altivec gcc-3 fixes by (Magnus Damm <damm at opensource dot se>)

Originally committed as revision 1896 to svn://svn.ffmpeg.org/ffmpeg/trunk

diff --git a/configure b/configure
index c02d43e..67b9fce 100755 (executable)
--- a/configure
+++ b/configure
@@ -343,8 +343,29 @@ if test $altivec = "default"; then
     fi
 fi
 
+# See if we have <altivec.h>
+cat > $TMPC << EOF
+#include <altivec.h>
+int main( void ) { return 0; }
+EOF
+
+_altivec_h="no"
+if $cc -o $TMPE $TMPC 2> /dev/null ; then
+_altivec_h="yes"
+fi
+
 # See does our compiler support Motorola AltiVec C API
 if test $altivec = "yes"; then
+if test $_altivec_h = "yes"; then
+cat > $TMPC << EOF
+#include <altivec.h>
+int main(void) {
+    vector signed int v1, v2, v3;
+    v1 = vec_add(v2,v3);
+    return 0;
+}
+EOF
+else
 cat > $TMPC << EOF
 int main(void) {
     vector signed int v1, v2, v3;
@@ -352,7 +373,12 @@ int main(void) {
     return 0;
 }
 EOF
+fi
+if test "$darwin" = "yes"; then
 $cc -o $TMPE $TMPC -faltivec 2> /dev/null || altivec="no"
+else
+$cc -o $TMPE $TMPC -maltivec -mabi=altivec 2> /dev/null || altivec="no"
+fi
 fi
 
 # Can only do mmi on mips
@@ -742,6 +768,11 @@ if test "$altivec" = "yes" ; then
   echo "#define HAVE_ALTIVEC 1" >> $TMPH
   echo "// Enable the next line to use the reference C code instead of AltiVec" >> $TMPH
   echo "// #define ALTIVEC_USE_REFERENCE_C_CODE 1" >> $TMPH
+  if test "$_altivec_h" = "yes" ; then
+    echo "#define HAVE_ALTIVEC_H 1" >> $TMPH
+  else
+    echo "#undef HAVE_ALTIVEC_H 1" >> $TMPH
+  fi
 fi
 if test "$gprof" = "yes" ; then
   echo "TARGET_GPROF=yes" >> config.mak

diff --git a/libavcodec/ppc/dsputil_altivec.c b/libavcodec/ppc/dsputil_altivec.c
index d48332b..32e881b 100644 (file)
--- a/libavcodec/ppc/dsputil_altivec.c
+++ b/libavcodec/ppc/dsputil_altivec.c
@@ -19,6 +19,9 @@
  */
  
 #include "../dsputil.h"
+
+#include "gcc_fixes.h"
+
 #include "dsputil_altivec.h"
 
 #ifdef CONFIG_DARWIN
@@ -303,11 +306,8 @@ int pix_abs8x8_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
     vector signed int sumdiffs;
 
     sad = (vector unsigned int)vec_splat_u32(0);
-#ifdef CONFIG_DARWIN
-    permclear = (vector unsigned char)(255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0);
-#else
-    permclear = (vector unsigned char){255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0};
-#endif
+
+    permclear = (vector unsigned char)AVV(255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0);
 
     for(i=0;i<8;i++) {
        /* Read potentially unaligned pixels into t1 and t2
@@ -387,11 +387,9 @@ int sse8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size)
     vector signed int sumsqr;
     
     sum = (vector unsigned int)vec_splat_u32(0);
-#ifdef CONFIG_DARWIN
-    permclear = (vector unsigned char)(255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0);
-#else
-    permclear = (vector unsigned char){255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0};
-#endif
+
+    permclear = (vector unsigned char)AVV(255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0);
+
     
     for(i=0;i<8;i++) {
        /* Read potentially unaligned pixels into t1 and t2

diff --git a/libavcodec/ppc/fft_altivec.c b/libavcodec/ppc/fft_altivec.c
index 992be5b..75c95bb 100644 (file)
--- a/libavcodec/ppc/fft_altivec.c
+++ b/libavcodec/ppc/fft_altivec.c
@@ -20,6 +20,8 @@
  */
 #include "../dsputil.h"
 
