                                                                             \n\
 # convert the luma part                                                     \n\
-psubusb   %5, %%mm6             # Y -= 16                                   \n\
+psubusb   mmx_10w"G", %%mm6     # Y -= 16                                   \n\
 movq      %%mm6, %%mm7          # Copy 8 Y        Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0   \n\
-pand      %6, %%mm6             # get Y even      00 Y6 00 Y4 00 Y2 00 Y0   \n\
+pand      mmx_00ffw"G", %%mm6   # get Y even      00 Y6 00 Y4 00 Y2 00 Y0   \n\
 psrlw     $8, %%mm7             # get Y odd       00 Y7 00 Y5 00 Y3 00 Y1   \n\
 psllw     $3, %%mm6             # Promote precision                         \n\
 psllw     $3, %%mm7             # Promote precision                         \n\
-pmulhw    %7, %%mm6             # Mul 4 Y even    00 y6 00 y4 00 y2 00 y0   \n\
-pmulhw    %7, %%mm7             # Mul 4 Y odd     00 y7 00 y5 00 y3 00 y1   \n\
+pmulhw    mmx_Y_coeff"G", %%mm6 # Mul 4 Y even    00 y6 00 y4 00 y2 00 y0   \n\
+pmulhw    mmx_Y_coeff"G", %%mm7 # Mul 4 Y odd     00 y7 00 y5 00 y3 00 y1   \n\
 "
 
 /*
@@ -167,14 +168,14 @@ punpcklbw %%mm5, %%mm2          #                 G7 G6 G5 G4 G3 G2 G1 G0   \n\
 
 #define MMX_YUV_GRAY "                                                      \n\
 # convert the luma part                                                     \n\
-psubusb   %5, %%mm6                                                         \n\
+psubusb   mmx_10w"G", %%mm6                                                 \n\
 movq      %%mm6, %%mm7                                                      \n\
-pand      %6, %%mm6                                                         \n\
+pand      mmx_00ffw"G", %%mm6                                               \n\
 psrlw     $8, %%mm7                                                         \n\
 psllw     $3, %%mm6                                                         \n\
 psllw     $3, %%mm7                                                         \n\
-pmulhw    %7, %%mm6                                                         \n\
-pmulhw    %7, %%mm7                                                         \n\
+pmulhw    mmx_Y_coeff"G", %%mm6                                             \n\
+pmulhw    mmx_Y_coeff"G", %%mm7                                             \n\
 packuswb  %%mm6, %%mm6                                                      \n\
 packuswb  %%mm7, %%mm7                                                      \n\
 punpcklbw %%mm7, %%mm6                                                      \n\
@@ -182,8 +183,8 @@ punpcklbw %%mm7, %%mm6                                                      \n\
 
 #define MMX_UNPACK_16_GRAY "                                                \n\
 movq      %%mm6, %%mm5                                                      \n\
-pand      %12, %%mm6                                                        \n\
-pand      %13, %%mm5                                                        \n\
+pand      mmx_mask_f8"G", %%mm6                                             \n\
+pand      mmx_mask_fc"G", %%mm5                                             \n\
 movq      %%mm6, %%mm7                                                      \n\
 psrlw     $3, %%mm7                                                         \n\
 pxor      %%mm3, %%mm3                                                      \n\
@@ -212,10 +213,10 @@ movq      %%mm2, 8(%3)                                                      \n\
 
 #define MMX_UNPACK_15 "                                                     \n\
 # mask unneeded bits off                                                    \n\
-pand      %12, %%mm0            # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
+pand      mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
 psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
-pand      %12, %%mm2            # g7g6g5g4 g3______ g7g6g5g4 g3______       \n\
-pand      %12, %%mm1            # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
+pand      mmx_mask_f8"G", %%mm2 # g7g6g5g4 g3______ g7g6g5g4 g3______       \n\
+pand      mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
 psrlw     $1,%%mm1              # __r7r6r5 r4r3____ __r7r6r5 r4r3____       \n\
 pxor      %%mm4, %%mm4          # zero mm4                                  \n\
 movq      %%mm0, %%mm5          # Copy B7-B0                                \n\
@@ -248,9 +249,9 @@ movq      %%mm5, 8(%3)          # store pixel 4-7                           \n\
 
 #define MMX_UNPACK_16 "                                                     \n\
 # mask unneeded bits off                                                    \n\
-pand      %12, %%mm0            # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
-pand      %13, %%mm2            # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____       \n\
-pand      %12, %%mm1            # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
+pand      mmx_mask_f8"G", %%mm0 # b7b6b5b4 b3______ b7b6b5b4 b3______       \n\
+pand      mmx_mask_fc"G", %%mm2 # g7g6g5g4 g3g2____ g7g6g5g4 g3g2____       \n\
+pand      mmx_mask_f8"G", %%mm1 # r7r6r5r4 r3______ r7r6r5r4 r3______       \n\
 psrlw     $3,%%mm0              # ______b7 b6b5b4b3 ______b7 b6b5b4b3       \n\
 pxor      %%mm4, %%mm4          # zero mm4                                  \n\
 movq      %%mm0, %%mm5          # Copy B7-B0                                \n\