+#include "gcc_fixes.h"
+
 #include "dsputil_altivec.h"
 
 /*

diff --git a/libavcodec/ppc/gcc_fixes.h b/libavcodec/ppc/gcc_fixes.h
new file mode 100644 (file)
index 0000000..855a5b4
--- /dev/null
+++ b/libavcodec/ppc/gcc_fixes.h
@@ -0,0 +1,85 @@
+/*
+ * gcc fixes for altivec.
+ * Used to workaround broken gcc (FSF gcc-3 pre gcc-3.3) 
+ * and to stay somewhat compatible with Darwin.
+ */
+
+#ifndef _GCC_FIXES_
+#define _GCC_FIXES_
+
+#ifdef HAVE_ALTIVEC_H
+#include <altivec.h>
+#endif
+
+#ifdef CONFIG_DARWIN
+#define AVV(x...) (x)
+#else
+#define AVV(x...) {x}
+
+#if (__GNUC__ * 100 + __GNUC_MINOR__ < 303)  
+
+/* This code was provided to me by Bartosch Pixa
+ * as a separate header file (broken_mergel.h).
+ * thanks to lu_zero for the workaround.
+ *
+ * See this mail for more information:
+ * http://gcc.gnu.org/ml/gcc/2003-04/msg00967.html
+ */
+
+static inline vector signed char my_vmrglb (vector signed char const A,
+                                         vector signed char const B)
+{
+    static const vector unsigned char lowbyte = {
+       0x08, 0x18, 0x09, 0x19, 0x0a, 0x1a, 0x0b,  0x1b,
+       0x0c, 0x1c, 0x0d, 0x1d, 0x0e, 0x1e, 0x0f, 0x1f
+    };
+    return vec_perm (A, B, lowbyte);
+}
+
+static inline vector signed short my_vmrglh (vector signed short const A,
+                                         vector signed short const B)
+{
+    static const vector unsigned char lowhalf = {
+       0x08, 0x09, 0x18, 0x19, 0x0a, 0x0b, 0x1a, 0x1b,
+       0x0c, 0x0d, 0x1c, 0x1d, 0x0e, 0x0f, 0x1e, 0x1f
+    };
+    return vec_perm (A, B, lowhalf);
+}
+
+static inline vector signed int my_vmrglw (vector signed int const A,
+                                         vector signed int const B)
+{
+    static const vector unsigned char lowword = {
+       0x08, 0x09, 0x0a, 0x0b, 0x18, 0x19, 0x1a, 0x1b,
+       0x0c, 0x0d, 0x0e, 0x0f, 0x1c, 0x1d, 0x1e, 0x1f
+    };
+    return vec_perm (A, B, lowword);
+}
+/*#define my_vmrglb my_vmrglb 
+#define my_vmrglh my_vmrglh 
+#define my_vmrglw my_vmrglw 
+*/
+#undef vec_mergel
+
+#define vec_mergel(a1, a2) \
+__ch (__bin_args_eq (vector signed char, (a1), vector signed char, (a2)), \
+      ((vector signed char) my_vmrglb ((vector signed char) (a1), (vector signed char) (a2))), \
+__ch (__bin_args_eq (vector unsigned char, (a1), vector unsigned char, (a2)), \
+      ((vector unsigned char) my_vmrglb ((vector signed char) (a1), (vector signed char) (a2))), \
+__ch (__bin_args_eq (vector signed short, (a1), vector signed short, (a2)), \
+      ((vector signed short) my_vmrglh ((vector signed short) (a1), (vector signed short) (a2))), \
+__ch (__bin_args_eq (vector unsigned short, (a1), vector unsigned short, (a2)), \
+      ((vector unsigned short) my_vmrglh ((vector signed short) (a1), (vector signed short) (a2))), \
+__ch (__bin_args_eq (vector float, (a1), vector float, (a2)), \
+      ((vector float) my_vmrglw ((vector signed int) (a1), (vector signed int) (a2))), \
+__ch (__bin_args_eq (vector signed int, (a1), vector signed int, (a2)), \
+      ((vector signed int) my_vmrglw ((vector signed int) (a1), (vector signed int) (a2))), \
+__ch (__bin_args_eq (vector unsigned int, (a1), vector unsigned int, (a2)), \
+      ((vector unsigned int) my_vmrglw ((vector signed int) (a1), (vector signed int) (a2))), \
+    __altivec_link_error_invalid_argument ())))))))
+
+#endif
+
+#endif /* CONFIG_DARWIN */
+
+#endif /* _GCC_FIXES_ */

diff --git a/libavcodec/ppc/gmc_altivec.c b/libavcodec/ppc/gmc_altivec.c
index 515a766..18d52bb 100644 (file)
--- a/libavcodec/ppc/gmc_altivec.c
+++ b/libavcodec/ppc/gmc_altivec.c
@@ -20,6 +20,8 @@
 
 #include "../dsputil.h"
 
+#include "gcc_fixes.h"
+
 #include "dsputil_altivec.h"
 
 /*

diff --git a/libavcodec/ppc/idct_altivec.c b/libavcodec/ppc/idct_altivec.c
index 13df78f..f8a8aa6 100644 (file)
--- a/libavcodec/ppc/idct_altivec.c
+++ b/libavcodec/ppc/idct_altivec.c
@@ -38,6 +38,9 @@
 #include <stdlib.h>                                      /* malloc(), free() */
 #include <string.h>
 #include "../dsputil.h"
+
+#include "gcc_fixes.h"
+
 #include "dsputil_altivec.h"
 
 #define vector_s16_t vector signed short
@@ -152,24 +155,13 @@
     vx7 = vec_sra (vy7, shift);
 
 
-#ifdef CONFIG_DARWIN
-static const vector_s16_t constants[5] = {
-    (vector_s16_t)(23170, 13573, 6518, 21895, -23170, -21895, 32, 31),
-    (vector_s16_t)(16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725),
-    (vector_s16_t)(22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521),
-    (vector_s16_t)(21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692),
-    (vector_s16_t)(19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722)
-};
-#else
-// broken gcc
 static const vector_s16_t constants[5] = {
-    (vector_s16_t){23170, 13573, 6518, 21895, -23170, -21895, 32, 31},
-    (vector_s16_t){16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725},
-    (vector_s16_t){22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521},
-    (vector_s16_t){21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692},
-    (vector_s16_t){19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722}
+    (vector_s16_t) AVV(23170, 13573, 6518, 21895, -23170, -21895, 32, 31),
+    (vector_s16_t) AVV(16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725),
+    (vector_s16_t) AVV(22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521),
+    (vector_s16_t) AVV(21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692),
+    (vector_s16_t) AVV(19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722)
 };
-#endif
 
 void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
 {

diff --git a/libavcodec/ppc/mpegvideo_altivec.c b/libavcodec/ppc/mpegvideo_altivec.c
index 95558a6..bbf9c44 100644 (file)
--- a/libavcodec/ppc/mpegvideo_altivec.c
+++ b/libavcodec/ppc/mpegvideo_altivec.c
@@ -15,11 +15,14 @@
  * License along with this library; if not, write to the Free Software
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  */
- 
+
 #include <stdlib.h>
 #include <stdio.h>
 #include "../dsputil.h"
 #include "../mpegvideo.h"
+
+#include "gcc_fixes.h"
+ 
 #include "dsputil_altivec.h"
 
 // Swaps two variables (used for altivec registers)